You are on page 1of 313

Abstract and Applied Analysis

Fractional and Time-Scales


Differential Equations
Guest Editors: Dumitru Baleanu, Ali H. Bhrawy, Delfim F. M. Torres,
and Soheil Salahshour
Fractional and Time-Scales
Differential Equations
Abstract and Applied Analysis

Fractional and Time-Scales


Differential Equations

Guest Editors: Dumitru Baleanu, Ali H. Bhrawy,


Delfim F. M. Torres, and Soheil Salahshour
Copyright © 2014 Hindawi Publishing Corporation. All rights reserved.

This is a special issue published in “Abstract and Applied Analysis.” All articles are open access articles distributed under the Creative
Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original
work is properly cited.
Editorial Board
Ravi P. Agarwal, USA Jean M. Combes, France Jaume Giné, Spain
Bashir Ahmad, Saudi Arabia Monica Conti, Italy Valery Y. Glizer, Israel
M. O. Ahmedou, Germany Diego Córdoba, Spain Laurent Gosse, Italy
Nicholas D. Alikakos, Greece Juan C. Cortés, Spain Jean P. Gossez, Belgium
Debora Amadori, Italy Graziano Crasta, Italy Jose L. Gracia, Spain
Pablo Amster, Argentina Bernard Dacorogna, Switzerland Maurizio Grasselli, Italy
Douglas R. Anderson, USA Vladimir Danilov, Russia Qian Guo, China
Jan Andres, Czech Republic Mohammad T. Darvishi, Iran Yuxia Guo, China
Giovanni Anello, Italy Luis F. Pinheiro de Castro, Portugal Chaitan P. Gupta, USA
Stanislav Antontsev, Portugal T. Diagana, USA Uno Hämarik, Estonia
Mohamed Kamal Aouf, Egypt Jesús I. Dı́az, Spain Ferenc Hartung, Hungary
Narcisa C. Apreutesei, Romania Josef Diblı́k, Czech Republic Behnam Hashemi, Iran
Natig M. Atakishiyev, Mexico Fasma Diele, Italy Norimichi Hirano, Japan
Ferhan M. Atici, USA Tomas Dominguez, Spain Jiaxin Hu, China
Ivan G. Avramidi, USA Alexander I. Domoshnitsky, Israel Zhongyi Huang, China
Soohyun Bae, Korea Marco Donatelli, Italy Chengming Huang, China
Chuanzhi Bai, China Bo-Qing Dong, China Gennaro Infante, Italy
Zhanbing Bai, China Ondr̆ej Došlý, Czech Republic Ivan Ivanov, Bulgaria
Dumitru Baleanu, Turkey Wei-Shih Du, Taiwan Hossein Jafari, Iran
Józef Banaś, Poland Luiz Duarte, Brazil Jaan Janno, Estonia
Gerassimos Barbatis, Greece Roman Dwilewicz, USA Aref Jeribi, Tunisia
Martino Bardi, Italy Paul W. Eloe, USA Un C. Ji, Korea
Roberto Barrio, Spain Ahmed El-Sayed, Egypt Zhongxiao Jia, China
Feyzi Başar, Turkey Luca Esposito, Italy L. Jódar, Spain
Abdelghani Bellouquid, Morocco Jose A. Ezquerro, Spain Jong Soo Jung, Republic of Korea
Daniele Bertaccini, Italy Khalil Ezzinbi, Morocco Henrik Kalisch, Norway
Lucio Boccardo, Italy Dashan Fan, USA Hamid Reza Karimi, Norway
Igor Boglaev, New Zealand Angelo Favini, Italy Satyanad Kichenassamy, France
Martin J. Bohner, USA Marcia Federson, Brazil Tero Kilpeläinen, Finland
Geraldo Botelho, Brazil Julian F. Bonder, Argentina Sung Guen Kim, Republic of Korea
Elena Braverman, Canada S. Filippas, Equatorial Guinea Ljubisa Kocinac, Serbia
Romeo Brunetti, Italy Alberto Fiorenza, Italy Andrei Korobeinikov, Spain
Janusz Brzdek, Poland Tore Flȧtten, Norway Pekka Koskela, Finland
Detlev Buchholz, Germany Ilaria Fragala, Italy Victor Kovtunenko, Austria
Sun-Sig Byun, Korea Bruno Franchi, Italy Ren-Jieh Kuo, Taiwan
Fabio M. Camilli, Italy Xianlong Fu, China Pavel Kurasov, Sweden
Jinde Cao, China Massimo Furi, Italy Miroslaw Lachowicz, Poland
Anna Capietto, Italy Giovanni P. Galdi, USA Kunquan Lan, Canada
Jianqing Chen, China Isaac Garcia, Spain Ruediger Landes, USA
Wing-Sum Cheung, Hong Kong Jesús Garcı́a Falset, Spain Irena Lasiecka, USA
Michel Chipot, Switzerland José A. Garcı́a-Rodrı́guez, Spain Matti Lassas, Finland
Changbum Chun, Korea Leszek Gasinski, Poland Chun-Kong Law, Taiwan
Soon Y. Chung, Korea György Gát, Hungary Ming-Yi Lee, Taiwan
Jaeyoung Chung, Korea Vladimir Georgiev, Italy Gongbao Li, China
Silvia Cingolani, Italy Lorenzo Giacomelli, Italy Elena Litsyn, Israel
Yansheng Liu, China Somyot Plubtieng, Thailand Gabriel Turinici, France
Shengqiang Liu, China Milan Pokorny, Czech Republic Milan Tvrdy, Czech Republic
Carlos Lizama, Chile Sergio Polidoro, Italy Mehmet Unal, Turkey
Milton C. Lopes Filho, Brazil Ziemowit Popowicz, Poland Csaba Varga, Romania
Julian López-Gómez, Spain Maria M. Porzio, Italy Carlos Vazquez, Spain
Guozhen Lu, USA Enrico Priola, Italy Gianmaria Verzini, Italy
Jinhu Lü, China Vladimir S. Rabinovich, Mexico Jesus Vigo-Aguiar, Spain
Grzegorz Lukaszewicz, Poland Irena Rachu̇nková, Czech Republic Qing-Wen Wang, China
Wanbiao Ma, China Maria Alessandra Ragusa, Italy Yushun Wang, China
Shiwang Ma, China Simeon Reich, Israel Shawn X. Wang, Canada
Eberhard Malkowsky, Turkey Abdelaziz Rhandi, Italy Jing Ping Wang, UK
Salvatore A. Marano, Italy Hassan Riahi, Malaysia Youyu Wang, China
Cristina Marcelli, Italy Juan P. Rincón-Zapatero, Spain Peixuan Weng, China
Paolo Marcellini, Italy Luigi Rodino, Italy Noemi Wolanski, Argentina
Jesús Marı́n-Solano, Spain Yuriy V. Rogovchenko, Norway Ngai-Ching Wong, Taiwan
Jose M. Martell, Spain Julio D. Rossi, Argentina Patricia J. Y. Wong, Singapore
Mieczysław Mastyło, Poland Wolfgang Ruess, Germany Yong Hong Wu, Australia
Ming Mei, Canada Bernhard Ruf, Italy Zili Wu, China
Taras Mel’nyk, Ukraine Marco Sabatini, Italy Shanhe Wu, China
Anna Mercaldo, Italy Satit Saejung, Thailand Tie-cheng Xia, China
Changxing Miao, China Stefan G. Samko, Portugal Xu Xian, China
Stanislaw Migorski, Poland Martin Schechter, USA Yanni Xiao, China
Mihai Mihǎilescu, Romania Javier Segura, Spain Gonang Xie, China
Feliz Minhós, Portugal Sigmund Selberg, Norway Fuding Xie, China
Dumitru Motreanu, France Valery Serov, Finland Naihua Xiu, China
Roberta Musina, Italy N. Shahzad, Saudi Arabia Daoyi Xu, China
G. M. N’Guérékata, USA Andrey Shishkov, Ukraine Zhenya Yan, China
Maria Grazia Naso, Italy Stefan Siegmund, Germany Xiaodong Yan, USA
Sylvia Novo, Spain Abdel-Maksoud A. Soliman, Egypt Norio Yoshida, Japan
Micah Osilike, Nigeria Pierpaolo Soravia, Italy Beong In Yun, Korea
Mitsuharu Ôtani, Japan Marco Squassina, Italy Vjacheslav Yurko, Russia
Turgut Öziş, Turkey Svatoslav Staněk, Czech Republic Agacik Zafer, Turkey
Filomena Pacella, Italy Stevo Stevic, Serbia Sergey V. Zelik, UK
Nikolaos S. Papageorgiou, Greece Antonio Suárez, Spain Jianming Zhan, China
Sehie Park, Korea Wenchang Sun, China Meirong Zhang, China
Alberto Parmeggiani, Italy Robert Szalai, UK Weinian Zhang, China
Kailash C. Patidar, South Africa Sanyi Tang, China Chengjian Zhang, China
Kevin R. Payne, Italy Chun-Lei Tang, China Zengqin Zhao, China
Ademir Fernando Pazoto, Brazil Youshan Tao, China Sining Zheng, China
Josip E. Pečarić, Croatia Gabriella Tarantello, Italy Tianshou Zhou, China
Shuangjie Peng, China Nasser-eddine Tatar, Saudi Arabia Yong Zhou, China
Sergei V. Pereverzyev, Austria Gerd Teschke, Germany Qiji J. Zhu, USA
Maria Eugenia Perez, Spain Bevan Thompson, Australia Chun-Gang Zhu, China
Josefina Perles, Spain Sergey Tikhonov, Spain Malisa R. Zizovic, Serbia
Allan Peterson, USA Claudia Timofte, Romania Wenming Zou, China
Andrew Pickering, Spain Thanh Tran, Australia
Cristina Pignotti, Italy Juan J. Trujillo, Spain
Contents
Fractional and Time-Scales Differential Equations, Dumitru Baleanu, Ali H. Bhrawy, Delfim F. M. Torres,
and Soheil Salahshour
Volume 2014, Article ID 365250, 2 pages

Fractional Cauchy Problem with Riemann-Liouville Fractional Delta Derivative on Time Scales,
Jiang Zhu and Ying Zhu
Volume 2013, Article ID 401596, 19 pages
On Solutions to Fractional Discrete Systems with Sequential ℎ-Differences, Małgorzata Wyrwas,
Dorota Mozyrska, and Ewa Girejko
Volume 2013, Article ID 475350, 11 pages

Numerical Solution of Fuzzy Fractional Pharmacokinetics Model Arising from Drug Assimilation into
the Bloodstream, Ali Ahmadian, Norazak Senu, Farhad Larki, Soheil Salahshour, Mohamed Suleiman,
and Md. Shabiul Islam
Volume 2013, Article ID 304739, 17 pages

Numerical Solutions of Fractional Fokker-Planck Equations Using Iterative Laplace Transform Method,
Limei Yan
Volume 2013, Article ID 465160, 7 pages

The Second Noether Theorem on Time Scales, Agnieszka B. Malinowska and Natália Martins
Volume 2013, Article ID 675127, 14 pages

On Solutions of Linear Fractional Differential Equations with Uncertainty, T. Allahviranloo,


S. Abbasbandy, M. R. Balooch Shahryari, S. Salahshour, and D. Baleanu
Volume 2013, Article ID 178378, 13 pages

Fractional-Order Total Variation Image Restoration Based on Primal-Dual Algorithm,


Dali Chen, YangQuan Chen, and Dingyu Xue
Volume 2013, Article ID 585310, 10 pages

Local Observability of Systems on Time Scales, Zbigniew Bartosiewicz


Volume 2013, Article ID 810625, 9 pages
Positive Solutions Using Bifurcation Techniques for Boundary Value Problems of Fractional
Differential Equations, Yansheng Liu
Volume 2013, Article ID 162418, 7 pages

Approximation of Eigenvalues of Sturm-Liouville Problems by Using Hermite Interpolation,


M. M. Tharwat and S. M. Al-Harbi
Volume 2013, Article ID 412028, 14 pages

Approximate Solutions of Fisher’s Type Equations with Variable Coefficients, A. H. Bhrawy


and M. A. Alghamdi
Volume 2013, Article ID 176730, 10 pages

A Jacobi Collocation Method for Solving Nonlinear Burgers-Type Equations, E. H. Doha, D. Baleanu,
A. H. Bhrawy, and M. A. Abdelkawy
Volume 2013, Article ID 760542, 12 pages
Mappings for Special Functions on Cantor Sets and Special Integral Transforms via Local Fractional
Operators, Yang Zhao, Dumitru Baleanu, Mihaela Cristina Baleanu, De-Fu Cheng, and Xiao-Jun Yang
Volume 2013, Article ID 316978, 6 pages

New Wavelets Collocation Method for Solving Second-Order Multipoint Boundary Value Problems
Using Chebyshev Polynomials of Third and Fourth Kinds, W. M. Abd-Elhameed, E. H. Doha,
and Y. H. Youssri
Volume 2013, Article ID 542839, 9 pages

Numerical Solution of a Class of Functional-Differential Equations Using Jacobi Pseudospectral


Method, A. H. Bhrawy, M. A. Alghamdi, and D. Baleanu
Volume 2013, Article ID 513808, 9 pages
Numerical Solution of a Kind of Fractional Parabolic Equations via Two Difference Schemes,
Abdon Atangana and Dumitru Baleanu
Volume 2013, Article ID 828764, 8 pages

Existence Results for a Class of Fractional Differential Equations with Periodic Boundary Value
Conditions and with Delay, Hadi Karami, Azizollah Babakhani, and Dumitru Baleanu
Volume 2013, Article ID 176180, 8 pages

Application of Fuzzy Fractional Kinetic Equations to Modelling of the Acid Hydrolysis Reaction,
Ferial Ghaemi, Robiah Yunus, Ali Ahmadian, Soheil Salahshour, Mohamed Suleiman,
and Shanti Faridah Saleh
Volume 2013, Article ID 610314, 19 pages

A Modified Generalized Laguerre Spectral Method for Fractional Differential Equations on the Half
Line, D. Baleanu, A. H. Bhrawy, and T. M. Taha
Volume 2013, Article ID 413529, 12 pages

Numerical Modeling of Fractional-Order Biological Systems, Fathalla A. Rihan


Volume 2013, Article ID 816803, 11 pages

Oscillation Criteria for Fourth-Order Nonlinear Dynamic Equations on Time Scales, Xin Wu,
Taixiang Sun, Hongjian Xi, and Changhong Chen
Volume 2013, Article ID 740568, 11 pages

The Bernstein Operational Matrices for Solving the Fractional Quadratic Riccati Differential Equations
with the Riemann-Liouville Derivative, Dumitru Baleanu, Mohsen Alipour, and Hossein Jafari
Volume 2013, Article ID 461970, 7 pages

Fractional Dynamics of Genetic Algorithms Using Hexagonal Space Tessellation, J. A. Tenreiro Machado
Volume 2013, Article ID 739464, 7 pages

An Operational Matrix Based on Legendre Polynomials for Solving Fuzzy Fractional-Order Differential
Equations, Ali Ahmadian, Mohamed Suleiman, and Soheil Salahshour
Volume 2013, Article ID 505903, 29 pages

Two Efficient Generalized Laguerre Spectral Algorithms for Fractional Initial Value Problems,
D. Baleanu, A. H. Bhrawy, and T. M. Taha
Volume 2013, Article ID 546502, 10 pages
Local Fractional Series Expansion Method for Solving Wave and Diffusion Equations on Cantor Sets,
Ai-Min Yang, Xiao-Jun Yang, and Zheng-Biao Li
Volume 2013, Article ID 351057, 5 pages

Nonlinear Fractional Jaulent-Miodek and Whitham-Broer-Kaup Equations within Sumudu Transform,


Abdon Atangana and Dumitru Baleanu
Volume 2013, Article ID 160681, 8 pages
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2014, Article ID 365250, 2 pages
http://dx.doi.org/10.1155/2014/365250

Editorial
Fractional and Time-Scales Differential Equations

Dumitru Baleanu,1 Ali H. Bhrawy,2 Delfim F. M. Torres,3 and Soheil Salahshour4


1
Department of Mathematics and Computer Sciences, Cankaya University, TR-06810 Yenimahalle, Ankara, Turkey
2
Department of Mathematics, Faculty of Science, King Abdulaziz University, Jeddah 21589, Saudi Arabia
3
Department of Mathematics, University of Aveiro, 3810-193 Aveiro, Portugal
4
Young Researchers and Elite Club, Mobarakeh Branch, Islamic Azad University, Mobarakeh, Iran

Correspondence should be addressed to Dumitru Baleanu; dumitru@cankaya.edu.tr

Received 30 December 2013; Accepted 30 December 2013; Published 9 February 2014

Copyright © 2014 Dumitru Baleanu et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

The theory and applications of fractional differential equa- expansion method for wave and diffusion equations and
tions (FDEs) are gaining more relevance since they are used some mappings for special functions on Cantor sets were
extensively in the modeling of various processes in physics, presented. Application of fuzzy fractional kinetic equations
chemistry, engineering, and other areas of science. As it is to modeling of the acid hydrolysis reaction, the solutions
known, the behavior of dynamics of most complex systems of linear fractional differential equations with uncertainty,
of the real world phenomena has memory. Therefore, the and an operational matrix based on Legendre polynomials
modeling of dynamics of these types of systems by FDEs has for solving fuzzy fractional-order differential equations were
more advantages than the classical ones, in which such effects also investigated. A class of fractional-order differential
are neglected. On the other hand, the time-scales formalism models of biological systems with memory to model the
unifies the theories of difference and differential equations. interaction of immune system with tumor cells and with
Accordingly, the time-scales analysis constitutes a good tool HIV infection of CD4+ T-cells was reported. Fractional-
to study both discrete and continuous systems. order total variation image restoration based on primal-dual
Advanced analytical and numerical techniques and com- algorithm and the fractional dynamics of genetic algorithms
putational methods are of important interest for classical, using hexagonal space tessellation were pointed out. The
fractional, fuzzy fractional, and time-scales differential equa- positive solution using bifurcation techniques for boundary
tions. value problems of fractional differential equations was a
The papers of this special issue contain some new algo- contribution of our special issue. The existence results for a
rithms and techniques designed to investigate classical, frac- class of fractional differential equations with boundary value
tional, fractal, fuzzy fractional, and time-scales differential conditions and with delay, a modified generalized Laguerre
equations of general interest. New insights of existence and spectral methods for fractional differential equations on the
uniqueness theorems of some differential equations were also half line, a Jacobi collocation method for solving nonlin-
presented. ear Burgers-type equations, and new wavelets collocation
In the following we summarize briefly the content of method for solving second-order multipoint boundary value
the special issue. The second Noether theorem, the local problems using Chebyshev polynomials of third and fourth
observability of systems, and the fractional Cauchy problem kinds were shown. The investigation of the nonlinear frac-
within Riemann-Liouville fractional delta derivative on time tional Jaulent-Miodek and Whitham-Broer-Kaup equations
scales were reported and the solutions of fractional discrete within Sumudu transform, the Bernstein operational matri-
systems with sequential h-differences were obtained. Besides, ces applied for solving the fractional quadratic Riccati dif-
the oscillation criteria for fourth-order nonlinear dynamic ferential equations, and the approximate solutions of Fisher’s
equations on time scales were depicted. Local fractional series type equations with variable coefficients are topics covered by
2 Abstract and Applied Analysis

our special issue. Another part of contributions was focussing


on two efficient generalized Laguerre spectral algorithms for
some fractional initial value problems, the approximation of
eigenvalues of Sturm-Liouville problems by using Hermite
interpolation and the presentation of the numerical solutions
of fractional Fokker-Planck equations by using the iterative
Laplace transform method.
Dumitru Baleanu
Ali H. Bhrawy
Delfim F. M. Torres
Soheil Salahshour
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 401596, 19 pages
http://dx.doi.org/10.1155/2013/401596

Research Article
Fractional Cauchy Problem with Riemann-Liouville
Fractional Delta Derivative on Time Scales

Jiang Zhu and Ying Zhu


School of Mathematics and Statistics, Jiangsu Normal University, Xuzhou 221116, China

Correspondence should be addressed to Jiang Zhu; jzhuccy@163.com

Received 9 August 2013; Accepted 6 October 2013

Academic Editor: Ali H. Bhrawy

Copyright © 2013 J. Zhu and Y. Zhu. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The Δ-power function and fractional Δ-integrals and fractional Δ-differential are defined, and then the definitions and properties
of Δ-Mittag-Leffler function are given. The properties of fractional Δ-integrals and fractional Δ-differential on time scales are
discussed in detail. After that, the existence of the solution and the dependency of the solution upon the initial value for Cauchy
type problem with fractional Δ-derivative are studied. Also the explicit solutions to homogeneous fractional Δ-differential equations
and nonhomogeneous fractional Δ-differential equations are derived by using Laplace transform method.

1. Introduction Δ-derivative on time scales by the inverse of Laplace trans-


form in [17].
The fractional differential equation theory is an important Inspired by these works, the aim of this paper is to give
subject of mathematics, which includes continuous frac- a new definition of fractional Δ-integral and Δ-derivative on
tional differential equations and discrete fractional difference general time scales and then study some fractional differential
equations. The theory of fractional differential equations equations on time scales. To define the fractional Δ-integral
has gained considerable popularity and importance dur- and fractional Δ-derivative, we would need to obtain a
ing the past three decades or so. Many applications in definition of fractional order power functions on time scales
numerous seemingly diverse and widespread fields of science to generalize the monomials. Different from definition of ∇-
and engineering have been gained. It does indeed provide power functions by axiomatization method in [16], we define
several potentially useful tools for solving differential and fractional Δ-power functions on general time scales by using
integral equations and various other problems involving inversion of time scale Laplace transform and shift transform
special functions of mathematical physics as well as their in Section 3, and Riemann-Liouville fractional Δ-integral and
extensions and generalizations in one and more variables. Riemann-Liouville fractional Δ-derivative on general time
About these advances, one can refer to [1, 2], the books scales are also given. In Section 4, we present the properties
[3, 4], and the references of them. For the recent devel- of fractional Δ-integrals and fractional Δ-differential on
opments about continuous fractional differential equations time scales. Then in Section 5, Cauchy type problem with
and discrete fractional difference equations, one can refer Riemann-Liouville fractional Δ-derivative is discussed. In
to [5–11]. To unify differential equations and difference Section 6, for the Riemann-Liouville fractional Δ-differential
equations, Higer proposed firstly the time scale and built the initial value problem, we discuss the dependency of the
relevant basic theories (see [12–15]). Recently, some authors solution upon the initial value. In Section 7, by applying
studied fractional calculus on time scales (see [16, 17]), where the Laplace transform method, we derive explicit solutions
Williams [16] gives a definition of fractional ∇-integral and ∇- to homogeneous fractional Δ-differential equations with
derivative on time scales to unify three cases of specific time constant coefficients. In Section 8, we also use the Laplace
scales. Bastos gives definitions of fractional Δ-integral and transform method to find particular solutions and general
2 Abstract and Applied Analysis

solutions of the corresponding fractional Δ-differential non- is given by


homogeneous equations.
𝑡
𝑦 (𝑡) = ∫ 𝑦 (𝑡, 𝑠) 𝑓 (𝑠) Δ𝑠, (7)
𝑡0
2. Preliminaries
where 𝑦(𝑡, 𝑠) is the Cauchy function for
First, we present some preliminaries about time scales in [12].
𝑛
𝑛 𝑛−𝑖
Definition 1 (see [12]). A time scale T is a nonempty closed 𝐿𝑦 = 0, 𝑤ℎ𝑒𝑟𝑒 𝐿𝑦 = 𝑦Δ + ∑𝑝𝑖 𝑦Δ . (8)
subset of the real numbers. Throughout this paper, T or 𝑖=1
T𝑖 (𝑖 = 1, 2, . . . , 𝑛) denotes a time scale.
Definition 8 (see [2]). The factorial polynomial is defined as
Definition 2 (see [12]). Let T be a time scale. For 𝑡 ∈ T one 𝑛−1
defines the forward jump operator 𝜎 : T → T by 𝜎(𝑡) := Γ (𝑡 + 1)
(𝑡)(𝑛) = ∏ (𝑡 − 𝑗) = . (9)
inf{𝑠 ∈ T : 𝑠 > 𝑡}, while the backward jump operator 𝜌 : T → 𝑗=0 Γ (𝑡 + 1 − 𝑛)
T is defined by 𝜌(𝑡) := sup{𝑠 ∈ T : 𝑠 < 𝑡}. If 𝜎(𝑡) > 𝑡, one says
that 𝑡 is right-scattered, while if 𝜌(𝑡) < 𝑡, one says that 𝑡 is left- For arbitrary ], define
scattered. Points that are right-scattered and left-scattered at
the same time are called isolated. Also, if 𝑡 < sup T and 𝜎(𝑡) = Γ (𝑡 + 1)
𝑡(]) = , (10)
𝑡, then 𝑡 is called right-dense, and if 𝑡 > inf T and 𝜌(𝑡) = 𝑡, Γ (𝑡 + 1 − 𝜐)
then 𝑡 is called left-dense.
where Γ denotes gamma function (see [3]).
Definition 3 (see [12]). A function 𝑓 : T → R is called
Definition 9 (see [12]). One says that a function 𝑝 : T → R
regulated provided that its right-sided limits exist (finite) at
is regressive provided that
all right-dense points in T and its left-sided limits exist (finite)
at all left-dense points in T. 1 + 𝜇 (𝑡) 𝑝 (𝑡) ≠
0 (11)
Definition 4 (see [12]). Let ℎ𝑘 : T 2 → R, 𝑘 ∈ N0 be defined for all 𝑡 ∈ T 𝑘 holds. The set of all regressive and rd-continuous
by functions 𝑓 : T → R will be denoted by
ℎ0 (𝑡, 𝑠) = 1 ∀𝑡, 𝑠 ∈ T (1) R = R (T) = R (T, R) . (12)
and then recursively by Theorem 10 (see [12]). If 𝑝 ∈ R, then the function ⊖𝑝 defined
𝑡
by
ℎ𝑘+1 (𝑡, 𝑠) = ∫ ℎ𝑘 (𝜏, 𝑠) Δ𝜏 ∀𝑡, 𝑠 ∈ T. (2)
𝑝 (𝑡)
𝑠
⊖𝑝 (𝑡) := − ∀𝑡 ∈ T 𝑘 (13)
1 + 𝜇 (𝑡) 𝑝 (𝑡)
Definition 5 (see [12]). One defines the Cauchy function 𝑦 :
𝑛
T × T 𝑘 → R for the linear dynamic equation is also an element of R.

𝑛
𝑛 𝑛−𝑖 Definition 11 (see [12]). If 𝑝 ∈ R, then one defines the Δ-
𝐿𝑦 = 0, where 𝐿𝑦 = 𝑦Δ + ∑𝑝𝑖 𝑦Δ , (3) exponential function by
𝑖=1

𝑛 𝑒𝑝 (𝑡, 𝑠)
to be for each fixed 𝑠 ∈ T 𝑘 the solution of the initial value
𝑡
problem 1
= exp (∫ Log (1 + 𝜇 (𝜏) 𝑝 (𝜏)) Δ𝜏) for 𝑠, 𝑡 ∈ T.
𝑠 𝜇 (𝜏)
Δ𝑖
𝐿𝑦 = 0, 𝑦 (𝜎 (𝑠) , 𝑠) = 0, 0 ≤ 𝑖 ≤ 𝑛 − 2, (14)
(4)
𝑛−1
𝑦Δ (𝜎 (𝑠) , 𝑠) = 1. Definition 12 (see [12]). If 𝑝 ∈ R, then the first order linear
dynamic equation
Remark 6 (see [12]). Note that
𝑦Δ = 𝑝 (𝑡) 𝑦 (15)
𝑦 (𝑡, 𝑠) := ℎ𝑛−1 (𝑡, 𝜎 (𝑠)) (5)
is called regressive.
𝑛
is the Cauchy function of 𝑦Δ = 0.
Theorem 13 (see [12]). Suppose that (15) is regressive and fix
Theorem 7 (variation of constants [12]). Let 𝑓 ∈ 𝐶rd ; then the 𝑡0 ∈ T. Then 𝑒𝑝 (⋅, 𝑡0 ) is a solution of the initial value problem
solution of the initial value problem
𝑦Δ = 𝑝 (𝑡) 𝑦, 𝑦 (𝑡0 ) = 1 (16)
Δ𝑖
𝐿𝑦 = 𝑓 (𝑡) , 𝑦 (𝑡0 ) = 0, 0 ≤ 𝑖 ≤ 𝑛 − 1, (6) on T.
Abstract and Applied Analysis 3

Theorem 14 (see [12]). If 𝑝 ∈ R, then Theorem 22 (see [18]). If 𝑓 and 𝑔 are infinitely often Δ-
differentiable, then for all 𝑘 ∈ 𝑁0
𝑒𝑝 (𝜎 (𝑡) , 𝑠) := 𝑒𝑝𝜎 (𝑡, 𝑠) = (1 + 𝜇 (𝑡) 𝑝 (𝑡)) 𝑒𝑝 (𝑡, 𝑠) . (17)
𝑘−1
Δ𝑘 𝑘 ] 𝑘−1−]
(𝑓 ∗ 𝑔) = 𝑓Δ ∗ 𝑔 + ∑ 𝑓Δ (𝑡0 ) 𝑔Δ
Definition 15 (see [12]). Assume that 𝑥 : T → R is regulated. ]=0
Then the Δ-Laplace transform of 𝑥 is defined by
𝑘
𝑘−1 ] 𝑘−1−]

𝜎
= 𝑓 ∗ 𝑔 Δ + ∑ 𝑓 Δ 𝑔Δ (𝑡0 ) , (24)
L {𝑥} (𝑧, 𝑡0 ) = ∫ 𝑥 (𝑡) 𝑒⊖𝑧 (𝑡, 𝑡0 ) Δ𝑡 (18) ]=0
𝑡0
𝑘−1
Δ𝑘 ] 𝑘−1−]
for 𝑧 ∈ D{𝑥}, where D{𝑥} consists of all complex numbers (𝑓 ∗ 𝑔) (𝑡0 ) = ∑ 𝑓Δ (𝑡0 ) 𝑔Δ (𝑡0 ) .
𝑧 ∈ C for which the improper integral exists. ]=0

Theorem 23 (convolution theorem [18]). Suppose that 𝑓, 𝑔 :


Definition 16 (uniqueness of the inverse [12]). If the functions
T → R are locally Δ-integrable functions on T and their
𝑓 : T → R and 𝑔 : T → R have the same Laplace transform,
convolution 𝑓 ∗ 𝑔 is defined by (20). Then,
then 𝑓 = 𝑔.
L {𝑓 ∗ 𝑔} (𝑧)
In order to give fractional integral and derivative on (25)
a time scale, we need to define fractional power function = L {𝑓} (𝑧) ⋅ L {𝑔} (𝑧) , 𝑧 ∈ D {𝑓} ∩ D {𝑔} .
ℎ𝛼 (𝑡, 𝑠) which is derived by the inverse of Laplace transform
and is introduced in the following section. Before this, we Theorem 24 (see [12]). Assume that 𝑥 : T → C is a mapping,
𝑘
need definitions of shift and convolution and some properties such that 𝑥Δ is regulated. Then
about convolution, such as convolution theorem and associa-
tivity, which are introduced in [18]. 𝑘
𝑘−1 𝑖
L {𝑥Δ } (𝑧, 𝑡0 ) = 𝑧𝑘 L {𝑥} (𝑧, 𝑡0 ) − ∑ 𝑧𝑘−𝑖−1 𝑥Δ (𝑡0 ) (26)
Definition 17 (see [18]). Let T be a time scale that sup T = ∞ 𝑖=0

and fix 𝑡0 ∈ T. For a given 𝑓 : [𝑡0 , ∞)T → C, the solution of for those regressive 𝑧 ∈ C satisfying
the shifting problem
𝑖
lim {𝑥Δ (𝑡) 𝑒⊖𝑧 (𝑡, 𝑡0 )} = 0, 𝑖 = 0, 1, . . . , 𝑘 − 1. (27)
Δ𝑡 Δ𝑠 𝑡→∞
𝑢 (𝑡, 𝜎 (𝑠)) = −𝑢 (𝑡, 𝑠) , 𝑡, 𝑠 ∈ T, 𝑡 ≥ 𝑠 ≥ 𝑡0 ,
(19) Theorem 25 (see [12]). Assume that ℎ𝑘 (𝑡, 𝑡0 ), 𝑘 ∈ N0 are
𝑢 (𝑡, 𝑡0 ) = 𝑓 (𝑡) , 𝑡 ∈ T, 𝑡 ≥ 𝑡0 defined as in Definition 4. Then

is denoted by 𝑓̂ and is called the shift (or delay) of 𝑓. 1


L {ℎ𝑘 (⋅, 𝑡0 )} (𝑧, 𝑡0 ) = (28)
𝑧𝑘+1
Example 18 (see [18]). Consider ℎ̂
𝑘 (⋅, 𝑟)(𝑡, 𝑠) = ℎ𝑘 (𝑡, 𝑠) for all for those regressive 𝑧 ∈ C satisfying
𝑡, 𝑠 ∈ T, independent of 𝑟.
lim {ℎ𝑘 (𝑡, 𝑡0 ) 𝑒⊖𝑧 (𝑡, 𝑡0 )} = 0. (29)
𝑡→∞
Definition 19 (see [18]). For given functions 𝑓, 𝑔 : T → R,
their convolution 𝑓 ∗ 𝑔 is defined by
3. Δ-Power Function and Fractional Integral
𝑡 and Derivative on Time Scales
(𝑓 ∗ 𝑔) (𝑡) = ∫ 𝑓̂ (𝑡, 𝜎 (𝑠)) 𝑔 (𝑠) Δ𝑠, (20)
𝑡0
In this section, inspired by property of ℎ𝑘 (⋅, 𝑡0 ) in Theorem 25
for 𝑘 ∈ N0 , we define fractional Δ-power functions ℎ𝛼 (𝑡, 𝑠)
where 𝑓̂ is the shift of 𝑓 introduced in Definition 17. for 𝛼 ∈ R by using inversion of Δ-Laplace transform and give
definitions of fractional integral and derivative on time scales.
Theorem 20 (associativity of the convolution [18]). The
convolution is associative; that is, Definition 26. One defines fractional generalized Δ-power
function ℎ𝛼 (𝑡, 𝑡0 ) on time scales
(𝑓 ∗ 𝑔) ∗ ℎ = 𝑓 ∗ (𝑔 ∗ ℎ) . (21)
1
ℎ𝛼 (𝑡, 𝑡0 ) = L−1 { } (𝑡) (30)
Theorem 21 (see [18]). If 𝑓 is delta differentiable, then 𝑧𝛼+1

Δ to those suitable regressive 𝑧 ∈ C \ {0} such that L−1 exist


(𝑓 ∗ 𝑔) = 𝑓Δ ∗ 𝑔 + 𝑓 (𝑡0 ) 𝑔, (22) for 𝛼 ∈ R, 𝑡 ≥ 𝑡0 . Fractional generalized Δ-power function
ℎ𝛼 (𝑡, 𝑠) on time scales is defined as the shift of ℎ𝛼 (𝑡, 𝑡0 ); that
and if 𝑔 is delta differentiable, then is,

ℎ𝛼 (𝑡, 𝑠) = ℎ̂
Δ
(𝑓 ∗ 𝑔) = 𝑓 ∗ 𝑔Δ + 𝑓𝑔 (𝑡0 ) . (23) 𝛼 (⋅, 𝑡0 ) (𝑡, 𝑠) (𝑡 ≥ 𝑠 ≥ 𝑡0 ) . (31)
4 Abstract and Applied Analysis

Applying the initial value theorem of Laplace transform satisfy


(see, e.g., [15, Theorem 1.3], for 𝛼 > 0, we have
̂
𝜕ℎ𝛼−1 (⋅, 𝑎) (𝑥, 𝑡) 𝜕ℎ ̂ (⋅, 𝑎) (𝑥, 𝑡)
1 = − 𝛼−1
ℎ𝛼 (𝑡0 , 𝑡0 ) = lim 𝑧 ⋅ = 0. (32) 𝜕𝑥 𝜕𝑡 (40)
𝑧→∞ 𝑧𝛼+1
̂
ℎ𝛼−1 (⋅, 𝑎) (𝑥, 𝑎) = ℎ𝛼−1 (𝑥, 𝑎) .
Theorem 27. For 𝛼, 𝛽 ∈ R, one has
When T = Z, ℎ𝛼−1 (𝑥, 𝑎) = (𝑥 − 𝑎)(𝛼−1) /Γ(𝛼), according
(ℎ𝛼 (⋅, 𝑡0 ) ∗ ℎ𝛽 (⋅, 𝑡0 )) (𝑡) = ℎ𝛼+𝛽+1 (𝑡, 𝑡0 ) . (33) to Definition 17,

Proof. According to convolution theorem, ̂ (𝑥 − 𝑡)(𝛼−1)


ℎ𝛼−1 (⋅, 𝑎) (𝑥, 𝑡) = (41)
Γ (𝛼)
L {(ℎ𝛼 (⋅, 𝑡0 ) ∗ ℎ𝛽 (⋅, 𝑡0 )) (𝑡)} (𝑧, 𝑡0 )
satisfy
= L {ℎ𝛼 (𝑡, 𝑡0 )} (𝑧, 𝑡0 ) L {ℎ𝛽 (𝑡, 𝑡0 )} (𝑧, 𝑡0 ) ̂ ̂
ℎ𝛼−1 (⋅, 𝑎) (𝑥 + 1, 𝑡 + 1) − ℎ𝛼−1 (⋅, 𝑎) (𝑥, 𝑡 + 1)
(34)
1 1 1
= ⋅ = ̂
= −ℎ𝛼−1 ̂
(⋅, 𝑎) (𝑥, 𝑡 + 1) + ℎ𝛼−1 (⋅, 𝑎) (𝑥, 𝑡) , (42)
𝑧𝛼+1 𝑧𝛽+1 𝑧𝛼+𝛽+2
= L {ℎ𝛼+𝛽+1 (𝑡, 𝑡0 )} (𝑧, 𝑡0 ) . ̂
ℎ𝛼−1 (⋅, 𝑎) (𝑥, 𝑎) = ℎ𝛼−1 (𝑥, 𝑎) .
As an especial case of Definition 28, we have the following
By the uniqueness of inverse transform for Laplace transform,
examples.
we obtain

(ℎ𝛼 (⋅, 𝑡0 ) ∗ ℎ𝛽 (⋅, 𝑡0 )) (𝑡) Example 29 (see [3]). When T = R, the fractional Δ-integral
of order 𝛼 is defined by
𝑡 (35)
= ∫ ℎ̂
𝑥
𝛼 (⋅, 𝑡0 ) (𝑡, 𝜎 (𝜏)) ℎ𝛽 (𝜏, 𝑠) Δ𝜏 = ℎ𝛼+𝛽+1 (𝑡, 𝑡0 ) . 𝛼 1 1
(𝐼𝑎+ 𝑓) (𝑥) := ∫ 𝑓 (𝑡) 𝑑𝑡 (𝑥 > 𝑎) . (43)
𝑡0 Γ (𝛼) 𝑎 (𝑥 − 𝑡)1−𝛼
Example 30. When T = Z, Consider the following.
Moreover, if we take 𝛼 = 0, then
𝑡
(1) The 𝑛th integral of 𝑓 is defined by
(1 ∗ ℎ𝛽 ) (𝑡, 𝑡0 ) = ∫ ℎ𝛽 (𝜏, 𝑠) Δ𝜏 = ℎ𝛽+1 (𝑡, 𝑡0 ) . (36)
−𝑛 1 𝑡−𝑛
𝑡0
𝑎 Δ 𝑥 (𝑡) = ∑(𝑡 − 𝜎 (𝑠))(𝑛−1) 𝑥 (𝑠) . (44)
Γ (𝑛) 𝑠=𝑎
That is,
Δ Here 𝑡 − 𝑛 ≥ 𝑎, 𝜎(𝑠) = 𝑠 + 1.
ℎ𝛽+1 (𝑡, 𝑡0 ) = ℎ𝛽 (𝑡, 𝑡0 ) . (37)
Note that power function (𝑡 − 𝜎(𝑠))(𝑛−1) /Γ(𝑛) vanishes
Now, we will give the definitions of fractional Δ-integral at 𝑠 ≥ 𝑡 − 𝑛 + 1. So
and Δ-derivative which are the main context in this section.
−𝑛 1 𝑡−1
𝑎 Δ 𝑥 (𝑡) = ∑(𝑡 − 𝜎 (𝑠))(𝑛−1) 𝑥 (𝑠)
Definition 28. Let Ω be a finite interval on a time scale T, Γ (𝑛) 𝑠=𝑎
𝑡0 , 𝑡 ∈ Ω. For 𝛼 ≥ 0 and for a function 𝑓 : T → R, the
Riemann-Liouville fractional Δ-integral of order 𝛼 is defined 1 𝑡−𝑛
= ∑(𝑡 − 𝜎 (𝑠))(𝑛−1) 𝑥 (𝑠) , for 𝑡 − 𝑛 ≥ 𝑎.
0
by 𝐼Δ,𝑡 𝑓(𝑡) = 𝑓(𝑡) and Γ (𝑛) 𝑠=𝑎
0
(45)
𝛼
(𝐼Δ,𝑡 𝑓) (𝑡) = (ℎ𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓) (𝑡)
0
(2) The 𝛾th fractional sum of 𝑓 is defined by
𝑡
̂
= ∫ ℎ𝛼−1 (⋅, 𝑡0 ) (𝑡, 𝜎 (𝜏)) 𝑓 (𝜏) Δ𝜏 −𝛾 1 𝑡−1
𝑡0 (38)
𝑎Δ 𝑥 (𝑡) = ∑(𝑡 − 𝜎 (𝑠))(𝛾−1) 𝑥 (𝑠) ,
Γ (𝛾) 𝑠=𝑎 (46)
𝑡
= ∫ ℎ𝛼−1 (𝑡, 𝜎 (𝜏)) 𝑓 (𝜏) Δ𝜏, for 𝑡 ≥ 𝑎 − [−𝛾] .
𝑡0

for 𝛼 > 0, 𝑡 > 𝑡0 . Definition 31. Let 𝛼 ≥ 0, 𝑚 = [𝛼] + 1, and 𝑓 : T → R.


𝑚
When T = R, ℎ𝛼−1 (𝑥, 𝑎) = (𝑥 − 𝑎)𝛼−1 /Γ(𝛼), according to For 𝑠, 𝑡 ∈ T 𝑘 with 𝑠 < 𝑡, the Riemann-Liouville fractional
Definition 17, Δ-derivative of order 𝛼 is defined by the expression
𝛼
(𝑥 − 𝑡)𝛼−1 𝐷Δ,𝑠 𝑓 (𝑡) := 𝐷Δ𝑚 𝐼Δ,𝑠
𝑚−𝛼
𝑓 (𝑡) , (47)
̂
ℎ𝛼−1 (⋅, 𝑎) (𝑥, 𝑡) = (39)
Γ (𝛼) if it exists.
Abstract and Applied Analysis 5

𝑛
Throughout this paper, we denote 𝑓Δ = 𝐷Δ𝑛 𝑓 = 𝐷Δ,𝑠 𝑛
𝑓, By differentiating 𝑘 times with respect to 𝜆 on both sides
𝛼 −𝛼 𝛼
𝑛 ∈ N, and, for 𝛼 < 0, 𝐷Δ,𝑠 𝑓(𝑡) means 𝐼Δ,𝑠 𝑓(𝑡) and 𝐼Δ,𝑠 𝑓(𝑡) of the formula in Theorem 35, we get the following result:
−𝛼
means 𝐷Δ,𝑠 𝑓(𝑡).
𝜕𝑘 𝑘!𝑧𝛼−𝛽
Finally, we present the definition of Δ-Mittag-Leffler L{ 𝑘 Δ 𝐹𝛼,𝛽 (𝜆, 𝑡, 𝑡0 )} (𝑧, 𝑡0 ) = . (51)
function which is an important tool for solving fractional 𝜕𝜆 (𝑧𝛼 − 𝜆)𝑘+1
difference equation.
4. Properties of Fractional Δ-Integral and
Definition 32. Δ-Mittag-Leffler function is defined by Δ-Derivative on Time Scales

In this section, we mainly give the properties of fractional Δ-
Δ 𝐹𝛼,𝛽 (𝜆, 𝑡, 𝑡0 ) = ∑𝜆𝑗 ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 ) , (48) integral and Δ-derivative on time scales which are needed in
𝑗=0
the following sections.
provided that the right series is convergent, where 𝛼, 𝛽 > 0, Theorem 36. Let 𝛼 > 0, 𝑚 = [𝑎] + 1, and 𝛽 ∈ R. Then
𝜆 ∈ R.
𝛼
(1) 𝐼Δ,𝑡 ℎ (𝑡, 𝑡0 ) = ℎ𝛽+𝛼−1 (𝑡, 𝑡0 ) ,
0 𝛽−1
Example 33. When 0 < 𝜆 < 1, for any 𝑗, |ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 )| < (52)
𝛼
𝑀, we can obtain that the series ∑∞ 𝑗 (2) 𝐷Δ,𝑡 ℎ (𝑡, 𝑡0 ) = ℎ𝛽−𝛼−1 (𝑡, 𝑡0 ) .
𝑗=0 𝜆 ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 ) is 0 𝛽−1

convergent. Proof. (1) According to Definition 28 and Theorem 27, we


have
Example 34. When T = R, ∑∞
𝜆 ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 ) =
𝑗=0
𝑗
𝛼
𝐼Δ,𝑡 ℎ (𝑡, 𝑡0 )
∑∞
𝑗=0
𝑗
𝜆 ((𝑡 − 𝑡0 ) 𝑗𝛼+𝛽−1 ∞
/Γ(𝑗𝛼 + 𝛽)). Since ∑𝑗=0 𝜆𝑗 ((𝑡 − 𝑡0 )𝑗𝛼+𝛽−1 / 0 𝛽−1
(53)
Γ(𝑗𝛼 + 𝛽)) is convergent for any 𝑡 ≥ 𝑡0 , ∑∞ 𝑗
𝑗=0 𝜆 ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 ) is = (ℎ𝛼−1 (⋅, 𝑡0 ) ∗ ℎ𝛽−1 (⋅, 𝑡0 )) (𝑡) = ℎ𝛼+𝛽−1 (𝑡, 𝑡0 ) .
convergent. That is, Δ 𝐹𝛼,𝛽 (𝜆, 𝑡, 𝑡0 ) is defined as 𝑡 ≥ 𝑡0 .
(2) By Definition 31, it is obtained that
As to the Laplace transform of Δ-Mittag-Leffler function, 𝛼
we have the following theorem. 𝐷Δ,𝑡 ℎ
0 𝛽−1
(𝑡, 𝑡0 ) = 𝐷Δ𝑚 𝐼Δ,𝑡
𝑚−𝛼
0
ℎ𝛽−1 (𝑡, 𝑡0 ) = 𝐷𝑚 ℎ𝑚+𝛽−𝛼−1 (𝑡, 𝑡0 ) .
(54)
Theorem 35. The Laplace transform of Δ-Mittag-Leffler func-
tion is Then
𝐷𝑚 ℎ𝑚+𝛽−𝛼−1 (𝑡, 𝑡0 ) = ℎ𝛽−𝛼−1 (𝑡, 𝑡0 ) . (55)
𝑧𝛼−𝛽
L {Δ 𝐹𝛼,𝛽 (𝜆, 𝑡, 𝑡0 )} (𝑧, 𝑡0 ) = 𝛼 (|𝜆| < |𝑧|𝛼 ) . (49)
𝑧 −𝜆

Proof. According to the definition of Laplace transform, it is In particular, if 𝛽 = 1, 𝛼 > 0, then the Riemann-Liouville
obtained that fractional Δ-derivatives of a constant are, in general, not equal
to zero:
L {Δ 𝐹𝛼,𝛽 (𝜆, 𝑡, 𝑡0 )} (𝑧, 𝑡0 ) 𝛼
𝐷Δ,𝑡 1 = ℎ−𝛼 (𝑡, 𝑡0 ) (0 < 𝛼 < 1) . (56)
0

𝜎 On the other hand, for 𝑗 = 1, 2, . . . , 𝑚,
=∫ 𝐹𝛼,𝛽 (𝜆, 𝑡, 𝑡0 ) ⋅ 𝑒⊖𝑧 (𝑡, 𝑡0 ) Δ𝑡
𝑡0 Δ 𝛼
𝐷Δ,𝑡 ℎ (𝑡, 𝑠) = 0.
0 𝛼−𝑗
(57)
∞ ∞
= ∫ ∑𝜆𝑗 ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 ) ⋅ 𝑒⊖𝑧
𝜎
(𝑡, 𝑡0 ) Δ𝑡 In fact,
𝑡0 𝑗=0
𝛼
𝐷Δ,𝑡 ℎ (𝑡, 𝑠) = 𝐷Δ𝑚 𝐼Δ,𝑡
0 𝛼−𝑗
𝑚−𝛼
0
ℎ𝛼−𝑗 (𝑡, 𝑠)
∞ ∞ (58)
= ∑ 𝜆𝑗 ∫ ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 ) ⋅ 𝑒⊖𝑧
𝜎
(𝑡, 𝑡0 ) Δ𝑡 = 𝐷Δ𝑚 ℎ𝑚−𝑗 (𝑡, 𝑠) = 0.
𝑗=0 𝑡0 (50)
From Theorem 36, we derive the following result in [3] when

𝑗 T = R.
= ∑ 𝜆 L {ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 )} (𝑧, 𝑡0 )
𝑗=0
Corollary 37 (see [3]). If 𝛼 ≥ 0 and 𝛽 > 0, then
∞ ∞
1 𝑗 Γ (𝛽)
= ∑ 𝜆𝑗 = 𝑧−𝛽 ∑𝜆𝑗 (𝑧−𝛼 ) 𝛼
(𝐼𝑎+ (𝑡 − 𝑎)𝛽−1 ) (𝑥) = (𝑥 − 𝑎)𝛽+𝛼−1 (𝛼 > 0) ,
𝑗=0 𝑧𝑗𝛼+𝛽 𝑗=0 Γ (𝛽 + 𝛼)

𝑧𝛼−𝛽 𝛼 Γ (𝛽)
= (|𝜆| < |𝑧|𝛼 ) . (𝐷𝑎+ (𝑡 − 𝑎)𝛽−1 ) (𝑥) = (𝑥 − 𝑎)𝛽−𝛼−1 (𝛼 ≥ 0) .
𝑧𝛼 − 𝜆 Γ (𝛽 − 𝛼)
(59)
6 Abstract and Applied Analysis

In particular, if 𝛽 = 1 and 𝛼 ≥ 0, then the Riemann-Liouville Lemma 40. (1) For 𝛼 > 0, 𝑛 = [𝛼]+1, let 𝑓 be a function which
𝑛 𝑛
fractional derivatives of a constant are, in general, not equal to is 𝑛 times Δ-differentiable on T 𝑘 with 𝑓Δ rd-continuous over
zero: T, and it is valid that
𝛼 (𝑥 − 𝑎)−𝛼 𝛼
(𝐷𝑎+ 1) (𝑥) = (0 < 𝛼 < 1) . (60) 𝐼Δ,𝑡 𝑓 (𝑡)
Γ (1 − 𝛼) 0

𝑛−1
On the other hand, for 𝑗 = 1, 2, . . . , [𝛼] + 1, 𝑘 𝑛
= ∑ ℎ𝑘+𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) + (ℎ𝑛+𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) .
𝛼
(𝐷𝑎+ (𝑡 − 𝑎)𝛼−𝑗 ) (𝑥) = 0. (61) 𝑘=0
(70)
As to the fractional sum and difference, we have the
following result, which is an improvement of Lemma 3.1 in (2) For 𝛼 ≥ 0, 𝑛 = [𝛼] + 1, let 𝑓 be a function which is
𝑛 𝑛
[1]. 𝑛 times Δ-differentiable on T 𝑘 with 𝑓Δ rd-continuous over T
𝛼
and 𝐷Δ,𝑡 𝑓 exists almost on T, and it is valid that
Corollary 38. Let 𝜇 ∈ R \ {. . . , −2, −1}. Then 0

𝛼
(1) Δ−] (𝜇)
= 𝜇(−]) (𝑡 − 𝑎)(𝑢+𝜐) , 𝐷Δ,𝑡 𝑓 (𝑡)
𝑎 (𝑡 − 𝑎) 𝑓𝑜𝑟 𝑡 ≥ 𝑎 − [−]] , 0

𝑛−1 (71)
(2) Δ]𝑎 (𝑡 − 𝑎)(𝜇) = 𝜇(]) (𝑡 − 𝑎)(𝑢−𝜐) , 𝑓𝑜𝑟 𝑡 ≥ 𝑎 + 1. 𝑘
= ∑ ℎ𝑘−𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) + (ℎ𝑛−𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) .
𝑛

(62) 𝑘=0

Lemma 39 (Taylor’s formula). Let 𝑛 ∈ N. Suppose that the


𝑛 𝑛−1 Proof. By Taylor’s formula
function 𝑓 is 𝑛 times differentiable on T 𝑘 . Let 𝛼 ∈ T 𝑘 , 𝑡 ∈ T,
and 𝑡 > 𝛼. Then one has 𝑛−1 𝑘 𝑡 𝑛
𝑛 𝑡 𝑓 (𝑡) = ∑ ℎ𝑘 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) + ∫ ℎ𝑛−1 (𝑡, 𝜎 (𝜏)) 𝑓Δ (𝜏) Δ𝜏,
𝑘 𝑛+1
𝑡0
𝑓 (𝑡) = ∑ ℎ𝑘 (𝑡, 𝛼) 𝑓Δ (𝛼) + ∫ ℎ𝑛 (𝑡, 𝜎 (𝜏)) 𝑓Δ (𝜏) Δ𝜏. 𝑘=0
𝑘=0 𝛼 (72)
(63)
we have
Δ𝑛+1
Proof. Let 𝑔(𝑡) := 𝑓 (𝑡). Then 𝑓 solves the initial value 𝑛−1 𝑘
problem 𝛼
𝐼Δ,𝑡 𝑓 (𝑡) = ∑ ℎ𝑘+𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 )
0
𝑛+1 𝑘 𝑘 𝑘=0
𝑥Δ = 𝑔 (𝑡) , 𝑥Δ (𝛼) = 𝑓Δ (𝛼) , 0 ≤ 𝑘 ≤ 𝑛. (64)
𝑛

Δ𝑛+1 + (ℎ𝛼−1 (⋅, 𝑡0 ) ∗ (ℎ𝑛−1 (⋅, 𝑡0 ) ∗ 𝑓Δ )) (𝑡)


Note that the Cauchy function for 𝑦 = 0 is
𝑦 (𝑡, 𝑠) = ℎ𝑛 (𝑡, 𝜎(𝑠)). By the variation of constants formula 𝑛−1 𝑘
in Theorem 7, = ∑ ℎ𝑘+𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 )
𝑡 𝑘=0 (73)
𝑓 (𝑡) = 𝑢 (𝑡) + ∫ 𝑦 (𝑡, 𝜎 (𝜏)) 𝑔 (𝜏) Δ𝜏, (65) Δ𝑛
𝛼 + ((ℎ𝛼−1 (⋅, 𝑡0 ) ∗ ℎ𝑛−1 (⋅, 𝑡0 )) ∗ 𝑓 ) (𝑡)
where 𝑢 solves the initial value problem 𝑛−1 𝑘
Δ𝑛+1 Δ𝑚 Δ𝑚 = ∑ ℎ𝑘+𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 )
𝑢 = 0, 𝑢 (𝛼) = 𝑓 (𝛼) , 0 ≤ 𝑚 ≤ 𝑛. (66)
𝑘=0
Δ𝑘
To validate the claim that 𝑢(𝑡) = ∑𝑛𝑘=0 ℎ𝑘 (𝑡, 𝛼)𝑓 (𝛼), set + (ℎ𝛼+𝑛−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) .
𝑛

𝑛 𝑘
𝜔 (𝑡) := ∑ ℎ𝑘 (𝑡, 𝛼) 𝑓Δ (𝛼) . (67) Besides,
𝑘=0
𝛼
𝑛+1 𝐷Δ,𝑡 𝑓 (𝑡) = 𝐷Δ𝑚 𝐼Δ,𝑡
𝑚−𝛼
𝑓 (𝑡)
By the properties of ℎ𝑘 (𝑡, 𝛼), 𝜔Δ (𝑡) = 0. We have moreover 0 0

that 𝑛−1 𝑘

𝑚
𝑛 𝑘
= 𝐷Δ𝑚 [ ∑ ℎ𝑘+𝑚−𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 )
𝜔Δ (𝑡) = ∑ ℎ𝑘−𝑚 (𝑡, 𝛼) 𝑓Δ (𝛼) , (68) 𝑘=0
𝑘=𝑚
𝑛
so that + (ℎ𝑚+𝑛−𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) ]
𝑚
𝑛 𝑘 𝑚
𝜔Δ (𝛼) = ∑ ℎ𝑘−𝑚 (𝛼, 𝛼) 𝑓Δ (𝛼) = 𝑓Δ (𝛼) , (69) 𝑛−1 𝑘 𝑛
𝑘=𝑚 = ∑ ℎ𝑘−𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 )+(ℎ𝑛−𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) ,
for 0 ≤ 𝑚 ≤ 𝑛. We consequently have that 𝜔 also solves (66), 𝑘=0
whence 𝑢 ≡ 𝜔 by uniqueness. (74)
Abstract and Applied Analysis 7

where Corollary 44. Let 𝑓 : N𝑎 → R be given. For any 𝑘 ∈ N0 and


Δ𝑛 𝜇 > 0 with 𝑀 − 1 < 𝜇 ≤ 𝑀, one has
𝐷Δ𝑚 (ℎ𝑚+𝑛−𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓 ) (𝑡)

𝑛
𝑚 𝑚−𝑘 Δ𝑘 Δ−𝜇 𝑘−𝜇
𝑎 𝑓 (𝑡) = Δ 𝑎 𝑓 (𝑡) , 𝑓𝑜𝑟 𝑡 ≥ 𝑎 − [−𝜇] ,
= (ℎ𝑛−𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) + ∑ ℎ𝑚+𝑛−𝛼−1−𝑘 (𝑡0 , 𝑡0 ) 𝑔Δ (80)
𝑘=0
Δ𝑘 Δ𝜇𝑎 𝑓 (𝑡) = Δ𝑘+𝜇
𝑎 𝑓 (𝑡) , 𝑓𝑜𝑟 𝑡 ≥ 𝑎 + 1.
𝑛
= (ℎ𝑛−𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) .
(75) Theorem 45. For 𝛼 > 0, 𝑛 is a positive integer; if 𝑓 is Δ-
differentiable and the highest order derivative is rd-continuous
over T, then it is valid that
When T = R, there is the following corollary.
𝑛−1 𝑘
𝑛+𝛼 𝛼 𝑛
Corollary 41 (see [3]). Let 𝛼 ≥ 0 and 𝑛 = [𝛼] + 1. If (1) 𝐷Δ,𝑡0
𝑓 (𝑡) = 𝐷Δ,𝑡0
𝐷Δ,𝑡0
𝑓 (𝑡) + ∑ ℎ𝑘−𝛼−𝑛 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) ,
𝑦(𝑥) ∈ 𝐴𝐶𝑛 [𝑎, 𝑏], then the fractional derivative 𝐷𝑎+
𝛼
𝑦 exists 𝑘=0

almost everywhere on [𝑎, 𝑏] and can be represented in the form 𝑛−1 𝑘


𝑛−𝛼 𝛼 𝑛
𝑛−1 (𝑘) (2) 𝐷Δ,𝑡 𝑓 (𝑡) = 𝐼Δ,𝑡 𝐷Δ,𝑡 𝑓 (𝑡) + ∑ ℎ𝑘+𝛼−𝑛 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) .
𝛼 𝑦 (𝑎) 0 0 0
(𝐷𝑎+ 𝑦) (𝑥) = ∑ (𝑥 − 𝑎)𝑘−𝛼 𝑘=0

𝑘=0
Γ (1 + 𝑘 − 𝛼) (81)
(76)
𝑥 𝑦(𝑛)
1 (𝑡) 𝑑𝑡
+ ∫ . Proof. (1) In the proof of Theorem 43(1), if we take 𝑠 = 𝑛 + 𝑚,
Γ (𝑛 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼−𝑛+1
then we have
𝛼 𝛽
Theorem 42. For 𝛼 > 0 and 𝛽 > 0, then (𝐼Δ,𝑡 𝐼 𝑓)(𝑡) =
0 Δ,𝑡0
𝛼+𝛽 𝐷Δ𝑛 𝐷Δ,𝑡
𝛼 𝑛+𝛼
𝑓 (𝑡) = 𝐷Δ,𝑡 𝑓 (𝑡)
(𝐼Δ,𝑡0 𝑓)(𝑡). 0 0

𝑛+𝑚−1 𝑘
Proof. According to Definition 28, Theorem 20 and 27, = ∑ ℎ𝑘−𝛼−𝑛 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) (82)
𝛼 𝛽 𝑘=0
(𝐼Δ,𝑡 𝐼 𝑓) (𝑡)
0 Δ,𝑡0
= (ℎ𝛼−1 (⋅, 𝑡0 ) ∗ (ℎ𝛽−1 (⋅, 𝑡0 ) ∗ 𝑓)) (𝑡) 𝑠
+ (ℎ𝑠−1−𝛼−𝑛 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) .
= ((ℎ𝛼−1 (⋅, 𝑡0 ) ∗ ℎ𝛽−1 (⋅, 𝑡0 )) ∗ 𝑓) (𝑡)
𝛼+𝛽
= (ℎ𝛼+𝛽−1 (⋅, 𝑡0 ) ∗ 𝑓) (𝑡) = (𝐼Δ,𝑡0 𝑓) (𝑡) . 𝑚+𝑛
As 𝐷Δ𝑛 𝑓(𝑡) is 𝑚 times Δ-differentiable on T 𝑘 , we have
(77)
𝑚−1 𝑛+𝑘
𝛼
𝐷Δ,𝑡0
(𝐷Δ𝑛 𝑓 (𝑡)) = ∑ ℎ𝑘−𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 )
Theorem 43. For 𝛼 > 0, 𝑛 is a positive integer; if 𝑓 is Δ- 𝑘=0 (83)
differentiable and the highest order derivative is rd-continuous 𝑠
over T, then it is valid that + (ℎ−𝛼+𝑚−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) .
(1) 𝐷Δ𝑛 𝐷Δ,𝑡
𝛼
0
𝑛+𝛼
𝑓 (𝑡) = 𝐷Δ,𝑡0
𝑓 (𝑡) ,
(78) By (82) and (83), if 𝑓 is at least 𝑛 times Δ-differentiable with
(2) 𝐷Δ𝑛 𝐼Δ,𝑡
𝛼
0
𝛼−𝑛
𝑓 (𝑡) = 𝐼Δ,𝑡0
𝑓 (𝑡) . the highest order derivative rd-continuous over T, then we
Proof. (1) Suppose that 𝑓 is a function which is 𝑠 times have
𝑠 𝑠
Δ-differentiable on T 𝑘 with 𝑓Δ rd-continuous over T. By
Lemma 40(2), 𝐷Δ𝑛 𝐷Δ,𝑡
𝛼
0
𝛼
𝑓 (𝑡) = 𝐷Δ,𝑡0
(𝐷Δ𝑛 𝑓 (𝑡))

𝐷Δ𝑛 𝐷Δ,𝑡
𝛼
𝑓 (𝑡) 𝑛−1 𝑘
(84)
0
𝑠−1
+ ∑ ℎ−𝛼−𝑛+𝑘 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) .
𝑘 𝑠
𝑘=0
= 𝐷Δ𝑛 [ ∑ ℎ𝑘−𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) + (ℎ𝑠−1−𝛼 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡)]
𝑘=0
𝑠−1 Thus
𝑘 𝑠
= ∑ ℎ𝑘−𝛼−𝑛 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) + (ℎ𝑠−1−𝛼−𝑛 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡)
𝑘=0 𝐷Δ𝑛 𝐷Δ,𝑡
𝛼
0
𝛼
𝑓 (𝑡) = 𝐷Δ,𝑡0
(𝐷Δ𝑛 𝑓 (𝑡)) (85)
𝑛+𝛼
= 𝐷Δ,𝑡0
𝑓 (𝑡) .
(79) is valid if and only if
By a similar way, we can get (2).
𝑘

When T = Z, we have the following corollary. 𝑓Δ (𝑡0 ) = 0, 𝑘 = 0, . . . , 𝑛 − 1. (86)


8 Abstract and Applied Analysis

(2) Similarly, we have Theorem 48. Let 𝑓 be Δ-differentiable and let its highest order
derivative be rd-continuous over T. When 𝛼, 𝛽 > 0, 𝑀 = [𝛽] +
𝑚−1 𝑘+𝑛
1, one has the following:
𝛼
𝐼Δ,𝑡 𝐷Δ𝑛 𝑓 (𝑡) = ∑ ℎ𝑘+𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 )
0 𝛽𝛼 𝛼−𝛽
𝑘=0 (1) (𝐷Δ,𝑡0 𝐼Δ,𝑡0
𝑓) (𝑡) = (𝐼Δ,𝑡0 ) 𝑓 (𝑡) ,
𝑚+𝑛
+ (ℎ𝛼+𝑚−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) , 𝛼 𝛽 𝛽−𝛼
(2) (𝐼Δ,𝑡 𝐷Δ,𝑡0 𝑓) (𝑡) = (𝐷Δ,𝑡0 ) 𝑓 (𝑡)
0 (93)
𝐷Δ𝑛 𝐼Δ,𝑡
𝛼
0
𝛼−𝑛
𝑓 (𝑡) = 𝐼Δ,𝑡0
𝑓 (𝑡) (87) 𝑀
𝛽−𝑘
𝑛+𝑚−1 − ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝐷Δ,𝑡0 𝑓 (𝑡0 ) .
𝑘
= ∑ ℎ𝑘+𝛼−𝑛 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) 𝑘=1

𝑘=0 Proof. According to Theorem 42 and (24), we have


Δ𝑚+𝑛
+ (ℎ𝛼+𝑚−1 (⋅, 𝑡0 ) ∗ 𝑓 ) (𝑡) . 𝛽 𝛼 𝑀−𝛽
(𝐷Δ,𝑡0 𝐼Δ,𝑡0
𝑓) (𝑡) = (𝐷Δ𝑀𝐼Δ,𝑡0 𝐼Δ,𝑡
𝛼
0
𝑓) (𝑡)

Therefore = 𝐷Δ𝑀 (ℎ𝑀+𝛼−𝛽−1 (⋅, 𝑡0 ) ∗ 𝑔) (𝑡)


(94)
𝑛−1
Δ𝑘 = (ℎ𝛼−𝛽−1 (⋅, 𝑡0 ) ∗ 𝑔) (𝑡)
𝐷Δ𝑛 𝐼Δ,𝑡
𝛼
0
𝑓 (𝑡) = 𝛼
𝐼Δ,𝑡0
𝐷Δ𝑛 𝑓 (𝑡) + ∑ ℎ𝑘+𝛼−𝑛 (𝑡, 𝑡0 ) 𝑓 (𝑡0 ) .
𝑘=0 𝛼−𝛽
(88) = (𝐼Δ,𝑡0 ) 𝑓 (𝑡) .

In addition,
Provided that 𝑓 is at least 𝑛 times Δ-differentiable with the
𝛼 𝛽 𝛼 𝑀−𝛽
highest order derivative rd-continuous over T. 𝐼Δ,𝑡0
𝐷Δ,𝑡0 𝑓 (𝑡) = 𝐼Δ,𝑡0
𝐷Δ𝑀𝐼Δ,𝑡0 𝑓 (𝑡)
Thus
𝑀−𝛽
= 𝐷Δ𝑀𝐼Δ,𝑡
𝛼
𝐼
0 Δ,𝑡0
𝑓 (𝑡)
𝐷Δ𝑛 𝐼Δ,𝑡
𝛼
0
𝑓 (𝑡) = 𝛼
𝐼Δ,𝑡0
𝐷Δ𝑛 𝑓 (𝑡) (89)
𝑀−1
𝑘+𝛽−𝑀 (95)
is valid if and only if − ∑ ℎ𝛼−𝑀+𝑘 (𝑡, 𝑡0 ) 𝐷Δ,𝑡0 𝑓 (𝑡0 )
𝑘=0

𝑘 𝑀
𝑓Δ (𝑡0 ) = 0, 𝑘 = 0, . . . , 𝑛 − 1. (90) 𝛼−𝛽 𝛽−𝑘
= 𝐼Δ,𝑡0 𝑓 (𝑡) − ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝐷Δ,𝑡0 𝑓 (𝑡0 ) .
𝑘=1

In particular, there are corollaries for T = R and for T = As a direct corollary of Theorem 48, we get Lemma 2.5 in
Z. [3].

Corollary 46 (see [3]). Let 𝛼 > 0 and 𝛽 > 0 be such that Corollary 49 (see [3]). Let 𝑅(𝛼) > 0, 𝑛 = [𝑅(𝛼)] + 1, and
𝑛−𝛼
𝑛 − 1 < 𝛼 ≤ 𝑛, 𝑚 − 1 < 𝛽 ≤ 𝑚 (𝑛, 𝑚 ∈ N) and 𝛼 + 𝛽 < 𝑛, let 𝑓𝑛−𝛼 (𝑥) = (𝐼𝑎+ 𝑓)(𝑥). If 𝑓(𝑥) ∈ 𝐿 1 (𝑎, 𝑏) and 𝑓𝑛−𝛼 (𝑥) ∈
𝑛
and let 𝑓 ∈ 𝐿 1 (𝑎, 𝑏) and 𝑓𝑚−𝛼 ∈ 𝐴𝐶𝑚 ([𝑎, 𝑏]). Then one has 𝐴𝐶 [𝑎, 𝑏], then
the following index rule: 𝑛 (𝑛−𝑗)
𝛼 𝛼 𝑓𝑛−𝛼 (𝑎)
(𝐼𝑎+ 𝐷𝑎+ 𝑓) (𝑥) = 𝑓 (𝑥) − ∑ (𝑥 − 𝑎)𝛼−𝑗 (96)
𝛼
(𝐷𝑎+ 𝛽
𝐷𝑎+ 𝛼+𝛽
𝑓) (𝑥) = (𝐷𝑎+ 𝑓) (𝑥) 𝑗=1 Γ (𝛼 − 𝑗 + 1)

𝑚 holds almost everywhere on [𝑎, 𝑏].


𝛽−𝑗 (𝑥 − 𝑎)−𝑗−𝛼 (91)
− ∑ (𝐷𝑎+ 𝑓) (𝑎+) .
𝑗=1 Γ (1 − 𝑗 − 𝛼) For fractional sum and difference, there is also the follow-
ing theorem in [1].
Corollary 47. Let 𝑓 : N𝑎 → R be given. For any 𝑘 ∈ N0 and It is different from Theorem 3.3 in [1], and from
] > 0, Theorem 48, we can get the following corollary.

𝑘−1 Corollary 50. Let 𝑓 : N𝑎 → R be given. For any 𝑘 ∈ N0 and


Δ𝑗 𝑓 (𝑎) ], 𝜇 > 0 with 𝑀 − 1 < 𝜇 ≤ 𝑀,
Δ−] 𝑘
𝑎 Δ 𝑓 (𝑡) = Δ𝑘−𝜐
𝑎 𝑓 (𝑡) −∑ (𝑡 − 𝑎)(𝜐−𝑘+𝑗) ,
𝑗=0 Γ (𝜐 − 𝑘 + 𝑗 + 1)
Δ−] 𝜇
𝑎 Δ 𝑎 𝑓 (𝑡)
𝑓𝑜𝑟 𝑡 ≥ 𝑎 − [−]] . 𝑀−1
Δ𝑗−(𝑀−𝜇) 𝑓 (𝑎) (97)
(92) = Δ𝜇−𝜐
𝑎 𝑓 (𝑡) − ∑
𝑎
(𝑡 − 𝑎)(𝜐−𝑀+𝑗) .
𝑗=0 Γ (𝜐 − 𝑘 + 𝑗 + 1)
Abstract and Applied Analysis 9

Next, we will give the Laplace transform of fractional Here 𝐿 Δ [𝑎, 𝑏) := 𝐿 Δ,1 [𝑎, 𝑏) is the space of Δ-Lebesgue
integral and derivative on time scales. summable functions in a finite interval T̃ := [𝑎, 𝑏)T .
In the following, we prove that Cauchy type problem and
Theorem 51. (1) Let 𝛼 > 0 and 𝑓 : T → R be locally Δ- nonlinear Volterra integral equation are equivalent in the
integrable. For 𝑠, 𝑡 ∈ T with 𝑡 > 𝑡0 , one has sense that if 𝑦(𝑡) ∈ 𝐿 Δ [𝑎, 𝑏) satisfies one of these relations,
𝛼 1 then it also satisfies the other.
L {𝐼Δ,𝑡 𝑓 (𝑡)} (𝑧, 𝑡0 ) = L {𝑓 (𝑡)} (𝑧, 𝑡0 ) . (98)
0
𝑧𝛼 𝑚
Theorem 52. Let 𝛼 > 0, 𝑚 = −[−𝛼], T̃ := [𝑎, 𝑏]T , 𝑡0 , 𝑡 ∈ T̃ 𝑘 .
(2) Let 𝛼 > 0 and 𝑓 : T → R be locally Δ-integrable. For
𝑚
Let 𝐺 be an open set in R and let 𝑓 : T̃ × 𝐺 → R be a function
𝑡0 , 𝑡 ∈ T 𝑘 with 𝑡 > 𝑠, one has such that 𝑓(𝑡, 𝑦) ∈ 𝐿 Δ [𝑎, 𝑏) for any 𝑦 ∈ 𝐺. If 𝑦(𝑡) ∈ 𝐿 Δ [𝑎, 𝑏),
𝛼
then Cauchy type problem (102) and (103) is equivalent to
L {𝐷Δ,𝑡0
𝑓 (𝑡)} (𝑧, 𝑡0 ) 𝑚
𝛼
𝑚 (99) 𝑦 (𝑡) = ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝑏𝑘 + 𝐼Δ,𝑡0
𝑓 (𝑡, 𝑦 (𝑡)) . (105)
𝛼−𝑗
= 𝑧𝛼 L {𝑓 (𝑡)} (𝑧, 𝑡0 ) − ∑𝑧𝑗−1 𝐷Δ,𝑡0 𝑓 (𝑡0 ) . 𝑘=1
𝛼
𝑗=1 Proof. First we prove the necessity. We apply 𝐼Δ,𝑡0
to both sides
Proof. (1) According to Definition 28 and convolution theo- of (102) and get by Theorem 48
rem, we have 𝑚
𝛼 𝛼 𝛼−𝑘
𝐼Δ,𝑡 𝐷Δ,𝑡 𝑦 (𝑡) = 𝑦 (𝑡) − ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝐷Δ,𝑡 𝑦 (𝑡0 )
𝛼 0 0 0
L {𝐼Δ,𝑡0
𝑓 (𝑡)} (𝑧, 𝑡0 ) 𝑘=1
= (ℎ𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓 (⋅, ⋅)) (𝑡) . (106)
= L {(ℎ𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓) (𝑡)} (𝑧, 𝑡0 )
𝛼
(100) = 𝐼Δ,𝑡 𝑓 (𝑡, 𝑦 (𝑡)) .
= L {ℎ𝛼−1 (𝑡, 𝑡0 )} (𝑧, 𝑡0 ) L {𝑓 (𝑡)} (𝑧, 𝑡0 ) 0

Thus
1 𝑚
= 𝛼 L {𝑓 (𝑡)} (𝑧, 𝑡0 ) . 𝛼
𝑧 𝑦 (𝑡) = ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝑏𝑘 + 𝐼Δ,𝑡0
𝑓 (𝑡, 𝑦 (𝑡)) . (107)
𝑘=1
(2) By Definition 31 and (26) and taking the Laplace
𝛼
transform of fractional integral into account, we get Now we prove the sufficiency. Applying the operator 𝐷Δ,𝑡0
to
both sides of (105) and by (57) and Theorem 48(1), we have
𝛼
L {𝐷Δ,𝑡0
𝑓 (𝑡)} (𝑧, 𝑡0 ) 𝑚
𝛼 𝛼
𝑚
𝐷Δ,𝑡 𝑦 (𝑡) = ∑ 𝐷Δ,𝑡 ℎ (𝑡, 𝑡0 ) 𝑏𝑘
0 𝛼−𝑘
= L {𝐷Δ,𝑡 𝐼𝑚−𝛼 𝑓 (𝑡)} (𝑧, 𝑡0 )
0 Δ,𝑡0
0
𝑘=1
𝛼 (108)
𝑚−1 + 𝐷Δ,𝑡 𝐼𝛼 𝑓 (𝑡, 𝑦 (𝑡))
0 Δ,𝑡0
= 𝑧𝑚 L {𝐼Δ,𝑡
𝑚−𝛼
0
𝑓 (𝑡)} (𝑧, 𝑡0 ) − ∑ 𝑧𝑚−𝑗−1 𝐷𝑗 𝐼Δ,𝑡
𝑚−𝛼
0
𝑓 (𝑡0 )
𝑗=0 = 𝑓 (𝑡, 𝑦 (𝑡)) .
𝑚−1 Now we show that the relations in (103) also hold. For this,
1 𝑗−𝑚+𝛼
= 𝑧𝑚 L {𝑓 (𝑡)} (𝑧, 𝑡0 ) − ∑ 𝑧𝑚−𝑗−1 𝐷Δ,𝑡0 𝑓 (𝑡0 ) 𝛼−𝑘
we apply the operators 𝐷Δ,𝑡 (𝑘 = 1, . . . , 𝑚) to both sides of
𝑧𝑚−𝛼 𝑗=0
0
(105):
𝑚 𝛼−𝑘
𝛼−𝑗 𝐷Δ,𝑡 𝑦 (𝑡)
= 𝑧𝛼 L {𝑓 (𝑡)} (𝑧, 𝑡0 ) − ∑𝑧𝑗−1 𝐷Δ,𝑡0 𝑓 (𝑡0 ) . 0

𝑗=1 𝑚
𝛼−𝑘 𝛼−𝑘 𝛼
(101) = ∑ 𝐷Δ,𝑡 ℎ (𝑡, 𝑡0 ) 𝑏𝑗 + 𝐷Δ,𝑡
0 𝛼−𝑗
𝐼 𝑓 (𝑡, 𝑦 (𝑡))
0 Δ,𝑡0
𝑗=1

𝑚 (109)
𝑚−𝑘 𝑚−𝛼 𝑘
5. Cauchy Type Problem with = ∑ 𝐷Δ,𝑡 𝐼 ℎ𝛼−𝑗 (𝑡, 𝑡0 ) 𝑏𝑗 + 𝐼Δ,𝑡
0 Δ,𝑡0 0
𝑓 (𝑡, 𝑦 (𝑡))
𝑗=1
Riemann-Liouville Fractional Derivative
𝑚
𝑚−𝑘 𝑘
In this section, we consider Cauchy type problem with = ∑ 𝐷Δ,𝑡0
ℎ𝑚−𝑗 (𝑡, 𝑡0 ) 𝑏𝑗 + 𝐼Δ,𝑡0
𝑓 (𝑡, 𝑦 (𝑡)) .
Riemann-Liouville fractional derivative 𝑗=1
𝛼
𝐷Δ,𝑡0
𝑦 (𝑡) = 𝑓 (𝑡, 𝑦 (𝑡)) (𝛼 > 0) , (102) Since
{ 0, 𝑘 < 𝑗,
𝛼−𝑘
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 𝑏𝑘 (𝑘 = 1, . . . , 𝑚 = − [−𝛼]) . (103) 𝑚−𝑘 {
0 𝐷Δ,𝑡 ℎ𝑚−𝑗 (𝑡, 𝑡0 ) = { 1, 𝑘 = 𝑗,
0 {
{ℎ𝑘−𝑗 (𝑡, 𝑡0 ) , 𝑘 > 𝑗,
In the space 𝐿𝛼Δ [𝑎, 𝑏) defined for 𝛼 > 0 by (110)

𝐿𝛼Δ [𝑎, 𝑏) := {𝑦 ∈ 𝐿 Δ [𝑎, 𝑏) : 𝐷Δ,𝑡


𝛼
0
∈ 𝐿 Δ [𝑎, 𝑏)} . (104) 𝛼−𝑘
𝐷Δ,𝑡0
𝑦 (𝑡0 ) = 𝑏𝑘 (𝑘 = 1, . . . , 𝑚) .
10 Abstract and Applied Analysis

By Weierstrass discriminance, we obtain 𝑦𝑙 (𝑡) convergent


In the following, we bring in Lipschitzian-type condition: uniformly. Next we will show the uniqueness. Assume that
󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨󵄨 𝑧(𝑡) is another solution to (105); that is,
󵄨󵄨𝑓 (𝑡, 𝑦1 (𝑡)) − 𝑓 (𝑡, 𝑦2 (𝑡))󵄨󵄨 ≤ 𝐴 󵄨󵄨𝑦1 (𝑡) − 𝑦2 (𝑡)󵄨󵄨 , (111)
𝛼
𝑧 (𝑡) = 𝑦0 (𝑡) + 𝐼Δ,𝑡 𝑓 (𝑡, 𝑧 (𝑡)) . (120)
where 𝐴 > 0 does not depend on 𝑡 ∈ [𝑎, 𝑏)T . We will derive 0

a unique solution to the Cauchy problem (102)-(103). As


󵄨󵄨 󵄨
Theorem 53. Let the condition of Theorem 52 be valid, let 󵄨󵄨𝑓 (𝑡, 𝑦 (𝑡))󵄨󵄨󵄨 ≤ 𝑀,
𝑓(𝑡, 𝑦) satisfy the Lipschitzian condition (111), and |𝑓(𝑡, 𝑦)| ≤ 󵄨󵄨 󵄨 𝛼 󵄨 󵄨 (121)
𝑀, Δ 𝐹𝛼,1 (𝐴, 𝑡, 𝑡0 ) is defined on T̃ := [𝑎, 𝑏]T , where 𝐴 is 󵄨󵄨𝑦0 (𝑡) − 𝑧 (𝑡)󵄨󵄨󵄨 ≤ 𝐼Δ,𝑡0 󵄨󵄨󵄨𝑓 (𝑡, 𝑧 (𝑡))󵄨󵄨󵄨 ≤ 𝑀ℎ𝛼 (𝑡, 𝑡0 ) .
the Lipschitzian constant in (111). Then there exists a unique If
solution 𝑦(𝑡) to initial value problem (102)-(103) in the space
𝐿𝛼Δ [𝑎, 𝑏). 󵄨󵄨 󵄨
󵄨󵄨𝑦𝑙−1 (𝑡) − 𝑧 (𝑡)󵄨󵄨󵄨 ≤ 𝑀𝐴 ℎ𝑙𝛼 (𝑡, 𝑡0 ) ,
𝑙−1
(122)
Proof. Since the Cauchy type problem (102)-(103) and the then
nonlinear Volterra integral equation (105) are equivalent, we 󵄨󵄨 󵄨 𝛼 󵄨 󵄨
only need to prove that there exists a unique solution to (105). 󵄨󵄨𝑦𝑙 (𝑡) − 𝑧 (𝑡)󵄨󵄨󵄨 ≤ (𝐼Δ,𝑡0 󵄨󵄨󵄨𝑓 (⋅, 𝑦𝑙−1 (⋅)) − 𝑓 (⋅, 𝑧 (⋅))󵄨󵄨󵄨) (𝑡)
We define function sequences: 󵄨 󵄨
𝛼
𝛼
≤ (𝐼Δ,𝑡 𝐴 󵄨󵄨󵄨𝑦𝑙−1 − 𝑧󵄨󵄨󵄨) (𝑡)
𝑦𝑙 (𝑡) = 𝑦0 (𝑡) + 𝐼Δ,𝑡0
𝑓 (𝑡, 𝑦𝑙−1 (𝑡)) (𝑙 = 1, 2, . . . , ) , (112) 0

𝛼
where ≤ (𝐼Δ,𝑡0
𝐴𝑀𝐴𝑙−1 ℎ𝑙𝛼 (⋅, 𝑡0 )) (𝑡) (123)
𝑚
𝑦0 (𝑡) = ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝑏𝑘 . (113) ≤ 𝑀𝐴𝑙 (𝐼Δ,𝑡
𝛼
ℎ (⋅, 𝑡0 )) (𝑡)
0 𝑙𝛼
𝑘=1

We obtain, by induction that, = 𝑀𝐴𝑙 ℎ(𝑙+1)𝛼 (𝑡, 𝑡0 ) .


󵄨󵄨󵄨𝑦𝑙 (𝑡) − 𝑦𝑙−1 (𝑡)󵄨󵄨󵄨 ≤ 𝑀𝐴𝑙−1 ℎ𝑙𝛼 (𝑡, 𝑡0 ) . (114) By mathematical induction, we have
󵄨 󵄨
In fact, for 𝑙 = 1, as |𝑓(𝑡, 𝑦(𝑡))| ≤ 𝑀, we have 󵄨󵄨 󵄨
󵄨󵄨𝑦𝑙 (𝑡) − 𝑧 (𝑡)󵄨󵄨󵄨 ≤ 𝑀𝐴 ℎ(𝑙+1)𝛼 (𝑡, 𝑡0 )
𝑙
(124)
󵄨󵄨 󵄨
󵄨󵄨𝑦1 (𝑡) − 𝑦0 (𝑡)󵄨󵄨󵄨 ≤ 𝑀𝐼Δ,𝑡0 ℎ0 (𝑡, 𝑡0 ) = 𝑀ℎ𝛼 (𝑡, 𝑡0 ) .
𝛼
(115)
and then get 𝑧(𝑡) = 𝑦(𝑡) owing to the uniqueness of the limit.
If This completes the proof of the theorem.
󵄨󵄨 󵄨󵄨 𝑙−2
󵄨󵄨𝑦𝑙−1 (𝑡) − 𝑦𝑙−2 (𝑡)󵄨󵄨 ≤ 𝑀𝐴 ℎ(𝑙−1)𝛼 (𝑡, 𝑡0 ) , (116) Next we consider the generalized Cauchy type problem:
then 𝛼 𝛼 𝛼
󵄨󵄨 󵄨󵄨 𝛼 󵄨󵄨 󵄨󵄨 𝐷Δ,𝑡 𝑦 (𝑡) = 𝑓 (𝑡, 𝑦 (𝑡) , 𝐷Δ,𝑡1 0 𝑦 (𝑡) , . . . , 𝐷Δ,𝑡𝑙 0 𝑦 (𝑡))
󵄨󵄨𝑦𝑙 (𝑡) − 𝑦𝑙−1 (𝑡)󵄨󵄨 ≤ 𝐴 (𝐼Δ,𝑡0 󵄨󵄨𝑦𝑙−1 − 𝑦𝑙−2 󵄨󵄨) (𝑡)
0
(125)
(0 = 𝛼0 ≤ 𝛼1 ≤ ⋅ ⋅ ⋅ ≤ 𝛼𝑙 < 𝛼) ,
𝛼
≤ 𝐴 (𝐼Δ,𝑡 𝑀𝐴𝑙−2 ℎ(𝑙−1)𝛼 (⋅, 𝑡0 )) (𝑡) 𝛼−𝑘
0
(117) 𝐷Δ,𝑡0
𝑦 (𝑡0 ) = 𝑏𝑘 (𝑘 = 1, . . . , 𝑛, 𝑛 = [𝛼] + 1) . (126)
𝑙−1 𝛼
= 𝑀𝐴 (𝐼Δ,𝑡 ℎ
0 (𝑙−1)𝛼
(⋅, 𝑡0 )) (𝑡) Theorem 54. Let 𝑓 : [𝑎, 𝑏)T × 𝐺 → R be a function such
that 𝑓(𝑡, 𝑦, 𝑦1 , . . . , 𝑦𝑙 ) ∈ 𝐿 Δ [𝑎, 𝑏) for any (𝑦, 𝑦1 , . . . , 𝑦𝑙 ) ∈ 𝐺. If
= 𝑀𝐴𝑙−1 ℎ𝑙𝛼 (𝑡, 𝑡0 ) . 𝑦(𝑡) ∈ 𝐿 Δ [𝑎, 𝑏), then 𝑦(𝑡) satisfies a.e. the relations (126) and
Let (125) if and only if it satisfies a.e. the integral equation

𝑛
𝑦 (𝑡) = lim (𝑦𝑙 (𝑡) − 𝑦0 (𝑡)) = ∑ (𝑦𝑙 (𝑡) − 𝑦𝑙−1 (𝑡)) , (118)
𝑙→∞ 𝑦 (𝑡) = ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝑏𝑘
𝑙=1
𝑘=1
and we have 𝛼 𝛼 𝛼
∞ ∞ + (𝐼Δ,𝑡0
𝑓 (𝜏, 𝑦 (𝜏) , 𝐷Δ,𝑡1 0 𝑦 (𝜏) , . . . , 𝐷Δ,𝑡𝑙 0 𝑦 (𝜏))) (𝑡) .
󵄨 󵄨
∑ 󵄨󵄨󵄨𝑦𝑙 (𝑡) − 𝑦𝑙−1 (𝑡)󵄨󵄨󵄨 ≤ ∑𝑀𝐴𝑙−1 ℎ𝑙𝛼 (𝑡, 𝑡0 ) (127)
𝑙=1 𝑙=1
Assume that 𝑓 satisfies generalized Lipschitzian condi-
𝑀∞ 𝑙 tion
= ∑𝐴 ℎ (𝑡, 𝑡0 )
𝐴 𝑙=1 𝑙𝛼 𝑙
(119) 󵄨󵄨 󵄨 󵄨 󵄨
󵄨󵄨𝑓 (𝑡, 𝑦, 𝑦1 , . . . , 𝑦𝑙 ) − 𝑓 (𝑡, 𝑧, 𝑧1 , . . . , 𝑧𝑙 )󵄨󵄨󵄨 ≤ 𝐴 [∑ 󵄨󵄨󵄨󵄨𝑦𝑗 − 𝑧𝑗 󵄨󵄨󵄨󵄨] .
𝑀∞ 𝑙
< ∑𝐴 ℎ (𝑏, 𝑡0 ) [𝑗=1 ]
𝐴 𝑙=1 𝑙𝛼 (128)
𝑀 According to Theorem 54 and by a similar proof to that
= [ 𝐹 (𝐴, 𝑏, 𝑡0 ) − 1] .
𝐴 Δ 𝛼,1 of Theorem 53, we have the following theorem.
Abstract and Applied Analysis 11

Theorem 55. Let the condition of Theorem 54 be valid and let Suppose that
𝑓(𝑡, 𝑦, 𝑦1 , . . . , 𝑦𝑙 ) satisfy the Lipschitzian condition (128). Then
there exists a unique solution 𝑦(𝑡) to the generalized Cauchy 󵄨󵄨 󵄨
󵄨󵄨𝑧𝑚−1 (𝑡) − 𝑦𝑚−1 (𝑡)󵄨󵄨󵄨
type problem.
𝑚−1 (136)
󵄨 󵄨
≤ 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ∑ 𝐴𝑗 ℎ𝑗𝛼+𝛼−1 (𝑡, 𝑡0 ) .
6. The Dependency of the Solution upon 𝑗=0
the Initial Value
We consider fractional differential initial value problem Then
again: 󵄨󵄨 󵄨
𝛼
󵄨󵄨𝑧𝑚 (𝑡) − 𝑦𝑚 (𝑡)󵄨󵄨󵄨
𝐷Δ,𝑡 𝑦 (𝑡) = 𝑓 (𝑡, 𝑦 (𝑡)) ,
󵄨 󵄨
≤ 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ℎ𝛼−1 (𝑡, 𝑡0 )
0
𝛼
(129)
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 𝜂,
𝛼 󵄨󵄨 󵄨
0

where 0 < 𝛼 < 1.


+ (𝐼Δ,𝑡 󵄨𝑓 (𝜏, 𝑧𝑚−1 (𝜏)) − 𝑓 (𝜏, 𝑦𝑚−1 (𝜏))󵄨󵄨󵄨) (𝑡)
0 󵄨

Using Theorem 52, we have 󵄨 󵄨


≤ 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ℎ𝛼−1 (𝑡, 𝑡0 )
𝛼
𝑦 (𝑡) = 𝜂ℎ𝛼−1 (𝑡, 𝑡0 ) + (𝐼Δ,𝑡 𝑓 (𝜏, 𝑦 (𝜏))) (𝑡) . (130) 󵄨 󵄨
0 𝛼
+ (𝐼Δ,𝑡 0
𝐴 󵄨󵄨󵄨𝑧𝑚−1 (𝜏) − 𝑦𝑚−1 (𝜏)󵄨󵄨󵄨) (𝑡)
Suppose that 𝑧(𝑡) is the solution to the initial value problem: 󵄨 󵄨
𝛼 ≤ 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ℎ𝛼−1 (𝑡, 𝑡0 )
𝐷Δ,𝑡0
𝑦 (𝑡) = 𝑓 (𝑡, 𝑦 (𝑡)) ,
𝛼
(131) 𝑚−1
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 𝜂. 󵄨 󵄨
0
𝛼
+ (𝐼Δ,𝑡0
𝐴 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ∑ 𝐴𝑗 ℎ𝑗𝛼+𝛼−1 (𝜏, 𝑡0 )) (𝑡)
𝑗=0
We can derive the dependency of the solution upon the initial
󵄨 󵄨
value. = 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ℎ𝛼−1 (𝑡, 𝑡0 )
Theorem 56. Let T̃ := [𝑎, 𝑏]T , 𝑡0 , 𝑡 ∈ T̃ 𝑘 and suppose that 𝑓 𝑚−1
󵄨 󵄨
satisfy the Lipschitz condition; that is, + 𝐴 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ∑ 𝐴𝑗 (𝐼Δ,𝑡
𝛼

0 𝑗𝛼+𝛼−1
(𝜏, 𝑡0 )) (𝑡)
󵄨󵄨 󵄨 󵄨 󵄨 𝑗=0
󵄨󵄨𝑓 (𝑡, 𝑧 (𝑡)) − 𝑓 (𝑡, 𝑦 (𝑡))󵄨󵄨󵄨 ≤ 𝐴 󵄨󵄨󵄨𝑧 (𝑡) − 𝑦 (𝑡)󵄨󵄨󵄨 . (132)
󵄨 󵄨
Then one has = 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ℎ𝛼−1 (𝑡, 𝑡0 )
󵄨󵄨 󵄨 󵄨 󵄨
󵄨󵄨𝑧 (𝑡) − 𝑦 (𝑡)󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 Δ 𝐹𝛼,𝛼 (𝐴, 𝑏, 𝑡0 ) . (133) 󵄨 󵄨
𝑚−1
+ 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ∑ 𝐴𝑗+1 ℎ(𝑗+1)𝛼+𝛼−1 (𝑡, 𝑡0 )
Proof. By the proof of Theorem 53, we know that 𝑦(𝑡) = 𝑗=0
lim𝑚 → ∞ 𝑦𝑚 (𝑡), 𝑧(𝑡) = lim𝑚 → ∞ 𝑧𝑚 (𝑡), where 𝑚
󵄨 󵄨
𝑦0 (𝑡) = 𝜂ℎ𝛼−1 (𝑡, 𝑡0 ) , = 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ∑ 𝐴𝑗 ℎ𝑗𝛼+𝛼−1 (𝑡, 𝑡0 ) .
𝑗=0
𝛼
𝑦𝑚 (𝑡) = 𝑦0 (𝑡) + (𝐼Δ,𝑡0
𝑓 (𝜏, 𝑦𝑚−1 (𝜏))) (𝑡) , (137)
(134)
𝑧0 (𝑡) = 𝜂ℎ𝛼−1 (𝑡, 𝑡0 ) , According to mathematical induction, we have
𝛼
𝑧𝑚 (𝑡) = 𝑧0 (𝑡) + (𝐼Δ,𝑡0
𝑓 (𝜏, 𝑧𝑚−1 (𝜏))) (𝑡) . 𝑚
󵄨󵄨 󵄨 󵄨 󵄨
󵄨󵄨𝑧𝑚 (𝑡) − 𝑦𝑚 (𝑡)󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ∑ 𝐴 ℎ𝑗𝛼+𝛼−1 (𝑡, 𝑡0 ) .
𝑗
Using the Lipschitz condition, we have (138)
𝑗=0
󵄨󵄨 󵄨
󵄨󵄨𝑧1 (𝑡) − 𝑦1 (𝑡)󵄨󵄨󵄨
󵄨 󵄨 Taking the limit 𝑚 → ∞, we obtain that
≤ 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ℎ𝛼−1 (𝑡, 𝑡0 )
𝛼 󵄨󵄨 󵄨 ∞
+ (𝐼Δ,𝑡 󵄨𝑓 (𝜏, 𝑧0 (𝜏)) − 𝑓 (𝜏, 𝑦0 (𝜏))󵄨󵄨󵄨) (𝑡) 󵄨󵄨 󵄨 󵄨 󵄨
0 󵄨 󵄨󵄨𝑧 (𝑡) − 𝑦 (𝑡)󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ∑𝐴 ℎ𝑗𝛼+𝛼−1 (𝑡, 𝑡0 )
𝑗

󵄨 󵄨 󵄨 󵄨 𝑗=0 (139)
≤ 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ℎ𝛼−1 (𝑡, 𝑡0 ) + (𝐼Δ,𝑡
𝛼
𝐴 󵄨󵄨󵄨𝑧0 (𝜏) − 𝑦0 (𝜏)󵄨󵄨󵄨) (𝑡)
0
󵄨 󵄨
≤ 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 Δ 𝐹𝛼,𝛼 (𝐴, 𝑏, 𝑡0 ) .
󵄨 󵄨 󵄨 󵄨
= 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ℎ𝛼−1 (𝑡, 𝑡0 ) + 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 𝐴 (𝐼Δ,𝑡
𝛼
ℎ (𝜏, 𝑡0 )) (𝑡)
0 𝛼−1

󵄨 󵄨 󵄨 󵄨
= 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 ℎ𝛼−1 (𝑡, 𝑡0 ) + 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 𝐴ℎ2𝛼−1 (𝑡, 𝑡0 )
󵄨 󵄨 As a special case, when fractional equation is linear, we
= 󵄨󵄨󵄨𝜂 − 𝜂󵄨󵄨󵄨 [ℎ𝛼−1 (𝑡, 𝑡0 ) + 𝐴ℎ2𝛼−1 (𝑡, 𝑡0 )] . can obtain its explicit solutions and we will explain it in next
(135) section.
12 Abstract and Applied Analysis

𝑐1
7. Homogeneous Equations with 𝑐2

Constant Coefficients where 𝑡 ∈ Ω, 𝐶 = ( .. ). As 𝑊𝛼 (𝑡∗ ) = 0, the equations have



.
𝑐𝑛
In this section, we apply the Laplace transform method to nontrivial solution 𝑐𝑗 (𝑗 = 1, 2, . . . , 𝑛). Now we construct a
derive explicit solutions to homogeneous equations of the function using these constants:
form
𝑛
𝑚 𝑦 (𝑡) = ∑ 𝑐𝑗 𝑦𝑗 (𝑡) , (147)
𝛼
∑ 𝐴 𝑘 [𝐷Δ,𝑡𝑘 0 𝑦 (𝑡)] + 𝐴 0 𝑦 (𝑡) = 0 𝑗=1
𝑘=1 (140)
and we get that 𝑦(𝑡) is a solution. From (146), we obtain that
(𝑡 > 𝑡0 ; 𝑚 ∈ N; 0 < 𝛼1 < ⋅ ⋅ ⋅ < 𝛼𝑚 ) 𝑦(𝑡) satisfies initial value condition
𝛼 𝛼−𝑘
with the Liouville fractional derivatives 𝐷Δ,𝑡𝑘 0 𝑦 (𝑘 = 1, . . . , (𝐷Δ,𝑡0
𝑦 (𝑡∗ )) = 0, 𝑘 = 1, . . . , 𝑛. (148)
𝑚). Here 𝐴 𝑘 ∈ R (𝑘 = 0, . . . , 𝑚) are real constants, and,
generally speaking, we can take 𝐴 𝑚 = 1. However, 𝑦(𝑡) = 0 is also a solution to equation satisfying the
In order to solve the equation, we need the following initial value condition. By the uniqueness of solution, we have
Laplace transform formula: 𝑛

𝛼
∑ 𝑐𝑗 𝑦𝑗 (𝑡) = 0, (149)
L {𝐷Δ,𝑡0
𝑦 (𝑡)} (𝑧, 𝑡0 ) = 𝑧𝛼 L {𝑦 (𝑡)} (𝑧, 𝑡0 ) 𝑗=1

𝑙 (141) and, thus 𝑦𝑗 (𝑡) (𝑗 = 1, 2, . . . , 𝑛) are linearly dependant,


− ∑ 𝑑𝑗 𝑧𝑗−1 (𝑙 − 1 < 𝛼 ≤ 𝑙 ∈ N) , which leads to a contradiction. Thus, if the solutions 𝑦1 (𝑡),
𝑗=1 𝑦2 (𝑡), . . . , 𝑦𝑛 (𝑡) are linearly independent, then 𝑊𝛼 (𝑡∗ ) ≠
0 at
𝛼−𝑗 some point 𝑡∗ ∈ Ω.
𝑑𝑗 = 𝐷Δ,𝑡0 𝑦 (𝑡0 ) (𝑗 = 1, . . . , 𝑙) . (142)
There hold the following statements.
First, we derive explicit solutions to (140) with 𝑚 = 1:
Theorem 59. Let 𝑙 − 1 < 𝛼 ≤ 𝑙 (𝑙 ∈ N) and 𝜆 ∈ R. Then the
𝛼
𝐷Δ,𝑡0
𝑦 (𝑡) − 𝜆𝑦 (𝑡) = 0 (𝑡 > 𝑡0 ; 𝑙 − 1 < 𝛼 ≤ 𝑙; 𝑙 ∈ N; 𝜆 ∈ R) . functions
(143) ∞
𝑦𝑗 (𝑡) = Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 ) (= ∑ 𝜆𝑘 ̂ℎ𝑘𝛼+𝛼−𝑗 (𝑡, 𝑡0 ))
In order to prove our result, we need the following definition 𝑘=0 (150)
and lemma.
(𝑗 = 1, . . . , 𝑙)
Definition 57. The function 𝑊𝛼 (𝑡) is defined by
yield the fundamental system of solutions to (143). Moreover,
𝑛
𝛼−𝑘
𝑊𝛼 (𝑡) = det ((𝐷Δ,𝑡 𝑦 ) (𝑡))𝑘,𝑗=1 (𝑛 = [𝛼] + 1, 𝑎 ≤ 𝑡 ≤ 𝑏) . 𝑦𝑗 (𝑡), 𝑗 = 1, 2, . . . , 𝑙, satisfy
0 𝑗

(144) 𝛼−𝑘
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 0 (𝑘, 𝑗 = 1, . . . , 𝑙; 𝑘 ≠
𝑗) ,
0 𝑗
(151)
Lemma 58. The solutions 𝑦1 (𝑡), 𝑦2 (𝑡), . . . , 𝑦𝑛 (𝑡) are linearly 𝛼−𝑘
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 1 (𝑘 = 1, . . . , 𝑙) .
independent if and only if 𝑊𝛼 (𝑡∗ ) ≠
0 at some point 𝑡∗ ∈ [𝑎, 𝑏]. 0 𝑘

Proof. Applying the Laplace transform to (143) and taking


Proof. We first prove sufficiency. If, to the contrary, 𝑦𝑗 (𝑡) (𝑗 =
(141) into account, we have
1, 2, . . . , 𝑛) are linearly dependent on Ω, then there exist 𝑛
constants {𝑐𝑗 }𝑛𝑗=1 , not all zero, such that 𝑙
𝑧𝑗−1
L {𝑦 (𝑡)} (𝑧, 𝑡0 ) = ∑𝑑𝑗 , (152)
𝑧𝛼 − 𝜆
𝑐1 𝑗=1

𝑛 𝑐 2
𝛼−𝑘
((𝐷Δ,𝑡 𝑦 ) (𝑡))𝑘,𝑗=1 ( .. ) ≡ 0 (145) where 𝑑𝑗 (𝑗 = 1, . . . , 𝑙) are given by (142).
0 𝑗
. Formula (49) with 𝛽 = 𝛼 + 1 − 𝑗 yields
𝑐𝑛
𝑧𝑗−1
L { Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 )} (𝑧, 𝑡0 ) = (|𝜆| < |𝑧|𝛼 ) .
holds, and thus 𝑊𝛼 (𝑡) ≡ 0 which leads to a contradiction. 𝑧𝛼 − 𝜆
Therefore, if 𝑊𝛼 (𝑡∗ ) ≠ 0 at some point 𝑡∗ ∈ Ω, then (153)
𝑦1 (𝑡), 𝑦2 (𝑡), . . . , 𝑦𝑛 (𝑡) are linearly independent. Now we prove
the necessity. Suppose, to the contrary, for 𝑡 ∈ Ω, 𝑊𝛼 (𝑡) = 0. Thus, from (152), we derive the following solution to (143):
Consider 𝑙
𝛼−𝑘 𝑛 𝑦 (𝑡) = ∑ 𝑑𝑗 𝑦𝑗 (𝑡) , 𝑦𝑗 (𝑡) = Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 ) . (154)
((𝐷Δ,𝑡 𝑦 ) (𝑡∗ ))𝑘,𝑗=1 𝐶 = 0,
0 𝑗
(146) 𝑗=1
Abstract and Applied Analysis 13

It is easily verified that the functions 𝑦𝑗 (𝑡) are solutions to Corollary 60. Consider that
(143):
𝛼
𝐷Δ,𝑡 𝑦 (𝑡) − 𝜆𝑦 (𝑡) = 0 (𝑡 > 𝑡0 ; 0 < 𝛼 ≤ 1; 𝜆 ∈ R) (161)
𝛼 0
𝐷Δ,𝑡0
[ Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 )]
(155) has its solution given by
= 𝜆 Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 ) (𝑗 = 1, . . . , 𝑙) .
𝑦 (𝑡) = Δ 𝐹𝛼,𝛼 (𝜆, 𝑡, 𝑡0 ) , (162)
In fact,
∞ while
𝛼 𝛼 𝑘
𝐷Δ,𝑡 [ Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 )] = 𝐷Δ,𝑡 [ ∑ 𝜆 ℎ𝑘𝛼+𝛼−𝑗 (𝑡, 𝑡0 )] 𝛼
0 0
𝑘=0
𝐷Δ,𝑡0
𝑦 (𝑡) − 𝜆𝑦 (𝑡) = 0 (𝑡 > 𝑡0 ; 1 < 𝛼 ≤ 2; 𝜆 ∈ R) (163)

has the fundamental system of solutions given by
= ∑ 𝜆𝑘 ℎ𝑘𝛼−𝑗 (𝑡, 𝑡0 )
𝑘=0
𝑦1 (𝑡) = Δ 𝐹𝛼,𝛼 (𝜆, 𝑡, 𝑡0 ) , 𝑦2 (𝑡) = Δ 𝐹𝛼,𝛼−1 (𝜆, 𝑡, 𝑡0 ) .
∞ (164)
= ∑ 𝜆𝑘+1 ℎ(𝑘+1)𝛼−𝑗 (𝑡, 𝑡0 )
𝑘=−1 Next we derive the explicit solutions to (140) with 𝑚 = 2
0 of the form
= 𝜆 ℎ−𝑗 (𝑡, 𝑡0 )
𝛼 𝛽
∞ 𝐷Δ,𝑡0
𝑦 (𝑡) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡) − 𝜇𝑦 (𝑡) = 0
𝑘+1 (165)
+ ∑𝜆 ℎ(𝑘+1)𝛼−𝑗 (𝑡, 𝑡0 )
(𝑡 > 𝑡0 ; 𝑙 − 1 < 𝛼 ≤ 𝑙; 𝑙 ∈ N; 0 < 𝛽 < 𝛼)
𝑘=0

∞ with 𝜆, 𝜇 ∈ R.
= 𝜆 ∑ 𝜆𝑘 ℎ𝑘𝛼+𝛼−𝑗 (𝑡, 𝑡0 )
𝑘=0 Theorem 61. Let 𝑙 − 1 < 𝛼 ≤ 𝑙 (𝑙 ∈ N), 0 < 𝛽 < 𝛼, and
= 𝜆 Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 ) . 𝜆, 𝜇 ∈ R. Then the functions

(156) 𝜇𝑘 𝜕𝑘
𝑦𝑗 (𝑡) = ∑ 𝐹 (𝜆, 𝑡, 𝑡0 ) (166)
Moreover, 𝑘=0
𝑘! 𝜕𝜆𝑘 Δ 𝛼−𝛽,𝛼+𝑘𝛽+1−𝑗

yield the fundamental system of solutions to (165) provided that
𝛼−𝑘
𝐷Δ,𝑡 𝛼−𝑘
𝑦 (𝑡) = 𝐷Δ,𝑡
0 𝑗 0
[∑𝜆𝑠 ̂ℎ𝑠𝛼+𝛼−𝑗 (𝑡, 𝑡0 )] the series in (166) are convergent. Moreover, if 𝛼 + 1 − 𝑙 > 𝛽 >
𝑠=0
(157) 𝑙 − 1, then 𝑦𝑗 (𝑡), 𝑗 = 1, 2, . . . , 𝑙, in (166) satisfy (151).

𝑠
= ∑𝜆 ℎ𝑠𝛼+𝑘−𝑗 (𝑡, 𝑡0 ) . Proof. Let 𝑚 − 1 < 𝛽 ≤ 𝑚 (𝑚 ≤ 𝑙; 𝑚 ∈ N). Applying
𝑠=0 the Laplace transform to (165) and using (141) as in (152), we
It follows from (157) that obtain
𝑙
𝛼−𝑘
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 0 (𝑘, 𝑗 = 1, . . . , 𝑙; 𝑘 > 𝑗) , 𝑧𝑗−1
0 𝑗 L {𝑦 (𝑡)} (𝑧, 𝑡0 ) = ∑ 𝑑𝑗 , (167)
(158) 𝑗=1 𝑧𝛼 − 𝜆𝑧𝛽 − 𝜇
𝛼−𝑘
𝐷Δ,𝑡 𝑦
0 𝑘
(𝑡0 ) = 1 (𝑘 = 1, . . . , 𝑙) .
𝛼−𝑗 𝛽−𝑗
where 𝑑𝑗 = 𝐷Δ,𝑡0 𝑦(𝑡0 ) − 𝜆𝐷Δ,𝑡0 𝑦(𝑡0 ) (𝑗 = 1, . . . , 𝑚), 𝑑𝑗 =
If 𝑘 < 𝑗, then 𝛼−𝑗
𝐷Δ,𝑡0 𝑦(𝑡0 ) (𝑗 = 𝑚 + 1, . . . , 𝑙).

𝛼−𝑘
𝐷Δ,𝑡 𝑦 (𝑡) = ∑𝜆𝑠 ℎ𝑠𝛼+𝑘−𝑗 (𝑡, 𝑡0 ) For 𝑧 ∈ C and |𝜇𝑧−𝛽 /(𝑧𝛼−𝛽 − 𝜆)| < 1, we have
0 𝑗
𝑠=1
(159) 1 𝑧−𝛽 1

𝛼 𝛽
= 𝛼−𝛽

= ∑𝜆𝑠+1 ℎ𝑠𝛼+𝛼+𝑘−𝑗 (𝑡, 𝑡0 ) , 𝑧 − 𝜆𝑧 − 𝜇 𝑧 − 𝜆 1 − 𝜇𝑧 / (𝑧𝛼−𝛽 − 𝜆)
−𝛽

∞ (168)
𝑠=0
𝜇𝑘 𝑧−𝛽−𝑘𝛽
=∑ ,
and since 𝛼 + 𝑘 − 𝑗 ≥ 𝛼 + 1 − 𝑙 > 0 for any 𝑘, 𝑗 = 1, . . . , 𝑙, the 𝛼−𝛽 − 𝜆)𝑘+1
𝑘=0 (𝑧
following relations hold:
𝛼−𝑘
and hence (167) has the following representation:
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 0
0 𝑗
(𝑘, 𝑗 = 1, . . . , 𝑙; 𝑘 < 𝑗) . (160)
𝑙 ∞
𝑧𝑗−1−𝛽−𝑘𝛽
By (158) and (160), 𝑊𝛼 (𝑡0 ) = 1. Then 𝑦𝑗 (𝑡), 𝑗 = 1, . . . , 𝑙, L {𝑦 (𝑡)} (𝑧, 𝑡0 ) = ∑𝑑𝑗 ∑ 𝜇𝑘 𝑘+1
. (169)
yield the fundamental system of solutions to (143). 𝑗=1 𝑘=0 (𝑧𝛼−𝛽 − 𝜆)
14 Abstract and Applied Analysis

By (51), for 𝑧 ∈ C and |𝜆𝑧𝛽−𝛼 | < 1, we have for 𝑠 = 0, 𝑞, 𝑗 = 1, . . . , 𝑙, 𝑘 ∈ N+ . Besides, we also have 𝑠(𝛼 −
𝛽) + 𝑘𝛽 + 𝑞 − 𝑗 ≥ (𝛼 − 𝛽) + 𝛽 + 1 − 𝑙 = 𝛼 + 1 − 𝑙 > 0 for
𝑧𝑗−1−𝛽−𝑘𝛽 𝑧(𝛼−𝛽)−(𝛼+𝑘𝛽+1−𝑗) 𝛼−𝑞
𝑞, 𝑗 = 1, . . . , 𝑙, 𝑠, 𝑘 ∈ N+ . These imply that 𝐷Δ,𝑡0 𝑦𝑗 (𝑡0 ) = 0.
𝑘+1
= 𝑘+1
(𝑧𝛼−𝛽 − 𝜆) (𝑧𝛼−𝛽 − 𝜆) Thus the relations in (151) are valid. The proof is finished.

1 𝜕𝑘 Corollary 62. Consider that


= L { 𝑘 Δ 𝐹𝛼−𝛽,𝛼+𝑘𝛽+1−𝑗 (𝜆, 𝑡, 𝑡0 )} (𝑧, 𝑡0 ) .
𝑘! 𝜕𝜆
(170) 𝛽
𝛼
𝐷Δ,𝑡0
𝑦 (𝑡) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡) = 0
From (169) and (170), we derive the solution to (165) as (174)
𝑙
(𝑡 > 𝑡0 ; 𝑙 − 1 < 𝛼 ≤ 𝑙; 𝑙 ∈ N; 0 < 𝛽 < 𝛼)
𝑦 (𝑡) = ∑ 𝑑𝑗 𝑦𝑗 (𝑡) , (171)
𝑗=1
has its fundamental system of solution given by
where 𝑦𝑗 (𝑡) (𝑗 = 1, . . . , 𝑙) are given by (166). For 𝑞, 𝑗 =
1, . . . , 𝑙, the direct evaluation yields
𝑦𝑗 (𝑡) = Δ 𝐹𝛼−𝛽,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 ) (𝑗 = 1, . . . , 𝑙) . (175)
𝛼−𝑞
𝐷Δ,𝑡0 𝑦𝑗 (𝑡)

𝛼−𝑞 𝜇𝑘 𝜕𝑘 Finally, we find the explicit solutions to (140) with any 𝑚 ∈
= 𝐷Δ,𝑡0 ∑ 𝐹 (𝜆, 𝑡, 𝑡0 )
𝑘=0
𝑘! 𝜕𝜆𝑘 Δ 𝛼−𝛽,𝛼+𝑘𝛽+1−𝑗 N \ {1, 2}. It is convenient to rewrite (140) in the form


𝜇𝑘 𝜕𝑘 𝛼−𝑞 ∞ 𝑠 ̂ (172)
=∑ 𝐷 [∑𝜆 ℎ (𝑡, 𝑡0 )] 𝑚−2
𝑘=0
𝑘! 𝜕𝜆𝑘 Δ,𝑡0 𝑠=0 𝑠(𝛼−𝛽)+𝛼+𝑘𝛽−𝑗 𝛼
𝐷Δ,𝑡
𝛽 𝛼
𝑦 (𝑡) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡) − ∑ 𝐴 𝑘 𝐷Δ,𝑡𝑘 0 𝑦 (𝑡) = 0
0
𝑘=0

𝜇𝑘 𝜕𝑘 ∞ 𝑠 ̂ (176)
=∑ ∑𝜆 ℎ (𝑡, 𝑡0 ) . (𝑡 > 𝑡0 ; 𝑚 ∈ N \ {1, 2} ; 0 = 𝛼0 < 𝛼1
𝑘=0
𝑘! 𝜕𝜆𝑘 𝑠=0 𝑠(𝛼−𝛽)+𝑘𝛽+𝑞−𝑗
𝛼−𝑞 𝛼−𝑞 < ⋅ ⋅ ⋅ < 𝛼𝑚−2 < 𝛽 < 𝛼; 𝜆, 𝐴 0 , . . . , 𝐴 𝑚−2 ∈ R) .
For 𝑞 > 𝑗, 𝐷Δ,𝑡0 𝑦𝑗 (𝑡0 ) = 0, and for 𝑞 = 𝑗, 𝐷Δ,𝑡0 𝑦𝑗 (𝑡0 ) =
1. Thus we have 𝑊𝛼 (𝑡0 ) = 1. It follows from Lemma 58
that the functions 𝑦𝑗 (𝑡), 𝑗 = 1, 2, . . . , 𝑙 in (166) are linearly Theorem 63. Let 𝑚 ∈ N\{1, 2}, 𝑙−1 < 𝛼 ≤ 𝑙 (𝑙 ∈ N), let 𝛽 and
independent solutions, and then they yield the fundamental 𝛼1 , . . . , 𝛼𝑚−2 be such that 𝛼 > 𝛽 > 𝛼𝑚−2 > ⋅ ⋅ ⋅ > 𝛼1 > 𝛼0 = 0,
system of solutions to (165). Furthermore, if 𝑞 < 𝑗, then we and let 𝜆, 𝐴 0 , . . . , 𝐴 𝑚−2 ∈ R. Then the functions
rewrite (172) as follows:

𝐷Δ,𝑡0 𝑦𝑗 (𝑡) = 𝐷Δ,𝑡0 ̂ℎ𝛼−𝑗 (𝑡, 𝑡0 )


𝛼−𝑞 𝛼−𝑞

1
𝑦𝑗 (𝑡) = ∑ ( ∑ )

𝑛=0 𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
+ ∑𝜆𝑠 ̂ℎ𝑠(𝛼−𝛽)+𝑞−𝑗 (𝑡, 𝑡0 )
𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛

𝑠=1 𝑚−2
𝑘 𝜕𝑛
∞ 𝑘 𝑘 × [ ∏ (𝐴 ] ) ] ] 𝐹 𝑚−2 (𝜆, 𝑡, 𝑡0 )
𝜇 𝜕 ̂ ]=0 𝜕𝜆𝑛 Δ 𝛼−𝛽,𝛼+1−𝑗+∑]=0 (𝛽−𝛼] )𝑘]
+∑ ℎ (𝑡, 𝑡0 )
𝑘=1
𝑘! 𝜕𝜆𝑘 𝑘𝛽+𝑞−𝑗 (177)

𝜇𝑘 𝜕𝑘 ∞ 𝑠 ̂
+∑ ∑𝜆 ℎ (𝑡, 𝑡0 )
𝑘=1
𝑘! 𝜕𝜆𝑘 𝑠=1 𝑠(𝛼−𝛽)+𝑘𝛽+𝑞−𝑗 (173) yield the fundamental system of solutions to (176) provided that
the series in (177) are convergent. The inner sum is taken over
∞ all 𝑘0 , . . . , 𝑘𝑚−2 ∈ N0 such that 𝑘0 + ⋅ ⋅ ⋅ + 𝑘𝑚−2 = 𝑛. Moreover,
= ∑𝜆𝑠 ̂ℎ𝑠(𝛼−𝛽)+𝑞−𝑗 (𝑡, 𝑡0 ) if 𝛼 + 1 − 𝑙 > 𝛽 > 𝛼𝑚−2 + 𝑙 − 1, then 𝑦𝑗 (𝑡), 𝑗 = 1, 2, . . . , 𝑙, in
𝑠=1 (177) satisfy (151).

𝜇𝑘 𝜕𝑘 ̂ Proof. Let 𝑙𝑚−1 − 1 < 𝛽 ≤ 𝑙𝑚−1 , 𝑙𝑘 − 1 < 𝛼𝑘 < 𝑙𝑘 (𝑘 = 1, . . . , 𝑚 −
+∑ ℎ (𝑡, 𝑡0 )
𝑘=1
𝑘! 𝜕𝜆𝑘 𝑘𝛽+𝑞−𝑗 2; 0 ≤ 𝑙1 ≤ ⋅ ⋅ ⋅ ≤ 𝑙𝑚−1 ≤ 𝑙). Applying the Laplace transform to
(176) and using (141) as in (167), we obtain

𝜇𝑘 𝜕𝑘 ∞ 𝑠 ̂
+∑ ∑𝜆 ℎ (𝑡, 𝑡0 ) .
𝑘=1
𝑘! 𝜕𝜆𝑘 𝑠=1 𝑠(𝛼−𝛽)+𝑘𝛽+𝑞−𝑗
𝑙
𝑧𝑗−1
If 𝛼 + 1 − 𝑙 > 𝛽 > 𝑙 − 1, then 𝑠(𝛼 − 𝛽) + 𝑞 − 𝑗 ≥ (𝛼 − 𝛽) + 1 − 𝑙 > 0 L {𝑦 (𝑡)} (𝑧, 𝑡0 ) = ∑𝑑𝑗 , (178)
for 𝑘 = 0, 𝑞, 𝑗 = 1, . . . , 𝑙, 𝑠 ∈ N+ , and 𝑘𝛽 + 𝑞 − 𝑗 ≥ 𝛽 + 1 − 𝑙 > 0 𝑗=1 𝑧𝛼 − 𝜆𝑧𝛽 − ∑𝑚−2
𝑘=0 𝐴 𝑘 𝑧
𝛼𝑘
Abstract and Applied Analysis 15

where From (178), (180), and (182), we derive the solution to (176),
𝛼−𝑗 𝛽−𝑗 as
𝑑𝑗 = 𝐷Δ,𝑡0 𝑦 (𝑡0 ) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡0 )
𝑙
𝑚−2
𝛼 −𝑗 𝑦 (𝑡) = ∑ 𝑑𝑗 𝑦𝑗 (𝑡) , (183)
− ∑ 𝐴 𝑘 𝐷Δ,𝑡𝑘 0 𝑦 (𝑡0 ) (𝑗 = 1, . . . , 𝑙1 ) , 𝑗=1
𝑘=1

𝛼−𝑗
𝑑𝑗 = 𝐷Δ,𝑡0 𝑦 (𝑡0 ) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡0 )
𝛽−𝑗 which shows that arbitrary solution 𝑦(𝑡) can be expressed by
𝑦𝑗 (𝑡), 𝑗 = 1, . . . , 𝑙, where 𝑦𝑗 (𝑡) (𝑗 = 1, . . . , 𝑙) are given by (177).
𝑚−2 For 𝑞, 𝑗 = 1, . . . , 𝑙, the direct evaluation yields
𝛼 −𝑗
− ∑ 𝐴 𝑘 𝐷Δ,𝑡𝑘 0 𝑦 (𝑡0 ) (𝑗 = 𝑙1 + 1 . . . , 𝑙2 ) ,
𝛼−𝑞
𝑘=2 𝐷Δ,𝑡0 𝑦𝑗 (𝑡)
..
. ∞ 𝑚−2
𝛼−𝑞 { 1 𝑘
= 𝐷Δ,𝑡0 { ∑ ( ∑ ) [ ∏ (𝐴 ] ) ] ]
𝛼−𝑗
𝑑𝑗 = 𝐷Δ,𝑡0 𝑦 (𝑡0 ) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡0 )
𝛽−𝑗
(𝑗 = 𝑙𝑚−2 + 1, . . . , 𝑙𝑚−1 ) , 𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
{𝑛=0 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛 ]=0

𝛼−𝑗
𝑑𝑗 = 𝐷Δ,𝑡0 𝑦 (𝑡0 ) (𝑗 = 𝑙𝑚−1 + 1, . . . , 𝑙) . 𝜕𝑛 }
× 𝑛
𝐹𝛼−𝛽,𝛼+1−𝑗+∑𝑚−2 (𝛽−𝛼] )𝑘] (𝜆; 𝑡, 𝑡0 ) }
(179) 𝜕𝜆 Δ ]=0
}
Here ∑𝑛𝑘=𝑚 𝐴 𝑘 := 0 (𝑚 > 𝑛). For 𝑧 ∈ C and
∞ 𝑚−2
| ∑𝑚−2
𝑘=0 𝐴 𝑘 𝑧
𝛼𝑘 −𝛽
/(𝑧𝛼−𝛽 − 𝜆)| < 1, we have 1 𝑘
= ∑( ∑ ) [ ∏ (𝐴 ] ) ] ]
1 𝑛=0 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 ! ]=0
𝑧𝛼 − 𝜆𝑧𝛽 − ∑𝑚−2
𝑘=0 𝐴 𝑘 𝑧𝛼𝑘 𝜕 𝑛 ∞
𝜆𝑠 ̂ℎ𝑠(𝛼−𝛽)+𝛼−𝑗+∑𝑚−2 (𝛽−𝛼] )𝑘] (𝑡, 𝑡0 )]
𝛼−𝑞
× 𝐷Δ,𝑡 [ ∑
𝜕𝜆𝑛 0 ]=0
𝑧−𝛽 1 𝑠=0
= ⋅
𝑧𝛼−𝛽 − 𝜆 (1 − ∑𝑚−2
𝑘=0 𝐴 𝑘 𝑧
𝛼𝑘 −𝛽 / (𝑧𝛼−𝛽 − 𝜆)) ∞
1 𝑚−2
𝑘
= ∑( ∑ ) [ ∏ (𝐴 ] ) ] ]
∞ 𝑚−2 𝑛 𝑛=0 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 ! ]=0
𝑧−𝛽
=∑ 𝑛+1
( ∑ 𝐴 𝑘 𝑧𝛼𝑘 −𝛽 ) (180)
𝑛=0 (𝑧
𝛼−𝛽 − 𝜆) 𝑘=0 𝜕𝑛 ∞ 𝑠 ̂
× ∑𝜆 ℎ 𝑚−2 (𝑡, 𝑡0 ) .
𝜕𝜆𝑛 𝑠=0 𝑠(𝛼−𝛽)+∑]=0 (𝛽−𝛼] )𝑘] +𝑞−𝑗

𝑛!
= ∑( ∑ ) (184)
𝑛=0 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
𝛼−𝑞 𝛼−𝑞
𝑚−2 −𝛽−∑𝑚−2
For 𝑞 > 𝑗, 𝐷Δ,𝑡0 𝑦𝑗 (𝑡0 ) = 0, and for 𝑞 = 𝑗, 𝐷Δ,𝑡0 𝑦𝑗 (𝑡0 ) =
𝑧 ]=0 (𝛽−𝛼] )𝑘]
𝑘 1. Thus we have 𝑊𝛼 (𝑡0 ) = 1. It follows from Lemma 58
× [ ∏ (𝐴 ] ) ] ] 𝑛+1
,
]=0 (𝑧𝛼−𝛽 − 𝜆) that the functions 𝑦𝑗 (𝑡), 𝑗 = 1, 2, . . . , 𝑙 in (177) are linearly
independent solutions and then they yield the fundamental
if we also take into account the following relation:
system of solutions to (176). Furthermore, if 𝑞 < 𝑗, then we
𝑛
(𝑥0 + ⋅ ⋅ ⋅ + 𝑥𝑚−2 ) rewrite (184) as follows:
𝑚−2 (181) 𝛼−𝑞
𝑛! 𝐷Δ,𝑡0 𝑦𝑗 (𝑡)
=( ∑ ) ∏ 𝑥𝑘] ,
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 ! ]=0 𝜐
= 𝐷Δ,𝑡0 ̂ℎ𝛼−𝑗 (𝑡, 𝑡0 )
𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛 𝛼−𝑞

where the summation is taken over all 𝑘0 , . . . , 𝑘𝑚−2 ∈ N0 such



that 𝑘0 + ⋅ ⋅ ⋅ + 𝑘𝑚−2 = 𝑛.
+ ∑𝜆𝑠 ̂ℎ𝑠(𝛼−𝛽)+∑𝑚−2 (𝛽−𝛼] )𝑘] +𝑞−𝑗 (𝑡, 𝑡0 )
In addition, for 𝑧 ∈ C and |𝜆𝑧𝛽−𝛼 | < 1, we have 𝑠=1
]=0

𝑗−1−𝛽−∑𝑚−2
]=0 (𝛽−𝛼] )𝑘]
𝑧 ∞
𝑛+1 1
(𝑧𝛼−𝛽 − 𝜆) +∑( ∑ )
𝑛=1 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
(𝛼−𝛽)−(𝛼+1−𝑗+∑𝑚−2
]=0 (𝛽−𝛼] )𝑘] )
𝑧 𝑚−2
= 𝑛+1 𝑘 𝜕𝑛 ̂
(𝑧𝛼−𝛽 − 𝜆) × [ ∏ (𝐴 ] ) ] ] ℎ 𝑚−2 (𝑡, 𝑡0 )
]=0 𝜕𝜆𝑛 ∑]=0 (𝛽−𝛼] )𝑘] +𝑞−𝑗
1 𝜕𝑛
= LΔ,𝑡0 { 𝑛 𝐹𝛼−𝛽,𝛼+1−𝑗+∑𝑚−2 (𝛽−𝛼] )𝑘] (𝜆; 𝑡, 𝑡0 )} (𝑧) . ∞
𝑛! 𝜕𝜆 ∇ ]=0 1
+∑( ∑ )
(182) 𝑛=1 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
16 Abstract and Applied Analysis

𝑚−2
𝑘 𝜕𝑛 ∞ 𝑠 ̂ Using the inverse Laplace transform L−1 Δ from here, we
× [ ∏ (𝐴 ] ) ] ] ∑𝜆 ℎ 𝑚−2 (𝑡, 𝑡0 )
]=0 𝜕𝜆𝑛 𝑠=1 𝑠(𝛼−𝛽)+∑]=0 (𝛽−𝛼] )𝑘] +𝑞−𝑗 obtain a particular solution to (186) in the form

∞ L {𝑓 (𝑡)} (𝑧, 𝑡0 )
= ∑𝜆𝑠 ̂ℎ𝑠(𝛼−𝛽)+𝑞−𝑗 (𝑡, 𝑡0 ) 𝑦 (𝑡) = L−1 [ ] (𝑡) . (189)
𝐴 0 + ∑𝑚
𝑘=1 𝐴 𝑘 𝑧
𝛼𝑘
𝑠=1
Using the Laplace convolution formula

1
+∑( ∑ ) L {𝑓 ∗ 𝑔} (𝑧, 𝑡0 ) = L {𝑓} (𝑧, 𝑡0 ) L {𝑔} (𝑧, 𝑡0 ) , (190)
𝑛=1 𝑘 +⋅⋅⋅+𝑘 =𝑛 𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
0 𝑚−2

we can introduce the Laplace fractional analog of the Green


𝑚−2
𝜕𝑛 𝑘] function as follows:
× [ ∏ (𝐴 ] ) ] 𝑛 ̂ℎ∑𝑚−2 (𝛽−𝛼] )𝑘] +𝑞−𝑗 (𝑡, 𝑡0 )
𝜕𝜆 ]=0
]=0 1
𝐺𝛼1 ,...,𝛼𝑚 (𝑡) = L−1 { } (𝑡) ,
∞ 𝑃𝛼1 ,...,𝛼𝑚 (𝑧)
1
+∑( ∑ ) (191)
𝑛=1 𝑘 +⋅⋅⋅+𝑘 =𝑛 𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 ! 𝑚
𝛼𝑘
0 𝑚−2
𝑃𝛼1 ,...,𝛼𝑚 (𝑧) = 𝐴 0 + ∑ 𝐴 𝑘 𝑧 ,
𝑚−2 𝑛 ∞ 𝑘=1
𝜕 𝑘]
× [ ∏ (𝐴 ] ) ] 𝑛 ∑𝜆𝑠 ̂ℎ𝑠(𝛼−𝛽)+∑𝑚−2 (𝛽−𝛼] )𝑘] +𝑞−𝑗 (𝑡, 𝑡0 ) . and we can express a particular solution of (152) in the form
]=0 𝜕𝜆 𝑠=1
]=0

(185) of the Laplace convolution 𝐺𝛼1 ,...,𝛼𝑚 (𝑡) and ℎ𝑟 (𝑡, 𝑡0 ):

𝑦 (𝑡) = (𝑓 ∗ 𝐺𝛼1 ,...,𝛼𝑚 ) (𝑡) . (192)


If 𝛼 + 1 − 𝑙 > 𝛽 > 𝛼𝑚−2 + 𝑙 − 1, then 𝑠(𝛼 − 𝛽) + 𝑞 − 𝑗 ≥
(𝛼 − 𝛽) + 1 − 𝑙 > 0 for 𝑛 = 0, 𝑞, 𝑗 = 1, . . . , 𝑙, 𝑠 ∈ N+ , and Generally speaking, we can consider (186) with 𝐴 𝑚 = 1. First
∑𝑚−2
]=0 (𝛽 − 𝛼] )𝑘] + 𝑞 − 𝑗 ≥ 𝛽 − 𝛼𝑖 + 1 − 𝑙 ≥ 𝛽 − 𝛼𝑚−2 + 1 − 𝑙 > we derive a particular solution to (186) with 𝑚 = 1 in the form
0 for 𝑠 = 0, 𝑞, 𝑗 = 1, . . . , 𝑙, 𝑛 ∈ N+ . Besides, we also have
𝛼
𝑠(𝛼 − 𝛽) + ∑𝑚−2 𝐷Δ,𝑡 𝑦 (𝑡) − 𝜆𝑦 (𝑡) = 𝑓 (𝑡) (𝑡 > 𝑡0 ; 𝛼 > 0) .
]=0 (𝛽 − 𝛼] )𝑘] + 𝑞 − 𝑗 ≥ (𝛼 − 𝛽) + 𝛽 − 𝛼𝑖 + 1 − 𝑙 ≥ 0
(193)
(𝛼 − 𝛽) + 𝛽 − 𝛼𝑚−2 + 1 − 𝑙 > 0 for 𝑞, 𝑗 = 1, . . . , 𝑙, 𝑠, 𝑛 ∈ N+ .
𝛼−𝑞 Theorem 64. Let 𝛼 > 0, 𝜆 ∈ R. Then (193) is solvable, and its
These imply that 𝐷Δ,𝑡0 𝑦𝑗 (𝑡0 ) = 0. Thus the relations in (151)
particular solution has the form
are valid. The result follows.
𝑦 (𝑡) = (𝑓 ∗ Δ 𝐹𝛼,𝛼 (𝜆, 𝑠, 𝑡0 )) (𝑡) (194)
8. Nonhomogeneous Equations with provided that the integral in the right-hand side of (194) is
Constant Coefficients convergent.
In Section 7, we have applied the Laplace transform method
Proof. Equation (193) is the same as (186) with 𝑚 = 1, 𝛼1 = 𝛼,
to derive explicit solutions to the homogeneous equations
and 𝐴 1 = 1, 𝐴 0 = −𝜆 and (191) takes the form
(140) with the Liouville fractional derivatives. Here we use
this approach to find particular solutions to the correspond- 1
𝐺𝛼 (𝑡) = L−1 { } (𝑡) = Δ 𝐹𝛼,𝛼 (𝜆, 𝑡, 𝑡0 ) . (195)
ing nonhomogeneous equations 𝑧𝛼 − 𝜆
𝑚 Thus (192), with 𝐺𝛼1 ,...,𝛼𝑚 (𝑡) = 𝐺𝛼 (𝑡), yields (194). Theorem is
𝛼
∑ 𝐴 𝑘 𝐷Δ,𝑡𝑘 0 𝑦 (𝑡) + 𝐴 0 𝑦 (𝑡) proved.
𝑘=1 (186)
Next we derive a particular solution to (186) with 𝑚 = 2
= 𝑓 (𝑡) (𝑡 > 𝑡0 ; 0 < 𝛼1 < ⋅ ⋅ ⋅ < 𝛼𝑚 ; 𝑚 ∈ N) of the form
𝛼 𝛽
with real 𝐴 𝑘 ∈ R (𝑘 = 0, . . . , 𝑚). 𝐷Δ,𝑡0
𝑦 (𝑡) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡) − 𝜇𝑦 (𝑡) = 𝑓 (𝑡) ,
By (141)-(142), for suitable functions 𝑦, the Laplace (196)
𝛼
transform of 𝐷Δ,𝑡 𝑦 is given by (𝑡 > 𝑡0 ; 𝛼 > 𝛽 > 0) .
0

Theorem 65. Let 𝛼 > 𝛽 > 0, 𝜆, 𝜇 ∈ R. Then (196) is solvable,


𝛼
L {𝐷Δ,𝑡0
𝑦 (𝑡)} (𝑧, 𝑡0 ) = 𝑧𝛼 L {𝑦 (𝑡)} (𝑧, 𝑡0 ) . (187) and its particular solution has the form

Applying the Laplace transform to (186) and taking (187) into 𝑦 (𝑡) = (𝑓 ∗ 𝐺𝛼,𝛽;𝜆,𝜇 (𝑠)) (𝑡) , (197)
account, we have ∞
𝜇𝑘 𝜕𝑘
𝐺𝛼,𝛽;𝜆,𝜇 (𝑡) = ∑ 𝐹 (𝜆, 𝑡, 𝑡0 ) (198)
𝑚
𝛼𝑘 𝑘=0
𝑘! 𝜕𝜆𝑘 Δ 𝛼−𝛽,𝛼+𝑘𝛽
[𝐴 0 + ∑ 𝐴 𝑘 𝑧 ] L {𝑦 (𝑡)} (𝑧, 𝑡0 ) = L {𝑓 (𝑡)} (𝑧, 𝑡0 ) .
𝑘=1 provided that the series in (198) and the integral in (197) are
(188) convergent.
Abstract and Applied Analysis 17

Proof. Equation (196) is the same as (186) with 𝑚 = 2, 𝛼2 = 𝛼, Proof. Equation (203) is the same equation as (186) with 𝛼𝑚 =
𝛼1 = 𝛽, 𝐴 2 = 1, 𝐴 1 = −𝜆, and 𝐴 0 = −𝜇, and (191) is given by 𝛼, 𝛼𝑚−1 = 𝛽, 𝐴 𝑚 = 1, . . . , 𝐴 𝑚−1 = −𝜆, and with −𝐴 𝑘 instead
of 𝐴 𝑘 for 𝑘 = 0, . . . , 𝑚 − 2. Since 𝛼0 = 0, (191) takes the form
1
𝐺𝛼,𝛽 (𝑡) = L−1 { } (𝑡) . (199) 1
𝑧𝛼 − 𝜆𝑧𝛽 − 𝜇 𝐺𝛼1 ,...,𝛼𝑚−2 ,𝛽,𝛼;𝜆 (𝑡) = L−1 { } (𝑡) .
𝑧𝛼 − 𝜆𝛼𝛽 − ∑𝑚−2
𝑘=0 𝐴 𝑘 𝑧
𝛼𝑘

According to (168) for 𝑧 ∈ C and |𝜇𝑧−𝛽 /(𝑧𝛼−𝛽 − 𝜆)| < 1, we (206)


have
For 𝑧 ∈ C and | ∑𝑚−2
𝑘=0 𝐴 𝑘 𝑧
𝛼𝑘 −𝛽
/(𝑧𝛼−𝛽 − 𝜆)| < 1, in accordance
∞ −𝛽−𝑛𝛽
𝑧 with (180), we have
𝐺𝛼,𝛽;𝜆,𝜇 (𝑡) = L−1 { ∑ 𝜇𝑛 𝑛+1
} (𝑡) . (200)
𝑛=0 (𝑧𝛼−𝛽 − 𝜆) 𝐺𝛼1 ,...,𝛼𝑚−2 ,𝛽,𝛼;𝜆 (𝑡)

In addition, for 𝑧 ∈ C and |𝜆𝑧𝛽−𝛼 | < 1, we have {∞ 𝑛!


= L−1 { ∑ ( ∑ )
𝑘 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
𝑧−𝛽−𝑛𝛽 1 𝜕𝑛 {𝑛=0 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛
0 (207)
𝑛+1
= L { 𝑛 Δ 𝐹𝛼−𝛽,𝛼+𝑛𝛽 (𝜆, 𝑡, 𝑡0 )} (𝑧, 𝑡0 ) ,
(𝑧𝛼−𝛽 − 𝜆) 𝑛! 𝜕𝜆 𝑚−2
𝑚−2
𝑘 𝑧−𝛽−∑𝜐=0 (𝛽−𝛼𝜐 )𝑘𝜐 }
(201) × [ ∏ (𝐴 𝜐 ) 𝜐 ] 𝑛+1 }
(𝑡) .
𝜐=0 (𝑧𝛼−𝛽 − 𝜆)
}
and hence (200) takes the following form:
For 𝑧 ∈ C and |𝜆𝑧𝛽−𝛼 | < 1, we have
∞ 𝑛 𝑛
𝜇 𝜕
𝐺𝛼,𝛽;𝜆,𝜇 (𝑡) = ∑ 𝐹
𝑛 Δ 𝛼−𝛽,𝛼+𝑛𝛽
(𝜆, 𝑡, 𝑡0 ) . (202) 𝑚−2
𝑧−𝛽−∑𝜐=0 (𝛽−𝛼𝜐 )𝑘𝜐
𝑛=0 𝑛! 𝜕𝜆
𝑛+1
(𝑧𝛼−𝛽 − 𝜆)
Thus the result in (197) follows from (192) with 𝐺𝛼1 ,...,𝛼𝑚 (𝑡) =
𝐺𝛼,𝛽;𝜆,𝜇 (𝑡). 1 𝜕𝑛
= L { 𝑛 𝐹𝛼−𝛽,𝛼+∑𝑚−2 (𝛽−𝛼𝜐 )𝑘𝜐 (𝜆, 𝑡, 𝑡0 )} (𝑧, 𝑡0 ) .
𝑛! 𝜕𝜆 Δ 𝜐=0

Finally, we find a particular solution to (186) with any 𝑚 ∈ (208)


N \ {1, 2}. It is convenient to rewrite (186) just as (176) in the
form The proof is finished.

𝑚−2
As in the case of ordinary differential equations, a general
𝛼 𝛽 𝛼 solution to the nonhomogeneous equation (186) is a sum
𝐷Δ,𝑡0
𝑦 (𝑡) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡) − ∑ 𝐴 𝑘 𝐷Δ,𝑡𝑘 0 𝑦 (𝑡) − 𝐴 0 𝑦 (𝑡)
𝑘=1 (203) of a particular solution to this equation and of the general
solution to the corresponding homogeneous equation (140).
= ̂ℎ𝑟 (𝑡, 𝑡0 ) (𝑡 > 𝑡0 ) Therefore, the results established in Section 7 and in Section 8
can be used to derive general solutions to the nonhomo-
with 𝑚 ∈ N \ {1, 2}, 0 < 𝛼1 < ⋅ ⋅ ⋅ < 𝛼𝑚−2 < 𝛽 < 𝛼, and geneous equations (193), (196), and (203). The following
𝜆, 𝐴 0 , . . . , 𝐴 𝑚−2 ∈ R. statements can thus be derived from Theorems 59, 64, 61, 65
and Theorems 63 and 66, respectively.
Theorem 66. Let 𝑚 ∈ N \ {1, 2}, 0 = 𝛼0 < 𝛼1 < ⋅ ⋅ ⋅ < 𝛼𝑚−2 <
𝛽 < 𝛼, and let 𝜆, 𝐴 0 , . . . , 𝐴 𝑚−2 ∈ R. Then (203) is solvable, Theorem 67. Let 𝑙 − 1 < 𝛼 ≤ 𝑙 (𝑙 ∈ N), 𝜆 ∈ R. Then (193) is
and its particular solution has the form solvable, and its general solution is given by
𝑙
𝑦 (𝑡) = (𝑓 ∗ 𝐺𝛼1 ,...,𝛼𝑚−2 ,𝛽,𝛼;𝜆 (𝑠)) (𝑡) , (204)
𝑦 (𝑡) = ∑ 𝑐𝑗Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 ) + 𝑓 (𝑡) , (209)
𝑗=1
𝐺𝛼1 ,...,𝛼𝑚−2 ,𝛽,𝛼;𝜆 (𝑡)
where 𝑐𝑗 (𝑗 = 1, . . . , 𝑙) are arbitrary real constants.

1
=∑( ∑ ) Theorem 68. Let 𝑙 − 1 < 𝛼 ≤ 𝑙 (𝑙 ∈ N), 0 < 𝛽 < 𝛼, 𝜆, 𝜇 ∈ R.
𝑛=0 𝑘 +⋅⋅⋅+𝑘 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
0 𝑚−2 Then (196) is solvable, and its general solution has the form
𝑚−2
𝑘 𝜕𝑛 𝑙 ∞
𝜇𝑛 𝜕𝑛
× [ ∏ (𝐴 𝜐 ) 𝜐 ] 𝐹 𝑚−2 (𝜆, 𝑡, 𝑡0 ) 𝑦 (𝑡) = ∑𝑐𝑗 ∑ 𝐹 (𝜆, 𝑡, 𝑡0 )
𝜕𝜆𝑛 Δ 𝛼−𝛽,𝛼+∑𝜐=0 (𝛽−𝛼𝜐 )𝑘𝜐 𝑛 Δ 𝛼−𝛽,𝛼+𝑛𝛽+1−𝑗
𝑛=0 𝑛! 𝜕𝜆
𝜐=0
𝑗=1 (210)
(205)
+ (𝑓 ∗ 𝐺𝛼,𝛽,𝜆,𝜇 (𝑠)) (𝑡) ,
provided that the series (205) and integral in (204) are
convergent. The inner sum is taken over all 𝑘0 , . . . , 𝑘𝑚−2 such where 𝐺𝛼,𝛽,𝜆,𝜇 (𝑡) is given by (198) and 𝑐𝑗 (𝑗 = 1, . . . , 𝑙) are
that 𝑘0 + ⋅ ⋅ ⋅ + 𝑘𝑚−2 = 𝑛. arbitrary real constants.
18 Abstract and Applied Analysis

Theorem 69. Let 𝑚 ∈ N \ {1, 2}, 𝑙 − 1 < 𝛼 ≤ 𝑙 (𝑙 ∈ N), let 𝛽 Conflict of Interests
and 𝛼1 , 𝛼2 , . . . , 𝛼𝑚−2 be such that 𝛼 > 𝛽 > 𝛼𝑚−2 > ⋅ ⋅ ⋅ > 𝛼1 >
𝛼0 = 0 and 𝛼 − 𝑙 + 1 ≥ 𝛽, and let 𝜆, 𝐴 0 , . . . , 𝐴 𝑚−2 ∈ R. Then The authors declare that there is no conflict of interests
(203) is solvable, and its general solution is given by regarding the publication of this paper.

Authors’ Contribution
𝑙 ∞
1
𝑦 (𝑡) = ∑𝑐𝑗 ∑ ( ∑ ) All authors contributed equally and significantly to the
𝑗=1 𝑛=0 𝑘 +⋅⋅⋅+𝑘 =𝑟 𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
0 𝑚−2 writing of this paper. All authors read and approved the final
𝑚−2 paper.
𝑘
× [ ∏ (𝐴 𝜐 ) 𝜐 ]
𝜐=0
(211)
Acknowledgments
𝑛
𝜕 First, the authors are very grateful to the referees for their
⋅ 𝐹 𝑚−2 (𝜆, 𝑡, 𝑡0 )
𝜕𝜆𝑛 Δ 𝛼−𝛽,𝛼+1−𝑗+∑𝜐=0 (𝛽−𝛼𝜐 )𝑘𝜐 careful reading of the paper, and lots of valuable comments
and suggestions, which greatly improve this manuscript.
+ (𝑓 ∗ 𝐺𝛼1 ,...,𝛼𝑚−2, 𝛽,𝛼;𝜆 (𝑠)) (𝑡) ,
Next, this work was supported by the National Natural Sci-
ence Foundation of China (11171286) and by Jiangsu Province
where 𝐺𝛼1 ,...,𝛼𝑚−2, 𝛽,𝛼;𝜆 (𝑡) is given by (205) and 𝑐𝑗 (𝑗 = 1, . . . , 𝑙) Colleges and Universities Graduate Scientific Research Inno-
vative Program (CXZZ12-0974).
are arbitrary real constant.

References
9. Conclusions
[1] M. Holm, “Sum and difference compositions in discrete frac-
In this paper, we first give a generalized definition of frac- tional calculus,” CUBO A Mathematical Journal, vol. 13, no. 3,
tional Δ-power function on general time scale by inversion pp. 153–184, 2011.
of Laplace transform and shift transform. The fractional Δ- [2] F. M. Atici and P. W. Eloe, “A transform method in discrete frac-
power function on a time scale is an important basis of tional calculus,” International Journal of Difference Equations,
fractional Δ-integral and fractional Δ-differential on time vol. 2, pp. 165–176, 2007.
scales. Then, based on the fractional Δ-power function, [3] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory and
Applications of Fractional Differentia Equation, Elsevier B.V.,
we give a new definition of Riemann-Liouville fractional
London, UK, 2006.
Δ-integral and Riemann-Liouville fractional Δ-derivative
[4] J. Cheng, Theory of Fractional Difference Equation, Xiamen
on time scales. Some of properties of Riemann-Liouville University Press, Fujian, China, 2011.
fractional Δ-integral and Riemann-Liouville fractional Δ- [5] C. Kou, H. Zhou, and Y. Yan, “Existence of solutions of initial
derivative on time scales are studied in detail. On this value problems for nonlinear fractional differential equations
basis, equivalencies of Cauchy type problem with Riemann- on the half-axis,” Nonlinear Analysis, Theory, Methods and
Liouville fractional Δ-derivative and nonlinear Volterra inte- Applications, vol. 74, no. 17, pp. 5975–5986, 2011.
gral equation are obtained. By employing Laplace transform, [6] R. P. Agarwal, V. Lakshmikantham, and J. J. Nieto, “On the
we derive explicit solutions to homogeneous and nonho- concept of solution for fractional differential equations with
mogeneous equations of Riemann-Liouville fractional Δ- uncertainty,” Nonlinear Analysis, Theory, Methods and Applica-
derivative with constant coefficient. We give the conditions tions, vol. 72, no. 6, pp. 2859–2862, 2009.
when the solutions of linear fractional differential equation [7] Y. Liu, W. Zhang, and X. Liu, “A sufficient condition for
will be linearly independent and when these linearly inde- the existence of a positive solution for a nonlinear fractional
pendent solutions form the fundamental system of solutions. differential equation with the Riemann-Liouville derivative,”
On the other hand, we know that continuous fractional Applied Mathematics Letters, vol. 25, no. 11, pp. 1986–1992, 2012.
differential equation theory and discrete fractional difference [8] V. Lakshmikantham and A. S. Vatsala, “Basic theory of frac-
equation theory have been studied by many authors, the tional differential equations,” Nonlinear Analysis, Theory, Meth-
ods and Applications, vol. 69, no. 8, pp. 2677–2682, 2008.
existence and uniqueness of the solution to the boundary
[9] K. Diethelm, The Analysis of Fractional Differential Equations,
value problems for fractional differential equations have
Springer, Berlin, Germany, 2010.
been studied a lot by many methods involving partial order
[10] R. A. C. Ferreira and D. M. Torres, “Fractional h-
method, fixed point method, lower and upper solutions
difference equations arising from the calculus of variations,”
method, transform method, and so on. For example, for the http://arxiv.org/abs/1101.5904v1 .
recent developments about continuous fractional differential [11] C. S. Goodrich, “Existence and uniqueness of solutions to
equations and discrete fractional difference equations, one a fractional difference equation with nonlocal conditions,”
can refer to [3, 5–11] and the references therein. However, the Computers and Mathematics with Applications, vol. 61, no. 2, pp.
fractional differential equation theory on time scales is still an 191–202, 2011.
open problem. Therefore, the authors believe that the present [12] M. Bohner and A. Peterson, Dynamic Equations on Time Scales,
work will potentiate further research in the study of fractional An Introduction with Applications, Birkhäauser, Boston, Mass,
differential equation theory on time scales in the future. USA, 2001.
Abstract and Applied Analysis 19

[13] M. Bohner and A. Peterson, Advances in Dynamic Equations on


Time Scales, Birkhäauser, Boston, Mass, USA, 2003.
[14] M. Bohner and G. S. Guseinov, “Partial differentiation on time
scales,” Dynamic Systems and Applications, vol. 13, no. 3-4, pp.
351–379, 2004.
[15] J. M. Davis, I. A. Gravagne, B. J. Jackson, R. J. Marks II, and
A. A. Ramos, “The Laplace transform on time scales revisited,”
Journal of Mathematical Analysis and Applications, vol. 332, no.
2, pp. 1291–1307, 2007.
[16] P. A. Williams, Unifying fractional calculus with time scales
[Ph.D. thesis], The University of Melbourne, Parkville, Australia,
2012.
[17] N. R. D. O. Bastos, Fractional calculus on time scales [Ph.D.
thesis], The University of Aveiro, Aveiro, Portugal, 2012.
[18] M. Bohner and G. S. Guseinov, “The convolution on time
scales,” Abstract and Applied Analysis, vol. 2007, Article ID
58373, 24 pages, 2007.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 475350, 11 pages
http://dx.doi.org/10.1155/2013/475350

Research Article
On Solutions to Fractional Discrete Systems with
Sequential ℎ-Differences

MaBgorzata Wyrwas, Dorota Mozyrska, and Ewa Girejko


Faculty of Computer Science, Bialystok University of Technology, Wiejska 45A, 15-351 Białystok, Poland

Correspondence should be addressed to Małgorzata Wyrwas; m.wyrwas@pb.edu.pl

Received 28 August 2013; Accepted 19 October 2013

Academic Editor: Ali H. Bhrawy

Copyright © 2013 Małgorzata Wyrwas et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

We study the subject of a behaviour of the solutions of systems with sequential fractional ℎ-differences. We give formulas for
the unique solutions to initial value problems for systems in linear and semilinear cases. Moreover, the sufficient condition that
guaranties the positivity of considered systems is presented.

1. Introduction in continuous case with different types of derivatives like


Caputo type, Riemann-Liouville type, and Hadamard type,
The first definition of the fractional derivative was introduced see, for example, [9, 12, 25–27]. In parallel with this paper
by Liouville and Riemann at the end of the 19th century. we developed the theory of fractional ℎ-differences with
Later on, in the late 1960s, this idea was used by engineers sequential operators in the paper [28], where the approx-
for modeling various processes. Thus the fractional calculus imation of continuous fractional sequential derivative is
started to be exploited since that time. This calculus is a field considered. We compute different formulas of solutions and
of mathematics that grows out of the traditional definitions of then we try to check the system’s behaviour, precisely the
integrals, derivatives, and difference operators and deals with positivity of solutions. As far as we know the subject of
fractional integrals, derivatives, and differences of any order. positivity is well developed for fractional linear systems
Many authors prove that fractional differential and difference with continuous time; see [8, 29–32]. However, positivity
equations are more adequate for modeling physical and of fractional discrete systems with sequential ℎ-differences
chemical processes than integer-order equations. Fractional is still a field to be examined. In the present paper we
differential and difference equations describe many phenom- open our studies in this field. We give formulas for the
ena arising in engineering, physics, economics, and science. unique solutions to initial problems for systems in linear and
In fact, several applications can be found in viscoelasticity, semilinear cases. Moreover, the positivity of systems with
electrochemistry, electromagnetic, and so forth. For example, sequential ℎ-differences is considered. We consider systems
Machado [1] gave a novel method for the design of fractional with sequential ℎ-differences of Caputo type, while in [32]
order digital controllers. Fractional difference calculus has systems with Grünwald-Letnikov operator are studied. It
been investigated by many authors, for example, [2–12] and is worth to add that in [32] (and references within) the
others. In particular, different delta and nabla type fractional considered systems are not of sequential type. Additionally,
differences have been studied in [13–15], where the authors we have the exact formulas of the solutions of the systems
relate these differences by deriving some dual identities. The both with the sequential differences and with the ordinary
calculus of fractional ℎ-differences was given, for instance, differences while in [32] the recurrence form of solution for
in [5, 10, 16–19]. The properties of systems defined by the discrete systems with Grünwald-Letnikov difference is given.
fractional difference equations where studied, for example, The paper is organized as follows. In Section 2 all
in [20–24]. There exist definitions of sequential operators preliminary definitions, facts, and notations are gathered.
2 Abstract and Applied Analysis

Section 3 presents systems with sequential fractional differ- The next definition with another notation was stated in
ences with results on uniqueness of solutions. We include [17]. Here we use more suitable summations.
semilinear systems in Section 4. The Section 5 concerns pos-
itivity of considered systems. Finally, the illustrative example Definition 4. For a function 𝑥 ∈ F(ℎN)𝑎 the fractional ℎ-sum
is presented. of order 𝛼 > 0 is given by

ℎ 𝑛
2. Preliminaries (𝑎 Δ−𝛼
ℎ 𝑥) (𝑡) := ∑ (𝑡 − 𝜎 (𝑎 + 𝑘ℎ))(𝛼−1)
ℎ 𝑥 (𝑎 + 𝑘ℎ) , (6)
Γ (𝛼) 𝑘=0
Let us denote by F𝐷 the set of all real valued functions
defined on 𝐷. Let ℎ > 0, 𝛼 > 0 and put (ℎN)𝑎 := {𝑎, 𝑎 + where 𝑡 = 𝑎 + (𝛼 + 𝑛)ℎ, 𝑛 ∈ N0 . Moreover, we define
ℎ, 𝑎 + 2ℎ, . . .}, where 𝑎 ∈ R. Let (𝑎 Δ0ℎ 𝑥)(𝑡) := 𝑥(𝑡).
It is important to notice that the operator 𝑎 Δ−𝛼
ℎ changes
R𝑁 𝑁
+ = {𝑥 ∈ R : 𝑥𝑖 ≥ 0, 1 ≤ 𝑖 ≤ 𝑁} , 𝑁 ∈ N1 . (1) the domains of functions.

Then the operator 𝜎 : (ℎN)𝑎 → (ℎN)𝑎 is defined by 𝜎(𝑡) := Remark 5. Note that 𝑎 Δ−𝛼
ℎ : F(ℎN)𝑎 → F(ℎN)𝑎+𝛼ℎ .
𝑡 + ℎ. The next two definitions of ℎ-difference operator were According to the definition of ℎ-factorial function the
originally given in [16, 17]. formula given in Definition 4 can be rewritten as
𝑛
Definition 1. For a function 𝑥 ∈ F(ℎN)𝑎 the forward ℎ- Γ (𝛼 + 𝑛 − 𝑘)
(𝑎 Δ−𝛼 𝛼
ℎ 𝑥) (𝑡) = ℎ ∑ 𝑥 (𝑎 + 𝑘ℎ)
difference operator is defined as 𝑘=0
Γ (𝛼) Γ (𝑛 − 𝑘 + 1)
(7)
𝑛
𝑥 (𝜎 (𝑡)) − 𝑥 (𝑡) 𝑛−𝑘+𝛼−1
(Δ ℎ 𝑥) (𝑡) := , 𝑡 = 𝑎 + 𝑛ℎ, 𝑛 ∈ N0 , (2) = ℎ𝛼 ∑ ( ) 𝑥 (𝑎 + 𝑘ℎ)
ℎ 𝑛−𝑘
𝑘=0
while the ℎ-difference sum is given by
for 𝑡 = 𝑎 + (𝛼 + 𝑛)ℎ, 𝑛 ∈ N0 . Observe that (𝑎 Δ−𝛼
ℎ 𝑥)(𝑎 + 𝛼ℎ) =
𝑛 ℎ𝛼 𝑥(𝑎) and for 𝛼 = 1 we have again (3).
(𝑎 Δ−1
ℎ 𝑥) (𝑡) := ℎ ∑ 𝑥 (𝑎 + 𝑘ℎ) , (3)
𝑘=0 Remark 6. In [7] one can find the following form of the
fractional ℎ-sum of order 𝛼 > 0:
where 𝑡 = 𝑎 + (𝑛 + 1)ℎ, 𝑛 ∈ N0 and (𝑎 Δ−1
ℎ 𝑥)(𝑎) := 0.
ℎ𝛼 𝑡−𝛼ℎ 𝑡 − 𝜎 (𝑘) (𝛼−1)
(𝑎 Δ−𝛼
ℎ 𝑥) (𝑡) = ∑( ) 𝑥 (𝑘) (8)
Definition 2. For arbitrary 𝛼 ∈ R the ℎ-factorial function is Γ (𝛼) 𝑘=𝑎 ℎ ℎ=1
defined by
that can be useful in implementation.
Γ ((𝑡/ℎ) + 1)
𝑡ℎ(𝛼) := ℎ𝛼 , (4)
Γ ((𝑡/ℎ) + 1 − 𝛼) The following definition can be found in [33] for ℎ = 1 or
in [10] for an arbitrary ℎ > 0.
where Γ is the Euler gamma function, that is, Γ(𝑧) =
+∞
∫0 𝑥𝑧−1 𝑒−𝑥 d𝑥 for all 𝑧 ∈ C such that Re 𝑧 > 0, (𝑡/ℎ) ∉ Z− := Definition 7. Let 𝛼 ∈ (0, 1]. The Caputo ℎ-difference operator
𝛼
{−1, −2, −3, . . .}, and we use the convention that division at a 𝑎 Δ ℎ,∗ 𝑥
of order 𝛼 for a function 𝑥 ∈ F(ℎN)𝑎 is defined by
pole yields zero.
(𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑡) := (𝑎 Δ−(1−𝛼)
ℎ (Δ ℎ 𝑥)) (𝑡) , 𝑡 ∈ (ℎN)𝑎+(1−𝛼)ℎ .
Notice that if we use the general binomial coefficient (9)
( 𝑎𝑏 ) := (Γ(𝑎 + 1))/(Γ(𝑏 + 1)Γ(𝑎 − 𝑏 + 1)), then (4) can be
𝛼
rewritten as Remark 8. Note that 𝑎 Δ ℎ,∗ : F(ℎN)𝑎 → F(ℎN)𝑎+(1−𝛼)ℎ , where
𝑡 𝛼 ∈ (0, 1].
𝑡ℎ(𝛼) 𝛼 ℎ
= ℎ Γ (𝛼 + 1) ( ) . (5) We need the power rule formulas in the sequel. Firstly,
𝛼
we easily notice that for 𝑝 ≠0 the well-defined ℎ-factorial
functions have the following property:
In the sequel we need the following technical properties.
(𝑝) (𝑝−1)
Δ ℎ (𝑡 − 𝑎)ℎ = 𝑝(𝑡 − 𝑎)ℎ . (10)
Proposition 3 (see [12]). Let 𝛼 ∈ R.
𝑗−𝛼−1
More properties of ℎ-factorial functions can be found in [10].
(1) For 𝑗 ∈ N0 one has (−1)𝑗 ( 𝛼𝑗 ) = ( 𝑗 ), where 𝛼, 𝑗 In our consideration the crucial role plays the power rule
are such that both sides are well defined; formula presented in [16], that is,
𝑗−𝛼−1
(2) For 𝑛 ∈ N0 one has ∑𝑛𝑗=0 ( 𝑗 ) = ( 𝑛−𝛼
𝑛 ); Γ (𝜇 + 1) (𝜇+𝛼)
(𝑎 Δ−𝛼
ℎ 𝜓) (𝑡) = (𝑡 − 𝑎 + 𝜇ℎ)ℎ , (11)
(3) For 𝑘 ∈ N1 one has ( 𝛼−1
𝑘 ) + ( 𝛼−1
𝑘−1 ) = ( 𝛼𝑘 ). Γ (𝜇 + 𝛼 + 1)
Abstract and Applied Analysis 3

(𝜇)
where 𝜓(𝑟) = (𝑟 − 𝑎 + 𝜇ℎ)ℎ , 𝑟 ∈ (ℎN)𝑎 , 𝑡 ∈ (ℎN)𝑎+𝛼ℎ . Note Let 𝜇 := 𝑘𝛼 + 𝑠𝛽. For 𝑟 ∈ (ℎN)𝑙ℎ we define the following
(𝜇)
that using the general binomial coefficient one can write (11) ℎ-factorial function 𝜓(𝑟) := (𝑟 + 𝜇ℎ)ℎ . Since
as
1
𝜑𝑘,𝑠 ((𝑛 − 𝑙) ℎ) =
𝑛 + 𝛼 + 𝜇 𝜇+𝛼 Γ (𝑘𝛼 + 𝑠𝛽 + 1)
(𝑎 Δ−𝛼
ℎ 𝜓) (𝑡) = Γ (𝜇 + 1) ( )ℎ . (12)
𝑛
(𝑘𝛼+𝑠𝛽)
⋅ ((𝑛 − 𝑙) ℎ + 𝑘𝛼ℎ + 𝑠𝛽ℎ)ℎ (17)
If 𝜓 ≡ 1, then we have for 𝜇 = 0, 𝑎 = (1−𝛼)ℎ and 𝑡 = 𝑛ℎ+𝑎+𝛼ℎ 1
= 𝜓 (𝑛ℎ − 𝑙ℎ)
Γ (𝜇 + 1)
1
(𝑎 Δ−𝛼
ℎ 1) (𝑡) = (𝑡 − 𝑎)(𝛼)
Γ (𝛼 + 1) ℎ for 𝑛 ≥ 𝑙 and 𝜑𝑘,𝑠 ((𝑚 − 𝑙)ℎ) = 0 for 𝑚 < 𝑙, by (11) we get
(13)
Γ (𝑛 + 𝛼 + 1) 𝑛+𝛼 𝛼 1
= ℎ𝛼 = ( )ℎ . (0 Δ−𝛼 −𝛼
ℎ 𝜑𝑘,𝑠 ) (𝑡) = (𝑙ℎ Δ ℎ 𝜑𝑘,𝑠 ) (𝑡) = ( Δ−𝛼 𝜓) (𝑡)
Γ (𝛼 + 1) Γ (𝑛 + 1) 𝑛 Γ (𝜇 + 1) 0 ℎ

Let us define special functions that we use in the next section 1 Γ (𝜇 + 1) (𝜇+𝛼)
= (𝑡 + 𝜇ℎ)ℎ (18)
to write the formula for solutions. Γ (𝜇 + 1) Γ (𝜇 + 𝛼 + 1)

Definition 9. For 𝛼, 𝛽 > 0 we define 1 (𝜇+𝛼)


= (𝑡 + 𝜇ℎ)ℎ ,
Γ (𝜇 + 𝛼 + 1)

{ 𝑛 + 𝑘𝛼 + 𝑠𝛽 𝑘𝛼+𝑠𝛽 where 𝑡 = 𝑎 − 𝑙ℎ + 𝑛ℎ. Hence


{( )ℎ , for 𝑛 ∈ N0
𝜑𝑘,𝑠 (𝑛ℎ) := { 𝑛
{ 𝑡 + 𝜇ℎ = 𝑛ℎ − (𝑙 + 1) ℎ + (𝑘 + 1) 𝛼ℎ + 𝑠𝛽ℎ,
{0, for 𝑛 < 0,
(14) (0 Δ−𝛼
ℎ 𝜑𝑘,𝑠 ) ((𝑛 − 𝑙) ℎ + 𝑎)
{ 𝑛+𝜇−1 𝜇 Γ (𝑛 + 𝜇)
{(
{ )ℎ = ℎ𝜇 , Γ (𝛼 + 𝑛 − (𝑙 + 1) + 𝜇 + 1)
𝑛 Γ (𝜇) Γ (𝑛 + 1)
𝜑̃𝑘,𝑠 (𝑛ℎ) := { = ℎ𝜇+𝛼
{
{ for 𝑛 ∈ N0 Γ (𝜇 + 𝛼 + 1) Γ (𝑛 − (𝑙 + 1) + 1)
{ 0, for 𝑛 < 0,
Γ (𝛼 + 𝑛 − 𝑙 + 𝜇)
= ℎ𝜇+𝛼
where 𝑛 belongs to the set of integers Z, 𝑘, 𝑠 ∈ N0 and 𝜇 = Γ (𝜇 + 𝛼 + 1) Γ (𝑛 − 𝑙)
𝑘𝛼 + 𝑠𝛽.
𝑛 − 𝑙 − 1 + 𝜇 + 𝛼 𝜇+𝛼
=( )ℎ
Remark 10. It is worthy to notice that for 𝑛 ∈ N0 𝑛−𝑙−1

𝑛 − (𝑙 + 1) + (𝑘 + 1) 𝛼 + 𝑠𝛽 (𝑘+1)𝛼+𝑠𝛽
(a) 𝜑0,0 (𝑛ℎ) = 1; =( )ℎ
𝑛 − (𝑙 + 1)
𝛼 −𝛼
(b) 𝜑1,0 (𝑛ℎ) = ( 𝑛+𝛼
𝑛 ) ℎ = (0 Δ ℎ 1)(𝑛ℎ+𝛼ℎ) and the values = 𝜑𝑘+1,𝑠 ((𝑛 − 𝑙 − 1) ℎ) .
𝜑1,0 ((𝑛 − 1)ℎ) = ( 𝑛−1 ) ℎ𝛼 = (0 Δ−𝛼
𝑛+𝛼−1
ℎ 1)((𝑛 − 1)ℎ + 𝛼ℎ) (19)
are neglected for 𝑛 = 0;

(c) 𝜑𝑘,𝑠 ((𝑛 − 𝑙)ℎ) = (Γ(𝑛 − 𝑙 + 1 + 𝑘𝛼 + 𝑠𝛽))/(Γ(𝑘𝛼 + 𝑠𝛽 +


1)Γ(𝑛 − 𝑙 + 1)) and as the division by pole gives zero, From the application of the power rule follows the rule
the formula works also for 𝑛 < 𝑙, 𝑙 ∈ N0 ; for composing two fractional ℎ-sums. The proof for the case
ℎ = 1 can be found in [7]. For any positive ℎ > 0 the proof is
(d) 𝜑𝑘,𝑠 ((𝑛 − 𝑙)ℎ) = (1/(Γ(𝑘𝛼 + 𝑠𝛽 + 1))) ⋅ presented in [10].
(𝑘𝛼+𝑠𝛽)
((𝑛 − 𝑙)ℎ + 𝑘𝛼ℎ + 𝑠𝛽ℎ)ℎ , 𝑙 ∈ N0 .
Proposition 12. Let 𝑥 be a real valued function defined on
We also need the following proposition. (ℎN)𝑎 , where 𝑎, ℎ ∈ R, ℎ > 0. For 𝛼, 𝛽 > 0 the following
equalities hold:
Proposition 11. Let 𝛼, 𝛽 ∈ (0, 1], ℎ > 0 and 𝑎 = (𝛼−1)ℎ, 𝑏 = −𝛽 −(𝛼+𝛽)
(𝛽 − 1)ℎ. Then for 𝑛 ∈ N𝑙+1 , 𝑙 ∈ N0 (𝑎+𝛽ℎ Δ−𝛼
ℎ (𝑎 Δ ℎ 𝑥)) (𝑡) = (𝑎 Δ ℎ 𝑥) (𝑡)
(20)
−𝛽
(0 Δ−𝛼 = (𝑎+𝛼ℎ Δ ℎ (𝑎 Δ−𝛼
ℎ 𝑥)) (𝑡) ,
ℎ 𝜑𝑘,𝑠 ) ((𝑛 − 𝑙) ℎ + 𝑎) = 𝜑𝑘+1,𝑠 ((𝑛 − 𝑙 − 1) ℎ) , (15)
−𝛽 where 𝑡 ∈ (ℎN)𝑎+(𝛼+𝛽)ℎ .
(0 Δ ℎ 𝜑𝑘,𝑠 ) ((𝑛 − 𝑙) ℎ + 𝑏) = 𝜑𝑘,𝑠+1 ((𝑛 − 𝑙 − 1) ℎ) . (16)
The next proposition gives a useful identity of transform-
Proof. We show only equality (15), as (16) is a symmetric one. ing Caputo fractional difference equations into fractional
4 Abstract and Applied Analysis

𝑛
summations for the case when an order is from the interval 𝛼−1
= ℎ1−𝛼 ∑ (−1)𝑗 ( )
(0, 1]. 𝑗
𝑗=0

Proposition 13 (see [10]). Let 𝛼 ∈ (0, 1], ℎ > 0, 𝑎 = (𝛼 − 𝑥 (𝑎 + (𝑛 − 𝑗) ℎ + ℎ) − 𝑥 (𝑎 + (𝑛 − 𝑗) ℎ)


1)ℎ and let 𝑥 be a real valued function defined on (ℎN)𝑎 . The ⋅

following formula holds:
= ℎ−𝛼 {𝑥 (𝑎 + (𝑛 + 1) ℎ)
(0 Δ−𝛼 𝛼
ℎ (𝑎 Δ ℎ,∗ 𝑥)) (𝑛ℎ + 𝑎) = 𝑥 (𝑛ℎ + 𝑎) − 𝑥 (𝑎) , 𝑛 ∈ N1 .
(21) 𝛼−1 𝛼−1
− [( )+( )] 𝑥 (𝑎 + 𝑛ℎ)
1 0
The operators presented above can be extended to vectors
𝛼−1 𝛼−1
in a componentwise manner. + [( )+( )] 𝑥 (𝑎 + (𝑛 − 1) ℎ)
2 1

3. Solutions of Systems with Sequential 𝛼−1 𝛼−1


− [( )+( )] 𝑥 (𝑎 + (𝑛 − 2) ℎ)
3 2
Fractional Differences
𝛼−1 𝛼−1
Let 𝛼, 𝛽 ∈ (0, 1] and 𝑥 : (ℎN)𝑎 → R𝑁. Moreover, let us take + ⋅ ⋅ ⋅ + (−1)𝑛−1 [( )+( )] 𝑥 (𝑎 + 2ℎ)
𝑛−1 𝑛−2
𝑎 = (𝛼 − 1)ℎ and 𝑏 = (𝛽 − 1)ℎ. Then we define
𝛼−1 𝛼−1
+ (−1)𝑛 [( )+( )]
𝑦 (𝑏 + 𝑛ℎ) := (𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) . (22) 𝑛 𝑛−1

𝛼−1
⋅ 𝑥 (𝑎 + ℎ) − (−1)𝑛 ( ) 𝑥 (𝑎) } .
Note that 𝑦 : (ℎN)𝑏 → R𝑁. Then we apply the next difference 𝑛
operator of order 𝛽 on the new function 𝑦 and consider here (27)
an initial value problem stated by the system

(𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) = 𝑦 (𝑏 + 𝑛ℎ) , (23) Since by Proposition 3 the following relation ( 𝛼−1 𝛼−1
𝑘 ) + ( 𝑘−1 ) =
𝛼
( 𝑘 ) holds, one gets
𝛽
(𝑏 Δ ℎ,∗ 𝑦) (𝑛ℎ) = 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) , (24)

𝛼
where 𝑓 : (ℎN)0 × R𝑁 → R𝑁, with initial values 𝑦 (𝑏 + 𝑛ℎ) = ℎ−𝛼 [𝑥 (𝑎 + (𝑛 + 1) ℎ) − ( ) 𝑥 (𝑎 + 𝑛ℎ)
1

(𝑎 Δ𝛼ℎ,∗ 𝑥) (0) = 𝑥0 , 𝛼
(25) + ( ) 𝑥 (𝑎 + (𝑛 − 1) ℎ)
2
𝑥 (𝑎) = 𝑥𝑎 , (26)
𝛼
− ( ) 𝑥 (𝑎 + (𝑛 − 2) ℎ)
3
where 𝑥𝑎 , 𝑥0 are constant vectors from R𝑁.
Solutions of the state equations (23) and (24) of the 𝛼
+ ( ) 𝑥 (𝑎 + (𝑛 − 3) ℎ)
sequential fractional discrete-time system can be computed 4
in the recursive way. From Definition 7 and by Proposition 3,
𝛼
(23) can be written as + ⋅ ⋅ ⋅ + (−1)𝑛−1 ( ) 𝑥 (𝑎 + 2ℎ)
𝑛−1
(28)
𝑦 (𝑏 + 𝑛ℎ) 𝛼
+ (−1)𝑛 ( ) 𝑥 (𝑎 + ℎ)
𝑛
= (𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) = (𝑎 Δ−(1−𝛼)
ℎ (Δ ℎ 𝑥)) (𝑛ℎ)
𝛼−1
−(−1)𝑛 ( ) 𝑥 (𝑎)]
ℎ 𝑛 𝑛
= ∑ (𝑛ℎ − 𝑎 − 𝑘ℎ − ℎ)(−𝛼)
ℎ (Δ ℎ 𝑥) (𝑎 + 𝑘ℎ)
Γ (1 − 𝛼) 𝑘=0
𝑛
𝛼
𝑛 = ℎ−𝛼 [∑ (−1)𝑗 ( ) 𝑥 (𝑎 + (𝑛 − 𝑗 + 1) ℎ)
ℎ Γ (𝑛 − 𝑘 − 𝛼 + 1) 𝑗
= ∑ ℎ−𝛼 (Δ ℎ 𝑥) (𝑎 + 𝑘ℎ) [𝑗=0
Γ (1 − 𝛼) 𝑘=0 Γ (𝑛 − 𝑘 + 1)
𝛼−1
𝑛
𝑛−𝑘−𝛼 +(−1)𝑛+1 ( ) 𝑥 (𝑎) ] .
= ℎ1−𝛼 ∑ ( ) (Δ ℎ 𝑥) (𝑎 + 𝑘ℎ) 𝑛
𝑛−𝑘 ]
𝑘=0
Abstract and Applied Analysis 5

𝑛
Repeating the same computation for (24) one gets 𝛼+𝛽
= ∑ (−1)𝑘 ( ) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ)
𝑘
𝛽 −(1−𝛽) 𝑘=0
𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) = (𝑏 Δ ℎ,∗ 𝑦) (𝑛ℎ) = (𝑏 Δ ℎ (Δ ℎ 𝑦)) (𝑛ℎ)
𝛼+𝛽 𝛽 𝛽−1
𝑛 + (−1)𝑛+1 [( )−( )+( )] 𝑥 (𝑎 + ℎ)
−𝛽 𝛽 𝑖 𝑛+1 𝑛+1 𝑛
= ℎ ∑(−1) ( ) 𝑦 (𝑏 + (𝑛 − 𝑖 + 1) ℎ)
𝑖
𝑖=0 𝑛
𝛼−1 𝛽−1
+ [∑(−1)𝑛−𝑖 ( ) + (−1)𝑛 ( )] 𝑥 (𝑎) .
𝛽−1 𝑛−𝑖+1 𝑛
+ ℎ−𝛽 (−1)𝑛+1 ( ) 𝑦 (𝑏) . 𝑖=0
𝑛
(31)
(29)

Hence, Then by Proposition 3, we have

𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ))


𝑛
𝛽 𝑛
𝛼+𝛽
= ℎ−𝛽 ∑(−1)𝑖 ( ) = ∑ (−1)𝑘 ( ) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ)
𝑖 𝑘
𝑖=0
𝑘=0

𝑛−𝑖+1
𝛼 𝛼+𝛽 𝛽 𝛽−1
⋅ [ℎ−𝛼 ∑ (−1)𝑗 ( ) + (−1)𝑛+1 [( )−( ) +( )] 𝑥 (𝑎 + ℎ)
𝑗 𝑛+1 𝑛+1 𝑛
𝑗=0
[
𝛼−2 𝛽−1
⋅ 𝑥 (𝑎 + (𝑛 − 𝑖 − 𝑗 + 2) ℎ) + [1 + (−1)𝑛 ( ) + (−1)𝑛 ⋅ ( )] 𝑥 (𝑎) .
𝑛+1 𝑛
(32)
𝛼−1
−𝛼
+ℎ (−1) 𝑛−𝑖+2
( ) 𝑥 (𝑎) ]
𝑛−𝑖+1
] Consequently, since ℎ𝛼 𝑥0 = 𝑥(𝑎 + ℎ) − 𝑥(𝑎), 𝑥(𝑎 + 2ℎ) =
𝛽−1 ℎ𝛼+𝛽 ⋅ 𝑓(0, 𝑥(𝑎)) − 𝛼𝑥(𝑎) + (1 + 𝛼) ⋅ 𝑥(𝑎 + ℎ) and for 𝑛 ≥ 1
+ ℎ−𝛽 (−1)𝑛+1 ( ) ⋅ 𝑥0
𝑛
𝑥 (𝑎 + (𝑛 + 2) ℎ)
𝑛 𝑛−𝑖+1
−𝛼−𝛽 [∑ ∑ (−1) 𝑖+𝑗 𝛽 𝛼
=ℎ ( )( ) = ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ))
𝑖 𝑗
[𝑖=0 𝑗=0
𝑛
𝛼+𝛽
⋅ 𝑥 (𝑎 + (𝑛 − 𝑖 − 𝑗 + 2) ℎ) + ∑ (−1)𝑘+1 ( ) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ)
𝑘
𝑘=1
𝑛
𝑛−𝑖+2 𝛼−1
+ ∑(−1) ( ) 𝑥 (𝑎) 𝛼+𝛽 𝛽 𝛽−1
𝑛−𝑖+1 + (−1)𝑛 [( )−( )+( )] 𝑥 (𝑎 + ℎ)
𝑖=0 𝑛+1 𝑛+1 𝑛

𝛽−1 𝛼−2 𝛽−1


+ (−1)𝑛+1 ( ) ⋅ [𝑥 (𝑎 + ℎ) − 𝑥 (𝑎)]] . − [1 + (−1)𝑛 ( ) + (−1)𝑛 ( )] 𝑥 (𝑎) .
𝑛 𝑛+1 𝑛
] (33)
(30)

Then using the Chu-Vandermonde identity, that is, Using Proposition 3 we get for 𝑛 ≥ 1
∑𝑘𝑖=0 ( 𝛽𝑖 ) ( 𝑘−𝑖
𝛼
) = ( 𝛼+𝛽
𝑘
), 𝛼, 𝛽 ∈ R, 𝑘 ∈ Z, one gets
𝑥 (𝑎 + (𝑛 + 2) ℎ)
𝛼+𝛽
ℎ ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ))
= ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ))
𝑛+1 𝑘
𝛽 𝛼 𝑛
= ∑ (−1)𝑘 ∑ ( ) ( ) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ) 𝛼+𝛽
𝑖 𝑘−𝑖 + ∑ (−1)𝑘+1 ( ) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ)
𝑘=0 𝑖=0 𝑘
𝑘=1
𝑛
𝛼−1 𝛽−1 𝑛−𝛽 𝑛−𝛽 𝑛−𝛼−𝛽
+ [∑(−1)𝑛−𝑖 ( ) + (−1)𝑛 ( )] 𝑥 (𝑎) + [( )+( )−( )] 𝑥 (𝑎 + ℎ)
𝑛−𝑖+1 𝑛 𝑛 𝑛+1 𝑛+1
𝑖=0

𝛽−1 𝑛−𝛼+2 𝑛−𝛽


+ (−1)𝑛+1 ( ) 𝑥 (𝑎 + ℎ) − [1 − ( )+( )] 𝑥 (𝑎)
𝑛 𝑛+1 𝑛
6 Abstract and Applied Analysis

= ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) with initial conditions (25) and (26), that is, (𝑎 Δ𝛼ℎ,∗ 𝑥)(0) =
𝑛 𝑥0 and 𝑥(𝑎) = 𝑥𝑎 , 𝑥0 , 𝑥𝑎 ∈ R𝑁, is given by the following
𝛼+𝛽
+ ∑ (−1)𝑘+1 ( ) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ) formula:
𝑘
𝑘=1 𝑛

𝑛−𝛽+1 𝑛−𝛼−𝛽 𝑥 (𝑎 + 𝑛ℎ) = ∑ 𝐴𝑘 𝜑𝑘,𝑘 ((𝑛 − 2𝑘) ℎ) 𝑥𝑎


+ [( )−( )] 𝑥 (𝑎 + ℎ) 𝑘=0
𝑛+1 𝑛+1 (41)
𝑛
𝑘
𝑛−𝛼+2 𝑛−𝛽 + ∑ 𝐴 𝜑𝑘+1,𝑘 ((𝑛 − (2𝑘 + 1)) ℎ) 𝑥0 ,
− [1 − ( )+( )] 𝑥 (𝑎) .
𝑛+1 𝑛 𝑘=0

(34) for 𝑛 ∈ N0 .
Therefore the solution of (23) and (24) is given recursively
Proof. Notice that for 𝑛 = 0 we get 𝑥(𝑎+0⋅ℎ) = 𝐴0 (𝜑0,0 (0)𝑥𝑎 +
by the following formula:
𝜑1,0 (−ℎ)𝑥0 ). Since 𝜑0,0 (0) = 1 and 𝜑1,0 (−ℎ) = 0, we get 𝑥(𝑎 +
𝑥 (𝑎 + ℎ) = ℎ𝛼 𝑥0 + 𝑥 (𝑎) , 0 ⋅ ℎ) = 𝑥𝑎 .
For 𝑛 > 0 let us define the sequence {𝑥𝑚 }𝑚≥1 in the
𝑥 (𝑎 + 2ℎ) = ℎ𝛼+𝛽 ⋅ 𝑓 (0, 𝑥 (𝑎)) − 𝛼𝑥 (𝑎) following way:

+ (1 + 𝛼) ⋅ 𝑥 (𝑎 + ℎ) , 𝑥𝑚+1 (𝑎 + 𝑛ℎ) = 𝑥𝑎 𝜑0,0 (𝑛ℎ) + 𝑥0 𝜑1,0 ((𝑛 − 1) ℎ)


(42)
𝑥 (𝑎 + (𝑛 + 2) ℎ) + (0 Δ−𝛼
ℎ 𝑔𝑚 ) (𝑎 + 𝑛ℎ) , 𝑚 ∈ N0 ,

where 𝑔𝑚 (𝑛ℎ) = (0 Δ ℎ 𝑓̃𝑚 )(𝑏 + 𝑛ℎ) and 𝑓̃𝑚 (𝑛ℎ) = 𝐴𝑥𝑚 (𝑎 + 𝑛ℎ)
−𝛽
= ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ))
(35) with 𝑥0 (𝑎 + 𝑛ℎ) = 𝑥𝑎 .
𝑛
+ ∑ (−1)𝑘+1 (
𝛼+𝛽
) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ) We calculate the first step. As 𝑓̃0 (𝑛ℎ) = 𝐴𝑥0 (𝑎 + 𝑛ℎ) =
𝑘 𝐴𝑥𝑎 , then 𝑔0 (𝑛ℎ) = 𝐴𝑥𝑎 (0 Δ−𝛽 1)(𝑏 + 𝑛ℎ) = 𝐴𝑥𝑎 𝜑0,1 ((𝑛 − 1)ℎ).
𝑘=1
Going further,
𝑛−𝛽+1 𝑛−𝛼−𝛽
+ [( )−( )] 𝑥 (𝑎 + ℎ)
𝑛+1 𝑛+1 𝑥1 (𝑎 + 𝑛ℎ) = 𝑥𝑎 𝜑0,0 (𝑛ℎ) + 𝑥0 𝜑1,0 ((𝑛 − 1) ℎ)
(43)
𝑛−𝛼+2 𝑛−𝛽 + (0 Δ−𝛼
ℎ 𝑔0 ) (𝑎 + 𝑛ℎ) ,
+ [( )−1−( )] 𝑥 (𝑎) .
𝑛+1 𝑛
which could be written as
Another possibility of computing the solution of (23) and
(24) is to use Proposition 13 twice and then, for 𝑛 ≥ 1, we get: 𝑥1 (𝑎 + 𝑛ℎ) = 𝑥𝑎 𝜑0,0 (𝑛ℎ) + 𝑥0 𝜑1,0 ((𝑛 − 1) ℎ)
(44)
−𝛽 𝛽 + 𝐴𝑥𝑎 𝜑1,1 ((𝑛 − 2) ℎ)
(0 Δ ℎ (𝑏 Δ ℎ,∗ 𝑦)) (𝑏 + 𝑛ℎ) = 𝑦 (𝑏 + 𝑛ℎ) − 𝑦 (𝑏)
and, using Proposition 11, we get
= (𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) − 𝑥0 , (36)
𝑥2 (𝑎 + 𝑛ℎ) = 𝑥𝑎 𝜑0,0 (𝑛ℎ) + 𝑥0 𝜑1,0 ((𝑛 − 1) ℎ)
(0 Δ−𝛼 𝛼
ℎ ( 𝑎 Δ ℎ,∗ 𝑥)) (𝑎 + 𝑛ℎ) = 𝑥 (𝑎 + 𝑛ℎ) − 𝑥𝑎 .
+ 𝐴𝑥𝑎 𝜑1,1 ((𝑛 − 2) ℎ)
Hence (45)
+ 𝐴𝑥0 𝜑2,1 ((𝑛 − 3) ℎ)
−𝛽 ̃
(𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) = 𝑥0 + (0 Δ ℎ 𝑓) (𝑏 + 𝑛ℎ) , (37)
+ 𝐴2 𝑥𝑎 𝜑2,2 ((𝑛 − 4) ℎ) .
̃
where 𝑓(𝑛ℎ) := 𝑓(𝑛ℎ, 𝑥(𝑎 + 𝑛ℎ)). Nextly,
Taking 𝑚 tending to +∞ we get formula (41) as the
𝑥 (𝑛ℎ + 𝑎) = 𝑥𝑎 + 𝑥0 (0 Δ−𝛼 −𝛼
ℎ 1) (𝑎 + 𝑛ℎ) + (0 Δ ℎ 𝑔) (𝑎 + 𝑛ℎ) ,
solution of (39) and (40) with initial conditions (25) and
(38) (26).

̃ + 𝑛ℎ).
where 𝑔(𝑛ℎ) = (0 Δ ℎ 𝑓)(𝑏
−𝛽
3.1. Semilinear Sequential Systems. Firstly we state a technical
Firstly we prove the formula for the unique solution in lemma and notations.
linear case of (23) and (24): 𝑓(𝑛ℎ, 𝑥(𝑛ℎ + 𝑎)) = 𝐴𝑥(𝑎 + 𝑛ℎ),
where 𝐴 is a constant square matrix of degree 𝑛. Lemma 15. Let 𝛾 : (ℎN)0 → R and 𝛼 > 0. Let
(0 Δ−𝑘𝛼
ℎ 𝛾)(𝑘𝛼ℎ + 𝑛ℎ) = 𝛾1 (𝑘𝛼ℎ + 𝑛ℎ) and 𝛾 ̃1 (𝑛ℎ) := 𝛾1 (𝑘𝛼ℎ +
Theorem 14. The solution to the system 𝑛ℎ) for 𝑘 ∈ N1 . Then for 𝑘 ∈ N1 one gets
(𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) = 𝑦 (𝑏 + 𝑛ℎ) , (39) (0 Δ−𝛼 ̃1 ) (𝑡) = (0 Δ−(𝑘+1)𝛼
ℎ 𝛾 ℎ 𝛾) (𝑘𝛼ℎ + 𝑡) , (46)
𝛽
(𝑏 Δ ℎ,∗ 𝑦) (𝑛ℎ) = 𝐴𝑥 (𝑎 + 𝑛ℎ) (40) where 𝑡 = 𝛼ℎ + 𝑛ℎ.
Abstract and Applied Analysis 7

Proof. First let us consider the case 𝑘 = 1. Then from Using the power rule formula for 𝜇 = 0 and by Remark 10 we
Proposition 12 we can write can write the recursive formula for the solution to nonlinear
problem given by (23) and (24) and conditions (25) and (26):
(𝛼ℎ Δ−𝛼 −𝛼 −2𝛼
ℎ (0 Δ ℎ 𝛾)) (𝑡) = (0 Δ ℎ 𝛾) (𝑡) , (47)
where 𝑡 = 2𝛼ℎ + 𝑛ℎ, 𝑛 ∈ N0 . 𝑥 (𝑎 + 𝑛ℎ) = 𝑥𝑎 + 𝑥0 𝜑1,0 ((𝑛 − 1) ℎ)
Let 𝛾1 (𝛼ℎ + 𝑛ℎ) = (0 Δ−𝛼 ̃1 (𝑛ℎ) := 𝛾1 (𝛼ℎ +
ℎ 𝛾)(𝛼ℎ + 𝑛ℎ) and 𝛾 𝑛−1
𝑛ℎ). Then
+ ∑ 𝜑̃1,1 (𝑛ℎ − ℎ − 𝜎 (𝑟ℎ)) 𝑓 (𝑟ℎ, 𝑥 (𝑎 + 𝑟ℎ)) .
(0 Δ−𝛼
ℎ 𝛾̃1 ) (𝛼ℎ + 𝑛ℎ) 𝑟=0
(53)
ℎ 𝑛
= ∑(𝑛ℎ + 𝛼ℎ − 𝜎 (𝑟ℎ))(𝛼−1) 𝛾̃1 (𝑟ℎ)
Γ (𝛼) 𝑟=0 ℎ The given formula (53) also works for 𝑛 = 0 as 𝜑̃1,1 (−2ℎ) = 0.
Then 𝑥(𝑎 + 0ℎ) = 𝑥𝑎 . We can check the next steps:
ℎ 𝑛+𝛼 (48)
= ∑ (𝑛ℎ + 2𝛼ℎ − 𝜎 (𝑠ℎ))(𝛼−1)
ℎ 𝛾1 (𝑠ℎ)
Γ (𝛼) 𝑠=𝛼 𝑥 (𝑎 + ℎ) = 𝑥𝑎 + 𝑥0 𝜑1,0 (0) + 𝜑̃1,1 (−ℎ) 𝑓 (0, 𝑥 (𝑎))

= (𝛼ℎ Δ−𝛼
ℎ 𝛾1 ) (2𝛼ℎ + 𝑛ℎ) = 𝑥𝑎 + 𝑥0 ℎ𝛼 ,

= (0 Δ−2𝛼
ℎ 𝛾) (2𝛼ℎ + 𝑛ℎ) .
𝑥 (𝑎 + 2ℎ) = 𝑥𝑎 + 𝑥0 𝜑1,0 (ℎ) + 𝜑̃1,1 (0ℎ) 𝑓 (0, 𝑥 (𝑎)) (54)

Equation (46) for 𝑘 > 1 follows inductively. + 𝜑̃1,1 (−ℎ) 𝑓 (ℎ, 𝑥 (𝑎 + ℎ))

Note that = 𝑥𝑎 + 𝑥0 ℎ𝛼 (1 + 𝛼) + ℎ𝛼+𝛽 𝑓 (0, 𝑥 (𝑎)) .


(0 Δ−𝑘𝛼
ℎ 𝛾) (𝑘𝛼ℎ + 𝑛ℎ)
For special semilinear case when 𝑓(𝑛ℎ, 𝑥(𝑛ℎ + 𝑎)) = 𝐴𝑥(𝑛ℎ +
ℎ 𝑛 (49) 𝑎) + 𝛾(𝑛ℎ) we have 𝑓(0, 𝑥(𝑎)) = 𝐴𝑥(𝑎) + 𝛾(0). Then
= ∑(𝑛ℎ + 𝑘𝛼ℎ − 𝜎 (𝑟ℎ))(𝑘𝛼−1)
ℎ 𝛾 (𝑟ℎ) .
Γ (𝑘𝛼) 𝑟=0
Similar to the procedure presented in the proof of Lemma 15 𝑥 (𝑎 + 2ℎ) = (𝐼 + ℎ𝛼+𝛽 𝐴) 𝑥𝑎 + (1 + 𝛼) ℎ𝛼 𝑥0 + ℎ𝛼+𝛽 𝛾 (0) .
we can prove that for 𝑘, 𝑠 ∈ N0 and 𝛼 > 0, 𝛽 > 0: (55)
−𝑘𝛼−𝑠𝛽
(0 Δ ℎ 𝛾) (𝑘𝛼ℎ + 𝑠𝛽ℎ + 𝑛ℎ) Theorem 16. The solution to the system
𝑛
ℎ (𝑘𝛼+𝑠𝛽−1)
= ∑((𝑛 + 𝑘𝛼 + 𝑠𝛽 − 𝑟 − 1) ℎ)ℎ 𝛾 (𝑟ℎ) (𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) = 𝑦 (𝑏 + 𝑛ℎ) , (56)
Γ (𝑘𝛼 + 𝑠𝛽) 𝑟=0
𝛽
𝑘𝛼+𝑠𝛽
𝑛
Γ (𝑛 − 𝑟 + 𝑘𝛼 + 𝑠𝛽) (𝑏 Δ ℎ,∗ 𝑦) (𝑛ℎ) = 𝐴𝑥 (𝑎 + 𝑛ℎ) + 𝛾 (𝑛ℎ) (57)
=ℎ ∑ 𝛾 (𝑟ℎ)
𝑟=0 Γ (𝑘𝛼 + 𝑠𝛽) Γ (𝑛 − 𝑟 + 1)
with initial conditions (25) and (26), that is, (𝑎 Δ𝛼ℎ,∗ 𝑥)(0) = 𝑥0
𝑛
𝑛 − 𝑟 + 𝑘𝛼 + 𝑠𝛽 − 1 𝑘𝛼+𝑠𝛽 and 𝑥(𝑎) = 𝑥𝑎 , 𝑥0 , 𝑥𝑎 ∈ R𝑁 is given by
= ∑( )ℎ 𝛾 (𝑟ℎ) .
𝑟=0
𝑛−𝑟
(50) 𝑥 (𝑎 + 𝑛ℎ)
Taking 𝜇 = 𝑘𝛼 + 𝑠𝛽 and using formula (38) we can write 𝑛
(50) shortly in the following way: = ∑ 𝐴𝑘 𝜑𝑘,𝑘 ((𝑛 − 2𝑘) ℎ) 𝑥𝑎
𝑛 𝑘=0
−𝜇
(0 Δ ℎ 𝛾) (𝜇ℎ + 𝑛ℎ) = ∑𝜑̃𝑘,𝑠 (𝑛ℎ − 𝑟ℎ) 𝛾 (𝑟ℎ) . (51) 𝑛
𝑟=0 + ∑ 𝐴𝑘 𝜑𝑘+1,𝑘 ((𝑛 − (2𝑘 + 1)) ℎ) 𝑥0
𝑘=0
Moreover, we can also write direct formula for values
(0 Δ−𝛼
ℎ 𝑔)(𝑛ℎ + 𝑎) given in (38) for nonlinear problem. In fact 𝑛−1 𝑛
using Definition 4 of fractional summation, formula (38) of + ∑ ( ∑ 𝐴𝑘 𝜑̃𝑘+1,𝑘+1 ((𝑛 − 1) ℎ − 𝜎 (𝑟ℎ))) 𝛾 (𝑟ℎ) ,
functions 𝜑̃𝑘,𝑠 and Proposition 12 we write (51) as follows: 𝑟=0 𝑘=0
(58)
𝑥 (𝑎 + 𝑛ℎ) = 𝑥𝑎 + 𝑥0 (0 Δ−𝛼
ℎ 1) (𝑎 + 𝑛ℎ)

𝑛−1 for 𝑛 ∈ N0 .
+ ∑ 𝜑̃1,1 (𝑛ℎ − ℎ − 𝜎 (𝑟ℎ)) 𝑓 (𝑟ℎ, 𝑥 (𝑎 + 𝑟ℎ)) .
𝑟=0 Proof. Note that 𝜑0,0 (0) = 1, 𝜑1,0 (−ℎ) = 0 and 𝜑𝑘+1,𝑘+1 (−ℎ)
(52) = 0 for 𝑘 ≥ 0, so we get 𝑥(𝑎 + 0ℎ) = 𝑥𝑎 .
8 Abstract and Applied Analysis

For 𝑛 > 0 based on the proof for linear case we can write Moreover,
the solution formula as follows:
𝑛 𝑥𝑖 (𝑎 + ℎ) ≥ 𝑥𝑎𝑖 = 𝑥𝑖 (𝑎) , (63)
𝑘
𝑥 (𝑎 + 𝑛ℎ) = ∑ 𝐴 𝜑𝑘,𝑘 ((𝑛 − 2𝑘) ℎ) 𝑥𝑎
𝑘=0 where 𝑥(𝑎 + ℎ) = [𝑥1 (𝑎 + ℎ), . . . , 𝑥𝑁(𝑎 + ℎ)]𝑇 and 𝑥(𝑎) =
𝑛 [𝑥1 (𝑎), . . . , 𝑥𝑁(𝑎)]𝑇 = [𝑥𝑎1 , . . . , 𝑥𝑎𝑁]𝑇 . Let 𝑥𝑖 (𝑎 + 𝑛ℎ) and
+ ∑ 𝐴𝑘 𝜑𝑘+1,𝑘 ((𝑛 − (2𝑘 + 1)) ℎ) 𝑥0 (59) 𝑓𝑖 (𝑛ℎ, 𝑥(𝑎 + 𝑛ℎ)) denote the 𝑖th coordinates of the vectors
𝑘=0 𝑥(𝑎 + 𝑛ℎ) and 𝑓(𝑛ℎ, 𝑥(𝑎 + 𝑛ℎ)), respectively. Then since 𝑥𝑖 (𝑎 +
𝑛
ℎ) ≥ 𝑥𝑖 (𝑎) ≥ 0 for 𝑖 = 1, . . . , 𝑁,
+ ∑ 𝐴𝑘 (0 Δ−𝜏
ℎ 𝛾) ((𝑛 − 1 + 𝜏) ℎ) ,
𝑘=0 𝑥𝑖 (𝑎 + 2ℎ) = ℎ𝛼+𝛽 ⋅ 𝑓𝑖 (0, 𝑥 (𝑎))
where 𝜏 = (𝑘 + 1)(𝛼 + 𝛽). Then taking into account formulas − 𝛼𝑥𝑖 (𝑎) + (1 + 𝛼) ⋅ 𝑥𝑖 (𝑎 + ℎ)
(50) and (51) we get the form (58) as the solution of (56) and
(57) with initial conditions (25) and (26). ≥ ℎ𝛼+𝛽 ⋅ 𝑓𝑖 (0, 𝑥 (𝑎)) − 𝛼𝑥𝑖 (𝑎) + (1 + 𝛼) ⋅ 𝑥𝑖 (𝑎)

= ℎ𝛼+𝛽 ⋅ 𝑓𝑖 (0, 𝑥 (𝑎)) + 𝑥𝑖 (𝑎) .


4. Positivity
(64)
Let R𝑁×𝑀
+ be the set of real 𝑁 × 𝑀 matrices with the
nonnegative entries and R𝑁 𝑁×1 Note that for 0 < 𝛼 + 𝛽 ≤ 1 we have −(1/8) ≤ ( 𝛼+𝛽 ) ≤ 0, so
+ = R+ . 2
𝛼+𝛽
Based on [8, 31, 32] we consider the following definitions. ( 2 ) + 1 ≥ 0. Then since 𝑥𝑖 (𝑎) ≥ 0 and (61) holds for 𝑛 = 0,
we get
Definition 17. The fractional system (23) and (24) is called
positive fractional system if and only if 𝑥(𝑎 + 𝑛ℎ) ∈ R𝑁
+ for 𝛼+𝛽
𝑥𝑖 (𝑎 + 2ℎ) ≥ [( ) + 1] 𝑥𝑖 (𝑎) ≥ 0. (65)
any initial conditions 𝑥𝑎 , 𝑥0 ∈ R𝑁
+ . 2

Similarly as in [32] we will use the recursive formula (35) For 𝑛 = 1 we have
to show the positivity of the considered systems. Observe that
the systems considered in this paper are of the sequential 𝛼+𝛽
𝑥𝑖 (𝑎 + 3ℎ) = ℎ𝛼+𝛽 ⋅ 𝑓𝑖 (ℎ, 𝑥 (𝑎 + ℎ)) − ( )
type while in [32] the sequential systems are not studied. 2
Moreover, the Grünwald-Letnikov operator with the step
2−𝛽
equal to one is used in [32], whereas we study the systems with ⋅ 𝑥𝑖 (𝑎 + ℎ) + (𝛼 + 𝛽) ⋅ 𝑥𝑖 (𝑎 + 2ℎ) + ( )
the ℎ-differences of Caputo type. So in our case the steps are 2
equal to ℎ. 3−𝛼
From [32] we have the following lemma. ⋅ 𝑥 (𝑎 + ℎ) + [( ) − 1 − (1 − 𝛽)] 𝑥𝑖 (𝑎) .
2
(66)
Lemma 18. If 0 < 𝛼 < 1, then (−1)𝑖+1 ( 𝛼𝑖 ) > 0, 𝑖 = 1, 2, 3, . . ..

Moreover, for 𝑘 ≥ 3 we have ( 𝑘1 ) = 0. Since ( 2−𝛽


2
) > 0, 𝑥𝑖 (𝑎 + ℎ) ≥ 𝑥𝑖 (𝑎) ≥ 0 for 𝑖 = 1, . . . , 𝑁, and
Using the properties of Euler gamma function one can (61) holds for 𝑛 = 1, we have
show that if 0 < 𝛼 + 𝛽 ≤ 1, then for 𝑛 ∈ N1 , one has the
following inequality: 2−𝛽 3−𝛼
𝑥𝑖 (𝑎 + 3ℎ) ≥ [( )+( ) + 𝛽] 𝑥𝑖 (𝑎) ≥ 0 (67)
2 2
𝑛−𝛽 𝑛−𝛼+2 𝑛−𝛼−𝛽 𝑛+1
( )+( )≥( )+( ). (60)
𝑛+1 𝑛+1 𝑛+1 𝑛+1 for 0 < 𝛼, 𝛽 ≤ 1.

Proposition 19. Let 0 < 𝛼 + 𝛽 ≤ 1. If 𝑥0 , 𝑥𝑎 ∈ R𝑁


+ and for all Now for 𝑛 = 2 we get
𝑛 ≥ 0, 𝑥 ∈ R𝑁
+
𝑥𝑖 (𝑎 + 4ℎ) = ℎ𝛼+𝛽 ⋅ 𝑓𝑖 (2ℎ, 𝑥 (𝑎 + 2ℎ))
𝛼+𝛽 𝛼+𝛽
ℎ 𝑓 (𝑛ℎ, 𝑥) − ( ) 𝑥 ∈ R𝑁
+, (61)
2 𝛼+𝛽
−( ) 𝑥𝑖 (𝑎 + 2ℎ) + (𝛼 + 𝛽) ⋅ 𝑥𝑖 (𝑎 + 3ℎ)
2
then 𝑥(𝑎 + 𝑛ℎ) ∈ R𝑁
+ for all 𝑛 ≥ 1.
3−𝛽 2−𝛼−𝛽
+ [( )−( )] 𝑥 (𝑎 + ℎ)
Proof. The proof is by the induction principle. Assume that 3 3
0 < 𝛼 + 𝛽 ≤ 1 and both 𝛼 > 0 and 𝛽 > 0. Since 𝑥0 , 𝑥𝑎 ∈ R𝑁
+
and ℎ > 0, by (35) we have 4−𝛼 2−𝛽
+ [( )−1−( )] 𝑥𝑖 (𝑎) .
3 2
𝑥 (𝑎 + ℎ) = ℎ𝛼 𝑥0 + 𝑥𝑎 ∈ R𝑁
+. (62) (68)
Abstract and Applied Analysis 9

Since ( 3−𝛽
3
) − ( 2−𝛼−𝛽
3
) > 0 for 0 < 𝛼, 𝛽 ≤ 1, 𝑥𝑖 (𝑎 + ℎ) ≥
𝑥𝑖 (𝑎) ≥ 0 for 𝑖 = 1, . . . , 𝑁, and (61) holds for 𝑛 = 2, using (60)
we have 0.4

3−𝛽 4−𝛼
𝑥𝑖 (𝑎 + 4ℎ) ≥ [( )+( )
3 3
(69) 0.3
2−𝛼−𝛽 3
−( ) − ( )] 𝑥𝑖 (𝑎) ≥ 0
3 3
for 0 < 𝛼, 𝛽 ≤ 1.
0.2
Assume that 𝑥(𝑎 + 𝑘ℎ) ∈ R𝑁 + for 𝑘 = 1, 2, 3, . . . , 𝑛 + 1.
Using the properties of gamma function one can show that
( 𝑛−𝛽+1
𝑛+1
) ≥ ( 𝑛−𝛼−𝛽
𝑛+1
) for 𝑛 ≥ 3. Then applying 𝑥(𝑎 + ℎ) ≥ 𝑥(𝑎)
to (35) we get for 𝑛 ≥ 3 0.1

𝑥𝑖 (𝑎 + (𝑛 + 2) ℎ)

𝛼+𝛽
= ℎ𝛼+𝛽 ⋅ 𝑓𝑖 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) − ( ) 𝑥𝑖 (𝑎 + 𝑛ℎ)
2 0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16

+ (𝛼 + 𝛽) 𝑥𝑖 (𝑎 + (𝑛 + 1) ℎ) Figure 1: Consider the following system (𝑎 Δ0.4 ℎ,∗ 𝑥)(𝑛ℎ) = 𝑦(𝑏 + 𝑛ℎ),
(𝑏 Δ0.5
ℎ,∗ 𝑦)(𝑛ℎ) = 𝐴𝑥(𝑎 + 𝑛ℎ), where 𝐴 is given in Example 22, with
𝑛
𝛼+𝛽 initial conditions: 𝑥0 = (0, 1), 𝑥𝑎 = (0, 0.1), ℎ0.9 𝐴 − ( 0.9 2×2
2 ) 𝐼2 ∈ R+ .
+ ∑ (−1)𝑘+1 ( ) 𝑥𝑖 (𝑎 + (𝑛 − 𝑘 + 2) ℎ)
𝑘
𝑘=3

+ [(
𝑛−𝛽+1
)−(
𝑛−𝛼−𝛽
)] 𝑥𝑖 (𝑎 + ℎ) using the induction principle we get that 𝑥(𝑎 + 𝑛ℎ) ∈ R𝑁
+ for
𝑛+1 𝑛+1 all 𝑛 ≥ 1.
𝑛−𝛼+2 𝑛−𝛽 Corollary 20. Let 0 < 𝛼 + 𝛽 ≤ 1 and 𝐼𝑁 denote the identity
+ [( )−1−( )] 𝑥𝑖 (𝑎)
𝑛+1 𝑛 matrix. If 𝑥0 , 𝑥𝑎 ∈ R𝑁 𝑁
+ and for all 𝑛 ≥ 1, 𝑥 ∈ R+

𝛼+𝛽
≥ ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) − ( ) 𝑥𝑖 (𝑎 + 𝑛ℎ) ℎ𝛼+𝛽 𝐴 − (
𝛼+𝛽
) 𝐼𝑁 ∈ R𝑁×𝑁 ,
2 2 + (71)

+ (𝛼 + 𝛽) 𝑥 (𝑎 + (𝑛 + 1) ℎ)
then 𝑥(𝑎 + 𝑛ℎ) ∈ R𝑁
+ for all 𝑛 ≥ 1 that is, (39) and (40) is
𝑛 positive.
𝛼+𝛽
+ ∑ (−1)𝑘+1 ( ) 𝑥𝑖 (𝑎 + (𝑛 − 𝑘 + 2) ℎ) (70)
𝑘
𝑘=3 Remark 21. In [32] the sufficient condition concerning the
positivity of the linear discrete systems with Grünwald-
𝑛−𝛽+1 𝑛−𝛼−𝛽
+ [(
𝑛+1
)−(
𝑛+1
) Letnikov operator is as follows: 𝐴 + 𝛼𝐼𝑁 ∈ R𝑁×𝑁 + . In our
case since we have systems with the sequential fractional ℎ-
𝑛−𝛼+2 𝑛−𝛽 difference, in our condition ℎ and both orders 𝛼 and 𝛽 appear.
+( )−1−( )] 𝑥𝑖 (𝑎) Note that taking ℎ = 1 the sufficient condition (71) has the
𝑛+1 𝑛
form 𝐴 − ( 𝛼+𝛽2
) 𝐼𝑁 ∈ R𝑁×𝑁
+ .
𝛼+𝛽
= ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) − ( ) 𝑥𝑖 (𝑎 + 𝑛ℎ)
2 Let us now consider some examples that illustrate the
solution of the considered systems.
+ (𝛼 + 𝛽) 𝑥 (𝑎 + (𝑛 + 1) ℎ)
𝑛 Example 22. Let 𝑁 = 2, 𝛼 = 0.4, 𝛽 = 0.5 and ℎ = 0.01. Then
𝑘+1 𝛼+𝛽 let us take 𝑎 = −0.006 and 𝑏 = −0.005 and consider the linear
+ ∑ (−1) ( ) 𝑥𝑖 (𝑎 + (𝑛 − 𝑘 + 2) ℎ)
𝑘 system with sequential difference in the following form:
𝑘=3

𝑛−𝛽 𝑛−𝛼−𝛽 (𝑎 Δ0.4


ℎ,∗ 𝑥) (𝑛ℎ) = 𝑦 (𝑏 + 𝑛ℎ) ,
+ [( )−( )
𝑛+1 𝑛+1 (72)
(𝑏 Δ0.5
ℎ,∗ 𝑦) (𝑛ℎ) = 𝐴𝑥 (𝑎 + 𝑛ℎ)
𝑛−𝛼+2 𝑛+1
+( )−( )] 𝑥𝑖 (𝑎) .
𝑛+1 𝑛+1 with initial conditions (25) and (26), that is, (𝑎 Δ𝛼ℎ,∗ 𝑥)(0) = 𝑥0
By inequality (60) and the assumptions, that is, 𝑥(𝑎 + 𝑘ℎ) ∈ and 𝑥(𝑎) = 𝑥𝑎 , 𝑥0 , 𝑥𝑎 ∈ R2 . Moreover let 𝐴 = [ −2 1
1 −2 ]
𝛼+𝛽
R𝑁 𝑁
+ , 𝑘 ≤ 𝑛+1, and by (61) one gets 𝑥(𝑎+(𝑛+2)ℎ) ∈ R+ . Hence and 𝐵 = [ −4 1
1 −2 ]. Then the matrix ℎ 𝐴 − ( 𝛼+𝛽 2
) 𝐼2 has
10 Abstract and Applied Analysis

4 [3] F. M. Atıcı and P. W. Eloe, “A transform method in discrete frac-


tional calculus,” International Journal of Difference Equations,
vol. 2, no. 2, pp. 165–176, 2007.
[4] F. M. Atıcı and P. W. Eloe, “Initial value problems in discrete
3 fractional calculus,” Proceedings of the American Mathematical
Society, vol. 137, no. 3, pp. 981–989, 2009.
[5] N. R. O. Bastos, R. A. C. Ferreira, and D. F. M. Torres, “Discrete-
time fractional variational problems,” Signal Processing, vol. 91,
2 no. 3, pp. 513–524, 2011.
[6] F. Chen, X. Luo, and Y. Zhou, “Existence results for nonlinear
fractional difference equation,” Advances in Difference Equa-
tions, vol. 2011, Article ID 713201, 12 pages, 2011.
1 [7] M. T. Holm, The theory of discrete fractional calculus: devel-
opment and application [Ph.D. thesis], University of Nebraska,
Lincoln, Neb, USA, 2011.
[8] T. Kaczorek, “Fractional positive continuous-time linear sys-
tems and their reachability,” International Journal of Applied
0 1 2 3 4 5 Mathematics and Computer Science, vol. 18, no. 2, pp. 223–228,
2008.
Figure 2: Consider the following system (𝑎 Δ0.4
ℎ,∗ 𝑥)(𝑛ℎ) = 𝑦(𝑏 + 𝑛ℎ),
[9] K. S. Miller and B. Ross, “Fractional difference calculus,”
(𝑏 Δ0.5
ℎ,∗ 𝑦)(𝑛ℎ) = 𝐵𝑥(𝑎 + 𝑛ℎ), where 𝐵 is given in Example 22, with in Proceedings of the International Symposium on Univalent
initial conditions: 𝑥0 = (0, 1), 𝑥𝑎 = (0, 0.1), ℎ0.9 𝐴 − ( 0.9
2 ) 𝐼2 ∉ R2×2
+ . Functions, Fractional Calculus and Their Applications, pp. 139–
152, Nihon University, Kōriyama, Japan, 1988.
[10] D. Mozyrska and E. Girejko, “Overview of the fractional ℎ-
nonnegative values and the fact of staring from nonnegative difference operators,” in Advances in Harmonic Analysis and
initial conditions provides the remaining in the same positive Operator Theory, vol. 229 of Operator Theory: Advances and
Applications, pp. 253–268, Springer, New York, NY, USA, 2013.
cone. In Figure 1 we present the trajectory for 𝑇 = 500 steps
on the (𝑥1 , 𝑥2 )-plane that start from 𝑥𝑎 = (0, 0.1) and with [11] M. D. Ortigueira, “Fractional discrete-time linear systems,” in
Proceedings of the IEEE International Conference on Acoustics,
the initial value 𝑥0 = (0, 1). Let us notice that in the case
Speech, and Signal Processing (ICASSP ’97), vol. 3, pp. 2241–2244,
ℎ𝛼+𝛽 𝐵 − ( 𝛼+𝛽
2
) 𝐼2 ∉ R2×2
+ one can see that in a few initial steps IEEE Computer Society, Munich, Germany, 1997.
points are escaping from the positive area (see Figure 2). All [12] I. Podlubny, Fractional Differential Equations, Academic Press,
calculations were done in the Maple program. San Diego, Calif, USA, 1999.
[13] T. Abdeljawad, “Dual identities in fractional difference calculus
5. Conclusions within Riemann,” Advances in Difference Equations, vol. 2013,
article 36, 2013.
The behaviour of the solutions of systems with sequential [14] T. Abdeljawad, “On delta and nabla Caputo fractional differ-
fractional difference is studied. We present recursive formulas ences and dual identities,” Discrete Dynamics in Nature and
for nonlinear systems and give the exact formulas for the Society, vol. 2013, Article ID 406910, 12 pages, 2013.
unique solutions to systems in linear and semilinear cases. We [15] T. Abdeljawad, D. Baleanu, F. Jarad, and R. Agarwal, “Frac-
prove the sufficient condition for the positivity of considered tional sums and difference with binomial coefficients,” Discrete
systems. Dynamic Systems in Nature and Society, vol. 2013, Article ID
Our future goal is to study stability of systems with se- 104173, 6 pages, 2013.
quential fractional differences. [16] R. A. C. Ferreira and D. F. M. Torres, “Fractional ℎ-difference
equations arising from the calculus of variations,” Applicable
Analysis and Discrete Mathematics, vol. 5, no. 1, pp. 110–121, 2011.
Acknowledgments [17] N. R. O. Bastos, R. A. C. Ferreira, and D. F. M. Torres, “Necessary
optimality conditions for fractional difference problems of the
This project was supported by the foundayion of the National calculus of variations,” Discrete and Continuous Dynamical
Science Centre granted on the bases of the decision. The Systems, vol. 29, no. 2, pp. 417–437, 2011.
DEC-2011/03/B/ST7/03476. This work was supported by Bia- [18] D. Mozyrska and E. Pawłuszewicz, “Fractional discrete-time
lystok University of Technology Grant no. G/WM/3/2012. linear control systems with initialisation,” International Journal
of Control, vol. 85, no. 2, pp. 213–219, 2012.
References [19] D. Mozyrska, E. Girejko, and M. Wyrwas, “Comparison of h-
difference fractional operators,” in Advances in the Thoery and
[1] J. A. T. Machado, “Analysis and design of fractional-order digital Applications of Non-Integer Order Systems, W. Mitkowski, J.
control systems,” Systems Analysis Modelling Simulation, vol. 27, Kacprzyk, and J. Baranowski, Eds., vol. 257 of Lecture Notes
no. 2-3, pp. 107–122, 1997. in Electrical Engineering, pp. 191–197, Springer, New York, NY,
[2] T. Abdeljawad and D. Baleanu, “Fractional differences and USA, 2013.
integration by parts,” Journal of Computational Analysis and [20] F. Jarad, T. Abdeljawad, D. Baleanu, and K. Biçen, “On the
Applications, vol. 13, no. 3, pp. 574–582, 2011. stability of some discrete fractional nonautonomous systems,”
Abstract and Applied Analysis 11

Abstract and Applied Analysis, vol. 2012, Article ID 476581, 9


pages, 2012.
[21] D. Mozyrska and E. Pawluszewicz, “Local controllability of
nonlinear discrete-time fractional order systems,” Bulletin of the
Polish Academy of Sciences: Technical Sciences, vol. 61, no. 1, pp.
251–256, 2013.
[22] D. Mozyrska and E. Pawluszewicz, “Controllability of h-dif-
ference linear control systems with two fractional orders,”
International Journal of Systems Science, 2013.
[23] E. Pawluszewicz and D. Mozyrska, “Constrained controllability
of h-difference linear systems with two fractional orders,” in
Advances in the Theory and Applications of Non-Integer Order
Systems, W. Mitkowski, J. Kacprzyk, and J. Baranowski, Eds.,
vol. 257 of Lecture Notes in Electrical Engineering, pp. 67–75,
Springer, New York, NY, USA, 2013.
[24] M. Wyrwas, E. Girejko, D. Mozyrska, and E. Pawłuszewicz,
“Stability of fractional difference systems with two orders,” in
Advances in the Thoery and Applications of Non-Integer Order
Systems, W. Mitkowski, J. Kacprzyk, and J. Baranowski, Eds.,
vol. 257 of Lecture Notes in Electrical Engineering, pp. 41–52,
Springer, New York, NY, USA, 2013.
[25] K. S. Miller and B. Ross, An Introduction to the Fractional
Calculus and Fractional Differential Equations, John Wiley &
Sons, New York, NY, USA, 1st edition, 1993.
[26] M. Klimek, “Sequential fractional differential equations with
hadamard derivative,” Communications in Nonlinear Science
and Numerical Simulation, vol. 16, no. 12, pp. 4689–4697, 2011.
[27] K. M. Furati, “A Cauchy-type problem with a sequential frac-
tional derivative in the space of continuous functions,” Bound-
ary Value Problems, vol. 2012, article 58, 2012.
[28] D. Mozyrska, E. Girejko, and M. Wyrwas, “Fractional nonlinear
systems with sequential operators,” Central European Journal of
Physics, 2013.
[29] T. Kaczorek, “Reachability and controllability to zero tests for
standard and positive fractional discrete-time systems,” Journal
Européan des Systemès Automatisés, vol. 42, no. 6–8, pp. 769–
787, 2008.
[30] T. Kaczorek, “Fractional positive linear systems,” Kybernetes,
vol. 38, no. 7-8, pp. 1059–1078, 2009.
[31] T. Kaczorek, “Reachability of cone fractional continuous-time
linear systems,” International Journal of Applied Mathematics
and Computer Science, vol. 19, no. 1, pp. 89–93, 2009.
[32] T. Kaczorek, Selected Problems of Fractional Systems Theory, vol.
411, Springer, Berlin, Germany, 2011.
[33] T. Abdeljawad, “On Riemann and Caputo fractional differ-
ences,” Computers & Mathematics with Applications, vol. 62, no.
3, pp. 1602–1611, 2011.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 304739, 17 pages
http://dx.doi.org/10.1155/2013/304739

Research Article
Numerical Solution of Fuzzy Fractional Pharmacokinetics Model
Arising from Drug Assimilation into the Bloodstream

Ali Ahmadian,1,2 Norazak Senu,1,2 Farhad Larki,3 Soheil Salahshour,4


Mohamed Suleiman,1 and Md. Shabiul Islam3
1
Institute for Mathematical Research, Universiti Putra Malaysia (UPM), 43400 Serdang, Selangor, Malaysia
2
Mathematics Department, Science Faculty, Universiti Putra Malaysia (UPM), 43400 Serdang, Selangor, Malaysia
3
Institute of Microengineering and Nanoelectronics (IMEN), Universiti Kebangsaan Malaysia (UKM), 43600 Bangi, Selangor, Malaysia
4
Department of Computer Engineering, Mashhad Branch, Islamic Azad University, Mashhad, Iran

Correspondence should be addressed to Ali Ahmadian; ahmadian.hosseini@gmail.com

Received 22 August 2013; Revised 8 October 2013; Accepted 9 October 2013

Academic Editor: Ali H. Bhrawy

Copyright © 2013 Ali Ahmadian et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We propose a Jacobi tau method for solving a fuzzy fractional pharmacokinetics. This problem can model the concentration of the
drug in the blood as time increases. The proposed approach is based on the Jacobi tau (JT) method. To illustrate the reliability of
the method, some special cases of the equations are solved as test examples. The method reduces the solution of the problem to the
solution of a system of algebraic equations. Error analysis included the fractional derivative error estimation, and the upper bound
of the absolute errors is introduced for this method.

1. Introduction Mathematics is widely used for the quantitative descrip-


tion of drug absorption, distribution, metabolism, and excre-
Pharmacokinetics is defined as the study of the time course tion (ADME). Some parameters in pharmacokinetics are
of drug absorption, distribution, metabolism, and excretion defined by mathematic equations which can be obtained
[1]. Pharmacokinetics determines bioavailability, volume of directly by measurement or through calculation using exper-
distribution, and clearance. Bioavailability is the fraction of imental data based on developed mathematical equations.
a drug absorbed into the systemic circulation, while volume For example, the first and simplest model used for drug
of distribution and clearance are measure of apparent space absorption from the gastrointestinal (GI) tract assumes
in the body available to contain the drug and measure of a “pseudosteady state” and utilizes the physicochemical
the body’s ability to eliminate the drug, respectively. In properties of the compound in conjunction with the “pH-
pharmacokinetics, as a basic hypothesis, always a relation- partition” hypothesis to predict the fraction of dose absorbed
ship exists between the pharmacologic or toxic response [3–5]. Tissue distribution is another important determinant
to a drug and the concentration of the drug in the blood of the pharmacokinetics profile of a drug. Hence, in drug
(plasma). However, for some drugs there is no straight- development, the prediction of tissue distribution would help
forward relationship between concentration in plasma and predict the in vivo pharmacokinetics of a compound prior
pharmacologic effects [2]. Generally, the concentration of to any experiments in animals or man. Two methods are
drug in the systemic circulation is related to the concentration available to predict tissue distribution. These methods predict
of drug at its sites of action. Absorption, drug distribution either tissue: plasma ratios or the volume of distribution
(target tissues), drug concentration in circulation system, at steady state (V). From molecular descriptors, properties
rate of elimination/plasma concentration, and elimination such as lipophilicity/hydrophobicity are estimated [6]. In
(metabolism, excretion) are some parameters that directly some approaches, plasma protein binding is also taken
modify the pharmacokinetics parameters. into consideration [7]. Physiological information on tissue
2 Abstract and Applied Analysis

composition, the blood composition, and blood flow to mainly two: first and foremost, PKPD models incorporating
the tissue is utilized to develop a partitioning model [8– fractional calculus have not been proposed; second is compu-
10]. Möhler et al. [11] proposed a model that describes cell tational: while the analytical solution of fractional differential
growth and detailed investigations on the metabolism and equations is available in special cases, it turns out that even
the kinetics of the influenza infection cycle which allow the simplest PKPD models that can be constructed using
for the optimization of influenza virus vaccine production. fractional calculus do not allow analytical solutions [23, 24].
In addition to these simple models, more complex math- In this paper, we propose new category of PKPD mod-
ematical manipulations called mathematical models have els incorporating fuzzy fractional calculus and investigate
also been used to describe pharmacokinetics [12]. There their behavior using purposely written algorithm. The main
are many classifications of pharmacokinetics models. In a purpose of this paper is to attract the attention of the
specific classification for pharmacokinetics modeling, two field into these possibly interesting families of PKPD mod-
general approaches can be considered: compartment-based els. In particular, we will focus on compartment models,
modeling [13] and noncompartment-based modeling [14]. direct concentration-response relationships describing mod-
Both types of modeling take advantage of the quantitative els relating drug concentration (in plasma or biophase) to a
structure-pharmacokinetics relationships that are described pharmacodynamic effect.
by empirical mathematical algorithms. They can be used to The concept of fractional or noninteger order derivation
estimate the activity of a compound based on its chemical and integration can be traced back to the genesis of integer
structure in a numeric format [15]. order calculus itself [25–27]. Due to its tremendous scope and
In this study, our focus is to find the approximate solution applications in several disciplines, a considerable attention
of the fuzzy fractional model of the compartment model for has been given to exact and numerical solutions of fractional
the flow of antihistamine in the blood. To this end, we use differential equations which is an extremely difficult task.
the operational matrix of the Caputo fractional derivative of Moreover, the solution techniques and their reliability are
the Jacobi tau approximation based on the Jacobi polynomials rather more important aspects. Several methods have been
to derive the fuzzy fractional approximate solution under proposed by many researchers to solve the fractional order
Hukuhara differentiability (H-differentiability). To the best differential equations such as Taylor series [28], variational
of our knowledge, there are no results on the JT method for iteration method [29–31], adomian decomposition method
solving pharmacokinetics equations arising in mathematical [32, 33], fractional differential transform method (FDTM)
physics. This partially motivated our interest in such model [34, 35], and homotopy analysis method [36, 37].
and solution method. As it is known, the spectral method is one of the flexible
methods of discretization for most types of differential
1.1. Compartment Model. The compartment model frame- equations [38–40]. Historically, spectral method has been
work is an extremely natural and valuable means with which relegated to fractional calculus, but in few years, it has been
to formulate models for processes which have inputs and successfully applied for the fractional equation models based
outputs over time [16]. Compartment model itself can be on the different types of orthogonal polynomials such as
classified in one-compartment model and two-compartment Block pulse functions [41, 42], Legendre polynomials [43–
model. In the former one, the body is depicted as a kinetically 46], Chebyshev polynomials [47–49], Laguerre polynomials
homogeneous unit, while in the latter one the body resolves [50–53], and Bernstein polynomials [54–56]. Doha et al.
into a central compartment and a peripheral compartment [57] introduced the shifted Jacobi operational matrix of
[17]. According to the one-compartment model, the drug fractional derivative which is based on Jacobi tau method for
instantaneously distributes throughout the body, and at the solving numerically linear multiterm fractional differential
same time it equilibrates between tissues. Thus, the drug equations with initial or boundary conditions. Subsequently,
concentration time profile shows a monophasic response. Kazem [58] generalized Jacobi integral operational matrix to
On the other hand, in two-compartment model the drug fractional calculus. Afterwards, Doha et al. [59] proposed a
does not achieve instantaneous equilibration between the two direct solution techniques for solving the linear multiorder
compartments. Although, in this model, two compartments fractional initial value problems with constant and variable
have no anatomical or physiological meaning, it is usually coefficients using shifted Jacobi tau method and quadrature
surmised that the central compartment is composed of tissues shifted Jacobi tau method, respectively. In this paper, we
that are highly perfused such as heart, lungs, kidneys, liver, intend to extend the application of the Jacobi polynomi-
and brain while the peripheral compartment comprises less als to solve fuzzy fractional pharmacokinetics model of
well-perfused tissues such as muscle, fat, and skin [17]. order [0, 1].
On the other hand, a considerable attention has been
1.2. Methods Have Been Proposed to Solve Fractional Pharma- made to the fractional differential equations in the sense
cokinetics Models. While an increasing number of fractional of fuzzy setting theory. Agarwal et al. [60] is among the
order integrals and differential equations applications have pioneers who presented the concept of solutions for fractional
been reported in the physics [18, 19], signal processing differential equations with uncertainty. Thereafter, the exis-
[20], engineering, and bioengineering literatures [21, 22], tence and uniqueness of the solution of the fuzzy fractional
little attention has been paid to this class of models in the differential equations (FFDEs) under the Riemann-Liouville
pharmacokinetics-pharmacodynamic (PKPD) literature. The and Caputo’s fuzzy fractional differentiability were investi-
reasons for the lack of application to pharmacodynamics are gated in the literature [61–64]. However, the application of
Abstract and Applied Analysis 3

the numerical methods for solving FFDEs is unknown and In this paper, the set of all fuzzy numbers is denoted
traceless in the literature of the fuzzy fractional calculus, and by RF .
only a few number of researches have been reported for the
approximate solution of FFDEs [64–69]. Definition 2. Let 𝑢 ∈ RF and 𝑟 ∈ [0, 1]. The 𝑟-cut of 𝑢 is
In the present paper, we intend to introduce new families the crisp set [𝑢]𝑟 that contains all elements with membership
of PKPD models based on the application of fractional degree in 𝑢 greater than or equal to 𝑟; that is
calculus to PKPD models. The aim of the paper is not to
[𝑢]𝑟 = {𝑥 ∈ R | 𝑢 (𝑥) ≥ 𝑟} . (3)
claim the superiority of fractional dynamics models with
respect to standard ones, but it is simply to define the new For a fuzzy number 𝑢, its 𝑟-cuts are closed intervals in R, and
families and provide some insights into their qualitative we denote them by
behavior. The main purpose is to apply the fuzzy logic in
differential equations of fractional order which has been used [𝑢]𝑟 = [𝑢1𝑟 , 𝑢2𝑟 ] . (4)
as an effective tool for considering uncertainty in modeling
the processes; FFDEs can also offer a more comprehensive According to Zadeh’s extension principle, the operation
account of the process or phenomenon, specifically for of addition on RF is defined as follows:
analyzing the behavior of the PKPD models. Furthermore, we (𝑢 + V) (𝑥) = sup min {𝑢 (𝑦) , V (𝑥 − 𝑦)} , 𝑥 ∈ R,
suggest possible applications and stimulate further research, 𝑦∈R
(5)
which might, or might not, demonstrate the applicability and
importance of spectral methods by using of the orthogonal and scalar multiplication of a fuzzy number is given by
polynomials for finding the approximate solutions of the 𝑥
PKPD models based on the fuzzy fractional calculus. {𝑢 ( ) , 𝑘 > 0,
(𝑘 ⊙ 𝑢) (𝑥) = { 𝑘 (6)
The structure of this paper is as follows. In the next ̃
section, we briefly recall the mathematical foundations of {0, 𝑘 = 0,
fractional calculus, required definitions of fuzzy setting the- where 0̃ ∈ RF .
ory, and summarize the properties of Jacobi polynomials.
In Section 3, we then provide illustrations of the govern- Definition 3 (see [70]). The distance 𝐷(𝑢, V) between two
ing fraction equation. In Section 4, the proposed method fuzzy numbers 𝑢 and V is defined as
is explained for numerical solution of the derived FFDE.
Section 5 is devoted to the numerical solution of the problem 𝐷 (𝑢, V) = sup 𝑑𝐻 ([𝑢]𝑟 , [V]𝑟 ) , (7)
with different dose of the drug, and the error analysis is 𝑟∈[0,1]

provided to demonstrate the applicability and validity of where


the method. A final discussion comments on the results are
󵄨 󵄨 󵄨 󵄨
presented in Section 6. 𝑑𝐻 ([𝑢]𝑟 , [V]𝑟 ) = max {󵄨󵄨󵄨𝑢1𝑟 − V1𝑟 󵄨󵄨󵄨 , 󵄨󵄨󵄨𝑢2𝑟 − V2𝑟 󵄨󵄨󵄨} (8)
𝑟 𝑟
is the Hausdorff distance between [𝑢] and [V] .
2. Preliminaries and Notations
It is easy to see that 𝐷 is a metric in RF and has the
In this section, we are going to state the definition and following properties (see [70, 72]):
preliminaries of fuzzy mathematics [63, 70, 71], fractional
calculus [25, 27], and some properties of shifted Jacobi (i) 𝐷(𝑢 ⊕ 𝑤, V ⊕ 𝑤) = 𝐷(𝑢, V), ∀𝑢, V, 𝑤 ∈ RF ,
polynomials. (ii) 𝐷(𝑘 ⊙ 𝑢, 𝑘 ⊙ V) = |𝑘|𝐷(𝑢, V), ∀𝑘 ∈ R, 𝑢, V ∈ RF ,
(iii) 𝐷(𝑢 ⊕ V, 𝑤 ⊕ 𝑒) ≤ 𝐷(𝑢, 𝑤) + 𝐷(V, 𝑒), ∀𝑢, V, 𝑤 ∈ RF ,
2.1. The Fuzzy Settings Definitions
(iv) 𝐷(𝑢 + V, 0) ≤ 𝐷(𝑢, 0) + 𝐷(V, 0), ∀𝑢, V ∈ RF ,
Definition 1. Let 𝑢 be a fuzzy set in R. 𝑢 is called a fuzzy (v) (RF , 𝐷) is a complete metric space.
number if:
Definition 4 (see [73]). Let 𝑓 and 𝑔 be the two fuzzy-
(i) 𝑢 is normal: there exists 𝑥0 ∈ R such that 𝑢(𝑥0 ) = 1; number-valued functions on the interval [𝑎, 𝑏], that is, 𝑓, 𝑔 :
(ii) 𝑢 is convex: for all 𝑥, 𝑦 ∈ R and 0 ≤ 𝜆 ≤ 1, it holds [𝑎, 𝑏] → RF . The uniform distance between fuzzy-number-
that valued functions is defined by

𝑢 (𝜆𝑥 + (1 − 𝜆) 𝑦 ≥ min {𝑢 (𝑥) , 𝑢 (𝑦)}) ; (1) 𝐷∗ (𝑓, 𝑔) := sup 𝐷 (𝑓 (𝑥) , 𝑔 (𝑥)) . (9)
𝑥∈[𝑎,𝑏]

(iii) 𝑢 is upper semicontinuous: for any 𝑥0 ∈ R, it holds Remark 5 (see [73]). Let 𝑓 : [𝑎, 𝑏] → RF be fuzzy contin-
that uous. Then from property (iv) of Hausdorff distance, we can
define
𝑢 (𝑥0 ) ≥ lim± 𝑢 (𝑥) ;
̃ = sup max {󵄨󵄨󵄨𝑓1𝑟 (𝑥)󵄨󵄨󵄨 , 󵄨󵄨󵄨𝑓2𝑟 (𝑥)󵄨󵄨󵄨} , ∀𝑥 ∈ [𝑎, 𝑏] .
𝑥 → 𝑥0 (2)
𝐷 (𝑓 (𝑥) , 0) 󵄨 󵄨 󵄨 󵄨
𝑟∈[0,1]
0 (10)
(iv) [𝑢] = supp(𝑢) is a compact subset of R.
4 Abstract and Applied Analysis

Definition 6 (see [74]). Let 𝑥, 𝑦 ∈ RF . If there exists 𝑧 ∈ In this part, we firstly give some basic definitions and
RF such that 𝑥 = 𝑦 ⊕ 𝑧, then 𝑧 is called the H-difference some properties of fractional calculus [27]. Afterwards, the
of 𝑥 and 𝑦, and it is denoted by 𝑥 ⊖ 𝑦. extension of the fractional differentiability in the sense of
fuzzy concept is provided, and some relevant properties
In this paper, the sign “⊖” always stands for H-difference, which are used in the rest of the paper are given [61, 63, 64].
and note that 𝑥 ⊕ 𝑦 ≠𝑥 + (−𝑦). Also throughout the paper, Let 𝑚 be the smallest integer that exceeds V, then
it is assumed that the Hukuhara difference and generalized Caputo’s fractional derivative operator of order V > 0 is
Hukuhara differentiability exist. defined as
Theorem 7 (see [75]). Let 𝐹 : (𝑎, 𝑏) → RF be a function 𝑐 𝐽𝑚−V 𝐷𝑚 𝑓 (𝑥) , if 𝑚 − 1 < V < 𝑚,
and denote [𝐹(𝑡)]𝑟 = [𝑓𝑟 (𝑡), 𝑔𝑟 (𝑡)], for each 𝑟 ∈ [0, 1]. Then, 𝐷V 𝑓 (𝑥) = { (17)
𝐷𝑚 𝑓 (𝑥) , if V = 𝑚, 𝑚 ∈ N,
(1) if 𝐹 is (1)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are dif- where
ferentiable functions and
𝑥
1
󸀠 𝑟 𝐽V 𝑓 (𝑥) = ∫ (𝑥 − 𝑡)V−1 𝑓 (𝑡) 𝑑𝑡, V > 0, 𝑥 > 0. (18)
[𝐹 (𝑡)] = [𝑓𝑟󸀠 (𝑡) , 𝑔𝑟󸀠 (𝑡)] ; (11) Γ (V) 0
For the Caputo derivative, we have
(2) if 𝐹 is (2)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are dif-
𝑐
ferentiable functions and 𝐷V 𝐶 = 0, (𝐶 is a constant) ,
𝑟 𝑐
[𝐹󸀠 (𝑡)] = [𝑔𝑟󸀠 (𝑡) , 𝑓𝑟󸀠 (𝑡)] . (12) 𝐷V 𝑥 𝛽

{ 0, for 𝛽 ∈ N0 , 𝛽 < ⌈V⌉ ,


Definition 8 (see [76]). Consider the 𝑛 × 𝑛 linear system of {
= { Γ (𝛽 + 1) 𝛽−V for 𝛽 ∈ N0 , 𝛽 ≥ ⌈V⌉
the following equations: { 𝑥 ,
{ Γ (𝛽 + 1 − V) or 𝛽 ∉ N, 𝛽 > ⌊V⌋ .
𝑎11 𝑥1 + 𝑎12 𝑥2 + ⋅ ⋅ ⋅ + 𝑎1𝑛 𝑥𝑛 = 𝑦1 , (19)
𝑎21 𝑥1 + 𝑎22 𝑥2 + ⋅ ⋅ ⋅ + 𝑎2𝑛 𝑥𝑛 = 𝑦2 , Caputo’s fractional differentiation is a linear operation,
(13) namely,
..
. 𝑐
𝐷V (𝜆𝑓 (𝑥) + 𝜇𝑔 (𝑥)) = 𝜆𝑐 𝐷V 𝑓 (𝑥) + 𝜇𝑐 𝐷V 𝑔 (𝑥) , (20)
𝑎𝑛1 𝑥1 + 𝑎𝑛2 𝑥2 + ⋅ ⋅ ⋅ + 𝑎𝑛𝑛 𝑥𝑛 = 𝑦𝑛 .
where 𝜆 and 𝜇 are constants.
The matrix form of the above equations is
Theorem 10 (generalized Taylor formula [78]). Suppose
𝐴𝑋 = 𝑌, (14) that 𝐷𝑎𝑘𝛼 𝑓(𝑡) ∈ 𝐶(𝑎, 𝑏] for 𝑘 = 0, 1, . . . , 𝑛 + 1 where 0 < 𝛼 ≤
1; then
where the coefficient matrix 𝐴 = (𝑎𝑖𝑗 ), 1 ≤ 𝑖, 𝑗 ≤ 𝑛 is a 𝑛
crisp 𝑛 × 𝑛 matrix and 𝑦𝑖 ∈ RF , 1 ≤ 𝑖 ≤ 𝑛. This system is (𝑡 − 𝑎)𝑖𝛼 𝑐 𝑘𝛼
𝑓 (𝑡) = ∑ [ 𝐷𝑎 𝑓 (𝑡)]𝑡=𝑎 + 𝑅𝑛𝛼 (𝑡, 𝑎) , (21)
called a fuzzy linear system (FLS). 𝑖=0 Γ (𝑖𝛼 + 1)

Definition 9 (see [76]). A fuzzy number vector (𝑥1 , 𝑥2 , . . . , with


𝑥𝑛 )𝑡 given by 𝑥𝑖 = (𝑥𝑖 𝑟− , 𝑥𝑖 𝑟+ ), 1 ≤ 𝑖 ≤ 𝑛, 0 ≤ 𝑟 ≤ 1 is called a
(𝑡 − 𝑎)(𝑛+1)𝛼 𝑐 (𝑛+1)𝛼
solution of the fuzzy linear system (2) if 𝑅𝑛𝛼 (𝑡, 𝑎) = [ 𝐷𝑎 𝑓 (𝑡)]𝑡=𝜉 ,
Γ ((𝑛 + 1) 𝛼 + 1) (22)
𝑟
𝑛 𝑛
( ∑𝑎𝑖𝑗 𝑥𝑗 ) = ∑ (𝑎𝑖𝑗 𝑥𝑗 )− = 𝑦𝑖𝑟− ,
𝑟 𝑎 ⩽ 𝜉 ⩽ 𝑡, 𝑡 ∈ [𝑎, 𝑏] ,
𝑗=1 − 𝑗=1
(15) where
𝑟
𝑐
𝑛 𝑛
𝑟 𝐷𝑎𝑛𝛼 = 𝑐⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
𝐷𝑎𝛼𝑐 𝑐 𝐷𝑎𝛼 , . . . ,𝑐 𝐷𝑎𝛼 .
( ∑𝑎𝑖𝑗 𝑥𝑗 ) = ∑ (𝑎𝑖𝑗 𝑥𝑗 )+ = 𝑦𝑖𝑟+ . (23)
𝑛 𝑡𝑖𝑚𝑒𝑠
𝑗=1 + 𝑗=1
Here, 𝑐 𝐷𝛼 shows fractional derivative operator in the
If for a particular 𝑘, 𝑎𝑘𝑗 > 0, 1 ≤ 𝑗 ≤ 𝑛, we simply get Caputo sense, and 𝑓𝑇𝑛𝑖 (𝑡) = ∑𝑛𝑖=0 ((𝑡 − 𝑎)𝑖𝛼 /Γ(𝑖𝛼 +
𝑛 𝑛 1))[𝑐 𝐷𝑎𝑘𝛼 𝑓(𝑡)]𝑡=𝑎 is called fractional Taylor expansion with
∑ 𝑎𝑘𝑗 𝑥𝑗𝑟− = 𝑦𝑘𝑟− , ∑ 𝑎𝑘𝑗 𝑥𝑗𝑟+ = 𝑦𝑘𝑟+ . (16) degree 𝑛𝑖 of 𝑓, and 𝑅𝑛𝛼 (𝑡, 𝑎) is reminder term in fractional
𝑗=1 𝑗=1 Taylor expansion of 𝑓.
Now, we present some fuzzy fractional notations which
To solve fuzzy linear systems, see [77]. are used later in the paper.
Abstract and Applied Analysis 5

(i) 𝐿R𝑝F (𝑎, 𝑏), 1 ≤ 𝑝 ≤ ∞ is the set of all fuzzy-valued For the sake of simplicity, we say that the fuzzy-valued
measurable functions 𝑓 on [𝑎, 𝑏] where ||𝑓||𝑝 = function 𝑓 is 𝑐 [(1) − V]-differentiable if it is differentiable as
1 1/𝑝 in Definition 13 case (i), and 𝑓 is 𝑐 [(2) − V]-differentiable if
(∫0 (𝑑(𝑓(𝑡), 0))𝑝 𝑑𝑡) . it is differentiable as in Definition 13 case (ii), and so on for
RF the other cases.
(ii) 𝐶 [𝑎, 𝑏] is a space of fuzzy-valued functions which
are continuous on [𝑎, 𝑏].
Theorem 14 (see [63]). Let 0 < V ≤ 1 and 𝑓 ∈ 𝐴𝐶RF [𝑎, 𝑏];
(iii) 𝐶𝑛RF [𝑎, 𝑏] indicates the set of all fuzzy-valued func- then the fuzzy Caputo fractional derivative exists almost
tions which are continuous up to order 𝑛. everywhere on (𝑎, 𝑏), and for all 0 ≤ 𝑟 ≤ 1 one has
(iv) 𝐴𝐶RF [𝑎, 𝑏] denotes the set of all fuzzy-valued func-
𝑥 𝑓󸀠𝑟 (𝑡) 𝑑𝑡
tions which are absolutely continuous. 1
(𝑐 𝐷𝑎+
V
𝑓) (𝑥; 𝑟) = [ ∫ − ,
Note that one can easily find these notations in the crisp Γ (1 − V) 𝑎 (𝑥 − 𝑡)V
context in [25, 27] and references therein. 𝑥 𝑓󸀠𝑟 (𝑡) 𝑑𝑡
1 (30)
RF RF ∫ + ]
Definition 11 (see [63]). Let 𝑓 ∈ 𝐶 [𝑎, 𝑏] ∩ 𝐿 [𝑎, 𝑏]. The Γ (1 − V) 𝑎 (𝑥 − 𝑡)V
Riemann-Liouville integral of fuzzy-valued function 𝑓 is
1−V
defined as = [(𝐼𝑎+ 𝐷𝑓−𝑟 ) (𝑥) , (𝐼𝑎+
1−V
𝐷𝑓+𝑟 ) (𝑥)] ,
𝑥 𝑓
1 (𝑡) 𝑑𝑡
(𝑅𝐿 𝐼𝑎+
V
𝑓) (𝑥) = ∫ , when 𝑓 is (1)-differentiable, and
Γ (V) 𝑎 (𝑥 − 𝑡)1−V (24)
𝑥 > 𝑎, 0 < V ≤ 1. (𝑐 𝐷𝑎+
V
𝑓) (𝑥; 𝑟)
𝑥 𝑓󸀠𝑟 (𝑡) 𝑑𝑡 𝑥 𝑓󸀠𝑟 (𝑡) 𝑑𝑡
Definition 12 (see [63]). Let 𝑓 ∈ 𝐶RF [𝑎, 𝑏] ∩ 𝐿RF [𝑎, 𝑏]. Then, 1 1
𝑓 is said to be Caputo’s H-differentiable at 𝑥 when =[ ∫ + , ∫ −
,]
Γ (1 − V) 𝑎 (𝑥 − 𝑡)V Γ (1 − V) 𝑎 (𝑥 − 𝑡)V
(i) (𝑐 𝐷𝑎+
V
𝑓) (𝑥) = (𝑅𝐿 𝐷𝑎+
𝛽
[𝑓 (𝑡) ⊖ 𝑓 (𝑎)]) (𝑥) , 1−V
= [(𝐼𝑎+ 𝐷𝑓+𝑟 ) (𝑥) , (𝐼𝑎+
1−V
𝐷𝑓−𝑟 ) (𝑥)] ,
(25)
(ii) (𝑐 𝐷𝑎+
V
𝑓) (𝑥) = (𝑅𝐿 𝐷𝑎+
𝛽
[−𝑓 (𝑎) ⊖ (−𝑓 (𝑡))]) (𝑥) . (31)

Definition 13 (see [63]). Let 𝑓 : 𝐿RF [𝑎,𝑏] ∩ 𝐶RF [𝑎, 𝑏] and 𝑥0 ∈ when f is (2)-differentiable.
𝑥
(𝑎, 𝑏) and Φ(𝑥) = (1/Γ(1 − V)) ∫𝑎 (𝑓(𝑡)/(𝑥 − 𝑡)V )𝑑𝑡. We say
Theorem 15 (fuzzy generalized Taylor’s formula [72]).
that 𝑓(𝑥) is fuzzy Caputo fractional differentiable of order
Let 𝑓(𝑥) ∈ 𝐴𝐶RF [𝑎,𝑏] (0, 𝑏], and suppose that 𝑐 𝐷𝑘𝛼 𝑓(𝑥) ∈
0 < V ≤ 1 at 𝑥0 , if there exists an element (𝑐 𝐷𝑎V + 𝑓)(𝑥0 ) ∈
𝐶RF [𝑎,𝑏] (0, 𝑏] for 𝑘 = 0, 1, . . . , 𝑛 + 1 where 0 < 𝛼 < 1, 0 ≤
𝐶RF [𝑎,𝑏] [𝑎, 𝑏] such that for all 0 ≤ 𝑟 ≤ 1, ℎ > 0,
𝑥0 ≤ 𝑥 and 𝑥 ∈ (0, 𝑏]. Then, one has
Φ (𝑥0 + ℎ) ⊖ Φ (𝑥0 )
(i) (𝑐 𝐷𝑎V + 𝑓) (𝑥0 ) = lim+ 𝑟 𝑟
ℎ→0 ℎ [𝑓 (𝑥)] = [𝑓𝑟 (𝑥) , 𝑓 (𝑥)] ,
(26)
Φ (𝑥0 ) ⊖ Φ (𝑥0 − ℎ) 𝑛
𝑥𝑖𝛼
= lim+ , 𝑓𝑟 (𝑥) = ∑ 𝑐 𝑖𝛼 𝑟
𝐷 𝑓 (0+ )
ℎ→0 ℎ Γ (𝑖𝛼 + 1)
𝑖=0
or
𝑐
Φ (𝑥0 ) ⊖ Φ (𝑥0 + ℎ) 𝐷(𝑛+1)𝛼 𝑓𝑟 (𝑥0 ) (𝑛+1)𝛼
(ii) (𝑐 𝐷𝑎V + 𝑓) (𝑥0 ) = lim+ + 𝑥 , (32)
ℎ→0 −ℎ Γ (𝑛𝛼 + 𝛼 + 1)
(27)
𝑛
Φ (𝑥0 − ℎ) ⊖ Φ (𝑥0 ) 𝑟 𝑥𝑖𝛼 𝑐 𝑖𝛼 𝑟
= lim+ , 𝑓 (𝑥) = ∑ 𝐷 𝑓 (0+ )
ℎ→0 −ℎ 𝑖=0 Γ (𝑖𝛼 + 1)
or
𝑐
Φ (𝑥0 + ℎ) ⊖ Φ (𝑥0 ) 𝐷(𝑛+1)𝛼 𝑓𝑟 (𝑥0 ) (𝑛+1)𝛼
(iii) ( 𝑐
𝐷𝑎V + 𝑓) (𝑥0 ) = lim+ + 𝑥 ,
ℎ→0 ℎ Γ (𝑛𝛼 + 𝛼 + 1)
(28) 𝑟 𝑟
Φ (𝑥0 − ℎ) ⊖ Φ (𝑥0 ) where 𝑐 𝐷𝛼 𝑓𝑟 (0) = 𝑐 𝐷𝛼 𝑓𝑟 (𝑥)|𝑥=0 , 𝑐 𝐷𝛼 𝑓 (0) = 𝑐 𝐷𝛼 𝑓 (𝑥)|𝑥=0 .
= lim+ ,
ℎ→0 −ℎ
or 2.2. Jacobi Polynomials. The well-known Jacobi polynomials
𝑐 Φ (𝑥0 ) ⊖ Φ (𝑥0 + ℎ) associated with the parameters (𝛼 > −1, 𝛽 > −1)
(iv) ( 𝐷𝑎V + 𝑓) (𝑥0 ) = lim+ (see, e.g., Luke [79] and Szegö [80]), are a sequence of
ℎ→0 −ℎ
(29) (𝛼,𝛽)
polynomials 𝑃𝑖 (𝑡) (𝑖 = 0, 1, . . .), each, respectively, of
Φ (𝑥0 ) ⊖ Φ (𝑥0 − ℎ) degree 𝑖. For using these polynomials on (0, 𝐿), we present
= lim+ .
ℎ→0 ℎ the shifted Jacobi polynomials by implementing the change
6 Abstract and Applied Analysis

of variable 𝑡 = (2𝑥/𝐿 − 1). Let the shifted Jacobi polyno- where the coefficients 𝑎𝑗 are
(𝛼,𝛽) (𝛼,𝛽)
mials 𝑃𝑖 (2𝑥/𝐿 − 1) be denoted by 𝑃𝐿,𝑖 (𝑥), satisfying the
orthogonality relation
1 𝐿 (𝛼,𝛽) (𝛼,𝛽)
𝐿 𝑎𝑗 = ∫ 𝑃 (𝑥) 𝑢 (𝑥) 𝑤𝐿 (𝑥) 𝑑𝑥, 𝑗 = 0, 1, . . . (39)
(𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) ℎ𝑗 0 𝐿,𝑗
∫ 𝑃𝐿,𝑗 (𝑥) 𝑃𝐿,𝑘 (𝑥) 𝑤𝐿 (𝑥) 𝑑𝑥 = ℎ𝑘 , (33)
0

where 𝑤𝐿
(𝛼,𝛽)
(𝑥) = 𝑥𝛽 (𝐿 − 𝑥)𝛼 and For a given particular problem, only the first (𝑁 + 1)-terms
shifted Jacobi polynomials are used. Therefore, we have
{ 𝐿𝛼+𝛽+1 Γ (𝑘 + 𝛼 + 1) Γ (𝑘 + 𝛽 + 1)
{ 𝑖 = 𝑗,
ℎ𝑘 = { (2𝑘 + 𝛼 + 𝛽 + 1) 𝑘!Γ (𝑘 + 𝛼 + 𝛽 + 1) (34) 𝑁
{ (𝛼,𝛽)
𝑢𝑁 (𝑥) ≃ ∑ 𝑎𝑗 𝑃𝐿,𝑗 (𝑥) .
{0, 𝑖 ≠𝑗.
𝑗=0
(40)
(𝛼,𝛽)
The shifted Jacobi polynomial 𝑃𝐿,𝑖 (𝑥) of degree 𝑖 has the
form Theorem 16 (see [59]). The Caputo fractional derivative of
(𝛼,𝛽) order V of the shifted Jacobi polynomials of degree 𝑖 is obtained
𝑃𝐿,𝑖 (𝑥)
from
𝑖
Γ (𝑖 + 𝛽 + 1) Γ (𝑖 + 𝑘 + 𝛼 + 𝛽 + 1)
= ∑ (−1)𝑖−𝑘 𝑥𝑘 , ∞
Γ (𝑘 + 𝛽 + 1) Γ (𝑖 + 𝛼 + 𝛽 + 1) (𝑖 − 𝑘)!𝑘!𝐿𝑘 (𝛼,𝛽) (𝛼,𝛽)
𝑘=0 𝐷V 𝑃𝐿,𝑖 (𝑥) = ∑ 𝑆V (𝑖, 𝑗, 𝛼, 𝛽) 𝑃𝐿,𝑖 (𝑥) ,
(35) 𝑗=0 (41)
where 𝑖 = ⌈V⌉ , ⌈V⌉ + 1, . . . ,
(𝛼,𝛽) Γ (𝑖 + 𝛽 + 1)
𝑃𝐿,𝑖 (0) = (−1)𝑖 ,
Γ (𝛽 + 1) 𝑖! where
(36)
(𝛼,𝛽) Γ (𝑖 + 𝛼 + 1)
𝑃𝐿,𝑖 (𝐿) = . 𝑆V (𝑖, 𝑗, 𝛼, 𝛽)
Γ (𝛼 + 1) 𝑖!
𝑖
Also we can state the shifted Jacobi polynomial by the
following recurrence relation: = ∑ (−1)𝑖−𝑘 𝐿𝛼+𝛽−V+1 Γ (𝑗 + 𝛽 + 1)
𝑘=⌈V⌉
(𝛼,𝛽)
𝑃𝐿,𝑖 (𝑥) × Γ (𝑖 + 𝛽 + 1) Γ (𝑖 + 𝑘 + 𝛼 + 𝛽 + 1)
2𝑥
= (𝛼 + 𝛽 + 2𝑖 − 1) {(𝛼2 − 𝛽2 + ( − 1)) × (ℎ𝑗 Γ (𝑗 + 𝑘 + 𝛼 + 𝛽 + 1) Γ (𝑘 + 𝛽 + 1)
𝐿
−1
× Γ (𝑖 + 𝛼 + 𝛽 + 1) Γ (𝑘 − V + 1) (𝑖 − 𝑘)!)
× (𝛼 + 𝛽 + 2𝑖) (𝛼 + 𝛽 + 2𝑖 − 2) }
𝑗

× (2𝑖 (𝛼 + 𝛽 + 𝑖) (𝛼 + 𝛽 + 2𝑖 − 2))
−1 × ∑(−1)𝑗−𝑙 Γ (𝑗 + 𝑙 + 𝛼 + 𝛽 + 1) Γ (𝛼 + 1)
𝑙=0
(𝛼,𝛽)
× 𝑃𝐿,𝑖−1 (𝑥) × Γ (𝑙 + 𝑘 + 𝛽 − V + 1)
(𝛼 + 𝑖 − 1) (𝛽 + 𝑖 − 1) (𝛼 + 𝛽 + 2𝑖) −1
− × (Γ (𝑙 + 𝛽 + 1) Γ (𝑙 + 𝑘 + 𝛼 + 𝛽 − V + 2) (𝑗 − 𝑙)!𝑙!) .
𝑖 (𝛼 + 𝛽 + 𝑖) (𝛼 + 𝛽 + 2𝑖 − 2) (42)
(𝛼,𝛽)
× 𝑃𝐿,𝑖−2 (𝑥) 𝑖 = 2, 3, . . . ,
(37) 3. Pharmacokinetics Model Equation
(𝛼,𝛽) (𝛼,𝛽) 3.1. Drug Assimilation into the Blood. The drug dissolves in
where 𝑃𝐿,0 (𝑥) = 1 and 𝑃𝐿,1 (𝑥) = ((𝛼 + 𝛽 + 2)/2)(2𝑥/𝐿 −
1) + (𝛼 − 𝛽)/2. the gastrointestinal tract (GI), and each ingredient is diffused
We notice that a function 𝑢(𝑥), square integrable into the bloodstream. They are carried to the locations in
in (0, 𝐿), can be expanded in terms of shifted Jacobi which they act and are removed from the blood by the
polynomials as kidneys and liver. Generally, the problem of drug assimilation
into the body can be considered as a two-compartment
+∞ model: GI-tract and the bloodstream [16]. Different com-
(𝛼,𝛽)
𝑢 (𝑥) = ∑ 𝑎𝑗 𝑃𝐿,𝑗 (𝑥) , (38) partments and the input/output of the model are depicted in
𝑗=0 Figure 1.
Abstract and Applied Analysis 7

Drug intake Digestion Tissue Insert this formula for 𝑥(𝑡) into the second rate equation,
GI tract Blood
which becomes
𝑑𝑦 (𝑡)
Figure 1: Schematic of input/output compartment for drug assimi- + 𝑘2 𝑦 (𝑡) = 𝑘1 𝐴𝑒−𝑘1 𝑡 𝑦 (0) = 0. (47)
𝑑𝑡
lation.
In this paper, we are concerned with fractional time deriva-
tives regarding (17) and (18). Having defined 𝑐 𝐷V , we can
For each compartment by applying the balance law we can now turn to fractional differential equations and systems
obtain of fractional differential equations, which will be used to
specify PKPD models and will need to be solved over an
rate of change of interval [0, 𝑡], in accordance with appropriate initial condi-
{ } tions.
drug in GI tract
A typical feature of differential equations (both classical
rate of drug rate drug leaves and fractional) is the need to specify additional conditions
={ }−{ }
intake GI-tract in order to produce a unique solution. For the case of
(43) Caputo fractional differential equations, these additional
rate of change of conditions are just the initial conditions which are simi-
{ }
drug in blood larly required by classical ODEs. In contrast, for Riemann-
Liouville fractional differential equations, these additional
rate drug rate drug leaves
={ }−{ }. conditions constitute certain fractional derivatives (and/or
enters blood blood
integrals) of the unknown solution at the initial point 𝑡 =
0 [81], which are functions of 𝑡. These initial conditions are
In this study, we consider a case of a single cold pill. Also, there
not physical; furthermore, it is not clear how such quantities
is no ingestion of the drug except that which occurs initially.
are to be measured from experiment, say, so that they can
be appropriately assigned in an analysis [82]. If for no other
3.2. Case of a Single Cold Pill. Let us consider 𝑥(𝑡) to be reason, the need to solve fractional differential equations
the amount of drug in the GI-tract at time 𝑡 and 𝑦(𝑡), the is justification enough for choosing Caputo’s definition for
amount in the bloodstream at time 𝑡. In the GI-tract we fractional differentiation over the more commonly used (at
suppose that the pill is to be swallowed, and so after this event least in mathematical analysis) definition of Liouville and
(over subsequent time) we have nothing more entering the Riemann, and this is the operator that we choose to use in
GI-tract. The pill dissolves and diffuses into the bloodstream the following.
from the GI-tract. So, the GI-tract is only an output term. We start by representing drug concentration in the
Assuming that the output rate is proportional to the amount effect compartment by the (Caputo) fractional differential
of drug in the bloodstream, which is in fact proportional to equation:
the GI-tract drug concentration, then [16],
𝑐
𝐷V 𝑦 (𝑡) + 𝑘2 𝑦 (𝑡) = 𝑘1 𝐴𝑒−𝑘1 𝑡 𝑦 (0) = 0. (48)
𝑑𝑦
= −𝑘1 𝑥, 𝑥 (0) = 𝑥0 , (44) In the standard direct action model, the effect at
𝑑𝑡 time 𝑡, 𝑌(𝑡), is expressed by an arbitrary (memory-less) func-
where 𝑥0 is the amount of a drug in the pill and 𝑘1 is a tion of drug concentration in the effect site at time 𝑡, 𝐺(𝑦(𝑡));
positive coefficient. We suppose that 𝑦(0) = 0, which means however to generate a wider class of relationships, we assume
that the initial amount of the drug in the bloodstream is zero. that the effect at time 𝑡 ∈ [0, 1] is related to the fuzzy Caputo
As the drug diffuses from the GI-tract, the level increases, and fractional derivative of 𝑦(𝑡). So, we have
as the kidneys and liver remove it, the level of drug gradually 𝑐
𝐷V 𝑦 (𝑡) + 𝑘2 𝑦 (𝑡) = 𝑘1 𝐴𝑒−𝑘1 𝑡 𝑦 (0; 𝑟) = [𝑦𝑟 , 𝑦𝑟0 ] , (49)
decreases. Thus, 0
RF RF
𝑑𝑦 in which 𝑦(𝑥) : 𝐿 [0, 1] ∩ 𝐶 [0, 1] is a continuous fuzzy-
= 𝑘1 𝑥 − 𝑘2 𝑦, 𝑦 (0) = 0, (45) valued function and 𝑐 𝐷0V+ denotes the fuzzy Caputo frac-
𝑑𝑡
tional derivative of order V ∈ [0, 1].
with 𝑘2 another positive constant. Decongestant and an
antihistamine are the component of the cold pill, and the Remark 17. In this paper, the drug concentration is modeled
coefficient of proportionality (𝑘1 and 𝑘2 ) is different for the by an oscillation-relaxation fuzzy fractional differential equa-
different component drugs in the pill. tion. So the right hand side of (49) is determined based on the
In order to obtain the growth and decay of antihistamine model parameters. It should mention here that the proposed
levels in the GI tract and bloodstream, rearrange the first rate method can be extended easily for solving other types of
FFDEs with more complicated right hand side functions. For
equation, multiply by the integrating factor 𝑒𝑘1 𝑡 , integrate,
more details, one can refer to (33) in [65] and [57–59].
and then use the initial data to obtain
We have assumed that 𝑘1 ≠𝑘2 , an assumption that is
𝑋 = 𝐴𝑒(−𝑘1 𝑡) . (46) justified by the pharmaceutical data. For the “average” person,
8 Abstract and Applied Analysis

a pharmaceutical company estimates that the values of the Remark 18. In the remainder of paper, for more simplicity, we
rate constants for the antihistamine in the cold pills are 𝑘1 = (𝛼,𝛽) (𝛼,𝛽)
consider 𝑃𝑖 (𝑥) instead of 𝑃1,𝑖 (𝑥) and 𝑤(𝛼,𝛽) (𝑥) = (1 −
0.6931 (hour)−1 and 𝑘2 = 0.0231 (hour)−1 . It can be observed (𝛼,𝛽)
𝑥)𝛼 ⊙ 𝑥𝛽 instead of 𝑤1 (𝑥) = (1 − 𝑥)𝛼 ⊙ 𝑥𝛽 .
from (46) that level of antihistamine in the bloodstream
increases as the time increases and saturate in a maximum Remark 19. Practically, only the first (𝑁 + 1)-terms shifted
value of antihistamine; however, (47) can conclude that as Jacobi polynomials are taken into consideration. So, we have
the time increases the amount of antihistamine in GI-tract
declines and reaches a minimum value. We now consider 𝑁
(𝛼,𝛽)
two different cases of patients who are not average. We 𝑢 (𝑥) ≃ 𝑢𝑁 (𝑥) = ∑ ∗ 𝑎𝑖 ⊙ 𝑃𝑖 (𝑥) = a𝑇 ⊙ Φ𝑁 (𝑥) , (52)
often define 𝑘2 as clearance coefficient of medication from 𝑖=0
the blood. The value of 𝑘2 often varies from old and sick
patients than young and healthy cases. This means that the where the fuzzy shifted Jacobi coefficient vector 𝐴𝑇 and
level of medication in the blood may become and then shifted Jacobi polynomials vector Φ𝑁+1 (𝑥) are presented by
remain excessively high with a standard dosage for the cases
that 𝑘2 is much lower, which is normally observed in old and a𝑇 = [𝑎0 , 𝑎1 , . . . , 𝑎𝑁] ,
sick cases. (53)
We investigate sensitivity of the medication over a 24- (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) 𝑇
Φ𝑁 (𝑥) = [𝑃0 (𝑥) , 𝑃1 (𝑥) , . . . , 𝑃𝑁 (𝑥)] .
hour period by keeping the value of 𝑘1 fixed at some value
such as 1.386 but setting 𝑘2 = 0.01386, 0.06386, 0.1386,
0.6386, 1.386 (𝐴 = 1). Different values of 𝑘2 correspond to We can state the fuzzy approximate function (52) in the
people of different ages and states of health. For all cases parametric form as follows.
as it is expected that, for the first few hours, the amount
of antihistamine in the bloodstream increases and then Definition 20 (see [65]). Let 𝑢(𝑥) ∈ 𝐿R𝑝F [0, 1] ∩ 𝐶RF [0, 1];
declines gradually. However, for the cases with lower value of the approximation of fuzzy-valued function 𝑢(𝑥) in the para-
clearance coefficient (old and sick), the maximum value of the metric form is
medication in bloodstream was much higher than the cases 𝑁 𝑁
with high 𝑘2 , and it did not decline for remaining times. This 𝑢𝑟 (𝑥) ≃ 𝑢𝑁
𝑟 𝑟
(𝑥) = [∑𝑎𝑖,− 𝑃𝑖
(𝛼,𝛽) 𝑟
(𝑥) , ∑𝑎𝑖,+ 𝑃𝑖
(𝛼,𝛽)
(𝑥)] ,
means that the level of medication in the bloodstream stays at 𝑖=0 𝑖=0
high level for a long time, and it could not be absorbed from
the blood. The same trend is observed for the case that 𝑘2 is 0 ≤ 𝑟 ≤ 1.
constant (𝑘2 = 0.0231) and the value of 𝑘1 varies (𝑘1 = (54)
0.06931, 0.11, 0.3, 0.6931, 1.0, and 1.5). In this case, the 𝑘1 is
a constant for the GI tract which is analogous to the 𝑘2 for We call that Span{𝑃𝑛(𝛼,𝛽) (𝑥) : 𝑛 ≥ 0} forms a complete
the bloodstream. RF
orthogonal system in 𝐿 2,𝑤 (𝛼,𝛽) (0, 1). Hence, we set

4. Description of the Method (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)


S𝑁 (0, 1) = Span {𝑃0 (𝑥) , 𝑃1 (𝑥) , . . . , 𝑃𝑁 (𝑥)} .
In this study, by developing the Jacobi polynomial approxi- (55)
mation [57, 58, 65] with the help of the matrix operations,
the tau method, and the fuzzy Caputo fractional derivative, The shifted Jacobi tau method to (49) is to obtain 𝑢𝑁 ∈
we obtain an approximate solution of the problem (49) for 𝑆𝑁(0, 1) such that
different values of 𝑘1 and 𝑘2 . As it is known, the existence
(𝛼,𝛽) (𝛼,𝛽)
and uniqueness of the FFDEs have been introduced in [60, (𝐷V 𝑢𝑁, 𝑃𝑘 (𝑥))𝑤(𝛼,𝛽) + 𝑘2 (𝑢𝑁, 𝑃𝑘 (𝑥))𝑤(𝛼,𝛽)
61, 63]. Here, we assume that the shifted Jacobi polynomials
are defined on (0, 1), so we have [65] = (𝑘1 𝐴𝑒−𝑘1 𝑡 , 𝑃𝑘
(𝛼,𝛽)
(𝑥))𝑤(𝛼,𝛽) , (56)
+∞
(𝛼,𝛽)
𝑢 (𝑥) = ∑ ∗ 𝑎𝑖 ⊙ 𝑃1,𝑖 (𝑥) , (50) 𝑘 = 0, 1, . . . , 𝑁 − 1, 𝑢𝑁 (0) = 𝑑0 ,
𝑖=0
1
where the fuzzy coefficients 𝑎𝑖 are gained by where 𝑤(𝛼,𝛽) (𝑥) = (1 − 𝑥)𝛼 ⊙ 𝑥𝛽 and (𝑢, V)𝑤(𝛼,𝛽) = ∫0 𝑤(𝛼,𝛽)
1 1 (𝛼,𝛽) (𝑥) ⊙ 𝑢(𝑥) ⊙ V(𝑥)𝑑𝑥 denotes the fuzzy inner product
(𝛼,𝛽) RF
𝑎𝑖 = ∫ 𝑃 (𝑥) ⊙ 𝑢 (𝑥) ⊙ 𝑤1 (𝑥) 𝑑𝑥, 𝑖 = 0, 1, . . . , in 𝐿 2,𝑤(𝛼,𝛽) (0, 1) and 𝑑0 is the fuzzy initial value of the prob-
ℎ𝑖 0 1,𝑖
(51) lem.
Let us define the following notations:
(𝛼,𝛽)
and 𝑤1 (𝑥) = (1 − 𝑥)𝛼 ⊙ 𝑥𝛽 , 𝑢 ∈ 𝐿R𝑝F [0, 1] ∩ 𝐶RF [0, 1],
(𝛼,𝛽)
(𝛼,𝛽)
and 𝑃1,𝑖 (𝑥) is as the same as the shifted Jacobi polyno- 𝑓𝑘 = (𝑘1 𝐴𝑒−𝑘1 𝑡 , 𝑃𝑘 (𝑥))𝑤(𝛼,𝛽) 𝑘 = 0, 1, . . . , 𝑁 − 1,
mials presented in Section 2.2, and ∑∗ means addition with (57)
𝑇
respect to ⊕ in RF . f = (𝑓0 , 𝑓1 , . . . , 𝑓𝑁−1 , 𝑑0 ) .
Abstract and Applied Analysis 9

𝑘V
Hence, the variational formulation of (56) according to Proof. Since 𝐷0+ 𝑓(𝑡) ∈ 𝐶(0, 𝑏], 𝑘 = 0, 1, . . . , 𝑁 + 1, 𝑓 can be
Relation (14) in [65], by means of a typical tau method like expanded to the fractional Taylor series
in the crisp context [29] and (52), is equivalent to
𝑁
𝑡𝑖V
𝑁 𝑓𝑇𝑁𝑖 (𝑡) = ∑ 𝑖V
[𝐷0+ 𝑓 (𝑡)]𝑡=0+ , (63)
∑𝑎𝑗 ⊙ [(𝐷(V) 𝑃𝑗
(𝛼,𝛽)
(𝑥) , 𝑃𝑘
(𝛼,𝛽)
(𝑥)) 𝑖=0 Γ (𝑖V + 1)
𝑤(𝛼,𝛽)
𝑗=0
and its reminder term is
(𝛼,𝛽) (𝛼,𝛽)
+ 𝑘2 (𝑃𝑗 (𝑥) , 𝑃𝑘 (𝑥)) ]
𝑤(𝛼,𝛽)
(58) V 𝑡(𝑁+1)V
−𝑘1 𝑡 (𝛼,𝛽) 𝑅𝑁 (𝑡, 0) = [𝐷(𝑁+1)V 𝑓 (𝑡)]𝑡=𝜉 ,
= (𝑘1 𝐴𝑒 , 𝑃𝑘 (𝑥))𝑤(𝛼,𝛽) , 𝑘 = 0, 1, . . . , 𝑁 − 1, Γ ((𝑁 + 1) V + 1) 0+ (64)
𝑁 0 ⩽ 𝜉 ⩽ 𝑡, ∀𝑡 ∈ (0, 𝑏] .
(𝛼,𝛽)
∑𝑎𝑗 ⊙ 𝑃𝑗 (0) = 𝑑0 .
𝑗=0 Therefore,
Denoting
𝑓 (𝑡) − 𝑓𝑇𝑁𝑖 (𝑡) = 𝑅𝑁
V
(𝑡, 0) . (65)
𝐴 = (𝑎𝑘𝑗 )0<𝑘,𝑗<𝑁, 𝐶 = (𝑐𝑘𝑗 )0<𝑘,𝑗<𝑁, (59) From (65) and triangle inequality, we obtain

󵄨󵄨 󵄨 󵄨 󵄨
we investigate that (58) is equivalent to the matrix system 󵄨󵄨𝑓 (𝑡) − 𝑦𝑁,V (𝑡)󵄨󵄨󵄨 = 󵄨󵄨󵄨󵄨𝑓 (𝑡) − 𝑦𝑁,V (𝑡) + 𝑓𝑇 (𝑡) − 𝑓𝑇 (𝑡)󵄨󵄨󵄨󵄨
𝑁𝑖 𝑁𝑖

(𝐴 + 𝑘2 𝐶) ⊙ a = f. 󵄨 󵄨 󵄨 󵄨
(60) ⩽ 󵄨󵄨󵄨󵄨𝑓 (𝑡) − 𝑓𝑇𝑁𝑖 (𝑡)󵄨󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑓𝑇𝑁𝑖 (𝑡) − 𝑦𝑁,V (𝑡)󵄨󵄨󵄨󵄨
󵄨 V 󵄨 󵄨 󵄨
The elements of the matrices mentioned above are deter- = 󵄨󵄨󵄨𝑅𝑁 (𝑡, 0)󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑓𝑇𝑁𝑖 (𝑡) − 𝑦𝑁,V (𝑡)󵄨󵄨󵄨󵄨 .
mined in the next theorem.
(66)
(𝛼,𝛽)
Theorem 21 (see [59]). Let us denote 𝑎𝑘𝑗 = (𝐷(V) 𝑃𝑗 (𝑥),
(𝛼,𝛽) Therefore, an upper bound of the absolute errors is obtained
𝑃𝑘 (𝑥))𝑤(𝛼,𝛽) (0 ⩽ 𝑘 ⩽ 𝑁 − 1, 0 ⩽ 𝑗 ⩽ 𝑁), 𝑎𝑘𝑗 = for the method in the crisp cases.
(𝛼,𝛽) (𝛼,𝛽)
𝐷𝑘−𝑁𝑃𝑗 (0) (𝑘 = 𝑁, 0 ⩽ 𝑗 ⩽ 𝑁) and 𝑐𝑘𝑗 = (𝑃𝑗 (𝑥),
(𝛼,𝛽)
𝑃𝑘 (𝑥))𝑤(𝛼,𝛽) (0 ⩽ 𝑘 ⩽ 𝑁 − 1, 0 ⩽ 𝑗 ⩽ 𝑁); then the nonzero Now, we provide an upper bound for the absolute errors
elements of 𝑎𝑘𝑗 and 𝑐𝑘𝑗 are given as of the fuzzy approximate solution by using the proposed
method.
𝑎𝑘𝑗 = ℎ𝑘 𝑆V (𝑗, 𝑘, 𝛼, 𝛽) , 0 ⩽ 𝑘 ⩽ 𝑁 − 1, 1 ⩽ 𝑗 ⩽ 𝑁, 𝑎𝑘𝑗
Theorem 23. Let 𝑢(𝑡) ∈ 𝐿R𝑝F [0, 1] ∩ 𝐶RF [0, 1] be the fuzzy
𝑗−𝑘+𝑁
(−1) Γ (𝑗 + 𝛽 + 1) (𝑗 + 𝛼 + 𝛽 + 1)𝑘−𝑁 exact solution of (49), and 𝑢𝑁(𝑡) is the best fuzzy Jacobi
= ,
Γ (𝑗 − 𝑘 + 𝑁 + 1) Γ (𝑘 − 𝑁 + 1 + 𝛽) approximate function (52), and suppose that 𝑐 𝐷𝑘V 𝑢(𝑡) ∈
𝐶RF [0, 𝑏] for 𝑘 = 0, 1, . . . , 𝑁 + 1 where 0 < V < 1 and 𝑡 ∈
𝑘 = 𝑁, 0 ⩽ 𝑗 ⩽ 𝑁, 𝑐𝑘𝑗 [0, 𝑏]. Then, one has
= ℎ𝑘 , 0 ⩽ 𝑘 = 𝑗 ⩽ 𝑁 − 1. 𝑟
[𝑢 (𝑡)]𝑟 = [𝑢𝑟 (𝑡) , 𝑢𝑟 (𝑡)] , [𝑢𝑁 (𝑡)] = [𝑢𝑟𝑁 (𝑡) , 𝑢𝑟𝑁 (𝑡)] ,
(61)
󵄨󵄨 󵄨 󵄨 V 󵄨 󵄨 𝑁𝑖 󵄨
󵄨󵄨𝑢 (𝑡) − 𝑢𝑁 (𝑡)󵄨󵄨󵄨 ⩽ 󵄨󵄨󵄨𝑅𝑁 (𝑡, 0)󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑢𝑇 (𝑡) − 𝑢𝑁 (𝑡)󵄨󵄨󵄨󵄨 ,
4.1. Error Analysis. In this section, error analysis of the
method will be presented for the FFDEs. Firstly, an upper 󵄨󵄨󵄨𝑢 (𝑡) − 𝑢𝑁 (𝑡)󵄨󵄨󵄨 ⩽ 󵄨󵄨󵄨󵄨𝑅V (𝑡, 0)󵄨󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑢𝑁𝑖 (𝑡) − 𝑢𝑁 (𝑡)󵄨󵄨󵄨󵄨 ,
󵄨 󵄨 󵄨 𝑁 󵄨 󵄨 𝑇 󵄨
bound of the absolute errors will be given for the technique (67)
by using generalized Taylor formula. Secondly, an error
bound will be introduced for the approximation of the fuzzy where regarding Theorem 15, one has
fractional Caputo’s derivative using Jacobi polynomials.
𝑟 𝑁𝑖
Lemma 22. Let 𝑦𝑁,V (𝑡) and 𝑓(𝑡) be the Jacobi approximate [𝑓𝑇𝑁𝑖 (𝑡)] = [𝑓𝑁𝑖 (𝑡; 𝑟) , 𝑓𝑇 (𝑡; 𝑟)]
𝑇
solution (40) and the exact solution of (48), respectively.
𝑁
𝑘V
If 𝐷0+ 𝑓(𝑡) ∈ 𝐶(0, 𝑏], 𝑘 = 0, 1, . . . , 𝑁 + 1, then 𝑡𝑖V 𝑖V
= [∑ [𝐷0+ 𝑓 (𝑡)] ,
Γ (𝑖V + 1) 𝑡=0+
󵄨󵄨 󵄨 󵄨 V 󵄨 󵄨 𝑁𝑖 󵄨 𝑖=0
󵄨󵄨𝑓 (𝑡) − 𝑦𝑁,V (𝑡)󵄨󵄨󵄨 ⩽ 󵄨󵄨󵄨𝑅𝑁 (𝑡, 0)󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑓𝑇 (𝑡) − 𝑦𝑁,V (𝑡)󵄨󵄨󵄨󵄨 , (62)
𝑁
𝑡𝑖V 𝑖V
∑ [𝐷0+ 𝑓 (𝑡)]𝑡=0+ ] ,
where 𝑓𝑇𝑁𝑖 and 𝑅𝑁
V
are defined according to Theorem 10. 𝑖=0 Γ (𝑖V + 1)
10 Abstract and Applied Analysis

V 𝑟 V,𝑟
[𝑅𝑁 (𝑡, 0)] = [𝑅V,𝑟
𝑁 (𝑡, 0) , 𝑅𝑁 (𝑡, 0)] assumption, namely, 𝑘2 = 0.0231 and Dose 𝐴 = 1, is as
follows:
𝑐 (𝑁+1)V 𝑟
𝐷 𝑓 (0+) (𝑁+1)V 𝑐
=[ 𝑡 , 𝐷V 𝑦 (𝑡) + 0.0231𝑦 (𝑡) = 0.6931𝑒−0.6931𝑡 ,
Γ (𝑁V + V + 1) (71)
[ 𝑦 (0; 𝑟) = [−1 + 𝑟, 1 − 𝑟] .
𝑐 𝑟
𝐷(𝑁+1)V 𝑓 (0+) (𝑁+1)V ] By using Theorem 7 for the above equation under 𝑐 [(1)−V]
𝑡 .
Γ (𝑁V + V + 1) differentiability, we have the following systems:
]
𝑐 V
(68) 𝐷 𝑦 (𝑡; 𝑟) + 0.0231𝑦 (𝑡; 𝑟)

Proof. It is straightforward from Lemma 22 and the fuzzy = 0.6931𝑒−0.6931𝑡 , 0 < V ≤ 1,


function definition. 𝑦 (0; 𝑟) = −1 + 𝑟, 0 < 𝑟 ≤ 1,
(72)
Also in the following theorem, according to the Relation 𝑐 V
𝐷 𝑦 (𝑡; 𝑟) + 0.0231𝑦 (𝑡; 𝑟)
(14) in [65], an upper bound for the absolute errors of the
approximate function of fuzzy fractional Caputo’s derivative = 0.6931𝑒−0.6931𝑡 , 0 < V ≤ 1,
is provided. 𝑦 (0; 𝑟) = 1 − 𝑟, 0 < 𝑟 ≤ 1.
Theorem 24 (see [65]). Assume that the error function of Solving (72) leads to determining the exact solution of
fuzzy Caputo fractional derivative operator by using the shifted (71) as follows:
Jacobi polynomials, 𝐸𝑘,V , is continuously fuzzy differentiable
for 0 < 𝑥0 ≤ 𝑥, 𝑥 ∈ (0, 1]. Additionally, 𝐸𝑘,V ∈ 𝑌 (𝑡; 𝑟) = (−1 + 𝑟) 𝐸V,1 [−0.0231𝑡V ]
𝐶RF [𝑥0 , 1] and 0 < V < 1; then the error bound is given by 𝑡
+ ∫ (𝑡 − 𝑥)V−1 𝐸V,V [−0.0231(𝑡 − 𝑥)V ]
𝐷∗ (RF 𝐷V Φ (𝑥) ,RF 𝐷(V) Φ (𝑥)) 0

(69) × (0.6931𝑒−0.6931𝑥 ) 𝑑𝑥, 0 < V ≤ 1,


𝑥0−V 𝛽+2 √
≤ (𝑆)2 ( ) 𝐵 (𝛼 + 1, 𝛽 + 1). 𝑌 (𝑡; 𝑟) = (1 − 𝑟) 𝐸V,1 [−0.0231𝑡 ] V (73)
(2)! |Γ (1 − V)| 2
𝑡
5. Numerical Results + ∫ (𝑡 − 𝑥)V−1 𝐸V,V [−0.0231(𝑡 − 𝑥)V ]
0

In this section, the fuzzy fractional PKPD model (49) is × (0.6931𝑒−0.6931𝑥 ) 𝑑𝑥, 0 < 𝑟 ≤ 1.
solved for different values of 𝑘1 and 𝑘2 by using the JT
method presented in Section 4. We have performed all We seek the fuzzy approximate solutions by applying the
numerical computations with a computer program written in technique described in Section 4 with 𝑁 = 2 and as
MATLAB. Also, absolute errors between fuzzy approximate
solution [𝑦𝑁]𝑟 = [𝑦𝑟 , 𝑦𝑟𝑁] and the corresponding exact solu- 𝑦2 (𝑡) = 𝑎0 ⊙ 𝑃0
(𝛼,𝛽) (𝛼,𝛽)
(𝑡) + 𝑎1 ⊙ 𝑃1 (𝑡) + 𝑎2 ⊙ 𝑃0
(𝛼,𝛽)
(𝑡) .
𝑁
tions 𝑦(𝑡; 𝑟) = [𝑦(𝑡; 𝑟), 𝑦(𝑡; 𝑟)], that is, [𝑁𝑒 ]𝑟 = [|𝑦𝑟 − 𝑦𝑟 |, (74)
𝑁
|𝑦𝑟𝑁 − 𝑦𝑟 |], are considered.
Now, we recall the FFDE (49) as follows: Here, we have

𝑐 0 0 0
𝐷V 𝑦 (𝑡) + 𝑘2 𝑦 (𝑡) = 𝑘1 𝐴𝑒−𝑘1 𝑡
(70) 𝐷0.85 = ( 1.8639 0.3901 −0.1755) , (75)
𝑦 (0; 𝑟) = [−1 + 𝑟, 1 − 𝑟] , −0.3901 4.5267 0.8696

in which 𝑦(𝑥) : 𝐿RF [0, 1] ∩ 𝐶RF [0, 1] is a continuous fuzzy- where 𝛼 = 𝛽 = 0. Regarding (54), we have
valued function and 𝑐 𝐷0V+ denotes the fuzzy Caputo frac-
𝑦𝑟 (𝑡) = 𝑎𝑟0 + 𝑎𝑟1 (2𝑥 − 1) + 𝑎𝑟2 (6𝑥2 − 6𝑥 + 1)
tional derivative of order V ∈ [0, 1]. 2
Two cases are considered for solving by the proposed (76)
technique. In the first case, we assume that 𝑘2 is unchange- 𝑦𝑟2 (𝑡) = 𝑎𝑟0 + 𝑎𝑟1 (2𝑥 − 1) + 𝑎𝑟2 (6𝑥2 − 6𝑥 + 1) .
able and different values of 𝑘1 are substituted in (49) to get
By solving the fuzzy linear system (60) in the parametric
the fuzzy approximate solution. Conversely, in the second
form, one can get fuzzy unknown coefficients in (76) as
case we try to solve the problem such that 𝑘1 has an invari-
able value and 𝑘2 is varied. The problem is analyzed for both 𝑎0 = −0.5757, 𝑎1 = 0.2624, 𝑎2 = −0.0619,
cases in details. (77)
𝑎0 = 1.2008, 𝑎1 = 0.2408, 𝑎2 = −0.0600,
Case I. Let us consider from Section 3 that 𝑘2 = 0.0231,
but let 𝑘1 vary (e.g., 0.6931, 0.11, and 0.3). So (49), with the where we assume that 𝑟-cut = 0.1 in (77).
Abstract and Applied Analysis 11

Table 1: The results of the proposed method for Case I with V = 0.85, 𝛼 = 𝛽 = 0, and 𝑁 = 8.

𝑘1 = 0.6931 𝑘1 = 0.11 𝑘1 = 0.3 𝑘1 = 0.6931 𝑘1 = 0.11 𝑘1 = 0.3


𝑟 𝑟 𝑟 𝑟
𝑁𝑟𝑒 𝑁𝑟𝑒 𝑁𝑟𝑒 𝑁𝑒 𝑁𝑒 𝑁𝑒
0 9.0140𝑒 − 5 1.6623𝑒 − 5 4.0452𝑒 − 5 8.4377𝑒 − 5 1.0860𝑒 − 5 3.4689𝑒 − 5
0.1 8.9852𝑒 − 5 1.6335𝑒 − 5 4.0164𝑒 − 5 8.4665𝑒 − 5 1.1148𝑒 − 5 3.4977𝑒 − 5
0.2 8.9564𝑒 − 5 1.6046𝑒 − 5 3.9876𝑒 − 5 8.4953𝑒 − 5 1.1436𝑒 − 5 3.5265𝑒 − 5
0.3 8.9276𝑒 − 5 1.5758𝑒 − 5 3.9588𝑒 − 5 8.5242𝑒 − 5 1.1724𝑒 − 5 3.5553𝑒 − 5
0.4 8.8988𝑒 − 5 1.5470𝑒 − 5 3.9300𝑒 − 5 8.5530𝑒 − 5 1.2012𝑒 − 5 3.5842𝑒 − 5
0.5 8.8700𝑒 − 5 1.5182𝑒 − 5 3.9011𝑒 − 5 8.5818𝑒 − 5 1.2300𝑒 − 5 3.6130𝑒 − 5
0.6 8.8411𝑒 − 5 1.4894𝑒 − 5 3.8723𝑒 − 5 8.6106𝑒 − 5 1.2588𝑒 − 5 3.6418𝑒 − 5
0.7 8.8123𝑒 − 5 1.4606𝑒 − 5 3.8435𝑒 − 5 8.6394𝑒 − 5 1.2877𝑒 − 5 3.6706𝑒 − 5
0.8 8.7835𝑒 − 5 1.4317𝑒 − 5 3.8147𝑒 − 5 8.6682𝑒 − 5 1.3165𝑒 − 5 3.6994𝑒 − 5
0.9 8.7547𝑒 − 5 1.4029𝑒 − 5 3.7859𝑒 − 5 8.6971𝑒 − 5 1.3453𝑒 − 5 3.7282𝑒 − 5
1 8.7259𝑒 − 5 1.3741𝑒 − 5 3.7571𝑒 − 5 8.7259𝑒 − 5 1.3741𝑒 − 5 3.7571𝑒 − 5

10−3.4

10−3.5
Absolute error (Ne )

10−3.6

10−3.7

10−3.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cuts

𝛼=𝛽=0 𝛼 = 0, 𝛽 = 0.5
𝛼 = 𝛽 = 0.5 𝛼 = 0.5, 𝛽 = 0

Figure 2: The absolute errors for different 𝛼 and 𝛽 with 𝑁 = 9, 𝑘1 = 0.3 and V = 0.75, Case I.

10−5
Absolute error (Ne )

10−6

10−7
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cuts

N=4 N=9
N=7 N = 11

Figure 3: The absolute errors for different values 𝑁 with V = 0.95, 𝑘1 = 0.11 and 𝛼 = 𝛽 = 0, Case I.
12 Abstract and Applied Analysis

Table 2: The fuzzy coefficients for Case II with V = 0.75, 𝛼 = 𝛽 = 0.5, 𝑁 = 2, and 𝑘2 = 0.01386.

𝑟 𝑎0 𝑎1 𝑎2 𝑎0 𝑎1 𝑎2
0 −0.42793 0.24218 −0.11135 1.55482 0.23234 −0.11003
0.1 −0.32879 0.24169 −0.11128 1.45568 0.23283 −0.11009
0.2 −0.22965 0.24120 −0.11121 1.35654 0.23332 −0.11016
0.3 −0.13052 0.24071 −0.11115 1.25741 0.23382 −0.11022
0.4 −0.03138 0.24021 −0.11108 1.15827 0.23431 −0.11029
0.5 0.06775 0.23972 −0.11102 1.05913 0.23480 −0.11036
0.6 0.16689 0.23923 −0.11095 0.95999 0.23529 −0.11042
0.7 0.26603 0.23874 −0.11088 0.86085 0.23578 −0.11049
0.8 0.36516 0.23824 −0.11082 0.76172 0.23628 −0.11055
0.9 0.46430 0.23775 −0.11075 0.66258 0.23677 −0.11062
1 0.56344 0.23726 −0.11069 0.56344 0.23726 −0.11069

Table 3: The results of the proposed method for Case II with V = 0.95, 𝛼 = 𝛽 = 0.5, and 𝑁 = 7.

𝑘2 = 0.01386 𝑘2 = 0.06386 𝑘2 = 0.1386 𝑘2 = 0.01386 𝑘2 = 0.6386 𝑘2 = 0.1386


𝑟 𝑟 𝑟 𝑟
𝑁𝑟𝑒 𝑁𝑟𝑒 𝑁𝑟𝑒 𝑁𝑒 𝑁𝑒 𝑁𝑒
0 5.2740𝑒 − 4 5.2151𝑒 − 4 5.1213𝑒 − 4 5.1711𝑒 − 4 4.7626𝑒 − 4 4.2036𝑒 − 4
0.1 5.2688𝑒 − 4 5.1925𝑒 − 4 5.0754𝑒 − 4 5.1763𝑒 − 4 4.7852𝑒 − 4 4.2495𝑒 − 4
0.2 5.2637𝑒 − 4 5.1698𝑒 − 4 5.0295𝑒 − 4 5.1814𝑒 − 4 4.8078𝑒 − 4 4.2953𝑒 − 4
0.3 5.2585𝑒 − 4 5.1472𝑒 − 4 4.9836𝑒 − 4 5.1866𝑒 − 4 4.8304𝑒 − 4 4.3412𝑒 − 4
0.4 5.2534𝑒 − 4 5.1246𝑒 − 4 4.9377𝑒 − 4 5.1917𝑒 − 4 4.8531𝑒 − 4 4.3871𝑒 − 4
0.5 5.2482𝑒 − 4 5.1019𝑒 − 4 4.8919𝑒 − 4 5.1968𝑒 − 4 4.8757𝑒 − 4 4.4330𝑒 − 4
0.6 5.2431𝑒 − 4 5.0793𝑒 − 4 4.8460𝑒 − 4 5.2020𝑒 − 4 4.8983𝑒 − 4 4.4789𝑒 − 4
0.7 5.2380𝑒 − 4 5.0567𝑒 − 4 4.8001𝑒 − 4 5.2071𝑒 − 4 4.9209𝑒 − 4 4.5248𝑒 − 4
0.8 5.2328𝑒 − 4 5.0341𝑒 − 4 4.7542𝑒 − 4 5.2123𝑒 − 4 4.9436𝑒 − 4 4.5707𝑒 − 4
0.9 5.2277𝑒 − 4 5.0114𝑒 − 4 4.7083𝑒 − 4 5.2174𝑒 − 4 4.9662𝑒 − 4 4.6165𝑒 − 4
1 5.2225𝑒 − 4 4.9888𝑒 − 4 4.6624𝑒 − 4 5.2225𝑒 − 4 4.9888𝑒 − 4 4.6624𝑒 − 4

k1 = 0.6931
Approximate solution

−2
1
0.8 1
0.5 0.6
0.2 0.4
x 0 0
r-cuts

k1 = 0.11 k1 = 0.3
Approximate solution

Approximate solution

2 2

0 0

−2 −2
1 1
1 1
0.5 0.5 0.5 0.5
x 0 0 r-cuts x 0 0 r-cuts

Figure 4: The fuzzy approximate solution of Case I, for different value of 𝑘1 , 𝛼 = 𝛽 = 0.5, 𝑁 = 8.
Abstract and Applied Analysis 13

10−3 reports for various values of the 𝑘1 [12]. It should be noted


here that the value of 𝑁𝑒𝑟 is directly proportional to the vari-
ation of exact solutions and corresponding fuzzy approximate
solution. As a result, increase of 𝑁𝑒𝑟 with increasing 𝑘1 is also
expected.
Absolute error (Ne )

10−4

Case II. In this case, we assume that 𝑘1 = 1.386 and


is not variable, but 𝑘2 takes different values (i.e., 𝑘2 =
0.01386, 0.06386, and 0.1386) and Dose 𝐴 = 1. So (70) alters
10−5 with, for example, 𝑘2 = 0.01386 as follows:
𝑐
𝐷V 𝑦 (𝑡) + 0.01386𝑦 (𝑡) = 1.386𝑒−1.386𝑡 ,
(78)
𝑦 (0; 𝑟) = [−1 + 𝑟, 1 − 𝑟] ,
−6
10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 with the exact solution as:
r-cuts
𝑌 (𝑡; 𝑟) = (−1 + 𝑟) 𝐸V,1 [−0.01386𝑡V ]
𝛼=𝛽=0 𝛼 = 0, 𝛽 = 0.5 𝑡
𝛼 = 𝛽 = 0.5 𝛼 = 0.5, 𝛽 = 0 + ∫ (𝑡 − 𝑥)V−1 𝐸V,V [−0.01386(𝑡 − 𝑥)V ]
0
Figure 5: The absolute errors for different 𝛼 and 𝛽 with 𝑁 = × (1.386𝑒−1.386𝑥 ) 𝑑𝑥, 0 < V ≤ 1,
8, 𝑘2 = 0.1386, and V = 0.95, Case II.
V
𝑌 (𝑡; 𝑟) = (1 − 𝑟) 𝐸V,1 [−0.01386𝑡 ]
𝑡
10−2 + ∫ (𝑡 − 𝑥)V−1 𝐸V,V [−0.01386(𝑡 − 𝑥)V ]
0
× (1.386𝑒−1.386𝑥 ) 𝑑𝑥, 0 < 𝑟 ≤ 1.

(79)
Absolute error (Ne )

10−3
By applying the JT method in Section 4, we can get the
fuzzy unknown coefficients {𝑎𝑗 }𝑁
𝑗=0 by using (58) which is in
the parametric form as follows:
10−4 𝑁
(𝛼,𝛽) (𝛼,𝛽)
∑𝑎𝑗 [(𝐷(V) 𝑃𝑗 (𝑥) , 𝑃𝑘 (𝑥))
𝑤(𝛼,𝛽)
𝑗=0

(𝛼,𝛽) (𝛼,𝛽)
10−5 + 0.01386(𝑃𝑗 (𝑥) , 𝑃𝑘 (𝑥)) ]
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 𝑤(𝛼,𝛽)
r-cuts
(𝛼,𝛽)
= (1.386𝐴𝑒−1.386𝑡 , 𝑃𝑘 (𝑥))𝑤(𝛼,𝛽) ,
m=4 m=9
m=7 m = 11
𝑘 = 0, 1, . . . , 𝑁 − 1,
Figure 6: The absolute errors for different values 𝑁 with V =
0.85, 𝑘2 = 0.1386, and 𝛼 = 𝛽 = 0, Case II. 𝑁
(𝛼,𝛽)
∑ 𝑎𝑗 𝑃𝑗 (0) = −1 + 𝑟,
𝑗=0

The comparison between absolute errors of different 𝑘1 𝑁


(𝛼,𝛽) (𝛼,𝛽)
obtained by our method is shown in Table 1. In Figure 2 ∑𝑎𝑗 [(𝐷(V) 𝑃𝑗 (𝑥) , 𝑃𝑘 (𝑥))
𝑤(𝛼,𝛽)
𝑗=0
logarithmic plot of absolute error with different values
of 𝛼 and 𝛽 was obtained, and different number of Jacobi (𝛼,𝛽) (𝛼,𝛽)
functions was experienced for solving Case I by using + 0.01386(𝑃𝑗 (𝑥) , 𝑃𝑘 (𝑥)) ]
𝑤(𝛼,𝛽)
the proposed method in Figure 3. Finally, the approximate (𝛼,𝛽)
fuzzy solutions are shown for different values of 𝑘1 with the = (1.386𝐴𝑒−1.386𝑡 , 𝑃𝑘 (𝑥))𝑤(𝛼,𝛽) ,
fractional order V = 0.85 in Figure 4.
The absolute error value for 𝑟-cut varied from 0 to 1 𝑘 = 0, 1, . . . , 𝑁 − 1,
for different value of 𝑘1 which is calculated in Table 1. As 𝑁
it can be observed at a constant 𝑟-cut by increasing the ∑ 𝑎𝑗 𝑃𝑗
(𝛼,𝛽)
(0) = 1 − 𝑟,
value of 𝑘1 which is the ratio of variation of the drug in the 𝑗=0
bloodstream to the amount of drug in the GI-tract the value
of absolute error increases. This is analogous to the previous (80)
14 Abstract and Applied Analysis

k2 = 0.01386

Approximate solution
2

−2
1
0.8 1
0.5 0.6
0.2 0.4
x 0 0
r-cuts

k2 = 0.06386 k2 = 0.1386
Approximate solution

Approximate solution
2 2

0 0

−2 −2
1 1
1 1
0.5 0.5 0.5 0.5
x 0 0 r-cuts x 0 0 r-cuts

Figure 7: The fuzzy approximate solution of Case II, for different value of 𝑘2 , 𝛼 = 0, 𝛽 = 0.5, and 𝑁 = 9.

where 𝐷(V) is given as drug stays in the bloodstream for longer time. Since the
constant value of 𝑘2 appears with negative sign in the main
0 0 0 differential equation as we expected, the value of 𝑁𝑒𝑟 which is
𝐷0.75 = ( 2.6929 0.5524 −0.1755) , (81) directly proportional to the exact solutions is decreased with
−1.2429 4.2241 1.1048 increasing the value of 𝑘2 .
with 𝛼 = 𝛽 = 0.5 and 𝑁 = 2. By substituting the above
matric in (80) we can reach the fuzzy linear algebraic system
(60) that can solve easily to determine the fuzzy coefficients
as shown in Table 2. 6. Conclusion
We compared JT method results for different values Fuzzy theory provides a suitable way to objectively account
of 𝑘2 and the outcomes are tabulated in Table 3. The absolute for parameter uncertainty in models. Fuzzy logic approaches
errors of the proposed method for this case are exhibited appear promising in preclinical applications and might be
in Figure 5 with four choices of 𝛼 and 𝛽. Clearly, the best useful in drug discovery and design. Considerable progress
approximation is achieved when the values of 𝛼 = 0 and 𝛽 = has been made in the last few years in the development
0. Additionally, we compared the approximate solutions of computational approaches for prediction of drug absorp-
obtained by the present method at 𝑁 = 4, 7, 9, and 11 which tion, distribution, metabolism, and excretion. Whilst several
is shown in Figure 6. From Figure 6, one can conclude approaches have been developed in pharmacokinetics, most
that with increasing the number of Jacobi polynomials, the of these approaches have not yet been adequately used in the
absolute errors are decreasing dramatically. Finally, Figure 7 complex process such as prediction of metabolism, and they
shows the numerical results for different values of 𝑘2 at V = require further improvement.
0.75. It can be seen that for all the values of 𝑘2 the approxi- In summary in this research, a tau method based on
mate solution is the fuzzy number. the Jacobi operational matrix was utilized to numerically
In Table 3, the value of absolute error for 𝑟-cut varied solve the PKPD equation, arising from drug assimilation
from 0 to 1 for different value of 𝑘2 is calculated. At a into the bloodstream. The comparison of the results shows
constant 𝑟-cut value, a descending trend is observed with that the present method is a powerful mathematical tool for
increasing the value of 𝑘2 . This trend is opposed to the trend finding the numerical solutions of a generalized linear fuzzy
observed in Table 1 for different value of 𝑘1 . 𝑘2 is clearance fractional PKPD equation.
constant and it is a coefficient for 𝑦(𝑡) which is the amount Although we concentrated on applying our algorithm to
of drug in the bloodstream in (44). We already explained solve fuzzy fractional PKPD equation, we show that such
that by decreasing the value of 𝑘2 ,which happens in the algorithm can be applied to solve other types of fractional
case of old and sick patient, the drug absorption from the equations models in science and engineering fields. Our
bloodstream by kidney and liver also decreases and the algorithm for the fuzzy fractional PKPD equations is efficient
Abstract and Applied Analysis 15

and numerically stable. Numerical results are presented [14] P. Veng-Pedersen, “Noncompartmentally-based pharmacoki-
which exhibit the high accuracy of the proposed algorithm. netic modeling,” Advanced Drug Delivery Reviews, vol. 48, no.
By using the obtained operational matrix, the following 2-3, pp. 265–300, 2001.
objectives were achieved. [15] D. E. Mager, “Quantitative structure-pharmacokinetic/pharma-
codynamic relationships,” Advanced Drug Delivery Reviews, vol.
(i) The given fuzzy fractional differential equation was 58, no. 12-13, pp. 1326–1356, 2006.
converted into a fuzzy algebraic system of equations, [16] B. Barnes and G. R. Fulford, Mathematical Modelling with Case
simplifying the solution procedure. Studies: A Differential Equation Approach Using Maple and
(ii) The method is computer oriented; thus, solving fuzzy MATLAB, CRC Press, 2011.
fractional differential equations of different fractional [17] S. Dhillon and K. Gill, “Basic pharmacokinetics,” in Clinical
orders V becomes a matter of changing V only. Pharmacokinetics, pp. 1–44, Pharmaceutical Press, 2006.
[18] L. Gaul, P. Klein, and S. Kemple, “Damping description
involving fractional operators,” Mechanical Systems and Signal
References Processing, vol. 5, no. 2, pp. 81–88, 1991.
[1] B. G. Katzung, S. B. Masters, and A. J. Trevor, Basic & Clinical [19] W. G. Glockle and T. F. Nonnenmacher, “A fractional calculus
Pharmacology, McGraw-Hill, 2004. approach to self-similar protein dynamics,” Biophysical Journal,
vol. 68, no. 1, pp. 46–53, 1995.
[2] H. Peng and B. Cheung, “A review on pharmacokinetic model-
ing and the effects of environmental stressors on pharmacoki- [20] R. J. Marks II and M. W. Hall, “Differintegral interpolation from
netics for operational medicine: operational pharmacokinetics,” a bandlimited signal’s samples,” IEEE Transactions on Acoustics,
Tech. Rep. ADA509469, DTIC, 2009. Speech, and Signal Processing, vol. 29, no. 4, pp. 872–877, 1981.
[3] A. Suzuki, W. I. Higuchi, and N. F. Ho, “Theoretical model [21] A. Atangana and D. Baleanu, “Nonlinear fractional Jaulent-
studies of drug absorption and transport in the gastrointestinal Miodek and Whitham-Broer-Kaup equations within Sumudu
tract. I,” Journal of Pharmaceutical Sciences, vol. 59, no. 5, pp. transform,” Abstract and Applied Analysis, vol. 2013, Article ID
644–651, 1970. 160681, 8 pages, 2013.
[4] H. Boxenbaum, “Absorption potential and its variants,” Phar- [22] B. Mehdinejadiani, A. A. Naseri, H. Jafari, A. Ghanbarzadeh,
maceutical Research, vol. 16, no. 12, p. 1893, 1999. and D. Baleanu, “A mathematical model for simulation of
a water table profile between two parallel subsurface drains
[5] A. Boobis, U. Gundert-Remy, P. Kremers, P. Macheras, and
using fractional derivatives,” Computers & Mathematics with
O. Pelkonen, “In silico prediction of ADME and pharmacoki-
Applications, vol. 66, no. 5, pp. 785–794, 2013.
netics: report of an expert meeting organised by COST B15,”
European Journal of Pharmaceutical Sciences, vol. 17, no. 4-5, pp. [23] C. Csajka and D. Verotta, “Pharmacokinetic-pharmacodynamic
183–193, 2002. modelling: history and perspectives,” Journal of Pharmacokinet-
[6] P. Poulin, K. Schoenlein, and F. P. Theil, “Prediction of adipose ics and Pharmacodynamics, vol. 33, no. 3, pp. 227–279, 2006.
tissue: plasma partition coefcients for structurally unrelated [24] D. Verotta, “Fractional dynamics pharmacokinetics-pharmaco-
drugs,” Journal of Pharmaceutical Sciences, vol. 90, pp. 436–447, dynamic models,” Journal of Pharmacokinetics and Pharmaco-
2001. dynamics, vol. 37, no. 3, pp. 257–276, 2010.
[7] F. Lombardo, R. S. Obach, M. Y. Shalaeva, and F. Gao, “Predic- [25] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional
tion of volume of distribution values in humans for neutral and Calculus: Models and Numerical Methods, vol. 3 of Series on
basic drugs using physicochemical measurements and plasma Complexity, Nonlinearity and Chaos, World Scientific Publish-
protein binding data,” Journal of Medicinal Chemistry, vol. 45, ing, Hackensack, NJ, USA, 2012.
no. 13, pp. 2867–2876, 2002. [26] D. Baleanu, Z. B. G. Güvenç, and J. A. Tenreiro Machado, New
[8] P. Poulin and F. P. Theil, “A priori prediction of tissue: Trends in Nanotechnology and Fractional Calculus Applications,
plasma partition coefcients of drugs to facilitate the use of Springer, New York, NY, USA, 2010.
physiologically-based pharmacokinetic models in drug discov- [27] I. Podlubny, Fractional Differential Equations, vol. 198 of Math-
ery,” Journal of Pharmaceutical Sciences, vol. 89, pp. 16–35, 1999. ematics in Science and Engineering, Academic Press, San Diego,
[9] P. Poulin and F.-P. Theil, “Prediction of pharmacokinetics prior Calif, USA, 1999.
to in vivo studies. 1. Mechanism-based prediction of volume of [28] M. Gülsu, Y. Öztürk, and A. Anapalı, “Numerical approach
distribution,” Journal of Pharmaceutical Sciences, vol. 91, no. 1, for solving fractional relaxation-oscillation equation,” Applied
pp. 129–156, 2002. Mathematical Modelling, vol. 37, no. 8, pp. 5927–5937, 2013.
[10] P. Poulin and F.-P. Theil, “Prediction of pharmacokinetics prior [29] M. Dehghan and A. Saadatmandi, “A tau method for the one-
to in vivo studies. II. Generic physiologically based pharma- dimensional parabolic inverse problem subject to temperature
cokinetic models of drug disposition,” Journal of Pharmaceu- overspecification,” Computers and Mathematics with Applica-
tical Sciences, vol. 91, no. 5, pp. 1358–1370, 2002. tions, vol. 52, no. 6-7, pp. 933–940, 2006.
[11] L. Möhler, D. Flockerzi, H. Sann, and U. Reichl, “Mathematical [30] H. Jafari, H. Tajadodi, and D. Baleanu, “A modified variational
model of influenza a virus production in large-scale microcar- iteration method for solving fractional Riccati differential equa-
rier culture,” Biotechnology and Bioengineering, vol. 90, no. 1, pp. tion by Adomian polynomials,” Fractional Calculus and Applied
46–58, 2005. Analysis, vol. 16, no. 1, pp. 109–122, 2013.
[12] P. L. Bonate, Pharmacokinetic-Pharmacodynamic Modeling and [31] G.-C. Wu and D. Baleanu, “Variational iteration method for
Simulation, Springer, 2011. fractional calculus—a universal approach by Laplace trans-
[13] M. Holz and A. Fahr, “Compartment modeling,” Advanced Drug form,” Advances in Difference Equations, vol. 2013, article 18,
Delivery Reviews, vol. 48, no. 2-3, pp. 249–264, 2001. 2013.
16 Abstract and Applied Analysis

[32] Y. Hu, Y. Luo, and Z. Lu, “Analytical solution of the linear [49] A. H. Bhrawy and A. S. Alofi, “The operational matrix of frac-
fractional differential equation by Adomian decomposition tional integration for shifted Chebyshev polynomials,” Applied
method,” Journal of Computational and Applied Mathematics, Mathematics Letters, vol. 26, no. 1, pp. 25–31, 2013.
vol. 215, no. 1, pp. 220–229, 2008. [50] D. Baleanu, A. H. Bhrawy, and T. M. Taha, “Two efficient
[33] S. Saha Ray and R. K. Bera, “Analytical solution of the Bagley generalized Laguerre spectral algorithms for fractional initial
Torvik equation by Adomian decomposition method,” Applied value problems,” Abstract and Applied Analysis, vol. 2013, Article
Mathematics and Computation, vol. 168, no. 1, pp. 398–410, 2005. ID 546502, 10 pages, 2013.
[34] A. Arikoglu and I. Ozkol, “Solution of fractional differential
[51] D. Baleanu, A. H. Bhrawy, and T. M. Taha, “A modified
equations by using differential transform method,” Chaos,
generalized Laguerre spectral method for fractional differential
Solitons & Fractals, vol. 34, no. 5, pp. 1473–1481, 2007.
equations on the half line,” Abstract and Applied Analysis, vol.
[35] P. Darania and A. Ebadian, “A method for the numerical solu- 2013, Article ID 413529, 12 pages, 2013.
tion of the integro-differential equations,” Applied Mathematics
and Computation, vol. 188, no. 1, pp. 657–668, 2007. [52] A. H. Bhrawy and T. M. Taha, “An operational matrix of
fractional integration of the Laguerre polynomials and its
[36] I. Hashim, O. Abdulaziz, and S. Momani, “Homotopy analysis
application on a semi-infinite interval,” Mathematical Sciences,
method for fractional IVPs,” Communications in Nonlinear
vol. 6, article 41, 2012.
Science and Numerical Simulation, vol. 14, no. 3, pp. 674–684,
2009. [53] A. H. Bhrawy, M. M. Alghamdi, and T. M. Taha, “A new
[37] H. M. Jaradat, F. Awawdeh, and E. A. Rawashdeh, “An analytical modified generalized Laguerre operational matrix of fractional
scheme for multi-order fractional differential equations,” Tam- integration for solving fractional differential equations on the
sui Oxford Journal of Mathematical Sciences, vol. 26, no. 3, pp. half line,” Advances in Difference Equations, vol. 2012, article 179,
305–320, 2010. 2012.
[38] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, [54] D. Rostamy, M. Alipour, H. Jafari, and D. Baleanu, “Solving
Spectral Methods in Fluid Dynamics, Springer Series in Com- multi-term orders fractional differential equations by opera-
putational Physics, Springer, New York, NY, USA, 1988. tional matrices of BPs with convergence analysis,” Romanian
[39] B. Fornberg, A Practical Guide to Pseudospectral Methods, Reports in Physics, vol. 65, pp. 334–349, 2013.
Cambridge University Press, Cambridge, Mass, USA, 1998. [55] S. Yüzbaşı, “Numerical solutions of fractional Riccati type
[40] E. L. Ortiz and H. Samara, “Numerical solution of differential differential equations by means of the Bernstein polynomials,”
eigenvalue problems with an operational approach to the Tau Applied Mathematics and Computation, vol. 219, no. 11, pp.
method,” Computing, vol. 31, no. 2, pp. 95–103, 1983. 6328–6343, 2013.
[41] Y. Li and N. Sun, “Numerical solution of fractional differential [56] D. Baleanu, M. Alipour, and H. Jafari, “The Bernstein oper-
equations using the generalized block pulse operational matrix,” ational matrices for solving the fractional quadratic Riccati
Computers and Mathematics with Applications, vol. 62, no. 3, pp. differential equations with the Riemann-Liouville derivative,”
1046–1054, 2011. Abstract and Applied Analysis, vol. 2013, Article ID 461970, 7
[42] M. Yi, J. Huang, and J. Wei, “Block pulse operational matrix pages, 2013.
method for solving fractional partial differential equation,”
Applied Mathematics and Computation, vol. 221, pp. 121–131, [57] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A new
2013. Jacobi operational matrix: an application for solving fractional
differential equations,” Applied Mathematical Modelling, vol. 36,
[43] A. Saadatmandi and M. Dehghan, “A new operational matrix
no. 10, pp. 4931–4943, 2012.
for solving fractional-order differential equations,” Computers
and Mathematics with Applications, vol. 59, no. 3, pp. 1326–1336, [58] S. Kazem, “An integral operational matrix based on Jacobi
2010. polynomials for solving fractional-order differential equations,”
[44] S. Kazem, S. Abbasbandy, and S. Kumar, “Fractional-order Applied Mathematical Modelling, vol. 37, no. 3, pp. 1126–1136,
Legendre functions for solving fractional-order differential 2013.
equations,” Applied Mathematical Modelling, vol. 37, no. 7, pp. [59] E. H. Doha, A. H. Bhrawy, D. Baleanu, and S. S. Ezz-Eldien, “On
5498–5510, 2013. shifted Jacobi spectral approximations for solving fractional
[45] A. H. Bhrawy and M. M. Al-Shomrani, “A shifted Legen- differential equations,” Applied Mathematics and Computation,
dre spectral method for fractional-order multipoint boundary vol. 219, no. 15, pp. 8042–8056, 2013.
value problems,” Advances in Difference Equations, vol. 2012, [60] R. P. Agarwal, V. Lakshmikantham, and J. J. Nieto, “On the
article 8, 2012. concept of solution for fractional differential equations with
[46] A. H. Bhrawy, A. S. Alofi, and S. S. Ezz-Eldien, “A quadrature uncertainty,” Nonlinear Analysis: Theory, Methods & Applica-
tau method for fractional differential equations with variable tions, vol. 72, no. 6, pp. 2859–2862, 2010.
coefficients,” Applied Mathematics Letters, vol. 24, no. 12, pp.
[61] T. Allahviranloo, S. Salahshour, and S. Abbasbandy, “Explicit
2146–2152, 2011.
solutions of fractional differential equations with uncertainty,”
[47] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “Efficient Soft Computing, vol. 16, no. 2, pp. 297–302, 2012.
Chebyshev spectral methods for solving multi-term fractional
orders differential equations,” Applied Mathematical Modelling, [62] T. Allahviranloo, Z. Gouyandeh, and A. Armand, “Fuzzy
vol. 35, no. 12, pp. 5662–5672, 2011. fractionaldifferential equations under generalized fuzzy Caputo
[48] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A Chebyshev derivative,” Journal of Intelligent and Fuzzy Systems. In press.
spectral method based on operational matrix for initial and [63] S. Salahshour, T. Allahviranloo, S. Abbasbandy, and D. Baleanu,
boundary value problems of fractional order,” Computers and “Existence and uniqueness results for fractional differential
Mathematics with Applications, vol. 62, no. 5, pp. 2364–2373, equations with uncertainty,” Advances in Difference Equations,
2011. vol. 2012, article 112, 2012.
Abstract and Applied Analysis 17

[64] S. Salahshour, T. Allahviranloo, and S. Abbasbandy, “Solving


fuzzy fractional differential equations by fuzzy Laplace trans-
forms,” Communications in Nonlinear Science and Numerical
Simulation, vol. 17, no. 3, pp. 1372–1381, 2012.
[65] A. Ahmadian, M. Suleiman, S. Salahshour, and D. Baleanu, “A
Jacobi operational matrix for solving a fuzzy linear fractional
differential equation,” Advances in Difference Equations, vol.
2013, article 104, 2013.
[66] A. Ahmadian, M. Suleiman, and S. Salahshour, “An opera-
tional matrix based on Legendre polynomials for solving fuzzy
fractional-order differential equations,” Abstract and Applied
Analysis, vol. 2013, Article ID 505903, 29 pages, 2013.
[67] M. R. Balooch Shahriyar, F. Ismail, S. Aghabeigi, A. Ahma-
dian, and S. Salahshour, “An eigenvalue-eigenvector method
for solving a system of fractional differential equations with
uncertainty,” Mathematical Problems in Engineering, vol. 2013,
Article ID 579761, 11 pages, 2013.
[68] F. Ghaemi, R. Yunus, A. Ahmadian, S. Salahshour, M. Suleiman,
and Sh. Faridah Saleh, “Application of fuzzy fractional kinetic
equations to modelling of the acid hydrolysis reaction,” Abstract
and Applied Analysis, vol. 2013, Article ID 610314, 19 pages, 2013.
[69] M. Mazandarani and A. Vahidian Kamyad, “Modified fractional
Euler method for solving fuzzy fractional initial value problem,”
Communications in Nonlinear Science and Numerical Simula-
tion, vol. 18, no. 1, pp. 12–21, 2013.
[70] D. Dubois and H. Prade, “Towards fuzzy differential calculus—
part 3: differentiation,” Fuzzy Sets and Systems, vol. 8, no. 3, pp.
225–233, 1982.
[71] R. Goetschel Jr. and W. Voxman, “Elementary fuzzy calculus,”
Fuzzy Sets and Systems, vol. 18, no. 1, pp. 31–43, 1986.
[72] G. A. Anastassiou, Fuzzy Mathematics: Approximation Theory,
vol. 251 of Studies in Fuzziness and Soft Computing, Springer,
Berlin, Germany, 2010.
[73] G. A. Anastassiou and S. G. Gal, “On a fuzzy trigonometric
approximation theorem of Weierstrass-type,” Journal of Fuzzy
Mathematics, vol. 9, no. 3, pp. 701–708, 2001.
[74] O. Kaleva, “Fuzzy differential equations,” Fuzzy Sets and Sys-
tems, vol. 24, no. 3, pp. 301–317, 1987.
[75] Y. Chalco-Cano and H. Román-Flores, “On new solutions of
fuzzy differential equations,” Chaos, Solitons & Fractals, vol. 38,
no. 1, pp. 112–119, 2008.
[76] T. Allahviranloo and M. Afshar Kermani, “Solution of a fuzzy
system of linear equation,” Applied Mathematics and Computa-
tion, vol. 175, no. 1, pp. 519–531, 2006.
[77] S. Abbasbandy and R. Ezzati, “Newton’s method for solving a
system of fuzzy nonlinear equations,” Applied Mathematics and
Computation, vol. 175, no. 2, pp. 1189–1199, 2006.
[78] Z. M. Odibat and N. T. Shawagfeh, “Generalized Taylor’s
formula,” Applied Mathematics and Computation, vol. 186, no.
1, pp. 286–293, 2007.
[79] Y. Luke, The Special Functions and Their Approximations, vol. 2,
Academic Press, New York, NY, USA, 1969.
[80] G. Szegö, Orthogonal Polynomials, vol. 23 of Colloquium Publi-
cations, American Mathematical Society, Providence, RI, USA,
1985.
[81] A. A. Kilbas and J. J. Trujillo, “Differential equations of
fractional order: methods, results and problems. I,” Applicable
Analysis, vol. 78, no. 1-2, pp. 153–192, 2001.
[82] K. S. Miller and B. Ross, An Introduction to the Fractional Calcu-
lus and Fractional Differential Equations, A Wiley-Interscience
Publication, John Wiley & Sons, New York, NY, USA, 1993.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 465160, 7 pages
http://dx.doi.org/10.1155/2013/465160

Research Article
Numerical Solutions of Fractional Fokker-Planck Equations
Using Iterative Laplace Transform Method

Limei Yan
School of Mathematical Sciences, Dezhou University, Dezhou 253023, China

Correspondence should be addressed to Limei Yan; yanlimei2013@163.com

Received 3 August 2013; Revised 17 October 2013; Accepted 2 November 2013

Academic Editor: Dumitru Baleanu

Copyright © 2013 Limei Yan. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A relatively new iterative Laplace transform method, which combines two methods; the iterative method and the Laplace transform
method, is applied to obtain the numerical solutions of fractional Fokker-Planck equations. The method gives numerical solutions in
the form of convergent series with easily computable components, requiring no linearization or small perturbation. The numerical
results show that the approach is easy to implement and straightforward when applied to space-time fractional Fokker-Planck
equations. The method provides a promising tool for solving space-time fractional partial differential equations.

1. Introduction and symbolic computation. By using this method, Jafari and


Seifi successfully obtained the numerical solutions of two
Fractional calculus has attracted much attention for its systems of space-time fractional differential equations. It has
potential applications in various scientific fields such as fluid been shown that, with this method, one can discover some
mechanics, biology, viscoelasticity, engineering, and other solutions found by the existing methods such as homo-
areas of science [1–4]. So it becomes important to find some topy perturbation method, Laplace Adomian decomposition
efficient methods for solving fractional differential equations. method, and variational iterative method [23].
A great deal of effort has been spent on constructing of It is well known that the choice of an appropriate ansatz
the numerical solutions and many effective methods have is of great importance when a method is applied to search
been developed such as fractional wavelet method [5–8], for numerical solutions of nonlinear partial differential equa-
fractional differential transform method [9], fractional oper- tions. In the present paper, we will use the iterative Laplace
ational matrix method [10, 11], fractional improved homotopy transform method to solve space-time fractional Fokker-
perturbation method [12, 13], fractional variational iteration Planck equations. The fractional derivatives described here
method [14, 15], and fractional Laplace Adomian decomposi- are in the Caputo sense.
tion method [16, 17]. Fokker-Planck equation has been applied in various nat-
In 2006, Daftardar-Gejji and Jafari proposed a new ural science fields such as quantum optics, solid-state physics,
iterative method to seek numerical solutions of nonlinear chemical physics, theoretical biology, and circuit theory. It
functional equations [18, 19]. By now, the iterative method is firstly proposed by Fokker and Planck to characterize
has been used to solve many nonlinear differential equations the Brownian motion of particles [24]. The general form of
of integer and fractional order [20] and fractional boundary Fokker-Planck equation is as follows:
value problem [21]. Most recently, Jafari et al. firstly applied
Laplace transform in the iterative method and proposed a 𝜕𝑢 𝜕 𝜕2
= [− 𝐴 (𝑥) + 2 𝐵 (𝑥)] 𝑢 (𝑥, 𝑡) (1)
new direct method called iterative Laplace transform method 𝜕𝑡 𝜕𝑥 𝜕𝑥
[22] to search for numerical solutions of a system of fractional with the initial condition
partial differential equations. The method is based on Laplace
transform, iterative method, Caputo fractional derivative, 𝑢 (𝑥, 0) = 𝑓 (𝑥) , (2)
2 Abstract and Applied Analysis

where 𝐴(𝑥), 𝐵(𝑥) > 0 are called diffusion coefficient and drift partial differential equation with initial conditions of the
coefficient, respectively. If 𝐴(𝑥), 𝐵(𝑥) > 0 depend on 𝑢(𝑥, 𝑡) form
and the time 𝑡, then (1) becomes the following generalized
nonlinear form [25]: 𝐷𝑡𝛼 𝑢 = A (𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .) , 𝑚 − 1 < 𝛼 ≤ 𝑚,
2 (8)
𝜕𝑢 𝜕 𝜕 𝑛 − 1 < 𝛽 ≤ 𝑛, 𝑚, 𝑛 ∈ 𝑁,
= [− 𝐴 (𝑥, 𝑡, 𝑢) + 2 𝐵 (𝑥, 𝑡, 𝑢)] 𝑢 (𝑥, 𝑡) . (3)
𝜕𝑡 𝜕𝑥 𝜕𝑥

The space-time fractional Fokker-Planck is as follows with initial value conditions


[26]:
𝑢(𝑘) (𝑥, 0) = ℎ𝑘 (𝑥) , 𝑘 = 0, 1, . . . , 𝑚 − 1, (9)
𝐷𝑡𝛼 𝑢 = [−𝐷𝑥𝛽 𝐴 (𝑥, 𝑡, 𝑢) + 𝐷𝑥2𝛽 𝐵 (𝑥, 𝑡, 𝑢)] 𝑢 (𝑥, 𝑡) (4)

which is the generalized fractional form of (3). Here where A(𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .) is a linear or nonlinear operator
𝐷𝑡𝛼 (⋅), 𝐷𝑥𝛽 (⋅), 𝐷𝑥2𝛽 (⋅) are the Caputo fractional derivative with of 𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . ., and 𝑢 = 𝑢(𝑥, 𝑡) is the unknown function
respect to 𝑡 and 𝑥 defined in Section 2. When 𝛼 = 𝛽 = 1, (4) that will be determined later.
reduces to (3). Taking Laplace transfer of both sides of (8) results in
The rest of this paper is organized as follows. In Section 2,
we review some basic definitions of Caputo fractional deriva-
𝑚−1
tive and Laplace transform. In Section 3, we describe the
𝑠𝛼 L [𝑢 (𝑥, 𝑡)] − ∑ 𝑠𝛼−1−𝑘 𝑢(𝑘) (𝑥, 0)
iterative Laplace transform method for solving fractional
𝑘=0 (10)
partial differential equations. In Section 4, we give three
applications of the method to Fokker-Planck equations. In = L [A (𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .)] .
Section 5, some conclusions and discussions are given.

Equivalently,
2. Preliminaries
Definition 1. The Caputo fractional derivative [27, 28] of 𝑚−1
function 𝑢(𝑥, 𝑡) is defined as L [𝑢 (𝑥, 𝑡)] = ∑ 𝑠−1−𝑘 𝑢(𝑘) (𝑥, 0)
𝑘=0 (11)
𝑡
1 𝑚−𝛼−1 (𝑚)
𝐷𝑡𝛼 𝑢 (𝑥, 𝑡) = ∫ (𝑡 − 𝜂) 𝑢 (𝑥, 𝜂) 𝑑𝜂, +𝑠 −𝛼
L [A (𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .)] .
Γ (𝑚 − 𝛼) 0 (5)
𝑚 − 1 < 𝛼 ≤ 𝑚, 𝑚 ∈ 𝑁,
Operating with Laplace inverse (denoted by L−1
where Γ(⋅) denotes the gamma function. throughout the present paper) on both sides of (11) gives

Definition 2. The Laplace transform of 𝑓(𝑡) is defined as [27, 𝑚−1


28] 𝑢 (𝑥, 𝑡) = L−1 [ ∑ 𝑠−1−𝑘 𝑢(𝑘) (𝑥, 0)]
∞ 𝑘=0 (12)
𝐹 (𝑠) = L [𝑓 (𝑡)] = ∫ 𝑒−𝑠𝑡 𝑓 (𝑡) 𝑑𝑡. (6) −1 −𝛼
0 + L [𝑠 L [A (𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .)]] ,
Definition 3. Laplace transform of 𝐷𝑡𝛼 𝑢(𝑥, 𝑡) is given as [27,
28] which can be rewritten as the form
𝑚−1 𝑚−1
L [𝐷𝑡𝛼 𝑢 (𝑥, 𝑡)] = 𝑠𝛼 L [𝑢 (𝑥, 𝑡)] − ∑ 𝑢(𝑘) (𝑥, 0) 𝑠𝛼−1−𝑘 , 𝑢 (𝑥, 𝑡) = L−1 [ ∑ 𝑠−1−𝑘 𝑢(𝑘) (𝑥, 0)]
𝑘=0 (7) 𝑘=0 (13)
𝑚 − 1 < 𝛼 ≤ 𝑚, + B (𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .) ,
where 𝑢(𝑘) (𝑥, 0) is the 𝑘-order derivative of 𝑢(𝑥, 𝑡) at 𝑡 = 0.
where B(𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .) = L−1 [𝑠−𝛼 L[A(𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢,
Further information about fractional derivative and its . . .)]].
properties can be found in [27–29]. The iterative Laplace transform method represents the
solution as an infinite series:
3. The Iterative Laplace Transform Method

To illustrate the basic idea of the iterative Laplace transform 𝑢 (𝑥, 𝑡) = ∑ 𝑢𝑛 , (14)
method [22], we consider the general space-time fractional 𝑛=0
Abstract and Applied Analysis 3

where the terms 𝑢𝑛 are to be recursively computed. The Example 1. Consider the Fokker-Planck equation in the case
linear or nonlinear operator B(𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .) can be that [30]
decomposed as follows:
𝜕𝑢 𝜕2 𝑢
∞ ∞ ∞ 𝐷𝑡𝛼 𝑢 = + , 0 < 𝛼 ≤ 1, (19)
𝜕𝑥 𝜕𝑥2
B ( ∑ 𝑢𝑛 , 𝐷𝑥𝛽 ∑ 𝑢𝑛 , 𝐷𝑥2𝛽 ∑ 𝑢𝑛 , . . .)
𝑛=0 𝑛=0 𝑛=0 subject to
= B (𝑢0 , 𝐷𝑥𝛽 𝑢0 , 𝐷𝑥2𝛽 𝑢0 , . . .) 𝑢 (𝑥, 0) = 𝑥. (20)
∞ 𝑗 𝑗 𝑗 (15) Taking Laplace transform on both sides of (19) gives
+ ∑ B ( ∑ 𝑢𝑘 , 𝐷𝑥𝛽 ∑ 𝑢𝑘 , 𝐷𝑥2𝛽 ∑ 𝑢𝑘 , . . .)
𝑗=1 𝑘=0 𝑘=0 𝑘=0 𝜕𝑢 𝜕2 𝑢
𝑠𝛼 L [𝑢 (𝑥, 𝑡)] − 𝑠𝛼−1 𝑢 (𝑥, 0) = L [ + ], (21)
∞ 𝑗−1 𝑗−1 𝑗−1 𝜕𝑥 𝜕𝑥2
− ∑ B ( ∑ 𝑢𝑘 , 𝐷𝑥𝛽 ∑ 𝑢𝑘 , 𝐷𝑥2𝛽 ∑ 𝑢𝑘 , . . .) .
𝑗=1 𝑘=0 𝑘=0 𝑘=0 𝑥 1 𝜕𝑢 𝜕2 𝑢
L [𝑢 (𝑥, 𝑡)] = + 𝛼L[ + ]. (22)
𝑠 𝑠 𝜕𝑥 𝜕𝑥2
Substituting (14) and (15) into (13) yields
∞ 𝑚−1
Operating with Laplace inverse on both sides of (22)
−1 −1−𝑘 (𝑘) results in
∑ 𝑢𝑛 = L [ ∑ 𝑠 𝑢 (𝑥, 0)]
𝑛=0 𝑘=0
1 𝜕𝑢 𝜕2 𝑢
𝑢 (𝑥, 𝑡) = 𝑥 + L−1 [ L [ + ]] . (23)
+ B (𝑢0 , 𝐷𝑥𝛽 𝑢0 , 𝐷𝑥2𝛽 𝑢0 , . . .) 𝑠𝛼 𝜕𝑥 𝜕𝑥2

∞ 𝑗 𝑗 𝑗 Substituting (14) and (15) into (23) and applying (17), we


+ ∑ [B ( ∑ 𝑢𝑘 , 𝐷𝑥𝛽 ∑ 𝑢𝑘 , 𝐷𝑥2𝛽 ∑ 𝑢𝑘 , . . .) obtain the components of the solution as follows:
𝑗=1 𝑘=0 𝑘=0 𝑘=0
𝑢0 (𝑥, 𝑡) = 𝑢 (𝑥, 0) = 𝑥,
𝑗−1 𝑗−1 𝑗−1
−B ( ∑ 𝑢𝑘 , 𝐷𝑥𝛽 ∑ 𝑢𝑘 , 𝐷𝑥2𝛽 ∑ 𝑢𝑘 , . . .)] . 1 𝜕𝑢 𝜕2 𝑢0 𝑡𝛼
𝑘=0 𝑘=0 𝑘=0
𝑢1 (𝑥, 𝑡) = L−1 [ 𝛼
L[ 0 + 2
]] = ,
𝑠 𝜕𝑥 𝜕𝑥 Γ (1 + 𝛼) (24)
(16)
𝑢2 (𝑥, 𝑡) = 0,
We set
𝑚−1 𝑢3 (𝑥, 𝑡) = ⋅ ⋅ ⋅ = 𝑢𝑛 (𝑥, 𝑡) = ⋅ ⋅ ⋅ = 0.
𝑢0 = L−1 [ ∑ 𝑠−1−𝑘 𝑢(𝑘) (𝑥, 0)] ,
𝑘=0 Therefore, the solution of (19)-(20) in a closed form can
be obtained as follows:
𝑢1 = B (𝑢0 , 𝐷𝑥𝛽 𝑢0 , 𝐷𝑥2𝛽 𝑢0 , . . .) ,
𝑡𝛼
𝑢 (𝑥, 𝑡) = 𝑥 + . (25)
𝑚 𝑚 𝑚 Γ (1 + 𝛼)
𝑢𝑚+1 = B ( ∑ 𝑢𝑘 , 𝐷𝑥𝛽 ∑ 𝑢𝑘 , 𝐷𝑥2𝛽 ∑ 𝑢𝑘 , . . .)
𝑘=0 𝑘=0 𝑘=0 If we take 𝛼 = 1, then (25) can be reduced to
𝑚−1 𝑚−1 𝑚−1
𝑢 (𝑥, 𝑡) = 𝑥 + 𝑡, (26)
− B ( ∑ 𝑢𝑘 , 𝐷𝑥𝛽 ∑ 𝑢𝑘 , 𝐷𝑥2𝛽 ∑ 𝑢𝑘 , . . .) , 𝑚 ≥ 1.
𝑘=0 𝑘=0 𝑘=0 which is exactly the same as that obtained by homotopy
(17) perturbation method in [30].
Therefore the 𝑚-term numerical solution of (8)-(9) is It should be pointed out that the iterative Laplace trans-
given by form method is the generalization algorithm of iterative
method proposed by Daftardar-Gejji and Jafari [18]. When
𝑢 (𝑥, 𝑡) ≅ 𝑢0 (𝑥, 𝑡) + 𝑢1 (𝑥, 𝑡) + ⋅ ⋅ ⋅ + 𝑢𝑚 (𝑥, 𝑡) , these two methods are used to solve differential equations
(18) with integer order derivatives, especially for linear cases, they
𝑚 = 1, 2, . . . . are not different from each other.

4. Numerical Solutions of Fractional Example 2. Consider the following space-time fractional


Fokker-Planck Equations Fokker-Planck equation with initial value conditions [26]:

𝑢𝑥 𝑢𝑥2
The iterative Laplace transform method, described in 𝐷𝑡𝛼 𝑢 = −𝐷𝑥𝛽 ( ) + 𝐷𝑥2𝛽 ( ),
Section 3, will be applied to solve three special cases of 6 12 (27)
space-time fractional Fokker-Planck equations with initial
conditions. 𝑡 > 0, 𝑥 > 0, 0 < 𝛼, 𝛽 ≤ 1,
4 Abstract and Applied Analysis

subject to The solution in series form is then given by

𝑢 (𝑥, 𝑡)

𝑢 (𝑥, 0) = 𝑥2 . (28) 1 2 𝑡𝛼
= 𝑥2 + (− 𝑥3−𝛽 + 𝑥4−2𝛽 ) ⋅
Γ (4 − 𝛽) Γ (5 − 2𝛽) Γ (1 + 𝛼)

4−𝛽 5 − 3𝛽
Taking Laplace transform on both sides of (27) gives +[ 𝑥4−2𝛽 − 𝑥5−3𝛽
6Γ (5 − 2𝛽) 3Γ (6 − 3𝛽)

(5 − 𝛽) (4 − 𝛽) 5−3𝛽 (6 − 2𝛽) (5 − 2𝛽) 6−4𝛽


𝛼 𝛼−1 − 𝑥 + 𝑥 ]
𝑠 L [𝑢 (𝑥, 𝑡)] − 𝑠 𝑢 (𝑥, 0) 12Γ (6 − 3𝛽) 6Γ (7 − 4𝛽)
𝑢𝑥2 (29)
𝑢𝑥 𝑡2𝛼
= L [−𝐷𝑥𝛽 ( ) + 𝐷𝑥2𝛽 ( )] , ⋅ + ⋅⋅⋅ .
6 12 Γ (1 + 2𝛼)
L [𝑢 (𝑥, 𝑡)]
(33)
2 (30)
𝑢 (𝑥, 0) 1 𝑢𝑥 𝑢𝑥
= + 𝛼 L [−𝐷𝑥𝛽 ( ) + 𝐷𝑥2𝛽 ( )] . Setting 𝛼 = 𝛽 = 1 in (33), we get the solution of the
𝑠 𝑠 6 12 problem by

𝑡 (𝑡/2)2
𝑢 (𝑥, 𝑡) = 𝑥2 (1 + + + ⋅⋅⋅) (34)
Operating with Laplace inverse on both sides of (30), we 2 2!
obtain the following Laplace equation:
and in a closed form by

𝑢 (𝑥, 𝑡) = 𝑥2 𝑒𝑡/2 (35)


𝑢 (𝑥, 𝑡)
which is in full agreement with the results by homotopy
1 𝑢𝑥 𝑢𝑥2
= 𝑢 (𝑥, 0) + L [ 𝛼 L [−𝐷𝑥𝛽 ( ) + 𝐷𝑥2𝛽 (
−1
)]] . perturbation method in [26].
𝑠 6 12
(31) Example 3. Consider the following space-time fractional
nonlinear initial value problem that describes Fokker-Planck
equation [12]:
Following the algorithm given in (17), the first three compo- 4𝑢2 𝑥𝑢
nents of the solution are as follows: 𝐷𝑡𝛼 𝑢 = −𝐷𝑥𝛽 ( − ) + 𝐷𝑥2𝛽 𝑢2 , 0 < 𝛼, 𝛽 ≤ 1, (36)
𝑥 3

subject to
𝑢0 (𝑥, 𝑡) = 𝑢 (𝑥, 0) = 𝑥2 ,
𝑢 (𝑥, 0) = 𝑥2 . (37)
𝑢1 (𝑥, 𝑡)
Taking Laplace transform on both sides of (35) gives
1 𝑢𝑥 𝑢 𝑥2
= L [ 𝛼 L [−𝐷𝑥𝛽 ( 0 ) + 𝐷𝑥2𝛽 ( 0 )]]
−1
𝑠𝛼 L [𝑢 (𝑥, 𝑡)] − 𝑠𝛼−1 𝑢 (𝑥, 0)
𝑠 6 12
4𝑢2 𝑥𝑢 (38)
1 2 𝑡𝛼 = L [−𝐷𝑥𝛽 ( − ) + 𝐷𝑥2𝛽 𝑢2 ] ,
= (− 𝑥3−𝛽 + 𝑥4−2𝛽 ) ⋅ , 𝑥 3
Γ (4 − 𝛽) Γ (5 − 2𝛽) Γ (1 + 𝛼)

𝑢2 (𝑥, 𝑡) (32) 𝑥2 1 4𝑢2 𝑥𝑢


L [𝑢 (𝑥, 𝑡)] = + 𝛼 L [−𝐷𝑥𝛽 ( − ) + 𝐷𝑥2𝛽 𝑢2 ] .
𝑠 𝑠 𝑥 3
4−𝛽 5 − 3𝛽 (39)
=[ 𝑥4−2𝛽 − 𝑥5−3𝛽
6Γ (5 − 2𝛽) 3Γ (6 − 3𝛽)
Operating with Laplace inverse on both sides of (38), we
(5 − 𝛽) (4 − 𝛽) 5−3𝛽 (6 − 2𝛽) (5 − 2𝛽) obtain the following Laplace equation:
− 𝑥 +
12Γ (6 − 3𝛽) 6Γ (7 − 4𝛽) 2
1 𝛽 4𝑢 𝑥𝑢
2𝛼 𝑢 (𝑥, 𝑡) = 𝑥2 + L−1 [ L [−𝐷𝑥 ( − ) + 𝐷𝑥2𝛽 𝑢2 ]] .
𝑡 𝑠𝛼 𝑥 3
× 𝑥6−4𝛽 ] ⋅ .
Γ (1 + 2𝛼) (40)
Abstract and Applied Analysis 5

Table 1: Several approximate values and exact solutions for (36) and (37) when 𝛼 = 𝛽 = 1.

𝑥 𝑡 SolutionILTM SolutionHPM SolutionADM Exact solution


0.1 0.0690729 0.0690729 0.0690729 0.0690732
0.2 0.0763333 0.0763333 0.0763333 0.0763377
0.25
0.4 0.0931667 0.0931667 0.0931667 0.093239
0.8 0.137833 0.137833 0.137833 0.139096
0.1 0.276292 0.276292 0.276292 0.276293
0.2 0.305333 0.305333 0.305333 0.305351
0.5
0.4 0.372667 0.372667 0.372667 0.372956
0.8 0.551333 0.551333 0.551333 0.556385
0.1 0.621656 0.621656 0.621656 0.621659
0.2 0.687 0.687 0.687 0.687039
0.75
0.4 8385 0.8385 0.8385 0.839151
0.8 1.2405 1.2405 1.2405 1.25187

Following the algorithm given in (17), the first few 506Γ (5 − 𝛽)


+ ⋅ 𝑡2𝛼 𝑥4−2𝛽
components of the solution are as follows: 3Γ (1 + 2𝛼) Γ (5 − 2𝛽) Γ (4 − 𝛽)
𝑢0 (𝑥, 𝑡) = 𝑥2 , 1936Γ (1 + 2𝛼) Γ (6 − 2𝛽)
− ⋅ 𝑡3𝛼 𝑥5−3𝛽
Γ2 (1 + 𝛼) Γ (1 + 3𝛼) Γ (6 − 3𝛽) Γ2 (4 − 𝛽)
1 4𝑢2 𝑥𝑢
𝑢1 (𝑥, 𝑡) = L−1 [ L [−𝐷𝑥𝛽 ( 0 − 0 ) + 𝐷𝑥2𝛽 𝑢02 ]]
𝑠𝛼 𝑥 3 576Γ (1 + 2𝛼) Γ (9 − 4𝛽)
+ ⋅ 𝑡3𝛼 𝑥8−6𝛽
Γ2 (1 + 𝛼) Γ (1 + 3𝛼) Γ (8 − 5𝛽) Γ2 (5 − 2𝛽)
22 24
= (− 𝑥3−𝛽 + 𝑥4−2𝛽 )
Γ (4 − 𝛽) Γ (5 − 2𝛽) 48Γ (7 − 2𝛽)
+ ⋅ 𝑡2𝛼 𝑥6−4𝛽
Γ (1 + 2𝛼) Γ (7 − 4𝛽) Γ (5 − 2𝛽)
𝑡𝛼
⋅ ,
Γ (1 + 𝛼) 22 24 𝑡𝛼
+( 𝑥3−𝛽 − 𝑥4−2𝛽 ) ⋅ ,
𝑢2 (𝑥, 𝑡) Γ (4 − 𝛽) Γ (5 − 2𝛽) Γ (1 + 𝛼)

2 ..
1 4(𝑢0 + 𝑢1 ) 𝑥 (𝑢0 + 𝑢1 ) .
= L−1 [ 𝛼
L [−𝐷𝑥𝛽 ( − )
𝑠 𝑥 3 (41)

2
The solution in series form is then given by
+𝐷𝑥2𝛽 (𝑢0 + 𝑢1 ) ]]
𝑢 (𝑥, 𝑡) = 𝑢0 (𝑥, 𝑡) + 𝑢1 (𝑥, 𝑡) + 𝑢2 (𝑥, 𝑡) + ⋅ ⋅ ⋅ . (42)
1 𝑥𝑢 4𝑢02 If we take 𝛼 = 𝛽 = 1, the first few components of the solution
− L−1 [ 𝛼
L [−𝐷𝑥𝛽 ( − 0 ) + 𝐷𝑥2𝛽 𝑢02 ]]
𝑠 𝑥 3 are as follows:
Γ (1 + 2𝛼) Γ (8 − 4𝛽) 𝑢0 (𝑥, 𝑡) = 𝑥2 ,
=−
Γ2 (1 + 𝛼) Γ (1 + 3𝛼) Γ (8 − 5𝛽) Γ (5 − 2𝛽)
𝑢1 (𝑥, 𝑡) = 𝑥2 𝑡,
2304 1056
⋅( + ) 𝑡3𝛼 𝑥7−5𝛽 𝑥2 𝑡2 (43)
Γ (5 − 2𝛽) Γ (4 − 𝛽) 𝑢2 (𝑥, 𝑡) = ,
2
1 184Γ (6 − 2𝛽) 44Γ (6 − 𝛽)
− ( + ) ..
Γ (1 + 2𝛼) Γ (6 − 3𝛽) Γ (5 − 2𝛽) Γ (4 − 𝛽) .

⋅ 𝑡2𝛼 𝑥5−3𝛽 For this special case, the exact solution of (36) and (37) is
therefore given by
Γ (1 + 2𝛼)
+
Γ2 (1 + 𝛼) Γ (1 + 3𝛼) Γ (7 − 4𝛽) Γ (4 − 𝛽) 𝑢 (𝑥, 𝑡) = 𝑥2 𝑒𝑡 (44)
484Γ (7 − 2𝛽) 4224Γ (7 − 3𝛽) which is exactly the result obtained by homotopy perturba-
×( + ) ⋅ 𝑡3𝛼 𝑥6−4𝛽
Γ (4 − 𝛽) Γ (5 − 2𝛽) tion transformation method in [12].
6 Abstract and Applied Analysis

Table 1 shows the numerical solutions for (36) and (37) [7] J. L. Wu, “A wavelet operational method for solving fractional
by using iterative Laplace transform method, homotopy partial differential equations numerically,” Applied Mathematics
perturbation transform method, Adomian decomposition and Computation, vol. 214, no. 1, pp. 31–40, 2009.
method, and the exact solution as 𝛼 = 𝛽 = 1. It should be [8] Ü. Lepik, “Solving fractional integral equations by the Haar
pointed out that only three terms of these methods are used wavelet method,” Applied Mathematics and Computation, vol.
to evaluate the numerical solutions in Table 1. It is obvious 214, no. 2, pp. 468–478, 2009.
that the iterative Laplace transform method used in the [9] V. S. Ertürk and S. Momani, “Solving systems of fractional
present paper has the same convergence as the convergence differential equations using differential transform method,”
of homotopy perturbation transform method and Adomian Journal of Computational and Applied Mathematics, vol. 215, no.
decomposition method for solving this fractional nonlinear 1, pp. 142–151, 2008.
Fokker-Planck equation. Therefore, iterative Laplace trans- [10] Y. Li and N. Sun, “Numerical solution of fractional differential
form method is an effective method for solving fractional equations using the generalized block pulse operational matrix,”
partial differential equations just as homotopy perturbation Computers & Mathematics with Applications, vol. 62, no. 3, pp.
transform method and Adomian decomposition method. 1046–1054, 2011.
[11] A. Saadatmandi and M. Dehghan, “A new operational matrix
for solving fractional-order differential equations,” Computers
5. Conclusions & Mathematics with Applications, vol. 59, no. 3, pp. 1326–1336,
2010.
With the aid of the symbolic computation system Math-
[12] Y. Liu, “Approximate solutions of fractional nonlinear equa-
ematica, the iterative Laplace transform method is first tions using homotopy perturbation transformation method,”
successfully applied to solve fractional Fokker-Planck equa- Abstract and Applied Analysis, vol. 2012, Article ID 752869, 14
tions. The results obtained by the iterative Laplace transform pages, 2012.
method are the same as those obtained by homotopy per- [13] Y. Liu, “Variational homotopy perturbation method for solv-
turbation transform method and Adomian decomposition ing fractional initial boundary value problems,” Abstract and
method. The method finds the solutions without unneces- Applied Analysis, vol. 2012, Article ID 727031, 10 pages, 2012.
sary linearization, small perturbation and other restrictive [14] N. H. Sweilam, M. M. Khader, and R. F. Al-Bar, “Numeri-
assumptions. Therefore, the method considerably reduces the cal studies for a multi-order fractional differential equation,”
computational work to a great extent. It is worth mentioning Physics Letters A, vol. 371, no. 1-2, pp. 26–33, 2007.
that the method can also be applied to solve other nonlinear [15] S. Das, “Analytical solution of a fractional diffusion equation by
fractional differential equations with initial value conditions. variational iteration method,” Computers & Mathematics with
Applications, vol. 57, no. 3, pp. 483–487, 2009.
Acknowledgments [16] H. Jafari, C. M. Khalique, and M. Nazari, “Application of the
Laplace decomposition method for solving linear and nonlin-
The author would like to express her sincere thanks to ear fractional diffusion-wave equations,” Applied Mathematics
the referees for their valuable suggestions which lead to an Letters, vol. 24, no. 11, pp. 1799–1805, 2011.
improved version. This research is supported by Excellent [17] M. Y. Ongun, “The Laplace adomian decomposition method for
Young Scientist Foundation of Shandong Province under solving a model for HIV infection of CD4+ cells,” Mathematical
Grant no. BS2013HZ026. and Computer Modelling, vol. 53, no. 5-6, pp. 597–603, 2011.
[18] V. Daftardar-Gejji and H. Jafari, “An iterative method for solving
nonlinear functional equations,” Journal of Mathematical Anal-
References ysis and Applications, vol. 316, no. 2, pp. 753–763, 2006.
[1] J. Sabatier, O. P. Agrawal, and J. A. Tenreiro Machado, Advances [19] H. Jafari, Iterative Methods for solving system of fractional
in Fractional Calculus: Theoretical Developments and Applica- differential equations [Ph.D. thesis], Pune University, 2006.
tions in Physics and Engineering, Springer, 2007. [20] S. Bhalekar and V. Daftardar-Gejji, “Solving evolution equations
[2] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional using a new iterative method,” Numerical Methods for Partial
Calculus, vol. 3 of Series on Complexity, Nonlinearity and Chaos, Differential Equations, vol. 26, no. 4, pp. 906–916, 2010.
World Scientific, Singapore, 2012. [21] V. Daftardar-Gejji and S. Bhalekar, “Solving fractional boundary
[3] Y. Liu and B. Xin, “Numerical solutions of a fractional predator- value problems with Dirichlet boundary conditions using a new
prey system,” Advances in Difference Equations, vol. 2011, Article iterative method,” Computers & Mathematics with Applications,
ID 190475, 11 pages, 2011. vol. 59, no. 5, pp. 1801–1809, 2010.
[4] J. Ma and Y. Liu, “Exact solutions for a generalized nonlinear [22] H. Jafari, M. Nazari, D. Baleanu, and C. M. Khalique, “A new
fractional Fokker-Planck equation,” Nonlinear Analysis: Real approach for solving a system of fractional partial differential
World Applications, vol. 11, no. 1, pp. 515–521, 2010. equations,” Computers & Mathematics with Applications, vol. 66,
[5] M. Rehman and R. Ali Khan, “The Legendre wavelet method no. 5, pp. 838–843, 2013.
for solving fractional differential equations,” Communications in [23] H. Jafari and S. Seifi, “Solving a system of nonlinear fractional
Nonlinear Science and Numerical Simulation, vol. 16, no. 11, pp. partial differential equations using homotopy analysis method,”
4163–4173, 2011. Communications in Nonlinear Science and Numerical Simula-
[6] Y. Li, “Solving a nonlinear fractional differential equation using tion, vol. 14, no. 5, pp. 1962–1969, 2009.
Chebyshev wavelets,” Communications in Nonlinear Science and [24] H. Risken, The Fokker-Planck Equation, Methods of Solution and
Numerical Simulation, vol. 15, no. 9, pp. 2284–2292, 2010. Applications, Springer, Berlin, Germany, 1989.
Abstract and Applied Analysis 7

[25] T. D. Frank, “Stochastic feedback, nonlinear families of Markov


processes, and nonlinear Fokker-Planck equations,” Physica A,
vol. 331, no. 3-4, pp. 391–408, 2004.
[26] A. Yiidirim, “Analytical approach to Fokker-Planck equation
with space- and time fractional derivatives by means of the
homotopy perturbation method,” Journal of King Saud Univer-
sity, vol. 22, no. 4, pp. 257–264, 2010.
[27] K. S. Miller and B. Ross, An Introduction to the Fractional
Calculus and Fractional Differential Equations, John Wiley &
Sons, New York, NY, USA, 1993.
[28] I. Podlubny, Fractional Differential Equations, vol. 198, Aca-
demic Press, New York, NY, USA, 1999.
[29] R. Hilfer, Applications of Fractional Calculus in Physics, World
Scientific, Singapore, 2000.
[30] J. Biazar, K. Hosseini, and P. Gholamin, “Homotopy perturba-
tion method Fokker-Planck equation,” International Mathemat-
ical Forum, vol. 3, no. 19, pp. 945–954, 2008.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 675127, 14 pages
http://dx.doi.org/10.1155/2013/675127

Research Article
The Second Noether Theorem on Time Scales

Agnieszka B. Malinowska1 and Natália Martins2


1
Faculty of Computer Science, Bialystok University of Technology, 15-351 Białystok, Poland
2
Center for Research and Development in Mathematics and Applications (CIDMA), Department of Mathematics,
University of Aveiro, 3810-193 Aveiro, Portugal

Correspondence should be addressed to Agnieszka B. Malinowska; a.malinowska@pb.edu.pl

Received 23 August 2013; Accepted 9 October 2013

Academic Editor: Delfim F. M. Torres

Copyright © 2013 A. B. Malinowska and N. Martins. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

We extend the second Noether theorem to variational problems on time scales. As corollaries we obtain the classical second Noether
theorem, the second Noether theorem for the ℎ-calculus and the second Noether theorem for the 𝑞-calculus.

1. Introduction time translations; conservation of linear momentum comes


from invariance of the system under spacial translations;
In 1915, general relativity was almost a finished theory, but still conservation of angular momentum reflects invariance with
there was a problem regarding the conservation of energy. respect to spatial rotations.
David Hilbert asked for help Emmy Noether. She solved The first Noether theorem is nowadays a well-known
the problem proving two remarkable theorems that relate to tool in modern theoretical physics, engineering, and the cal-
the invariance of a variational integral with properties of its culus of variations [3–9]. Inexplicably, it is still not well
Euler-Lagrange equations. These results were published in known that the famous paper of Emmy Noether includes
1918 in the paper Invariante Variations probleme [1]. Noether another important result, the second Noether theorem, which
was described by many important scientists, such as Pavel applies to variational problems that are invariant under a
Alexandrov, Albert Einstein, Jean Dieudonné, and David certain group of transformations, a so-called infinite contin-
Hilbert, as the most important woman in the history of math- uous group, which depends on arbitrary functions and their
ematics. In order to get a good exposition of the history of derivatives (see also [10]). Such transformations are local
Emmy Noether and her important contributions to funda- transformations because they can affect every part of the
mental physics and mathematics, we refer the reader to the system differently. Noether’s second theorem states that if
recent book [2]. This book explains very clearly that it took a variational integral has an infinite-dimensional Lie alge-
too much time before mathematicians and physicists began bra of infinitesimal symmetries parameterized linearly by 𝑟
to recognize the importance of Noether’s theorems: until 1950 arbitrary functions and their derivatives up to a given order
Noether’s theorems were poorly understood and Noether’s 𝑚, then there are 𝑟 identities between Euler-Lagrange expre-
name disappeared almost entirely. ssions and their derivatives up to order 𝑚. These identi-
The first theorem in [1], usually known as Noether’s theo- ties Noether called them “dependencies.” For example, the
rem, guarantees that the invariance of a variational integral Bianchi identities, in the general theory of relativity, are
with respect to continuous symmetry transformations that examples of such “dependencies.” Noether’s second theo-
depend on 𝜌 parameters implies the existence of 𝜌 con- rem has applications in general relativity, electrodynamics,
served quantities along the Euler-Lagrange extremals. Such hydromechanics, quantum chromodynamics, and other
transformations are global transformations. Noether’s theo- gauge field theories. Motivated by the important applications
rem explains all conservation laws of mechanics: conserva- of the second Noether theorem, our goal in this paper is
tion of energy comes from invariance of the system under to generalize this result proving that the second Noether
2 Abstract and Applied Analysis

theorem is valid for an infinite number of time scales. As 𝑎 < 𝑏. We consider the following optimization problem on
we will see, in the particular case where the time scale T T:
is R, we get from our result the classical second Noether
𝑏
theorem; when T = Z, we obtain the analogue of the second
Noether theorem for the difference calculus of variations; L [𝑦] = ∫ 𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) Δ𝑡 󳨀→ extremize, (1)
𝑎
when T = 𝑞N0 (for some 𝑞 > 1), we obtain a new result:
the second Noether theorem for the 𝑞-calculus (quantum where the set of admissible functions are
calculus). For more on the theory of quantum calculus and
quantum calculus of variation, we refer to [11–21]. D = {𝑦 | 𝑦 : [𝑎, 𝑏] ∩ T 󳨀→ R𝑛 , 𝑦 ∈ 𝐶rd
1
([𝑎, 𝑏] ∩ T, R𝑛 ) ,
The theory of time scales was introduced in 1988 by
Hilger in his Ph.D. thesis [22] as a means of unifying theories 𝑦 (𝑎) = 𝛼, 𝑦 (𝑏) = 𝛽} ,
of differential calculus and difference calculus into a single (2)
theory. With a short time this unification aspect has been
supplemented by the extension and generalization features.
The time scale calculus allows to consider more complex time for some 𝛼, 𝛽 ∈ R𝑛 , and where 𝜎 is the forward jump operator,
domains, such as ℎZ, T = 𝑞N0 , or hybrid domains. The study 𝑦Δ is the delta-derivative of 𝑦, and, for 𝑖 ∈ N,
of the calculus of variations in the context of time scales had
𝑖
its beginning only in 2004 with the paper [23] of Bohner 𝐶rd ([𝑎, 𝑏] ∩ T, R𝑛 ) := {𝑦 | 𝑦 : [𝑎, 𝑏] ∩ T 󳨀→ R𝑛 ,
(see also [24]). Since then, the variational calculus on time
𝑖 𝑖
scales advanced fairly quickly, as can be verified with the large 𝑦Δ is rd-continuous on [𝑎, 𝑏]𝜅 } .
number of published papers on the subject [25–35]. Noether’s
first theorem has been extended to the variational calculus (3)
on time scales using several approaches [4, 9, 36], while the
𝑖
second Noether theorem on times scales is still not available As usual, 𝑦𝜎 (𝑡) denotes 𝑦(𝜎(𝑡)) and [𝑎, 𝑏]𝜅 := [𝑎, 𝜌𝑖 (𝑏)],
in the literature. So, there is evidently a need for a time scale where 𝜌 is the backward jump operator. By extremize, we
analogue of Noether’s second theorem. mean maximize or minimize.
The paper is organized as follows. In Section 2, we In what follows, all intervals are time scales intervals; that
review some preliminaries about single-variable variational is, we simply write [𝑎, 𝑏] to denote the set [𝑎, 𝑏] ∩ T. Let 𝑦 :=
calculus on time scales; for example, we recall the Euler- (𝑦1 , . . . , 𝑦𝑛 ) and denote by 𝜕𝐿/𝜕𝑦𝑘 the partial derivative of 𝐿
Lagrange equation for a delta variational problem. Our main with respect to 𝑦𝑘 .
results are stated in Section 3. Namely, in Subsection 3.1, we
1
prove Noether’s second theorem for variational problems Definition 1. We say that 𝑦∗ ∈ 𝐶rd ([𝑎, 𝑏], R𝑛 ) is a local mini-
involving a single delta integral (with and without trans- mizer (resp., local maximizer) for problem (1) if there exists
formation of time) and in Section 3.2 we prove Noether’s 𝛿 > 0 such that
second theorem for variational problems involving multiple
delta integrals (without transformation of time). Section 4 L [𝑦∗ ] ≤ L [𝑦] (resp., L [𝑦∗ ] ≥ L [𝑦]) , (4)
provides a concrete example of application of our results.
Finally, in Section 5, we present some concluding remarks. 1
for all 𝑦 ∈ 𝐶rd ([𝑎, 𝑏], R𝑛 ) satisfying the boundary conditions
𝑦(𝑎) = 𝛼, 𝑦(𝑏) = 𝛽, and
2. Preliminaries
󵄩󵄩 󵄩 󵄨 𝜎 󵄨
󵄩󵄩𝑦 − 𝑦∗ 󵄩󵄩󵄩 := sup 󵄨󵄨󵄨𝑦 (𝑡) − 𝑦∗ (𝑡)󵄨󵄨󵄨
𝜎
In this paper, we assume the reader to be familiar with the 𝜅
𝑡∈[𝑎,𝑏]
calculus on time scales. For a good introduction to the theory
󵄨 󵄨 (5)
+ sup 󵄨󵄨󵄨󵄨𝑦Δ (𝑡) − 𝑦∗Δ (𝑡)󵄨󵄨󵄨󵄨 < 𝛿,
of time scales, we refer to the well-known books in this field
[37, 38]. The first developments on time scale calculus were 𝑡∈[𝑎,𝑏] 𝜅

done essentially using the delta-calculus. However, for some


applications, in particular to solve problems of the calculus where | ⋅ | denotes a norm in R𝑛 .
of variations and control theory in economics, it is often
1
more convenient to work backwards in time, that is, using Definition 2. We say that 𝜂 ∈ 𝐶rd ([𝑎, 𝑏], R𝑛 ) is an admissible
the nabla-calculus. In this paper, we are concerned with the variation for problem (1) provided that 𝜂(𝑎) = 𝜂(𝑏) = 0.
delta-calculus. It is clear that all the arguments used in the
proofs of our results can be modified to work for the nabla- Definition 3. A function 𝑓 : [𝑎, 𝑏] × R → R is called conti-
calculus. In what follows, we review some preliminaries about nuous in the second variable, uniformly in the first variable,
the variational calculus on time scales needed in this paper. if for each 𝜖 > 0, there exists 𝛿 > 0 such that |𝑥1 − 𝑥2 | < 𝛿
Let T be a given time scale, 𝑛 ∈ N, and let 𝐿 : T × implies |𝑓(𝑡, 𝑥1 ) − 𝑓(𝑡, 𝑥2 )| < 𝜖 for all 𝑡 ∈ [𝑎, 𝑏].
R𝑛 × R𝑛 → R be continuous, together with its partial delta
derivatives of first and second order with respect to 𝑡 and Lemma 4 (see [23]). Suppose that 𝜂 := (𝜂1 , . . . , 𝜂𝑛 ) is an admi-
partial usual derivatives of the first and second order with ssible variation for problem (1) and 𝑦 := (𝑦1 , . . . , 𝑦𝑛 ) ∈ D. Let
respect to the other variables. Suppose that 𝑎, 𝑏 ∈ T, and 𝜙 : R → R and 𝑓 : [𝑎, 𝑏] × R → R be defined, respectively,
Abstract and Applied Analysis 3

by 𝜙(𝜖) := L[𝑦 + 𝜖𝜂] and 𝑓(𝑡, 𝜖) := 𝐿(𝑡, 𝑦𝜎 (𝑡) + 𝜖𝜂𝜎 (𝑡), 𝑦Δ (𝑡) + 3.1. Noether’s Second Theorem: Single Delta Integral Case. In
𝜖𝜂Δ (𝑡)). If 𝜕𝑓/𝜕𝜖 is continuous in 𝜖, uniformly in 𝑡, then this subsection, we suppose that the time scale T satisfies
condition (H) and that 𝐿 satisfies the assumption of Lemma 4,
0 0
𝑏 𝑛
𝜕𝐿 for every 𝑦 and 𝜂. As usual, 𝜂Δ and 𝜂𝜎 denote 𝜂. Let 𝑚 be
𝜙 ̇(0) = ∫ ∑ ( (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) 𝜂𝑘𝜎 (𝑡)
𝑎 𝑘=1 𝜕𝑦𝑘𝜎 a fixed natural number. We also assume that the time scale
(6) interval [𝑎, 𝑏] has, at least, 2𝑚 + 1 points.
𝜕𝐿 We begin with some technical results that will be useful
+ Δ (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) 𝜂𝑘Δ (𝑡)) Δ𝑡. in the proofs of Theorems 16 and 22.
𝜕𝑦𝑘
Lemma 9 (higher-order fundamental lemma of the calculus
Next, we present the following result that is a fundamental
of variations). Let T be a time scale satisfying condition (H)
tool in the calculus of variations on time scales.
and 𝑓0 , 𝑓1 , . . . , 𝑓𝑚 ∈ 𝐶rd ([𝑎, 𝑏], R). If
Theorem 5 (Euler-Lagrange equation on time scales [23]). If 𝜌𝑚−1 (𝑏) 𝑚 𝑚−𝑖
Δ𝑖
𝑦∗ is a weak local extremizer for problem (1) and 𝐿 satisfies ∫ (∑𝑓𝑖 (𝑡) 𝜂𝜎 (𝑡)) Δ𝑡 = 0, (9)
the assumption of Lemma 4, for every 𝑦 and 𝜂, then the 𝑎 𝑖=0
components of 𝑦∗ satisfy the 𝑛 Euler-Lagrange equations: 2𝑚
for all 𝜂 ∈ 𝐶rd ([𝑎, 𝑏], R) such that
Δ 𝜕𝐿 𝜕𝐿 𝜂 (𝜌𝑚−1 (𝑏)) = 0,
Δ
(𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) = 𝜎 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) , 𝜂 (𝑎) = 0,
Δ𝑡 𝜕𝑦𝑘 𝜕𝑦𝑘 (7)
..
𝑘 = 1, . . . , 𝑛 . (10)
𝑚−1 𝑚−1
for 𝑡 ∈ [𝑎, 𝑏]𝜅 . 𝜂Δ (𝑎) = 0, 𝜂Δ (𝜌𝑚−1 (𝑏)) = 0,

It is well known that the forward jump operator, 𝜎, is not then


delta differentiable for certain time scales. Also, the chain rule 𝑚
1 𝑖(𝑖−1)/2 Δ𝑖 𝑚
as we know it from the classical calculus (i.e., when T = R) is ∑(−1)𝑖 ( ) 𝑓𝑖 (𝑡) = 0, 𝑡 ∈ [𝑎, 𝑏]𝜅 . (11)
𝑖=0 𝑏1
not valid in general. For this reason, we suppose that the time
scale T satisfies the following condition Proof. The proof is similar to the proof of Lemma 16 of [33].
(H) for each 𝑡 ∈ T, 𝜎(𝑡) = 𝑏1 𝑡 + 𝑏0 for some 𝑏1 ∈ R+ and
𝑏0 ∈ R. Remark 10. We emphasize that the delta differentiability of
the functions 𝑓0 , 𝑓1 , . . . , 𝑓𝑚 was not assumed in advance.
Remark 6. Note that condition (H) implies that 𝜎 is delta
differentiable and 𝜎Δ (𝑡) = 𝑏1 , 𝑡 ∈ T 𝜅 . Also note that condition Lemma 11. Assume that the time scale T satisfies condition (H)
(H) describes, in particular, the differential calculus (T = R, 2𝑚
and 𝜂 ∈ 𝐶rd ([𝑎, 𝑏], R) is such that,
𝑏1 = 1, 𝑏0 = 0), the difference calculus (T = Z, 𝑏1 = 1, 𝑏0 = 1),
𝑖
the ℎ-calculus (T = ℎZ := {ℎ𝑧 : 𝑧 ∈ Z}, 𝑏1 = 1, 𝑏0 = ℎ for 𝜂Δ (𝑎) = 0, 𝑖 = 0, 1, . . . , 𝑚. (12)
some ℎ > 0), and the 𝑞-calculus (T = 𝑞N0 := {𝑞𝑘 : 𝑘 ∈ N0 } for
some 𝑞 > 1, 𝑏1 = 𝑞, 𝑏0 = 0). Then,
𝑖

Lemma 7 (see [27]). Let T be a time scale satisfying condition 𝜂𝜎 (𝑎) = 0, 𝑖 = 0, 1, . . . , 𝑚. (13)
(𝐻). If 𝑓 : T → R is two times delta differentiable, then Proof. If 𝑎 is right-dense, the result is trivial. Suppose that 𝑎
2 is right-scattered. Since 𝜂Δ (𝑎) = (𝜂𝜎 (𝑎) − 𝜂(𝑎))/(𝜎(𝑎) − 𝑎) = 0
𝑓𝜎Δ (𝑡) = 𝑏1 𝑓Δ𝜎 (𝑡) , 𝑡 ∈ T𝜅 . (8)
and 𝜂(𝑎) = 0, we conclude that 𝜂𝜎 (𝑎) = 0. Since 𝜂Δ (𝑎) =
2

(𝜂Δ )Δ (𝑎) = ((𝜂Δ )𝜎 (𝑎) − 𝜂Δ (𝑎))/(𝜎(𝑎) − 𝑎) = 0 and 𝜂Δ (𝑎) = 0,


The next result is also useful for the proofs of our main
results. then (𝜂Δ )𝜎 (𝑎) = 0. Using Lemma 7, we get (𝜂𝜎 )Δ (𝑎) = 0. Since
2
(𝜂𝜎 )Δ (𝑎) = (𝜂𝜎 (𝑎) − 𝜂𝜎 (𝑎))/(𝜎(𝑎) − 𝑎) = 0 and 𝜂𝜎 (𝑎) = 0, we
2
Lemma 8 (cf. [35]). Assume that the time scale T satisfies conclude that 𝜂𝜎 (𝑎) = 0.
2𝑚
condition (𝐻), 𝑚 ∈ N, and 𝜂 ∈ 𝐶rd ([𝑎, 𝑏], R) is such that 3 2 2 2
Since 𝜂Δ (𝑎) = (𝜂Δ )Δ (𝑎) = ((𝜂Δ )𝜎 (𝑎) − 𝜂Δ (𝑎))/(𝜎(𝑎) −
𝑖 𝑖−1
𝜂 (𝑎) = 0 for all 𝑖 = 0, 1, . . . , 𝑚. Then, 𝜂𝜎Δ (𝑎) = 0 for each
Δ 2 2
𝑎) = 0 and 𝜂Δ (𝑎) = 0, then 𝜂Δ 𝜎 (𝑎) = 0. Using Lemma 7, we
𝑖 = 1, . . . , 𝑚. 2 2
get (𝜂𝜎 )Δ (𝑎) = 0. Since (𝜂𝜎 )Δ (𝑎) = (𝜂𝜎Δ𝜎 (𝑎)−𝜂𝜎Δ (𝑎))/(𝜎(𝑎)−
𝑎) = 0 and 𝜂𝜎Δ (𝑎) = 0, then 𝜂𝜎Δ𝜎 (𝑎) = 0. Lemma 7 proves that
3. Main Results 2 2 3 2
𝜂𝜎 Δ (𝑎) = 0. Since 𝜂𝜎 Δ (𝑎) = (𝜂𝜎 (𝑎) − 𝜂𝜎 (𝑎))/(𝜎(𝑎) − 𝑎) = 0
2 3
In this section, we formulate and prove the second Noether and 𝜂𝜎 (𝑎) = 0, we get 𝜂𝜎 (𝑎) = 0. Repeating recursively this
theorem for single and multiple integral variational problems. process, we conclude the proof.
4 Abstract and Applied Analysis

Lemma 12. Assume that the time scale T satisfies condition Theorem 15 (necessary condition of invariance). If func-
2(𝑚−1) tional L is invariant under transformations (16), then
(H) and 𝜂 ∈ 𝐶rd ([𝑎, 𝑏], R) is such that,

𝑖 𝜎
𝜂Δ (𝑎) = 0, 𝑖 = 0, 1, . . . , 𝑚 − 1. (14) 𝑛
𝜕𝐿 𝑏 𝑟
∑ ∫ ( 𝜎 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) ⋅ ( ∑ 𝑇𝑘𝑗 (𝑝𝑗 )) (𝑡)
𝑘=1 𝑎
𝜕𝑦𝑘 𝑗=1
Then,
Δ
𝑟
𝜎Δ𝑚−2 𝜎2 Δ𝑚−3 𝜎3 Δ𝑚−4 𝜎𝑚−2 Δ 𝜕𝐿
𝜂 (𝑎) = 𝜂 (𝑎) = 𝜂 (𝑎) = . . . = 𝜂 (𝑎) = 0. + Δ (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) ⋅ ( ∑ 𝑇𝑘𝑗 (𝑝𝑗 )) (𝑡)) Δ𝑡 = 0.
(15) 𝜕𝑦𝑘 𝑗=1
(19)
Proof. If 𝑎 is right-dense, the result is trivial. Suppose that 𝑎
𝑚−1 𝑚−2 𝑚−2
is right-scattered. Since 𝜂Δ (𝑎) = (𝜂Δ 𝜎 (𝑎) − 𝜂Δ (𝑎))/ Proof. Using the definition of invariance and noting that the
𝑚−2 𝑚−2
(𝜎(𝑎) − 𝑎) = 0 and 𝜂Δ (𝑎) = 0, then 𝜂Δ 𝜎 (𝑎) = 0. Using family of transformations (16) depends upon arbitrary func-
𝑚−2
Lemma 7, we get 𝜂𝜎Δ (𝑎) = 0 (or use Lemma 8). Note that tions 𝑝1 , 𝑝2 , . . . , 𝑝𝑟 , we conclude that, for any real number 𝜀,
𝑚−2 𝑚−3 𝑚−3
𝜂𝜎Δ (𝑎) = (𝜂𝜎Δ 𝜎 (𝑎) − 𝜂𝜎Δ (𝑎))/(𝜎(𝑎) − 𝑎) = 0. Lemma 8
𝑚−3 𝑚−3 𝑏
shows that 𝜂𝜎Δ (𝑎) = 0; hence, 𝜂𝜎Δ 𝜎 (𝑎) = 0. Using ∫ 𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) Δ𝑡
2 𝑚−3 2 𝑚−4
Lemma 7, we get 𝜂𝜎 Δ (𝑎) = 0. Next, we prove that 𝜂𝜎 Δ 𝑎
𝑚−3 𝑚−4 𝑚−4
(𝑎) = 0. Since 𝜂𝜎Δ (𝑎) = (𝜂𝜎Δ 𝜎 (𝑎) − 𝜂𝜎Δ (𝑎))/ (𝜎(𝑎) − 𝑏 𝑟
𝜎
𝑚−4
𝑎) = 0 and 𝜂𝜎Δ (𝑎) = 0 (by Lemma 8), then 𝜂𝜎Δ 𝜎 (𝑎) =
𝑚−4
=∫ 𝐿 (𝑡, 𝑦1𝜎 (𝑡) + 𝜀( ∑ 𝑇 (𝑝𝑗 )) (𝑡) , . . . , 𝑦𝑛𝜎 (𝑡)
1𝑗

2 𝑚−4 2 𝑚−3 𝑎 𝑗=1


0. Lemma 7 shows that 𝜂𝜎 Δ (𝑎) = 0. Since 𝜂𝜎 Δ (𝑎) =
2 𝑚−4 2 𝑚−4 2 𝑚−4 𝜎
(𝜂𝜎 Δ 𝜎 (𝑎) − 𝜂𝜎 Δ (𝑎))/(𝜎(𝑎) − 𝑎) = 0 and 𝜂𝜎 Δ (𝑎) = 0, 𝑟
𝑛𝑗
2 𝑚−4
then 𝜂𝜎 Δ 𝜎 (𝑎) = 0. Lemma 7 shows that 𝜂𝜎 Δ (𝑎) = 0.
3 𝑚−4
+ 𝜀( ∑ 𝑇 (𝑝𝑗 )) (𝑡) ,
𝑗=1
Repeating recursively this process, we prove the intended
result. 𝑟
Δ

𝑦1Δ (𝑡) + 𝜀( ∑𝑇 (𝑝𝑗 )) (𝑡) , . . . , 𝑦𝑛Δ (𝑡)


1𝑗

𝑗=1
Let 𝑦 := (𝑦1 , 𝑦2 , . . . , 𝑦𝑛 ). Firstly, we will prove the
second Noether theorem without transformation of time. For Δ
𝑟
that consider the following transformations that depend on 𝑛𝑗
+𝜀( ∑ 𝑇 (𝑝𝑗 )) (𝑡)) Δ𝑡.
arbitrary functions 𝑝1 , 𝑝2 , . . . , 𝑝𝑟 and their delta-derivatives 𝑗=1
up to order 𝑚:
(20)
𝑡=𝑡
Differentiating with respect to 𝜀 (use Lemma 4) and taking
𝑟 (16)
𝑘𝑗 𝜀 = 0, we get equality (19).
𝑦𝑘 (𝑡) = 𝑦𝑘 (𝑡) + ∑ 𝑇 (𝑝𝑗 ) (𝑡) , 𝑘 = 1, 2, . . . , 𝑛,
𝑗=1

2𝑚
where, for each 𝑗 = 1, 2, . . . , 𝑟, 𝑝𝑗 ∈ 𝐶rd ([𝑎, 𝜎𝑚 (𝑏)], R), Define

𝑚 𝑚−(𝑖+1)
Δ𝑖
𝜕𝐿 Δ 𝜕𝐿
𝑇𝑘𝑗 (𝑝𝑗 ) := ∑𝑔𝑖𝑗𝑘 𝑝𝑗𝜎 , (17) 𝐸𝑘 (𝐿) := − , 𝑘 = 1, 2, . . . , 𝑛. (21)
𝜕𝑦𝑘𝜎 Δ𝑡 𝜕𝑦𝑘Δ
𝑖=0

and 𝑔𝑖𝑗𝑘 ∈ 𝐶rd


1
([𝑎, 𝑏], R). We will call 𝐸𝑘 (𝐿), 𝑘 = 1, 2, . . . , 𝑛, the Euler-Lagrange expre-
ssions associated to the Lagrangian 𝐿.
Definition 13. Functional L is invariant under transforma-
1
tions (16) if, and only if, for all 𝑦 ∈ 𝐶rd ([𝑎, 𝑏], R), we have Theorem 16 (Noether’s second theorem without transform-
ing time). If functional L is invariant under transformations
𝑏 𝑏 (16), then there exist the following identities:
∫ 𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) Δ𝑡 = ∫ 𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) Δ𝑡. (18)
𝑎 𝑎
𝑛 𝑚
1 𝑖(𝑖+1)/2 𝑘 𝜎 Δ𝑖
∑ ∑(−1)𝑖 ( ) ((𝑔𝑖𝑗 ) 𝐸𝑘 (𝐿)) ≡ 0,
Remark 14. Note that the most common definition of invari-
𝑘=1 𝑖=0
𝑏1 (22)
ance (with equality of the integrals for any subinterval
[𝑡𝑎 , 𝑡𝑏 ] ⊆ [𝑎, 𝑏] with 𝑎, 𝑏 ∈ T) implies Definition 13. 𝑗 = 1, 2, . . . , 𝑟.
Abstract and Applied Analysis 5

Proof. Using the necessary condition of invariance Hence,


(Theorem 15), we conclude that 𝜎
𝑛 𝑏 𝑚
𝜎 𝜎 Δ 𝑘 𝜎𝑚−(𝑖+1) Δ𝑖
𝑛 𝑏
𝜕𝐿 𝑟 ∑ ∫ 𝐸𝑘 (𝐿) (𝑡, 𝑦 (𝑡) , 𝑦 (𝑡)) ⋅ (∑𝑔𝑖𝑗 𝑝𝑗 ) (𝑡) Δ𝑡
𝜎 Δ
∑∫ ( (𝑡, 𝑦 (𝑡) , 𝑦 (𝑡)) ⋅ ( ∑ 𝑇𝑘𝑗 (𝑝𝑗 )) (𝑡) 𝑘=1 𝑎 𝑖=0
𝑘=1 𝑎
𝜕𝑦𝑘𝜎 𝑗=1
= 0.
Δ (29)
𝑟
𝜕𝐿
+ Δ
(𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) ⋅ ( ∑ 𝑇𝑘𝑗 (𝑝𝑗 )) (𝑡)) Δ𝑡 = 0.
𝜕𝑦𝑘 𝑗=1 Therefore, by Lemma 7, we get
(23) 𝑏 𝑚 𝑛
∫ ∑ ∑ 𝐸𝑘 (𝐿) (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡))
Fix 𝑗 = 1, 2, . . . , 𝑟. By the arbitrariness of 𝑝1 , 𝑝2 , . . . , 𝑝𝑟 , we 𝑎 𝑖=0 𝑘=1
can suppose that 𝑝ℎ ≡ 0 for ℎ ≠ 𝑗. Therefore, (30)
𝑛 𝜎 1 𝑖 𝑚−𝑖 𝑖
𝑏
𝜕𝐿 𝜎 Δ 𝑘𝑗 𝜎 ⋅ (𝑔𝑖𝑗𝑘 ) (𝑡) ( ) 𝑝𝑗𝜎 Δ (𝑡) Δ𝑡 = 0.
∑∫ ( 𝜎 (𝑡, 𝑦 (𝑡) , 𝑦 (𝑡)) ⋅ (𝑇 (𝑝𝑗 )) (𝑡) 𝑏1
𝑘=1 𝑎
𝜕𝑦𝑘
Applying the higher-order fundamental lemma of the calcu-
𝜕𝐿 Δ lus of variations (Lemma 9), we obtain
+ Δ (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) ⋅ (𝑇𝑘𝑗 (𝑝𝑗 )) (𝑡)) Δ𝑡 = 0.
𝜕𝑦𝑘
Δ𝑖
(24) 𝑚 𝑛
1 𝑖(𝑖−1)/2𝑖 𝜎 1
𝑖
∑ ∑ (−1) ( ) (𝐸𝑘 (𝐿)(𝑔𝑖𝑗𝑘 ) ( ) ) ≡ 0, (31)
Integrating by parts, we obtain 𝑖=0 𝑘=1 𝑏1 𝑏1
𝑛 𝑏
𝜕𝐿 Δ 𝜕𝐿 which is equivalent to
∑ (∫ ( (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) − (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)))
𝑎 𝜕𝑦𝑘𝜎 Δ𝑡 𝜕𝑦𝑘Δ
𝑘=1 𝑛 𝑚
1 𝑖(𝑖+1)/2
𝑖
𝜎 Δ
𝜎 ∑ ∑(−1)𝑖 ( ) ((𝑔𝑖𝑗𝑘 ) 𝐸𝑘 (𝐿)) ≡ 0 (32)
⋅ (𝑇𝑘𝑗 (𝑝𝑗 )) (𝑡) Δ𝑡 𝑘=1 𝑖=0
𝑏1

𝑏 proving the desired result.


𝜕𝐿
+[ (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) ⋅ 𝑇𝑘𝑗 (𝑝𝑗 ) (𝑡)] ) = 0.
𝜕𝑦𝑘Δ 𝑎
We present some particular results that follow from
(25) Theorem 16 in the case where T = R, T = ℎZ (for some
ℎ > 0), and T = 𝑞N0 (for some 𝑞 > 1). If T = R, then
By the arbitrariness of 𝑝𝑗 , we can restrict to those functions ̇ and we obtain the classical second
𝜎(𝑡) = 𝑡, 𝑓Δ (𝑡) = 𝑓(𝑡),
such that Noether theorem without transformation of time.
𝑝𝑗 (𝑎) = 0, 𝑝𝑗 (𝑏) = 0,
Corollary 17 (cf. [1]). Let 𝐿 : R × R𝑛 × R𝑛 → R be a 𝐶2
.. function. If functional L defined by
.
(26) 𝑏
𝑝𝑗Δ
𝑚−1
(𝑎) = 0, 𝑝𝑗Δ
𝑚−1
(𝑏) = 0, L [𝑦] = ∫ 𝐿 (𝑡, 𝑦 (𝑡) , 𝑦̇(𝑡)) 𝑑𝑡 (33)
𝑎
−1
Δ𝑚 −1
Δ𝑚 is invariant under transformations (16) (where 𝜎 denotes in
𝑝𝑗𝜎 (𝑎) = 0, 𝑝𝑗𝜎 (𝑏) = 0.
this context the identity function and Δ denotes the usual
Using Lemmas 8, 11, and 12, we conclude that 𝑇𝑘𝑗 (𝑝𝑗 )(𝑎) = 0 derivative), then there exist the following identities:
and 𝑇𝑘𝑗 (𝑝𝑗 )(𝑏) = 0, 𝑘 = 1, 2, . . . , 𝑛. Then, 𝑛 𝑚
(𝑖)
𝑛
∑ ∑(−1)𝑖 [𝑔𝑖𝑗𝑘 𝐸𝑘 (𝐿)] ≡ 0, 𝑗 = 1, 2, . . . , 𝑟. (34)
𝑏
𝜕𝐿 Δ 𝜕𝐿 𝑘=1 𝑖=0
∑ ∫ ( 𝜎 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) − Δ
(𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)))
𝑎 𝜕𝑦 Δ𝑡 𝜕𝑦
𝑘=1 𝑘 𝑘 Choosing ℎZ, we obtain Noether’s second theorem with-
𝜎 out transformation of time for the ℎ-calculus.
⋅ (𝑇𝑘𝑗 (𝑝𝑗 )) (𝑡) Δ𝑡 = 0;
Corollary 18. Let ℎ > 0, 𝐿 : ℎZ × R𝑛 × R𝑛 → R and 𝑎, 𝑏 ∈
ℎZ, 𝑎 < 𝑏. If functional L defined by
(27)
𝑏−ℎ
that is,
L [𝑦] = ∑ 𝐿 (𝑡, 𝑦 (𝑡 + ℎ) , Δ ℎ [𝑦] (𝑡)) (35)
𝑛 𝑏 𝑡=𝑎
𝜎 Δ 𝑘𝑗 𝜎
∑ ∫ 𝐸𝑘 (𝐿) (𝑡, 𝑦 (𝑡) , 𝑦 (𝑡)) ⋅ (𝑇 (𝑝𝑗 )) (𝑡) Δ𝑡 = 0.
𝑘=1 𝑎 is invariant under transformations (16) (where 𝜎 denotes in this
(28) context the function 𝜎(𝑡) = 𝑡+ℎ and Δ denotes the ℎ-difference,
6 Abstract and Applied Analysis

that is, 𝑦Δ (𝑡) = Δ ℎ [𝑦](𝑡) = (𝑦(𝑡+ℎ)−𝑦(𝑡))/ℎ), then there exist Moreover, we assume that the map
the following identities:
𝑟
𝑛 𝑚 𝑡 󳨃󳨀→ 𝛼 (𝑡) := 𝑡 + ∑ 𝐻𝑗 (𝑝𝑗 ) (𝑡) (41)
∑ ∑(−1)𝑖 [𝑔𝑖𝑗𝑘 (𝑡 + ℎ) 𝑗=1
𝑘=1 𝑖=0
1
Δ𝑖ℎ (36) is a strictly increasing 𝐶rd function and its image is again a
⋅𝐸𝑘 (𝐿) (𝑡, 𝑦 (𝑡 + ℎ) , Δ ℎ [𝑦] (𝑡)) ] = 0, time scale, T. We denote the forward shift operator relative
𝑡 ∈ [𝑎, 𝑏 − 𝑚ℎ] , 𝑗 = 1, 2, . . . , 𝑟. to T by 𝜎 and the delta derivative by Δ. We remark that the
following holds [40]:
For ℎ = 1, we obtain the analogue of Noether’s second
𝜎 ∘ 𝛼 = 𝛼 ∘ 𝜎. (42)
theorem for the difference calculus of variations recently
proved in [39]. In the case T = 𝑞N0 , we obtain the new result. Definition 20. Functional L is invariant under transforma-
1
tions (39) if, and only if, for all 𝑦 ∈ 𝐶rd ([𝑎, 𝑏], R), we have
Corollary 19. Let 𝑞 > 1, 𝐿 : 𝑞N0 × R𝑛 × R𝑛 → R and 𝑎, 𝑏 ∈
𝑞N0 , 𝑎 < 𝑏. If functional L defined by 𝑏 𝑏
∫ 𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) Δ𝑡 = ∫ 𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) Δ𝑡.
𝑏/𝑞 𝑎 𝑎
L [𝑦] = ∑𝐿 (𝑡, 𝑦 (𝑞𝑡) , Δ 𝑞 [𝑦] (𝑡)) (37) (43)
𝑡=𝑎
We recall the following results that will be very useful in
is invariant under transformations (16) (where 𝜎 denotes in this the proof of Theorem 22.
context the function 𝜎(𝑡) = 𝑞𝑡 and Δ denotes the 𝑞-derivative,
that is, 𝑦Δ (𝑡) = Δ 𝑞 [𝑦](𝑡) = (𝑦(𝑞𝑡) − 𝑦(𝑡))/(𝑞 − 1)𝑡), then there Theorem 21 (see [37]). Assume that ] : T → R is strictly
exist the following identities: increasing and T̃ := ](T) is a time scale.
̃
(1) Chain rule: let 𝜔 : T̃ → R. If ]Δ (𝑡) and 𝜔Δ (](𝑡)) exist
𝑛
1 𝑖(𝑖+1)/2 𝑘
𝑚
𝑖
∑ ∑(−1) ( ) [𝑔𝑖𝑗 (𝑞𝑡) for all 𝑡 ∈ T 𝜅 , then
𝑘=1 𝑖=0
𝑞
̃
Δ𝑖𝑞 (𝜔 ∘ ])Δ = (𝜔Δ ∘ ]) ]Δ . (44)
⋅𝐸𝑘 (𝐿) (𝑡, 𝑦 (𝑞𝑡) , Δ 𝑞 [𝑦] (𝑡))] = 0,

𝑏 (2) Substitution in the integral: if 𝑓 : T̃ → R is a 𝐶rd fun-


𝑡 ∈ [𝑎, ] , 𝑗 = 1, 2, . . . , 𝑟.
𝑞𝑚 1
ction and ] is a 𝐶rd function, then for 𝑎, 𝑏 ∈ T,
(38)
𝑏 ](𝑏)
∫ 𝑓 (] (𝑡)) ]Δ (𝑡) Δ𝑡 = ∫ ̃
𝑓 (𝑠) Δ𝑠. (45)
In order to prove the second Noether theorem with
𝑎 ](𝑎)
transformation of time, we will consider that the Lagrangian
𝐿 is defined for all 𝑡 ∈ R (not only for 𝑡 from the initial
Now, we are ready to state and prove Noether’s second
time scale T), 𝐿 : R × R𝑛 × R𝑛 → R. Consider the
theorem with transformation of time.
following transformations that depend on arbitrary functions
𝑝1 , 𝑝2 , . . . , 𝑝𝑟 and their delta-derivatives up to order 𝑚: Theorem 22 (Noether’s second theorem with transformation
𝑟 of time). If functional L is invariant under transformations
𝑡 = 𝑡 + ∑ 𝐻𝑗 (𝑝𝑗 ) (𝑡) , (39), then there exist the following identities:
𝑗=1
(39) 𝑛 𝑚
1 𝑖(𝑖+1)/2 [
𝑖
𝜎 Δ
𝑟 ∑ ∑(−1)𝑖 ( ) ((𝑔𝑖𝑗𝑘 ) 𝐸𝑘 (𝐿))
𝑦𝑘 (𝑡) = 𝑦𝑘 (𝑡) + ∑𝑇𝑘𝑗 (𝑝𝑗 ) (𝑡) , 𝑘 = 1, 2, . . . , 𝑛, 𝑘=1 𝑖=0
𝑏1
𝑗=1
[
Δ𝑖
2𝑚 𝜎 𝜕𝐿 Δ 𝜕𝐿 𝜕𝐿
where, for each 𝑗 = 1, 2, . . . , 𝑟, 𝑝𝑗 ∈ 𝐶rd ([𝑎, 𝜎𝑚 (𝑏)], R), +((𝑓𝑖𝑗 ) ( − (𝐿 − 𝑦𝑘Δ Δ − 𝜇 ))) ] ≡ 0
𝜕𝑡 Δ𝑡 𝜕𝑦𝑘 𝜕𝑡
𝑚
]
𝑚−(𝑖+1)
Δ𝑖 (46)
𝐻𝑗 (𝑝𝑗 ) := ∑𝑓𝑖𝑗 𝑝𝑗𝜎 ,
𝑖=0
(40) for 𝑗 = 1, 2, . . . , 𝑟.
𝑚 𝑚−(𝑖+1) 𝑖
𝑇𝑘𝑗 (𝑝𝑗 ) := ∑𝑔𝑖𝑗𝑘 𝑝𝑗𝜎 Δ
, Proof. The idea of the proof is to reduce the statement of
𝑖=0 this result to the one of Theorem 16 using a technique of
reparametrization of time: artificially, we will consider 𝑡 as a
1
𝑓𝑖𝑗 ∈ 𝐶rd ([𝑎, 𝑏], R) and 𝑔𝑖𝑗𝑘 ∈ 𝐶rd
1
([𝑎, 𝑏], R). dependent variable of the same footing with 𝑦.
Abstract and Applied Analysis 7

Let 𝑟 ≠
0 and define Hence, using (50) and (52), we get

̃ (𝑡, 𝑠, 𝑦, 𝑟, V) := 𝐿 (𝑠 − 𝜇 (𝑡) 𝑟, 𝑦, V ) 𝑟.
𝐿 (47)
̃ [𝑠 (⋅) , 𝑦 (⋅)] = L
L ̃ [𝐻 (𝑠 (⋅) , 𝑦 (⋅)) , 𝑇 (𝑠 (⋅) , 𝑦 (⋅))] . (53)
𝑟
̃ is invariant on
This means that L
1
Note that, for 𝑠(𝑡) = 𝑡 and any 𝑦 ∈ 𝐶rd ([𝑎, 𝑏], R𝑛 ), we have
̃ = {(𝑠, 𝑦) | 𝑠 (𝑡) = 𝑡 ∧ 𝑦 ∈ 𝐶1 ([𝑎, 𝑏] , R𝑛 )}
𝑈 (54)
rd
𝜎 Δ̃ (𝑡, 𝑠 (𝑡) , 𝑦 (𝑡) , 𝑠 (𝑡) , 𝑦 (𝑡)) .
𝐿 (𝑡, 𝑦 (𝑡) , 𝑦 (𝑡)) = 𝐿 𝜎 𝜎 Δ Δ
under the group of state transformations
(48)
(𝑠, 𝑦) = (𝐻 (𝑠, 𝑦) , 𝑇 (𝑠, 𝑦)) (55)
Therefore, for 𝑠(𝑡) = 𝑡,
in the sense of Definition 13.
𝑏
𝜎 Δ Using Theorem 16, we can conclude that there exist the
L [𝑦] := ∫ 𝐿 (𝑡, 𝑦 (𝑡) , 𝑦 (𝑡)) Δ𝑡
𝑎 following 𝑟 identities (𝑗 = 1, 2, . . . , 𝑟)
𝑏 𝑛 𝑚
1 𝑖(𝑖+1)/2
𝑖
𝜎 Δ
=∫ 𝐿 ̃ [𝑠, 𝑦] .
̃ (𝑡, 𝑠𝜎 (𝑡) , 𝑦𝜎 (𝑡) , 𝑠Δ (𝑡) , 𝑦Δ (𝑡)) Δ𝑡 := L ∑ ∑(−1)𝑖 ( ) ̃
((𝑔𝑖𝑗𝑘 ) 𝐸𝑘 (𝐿))
𝑎 𝑘=1 𝑖=0
𝑏1
(49) (56)
𝑚
1 𝑖(𝑖+1)/2
𝑖
𝜎 Δ
Note that, for 𝑠(𝑡) = 𝑡, + ∑(−1) ( ) 𝑖 ̃
((𝑓𝑖𝑗 ) 𝐸𝑠 (𝐿)) ≡ 0,
𝑖=0 𝑏1
̃ [𝑠 (⋅) , 𝑦 (⋅)]
L
̃ := (𝜕𝐿/𝜕𝑠
where we denote 𝐸𝑠 (𝐿) ̃ 𝜎 ) − (Δ/Δ𝑡)(𝜕𝐿/𝜕𝑠
̃ Δ ).
= L [𝑦 (⋅)] Note that, for 𝑠(𝑡) = 𝑡,
𝑏 ̃
𝜕𝐿
𝜎
= ∫ 𝐿 (𝑡, 𝑦 (𝑡) , 𝑦 (𝑡)) Δ𝑡 Δ (𝑡, 𝑠𝜎 (𝑡) , 𝑦𝜎 (𝑡) , 𝑠Δ (𝑡) , 𝑦Δ (𝑡))
𝑎
𝜕𝑠𝜎
𝛼(𝑏) 𝜕𝐿 𝜎 𝑦Δ (𝑡)
=∫ Δ
𝐿 (𝑡, (𝑦 ∘ 𝜎) (𝑡) , 𝑦 (𝑡)) Δ𝑡 = (𝑠 (𝑡) − 𝜇 (𝑡) 𝑠Δ (𝑡) , 𝑦𝜎 (𝑡) , Δ ) 𝑠Δ (𝑡) ,
𝛼(𝑎)
𝜕𝑡 𝑠 (𝑡)

𝑏 𝜕𝐿̃
= ∫ 𝐿 (𝛼 (𝑡) , (𝑦 ∘ 𝜎 ∘ 𝛼) (𝑡) , 𝑦Δ (𝛼 (𝑡))) 𝛼Δ (𝑡) Δ𝑡 (𝑡, 𝑠𝜎 (𝑡) , 𝑦𝜎 (𝑡) , 𝑠Δ (𝑡) , 𝑦Δ (𝑡))
𝑎
𝜕𝑠Δ
𝑏 (𝑦 ∘ 𝛼) (𝑡)
Δ 𝑦Δ (𝑡)
= 𝐿 (𝑠𝜎 (𝑡) − 𝜇 (𝑡) 𝑠Δ (𝑡) , 𝑦𝜎 (𝑡) , )
= ∫ 𝐿 (𝛼 (𝑡) , (𝑦 ∘ 𝛼 ∘ 𝜎) (𝑡) , Δ (𝑡)
) 𝛼Δ (𝑡) Δ𝑡 𝑠Δ (𝑡)
𝑎 𝛼
Δ
𝑛
𝑦𝑘Δ (𝑡) 𝜕𝐿 𝜎 Δ 𝜎 𝑦Δ (𝑡)
𝑏
𝜎 (𝑦 ∘ 𝛼) (𝑡) −∑ (𝑠 (𝑡) − 𝜇 (𝑡) 𝑠 (𝑡) , 𝑦 (𝑡) , )
= ∫ 𝐿 (𝛼𝜎 (𝑡)−𝜇 (𝑡) 𝛼Δ (𝑡) , (𝑦 ∘ 𝛼) (𝑡) , ) 𝑘=1
𝑠Δ (𝑡) 𝜕𝑦𝑘Δ 𝑠Δ (𝑡)
𝑎 𝛼Δ (𝑡)
𝜕𝐿 𝜎 𝑦Δ (𝑡)
⋅ 𝛼Δ (𝑡) Δ𝑡 − (𝑠 (𝑡) − 𝜇 (𝑡) 𝑠Δ (𝑡) , 𝑦𝜎 (𝑡) , Δ )
𝜕𝑡 𝑠 (𝑡)
𝑏
̃ (𝑡, 𝛼𝜎 (𝑡) , (𝑦 ∘ 𝛼)𝜎 (𝑡) , 𝛼Δ (𝑡) , (𝑦 ∘ 𝛼)Δ (𝑡)) Δ𝑡
=∫ 𝐿 ⋅ 𝜇 (𝑡) 𝑠Δ (𝑡) .
𝑎
(57)
̃ [𝛼 (⋅) , (𝑦 ∘ 𝛼) (⋅)] .
=L Hence, for 𝑠(𝑡) = 𝑡,
(50)
̃ (𝑡, 𝑠𝜎 (𝑡) , 𝑦𝜎 (𝑡) , 𝑠Δ (𝑡) , 𝑦Δ (𝑡))
𝐸𝑠 (𝐿)
1 2 𝑛
Let 𝐻(𝑡, 𝑦(𝑡)) := 𝛼(𝑡) and 𝑇 = (𝑇 , 𝑇 , . . . , 𝑇 ) where
𝜕𝐿
𝑟 = (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡))
𝜕𝑡
𝑇𝑘 (𝑡, 𝑦 (𝑡)) := 𝑦𝑘 (𝑡) + ∑ 𝑇𝑘𝑗 (𝑝𝑗 ) (𝑡) , 𝑘 = 1, 2, . . . , 𝑛.
𝑗=1
Δ
(51) − (𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡))
Δ𝑡 (58)
Then, for 𝑠(𝑡) = 𝑡, 𝑛
𝜕𝐿
− ∑ 𝑦𝑘Δ (𝑡) (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡))
(𝛼 (𝑡) , (𝑦 ∘ 𝛼) (𝑡)) = (𝑡, 𝑦 (𝑡)) 𝜕𝑦𝑘Δ
𝑘=1

= (𝐻 (𝑡, 𝑦 (𝑡)) , 𝑇 (𝑡, 𝑦 (𝑡))) (52)


𝜕𝐿
−𝜇 (𝑡) (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) ) .
= (𝐻 (𝑠 (𝑡) , 𝑦 (𝑡)) , 𝑇 (𝑠 (𝑡) , 𝑦 (𝑡))) . 𝜕𝑡
8 Abstract and Applied Analysis

Also note that, for 𝑠(𝑡) = 𝑡 and 𝑘 = 1, 2, . . . , 𝑛, 3.2. Noether’s Second Theorem: Multiple Delta Integral Case.
In this subsection, we extend the second Noether theorem
̃ (𝑡, 𝑠𝜎 (𝑡) , 𝑦𝜎 (𝑡) , 𝑠Δ (𝑡) , 𝑦Δ (𝑡))
𝐸𝑘 (𝐿) (without transformation of time) to multiple integral vari-
(59) ational problems in the time scale setting. For simplicity
= 𝐸𝑘 (𝐿) (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) . of presentation, we prove the result for the case of two
independent variables and transformations that depend on an
Substituting the above equalities into (56), we conclude the arbitrary function and its first-order partial delta derivatives.
desired result: Clearly, our result can be generalized for 𝑛 independent
variables and 𝑟 arbitrary functions and their higher-order
𝑛 𝑚
1 𝑖(𝑖+1)/2 [
𝑖
𝜎 Δ partial delta derivatives.
∑ ∑(−1)𝑖 ( ) ((𝑔𝑖𝑗𝑘 ) 𝐸𝑘 (𝐿))
𝑘=1 𝑖=0
𝑏1 For the convenience of the reader, we recall notions and
[ results that are needed in the sequel. A general introduction to
Δ𝑖 differential calculus and integration theory for multivariable
𝜕𝐿 Δ
𝜎 𝜕𝐿 𝜕𝐿
+((𝑓𝑖𝑗 ) ( − (𝐿 − 𝑦𝑘Δ Δ − 𝜇 ))) ] ≡ 0 functions on time scales is presented, respectively, in [41]
𝜕𝑡 Δ𝑡 𝜕𝑦𝑘 𝜕𝑡 (see also [42]) and [43]. For the double integral calculus of
] variations on time scales, we refer the reader to [44].
(60)
Let T1 and T2 be two given time scales. For 𝑖 = 1, 2,
for 𝑗 = 1, 2, . . . , 𝑟. denote by 𝜎𝑖 and Δ 𝑖 the forward jump operator and the delta
(1)
derivative on T𝑖 , respectively. Let 𝐶rd denote the set of all
Remark 23. Define continuous functions defined on T1 × T2 for which both the
𝑛
Δ 1 -partial derivative and the Δ 2 -partial derivative exist and
𝜕𝐿 Δ 𝜕𝐿 𝜕𝐿 are of class 𝐶rd (for a definition see [44]).
𝐸𝑘𝑠 (𝐿) := − (𝐿 − ∑ 𝑦𝑘Δ Δ − 𝜇 ) , (61)
𝜕𝑡 Δ𝑡 𝑘=1 𝜕𝑦𝑘 𝜕𝑡 Let Ω ⊆ T1 × T2 be an 𝜔-type set and let Γ be its positively
fence (see [44]). Denote
for 𝑘 = 1, 2, . . . , 𝑛. Then, 𝐸𝑘𝑠 (𝐿) = 0 are the second Euler-
Lagrange equations for problem (1) [36]. Therefore, expres- Ω∘ := {(𝑥, 𝑦) ∈ Ω : (𝜎1 (𝑥) , 𝜎2 (𝑦)) ∈ Ω} . (65)
sion (46) provides “dependencies” between two types of the Let a function 𝐿(𝑥, 𝑦, 𝑢, 𝑝, 𝑞), where (𝑥, 𝑦) ∈ Ω ∪ Γ and
Euler-Lagrange expressions. (𝑢, 𝑝, 𝑞) ∈ R3𝑛 , be given. We will suppose that 𝐿 is continuous,
together with its partial delta derivatives of first and second
Note that if T = R, Noether’s identity (46) simplifies
order with respect to 𝑥, 𝑦 and partial usual derivatives of the
because
first and second order with respect to 𝑢, 𝑝, 𝑞. In what follows,
𝜕𝐿 Δ 𝑛
𝜕𝐿 𝜕𝐿 𝑢Δ 1 and 𝑢Δ 2 denote, respectively, 𝜕𝑢/Δ 1 𝑥 and 𝜕𝑢/Δ 2 𝑦.
− (𝐿 − ∑ 𝑦𝑘Δ Δ − 𝜇 ) Consider the following optimization problem:
𝜕𝑡 Δ𝑡 𝑘=1 𝜕𝑦𝑘 𝜕𝑡
(62)
𝑛 𝑛 L [𝑢] = ∫ ∫ 𝐿 (𝑥, 𝑦, 𝑢 (𝜎1 (𝑥) , 𝜎2 (𝑦)) , 𝑢Δ 1 (𝑥, 𝜎2 (𝑦)) ,
𝜕𝐿 𝑑 𝜕𝐿 Ω
= − (𝐿 − ∑ 𝑦𝑘̇ ) = − ∑ 𝑦𝑘̇𝐸𝑘 (𝐿) ,
𝜕𝑡 𝑑𝑡 𝜕𝑦𝑘̇
𝑘=1 𝑘=1 𝑢Δ 2 (𝜎1 (𝑥) , 𝑦)) Δ1 𝑥Δ 2 𝑦 󳨀→ extremize,
and we obtain the following corollary. (66)

Corollary 24 (classical Noether’s second theorem, cf. [1]). If where the set of admissible functions are
functional L defined by D = {𝑢 | 𝑢 : Ω ∪ Γ 󳨀→ R𝑛 , 𝑢 ∈ 𝐶rd
(1)
, 𝑢 = 𝑔 on Γ} , (67)
𝑏
where 𝑔 is a fixed function defined and continuous on the
L [𝑦] = ∫ 𝐿 (𝑡, 𝑦 (𝑡) , 𝑦̇(𝑡)) 𝑑𝑡 (63)
𝑎 fence Γ of Ω.
As noticed in [44], for the variational problem (66) being
is invariant under transformations (39) (where 𝜎 denotes in well posed, we have to assume that there exists at least one
this context the identity function and Δ denotes the usual admissible function 𝑢0 ∈ D because it is possible to choose a
derivative), then there exist the following identities: continuous function 𝑔 such that no function 𝑢 is admissible.
𝑛 𝑚 Note that if there exists an admissible function 𝑢0 , then the set
(𝑖) (𝑖)
∑ ∑(−1)𝑖 [(𝑔𝑖𝑗𝑘 𝐸𝑘 (𝐿)) − (𝑓𝑖𝑗 ⋅ 𝑦𝑘̇𝐸𝑘 (𝐿)) ] ≡ 0, D contains a set of functions of the form 𝑢 = 𝑢0 + 𝜂, where
(1)
𝑘=1 𝑖=0 (64) 𝜂 : Ω ∪ Γ → R𝑛 is 𝐶rd and 𝜂 = 0 on Γ. Any such 𝜂 is called
an admissible variation for problem (66).
𝑗 = 1, 2, . . . , 𝑟.
Definition 25. We say that 𝑢∗ ∈ D is a local minimizer (resp.,
In the case T = ℎZ (for some ℎ > 0), we obtain from local maximizer) for problem (66) if there exists 𝛿 > 0 such
Theorem 22 the second Noether theorem for the ℎ-calculus; that
whereas if T = 𝑞N0 (for some 𝑞 > 1), we get the second
Noether theorem for the 𝑞-calculus. L [𝑢∗ ] ≤ L [𝑢] (resp., L [𝑢∗ ] ≥ L [𝑢]) , (68)
Abstract and Applied Analysis 9

for all 𝑢 ∈ D with ℎ > 0), 𝑞N0 (for some 𝑞 > 1), and many other interesting time
󵄩󵄩 󵄩 󵄨 󵄨 scales satisfy property (74).
󵄩󵄩𝑢 − 𝑢∗ 󵄩󵄩󵄩 := sup 󵄨󵄨󵄨𝑢 (𝑥, 𝑦) − 𝑢∗ (𝑥, 𝑦)󵄨󵄨󵄨 Let 𝑢(𝑥, 𝑦) = (𝑢1 (𝑥, 𝑦), 𝑢2 (𝑥, 𝑦), . . . , 𝑢𝑛 (𝑥, 𝑦)) and con-
(𝑥,𝑦)∈Ω∪Γ
sider the following transformations that depend on an arbi-
󵄨 󵄨 trary continuous function 𝑝 and the partial delta derivatives
+ sup 󵄨󵄨󵄨󵄨𝑢Δ 1 (𝑥, 𝜎2 (𝑦)) − 𝑢∗Δ 1 (𝑥, 𝜎2 (𝑦))󵄨󵄨󵄨󵄨
(𝑥,𝑦)∈Ω of 𝑝:
󵄨 󵄨
+ sup 󵄨󵄨󵄨󵄨𝑢Δ 2 (𝜎1 (𝑥) , 𝑦) − 𝑢∗Δ 2 (𝜎1 (𝑥) , 𝑦)󵄨󵄨󵄨󵄨 < 𝛿, 𝑥=𝑥
(𝑥,𝑦)∈Ω 𝑦=𝑦 (75)
(69)
𝑢𝑘 (𝑥, 𝑦) = 𝑢𝑘 (𝑥, 𝑦) + 𝑇𝑘 (𝑝) (𝑥, 𝑦) ,
where | ⋅ | denotes a norm in R𝑛 .

We recall the following results which will play an impor- where, for each 𝑘 = 1, 2, . . . , 𝑛,
tant role in the proofs of our results.
𝑇𝑘 (𝑝) (𝑥, 𝑦) := 𝑎0𝑘 (𝑥, 𝑦) 𝑝 (𝑥, 𝑦)
Theorem 26 (Green’s theorem [45]). If the functions 𝑀 and 𝜕
𝑁 are continuous and have continuous partial delta derivatives + 𝑎1𝑘 (𝑥, 𝑦) 𝑝 (𝜌1 (𝑥) , 𝑦)
𝜕𝑀/Δ 2 𝑦 and 𝜕𝑁/Δ 1 𝑥 on Ω ∪ Γ, then Δ 1𝑥 (76)

𝜕
𝜕𝑁 𝜕𝑀 + 𝑎2𝑘 (𝑥, 𝑦) 𝑝 (𝑥, 𝜌2 (𝑦)) ,
∫∫ ( − ) Δ 1 𝑥Δ 2 𝑦 = ∫ 𝑀𝑑∗ 𝑥 + 𝑁𝑑∗ 𝑦, Δ 2𝑦
Ω Δ 1𝑥 Δ 2𝑦 Γ
(70)
𝑎0 , 𝑎1 , and 𝑎2 are 𝐶1 functions and we assume that 𝑝 has
where the “star line integrals” on the right side in (70) denote continuous partial delta derivatives of the first and second
the sum of line delta integrals taken over the line segment order.
constituents of Γ directed to the right or upwards and line nabla
integrals taken over the line segment constituents of Γ directed Definition 29. Functional L is invariant under transforma-
to the left or downwards. tions (75) if, and only if, for all 𝑢 ∈ D, we have

Lemma 27 (fundamental lemma of the double variational ∫ ∫ 𝐿 (𝑥, 𝑦, 𝑢 (𝜎1 (𝑥) , 𝜎2 (𝑦)) , 𝑢Δ 1 (𝑥, 𝜎2 (𝑦)) ,
calculus [44]). If 𝑀 is continuous on Ω ∪ Γ with Ω

𝑢Δ 2 (𝜎1 (𝑥) , 𝑦)) Δ 1 𝑥Δ 2 𝑦


∫ ∫ 𝑀 (𝑥, 𝑦) 𝜂 (𝜎1 (𝑥) , 𝜎2 (𝑦)) Δ 1 𝑥Δ 2 𝑦 = 0 (71)
Ω
= ∫ ∫ 𝐿 (𝑥, 𝑦, 𝑢 (𝜎1 (𝑥) , 𝜎2 (𝑦)) , 𝑢Δ 1 (𝑥, 𝜎2 (𝑦)) ,
for any admissible variation 𝜂, then Ω

𝑀 (𝑥, 𝑦) = 0 ∀ (𝑥, 𝑦) ∈ Ω . (72) 𝑢Δ 2 (𝜎1 (𝑥) , 𝑦)) Δ 1 𝑥Δ 2 𝑦.
(77)
Theorem 28 (Euler-Lagrange equation of the double varia-
tional calculus [44]). Suppose that an admissible function 𝑢∗ In what follows, we use the notations
provides a local minimum for L and that 𝑢∗ has continuous
partial delta derivatives of the second order. Then 𝑢∗ satisfies 𝑇𝑘 (𝑝𝜎 ) (𝑥, 𝑦) := 𝑇𝑘 (𝜎1 (𝑥) , 𝜎2 (𝑦)) ,
the Euler-Lagrange equation
𝜕𝐿 𝜕 𝜕𝐿 𝜕 𝜕𝐿 𝑇𝑘 (𝑝𝜎1 ) (𝑥, 𝑦) := 𝑇𝑘 (𝜎1 (𝑥) , 𝑦) , (78)
(⋅) − (⋅) − (⋅) = 0, (73)
𝜕𝑢 Δ 1 𝑥 𝜕𝑝 Δ 2 𝑦 𝜕𝑞 𝑇𝑘 (𝑝𝜎2 ) (𝑥, 𝑦) := 𝑇𝑘 (𝑥, 𝜎2 (𝑦)) .
where (⋅) = (𝑥, 𝑦, 𝑢(𝜎1 (𝑥), 𝜎2 (𝑦)), 𝑢Δ 1 (𝑥, 𝜎2 (𝑦)), 𝑢Δ 2 (𝜎1 (𝑥),
Using similar arguments as the ones used in the proof of
𝑦)) for (𝑥, 𝑦) ∈ Ω∘ .
Theorem 15, we can prove the following result.
Let us denote by 𝜌1 and 𝜌2 the backward jump operator of
Theorem 30 (necessary condition of invariance). If func-
T1 and T2 , respectively. In what follows, we will suppose that
tional L is invariant under transformations (75), then
T1 and T2 are such that
𝑛
𝜎1 (𝜌1 (𝑥)) = 𝑥, ∀𝑥 ∈ (T1 )𝜅 , 𝜕𝐿 𝑘 𝜎 𝜕𝐿 𝜕 𝑘 𝜎2
∑ ∫∫ ( 𝜎 ⋅ 𝑇 (𝑝 ) + Δ
𝑇 (𝑝 )
(74)
𝑘=1 Ω 𝜕𝑢𝑘 𝜕𝑢𝑘 1 Δ 1 𝑥
𝜎2 (𝜌2 (𝑦)) = 𝑦, ∀𝑦 ∈ (T2 )𝜅 , (79)
where T𝜅 := T \ {𝑚} if T has a right-scattered minimum 𝑚; 𝜕𝐿 𝜕 𝑘 𝜎1
+ Δ
𝑇 (𝑝 )) Δ 1 𝑥Δ 2 𝑦 = 0.
otherwise, T𝜅 = T. We recall the fact that R, ℎZ (for some 𝜕𝑢𝑘 2 Δ 2𝑦
10 Abstract and Applied Analysis

Define and therefore,


𝜕𝐿 𝜕 𝜕𝐿 𝜕 𝜕𝐿 𝜕 𝜕𝐿
𝐸̂𝑘 (𝐿) := 𝜎 − − , 𝑘 = 1, 2, . . . , 𝑛. ∫∫ [ ( Δ ⋅ 𝑇𝑖 (𝑝𝜎2 ))
𝜕𝑢𝑘 Δ 1 𝑥 𝜕𝑢Δ 1 Δ 2 𝑦 𝜕𝑢Δ 2 Ω Δ 1 𝑥 𝜕𝑢 1
𝑘 𝑘 𝑖
(80) (85)
𝜕 𝜕𝐿
We call 𝐸̂𝑘 (𝐿), 𝑘 = 1, 2, . . . , 𝑛, the Euler-Lagrange expressions + ( ⋅ 𝑇𝑖 (𝑝𝜎1 ))] Δ 1 𝑥Δ 2 𝑦 = 0.
Δ 2 𝑦 𝜕𝑢Δ 2
associated to the Lagrangian 𝐿 relative to problem (66). 𝑖

The following lemmas will be used in the proof of By Theorem 30, we obtain
Theorem 34. 𝑛
𝜕𝐿 𝜕 𝜕𝐿
∑ ∫∫ ( ⋅ 𝑇𝑘 (𝑝𝜎 ) − ⋅ 𝑇𝑘 (𝑝𝜎 )
Lemma 31. If L is invariant under transformations (75), then 𝑘=1 Ω 𝜕𝑢𝑘𝜎 Δ 1 𝑥 𝜕𝑢Δ 1
𝑘
𝑛 (86)
∑ ∫ ∫ 𝐸̂𝑘 (𝐿) ⋅ 𝑇𝑘 (𝑝𝜎 ) Δ 1 𝑥Δ 2 𝑦 = 0. (81) −
𝜕 𝜕𝐿
⋅ 𝑇𝑘 (𝑝𝜎 )) Δ 1 𝑥Δ 2 𝑦 = 0,
𝑘=1 Ω Δ 2 𝑦 𝜕𝑢Δ 2
𝑘
Proof. Fix 𝑖 ∈ {1, 2, . . . , 𝑛}. Observe that
which proves that ∑𝑛𝑘=1 ∫ ∫Ω 𝐸̂𝑘 (𝐿) ⋅ 𝑇𝑘 (𝑝𝜎 )Δ 1 𝑥Δ 2 𝑦 = 0.
𝜕𝐿 𝜕 𝑖 𝜎2 𝜕𝐿 𝜕 𝑖 𝜎1 Lemma 32. For each 𝑘 = 1, 2, . . . , 𝑛,
∫∫ ( Δ
𝑇 (𝑝 ) + Δ 𝑇 (𝑝 )) Δ 1 𝑥Δ 2 𝑦
Ω 𝜕𝑢𝑖 1 Δ 1𝑥 𝜕𝑢𝑖 2 Δ 2 𝑦
∫ ∫ 𝑞 ⋅ 𝑇𝑘 (𝑝𝜎 ) Δ 1 𝑥Δ 2 𝑦
Ω
𝜕 𝜕𝐿
= ∫∫ [ ( ⋅ 𝑇𝑖 (𝑝𝜎2 ))
Δ 1 𝑥 𝜕𝑢Δ 1 𝜕 𝜕
Ω
𝑖 = ∫ ∫ (𝑞𝑎0𝑘 − (𝑞𝑎1𝑘 ) − (𝑞𝑎2𝑘 )) ⋅ 𝑝𝜎 Δ 1 𝑥Δ 2 𝑦
Ω Δ 1𝑥 Δ 2𝑦
𝜕 𝜕𝐿 (87)
+ ( ⋅ 𝑇𝑖 (𝑝𝜎1 ))] Δ 1 𝑥Δ 2 𝑦
Δ 2 𝑦 𝜕𝑢Δ 2 holds.
𝑖

𝜕 𝜕𝐿 𝜕 𝜕𝐿 Proof. Note that


−∫∫ [ ⋅𝑇𝑖 (𝑝𝜎 )+ ⋅𝑇𝑖 (𝑝𝜎 )] Δ 1 𝑥Δ 2 𝑦.
Ω Δ 1 𝑥 𝜕𝑢Δ 1 Δ 2 𝑦 𝜕𝑢Δ 2
𝑖 𝑖
(82) ∫ ∫ 𝑞 ⋅ 𝑇𝑘 (𝑝𝜎 ) Δ 1 𝑥Δ 2 𝑦
Ω

Using Green’s theorem we get


= ∫ ∫ [𝑞𝑎0𝑘 (𝑥, 𝑦) 𝑝𝜎 (𝑥, 𝑦)
Ω
𝜕 𝜕𝐿
∫∫ [ ( Δ ⋅ 𝑇𝑖 (𝑝𝜎2 )) 𝜕
Ω Δ 1𝑥 𝜕𝑢 1 + 𝑞𝑎1𝑘 (𝑥, 𝑦) 𝑝 (𝑥, 𝜎2 (𝑦))
𝑖
Δ 1𝑥
𝜕 𝜕𝐿 𝜕
+ ( ⋅ 𝑇𝑖 (𝑝𝜎1 ))] Δ 1 𝑥Δ 2 𝑦 (83)
+𝑞𝑎2𝑘 (𝑥, 𝑦) 𝑝 (𝜎1 (𝑥) , 𝑦)] Δ 1 𝑥Δ 2 𝑦,
Δ 2 𝑦 𝜕𝑢Δ 2 Δ 2𝑦
𝑖
(88)
𝜕𝐿 𝜕𝐿
=∫ Δ1
⋅ 𝑇𝑖 (𝑝𝜎2 ) 𝑑∗ 𝑦 − Δ2
⋅ 𝑇𝑖 (𝑝𝜎1 ) 𝑑∗ 𝑥.
Γ 𝜕𝑢𝑖 𝜕𝑢𝑖 𝜕
∫ ∫ [𝑞𝑎1𝑘 (𝑥, 𝑦) 𝑝 (𝑥, 𝜎2 (𝑦))
Since 𝑝 is arbitrary, we can choose 𝑝 such that Ω Δ 1𝑥
󵄨 𝜕
𝑝 (𝑥, 𝜎2 (𝑦))󵄨󵄨󵄨Γ = 0, +𝑞𝑎2𝑘 (𝑥, 𝑦) 𝑝 (𝜎1 (𝑥) , 𝑦)] Δ 1 𝑥Δ 2 𝑦
Δ 2𝑦
󵄨
𝑝 (𝜎1 (𝑥) , 𝑦)󵄨󵄨󵄨Γ = 0,
𝜕
󵄨󵄨 = ∫∫ [ (𝑞𝑎1𝑘 (𝑥, 𝑦) 𝑝 (𝑥, 𝜎2 (𝑦)))
𝜕 󵄨 Δ 1𝑥
𝑝 (𝜌1 (𝑥) , 𝜎2 (𝑦))󵄨󵄨󵄨 = 0, Ω
Δ 1𝑥 󵄨󵄨Γ
𝜕
󵄨󵄨 + (𝑞𝑎2𝑘 (𝑥, 𝑦) 𝑝 (𝜎1 (𝑥) , 𝑦))] Δ 1 𝑥Δ 2 𝑦
𝜕 󵄨 Δ 2𝑦
𝑝 (𝑥, 𝑦)󵄨󵄨󵄨 = 0, (84)
Δ 1𝑥 󵄨󵄨Γ
𝜕
󵄨󵄨 − ∫∫ [ (𝑞𝑎1𝑘 (𝑥, 𝑦)) ⋅ 𝑝 (𝜎1 (𝑥) , 𝜎2 (𝑦))
𝜕 󵄨 Δ 1𝑥
𝑝 (𝑥, 𝑦)󵄨󵄨󵄨 = 0, Ω
Δ 2𝑦 󵄨󵄨Γ
𝜕
󵄨󵄨 + (𝑞𝑎2𝑘 (𝑥, 𝑦)) ⋅ 𝑝 (𝜎1 (𝑥) , 𝜎2 (𝑦))] Δ 1 𝑥Δ 2 𝑦.
𝜕 󵄨 Δ 2𝑦
𝑝 (𝜎1 (𝑥) , 𝜌2 (𝑦))󵄨󵄨󵄨 = 0,
Δ 2𝑦 󵄨󵄨Γ (89)
Abstract and Applied Analysis 11

Using Green’s theorem, we can conclude that Corollary 35 (classical Noether’s second theorem for double
integrals problems, cf. [1]). Let Ω ⊆ R2 be an 𝜔-type set and
𝜕 let Γ be its positive fence. Let 𝐿(𝑥, 𝑦, 𝑢, 𝑝, 𝑞) be a function of
∫∫ [ (𝑞𝑎1𝑘 (𝑥, 𝑦) 𝑝 (𝑥, 𝜎2 (𝑦)))
Ω Δ 1𝑥 class 𝐶2 , (𝑥, 𝑦) ∈ Ω ∪ Γ, 𝑢 = (𝑢1 , 𝑢2 , . . . , 𝑢𝑛 ). If functional L
defined by
𝜕
+ (𝑞𝑎2𝑘 (𝑥, 𝑦) 𝑝 (𝜎1 (𝑥) , 𝑦))] Δ 1 𝑥Δ 2 𝑦 = 0.
Δ 2𝑦 𝜕𝑢 𝜕𝑢
(90) L [𝑦] = ∫ ∫ 𝐿 (𝑥, 𝑦, 𝑢 (𝑥, 𝑦) , (𝑥, 𝑦) , (𝑥, 𝑦)) 𝑑𝑥 𝑑𝑦
Ω 𝜕𝑥 𝜕𝑦
(96)
Hence,
is invariant under transformations (75) (where 𝜌1 and 𝜌2
∫ ∫ 𝑞 ⋅ 𝑇𝑘 (𝑝𝜎 ) Δ 1 𝑥Δ 2 𝑦 denote in this context the identity function, and Δ 1 and Δ 2
Ω denote the usual derivative), then
𝜕
= ∫ ∫ [𝑞𝑎0𝑘 ⋅ 𝑝𝜎 − (𝑞𝑎1𝑘 ) ⋅ 𝑝𝜎 (91) 𝑛
Ω Δ 1𝑥 ̃ 𝑘 (𝐸̂𝑘 (𝐿)) ≡ 0 𝑜𝑛 Ω,
∑𝑇 (97)
𝑘=1
𝜕
− (𝑞𝑎2𝑘 ) ⋅ 𝑝𝜎 ] Δ 1 𝑥Δ 2 𝑦
Δ 2𝑦 where
𝜕𝐿 𝜕 𝜕𝐿 𝜕 𝜕𝐿
proving the desired result. 𝐸̂𝑘 (𝐿) := − − , 𝑘 = 1, 2, . . . , 𝑛 (98)
𝜕𝑢𝑘 𝜕𝑥 𝜕𝑝𝑘 𝜕𝑦 𝜕𝑞𝑘
Remark 33. Lemma 32 shows that we can define an adjoint
̃ 𝑘 , by
operator of 𝑇𝑘 , 𝑇 ̃ 𝑘 is the adjoint operator of 𝑇𝑘 .
and 𝑇

Similarly to the single delta integral case choosing T =


̃ 𝑘 (𝑞) = 𝑞𝑎𝑘 − 𝜕 (𝑞𝑎𝑘 ) − 𝜕 (𝑞𝑎𝑘 ) .
𝑇 (92)
0
Δ 1𝑥 1
Δ 2𝑦 2 ℎZ (for some ℎ > 0), we obtain from Theorem 34 the sec-
ond Noether theorem for the double variational ℎ-calculus;
We are now ready to state and prove the Noether second whereas choosing T = 𝑞N0 (for some 𝑞 > 1), we obtain
theorem without transformation of time for multiple integral the second Noether theorem for the double variational 𝑞-
problems on time scales. calculus.

Theorem 34 (Noether’s second theorem without transform- 4. Example


ing time). If functional L is invariant under transformations
(75), then In order to illustrate the second Noether Theorem for the
𝑛
multiple integral case, we will present the following example.
̃ 𝑘 (𝐸̂𝑘 (𝐿)) ≡ 0 𝑜𝑛 Ω∘ ,
∑𝑇 (93) Let T0 , T1 , T2 , and T3 be time scales and let Ω ⊆ T0 ×T1 ×T2 ×T3
𝑘=1 be an 𝜔-type set. For 𝑖 = 0, 1, 2, 3, denote by 𝜎𝑖 , 𝜌𝑖 , and Δ 𝑖 the
forward jump operator, the backward jump operator, and the
̃ 𝑘 is the
where 𝐸̂𝑘 (𝐿) are the 𝑛 Euler-Lagrange expressions and 𝑇 delta derivative on T𝑖 , respectively.
adjoint operator of 𝑇 .𝑘 Let 𝑡 := (𝑡0 , 𝑡1 , 𝑡2 , 𝑡3 ) ∈ Ω and consider the following
real functions defined on Ω: 𝐴 0 , 𝐴 1 , 𝐴 2 , 𝐴 3 . Let A :=
Proof. Using Lemmas 31 and 32, we conclude that if L is (𝐴 1 , 𝐴 2 , 𝐴 3 ) and denote
invariant under transformations (75), then
𝜕𝐴 0
∇𝐴 0 (𝑡) := ( (𝜎 (𝑡 ) , 𝑡 , 𝜎 (𝑡 ) , 𝜎3 (𝑡3 )) ,
𝑛 Δ 1 𝑡1 0 0 1 2 2
∑ ∫ ∫ 𝐸̂𝑘 (𝐿) ⋅ 𝑇𝑘 (𝑝𝜎 ) Δ 1 𝑥Δ 2 𝑦
𝑘=1 Ω 𝜕𝐴 0
(94) (𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝑡2 , 𝜎3 (𝑡3 )) ,
𝑛 Δ 2 𝑡2 0 0
̃ 𝑘 (𝐸̂𝑘 (𝐿)) ⋅ 𝑝𝜎 Δ 1 𝑥Δ 2 𝑦 = 0,
= ∑ ∫∫ 𝑇
Ω 𝜕𝐴 0
𝑘=1 (𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝜎2 (𝑡2 ) , 𝑡3 )) ,
Δ 3 𝑡3 0 0
where 𝑇̃ 𝑘 is the adjoint operator of 𝑇𝑘 . Applying the funda-
𝜕A 𝜕𝐴
mental lemma of the double variational calculus (Lemma 27), (𝑡) := ( 1 (𝑡0 , 𝜎1 (𝑡1 ) , 𝜎2 (𝑡2 ) , 𝜎3 (𝑡3 )) ,
we get Δ 0 𝑡0 Δ 0 𝑡0

𝑛 𝜕𝐴 2
(𝑡 , 𝜎 (𝑡 ) , 𝜎2 (𝑡2 ) , 𝜎3 (𝑡3 )) ,
̃ 𝑘 (𝐸̂𝑘 (𝐿)) ≡ 0
∑𝑇 on Ω , ∘
(95) Δ 0 𝑡0 0 1 1
𝑘=1
𝜕𝐴 3
(𝑡 , 𝜎 (𝑡 ) , 𝜎2 (𝑡2 ) , 𝜎3 (𝑡3 ))) ,
proving the desired result. Δ 0 𝑡0 0 1 1
12 Abstract and Applied Analysis

𝜕𝐴 3 order and that 𝐴 0 and the vector field A satisfy the so-called
curl A (𝑡) := ( (𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝑡2 , 𝜎3 (𝑡3 ))
Δ 2 𝑡2 0 0 Lorentz conditions on time scales:
𝜕𝐴 2 𝜕𝐴 0 󵄨󵄨󵄨󵄨
− (𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝜎2 (𝑡2 ) , 𝑡3 ) , div A|(𝑡0 ,𝜎1 (𝑡1 ),𝜎2 (𝑡2 ),𝜎3 (𝑡3 )) = 󵄨
Δ 3 𝑡3 0 0 Δ 0 𝑡0 󵄨󵄨󵄨(𝑡0 ,𝜎1 (𝑡1 ),𝜎2 (𝑡2 ),𝜎3 (𝑡3 ))
𝜕𝐴 1 𝜕𝐴 0 󵄨󵄨󵄨󵄨
(𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝜎2 (𝑡2 ) , 𝑡3 ) div A|(𝜎0 (𝑡0 ),𝑡1 ,𝜎2 (𝑡2 ),𝜎3 (𝑡3 )) = 󵄨
Δ 3 𝑡3 0 0 Δ 0 𝑡0 󵄨󵄨󵄨(𝜎0 (𝑡0 ),𝑡1 ,𝜎2 (𝑡2 ),𝜎3 (𝑡3 ))
(106)

𝜕𝐴 3
(𝜎 (𝑡 ) , 𝑡 , 𝜎 (𝑡 ) , 𝜎3 (𝑡3 )) , 𝜕𝐴 0 󵄨󵄨󵄨󵄨
Δ 1 𝑡1 0 0 1 2 2 div A|(𝜎0 (𝑡0 ),𝜎1 (𝑡1 ),𝑡2 ,𝜎3 (𝑡3 )) = 󵄨
Δ 0 𝑡0 󵄨󵄨󵄨(𝜎0 (𝑡0 ),𝜎1 (𝑡1 ),𝑡2 ,𝜎3 (𝑡3 ))
𝜕𝐴 2
(𝜎 (𝑡 ) , 𝑡 , 𝜎 (𝑡 ) , 𝜎3 (𝑡3 )) 𝜕𝐴 0 󵄨󵄨󵄨󵄨
Δ 1 𝑡1 0 0 1 2 2 div A|(𝜎0 (𝑡0 ),𝜎1 (𝑡1 ),𝜎2 (𝑡2 ),𝑡3 ) = 󵄨 ,
Δ 0 𝑡0 󵄨󵄨󵄨(𝜎0 (𝑡0 ),𝜎1 (𝑡1 ),𝜎2 (𝑡2 ),𝑡3 )
𝜕𝐴 1
− (𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝑡2 , 𝜎3 (𝑡3 ))) . where div A denotes the divergence of a vector field A, that is,
Δ 2 𝑡2 0 0
(99) 𝜕𝐴 1 𝜕𝐴 2 𝜕𝐴 3
div A := + + , (107)
Δ 1 𝑡1 Δ 2 𝑡2 Δ 3 𝑡3
We will consider the following Lagrangian function:
then the Euler-Lagrange expressions can be written in the
1 󵄩󵄩󵄩 𝜕A 󵄩󵄩󵄩󵄩2 1 following way:
𝐿 = 󵄩󵄩󵄩∇𝐴 0 − 󵄩 − ‖curl A‖
2
(100)
2 󵄩󵄩 Δ 0 𝑡0 󵄩󵄩󵄩 2
𝜕2 𝐴 𝑘
𝐸̂𝑘 (𝐿) = (𝑡 , 𝜎 (𝑡 ) , 𝜎2 (𝑡2 ) , 𝜎3 (𝑡3 ))
that is the time scale version of the Lagrangian density for the Δ 0 𝑡02 0 1 1 (108)
electromagnetic field (see, e.g., [46]). 2
It can be proved that the functional − ∇ 𝐴 𝑘 (𝑡0 , 𝑡1 , 𝑡2 , 𝑡3 ) , 𝑘 = 0, 1, 2, 3,

where
L = ∫ ⋅ ⋅ ⋅ ∫ 𝐿 Δ 0Δ 1Δ 2Δ 3 (101)
Ω 𝜕2 𝐴 𝑘
∇2 𝐴 𝑘 (𝑡) := (𝜎 (𝑡 ) , 𝑡 , 𝜎 (𝑡 ) , 𝜎3 (𝑡3 ))
is invariant under the gauge transformations: Δ 1 𝑡12 0 0 1 2 2

𝜕 𝜌𝑘 𝜕2 𝐴 𝑘
𝐴𝑘 = 𝐴 𝑘 + 𝑝 , 𝑘 = 0, 1, 2, 3, + (𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝑡2 , 𝜎3 (𝑡3 )) (109)
Δ 𝑘 𝑡𝑘
(102) Δ 2 𝑡22 0 0

where 𝑝 : Ω → 𝑅 is an arbitrary continuous function 𝜕2 𝐴 𝑘


+ (𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝜎2 (𝑡2 ) , 𝑡3 ) .
that has continuous partial delta derivatives of the first and Δ 3 𝑡32 0 0
second order (hence, we have equality of mixed partial delta
derivatives; see [41]). Hence, under these assumptions, we can conclude that
Since, for each 𝑘 = 0, 1, 2, 3, 3
𝜕 𝜕2 𝐴 𝑘
∑ ( (𝑡 , 𝜎 (𝑡 ) , 𝜎2 (𝑡2 ) , 𝜎3 (𝑡3 ))
𝑇𝑘 (𝑝) =
𝜕 𝜌𝑘
𝑝 , Δ 𝑡
𝑘=0 𝑘 𝑘
Δ 0 𝑡02 0 1 1
(103)
Δ 𝑘 𝑡𝑘 (110)

then, by Lemma 32, we conclude that −∇2 𝐴 𝑘 (𝑡0 , 𝑡1 , 𝑡2 , 𝑡3 ) ) = 0 on Ω∘ .

̃ 𝑘 (𝑞) = − 𝜕 𝑞.
𝑇 (104)
Δ 𝑘 𝑡𝑘 5. Concluding Remarks
We proved that the important Noether’s second theorem is
Hence, from the second Noether theorem (Theorem 34), we
valid not only for the continuous and discrete calculus but
get
also for the quantum calculus. Moreover, in our opinion, the
3 proofs presented in this paper are elegant and clear to follow.
𝜕 ̂
∑ 𝐸 (𝐿) ≡ 0 on Ω∘ , (105) The question of obtaining Noether’s second theorem for
Δ 𝑡 𝑘
𝑘=0 𝑘 𝑘 multiple integrals with transformation of time in the time
scale setting remains an interesting open question. To the
where 𝐸̂𝑘 (𝐿), 𝑘 = 0, 1, 2, 3 are the Euler-Lagrange expressions best of the authors’ knowledge, to extend the second Noether
associated to functional L. theorem to multiple integrals with transformation of time,
If we suppose that, for each 𝑘 = 0, 1, 2, 3, 𝐴 𝑘 has substitution in the multiple integral is a fundamental tool and
continuous partial delta derivatives of the first and second this result is not yet available in the literature.
Abstract and Applied Analysis 13

For other generalizations of the second Noether theorem, [12] G. Bangerezako, “Variational 𝑞-calculus,” Journal of Mathemati-
we refer the reader to [47] (in the context of the fractional cal Analysis and Applications, vol. 289, no. 2, pp. 650–665, 2004.
calculus of variations) and [48] (in the context of the optimal [13] J. Baoguo, L. Erbe, and A. Peterson, “Oscillation of a family of 𝑞-
control). difference equations,” Applied Mathematics Letters, vol. 22, no.
6, pp. 871–875, 2009.
[14] A. M. C. Brito da Cruz, N. Martins, and D. F. M. Torres, “Higher-
Acknowledgments order Hahn’s quantum variational calculus,” Nonlinear Analysis:
Theory, Methods & Applications, vol. 75, no. 3, pp. 1147–1157,
This work was supported by FEDER funds through COM- 2012.
PETE-Operational Programme Factors of Competitiveness
[15] A. M. C. Brito da Cruz and N. Martins, “The 𝑞-symmetric vari-
(“Programa Operacional Factores de Competitividade”) and ational calculus,” Computers & Mathematics with Applications,
by Portuguese funds through the Center for Research and vol. 64, no. 7, pp. 2241–2250, 2012.
Development in Mathematics and Applications (University [16] A. M. C. Brito da Cruz, N. Martins, and D. F. M. Torres, “Hahn’s
of Aveiro) and the Portuguese Foundation for Science and symmetric quantum variational calculus,” Numerical Algebra,
Technology (“FCT-Fundação para a Ciência e a Tecnologia”), Control and Optimization, vol. 3, no. 1, pp. 77–94, 2013.
within Project PEst-C/MAT/UI4106/2011 with COMPETE [17] J. Cresson, G. S. F. Frederico, and D. F. M. Torres, “Constants of
no. FCOMP-01-0124-FEDER-022690. Agnieszka B. Malino- motion for non-differentiable quantum variational problems,”
wska was also supported by Bialystok University of Technol- Topological Methods in Nonlinear Analysis, vol. 33, no. 2, pp. 217–
ogy Grant S/WI/02/2011. 231, 2009.
[18] A. B. Malinowska and N. Martins, “Generalized transversality
conditions for the Hahn quantum variational calculus,” Opti-
References mization, vol. 62, no. 3, pp. 323–344, 2013.
[1] E. Noether, “Invariante variationsprobleme,” Nachrichten von [19] A. B. Malinowska and D. F. M. Torres, “The Hahn quantum
der Gesellschaft der Wissenschaften zu Göttingen, pp. 235–257, variational calculus,” Journal of Optimization Theory and Appli-
1918. cations, vol. 147, no. 3, pp. 419–442, 2010.
[2] Y. Kosmann-Schwarzbach, The Noether Theorems: Invariance [20] N. Martins and D. F. M. Torres, “Higher-order infinite horizon
and Conservation Laws in the Twentieth Century, Sources and variational problems in discrete quantum calculus,” Computers
Studies in the History of Mathematics and Physical Sciences, & Mathematics with Applications, vol. 64, no. 7, pp. 2166–2175,
Springer, New York, NY, USA, 2011. 2012.
[3] T. M. Atanacković, S. Konjik, S. Pilipović, and S. Simić, “Vari- [21] V. Kac and P. Cheung, Quantum Calculus, Springer, New York,
ational problems with fractional derivatives: invariance condi- NY, USA, 2002.
tions and Noether’s theorem,” Nonlinear Analysis: Theory, Meth- [22] S. Hilger, Ein maÿkettenkalkül mit anwendung auf zentrums-
ods & Applications, vol. 71, no. 5-6, pp. 1504–1517, 2009. mannigfaltigkeiten [Ph.D. thesis], Universität Würzburg, Wür-
zburg, Germany, 1988.
[4] Z. Bartosiewicz and D. F. M. Torres, “Noether’s theorem on time
scales,” Journal of Mathematical Analysis and Applications, vol. [23] M. Bohner, “Calculus of variations on time scales,” Dynamic
342, no. 2, pp. 1220–1226, 2008. Systems and Applications, vol. 13, no. 3-4, pp. 339–349, 2004.
[5] G. S. F. Frederico and D. F. M. Torres, “A formulation of Noe- [24] R. Hilscher and V. Zeidan, “Calculus of variations on time sca-
1
ther’s theorem for fractional problems of the calculus of vari- les: weak local piecewise 𝐶𝑟𝑑 solutions with variable endpoints,”
ations,” Journal of Mathematical Analysis and Applications, vol. Journal of Mathematical Analysis and Applications, vol. 289, no.
334, no. 2, pp. 834–846, 2007. 1, pp. 143–166, 2004.
[25] R. Almeida and D. F. M. Torres, “Isoperimetric problems on
[6] G. S. F. Frederico and D. F. M. Torres, “Noether’s symmetry
time scales with nabla derivatives,” Journal of Vibration and
theorem for variational and optimal control problems with time
Control, vol. 15, no. 6, pp. 951–958, 2009.
delay,” Numerical Algebra, Control and Optimization, vol. 2, no.
3, pp. 619–630, 2012. [26] M. J. Bohner, R. A. C. Ferreira, and D. F. M. Torres, “Integral
inequalities and their applications to the calculus of variations
[7] A. B. Malinowska, “A formulation of the fractional Noether- on time scales,” Mathematical Inequalities & Applications, vol.
type theorem for multidimensional Lagrangians,” Applied 13, no. 3, pp. 511–522, 2010.
Mathematics Letters, vol. 25, no. 11, pp. 1941–1946, 2012.
[27] R. A. C. Ferreira and D. F. M. Torres, “Higher-order calculus of
[8] A. B. Malinowska and D. F. M. Torres, Introduction to the Frac- variations on time scales,” in Mathematical Control Theory and
tional Calculus of Variations, Imperial College Press, London, Finance, pp. 149–159, Springer, Berlin, Germany, 2008.
UK, 2012.
[28] R. A. C. Ferreira and D. F. M. Torres, “Isoperimetric problems of
[9] N. Martins and D. F. M. Torres, “Noether’s symmetry theorem the calculus of variations on time scales,” in Nonlinear Analysis
for nabla problems of the calculus of variations,” Applied Math- and Optimization II. Optimization, vol. 514 of Contemporary
ematics Letters, vol. 23, no. 12, pp. 1432–1438, 2010. Mathematics, pp. 123–131, American Mathematical Society,
[10] J. D. Logan, “On variational problems which admit an infinite Providence, RI, USA, 2010.
continuous group,” Yokohama Mathematical Journal, vol. 22, pp. [29] E. Girejko, A. B. Malinowska, and D. F. M. Torres, “The con-
31–42, 1974. tingent epiderivative and the calculus of variations on time
[11] K. A. Aldwoah, A. B. Malinowska, and D. F. M. Torres, “The scales,” Optimization, vol. 61, no. 3, pp. 251–264, 2012.
power quantum calculus and variational problems,” Dynamics [30] A. B. Malinowska, N. Martins, and D. F. M. Torres, “Transversal-
of Continuous, Discrete and Impulsive Systems B, vol. 19, no. 1-2, ity conditions for infinite horizon variational problems on time
pp. 93–116, 2012. scales,” Optimization Letters, vol. 5, no. 1, pp. 41–53, 2011.
14 Abstract and Applied Analysis

[31] A. B. Malinowska and D. F. M. Torres, “Leitmann’s direct


method of optimization for absolute extrema of certain prob-
lems of the calculus of variations on time scales,” Applied Math-
ematics and Computation, vol. 217, no. 3, pp. 1158–1162, 2010.
[32] A. B. Malinowska and D. F. M. Torres, “A general backwards
calculus of variations via duality,” Optimization Letters, vol. 5,
no. 4, pp. 587–599, 2011.
[33] N. Martins and D. F. M. Torres, “Calculus of variations on time
scales with nabla derivatives,” Nonlinear Analysis: Theory, Meth-
ods & Applications, vol. 71, no. 12, pp. e763–e773, 2009.
[34] N. Martins and D. F. M. Torres, “Generalizing the variational
theory on time scales to include the delta indefinite integral,”
Computers & Mathematics with Applications, vol. 61, no. 9, pp.
2424–2435, 2011.
[35] N. Martins and D. F. M. Torres, “Necessary optimality condi-
tions for higher-order infinite horizon variational problems on
time scales,” Journal of Optimization Theory and Applications,
vol. 155, no. 2, pp. 453–476, 2012.
[36] Z. Bartosiewicz, N. Martins, and D. F. M. Torres, “The second
Euler-Lagrange equation of variational calculus on time scales,”
European Journal of Control, vol. 17, no. 1, pp. 9–18, 2011.
[37] M. Bohner and A. Peterson, Dynamic Equations on Time Scales,
Birkhäauser, Boston, Mass, USA, 2001.
[38] M. Bohner and A. Peterso, Eds., Advances in Dynamic Equations
on Time Scales, Birkhäauser, Boston, Mass, USA, 2003.
[39] P. E. Hydon and E. L. Mansfield, “Extensions of Noether’s sec-
ond theorem: from continuous to discrete systems,” Proceedings
of The Royal Society of London A, vol. 467, no. 2135, pp. 3206–
3221, 2011.
[40] C. D. Ahlbrandt, M. Bohner, and J. Ridenhour, “Hamiltonian
systems on time scales,” Journal of Mathematical Analysis and
Applications, vol. 250, no. 2, pp. 561–578, 2000.
[41] M. Bohner and G. Sh. Guseinov, “Partial differentiation on time
scales,” Dynamic Systems and Applications, vol. 13, no. 3-4, pp.
351–379, 2004.
[42] M. Z. Sarikaya, N. Aktan, H. Yıldırım, and K. İlarslan, “Partial
Δ-differentiation for multivariable functions on 𝑛-dimensional
time scales,” Journal of Mathematical Inequalities, vol. 3, no. 2,
pp. 277–291, 2009.
[43] M. Bohner and G. Sh. Guseinov, “Multiple integration on time
scales,” Dynamic Systems and Applications, vol. 14, no. 3-4, pp.
579–606, 2005.
[44] M. Bohner and G. Sh. Guseinov, “Double integral calculus
of variations on time scales,” Computers & Mathematics with
Applications, vol. 54, no. 1, pp. 45–57, 2007.
[45] M. Bohner and G. Sh. Guseinov, “Line integrals and Green’s
formula on time scales,” Journal of Mathematical Analysis and
Applications, vol. 326, no. 2, pp. 1124–1141, 2007.
[46] J. D. Logan, Invariant Variational Principles, Academic Press,
New York, NY, USA, 1977.
[47] A. B. Malinowska, “On fractional variational problems which
admit local transformations,” Journal of Vibration and Control,
vol. 19, no. 8, pp. 1161–1169, 2013.
[48] D. F. M. Torres, “Gauge symmetries and Noether currents in
optimal control,” Applied Mathematics E-Notes, vol. 3, pp. 49–
57, 2003.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 178378, 13 pages
http://dx.doi.org/10.1155/2013/178378

Research Article
On Solutions of Linear Fractional Differential
Equations with Uncertainty

T. Allahviranloo,1 S. Abbasbandy,2 M. R. Balooch Shahryari,2


S. Salahshour,3 and D. Baleanu4,5,6
1
Department of Electronic and Communications, Faculty of Engineering, Izmir University, Izmir, Turkey
2
Department of Mathematics, Science and Research Branch, Islamic Azad University, Tehran, Iran
3
Young Researchers and Elite Club, Mobarakeh Branch, Islamic Azad University, Mobarakeh, Iran
4
Department of Chemical and Materials Engineering, Faculty of Engineering, King Abdulaziz University, Jeddah 21589, Saudi Arabia
5
Department of Mathematics and Computer Sciences, Faculty of Arts and Sciences, Cankaya University, 06530 Ankara, Turkey
6
Institute of Space Sciences, Magurele-Bucharest, Bucharest, RO 7690, Romania

Correspondence should be addressed to S. Salahshour; soheilsalahshour@yahoo.com

Received 8 June 2013; Revised 29 August 2013; Accepted 30 August 2013

Academic Editor: Ali H. Bhrawy

Copyright © 2013 T. Allahviranloo et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The solutions of linear fuzzy fractional differential equations (FFDEs) under the Caputo differentiability have been investigated.
To this end, the fuzzy Laplace transform was used to obtain the solutions of FFDEs. Then, some new results regarding the relation
between some types of differentiability have been obtained. Finally, some applicable examples are solved in order to show the ability
of the proposed method.

1. Introduction So, we adopt a generalization of a strongly generalized


differentiability to a fractional case.
The topic of fuzzy differential equations (FDEs) has been In the following, we propose the Riemann-Liouville dif-
rapidly growing in recent years (see, e.g., [1–8] and the ferentiability by using Hukuhara difference which is named
references therein). Recently, applying fractional differential as the Riemann-Liouville H-differentiability. Therefore, a
equations has grown increasingly and have excited as a direct procedure is adopted to derive such concept which is
considerable interest both in mathematics and applications, constructed based on the combination of strongly general-
such that they were used in modeling of many physical and ized differentiability [21] and the Riemann-Liouville deriva-
chemical processes and in engineering [9–15]. tive [14, 15]. Moreover, we suggest the concept of frac-
In addition, many research papers are published to tional derivatives under Caputo’s differentiability by applying
consider solutions of fractional differential equations (see, the Hukuhara difference which is named as Caputo’s H-
e.g., [16–18] and the references therein). differentiability. Similar to the deterministic cases, construc-
Recently, Agarwal et al. [19] proposed the concept of solu- tion of Caputo’s derivatives is based on the definitions of the
tions for fractional differential equations with uncertainty. Riemann-Liouville derivatives in fuzzy cases.
They considered the Riemann-Liouville differentiability to Consequently, we intend to propose an analytical method
solve FFDEs which is a combination of Hukuhara difference to solve FFDEs. Since, considering solutions of FFDEs is
and Riemann-Liouville derivative. There are some basic a new subject, presets the fractional Green’s functions for
papers which are written by Bede and Gal [20] who discussed fuzzy fractional differential equations is considered and, as
on shortcomings of applications of Hukuhara difference. particular cases, we obtain the classical harmonic oscillator,
2 Abstract and Applied Analysis

the damped harmonic oscillator, relaxation equation, all in According to Zadeh’s extension principle, operation of
the fuzzy fractional versions by using fuzzy laplace trans- addition on E is defined by
forms; so we should first implement analytical method to
solve it; then numerical methods can be applied. To this end, (𝑢 ⊕ V) (𝑥) = sup min {𝑢 (𝑦) , V (𝑥 − 𝑦)} , 𝑥 ∈ R, (1)
we adopt fractional Green’s functions to solve FFDEs by using 𝑦∈R

fuzzy Laplace transforms method. One can see some useful


papers about fuzzy Laplace transforms in [2, 22, 23]. and scalar multiplication of a fuzzy number is given by
Recently, Salahshour et al. proposed some new results 𝑥
toward existence and uniqueness of the solutions of FFDE {𝑢 ( ) , 𝑘 > 0,
[24]. Then, Mazandarani and Kamyad numerically solved (𝑘 ⊙ 𝑢) (𝑥) = { 𝑘 (2)
̃
the FFDE using the Euler method [25]. Also, Agarwal et al. {0, 𝑘 = 0,
[26] investigated fuzzy fractional integral equation under
compactness type condition. where 0̃ ∈ E.
This paper is structured as follows. In Section 1, we The Hausdorff distance between fuzzy numbers is given
recall some well-known definitions of fuzzy numbers and by 𝑑 : E × E → R+ ∪ {0},
present some needed concepts. In Section 2, Caputo’s H- 󵄨 󵄨
differentiability is introduced, and the relation between 𝑑 (𝑢, V) = sup max {󵄨󵄨󵄨𝑢 (𝑟) − V (𝑟)󵄨󵄨󵄨 , |𝑢 (𝑟) − V (𝛼)|} , (3)
𝑟∈[0,1]
the Riemann-Liouville differentiability and Caputo’s H-
differentiability and some of their properties is considered. where 𝑢 = (𝑢(𝑟), 𝑢(𝑟)), V = (V(𝑟), V(𝑟)) ⊂ R is utilized in
Consequently, the fuzzy Laplace transforms are considered [20]. Then, it is easy to see that 𝑑 is a metric in E and has the
for fuzzy-valued function, and an essential theorem for the following properties (see [31]):
Laplace transform of 𝑓 under Caputo’s H-derivative is given
in Section 3. The solutions of FFDEs are investigated by using (1) 𝑑(𝑢 + 𝑤, V + 𝑤) = 𝑑(𝑢, V), for all 𝑢, V, 𝑤 ∈ E,
the fuzzy Laplace transforms and their inverses in Section 4. (2) 𝑑(𝑘𝑢, 𝑘V) = |𝑘|𝑑(𝑢, V), for all 𝑘 ∈ R, 𝑢, V ∈ E,
In Section 5, some examples are solved to illustrate the
method. Finally, conclusion is drawn in Section 6. (3) 𝑑(𝑢 + V, 𝑤 + 𝑒) ≤ 𝑑(𝑢, 𝑤) + 𝑑(V, 𝑒), for all 𝑢, V, 𝑤, 𝑒 ∈ E,
(4) (𝑑, E) is a complete metric space.
2. Preliminaries Theorem 2 (see [32]). Let 𝑓(𝑥) be a fuzzy-valued function
The basic definition of fuzzy numbers can be seen in [27]. on [𝑎, ∞), and it is represented by (𝑓(𝑥; 𝑟), 𝑓(𝑥; 𝑟)). For any
We denote by R the set of all real numbers and the set of fixed 𝑟 ∈ [0, 1], assume that 𝑓(𝑥; 𝑟) and 𝑓(𝑥; 𝑟) are Riemann-
all fuzzy numbers on R is indicated by E. A fuzzy number is integrable on [𝑎, 𝑏] for every 𝑏 ≥ 𝑎, and assume that there
a mapping 𝑢 : R → [0, 1] with the following properties. 𝑏
are two positive 𝑀(𝑟) and 𝑀(𝑟) such that ∫𝑎 |𝑓(𝑥; 𝑟)|𝑑𝑥 ≤
(a) 𝑢 is upper semicontinuous. 𝑏
𝑀(𝑟) and ∫𝑎 |𝑓(𝑥; 𝑟)|𝑑𝑥 ≤ 𝑀(𝑟) for every 𝑏 ≥ 𝑎. Then
(b) 𝑢 is fuzzy convex; that is, 𝑢(𝜆𝑥 + (1 − 𝜆)𝑦) ≥ 𝑓(𝑥) is improper fuzzy Riemann-integrable on [𝑎, ∞), and
min{𝑢(𝑥), 𝑢(𝑦)} for all 𝑥, 𝑦 ∈ R, 𝜆 ∈ [0, 1]. the improper fuzzy Riemann-integral is a fuzzy number.
Furthermore, one has:
(c) 𝑢 is normal; that is; ∃𝑥0 ∈ R for which 𝑢(𝑥0 ) = 1.
∞ ∞ ∞
(d) supp 𝑢 = {𝑥 ∈ R | 𝑢(𝑥) > 0} is the support of the 𝑢, ∫ 𝑓 (𝑥) 𝑑𝑥 = (∫ 𝑓 (𝑥; 𝑟) 𝑑𝑥, ∫ 𝑓 (𝑥; 𝑟) 𝑑𝑥) . (4)
𝑎 𝑎 𝑎
and its closure cl(supp 𝑢) is compact.
Definition 3. Let 𝑥, 𝑦 ∈ E. If there exists 𝑧 ∈ E such that
An equivalent parametric definition is also given in [28– 𝑥 = 𝑦 + 𝑧, then 𝑧 is called the H-difference of 𝑥 and 𝑦, and it
30] as follows. is denoted by 𝑥 ⊖ 𝑦.
Definition 1. A fuzzy number 𝑢 in parametric form is a pair In this paper, the sign “⊖” always stands for H-difference,
(𝑢, 𝑢) of functions 𝑢(𝑟), 𝑢(𝑟), 0 ≤ 𝑟 ≤ 1, which satisfy the and also note that 𝑥 ⊖ 𝑦 ≠𝑥 + (−1)𝑦.
following requirements:

(1) 𝑢(𝑟) is a bounded nondecreasing left continuous 2.1. Caputo’s H-Differentiability. In this section, the concept
function in (0, 1]; and right continuous at 0; of the fuzzy Caputo derivatives has been reviewed [33–
35]. Also, we denote by 𝐶F [𝑎, 𝑏] a space of all fuzzy-valued
(2) 𝑢(𝑟) is a bounded nonincreasing left continuous functions which are continuous on [𝑎, 𝑏]. Also, we denote the
function in (0, 1]; and right continuous at 0; space of all Lebesgue integrable fuzzy-value functions on the
(3) 𝑢(𝑟) ≤ 𝑢(𝑟), 0 ≤ 𝑟 ≤ 1. bounded interval [𝑎, 𝑏] ⊂ R by 𝐿F [𝑎, 𝑏], and we denote the
space of fuzzy-value functions 𝑓(𝑥) which have continuous
Moreover, we also can present the 𝑟-cut representation of H-derivative up to order 𝑛 − 1 on [𝑎, 𝑏] such that 𝑓(𝑛−1) (𝑥) ∈
fuzzy number as [𝑢]𝑟 = [𝑢(𝑟), 𝑢(𝑟)] for all 0 ≤ 𝑟 ≤ 1. 𝐴𝐶F ([𝑎, 𝑏]) by 𝐴𝐶(𝑛)F ([𝑎, 𝑏]).
Abstract and Applied Analysis 3

Definition 4. Let 𝑓 ∈ 𝐶F [𝑎, 𝑏] ∩ 𝐿F [𝑎, 𝑏]; the fuzzy Riemann- Theorem 7. Let 𝑓(𝑥) ∈ 𝐶(𝑛)F [𝑎, 𝑏] ∩ 𝐿F [𝑎, 𝑏], 𝑥0 ∈ (𝑎, 𝑏),
Liouville integral of fuzzy-valued function 𝑓 is defined as (𝛽 > 0), and (𝑛 = [𝛽] + 1) such that for all 0 ≤ 𝑟 ≤ 1, then
follows: (i) if 𝑓(𝑥) is a 𝑅𝐿 [(𝑖) − 𝛽]-differentiable fuzzy-valued
𝛽 1 𝑥 𝑓
(𝑡) 𝑑𝑡 function, then
(𝐼𝑎+ 𝑓) (𝑥) = ∫ , 𝑥 > 𝑎, 0 < 𝛽 < 1.
Γ (𝛽) 𝑎 (𝑥 − 𝑡)1−𝛽
(5) 𝑅𝐿 𝛽 𝑅𝐿 𝛽 𝛽
( 𝐷𝑎+ 𝑓) (𝑥0 ; 𝑟) = [ 𝐷𝑎+ 𝑓 (𝑥0 , 𝑟) ,𝑅𝐿 𝐷𝑎+ 𝑓 (𝑥0 , 𝑟)] ;
Since 𝑓(𝑥; 𝑟) = [𝑓(𝑥; 𝑟), 𝑓(𝑥; 𝑟)], for all 0 ≤ 𝑟 ≤ 1, then one (10)
can indicate the fuzzy Riemann-Liouville integral of fuzzy-
valued function 𝑓 based on the lower and upper functions as
follows. (ii) if 𝑓(𝑥) is a 𝑅𝐿 [(𝑖𝑖)−𝛽]-differentiable fuzzy-valued func-
tion, then
Theorem 5. Let 𝑓(𝑥) ∈ 𝐶F [𝑎, 𝑏] ∩ 𝐿F [𝑎, 𝑏]; the fuzzy Riem-
ann-Liouville integral of fuzzy-valued function 𝑓 is defined by
𝛽 𝑅𝐿 𝛽 𝛽
𝛽 𝛽 𝛽 (𝑅𝐿 𝐷𝑎+ 𝑓) (𝑥0 ; 𝑟) = [ 𝐷𝑎+ 𝑓 (𝑥0 ; 𝑟) ,𝑅𝐿 𝐷𝑎+ 𝑓 (𝑥0 , 𝑟)] ,
(𝐼𝑎+ 𝑓) (𝑥; 𝑟) = [(𝐼𝑎+ 𝑓) (𝑥; 𝑟) , (𝐼𝑎+ 𝑓) (𝑥; 𝑟)] , 0 ≤ 𝑟 ≤ 1,
(11)
(6)
where
where
𝑥 𝑓 (𝑡; 𝑟) 𝑑𝑡
𝛽 1
(𝐼𝑎+ 𝑓) (𝑥; 𝑟) = ∫ , 0 ≤ 𝑟 ≤ 1,
Γ (𝛽) 𝑎 (𝑥 − 𝑡)1−𝛽
(7) 𝑅𝐿 𝛽 1 𝑑𝑛 𝑥 𝑓 (𝑡; 𝑟) 𝑑𝑡
( 𝐷𝑎+ 𝑓) (𝑥0 ; 𝑟) = [ ∫ ] ,
𝛽 1 𝑥 𝑓
(𝑡; 𝑟) 𝑑𝑡 Γ (𝑛 − 𝛽) 𝑑𝑥𝑛 𝑎 (𝑥 − 𝑡)𝛽−𝑛+1 𝑥=𝑥
(𝐼𝑎+ 𝑓) (𝑥; 𝑟) = ∫ , 0 ≤ 𝑟 ≤ 1. 0

Γ (𝛽) 𝑎 (𝑥 − 𝑡)1−𝛽
𝑅𝐿 𝛽 1 𝑑𝑛 𝑥 𝑓 (𝑡, 𝑟) 𝑑𝑡
Now, we define the fuzzy Riemann-Liouville fractional ( 𝐷𝑎+ 𝑓) (𝑥0 ; 𝑟) = [ ∫ ] .
Γ (𝑛 − 𝛽) 𝑑𝑥𝑛 𝑎 (𝑥 − 𝑡)𝛽−𝑛+1 𝑥=𝑥
derivatives of order 𝑛 − 1 < 𝛽 < 𝑛 for fuzzy-valued function 0
(12)
𝑓 (which is a direct extension of strongly generalized H-
differentiability [20] and lateral type of H-differentiability
[36] in the fractional literature) as follows.
Definition 8. Let 𝑓 ∈ 𝐶(𝑛)F [𝑎, 𝑏] ∩ 𝐿F1 [𝑎, 𝑏], and 𝑓(𝑛) is
𝑛
Definition 6. Let 𝑓 ∈ 𝐶 [𝑎, 𝑏] ∩ (𝑛)F
𝐿F1 [𝑎, 𝑏],
let 𝑥0 ∈ (𝑎, 𝑏), integrable; then the right fuzzy Caputo derivative of 𝑓 for
𝛽
𝑥
and let Φ(𝑥) = (1/Γ(𝑛 − 𝛽)) ∫𝑎 (𝑓(𝑡)𝑑𝑡/(𝑥 − 𝑡)𝛽−𝑛+1 ) (𝑛 = 𝑛 − 1 < 𝛽 < 𝑛 and 𝑥 ∈ [𝑎, 𝑏] is denoted by (𝐶𝐷𝑎+ 𝑓)(𝑥) ∈ E
[𝛽] + 1; 𝑥 > 𝑎). We say that 𝑓 is fuzzy Riemann-Liouville and defined by
fractional differentiable of order 𝛽, at 𝑥0 , if there exists an
𝛽
element (𝐷𝑎+ 𝑓)(𝑥0 ) ∈ E, such that for all ℎ > 0 sufficiently 𝛽 1 𝑥
small ( 𝐶𝐷𝑎+ 𝑓) (𝑥) = ⊙ ∫ (𝑥 − 𝑡)−𝛽+𝑛−1 ⊙ 𝑓(𝑛) (𝑡) 𝑑𝑡.
Γ (𝑛 − 𝛽) 𝑎
(13)
(i)

𝛽 Φ(𝑛−1) (𝑥0 + ℎ) ⊖ Φ(𝑛−1) (𝑥0 )


(𝐷𝑎+ 𝑓) (𝑥0 ) = lim Theorem 9. Let 𝑓(𝑥) ∈ 𝐶(𝑛)F [𝑎, 𝑏] ∩ 𝐿F [𝑎, 𝑏], 𝑥 ∈ (𝑎, 𝑏), and
ℎ→0 ℎ
(8) 𝑛 − 1 < 𝛽 < 𝑛, such that for all 0 ≤ 𝑟 ≤ 1, then
Φ(𝑛−1) (𝑥0 ) ⊖ Φ(𝑛−1) (𝑥0 − ℎ) (i) if 𝑓 is a 𝐶[(𝑖) − 𝛽]-differentiable fuzzy-valued function,
= lim then
ℎ→0 ℎ
or
(ii) 𝐶 𝛽 𝐶 𝛽 𝐶 𝛽
( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) = [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) , ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)] ;
(𝑛−1) (𝑛−1)
𝛽 Φ (𝑥0 ) ⊖ Φ (𝑥0 + ℎ) (14)
(𝐷𝑎+ 𝑓) (𝑥0 ) = lim
ℎ→0 −ℎ
(9)
Φ(𝑛−1) (𝑥0 − ℎ) ⊖ Φ(𝑛−1) (𝑥0 ) (ii) if 𝑓(𝑥) is a 𝐶[(𝑖𝑖) − 𝛽]-differentiable fuzzy-valued
= lim .
ℎ→0 −ℎ function, then
For sake of simplicity, we say that a fuzzy-valued function 𝑓 is
𝑅𝐿
[(i)−𝛽]-differentiable if it is differentiable as in Definition 6 𝐶 𝛽 𝐶 𝛽 𝐶 𝛽
case (i) and is 𝑅𝐿 [(ii) − 𝛽]-differentiable if it is differentiable ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) = [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) , ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)] ,
as in Definition 6 case (ii). (15)
4 Abstract and Applied Analysis

where Similarly, we get the following:


𝑟
[Φ(𝑛−1) (𝑥) ⊖ Φ(𝑛−1) (𝑥 − ℎ)]
𝐶 𝛽
( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) ℎ

𝑛−1 𝑘 Φ(𝑛−1) (𝑥; 𝑟) − Φ(𝑛−1) (𝑥 − ℎ; 𝑟)


𝑅𝐿 𝛽 𝑥 =[ , (19)
= 𝐷𝑎+ [𝑓 (𝑡; 𝑟) − ∑ 𝑓(𝑘) (𝑎; 𝑟)] (𝑥; 𝑟) ℎ
𝑘=0
𝑘!
(𝑛−1) (𝑛−1)
Φ (𝑥; 𝑟) − Φ (𝑥 − ℎ; 𝑟)
1 𝑑𝑛 ].
= ℎ
Γ (𝑛 − 𝛽) 𝑑𝑥𝑛
Passing to the limit we obtain
𝑥 𝑓 (𝑡; 𝑟) − ∑𝑛−1 𝑘
𝑘=0 (𝑡 /𝑘!) 𝑓
(𝑘)
(𝑎; 𝑟)
×∫ 𝑑𝑡, 𝛽 𝛽 𝛽
𝑎 (𝑥 − 𝑡)𝛽−𝑛+1 (𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟) = [ 𝐶𝐷𝑎+ 𝑓 (𝑥; 𝑟) , 𝐶𝐷𝑎+ 𝑓 (𝑥; 𝑟)] , (20)
(16)
𝐶 𝛽 which proves the theorem.
( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)

𝑛−1 𝑘 3. The Fuzzy Laplace Transforms


𝑅𝐿 𝛽 𝑥 (𝑘)
= 𝐷𝑎+ [𝑓 (𝑡; 𝑟) − ∑ 𝑓 (𝑎; 𝑟)] (𝑥; 𝑟)
𝑘=0
𝑘! In this section, we consider the fuzzy Laplace transform
for fuzzy-valued function; then derivative theorem is given
1 𝑑𝑛 which is essential to determine solutions of FFDEs.
= In this way, Allahviranloo and Ahmadi [2] suggested the
Γ (𝑛 − 𝛽) 𝑑𝑥𝑛
concept of Laplace transforms for fuzzy-valued function as
(𝑘) follows.
𝑥 𝑓 (𝑡, 𝑟) − ∑𝑛−1 𝑘
𝑘=0 (𝑡 /𝑘!) 𝑓 (𝑎; 𝑟)
×∫ 𝑑𝑡.
𝑎 (𝑥 − 𝑡)𝛽−𝑛+1 Definition 10 (see [2]). Let 𝑓(𝑥) be continuous fuzzy-value
function. Suppose that 𝑓(𝑥) ⊙ 𝑒−𝑝𝑥 is improper fuzzy

Riemann-integrable on [0, ∞); then ∫0 𝑓(𝑥) ⊙ 𝑒−𝑝𝑥 𝑑𝑥 is
Proof. We prove the case of 𝐶[(i) − 𝛽]-differentiability, (𝛽) > called fuzzy Laplace transforms and is denoted by
0, (𝑛 = [𝛽] + 1), and for the case of 𝐶[(ii) − 𝛽]-differentiability ∞
the proof is completely similar to previous one, and hence it L [𝑓 (𝑥)] = ∫ 𝑓 (𝑥) ⊙ 𝑒−𝑝𝑥 𝑑𝑥 (𝑝 > 0 and integer) .
is omitted. 0
Let us consider that 𝑓 is 𝐶[(i) − 𝛽]-differentiability and (21)
that 𝑥 ∈ (𝑎, 𝑏); then we have the following:
Remark 11. In [8], the authors have investigated under what
conditions the fuzzy-valued functions can possess the fuzzy
𝑟 Laplace transform. So, we suppose that the given fuzzy-
[Φ(𝑛−1) (𝑥 + ℎ) ⊖ Φ(𝑛−1)(𝑥) ] valued functions have mentioned conditions throughout the
paper.
= [Φ(𝑛−1) (𝑥 + ℎ; 𝑟)
From Theorem 2, we have
(𝑛−1) (𝑛−1) (𝑛−1) ∞
−Φ (𝑥; 𝑟) , Φ (𝑥 + ℎ; 𝑟) − Φ (𝑥; 𝑟)] ,
∫ 𝑓 (𝑥) ⊙ 𝑒−𝑝𝑥 𝑑𝑥
(17) 0
(22)
∞ ∞
= (∫ 𝑓 (𝑥; 𝑟) 𝑒−𝑝𝑥 𝑑𝑥, ∫ 𝑓 (𝑥; 𝑟) 𝑒−𝑝𝑥 𝑑𝑥) .
0 0
and multiplying by 1/ℎ, we get the following:
Also by using the definition of classical Laplace transform one
𝑟
has
[Φ(𝑛−1) (𝑥 + ℎ) ⊖ Φ(𝑛−1) (𝑥)] ∞

ℎ ℓ [𝑓 (𝑥; 𝑟)] = ∫ 𝑓 (𝑥; 𝑟) 𝑒−𝑝𝑥 𝑑𝑥,


0
(𝑛−1) (𝑛−1)
(23)

Φ (𝑥 + ℎ; 𝑟) − Φ (𝑥; 𝑟) −𝑝𝑥
=[ , (18) ℓ [𝑓 (𝑥, 𝑟)] = ∫ 𝑓 (𝑥; 𝑟) 𝑒 𝑑𝑥;
ℎ 0

(𝑛−1) (𝑛−1) then we get


Φ (𝑥 + ℎ; 𝑟) − Φ (𝑥; 𝑟)
].
ℎ L [𝑓 (𝑥)] = (ℓ [𝑓 (𝑥; 𝑟)] , ℓ [𝑓 (𝑥; 𝑟)]) . (24)
Abstract and Applied Analysis 5

Theorem 12 (see [2]). Let 𝑓(𝑥), 𝑔(𝑥) be continuous-fuzzy- Since 𝐷(𝑘−1) 𝑓 is 𝐶[(i) − 𝛽], we get the following:
valued functions; suppose that 𝑐1 , 𝑐2 are constant; then
L [(𝑐1 ⊙ 𝑓 (𝑥)) ⊕ (𝑐2 ⊙ 𝑔 (𝑥))] 𝐶 𝐶 𝐶
𝛽 𝛽 𝛽
(25) ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) = [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) , ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)]
= (𝑐1 ⊙ L [𝑓 (𝑥)]) ⊕ (𝑐2 ⊙ L [𝑔 (𝑥)]) .
𝐶 𝛽 𝐶 𝛽
Lemma 13 (see [2]). Let 𝑓(𝑥) be continuous fuzzy-value = [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) , ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)] .
function on [0, ∞) and let 𝜆 ∈ R; then
(32)
L [𝜆 ⊙ 𝑓 (𝑥)] = 𝜆 ⊙ L [𝑓 (𝑥)] . (26)
Lemma 14 (see [2]). Let 𝑓(𝑥) be continuous fuzzy-value Hence, we get
function and 𝑔(𝑥) ≥ 0. Suppose that (𝑓(𝑥) ⊙ 𝑔(𝑥)) ⊙ 𝑒−𝑝𝑥 is
improper fuzzy Riemann-integrable on [0, ∞); then
∞ 𝐶 𝛽 𝐶 𝛽
∫ (𝑓 (𝑥) ⊙ 𝑔 (𝑥)) ⊙ 𝑒−𝑝𝑥 𝑑𝑥 ℓ [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)] = ℓ [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)]
0
∞ ∞
= 𝑝𝛽 ℓ [𝑓 (𝑥; 𝑟)]
= (∫ 𝑔 (𝑥) 𝑓 (𝑥, 𝑟) 𝑒−𝑝𝑥 𝑑𝑥, ∫ 𝑔 (𝑥) 𝑓 (𝑥, 𝑟) 𝑒−𝑝𝑥 𝑑𝑥) .
0 0
𝑛−1
(27) (𝑘)
− ( ∑ 𝑝𝛽−𝑘−1 𝑓 ) (0; 𝑟) ,
Theorem 15 (first translation theorem (see [2])). Let 𝑓 be 𝑘=0
continuous fuzzy value function and let L[𝑓(𝑥)] = 𝐹(𝑝). Then 𝐶 𝛽 𝐶 𝛽
𝑎𝑥 ℓ [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)] = ℓ [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)]
L [𝑒 ⊙ 𝑓 (𝑥)] = 𝐹 (𝑝 − 𝑎) , (28)
where 𝑒𝑎𝑥 is real value function. = 𝑝𝛽 ℓ [𝑓 (𝑥; 𝑟)]

In order to solve FFDEs, it is necessary to know the fuzzy 𝑛−1


𝐶 𝛽 − ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0; 𝑟) .
Laplace transform of Caputo’s H-derivative of 𝑓, ( 𝐷𝑎+ 𝑓)(𝑥). 𝑘=0
𝛽
The virtue of L[( 𝐶𝐷𝑎+ 𝑓)(𝑥)] is that it can be written in terms (33)
of L[𝑓(𝑥)].

Theorem 16 (derivative theorem). Suppose that 𝑓(𝑥) ∈ Then, we obtained the following:
𝐶(𝑛)F [0, ∞) ∩ 𝐿F [0, ∞); then
𝛽 𝑛−1
L [( 𝐶𝐷𝑎+ 𝑓) (𝑥)]
𝑝𝛽 L [𝑓 (𝑥)] ⊖ ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0)
(29) 𝑘=0
𝑛−1
𝛽 𝛽−𝑘−1 (𝑘)
= 𝑝 L [𝑓 (𝑡)] ⊖ ( ∑ 𝑝 𝑓 ) (0) 𝛽 𝛽
𝑘=0
= (ℓ [(𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟)] , ℓ [(𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟)]) .
(34)
if 𝐷(𝑘−1) 𝑓 is 𝐶[(𝑖) − 𝛽]-differentiable, and also
𝛽
L [( 𝐶𝐷𝑎+ 𝑓) (𝑥)] Then, we have
𝑛−1 (30)
= − ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0) ⊖ (−𝑝𝛽 L [𝑓 (𝑡)]) 𝑛−1
𝑘=0 𝑝𝛽 L [𝑓 (𝑥)] ⊖ ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0)
𝑘=0
if 𝐷(𝑘−1) 𝑓 is 𝐶[(𝑖𝑖) − 𝛽]-differentiable. (35)
𝐶 𝛽 𝐶 𝛽
Proof. For arbitrary fixed 𝑟 ∈ [0, 1] we have = L [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) , ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)] .

𝑝𝛽 L [𝑓 (𝑡)] ⊖ 𝑝𝛽−1 𝑓 (0)


Using (32) leads to obtain
𝑛−1
𝛽 𝛽−𝑘−1 (𝑘)
= [𝑝 ℓ [𝑓 (𝑥; 𝑟)] − ( ∑ 𝑝 𝑓 ) (0; 𝑟) ,
𝑘=0 (31)
𝑛−1
𝛽
𝑛−1 𝑝𝛽 L [𝑓 (𝑥)] ⊖ ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0) = L [(𝐶𝐷𝑎+ 𝑓) (𝑥)] .
𝛽 𝛽−𝑘−1 (𝑘)
𝑝 ℓ [𝑓 (𝑥; 𝑟)] − ( ∑ 𝑝 𝑓 ) (0; 𝑟)] . 𝑘=0
𝑘=0 (36)
6 Abstract and Applied Analysis

Now, we assume that 𝐷(𝑘−1) 𝑓 is the 𝐶[(ii) − 𝛽]; then for differential equations under Caputo’s H-differentiability. First
arbitrary fixed 𝑟 ∈ [0, 1] we have we consider the nonhomogeneous fuzzy fractional differen-
tial equation of order 0 < 𝛽 < 1 together with nonhomoge-
𝑛−1
neous initial condition
− ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0) ⊖ (−𝑝𝛽 L [𝑓 (𝑡)])
𝑘=0 2𝛽 2𝛽
(𝐶𝐷0+ Φ) (𝑥) ⊕ (𝐶𝐷0+ ) (𝑥)
= (−𝑓2 (𝑡; 𝑟) , −𝑓1 (𝑡; 𝑟)) ,
= 𝜑 [𝑥; Φ (𝑥)] , 0 < 𝛽 < 1, (42)
𝑛−1
𝛽−𝑘−1 (𝑘)
−𝑓2 (𝑡; 𝑟) = − ( ∑ 𝑝 𝑓 ) (0; 𝑟) + 𝑝𝛽 ℓ [𝑓 (𝑥; 𝑟)] , Φ (0) = 𝑓 (𝑥) ∈ E,
𝑘=0

𝑛−1 where 𝑡 > 0. Fractional Green’s functions for case so-called


−𝑓1 (𝑡; 𝑟) = − ( ∑ 𝑝 𝛽−𝑘−1 (𝑘) 𝛽
𝑓 ) (0; 𝑟) + 𝑝 ℓ [𝑓 (𝑥; 𝑟)] . the classical Green’s functions, that is, considering 𝜑(𝑥) =
𝑘=0 𝛿(𝑥) and Φ(0) = 0̃, are discussed.
(37)
4.1. Determining Algebraic Solutions. The algebraic solutions
Since 𝐷(𝑘−1) 𝑓 is 𝐶[(ii) − 𝛽], we get of fuzzy fractional differential equations have been inves-
tigated under Caputo’s H-differentiability. We provided the
𝛽 𝛽 𝛽 fuzzy Laplace transform to obtain the solutions of FFDE (42).
( 𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟) = [( 𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟) , ( 𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟)] By taking Laplace transform on the both sides of (42), we get
𝛽 𝛽 𝐶 2𝛽 𝐶 𝛽
= [( 𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟) , ( 𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟)] , L [( 𝐷0+ Φ) (𝑥) ⊕ ( 𝐷0+ Φ) (𝑥)] = L [𝜑 (𝑥, Φ (𝑥))] .
(43)
(38)
After that, based on the types of Caputo’s H-differentiability
and, hence, we get we have the following cases.
𝑛−1
− ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0) ⊖ (−𝑝𝛽 L [𝑓 (𝑡)]) Case I. Let us consider that 𝐷Φ(𝑥) is a 𝐶[(i)−𝛽]-differentiable
𝑘=0
function; then (43) is extended based on the its lower and
upper functions as follows:
𝑛−1
(𝑘)
= 𝑝𝛽 ℓ [𝑓 (𝑥; 𝑟)] − ( ∑ 𝑝𝛽−𝑘−1 𝑓 ) (0; 𝑟) , 𝑝𝛽 ℓ [𝑓 (𝑥; 𝑟)] ℓ [𝜑 (𝑥, Φ (𝑥) ; 𝑟)] = 𝑝2𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝2𝛽−1 Φ (0, 𝑟)
𝑘=0

𝑛−1 + 𝑝𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝𝛽−1 Φ (0, 𝑟) ,


𝛽−𝑘−1 (𝑘)
− (∑𝑝 𝑓 ) (0; 𝑟) .
𝑘=0 ℓ [𝜑 (𝑥, Φ (𝑥) ; 𝑟)] = 𝑝2𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝2𝛽−1 Φ (0, 𝑟)
(39)
+ 𝑝𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝𝛽−1 Φ (0, 𝑟) ,
So, we get
(44)
𝑛−1
− ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0) ⊖ (−𝑝𝛽 L [𝑓 (𝑡)]) where
𝑘=0
(40) 𝜑 (𝑥, Φ(𝑥) ; 𝑟) = min {𝜑 (𝑥, 𝑢) |
𝛽 𝛽
= [ℓ [( 𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟) , ℓ ( 𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟)]] .
𝑢 ∈ [Φ (𝑥; 𝑟) , Φ (𝑥; 𝑟)]} ,
Thus, we obtain the following:
0 ≤ 𝑟 ≤ 1,
𝑛−1 (45)
− ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0) ⊖ (−𝑝𝛽 L [𝑓 (𝑡)]) 𝜑 (𝑥, Φ (𝑥) ; 𝑟) = max {𝜑 (𝑥, 𝑢) |
𝑘=0
(41)
𝛽
𝑢 ∈ [Φ (𝑥; 𝑟) , Φ (𝑥; 𝑟)]} ,
= L [(𝐶𝐷𝑎+ 𝑓) (𝑥)] ,
0 ≤ 𝑟 ≤ 1.
which completes the proof.
In order to solve the linear system (44), for simplify we
assume that
4. Fractional Green’s Functions
ℓ [Φ (𝑥; 𝑟)] = 𝐻1 (𝑝; 𝑟) , 0 ≤ 𝑟 ≤ 1,
In the following, we present the so-called fundamental (46)
solutions of fractional Green’s functions for fuzzy fractional ℓ [Φ (𝑥; 𝑟)] = 𝐾1 (𝑝; 𝑟) , 0 ≤ 𝑟 ≤ 1,
Abstract and Applied Analysis 7

where 𝐻1 (𝑝; 𝑟) and 𝐾1 (𝑝; 𝑟) are solutions of system (44). where 𝐻2 (𝑝; 𝑟) and 𝐾2 (𝑝; 𝑟) are solutions of system (49). Then
Then, by using the inverse Laplace transform, Φ(𝑥; 𝑟) and by making use of the inverse Laplace transform, Φ(𝑥; 𝑟) and
Φ(𝑥; 𝑟) are computed as follows: Φ(𝑥; 𝑟) are computed as follows:

Φ (𝑥, 𝑟) = ℓ−1 [𝐻1 (𝑝; 𝑟)] Φ (𝑥; 𝑟) = ℓ−1 [𝐻2 (𝑝; 𝑟)]
𝑡 = Φ (0; 𝑟) 𝐺2 (𝑡)
= Φ (0; 𝑟) 𝐺1 (𝑡) + ∫ 𝐺1 (𝑡; 𝜏) 𝜑 (𝑡 − 𝜏; 𝑟) 𝑑𝜏,
0 𝑡
+ ∫ 𝐺2 (𝑡; 𝜏) 𝜑 (𝑡 − 𝜏; 𝑟) 𝑑𝜏,
0 ≤ 𝑟 ≤ 1, 0

−1
Φ (𝑥; 𝑟) = ℓ [𝐾1 (𝑝; 𝑟)] 0 ≤ 𝑟 ≤ 1,
(52)
𝑡 Φ (𝑥; 𝑟) = ℓ−1 [𝐾2 (𝑝; 𝑟)]
= Φ (0; 𝑟) 𝐺1 (𝑡) + ∫ 𝐺1 (𝑡; 𝜏) 𝜑 (𝑡 − 𝜏; 𝑟) 𝑑𝜏,
0
= Φ (0; 𝑟) 𝐺2 (𝑡)
0 ≤ 𝑟 ≤ 1.
𝑡
(47) + ∫ 𝐺2 (𝑡; 𝜏) 𝜑 (𝑡 − 𝜏; 𝑟) 𝑑𝜏,
0

To explicit the corresponding Green’s function, which is 0 ≤ 𝑟 ≤ 1.


solution of the above relation satisfying the condition Φ(0) =
0̃ and taking 𝜑(𝑥) = 𝛿(𝑥), we use the linearity of the inverse To show the corresponding Green’s function, which is solu-
Laplace transform to obtain
tion of (42) satisfying the condition Φ(0) = 0̃ and taking
𝜑(𝑥) = 𝛿(𝑥), we use the linearity of the inverse Laplace
𝐺1 (𝑡 | 0) = ℓ−1 [Φ (𝑥; 𝑟)] . (48) transform to obtain

Case II. Let us consider that 𝐷Φ(𝑥) is 𝐶[(ii) − 𝛽]- 𝐺2 (𝑡 | 0) = ℓ−1 [Φ (𝑥; 𝑟)] . (53)
differentiable; then (43) can be written as follows:
5. Examples
ℓ [𝜑 (𝑥, Φ (𝑥) ; 𝑟)] = 𝑝2𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝2𝛽−1 Φ (0, 𝑟)
In this section, we consider several examples, and we
𝛽 𝛽−1 find explicit expressions for fractional Green’s functions,
+ 𝑝 ℓ [Φ (𝑥, 𝑟)] − 𝑝 Φ (0, 𝑟) ,
including some versions of the fuzzy harmonic oscillator
ℓ [𝜑 (𝑥, Φ (𝑥) ; 𝑟)] = 𝑝2𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝2𝛽−1 Φ (0, 𝑟) equations and fuzzy relaxation equation under Caputo’s H-
differentiability.
+ 𝑝𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝𝛽−1 Φ (0, 𝑟) ,
Example 1 (Green’s function for fuzzy damped harmonic
(49) oscillator). Let us examine the following FFDE:

where 𝐶 2𝛽 𝐶 𝛽
𝑎 ⊙ ( 𝐷0+ Φ) (𝑥) + 𝑏 ⊙ ( 𝐷0+ Φ) (𝑥)
𝜑 (𝑥, Φ (𝑥) ; 𝑟) = min {𝜑 (𝑥, 𝑢) | 𝑢 ∈ [Φ (𝑥; 𝑟) , Φ (𝑥; 𝑟)]} , (54)
= 𝜆𝑤2 ⊙ Φ (𝑥) + 𝜑 (𝑥) , 0 < 𝛽 < 1,
0 ≤ 𝑟 ≤ 1, Φ (0) ∈ E,
𝜑 (𝑥, Φ (𝑥) ; 𝑟) = max {𝜑 (𝑥, 𝑢) | 𝑢 ∈ [Φ (𝑥; 𝑟) , Φ (𝑥; 𝑟)]} ,
where 𝑎, 𝑏 are real positive constants and 𝑤 is the frequency of
0 ≤ 𝑟 ≤ 1. the harmonic oscillator. We will solve this example according
(50) to the two following cases for 𝜆 ∈ R.

Case I. Suppose that 𝜆 ∈ R+ = (0, ∞); then taking Laplace


Consequently, to solve the linear system (49), we set for transform on the both sides of above equation, we have
simplify
𝐶 2𝛽 𝐶 𝛽
ℓ [Φ (𝑥; 𝑟)] = 𝐻2 (𝑝; 𝑟) , L [𝑎 ⊙ ( 𝐷0+ Φ) (𝑥) + 𝑏 ⊙ ( 𝐷0+ Φ) (𝑥)]
(51) (55)
2
ℓ [Φ (𝑥; 𝑟)] = 𝐾2 (𝑝; 𝑟) , = L [𝜆𝑤 ⊙ Φ (𝑥) + 𝜑 (𝑥)] .
8 Abstract and Applied Analysis

Using 𝐶[(i) − 𝛽]-differentiability, we get



−𝑏 𝑗 𝛽𝑗 (𝑗+1) 𝜆𝑤2 2𝛽
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ∑( ) 𝑥 𝐸2𝛽, 𝛽 +1 ( 𝑥 )
𝑗=0 𝑎 𝑗 𝑎
𝜆𝑤2 ℓ [Φ (𝑥; 𝑟)] + ℓ [𝜑 (𝑥; 𝑟)]
𝑏 (𝑗+1) 𝜆𝑤2 2𝛽
2𝛽 2𝛽−1 + 𝑥𝛽 𝐸2𝛽, 𝛽 +𝛽+1 ( 𝑥 )
= 𝑎𝑝 ℓ [Φ (𝑥, 𝑟)] − 𝑎𝑝 Φ (0, 𝑟) 𝑎 𝑗 𝑎

+ 𝑏𝑝𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑏𝑝𝛽−1 Φ (0, 𝑟) , 1 ∞ −𝑏 𝑗 𝑥 (𝑗+1)


+ ∑ ( ) ∫ 𝜑 (𝑡 − 𝑦; 𝑟) 𝑦𝛽𝑗+2𝛽−1 𝐸2𝛽,𝛽 +2𝛽
(56) 𝑎 𝑗=0 𝑎 0 𝑗

𝜆𝑤2 ℓ [Φ (𝑥; 𝑟)] + ℓ [𝜑 (𝑥; 𝑟)]


𝜆𝑤2 2𝛽
= 𝑎𝑝2𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑎𝑝2𝛽−1 Φ (0, 𝑟) ×( 𝑦 ) 𝑑𝑦, 0 ≤ 𝑟 ≤ 1.
𝑎
+ 𝑏𝑝𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑏𝑝𝛽−1 Φ (0, 𝑟) . (59)
To explicit the corresponding Green’s function under 𝐶[(i) −
Then, after some calculations, we obtain the following: 𝛽]-differentiability, which is solution of (54) satisfying the
condition Φ(0) = 0̃ and taking 𝜑(𝑥) = 𝛿(𝑥), we use the
(𝑎𝑝2𝛽 + 𝑏𝑝𝛽 − 𝜆𝑤2 ) ℓ [Φ (𝑥; 𝑟)] linearity of the inverse Laplace transform to obtain
𝐺1 (𝑥 | 0) ≡ ℓ−1 [ℓ [Φ (𝑥; 𝑟)]]
= (𝑎𝑝2𝛽−1 + 𝑏𝑝𝛽−1 ) Φ (0; 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] ,
(57)
1 ∞ −𝑏 𝑗 𝛽𝑗+2𝛽−1 (𝑗+1) 𝜆𝑤2 2𝛽
(𝑎𝑝2𝛽 + 𝑏𝑝𝛽 − 𝜆𝑤2 ) ℓ [Φ (𝑥; 𝑟)] = ∑( ) 𝑡 𝐸2𝛽, 𝛽 +2𝛽 ( 𝑡 ).
𝑎 𝑗=0 𝑎 𝑗 𝑎
= (𝑎𝑝2𝛽−1 + 𝑏𝑝𝛽−1 ) Φ (0; 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] . (60)
On the other hand, taking 𝜑(𝑥) = 0, ̃ we have a homogeneous
Consequently, taking the inverse of Laplace on both sides of fuzzy fractional differential equation whose solution under
(57) we have 𝐶
[(i) − 𝛽]-H-differentiability is given by

Φ (𝑥; 𝑟) = Φ (0; 𝑟) ℓ−1



−𝑏 𝑗 𝛽𝑗 (𝑗+1) 𝜆𝑤2 2𝛽
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ∑( ) 𝑥 𝐸2𝛽, 𝛽 +1 ( 𝑥 )
𝑗=0 𝑎 𝑗 𝑎
(𝑎𝑝2𝛽−1 + 𝑏𝑝𝛽−1 ) Φ (0; 𝑟) + ℓ [𝜑 (𝑥; 𝑟)]
×[ ],
𝑏 𝜆𝑤2 2𝛽
𝑎𝑝2𝛽 + 𝑏𝑝𝛽 − 𝜆𝑤2 (𝑗+1)
+ 𝑥𝛽 𝐸2𝛽,𝛽 +𝛽+1 ( 𝑥 ), 0 ≤ 𝑟 ≤ 1,
[ ] 𝑎 𝑗 𝑎
0 ≤ 𝑟 ≤ 1, ∞
−𝑏 𝑗 𝛽𝑗 (𝑗+1) 𝜆𝑤2 2𝛽
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ∑( ) 𝑥 𝐸2𝛽, 𝛽 +1 ( 𝑥 )
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ℓ−1 𝑗=0 𝑎 𝑗 𝑎
2𝛽−1 𝛽−1
(𝑎𝑝 + 𝑏𝑝 ) Φ (0; 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] 𝑏 (𝑗+1) 𝜆𝑤2 2𝛽
×[ ], + 𝑥𝛽 𝐸2𝛽, 𝛽 +𝛽+1 ( 𝑥 ), 0 ≤ 𝑟 ≤ 1.
𝑎𝑝2𝛽 + 𝑏𝑝𝛽 − 𝜆𝑤2 𝑎 𝑗 𝑎
(61)
0 ≤ 𝑟 ≤ 1.
(58) Case II. If 𝜆 ∈ R− = (−∞, 0), then by using 𝐶[(ii) −
𝛽]-differentiability and Theorem 16, and their inverses, frac-
tional Green’s function will be obtained similar to (60).
Finally, we determine the solution of FFDE as follows:
Example 2 (Green’s function for fuzzy the driven harmonic

−𝑏 𝑗 (𝑗+1) 𝜆𝑤2 2𝛽 oscillator). Let us investigate the following FFDEs; namely,
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ∑( ) 𝑥𝛽𝑗 𝐸2𝛽, 𝛽 +1 ( 𝑥 )
𝑗=0 𝑎 𝑎
𝑗
𝐶 2𝛽
( 𝐷0+ Φ) (𝑥) = 𝜆𝑤2𝛽 ⊙ Φ (𝑥) + 𝜑 (𝑥) , 0<𝛽<1
𝑏 (𝑗+1) 𝜆𝑤2 2𝛽
+ 𝑥𝛽 𝐸2𝛽, 𝛽 +𝛽+1 ( 𝑥 ) Φ (0) ∈ E,
𝑎 𝑗 𝑎
(62)
1 ∞ −𝑏 𝑗 𝑥 (𝑗+1)
+ ∑ ( ) ∫ 𝜑 (𝑡 − 𝑦; 𝑟) 𝑦𝛽𝑗+2𝛽−1 𝐸2𝛽, 𝛽 +2𝛽 with 𝑏 = 0 and 1/2 < 𝛽 < 1; the fuzzy fractional differential
𝑎 𝑗=0 𝑎 0 𝑗
equation associated with the driven harmonic oscillator is
obtained. We obtain the same result taking 𝑏 = 0, 𝑎 = 1 and
𝜆𝑤2 2𝛽 solve this example according to the two following cases for
×( 𝑦 ) 𝑑𝑦, 0 ≤ 𝑟 ≤ 1,
𝑎 𝜆 ∈ R.
Abstract and Applied Analysis 9

Case I. Suppose that 𝜆 ∈ R+ = (0, +∞); then taking Laplace In the case 𝛽 = 1, we report
transform on the both sides of above equation, we have the
following: 𝐺1 (𝑥 | 0) = 𝑥𝐸2, 2 (𝜆𝑤2 𝑥2 ) (69)
𝐶 2𝛽
L [( 𝐷𝑎+ Φ) (𝑥)] = L (𝜆𝑤2𝛽 ⊙ Φ (𝑥) + 𝜑 (𝑥)) . (63) which is the classic Green’s function associated with fuzzy the
driven harmonic oscillator.
Using 𝐶[(i) − 𝛽]-differentiability, we obtain
Taking 𝜑(𝑥) = 0̃, we have a homogeneous fuzzy fractional
2𝛽
𝜆𝑤 ℓ [Φ (𝑥; 𝑟)] + ℓ [𝜑 (𝑥; 𝑟)] differential equation whose solution under 𝐶[(𝑖) − 𝛽]-H-
differentiability is given by
= 𝑝2𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝2𝛽−1 Φ (0, 𝑟) ,
(64) Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸2𝛽 (𝜆𝑤2𝛽 𝑥2𝛽 ) , 0 ≤ 𝑟 ≤ 1,
2𝛽
𝜆𝑤 ℓ [Φ (𝑥; 𝑟)] + ℓ [𝜑 (𝑥; 𝑟)] (70)
Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸2𝛽 (𝜆𝑤2𝛽 𝑥2𝛽 ) , 0 ≤ 𝑟 ≤ 1.
= 𝑝2𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝2𝛽−1 Φ (0, 𝑟) .
For 𝛽 = 1, we get
Then, after some manipulations the following result is
reported: Φ (𝑥; 𝑟) = Φ (0; 𝑟) cos 𝑤𝑡, 0 ≤ 𝑟 ≤ 1,
(71)
(𝑝2𝛽 − 𝜆𝑤2𝛽 ) ℓ [Φ (𝑥; 𝑟)] = 𝑝2𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] , Φ (𝑥; 𝑟) = Φ (0; 𝑟) cos 𝑤𝑡, 0 ≤ 𝑟 ≤ 1.

(𝑝2𝛽 − 𝜆𝑤2𝛽 ) ℓ [Φ (𝑥; 𝑟)] = 𝑝2𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] . Case II. Let us suppose that 𝜆 ∈ R− = (−∞, 0); then using
𝐶
(65) [(ii)−𝛽]-differentiability and Theorem 16 and their inverses,
fractional Green’s function will be obtained similar to (68).
By taking inverse of Laplace on the both sides of (75) we have
Example 3 (Green’s functions for fuzzy fractional relaxation
𝑝2𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] equation). The next step is to consider the following FFDE:
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ℓ [−1 ],
𝑝2𝛽 − 𝜆𝑤2𝛽 𝛽
[ ] 𝑏 ⊙ ( 𝐶𝐷0+ Φ) (𝑥) = 𝜆𝑤2 ⊙ Φ (𝑥) + 𝜑 (𝑥) ,
0 ≤ 𝑟 ≤ 1,
0 < 𝛽 < 1, (72)
2𝛽−1
𝑝 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)]
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ℓ−1 [ ], Φ (0) ∈ E,
𝑝2𝛽 − 𝜆𝑤2𝛽
0 ≤ 𝑟 ≤ 1. where 𝑎 = 0 in (54) and 0 < 𝛽 < 1. We solve this example
(66) according to the two following cases corresponding to 𝜆 ∈ R.

Finally, we determine the solution of FFDE as follows: Case I. Suppose that 𝜆 ∈ R+ = (0, +∞); then taking Laplace
transform on the both sides of above equation, we have
Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸2𝛽 (𝜆𝑤2𝛽 𝑥2𝛽 )
𝐶 𝛽
𝑥 L [𝑏 ⊙ ( 𝐷𝑎+ Φ) (𝑥)] = L [𝜆𝑤2 ⊙ Φ (𝑥) + 𝜑 (𝑥)] . (73)
2𝛽−1 2𝛽 2𝛽
+ ∫ 𝜑 (𝑡 − 𝑦; 𝑟) 𝑦 𝐸2𝛽, 2𝛽 (𝜆𝑤 𝑦 ) 𝑑𝑦,
0
Using 𝐶[(i) − 𝛽]-differentiability, we get
0 ≤ 𝑟 ≤ 1,
𝜆𝑤𝛽 ℓ [Φ (𝑥; 𝑟)] + ℓ [𝜑 (𝑥; 𝑟)]
Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸2𝛽 (𝜆𝑤2𝛽 𝑥2𝛽 )
𝑥 = 𝑏𝑝𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑏𝑝𝛽−1 Φ (0, 𝑟) ,
+ ∫ 𝜑 (𝑡 − 𝑦; 𝑟) 𝑦2𝛽−1 𝐸2𝛽, 2𝛽 (𝜆𝑤2𝛽 𝑦2𝛽 ) 𝑑𝑦, (74)
0 𝜆𝑤𝛽 ℓ [Φ (𝑥; 𝑟)] + ℓ [𝜑 (𝑥; 𝑟)]
0 ≤ 𝑟 ≤ 1.
(67) = 𝑏𝑝𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑏𝑝𝛽−1 Φ (0, 𝑟) .

To show the corresponding Green’s function under 𝐶[(i) − Then, after some manipulations we get the following:
𝛽]-differentiability, which is solution of (62) satisfying the
condition Φ(0) = 0̃ and taking 𝜑(𝑥) = 𝛿(𝑥), we use the (𝑏𝑝𝛽 − 𝜆𝑤2 ) ℓ [Φ (𝑥; 𝑟)] = 𝑏𝑝𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] ,
linearity of the inverse Laplace transform to obtain
(𝑏𝑝𝛽 − 𝜆𝑤2 ) ℓ [Φ (𝑥; 𝑟)] = 𝑏𝑝𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] .
2𝛽−1 2𝛽 2𝛽
𝐺1 (𝑥 | 0) ≡ 𝑥 𝐸2𝛽, 2𝛽 (𝜆𝑤 𝑥 ) . (68) (75)
10 Abstract and Applied Analysis

By taking inverse of Laplace on the both sides of (75) we have Case II. Suppose that 𝜆 ∈ R− = (−∞, 0); then using
𝐶
[(ii)−𝛽]-differentiability and Theorem 16, and their inverses,
𝑏𝑝𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] fractional Green’s function will be obtained similar to (78).
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ℓ−1 [ ],
𝑏𝑝𝛽 − 𝜆𝑤2
[ ] Now, we provide some examples such that the obtained
0 ≤ 𝑟 ≤ 1, solutions coincide with the previously reported solution in
integer case [37].
𝑏𝑝𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)]
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ℓ−1 [ ], Example 4. For a special case, let us consider the following
𝑏𝑝𝛽 − 𝜆𝑤2 FFDE:
𝐶
0 ≤ 𝑟 ≤ 1. 𝛽
( 𝐷0+ Φ) (𝑥) = 𝜆 ⊙ Φ (𝑥) , 0 < 𝛽, 𝑥 < 1,
(76) (82)
Thus, we determine the solution of FFDE as follows: Φ (0) ∈ E,
2
𝜆𝑤 𝛽 where 𝑦 represents the number of radionuclides present in a
Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸𝛽 ( 𝑥 ) given radioactive and 𝜆 is a decay constant.
𝑏

1 𝑥 𝜆𝑤2 𝛽 Case I. Suppose that 𝜆 ∈ R+ = (0, +∞); then using 𝐶[(i) −


+ ∫ 𝜑 (𝑡 − 𝑦; 𝑟) 𝑦𝛽−1 𝐸𝛽, 𝛽 ( 𝑦 ) 𝑑𝑦, 𝛽]-differentiability and Theorem 16, the solution is derived as
𝑏 0 𝑏
follows:
0 ≤ 𝑟 ≤ 1,
Φ (𝑥; 𝑟) = Φ (0, 𝑟) 𝐸𝛽, 𝛽 [𝜆𝑥𝛽 ] , 0 ≤ 𝑟 ≤ 1,
𝜆𝑤 𝛽 2 (83)
Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸𝛽 ( 𝑥 ) Φ (𝑥; 𝑟) = Φ (0, 𝑟) 𝐸𝛽, 𝛽 [𝜆𝑥𝛽 ] , 0 ≤ 𝑟 ≤ 1.
𝑏
Case II. Suppose that 𝜆 ∈ R− = (−∞, 0); then using
1 𝑥 𝜆𝑤2 𝛽 𝐶
+ ∫ 𝜑 (𝑡 − 𝑦; 𝑟) 𝑦𝛽−1 𝐸𝛽,𝛽 ( 𝑦 ) 𝑑𝑦, [(ii) − 𝛽]-differentiability and Theorem 16, the solution will
𝑏 0 𝑏 be obtained similar to (83). For more details see Figures 1, 2,
3, and 4.
0 ≤ 𝑟 ≤ 1.
(77) Example 5. Let us discuss the following FFDE:
𝐶
Here, we obtain explicitly Green’s functions under [(i) − 𝛽]- 𝛽
( 𝐶𝐷0+ Φ) (𝑥) = (−1) ⊙ Φ (𝑥) + 𝑥 + 1, 0 < 𝛽, 𝑥 < 1,
differentiability, introducing 𝜑(𝑥) = 𝛿(𝑥) and Φ(0) = 0̃; that
is,
Φ (0) ∈ E.
1 𝜆𝑤𝛽 𝛽 (84)
𝐺1 (𝑥 | 0) ≡ 𝑥𝛽−1 𝐸𝛽, 𝛽 ( 𝑥 ). (78)
𝑏 𝑏
Then, using 𝐶[(ii) − 𝛽]-differentiability and Theorem 16, the
When 𝛽 = 1, we get the classic Green’s function under solution becomes as follows:
strongly generalized differentiability
𝑦 (𝑥; 𝑟) = Φ (0, 𝑟) 𝐸𝛽,𝛽 [𝜆𝑥𝛽 ]
1 𝜆𝑤2
𝐺1 (𝑥 | 0) = exp ( 𝑥) . (79) 𝑥
𝑏 𝑏 + ∫ (𝑥 − 𝑡)𝛽−1 𝐸𝛽,𝛽 [𝜆(𝑥 − 𝑡)𝛽 ] (𝑡 + 1) 𝑑𝑡,
0
On the other hand, taking 𝜑(𝑥) = 0̃, we have a homogeneous
fuzzy fractional differential equation whose solution under 0 ≤ 𝑟 ≤ 1,
𝐶
[(i) − 𝛽]-H-differentiability is given by
𝑦 (𝑥; 𝑟) = Φ (0, 𝑟) 𝐸𝛽,𝛽 [𝜆𝑥𝛽 ]
2
𝜆𝑤 𝛽
Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸𝛽 ( 𝑥 ), 0 ≤ 𝑟 ≤ 1, 𝑥
𝑏 + ∫ (𝑥 − 𝑡)𝛽−1 𝐸𝛽,𝛽 [𝜆(𝑥 − 𝑡)𝛽 ] (𝑡 + 1) 𝑑𝑡,
(80) 0
𝜆𝑤2 𝛽
Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸𝛽 ( 𝑥 ), 0 ≤ 𝑟 ≤ 1. 0 ≤ 𝑟 ≤ 1.
𝑏
(85)
Finally, in the case 𝛽 = 1 we get
For more details see Figures 5 and 6.
𝜆𝑤2
Φ (𝑥; 𝑟) = Φ (0; 𝑟) exp ( 𝑥) , 0 ≤ 𝑟 ≤ 1,
𝑏 6. Conclusion
(81)
2
𝜆𝑤 In this paper, the fractional Green’s functions for the solutions
Φ (𝑥; 𝑟) = Φ (0; 𝑟) exp ( 𝑥) , 0 ≤ 𝑟 ≤ 1.
𝑏 of FFDEs by using the fuzzy Laplace transforms under
Abstract and Applied Analysis 11

9 1020

8
1015
7

6
1010

Log(y(x))
y(x)

4 105

3
100
2

1 10−5
0 0.2 0.4 0.6 0.8 1 10−30 10−25 10−20 10−15 10−10 10−5 100
x
Log(x)
Figure 1: Solution of Example 4, Case I, 𝛽 = 1.
Figure 4: Solution of Example 4, Case II, 𝛽 = 0.5.

1016 3
14
10 2.8
2.6
1012
2.4
1010
2.2
Log(y(x))

108
y(x)

2
6
10 1.8

104 1.6
1.4
102
1.2
100 1
10−30 10−25 10−20 10−15 10−10 10−5 100 0 0.2 0.4 0.6 0.8 1
Log(x) x

Figure 2: Solution of Example 4, Case I, 𝛽 = 0.5. Figure 5: Solution of Example 5, 𝛽 = 1.

3 1020

2.5
1015

2
1010
Log(y(x))
y(x)

1.5

105
1

100
0.5

0 10−5
0 0.2 0.4 0.6 0.8 1 10−30 10−25 10−20 10−15 10−10 10−5 100
x Log(x)

Figure 3: Solution of Example 4, Case II, 𝛽 = 1. Figure 6: Solution of Example 5, 𝛽 = 0.5.


12 Abstract and Applied Analysis

Caputo’s differentiability were applied. Some particular cases [15] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory
were also presented, involving some fuzzy versions of the and Applications of Fractional Differential Equations, vol. 204
harmonic oscillator. To this end, some needed concepts are of North-Holland Mathematics Studies, Elsevier Science B.V.,
given like the Caputo derivatives, fuzzy Laplace transforms, Amsterdam, The Netherland, 2006.
and their inverses, and so forth. For future research, we [16] V. Lakshmikantham, S. Leela, and J. Vasundhara Devi, Theory of
extend the boundary value problem for FDE [38] to the Fractional Dynamic Systems, Cambridge Scientific, Cambridge,
fractional case. UK, 2009.
[17] V. Lakshmikantham and R. N. Mohapatra, Theory of Fuzzy
Differential Equations and Inclusions, vol. 6 of Series in Math-
Acknowledgments ematical Analysis and Applications, Taylor & Francis, London,
UK, 2003.
The authors thank the referees for careful reading and helpful [18] V. Lakshmikantham and A. S. Vatsala, “Basic theory of frac-
suggestions on the improvement of the paper. tional differential equations,” Nonlinear Analysis: Theory, Meth-
ods & Applications, vol. 69, no. 8, pp. 2677–2682, 2008.
References [19] R. P. Agarwal, V. Lakshmikantham, and J. J. Nieto, “On the
concept of solution for fractional differential equations with
[1] A. Ahmadian, M. Suleiman, S. Salahshour, and D. Baleanu, “A uncertainty,” Nonlinear Analysis: Theory, Methods & Applica-
Jacobi operational matrix for solving a fuzzy linear fractional tions, vol. 72, no. 6, pp. 2859–2862, 2010.
differential equation,” Advances in Difference Equations, vol. [20] B. Bede and S. G. Gal, “Generalizations of the differentiability
2013, article 104, pp. 1–29, 2013. of fuzzy-number-valued functions with applications to fuzzy
[2] T. Allahviranloo and M. B. Ahmadi, “Fuzzy Laplace trans- differential equations,” Fuzzy Sets and Systems, vol. 151, no. 3,
forms,” Soft Computing, vol. 14, no. 3, pp. 235–243, 2010. pp. 581–599, 2005.
[3] T. Allahviranloo, S. Salahshour, and S. Abbasbandy, “Explicit [21] B. Bede, I. J. Rudas, and A. L. Bencsik, “First order linear
solutions of fractional differential equations with uncertainty,” fuzzy differential equations under generalized differentiability,”
Soft Computing, vol. 16, no. 2, pp. 297–302, 2012. Information Sciences, vol. 177, no. 7, pp. 1648–1662, 2007.
[4] T. Allahviranloo, S. Abbasbandy, S. Salahshour, and A. [22] I. Perfilieva, “Fuzzy transforms: theory and applications,” Fuzzy
Hakimzadeh, “A new method for solving fuzzy linear differen- Sets and Systems, vol. 157, no. 8, pp. 993–1023, 2006.
tial equations,” Computing, vol. 92, no. 2, pp. 181–197, 2011. [23] I. Perfilieva, H. De Meyer, B. De Baets, and D. Plšková, “Cauchy
[5] T. Allahviranloo and S. Salahshour, “A new approach for solving problem with fuzzy initial condition and its approximate
first order fuzzy differential equations,” Communications in solution with the help of fuzzy transform,” in Proceedings of
Computer and Information Science, vol. 81, part 5, part 6, pp. IEEE International Conference on Fuzzy Systems (FUZZ ’08), pp.
522–531, 2010. 2285–2290, June 2008.
[6] J. Li, A. Zhao, and J. Yan, “The Cauchy problem of fuzzy [24] S. Salahshour, T. Allahviranloo, S. Abbasbandy, and D. Baleanu,
differential equations under generalized differentiability,” Fuzzy “Existence and uniqueness results for fractional differential
Sets and Systems, vol. 200, pp. 1–24, 2012. equations with uncertainty,” Advances in Difference Equations,
[7] S. Salahshour and T. Allahviranloo, “Application of fuzzy dif- vol. 2012, article 112, pp. 1–12, 2012.
ferential transform method for solving fuzzy Volterra integral [25] M. Mazandarani and A. V. Kamyad, “Modified fractional Euler
equations,” Applied Mathematical Modelling, vol. 37, no. 3, pp. method for solving fuzzy fractional initial value problem,” Com-
1016–1027, 2013. munications in Nonlinear Science and Numerical Simulation, vol.
[8] S. Salahshour and A. Allahviranloo, “Applications of fuzzy 18, no. 1, pp. 12–21, 2013.
Laplace transforms,” Soft Computing, vol. 17, no. 1, pp. 145–178, [26] R. P. Agarwal, S. Arshad, D. O’Regan, and V. Lupulescu,
2013. “Fuzzy fractional integral equations under compactness type
[9] A. Arara, M. Benchohra, N. Hamidi, and J. J. Nieto, “Fractional condition,” Fractional Calculus and Applied Analysis, vol. 15, no.
order differential equations on an unbounded domain,” Nonlin- 4, pp. 572–590, 2012.
ear Analysis: Theory, Methods & Applications, vol. 72, no. 2, pp. [27] J. Xu, Z. Liao, and Z. Hu, “A class of linear differential dynamical
580–586, 2010. systems with fuzzy initial condition,” Fuzzy Sets and Systems,
[10] Y. I. Babenko, Heat and Mass Transfer, Chemia, Leningrad, vol. 158, no. 21, pp. 2339–2358, 2007.
Russia, 1986. [28] M. Friedman, M. Ma, and A. Kandel, “Numerical solutions
[11] R. L. Bagley, “On the fractional order initial value problem of fuzzy differential and integral equations,” Fuzzy Sets and
and its engineering applications,” in Fractional Calculus and Systems, vol. 106, no. 1, pp. 35–48, 1999.
Its Applications, K. Nishimoto, Ed., pp. 12–20, College of [29] M. Ma, M. Friedman, and A. Kandel, “Numerical solutions of
Engineering, Nihon University, Tokyo, Japan, 1990. fuzzy differential equations,” Fuzzy Sets and Systems, vol. 105,
[12] H. Beyer and S. Kempfle, “Definition of physically consistent no. 1, pp. 133–138, 1999.
damping laws with fractional derivatives,” Zeitschrift für Ange- [30] H.-J. Zimmermann, Fuzzy Set Theory and Its Applications,
wandte Mathematik und Mechanik, vol. 75, no. 8, pp. 623–635, Kluwer Academic Publishers, Boston, Mass, USA, 2nd edition,
1995. 1992.
[13] K. Diethelm and N. J. Ford, “Analysis of fractional differential [31] M. L. Puri and D. A. Ralescu, “Fuzzy random variables,” Journal
equations,” Journal of Mathematical Analysis and Applications, of Mathematical Analysis and Applications, vol. 114, no. 2, pp.
vol. 265, no. 2, pp. 229–248, 2002. 409–422, 1986.
[14] I. Podlubny, Fractional Differential Equations, vol. 198 of Math- [32] H.-C. Wu, “The improper fuzzy Riemann integral and its
ematics in Science and Engineering, Academic Press, San Diego, numerical integration,” Information Sciences, vol. 111, no. 1–4, pp.
Calif, USA, 1999. 109–137, 1998.
Abstract and Applied Analysis 13

[33] G. A. Anastassiou, Fuzzy Mathematics: Approximation Theory,


vol. 251 of Studies in Fuzziness and Soft Computing, Springer,
Berlin, Germany, 2010.
[34] G. A. Anastassiou, Intelligent Mathematics: Computational
Analysis, vol. 5 of Intelligent Systems Reference Library, Springer,
Berlin, Germany, 2011.
[35] S. Salahshour, T. Allahviranloo, and S. Abbasbandy, “Solving
fuzzy fractional differential equations by fuzzy Laplace trans-
forms,” Communications in Nonlinear Science and Numerical
Simulation, vol. 17, no. 3, pp. 1372–1381, 2012.
[36] Y. Chalco-Cano and H. Román-Flores, “On new solutions of
fuzzy differential equations,” Chaos, Solitons and Fractals, vol.
38, no. 1, pp. 112–119, 2008.
[37] A. Khastan, J. J. Nieto, and R. Rodrı́guez-López, “Variation of
constant formula for first order fuzzy differential equations,”
Fuzzy Sets and Systems, vol. 177, pp. 20–33, 2011.
[38] A. Khastan, J. J. Nieto, and R. Rodrı́guez-López, “Periodic
boundary value problems for first-order linear differential
equations with uncertainty under generalized differentiability,”
Information Sciences, vol. 222, pp. 544–558, 2013.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 585310, 10 pages
http://dx.doi.org/10.1155/2013/585310

Research Article
Fractional-Order Total Variation Image Restoration Based on
Primal-Dual Algorithm

Dali Chen,1 YangQuan Chen,2 and Dingyu Xue1


1
College of Information Science and Engineering, Northeastern University, Shenyang, Liaoning 110006, China
2
MESA Lab, University of California, Merced, 5200 North Lake Road, Merced, CA 95343, USA

Correspondence should be addressed to Dali Chen; chendali@ise.neu.edu.cn

Received 7 August 2013; Accepted 27 September 2013

Academic Editor: Dumitru Baleanu

Copyright © 2013 Dali Chen et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper proposes a fractional-order total variation image denoising algorithm based on the primal-dual method, which provides
a much more elegant and effective way of treating problems of the algorithm implementation, ill-posed inverse, convergence
rate, and blocky effect. The fractional-order total variation model is introduced by generalizing the first-order model, and the
corresponding saddle-point and dual formulation are constructed in theory. In order to guarantee 𝑂(1/𝑁2 ) convergence rate, the
primal-dual algorithm was used to solve the constructed saddle-point problem, and the final numerical procedure is given for
image denoising. Finally, the experimental results demonstrate that the proposed methodology avoids the blocky effect, achieves
state-of-the-art performance, and guarantees 𝑂(1/𝑁2 ) convergence rate.

1. Introduction It has been proved that this model is able to deal with
the blocky problem. However, it tends to cause the sign
Since the work of Rudin et al. [1], total variation (TV) of uplifting effect and formation of artifacts around edges.
minimization problems arise in many image processing For this problem, an improved fourth-order PDE model,
applications for regularizing inverse problems where one replacing the Laplacian operator of diffusivity function by the
expects the recovered image or signal to be piecewise constant gradient operator, was proposed in [8].
[2–5]. The typical total variation model, for example, ROF In this paper, our interest focuses on the second gen-
model, has been proved to be able to achieve a good tradeoff eralization which deals with fractional-order differentiation.
between edge preservation and noise removal [1]. However, Fractional calculus is a rapidly growing mathematical disci-
it tends to produce the so-called blocky (staircase) effects on pline, which provides an important tool for nonlocal field
the images because it favors a piecewise constant solution theories [9]. Recently, it has been greatly studied in computer
in bounded variation (BV) space [6]. In order to deal vision [10–16], and the main reason for this development
with blocky effects, the modification of TV model, which is the expectation that the use of this theory will lead to a
generalizes the differential order in regularization term, has much more elegant and effective way of treating problems of
aroused the more and more attentions of numerous scholars. blocky effect and detailed information protection. Specially,
The improved methods of TV model are divided into the fractional-order total variation (TV) models play an
two kinds: the high-order derivative and the fractional-order important role for image denoising, inpainting, and motion
derivative. The first one replaces the first-order derivative estimation [17–21]. So far, the methods adopted to deal
in regularization term by the high-order derivative. For with the fractional-order total variation problem are divided
example, a fourth-order partial differential equation-(PDE- into three kinds: (1) solving the associated Euler-Lagrange
based) denoising mode was proposed by [7], in which the equation, which is a nonlinear partial differential equation
regularized solution is obtained by solving the minimization (PDE) [17, 18]; (2) using the methods based on duality [6];
of potential function of second-order derivative of the image. (3) using the method based on majorization-minimization
2 Abstract and Applied Analysis

(MM) algorithm [22]. Although the convergence of these where √(𝐷1𝛼 𝑢)2 + (𝐷2𝛼 𝑢)2 is the discretization of |𝐷𝛼 𝑢|, 𝐷𝛼
methods has been studied in the corresponding literatures, is the fractional-order derivative operator, and 𝐷𝛼 𝑢 :=
the rate of convergence for them has not been discussed yet.
(𝐷1𝛼 𝑢, 𝐷2𝛼 𝑢)𝑇 . 𝐷1𝛼 and 𝐷2𝛼 are linear operators corresponding
For this problem, in this paper, a primal-dual algorithm is
to horizontal and vertical fractional-order derivative.
used to solve the fractional-order total variation problem,
From Grünwald-Letnikov fractional derivative definition
which is able to guarantee 𝑂(1/𝑁2 ) convergence rate.
[31], the finite fractional-order forward difference can be
The primal-dual algorithm was first presented by [23]
obtained by
and named as the primal-dual hybrid gradient (PDHG)
method in [24]. In this method, each iteration updates both
a primal and a dual variable. It is thus able to avoid some of 𝐾−1 𝐾−1
the difficulties that arise when working only on the primal 𝐷1𝛼 𝑢𝑖,𝑗 = ∑ 𝑤𝑘(𝛼) 𝑢𝑖+𝑘,𝑗 , 𝐷2𝛼 𝑢𝑖,𝑗 = ∑ 𝑤𝑘(𝛼) 𝑢𝑖,𝑗+𝑘 , (4)
or dual side [25, 26]. The convergence of the primal-dual 𝑘=0 𝑘=0
algorithm has been studied in [27, 28]. Recently a unified
form of primal-dual algorithm was presented by [29, 30],
which demonstrated that, in some case, these algorithms can where 𝑤𝑘(𝛼) = (−1)𝑘+1 C𝛼𝑘 , C𝛼𝑘 = Γ(𝛼 + 1)/(Γ(𝑘 + 1)Γ(𝛼 − 𝑘 + 1))
also achieve the 𝑂(1/𝑁2 ) rate of convergence. In our paper, denotes the generalized binomial coefficient, and Γ(𝑥) is the
a new image denoising method is proposed, in which the Gamma function. In addition, the coefficients 𝑤(𝛼) can also
primal-dual algorithm is used to solve the fractional-order be obtained recursively from
total variation denoising model. The proposed denoising
method is able to avoid the blocky effect, achieves state- 𝛼+1
of-the-art performance, and guarantees the 𝑂(1/𝑁2 ) rate of 𝑤0(𝛼) = −1, 𝑤𝑘(𝛼) = (1 − (𝛼)
) 𝑤𝑘−1 , 𝑘 = 1, 2, . . . .
𝑘
convergence. (5)
This paper is organized as follows. Section 1 introduces
prior work, focusing on the main problems with existing
methods that are addressed by our model. In Section 2, the When 𝛼 = 1, 𝑤𝑘1 = 0 for 𝑘 > 1 and (4) is the first-order
fractional-order total variation denoising model is described, forward derivative as usual.
and the corresponding saddle-point and dual formulation are To describe (2) in matrix algebra language, we reord
constructed in theory. Based on this, the primal-dual algo- the image matrix 𝑢 and 𝑓 row wisely into the vector 𝑥
rithm was used to solve the constructed saddle-point prob- and 𝑧, associating the (𝑖, 𝑗) element of the two-dimensional
lem, and the final numerical implementation is presented structure with the element (𝑗 − 1)𝑛 + 𝑖 of the vector structure,
for image denoising. Experimental evaluation is presented in 𝑥(𝑗−1)𝑛+𝑖 = 𝑢𝑖,𝑗 , and 𝑧(𝑗−1)𝑛+𝑖 = 𝑓𝑖,𝑗 . We have 𝑥 ∈ R𝑁
Section 3 and the paper is concluded in Section 4. and 𝑧 ∈ R𝑁, where 𝑁 = 𝑛2 . The (𝑖, 𝑗) component of the
fractional-order derivative 𝐷𝛼 𝑢 can thus be represented as a
2. Fractional-Order Total Variation Model and multiplication of the vector 𝑥 ∈ R𝑁 by a matrix 𝐴 𝑚 ∈ R2×𝑁,
Primal-Dual Algorithm for 𝑚 = 1, 2, . . . , 𝑁:

𝑇
2.1. Model Description. Let 𝑓𝑖,𝑗 = 𝑢𝑖,𝑗 + V𝑖,𝑗 denote the {
{
𝐾−1 𝐾−1
{
{ ( ∑ 𝑤𝑘(𝛼) 𝑥𝑚+𝑘 , ∑ 𝑤𝑘(𝛼) 𝑥𝑚+𝑛𝑘 , )
observed noisy image, where (𝑖, 𝑗)𝑇 denotes the location with {
{
{ 𝑘=0 𝑘=0
a rectangular image domain Ω ∈ R𝑛×𝑛 and V is white {
{ if (𝑚 mod 𝑛) ≠0 and 𝑚 ≤ 𝑁 − 𝑛
{
{
Gaussian noise. The typical total variation (TV) denoising {
{ 𝑇
{
{ 𝐾−1
model estimates the desired clean image 𝑢𝑖,𝑗 by solving the {
{(0, ∑ 𝑤 (𝛼)
𝑥 ) ,
{
{ 𝑘 𝑚+𝑛𝑘
following finite-dimensional optimization problem: {
{ 𝑘=0
𝐴 𝑚 𝑥 = {if (𝑚 mod 𝑛) = 0 and 𝑚 ≤ 𝑁 − 𝑛 (6)
𝜆󵄩 {
{ 𝑇
󵄩2 {
{
𝑢̂ = arg min {𝐸 (𝑢) := TV (𝑢) + 󵄩󵄩󵄩𝑓 − 𝑢󵄩󵄩󵄩2 } , (1) {
{ (
𝐾−1
∑ 𝑤(𝛼)
𝑥 , 0) ,
𝑢∈Ω 2 {
{ 𝑘 𝑚+𝑘
{
{ 𝑘=0
{
{
where TV(𝑢) is the regularization term, ‖𝑓 − 𝑢‖22 is the data {
{ if (𝑚 mod 𝑛) ≠0 and 𝑚 > 𝑁 − 𝑛
{
{
fidelity term, ‖𝑢‖] is ]-norm of 𝑢, and 𝜆 is regularization {
{(0, 0)𝑇 ,
{
{
{if (𝑚 mod 𝑛) = 0 and 𝑚 > 𝑁 − 𝑛.
parameter which controls the degree of smoothing. In this
paper, we consider a fractional-order total variation model,
defined as
From this definition, the discrete version of the prime
𝜆 󵄩󵄩 󵄩2
𝑢̂ = arg min {𝐸 (𝑢) := TV𝛼 (𝑢) + 󵄩󵄩𝑓 − 𝑢󵄩󵄩󵄩2 } , (2) fractional-order total variation (FOTV) model (2) can be
𝑢∈Ω 2 written as
where TV𝛼 (𝑢) is obtained by the following formula:
𝑁
󵄩󵄩󵄩 2󵄩
󵄩 󵄩 󵄩 𝜆
2 󵄩
TV𝛼 (𝑢) = 󵄩󵄩󵄩√(𝐷1𝛼 𝑢) + (𝐷2𝛼 𝑢) 󵄩󵄩󵄩 , (3) min𝑁 {𝐸 (𝑥) := ∑ 󵄩󵄩󵄩𝐴 𝑚 𝑥󵄩󵄩󵄩2 + ‖𝑥 − 𝑧‖22 } . (7)
󵄩󵄩 󵄩󵄩1 𝑥∈𝑅 𝑚=1 2
Abstract and Applied Analysis 3

Step 1. Choose 𝜏0 , 𝜎0 > 0 with 𝜏0 𝜎0 𝐿2 ≤ 1, 𝑥0 ∈ 𝑅𝑁 , 𝑦0 ∈ 𝑅2𝑁 , and 𝑥0 = 𝑥0 .


Step 2. Update 𝑥𝑡 , 𝑦𝑡 , 𝑥𝑡 , 𝜃𝑡 , 𝜏𝑡 , 𝜎𝑡 as follows:
1 󵄩󵄩 󵄩2
𝑦𝑡+1 = arg min ⟨𝑦, −𝐴𝑥𝑡 ⟩ + 𝛿 (𝑦) + 󵄩𝑦 − 𝑦𝑡 󵄩󵄩󵄩2 , (A)
𝑦∈𝑅2𝑁 2𝜎𝑡 󵄩
𝜆 1 󵄩󵄩 󵄩2
𝑥𝑡+1 = arg min ‖𝑥 − 𝑧‖22 + ⟨𝐴𝑥, 𝑦𝑡+1 ⟩ + 󵄩𝑥 − 𝑥𝑡 󵄩󵄩󵄩2 , (B)
𝑥∈𝑅𝑁 2 2𝜏𝑡 󵄩
1 𝜎
𝜃𝑡 = , 𝜏𝑡+1 = 𝜃𝑡 𝜏𝑡 , 𝜎𝑡+1 = 𝑡 ,
√1 + 2𝛾𝜏𝑡 𝜃𝑡
𝑥𝑡+1 = 𝑥𝑡+1 + 𝜃𝑡 (𝑥𝑡+1 − 𝑥𝑡 ) ,

Algorithm 1

Define a norm ‖V‖𝐸 = ∑𝑁 2 2 𝑇


𝑚=1 √V1 + V2 where V = (V1 , V2 ) , and
where −𝐴𝑇 is the corresponding discretization of the
(7) can be rewritten in term of ‖ ⋅ ‖𝐸 as fractional-order divergence, and when 𝛼 = 1, it is the discrete
divergence as usual.
𝜆 For primal-dual feasible pair (𝑥, 𝑦), the partial primal-
min𝑁 {𝐸 (𝑥) := ‖𝐴𝑥‖𝐸 + ‖𝑥 − 𝑧‖22 } , (8) dual gap 𝐺(𝑥, 𝑦) is defined by
𝑥∈𝑅 2
Here 𝐴 = [𝐴 1 , 𝐴 2 , . . . , 𝐴 𝑁]𝑇 ∈ 𝑅2𝑁×𝑁.
𝜆
𝐺 (𝑥, 𝑦) = { max ⟨𝑦󸀠 , 𝐴𝑥⟩ − 𝛿 (𝑦󸀠 ) + ‖𝑥 − 𝑧‖22 }
2.2. Dual Problem and Saddle-Point Problem. For any vector, 𝑦󸀠 ∈𝑅2𝑁 2
2 𝑇 (14)
𝑝𝑚 = (𝑝𝑚 1
, 𝑝𝑚 ) ∈ 𝑅2 and 𝑝 = (𝑝1 , 𝑝2 , . . . , 𝑝𝑁)𝑇 ∈ 𝑅2𝑁. We 𝜆󵄩 󵄩2
have ‖𝐴𝑥‖𝐸 = max‖𝑝‖𝐸∗ ≤1 ⟨𝐴𝑥, 𝑝⟩, where ‖ ⋅ ‖𝐸∗ is the dual − { min ⟨𝑦, 𝐴𝑥 ⟩ − 𝛿 (𝑦) + 󵄩󵄩󵄩󵄩𝑥󸀠 − 𝑧󵄩󵄩󵄩󵄩2 } .
󸀠

norm of norm ‖ ⋅ ‖𝐸 , and the ‖ ⋅ ‖𝐸 and ‖ ⋅ ‖𝐸∗ are analogous


󸀠
𝑥 ∈𝑅𝑁 2
to ‖ ⋅ ‖1 and ‖ ⋅ ‖∞ , respectively.
Let the function 𝛿(𝑦) : 𝑅2𝑁 → 𝑅 be the conjugate of ̂ 𝑦)
If (𝑥, ̂ is the saddle point of problem (11), then 𝑥̂ is primal
‖𝐴𝑥‖𝐸 ; we have optimal, 𝑦̂ is dual optimal [32], and we have
𝛿 (𝑦) = sup ⟨𝑦, 𝐴𝑥⟩ − ‖𝐴𝑥‖𝐸
𝑥
𝜆
̂ 𝐴𝑥⟩ − 𝛿 (𝑦)
𝐺 (𝑥, 𝑦) ≥ {⟨𝑦, ̂ + ‖𝑥 − 𝑧‖22 }
= sup ⟨𝑦, 𝐴𝑥⟩ − max ⟨𝐴𝑥, 𝑝⟩ 2
𝑥 ‖𝑝‖𝐸∗ ≤1 (9) (15)
𝜆
0, if ‖V‖𝐸∗ ≤ 1 ̂ − 𝛿 (𝑦) + ‖𝑥̂ − 𝑧‖22 } ≥ 0,
− {⟨𝑦, 𝐴𝑥⟩
={ 2
∞, if otherwise,

in which case we can interpret 𝛿(⋅) as the indicator function which vanishes only if (𝑥, 𝑦) is itself the saddle point [29].
for the unit ball in the dual norm, and 𝑦 = (𝑦1 , 𝑦2 , . . . , 𝑦𝑁)𝑇 ∈ The primal-dual gap 𝐺(𝑥, 𝑦) is a measure of closeness of the
2 𝑇 primal-dual (𝑥, 𝑦) to the primal-dual solution, and we use it
𝑅2𝑁 with 𝑦𝑚 = (𝑦𝑚 1
, 𝑦𝑚 ) ∈ 𝑅2 . Since 𝐹 = 𝐹∗∗ if 𝐹 is closed
to design the stopping criterion for our numerical algorithm
and convex, we have
in this paper.
‖𝐴𝑥‖𝐸 = sup ⟨𝑦, 𝐴𝑥⟩ − 𝛿 (𝑦) . (10)
𝑦∈𝑅2𝑁
2.3. Primal-Dual Method for Solving Saddle-Point Problem.
Substituting (10) into (8), we can obtain the saddle-point The first-order primal-dual method summarized in [29]
formulation of FOTV model defined by for convex problem was applied in this paper to solve the
saddle-point problem described by (11), which is described
𝜆
min sup ⟨𝑦, 𝐴𝑥⟩ − 𝛿 (𝑦) + ‖𝑥 − 𝑧‖22 . (11) by Algorithm 1.
𝑥∈𝑅𝑁 𝑦∈𝑅2𝑁 2 Let 𝐿 = ‖𝐴‖ and 𝜏𝑡 𝜎𝑡 𝐿2 ≤ 1; then there exists a saddle-point
̂ 𝑦)
(𝑥, ̂ such that 𝑥𝑡 → 𝑥̂ and 𝑦𝑡 → 𝑦. ̂ Since (𝜆/2)‖𝑥 − 𝑧‖22
The minimization problem of (11) can be solved exactly as
is uniformly convex such that it has a Lipschitz continuous
1 𝑇 gradient, the output (𝑥𝑡 , 𝑦𝑡 ) possesses a rate of convergence of
𝑥=𝑧− 𝐴 𝑦. (12)
𝜆 𝑂(1/𝑁2 ) [29]. More specifically, we assume that the constant
stepsizes are used; that is, 𝜏𝑡 = 𝜏 > 0, 𝜎𝑡 = 𝜎 > 0, and 𝜃𝑡 = 𝜃
Substituting (12) into (11) yields the following dual problem:
for all 𝑡 ≥ 0. If 𝜏𝜎𝐿2 < 1; then the 𝑂(1/𝑁) convergence rate
󵄩 󵄩2 can be guaranteed for 𝜃 = 1, and 𝑂(1/√𝑁) convergence rate
max − (󵄩󵄩󵄩󵄩𝐴𝑇 𝑦 − 𝜆𝑧󵄩󵄩󵄩󵄩2 + 𝛿 (𝑦)) , (13)
𝑦∈𝑅2𝑁 can be guaranteed for 𝜃 = 0 [30].
4 Abstract and Applied Analysis

To summarize, our entire noise removal algorithm in a iteration of our method performs the following updates on
form of a pseudocode is done in following: prime variable 𝑢 ∈ R𝑛×𝑛 and dual variable 𝑝 ∈ R𝑛×𝑛×2 :
(1) initialization: 𝑝𝑘 + 𝜎𝑘 𝐷𝛼 𝑢𝑘
𝑝𝑘+1 = 󵄨 󵄨 , (19)
max {1, 󵄨󵄨󵄨󵄨𝑝𝑘 + 𝜎𝑘 𝐷𝛼 𝑢𝑘 󵄨󵄨󵄨󵄨}
(i) for a given noisy image 𝑓 of size 𝑛 × 𝑛, we reord
the image 𝑓 row wisely into the vector 𝑧 ∈ 𝑅𝑁 𝑢𝑘 − 𝜏𝑘 div𝛼 𝑝𝑘+1 + 𝜏𝑘 𝜆𝑓
with 𝑁 = 𝑛 × 𝑛, 𝑧(𝑗−1)𝑛−𝑖 = 𝑓𝑖,𝑗 ; 𝑢𝑘+1 = , (20)
1 + 𝜏𝑘 𝜆
(ii) compute the fractional-order discrete operator
𝐴 according to (6); where the dual variable 𝑝 = (𝑝1 , 𝑝2 )𝑇 corresponds with 𝑦
(iii) initialize the regularization parameter 𝜆, itera- and the discrete fractional divergence operator div𝛼 is adjoint
tion number 𝑇, and maximum permissible error to the discrete fractional gradient operator 𝐷𝛼 . Note that
𝐸mpe ; (19) and (20), respectively, correspond with the solutions of
(iv) set 𝑡 = 0, 𝑥0 = 𝑧, 𝐿 = ‖𝐴‖, 𝜏0 = 1/𝐿2 , 𝜎0 = 1/𝐿2 , problem (A) and (B). The updates of the other parameters
and 𝛾 = 0.7𝜆; are the same as the description in the foregoing pseudocode,
and the number of iteration is fixed to ensure the correct
timing sequence. After the iteration, the denoised image 𝑢 is
(2) iteration: compute 𝑥 by the following steps.
outputted, the next image frame is captured, and the same
Step 1. Solve the problem (A) by process is repeated until the assignment is over.

𝑡+1 𝑦𝑚𝑡
+ 𝜎𝑡 𝐴 𝑚 𝑥𝑡 3. Experiments and Analysis
𝑦𝑚 = 󵄩 𝑡 󵄩 , 𝑚 = 1, 2, . . . , 𝑁. (16)
max {1, 󵄩󵄩󵄩𝑦𝑚 + 𝜎𝑡 𝐴 𝑚 𝑥𝑡 󵄩󵄩󵄩2 }
3.1. Restraint of Block Effect. The blocky effect is the main
drawback produced by the typical first-order total variation
Step 2. Substitute 𝑦𝑡+1 into the (B) and solve the problem (B) denoising algorithm. In this section, some experiments are
by given to assess the capability of reducing block effect of our
proposed fractional-order total variation (FOTV) denoising
𝑥𝑡 − 𝜏𝑡 𝐴𝑇 𝑦𝑡+1 + 𝜏𝑡 𝜆𝑧 algorithm. Firstly, the one-dimensional signal is used as the
𝑥𝑡+1 = . (17)
1 + 𝜏𝑡 𝜆 test signal, 𝑦(𝑡) = 4 sin(2𝜋𝑡) + 8 sin(3𝜋𝑡). We contaminate
the given signal using the additive white Gaussian noise
Step 3. Compute 𝜃𝑡 = 1/√1 + 2𝛾𝜏𝑡 , 𝜏𝑡+1 = 𝜏𝑡 𝜃𝑡 , and 𝜎𝑡+1 = (AWGN) with standard deviation (SD) of 0.5. The corrupted
𝜎𝑡 /𝜃𝑡 . signal is shown in Figure 1(a). The proposed FOTV denoising
algorithm with 𝛼 = 1 and 𝛼 = 1.5 is used to process the
Step 4. Update 𝑥 by 𝑥𝑡+1 = (1 + 𝜃𝑡 )𝑥𝑡+1 − 𝜃𝑡 𝑥𝑡 . contaminated signal, respectively. When 𝛼 = 1, our FOTV
algorithm is the typical ROF denoising algorithm based on
Step 5. Compute the primal-dual gap 𝑔𝑡+1 according to (15) primal-dual method. In this experiment, we set 𝜆 = 0.01, and
as follows: the results at the 1000th iteration are, respectively, shown in
Figures 1(b) and 1(c). The difference is obvious: while the first-
𝑁
󵄩 󵄩 𝑇 order TV denoising algorithm approximates the observed
𝑔𝑡+1 = ∑ 󵄩󵄩󵄩󵄩𝐴 𝑚 𝑥𝑡+1 󵄩󵄩󵄩󵄩2 − (𝑦𝑡+1 ) 𝐴𝑥𝑡+1
signal with a step signal, the fractional-order algorithm with
𝑚=1
(18) a piecewise planar signal which looks more natural and does
𝜆 󵄩󵄩 1 󵄩󵄩2 not produce false edges.
+ 󵄩󵄩󵄩󵄩𝑥𝑡+1 − 𝑧 + 𝐴𝑇 𝑦𝑡+1 󵄩󵄩󵄩󵄩 . We now consider a two-dimensional image “Lena” of
2󵄩 𝜆 󵄩2
size 512 × 512. The image is degraded by using AWGN
Step 6. If (𝑔𝑡+1 < 𝐸mpe ) or (𝑡 > 𝑇), then we terminate the with standard deviation of 10 and the result is shown in
iteration and output 𝑥𝑡+1 ; otherwise, go back to Step 1. Figure 2(a). This degraded image is, respectively, fed into the
proposed FOTV denoising algorithm with 𝛼 = 1 and 𝛼 = 1.5
Due to the simplicity of our model, it is easy to be as initial condition, and the time evolution of the algorithms
implemented and can be effectively accelerated on parallel begins. The results at the 1000th iteration are, respectively,
hardware such as field programmable gata array (FPGA) and shown in Figures 2(b) and 2(c). The blocky effects are obvious
graphics processing unit (GPU). In order to employ the huge in Figure 2(b), while the Figure 2(c) looks more natural and
computational power and the parallel processing capabilities does not produce blocky effect.
of FPGA to obtain a fully accelerated implementation of our In conclusion, the fractional-order TV algorithm can
denoising method, the numerical method should be working reduce blocky effect effectively comparing with the tradi-
on the regular grids. For this purpose, the implementation of tional first-order TV algorithm.
our algorithm on FPGA is described as the following strategy.
First, an image frame is captured and read in the processor 3.2. Analysis of Denoising Performance. The aim of this
from a live camera. For the captured frame 𝑓 ∈ R𝑛×𝑛 , the 𝑘th section is to analyze the denoising performance of the FOTV
Abstract and Applied Analysis 5

10 10

5 5

y
0 y 0

−5 −5

−10 −10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t t
(a) Noisy signal (b) Processed by FOTV with 𝛼 = 1

10

y
0

−5

−10
0 0.2 0.4 0.6 0.8 1
t
(c) Processed by FOTV with 𝛼 = 1.5

Figure 1: Signal processed by FOTV with 𝛼 = 1 and 𝛼 = 1.5.

algorithm. For this purpose, three famous 512 × 512 test Peppers image. From the figure, we can obtain the following
images, called Barbara, Lena, and Peppers, are used in the conclusions. Firstly, the PSNR reaches a maximum between
experiments, as shown in Figure 3. In order to quantify the 𝛼 = 1 and 𝛼 = 2. Secondly, the PSNR at 𝛼 = 1 is lower than
denoised image, we consider the peak-signal-to-noise ratio PSNR at 𝛼 > 1, which is owing to the blocky effect. Finally, the
(PSNR), which has been largely used in the literature and PSNR decreases rapidly as 𝛼 tends to zero, and when 𝛼 = 0,
commonly applied to determine the quality of a processed the PSNR reaches a minimum. According to these facts, we
image. It can be calculated by the following formula: can select the fractional-order 𝛼 between 𝛼 = 1.2 and 𝛼 = 2.
𝑉𝐿2 𝑀𝑁 In order to analyze the denoising performance, the addi-
PSNR = 10log10 2
, (21) tive white Gaussian noise (AWGN) with standard deviation
∑𝑁 𝑀
̂ (𝑖, 𝑗))
𝑖=1 ∑𝑗=1 (𝑢 (𝑖, 𝑗) − 𝑢 (SD) of 10, 20, and 30 is, respectively, added in the three
where 𝑢 is the original image, 𝑢̂ is the denoised image, and 𝑉𝐿 test images. Five denoising algorithms are used to process
is the maximal gray level of the image. these given noisy images, which are the improved Perona and
Firstly, in order to decide the value of the fractional- Malik (IP-M) model [33], fourth order (F-O) PDE model [7],
order 𝛼, we study the relation between the PSNR and 𝛼. improved fourth order (IF-O) PDE model [8], ROF model
Figure 4 shows the relation between the PSNR and 𝛼 on Lena [1], and the proposed FOTV denoising algorithm. In this
and Peppers image corrupted by the additive white Gaussian experiment, we set 𝜎 = 5 and Δ𝑡 = 0.25 for IP-M, F-O-PDE,
noise with standard deviation (SD) of 20. The left image is and IF-O-PDE models, and we set 𝜆 = 0.07, 𝜏 = 0.1, and
the result of Lena image and the right one is the result of 𝐾 = 20 for ROF model. These parameters are able to ensure
6 Abstract and Applied Analysis

(a) Noisy image (b) Processed by FOTV with 𝛼 = 1

(c) Processed by FOTV with 𝛼 = 1.5

Figure 2: Image processed by FOTV with 𝛼 = 1 and 𝛼 = 1.5.

Table 1: PSNR quantitative comparison among five denoising models.

Image SD IP-M F-O-PDE IF-O-PDE ROF FOTV


10 31.2427 29.3776 29.3777 31.0871 31.2850
Barbara 20 26.6231 24.8155 24.8156 26.8212 27.0976
30 24.4496 22.9267 22.9273 24.7415 24.9479
10 33.6393 31.5440 31.5442 33.8378 34.4747
Lena 20 29.7665 27.8080 27.8098 30.4101 31.2153
30 27.4437 26.0883 26.1019 28.5889 29.3855
10 33.6967 31.8199 31.8205 33.8715 34.0619
Peppers 20 30.0275 28.0437 28.0457 30.1768 31.0285
30 27.4982 26.1613 26.1694 28.2689 29.0046

the best denoising performance of the corresponding denois- the PSNR is, the better the denoising performance is. For easy
ing model. Based on the conclusion of the foregoing experi- observation, the biggest PSNR values are shown in boldface.
ment, we set 𝛼 = 1.8 for FOTV denoising algorithm, and the From the table, it is obvious that the PSNR of our proposed
denoising results are shown in Table 1. In the table, the first algorithm is bigger than that of the other four algorithms,
column lists the given image and the second column lists the so we can conclude that our FOTV denoising algorithm
standard deviation of noisy image. The PSNRs of the denoised outperforms the other algorithms.
images processed by the different denoising algorithms are In order to further verify the denoising performance of
listed under the corresponding denoising model. The bigger our proposed algorithm, Figure 5 shows the denoised images
Abstract and Applied Analysis 7

(a) Barbara (b) Lena

(c) Peppers

Figure 3: Original image: Barbara, Lena, and Peppers.

32 X: 1 32
Y: 30.75
X: 1.7 X: 2
X: 1.5 X: 2
30 Y: 31.32 Y: 31.27
30 X: 1
Y: 31.04 Y: 30.96
Y: 30.66

28
28
26
PSNR

PSNR

26
24
24
22

20 22

18 20
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4
𝛼 𝛼
(a) Lena (b) Peppers

Figure 4: Relation between the fractional-order 𝛼 and PSNR on Lena (a) and Peppers (b) image corrupted by the additive white Gaussian
noise with standard deviation (SD) of 20.
8 Abstract and Applied Analysis

(a) Barbara with 20 SD Gaussian noise (b) Partial view of (a) (c) IP-M algorithm

(d) IF-O PDE algorithm (e) ROF algorithm (f) FOTV algorithm

Figure 5: Comparison of four algorithms for the noisy image with additive Gaussian white noise of 𝜎 = 20.

processed by four denoising algorithms. The first figure is iteration. Figure 6(a) shows the relation between the iteration
the famous image “Barbara” corrupted by the additive white and PSNR. The horizontal axis of this figure is iteration
Gaussian noise with standard deviation (SD) of 20. In order number and the vertical axis is PSNR. The different colorful
to show the comparison clearly, the partial enlarged view curve denotes the result obtained from the different noisy
of the noisy image is shown in the second figure. The image. The red dotted line is the result of the noisy image
denoised result of IP-M algorithm is shown in the right-top with 𝜎 = 30, the green dashed line is the result of the noisy
figure. The second row shows the denoised images processed image with 𝜎 = 20, and the red real line is the result of the
by IF-O PDE algorithm, ROF algorithm, and our FOTV noisy image with 𝜎 = 10. From the figure, we can see that
algorithm. In Figures 5(c) and 5(d), although the detailed the PSNR values are stable after 50 iterations, which indicate
information is preserved, there are a lot of noises unremoved. that our proposed FOTV denoising algorithm is convergent.
The result in Figure 5(e) looks blocky. Only the result of our In addition, it can be seen that the bigger the variance of white
model, as shown in Figure 5(f), looks natural and does not noise is, the longer the stopping times are and the lower the
produce false edges. So we conclude that our proposed FOTV PSNRs are.
algorithm is able to achieve a better tradeoff between edge Figure 6 plots the convergence of the FOTV algorithm for
preservation and noise removal. the above experiment together with the theoretical 𝑂(1/𝑁2 )
and 𝑂(1/𝑁4 ) rate. For easy observation, the horizontal axis
3.3. The Convergence and the Rate of Convergence. The is the log curve of iteration number and the vertical axis is
convergence and the rate of convergence are two important the log curve of primal-dual gap calculated in Step 5. The
factors for evaluating the performance of the denoising dashed line shows the theoretical 𝑂(1/𝑁2 ) rate and the dotted
method. In this section, some experiments are given to line shows the theoretical 𝑂(1/𝑁4 ) rate. The red line shows
show the convergence and convergence rate of our proposed the convergence of FOTV algorithm in the noisy image with
fractional-order total variation (FOTV) denoising algorithm. 𝜎 = 30, the green line shows the convergence of FOTV
For this purpose, the proposed FOTV algorithm is, algorithm in the noisy image with 𝜎 = 20, and the blue line
respectively, used to process the image “Lena” corrupted by shows the convergence of FOTV algorithm in the noisy image
the additive white Gaussian noise with standard deviation with 𝜎 = 10. From this figure, it can be seen that the smaller
(SD) of 10, 20, and 30, and the PSNR is recorded at each the variance of white noise is, the faster convergence is. In
Abstract and Applied Analysis 9

36 106

34

32 104

30

28 102

Error
PSNR

26

24 100

22

20 10−2

18
0 50 100 150 200 100 101 102 103
Iteration number Iterations

𝜎 = 10 FOTV(𝜎 = 10) O(1/N2 )


𝜎 = 20 FOTV(𝜎 = 20) O(1/N4 )
𝜎 = 30 FOTV(𝜎 = 30)
(a) Relation between the iteration and PSNR (b) Relation between the iteration and primal-dual gap

Figure 6: Relation among the iteration, PSNR, and primal-dual gap on the noisy Lena image with 𝜎 = 10, 𝜎 = 20, and 𝜎 = 30.

addition, we can see that the FOTV algorithm converges at [3] C. Vogel and M. Oman, “Fast, robust total variation-based
𝑂(1/𝑁2 ) rate firstly and 𝑂(1/𝑁4 ) rate in the end. reconstruction of noisy, blurred images,” IEEE Transactions on
Image Processing, vol. 7, no. 6, pp. 813–824, 1998.
[4] F. Alter, S. Durand, and J. Froment, “Adapted total variation
4. Conclusion for artifact free decompression of JPEG images,” Journal of
Mathematical Imaging and Vision, vol. 23, no. 2, pp. 199–211,
A fractional-order total variation image denoising algorithm 2005.
based on primal-dual method was proposed in this paper. [5] F. Li, C. Shen, and C. Li, “Multiphase soft segmentation with
The main contributions are as follows: (1) the saddle point total variation and H 1 regularization,” Journal of Mathematical
and dual formulation of the proposed fractional-order total Imaging and Vision, vol. 37, no. 2, pp. 98–111, 2010.
variation model are constructed in theory; (2) the primal- [6] J. Zhang, Z. Wei, and L. Xiao, “Adaptive fractional-order multi-
dual algorithm was used to solve the constructed saddle- scale method for image denoising,” Journal of Mathematical
point problem, in which 𝑂(1/𝑁2 ) convergence rate can Imaging and Vision, vol. 43, no. 1, pp. 39–49, 2012.
be guaranteed; (3) the experiments demonstrate that the [7] Y. L. You and M. Kaveh, “Fourth-order partial differential
proposed methodology is able to avoid the blocky effect, equations for noise removal,” IEEE Transactions on Image
achieve state-of-the-art performance, and guarantee 𝑂(1/𝑁2 ) Processing, vol. 9, no. 10, pp. 1723–1730, 2000.
convergence rate. [8] M. Hajiaboli, “A self-governing fourth-order nonlinear dif-
fusion filter for image noise removal,” IPSJ Transactions on
Computer Vision and Applications, vol. 2, pp. 94–103, 2010.
Acknowledgments [9] R. Herrmann, Fractional Calculus: An Introduction for Physi-
cists, World Scientific, New Jersey, NJ, USA, 2011.
This work was supported by National Natural Science Foun-
dation of China (no. 61201378), Scientific Research Fund of [10] S.-C. Liu and S. Chang, “Dimension estimation of discrete-time
fractional Brownian motion with applications to image texture
Liaoning Provincial Education Department (L2012073), and
classification,” IEEE Transactions on Image Processing, vol. 6, no.
Fundamental Research Funds for the Central Universities 8, pp. 1176–1184, 1997.
(N110304001).
[11] D. Chen, Y. Chen, and D. Xue, “Digital fractional order
Savitzky-Golay differentiator,” IEEE Transactions on Circuits
References and Systems II, vol. 58, no. 11, pp. 758–762, 2011.
[12] S. Didas, B. Burgeth, A. Imiya, and J. Weickert, “Regularity and
[1] L. Rudin, S. Osher, and E. Fatemi, “Nonlinear total variation scale-space properties of fractional high order linear filtering,”
based noise removal algorithms,” Physica D, vol. 60, no. 1–4, pp. in 5th International Conference on Scale Space and PDE Methods
259–268, 1992. in Computer Vision, Scale-Space 2005, pp. 13–25, deu, April
[2] J.-F. Aujol, “Some first-order algorithms for total variation based 2005.
image restoration,” Journal of Mathematical Imaging and Vision, [13] B. Ninness, “Estimation of 1/𝑓 Noise,” IEEE Transactions on
vol. 34, no. 3, pp. 307–327, 2009. Information Theory, vol. 44, no. 1, pp. 32–46, 1998.
10 Abstract and Applied Analysis

[14] I. Petras, D. Sierociuk, and I. Podlubny, “Identification of [32] S. Boyd and L. Vandenberghe, Convex Optimization, Cambridge
parameters of a half-order system,” IEEE Transactions on Signal University Press, New York, NY, USA, 2004.
Processing, vol. 60, no. 10, pp. 5561–5566, 2012. [33] P. Perona and J. Malik, “Scale-space and edge detection using
[15] D. Chen, Y. Chen, and D. Xue, “1-D and 2-D digital fractional- anisotropic diffusion,” IEEE Transactions on Pattern Analysis
order Savitzky-Golay differentiator,” Signal, Image and Video and Machine Intelligence, vol. 12, no. 7, pp. 629–639, 1990.
Processing, vol. 6, no. 3, pp. 503–511, 2012.
[16] Y. F. Pu, J. L. Zhou, and X. Yuan, “Fractional differential mask:
a fractional differential-based approach for multiscale texture
enhancement,” IEEE Transactions on Image Processing, vol. 19,
no. 2, pp. 491–511, 2010.
[17] D. Chen, H. Sheng, Y. Chen, and D. Xue, “Fractional-order vari-
ational optical flow model for motion estimation,” Philosophical
Transactions of the Royal Society A, vol. 371, no. 1990, 2013.
[18] J. Bai and X. Feng, “Fractional-order anisotropic diffusion for
image denoising,” IEEE Transactions on Image Processing, vol.
16, no. 10, pp. 2492–2502, 2007.
[19] P. Guidotti and J. V. Lambers, “Two new nonlinear nonlocal
diffusions for noise reduction,” Journal of Mathematical Imaging
and Vision, vol. 33, no. 1, pp. 25–37, 2009.
[20] E. Cuesta, M. Kirane, and S. A. Malik, “Image structure preserv-
ing denoising using generalized fractional time integrals,” Signal
Processing, vol. 92, no. 2, pp. 553–563, 2012.
[21] M. Janev, S. Pilipović, T. Atanacković, R. Obradović, and
N. Ralević, “Fully fractional anisotropic diffusion for image
denoising,” Mathematical and Computer Modelling, vol. 54, no.
1-2, pp. 729–741, 2011.
[22] D. Chen, S. Sun, C. Zhang, Y. Chen, and D. Xue, “Fractional-
order TV-L 2 model for image denoising,” Central European
Journal of Physics, 2013.
[23] K. Arrow, L. Hurwicz, and H. Uzawa, “Studies in linear and non-
linear programming,” in Stanford Mathematical Studies in the
Social Sciences, Stanford University Press, Palo Alto, Calif, USA,
1958.
[24] M. Zhu and T. Chan, “An efficient primal-dual hybrid gradient
algorithm for total variation image restoration,” CAM Report
08-34, UCLA, Los Angeles, Calif, USA, 2008.
[25] A. Chambolle, “An Algorithm for Total Variation Minimization
and Applications,” Journal of Mathematical Imaging and Vision,
vol. 20, no. 1-2, pp. 89–97, 2004.
[26] E. Esser, X. Zhang, and T. F. Chan, “A general framework
for a class of first order primal-dual algorithms for convex
optimization in imaging science,” SIAM Journal on Imaging
Sciences, vol. 3, no. 4, pp. 1015–1046, 2010.
[27] S. Bonettini and V. Ruggiero, “On the convergence of primal-
dual hybrid gradient algorithms for total variation image
restoration,” Journal of Mathematical Imaging and Vision, vol.
44, no. 3, pp. 236–253, 2012.
[28] B. He and X. Yuan, “Convergence analysis of primal-dual
algorithms for a saddle-point problem: from contraction per-
spective,” SIAM Journal on Imaging Sciences, vol. 5, no. 1, pp.
119–149, 2012.
[29] A. Chambolle and T. Pock, “A first-order primal-dual algorithm
for convex problems with applications to imaging,” Journal of
Mathematical Imaging and Vision, vol. 40, no. 1, pp. 120–145,
2011.
[30] Y. Chen, G. Lan, and Y. Ouyang, “Optimal primal-dual methods
for a class of saddle point problems,” CAM Report 13-31, UCLA,
Los Angeles, Calif, USA, 2013.
[31] I. Podlubny, Fractional Differential Equations, Academic Press,
New York, NY, USA, 1999.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 810625, 9 pages
http://dx.doi.org/10.1155/2013/810625

Research Article
Local Observability of Systems on Time Scales

Zbigniew Bartosiewicz
Faculty of Computer Science, Bialystok University of Technology, Wiejska 45a, 15-351 Bialystok, Poland

Correspondence should be addressed to Zbigniew Bartosiewicz; z.bartosiewicz@pb.edu.pl

Received 9 August 2013; Revised 10 October 2013; Accepted 10 October 2013

Academic Editor: Delfim F. M. Torres

Copyright © 2013 Zbigniew Bartosiewicz. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

Analytic systems on an arbitrary time-scale are studied. As particular cases they include continuous-time and discrete-time systems.
Several local observability properties are considered. They are characterized in a unified way using the language of real analytic
geometry, ideals of germs of analytic functions, and their real radicals. It is shown that some properties related to observability are
preserved under various discretizations of continuous-time systems.

1. Introduction Ideals of germs of analytic functions and real radicals of these


ideals are used to express the criteria.
Local observability for nonlinear systems is defined in various As an application of time-scale approach to local observ-
ways [1–4]. As most of these concepts are introduced in ability we consider discretization of continuous-time systems.
the general system-theoretic setting via indistinguishability This means replacing the standard derivative by the delta
relation, they mean the same for continuous- and discrete- derivative on an appropriate discrete time-scale. We show
time systems. We show here that some of these concepts may that some of the properties related to observability are
be studied in a unified way in the framework of systems preserved under this operation. We will allow arbitrary
on time-scales. A time-scale is a model of time, which can discretizations: the discrete time-scale will not have to be
be continuous, discrete, or even mixed. Calculus on time- homogeneous. Such nonuniform discretizations behave in a
scale is a unification of ordinary differential calculus and better way in many computations.
the calculus of finite differences. Delta differential equations In Appendices we provide necessary information on
may be used to model continuous- and discrete-time systems. time-scale calculus, local analytic geometry, and real algebra.
In the discrete case, time-scale may be nonhomogeneous. It
may be applied to systems that are obtained by nonuniform
2. Preliminaries
sampling or nonuniform Euler discretization of continuous-
time systems. Let T be a time-scale. We will assume that T is forward infinite;
We concentrate on strong, weak, and robust local observ- that is, for every 𝑡0 ∈ T there are infinitely many elements of
ability of analytic systems. We show that the results obtained T that are greater than 𝑡0 . This will allow us to compute delta
in [2, 3, 5] may be extended to systems on arbitrary time- derivatives of arbitrary order at 𝑡0 . Let us consider a control
scales. This is due to the fact that all the properties are system with output
characterized with the aid of the observation algebra of the
𝑥Δ (𝑡) = 𝑓 (𝑥 (𝑡) , 𝑢 (𝑡)) ,
system, which may be introduced in a universal way on all Σ: (1)
𝑦 (𝑡) = ℎ (𝑥 (𝑡)) ,
time-scales. Since the observation algebra consists of real
functions defined on the state space, we can use the common where 𝑡 ∈ T, 𝑥(𝑡) ∈ R𝑛 , 𝑦(𝑡) ∈ R𝑟 , and 𝑢(𝑡) ∈ Ω—arbitrary
procedures to derive the criteria of local observability. As in set. For 𝜔 ∈ Ω, let 𝑓𝜔 be defined by 𝑓𝜔 (𝑥) := 𝑓(𝑥, 𝜔). We
[2, 3] we use the language of local analytic geometry and real assume that the maps ℎ and 𝑓𝜔 for every 𝜔 ∈ Ω are analytic
algebra to characterize weak and robust local observability. and that controls 𝑢 are piecewise constant functions of time.
2 Abstract and Applied Analysis

If T = R, then (1) is the standard continuous-time system Example 3. Let 𝜑 = 𝑥𝑖 be the 𝑖th coordinate function on R𝑛 .
Then 𝜑󸀠 = 𝑒𝑖 —the (row) vector of the standard basis of R𝑛
𝑥̇(𝑡) = 𝑓 (𝑥 (𝑡) , 𝑢 (𝑡)) , with 1 at the 𝑖-th position. For any 𝑡0 ∈ T we have
(2)
𝑦 (𝑡) = ℎ (𝑥 (𝑡)) . (Γ𝑔𝑡0 𝑥𝑖 ) (𝑥) = 𝑒𝑖 𝑔 (𝑥) = 𝑔𝑖 (𝑥) . (8)
For T = Z (1) takes the form Observe that if 𝜇(𝑡0 ) > 0 then
𝑥 (𝑡 + 1) − 𝑥 (𝑡) = 𝑓 (𝑥 (𝑡) , 𝑢 (𝑡)) , 1
(3) (Γ𝑔𝑡0 𝜑) (𝑥) = (𝜑 (𝑥 + 𝜇 (𝑡0 ) 𝑔 (𝑥)) − 𝜑 (𝑥)) . (9)
𝑦 (𝑡) = ℎ (𝑥 (𝑡)) . 𝜇 (𝑡0 )

This can be rewritten in a more standard shift form as On the other hand for 𝜇(𝑡0 ) = 0 we obtain (Γ𝑔𝑡0 𝜑)(𝑥) =
𝜑󸀠 (𝑥)𝑔(𝑥), which is the standard Lie derivative of the function
𝑥 (𝑡 + 1) = 𝑓 (𝑥 (𝑡) , 𝑢 (𝑡)) + 𝑥 (𝑡) =: 𝑔 (𝑥 (𝑡) , 𝑢 (𝑡)) , 𝜑 along the vector field 𝑔. In general, when operator Γ𝑔𝑡0 does
(4)
𝑦 (𝑡) = ℎ (𝑥 (𝑡)) . not depend on 𝑡0 , we will denote it by Γ𝑔 . This happens, for
instance, if the time-scale is homogeneous.
As there is a simple passage from 𝑓 to 𝑔 and vice versa, all Let 𝜎𝑔𝑡0 : 𝐶𝜔 (R𝑛 ) → 𝐶𝜔 (R𝑛 ) be another map related to
statements for (3) may be translated to statements for (4). the function 𝑔 and the time 𝑡0 defined by

Remark 1. The equation 𝑥(𝑡 + 1) = 𝑔(𝑥(𝑡), 𝑢(𝑡)) may be (𝜎𝑔𝑡0 𝜑) (𝑥) := 𝜑 (𝑥 + 𝜇 (𝑡0 ) 𝑔 (𝑥)) . (10)
studied on an arbitrary set 𝑋 or on an analytic manifold
𝑀, if analyticity of the system is essential. But then we If 𝜇(𝑡0 ) = 0, then 𝜎𝑔𝑡0 is the identity map. In general we have
cannot pass to form (3), as to do this we need a linear space an obvious property
structure. Thus, one can argue that (4) is more general than
(3). However, we concentrate here on local analytic problems, Proposition 4. The map 𝜎𝑔𝑡0 is an endomorphism of the
for which R𝑛 is general enough. algebra 𝐶𝜔 (R𝑛 ).
By 𝛾(𝑡, 𝑡0 , 𝑥0 , 𝑢) we denote the solution of the equation
𝑥Δ = 𝑓(𝑥, 𝑢) corresponding to control 𝑢 and the initial For 𝜇(𝑡0 ) > 0 the operators Γ𝑔𝑡0 and 𝜎𝑔𝑡0 are related by the
condition 𝑥(𝑡0 ) = 𝑥0 and evaluated at time 𝑡. following equality:
Let 𝑥1 , 𝑥2 ∈ R𝑛 and 𝑡0 ∈ T. Then 𝑥1 and 𝑥2 are called
indistinguishable at time 𝑡0 if 𝜎𝑔𝑡0 𝜑 = 𝜑 + 𝜇 (𝑡0 ) Γ𝑔𝑡0 𝜑. (11)

ℎ (𝛾 (𝑡, 𝑡0 , 𝑥1 , 𝑢)) = ℎ (𝛾 (𝑡, 𝑡0 , 𝑥2 , 𝑢)) (5) We also have the following generalization of the Leibniz rule.

for every control 𝑢 defined on [𝑡0 , 𝑡1 )T for some 𝑡1 ∈ T, and Proposition 5.


every 𝑡 ∈ [𝑡0 , 𝑡1 ]T for which both sides of the equation are
defined. Otherwise 𝑥1 and 𝑥2 are distinguishable at time 𝑡0 . Γ𝑔𝑡0 (𝜑𝜓) = (Γ𝑔𝑡0 𝜑) (𝜎𝑔𝑡0 𝜓) + 𝜑 (Γ𝑔𝑡0 𝜓) . (12)
The states 𝑥1 and 𝑥2 are called indistinguishable if they are
indistinguishable at time 𝑡0 for every 𝑡0 ∈ T. Otherwise 𝑥1 and This property means that Γ𝑔𝑡0 is a skew derivation of the
𝑥2 are distinguishable. Thus, 𝑥1 and 𝑥2 are distinguishable if algebra 𝐶𝜔 (R𝑛 ) with respect to 𝜎𝑔𝑡0 or, in other words, that
they are distinguishable at some time 𝑡0 . Γ𝑔𝑡0 is a 𝜎𝑔𝑡0 -derivation of this algebra. For 𝜇(𝑡0 ) = 0 it is an
ordinary derivation.
Remark 2. If the time-scale is not homogeneous, indistin-
guishability at time 𝑡0 may depend on 𝑡0 . Though the systems
we consider have “constant coefficients,” that is, the map 𝑓 3. Local Observability
does not depend on time, inhomogeneity of the time-scale
Let 𝐻𝑡0 be a subset of 𝐶𝜔 (R𝑛 ) consisting of functions of the
results in the behavior found in time-variant systems. This
form
may be observed even for linear systems (see [6]).
Let 𝐶𝜔 (R𝑛 ) denote the algebra of all real analytic func- Γ𝑔𝑡0𝑘 . . . Γ𝑔𝑡01 ℎ𝑖 , (13)
tions on R𝑛 and let 𝑔 : R𝑛 → R𝑛 be analytic. Let us fix
𝑡0 ∈ R𝑛 . In [7] the following operator where 𝑖 = 1, . . . , 𝑝, 𝑘 ≥ 0 and 𝑔𝑖 = 𝑓𝜔𝑖 for some 𝜔𝑖 ∈ Ω and
𝑖 = 1, . . . , 𝑘. If 𝑘 = 0 then this function is just ℎ𝑖 . Let
Γ𝑔𝑡0 : 𝐶𝜔 (R𝑛 ) 󳨀→ 𝐶𝜔 (R𝑛 ) (6)
𝐻 = ⋃ 𝐻𝑡0 . (14)
was introduced as 𝑡0 ∈T

1
Proposition 6. (i) The states 𝑥1 and 𝑥2 are indistinguishable
(Γ𝑔𝑡0 𝜑) (𝑥) := ∫ 𝜑󸀠 (𝑥 + 𝑠𝜇 (𝑡0 ) 𝑔 (𝑥)) 𝑑𝑠 ⋅ 𝑔 (𝑥) , (7)
at time 𝑡0 if and only if for every 𝜑 ∈ 𝐻𝑡0 , 𝜑(𝑥1 ) = 𝜑(𝑥2 ).
0
(ii) The states 𝑥1 and 𝑥2 are indistinguishable if and only if
where 𝜑 ∈ 𝐶𝜔 (R𝑛 ) and 𝜑󸀠 is the gradient of 𝜑 (a row vector). for every 𝜑 ∈ 𝐻, 𝜑(𝑥1 ) = 𝜑(𝑥2 ).
Abstract and Applied Analysis 3

Proof. The statement (ii) was shown in [1] for continuous- We say that Σ is robustly locally observable at 𝑥0 (RLO(𝑥0 ))
time systems (T = R). In this case (i) and (ii) mean the if there is a neighborhood 𝑈 of 𝑥0 such that Σ is weakly locally
same since indistinguishability at some 𝑡0 is equivalent to observable at 𝑥 for every 𝑥 ∈ 𝑈.
indistinguishability. For an arbitrary time-scale (ii) follows Robust local observability was introduced in [3] for
from (i). The statement (i) for an arbitrary time-scale was continuous-time systems under the name “stable local
shown in [8]. Analyticity of the control system and the observability.” It means that the weak local observability at 𝑥0
functions from 𝐻 is essential in the proof. is stable or robust with respect to small perturbations of the
initial condition 𝑥0 .
Remark 7. Proposition 6 allows us to use the same language We call Σ weakly locally observable (WLO) (strongly locally
of analytic functions on R𝑛 to study different observability observable (SLO), and robustly locally observable (RLO),
properties of analytic systems on arbitrary time-scales as long resp.), when it is weakly locally observable (strongly locally
as these properties are defined via the indistinguishability observable and robustly locally observable, resp.) at every
relation. It also implies that indistinguishability is an equiv- 𝑥 ∈ R𝑛 .
alence relation. This is not true for smooth systems (see [9]) Let 𝑑H(𝑥0 ) denote the linear space of the differentials of
or for analytic partially defined systems (see [10], where a functions from H taken at 𝑥0 . The following theorem is a
different definition was developed to preserve this property simple extension of the result from [1].
for partially defined systems).
Let H denote the subalgebra of 𝐶𝜔 (R𝑛 ) generated by 𝐻. Theorem 11. (a) If dim 𝑑H(𝑥0 ) = 𝑛, then Σ is SLO(𝑥0 ).
It will be called the observation algebra of the system Σ. The (b) (𝑓𝑜𝑟 𝑎𝑙𝑙 𝑥0 ∈ R𝑛 : dim 𝑑H(𝑥0 ) = 𝑛) ⇒ (𝑓𝑜𝑟
elements of H are obtained by substituting functions from 𝐻 𝑎𝑙𝑙 𝑥0 ∈ R𝑛 : Σ is SLO (𝑥0 )) ⇒ (∃𝑋—a real analytic set in
into polynomials of several variables with real coefficients. In R𝑛 : 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑥0 ∉ 𝑋 : dim 𝑑H(𝑥0 ) = 𝑛).
particular, all constant functions belong to H.
From Proposition 6 we get the following. Let us denote the condition dim 𝑑H(𝑥0 ) = 𝑛 by HK(𝑥0 )
(Hermann-Krener condition at 𝑥0 ). The second part of
Proposition 8. The states 𝑥1 and 𝑥2 are indistinguishable if Theorem 11 says that if we are interested in strong local
and only if for every 𝜑 ∈ H, 𝜑(𝑥1 ) = 𝜑(𝑥2 ). observability at large, that is, at each point, then condition
HK(𝑥0 ) is satisfied almost everywhere, so the gap between
We say that Σ is observable if any two distinct states are sufficient condition and necessary condition for strong local
distinguishable. observability at large is quite narrow. However, when one is
From the definition and Proposition 8 we obtain the interested in local observability at a particular point of the
following characterization. state space, the Hermann-Krener condition may be far from
being necessary (see [2]).
Proposition 9. Σ is observable if and only if for any distinct We have a nice gradation of different local observability
𝑥1 , 𝑥2 ∈ R𝑛 there is 𝜑 ∈ H such that 𝜑(𝑥1 ) ≠𝜑(𝑥2 ). concepts.
The condition stated in Proposition 9 is difficult to check. Proposition 12. HK(𝑥0 ) ⇒ Σ is SLO(𝑥0 ) ⇒ Σ is RLO(𝑥0 ) ⇒
This is one of the reasons that a weaker concept of local Σ is WLO(𝑥0 ).
observability seems to be more interesting. There are many None of the implications in Proposition 12 may be, in
different concepts of local observability and one concept has general, reversed.
often a few different names. For the first two concepts we
follow the terminology used in [4]. Example 13. (a) Let 𝑥 ∈ R, 𝑥Δ = 0, and 𝑦 = 𝑥3 . Thus, 𝐻 =
We say that Σ is weakly locally observable at 𝑥0 (WLO(𝑥0 )) {𝑥3 }. The system is observable, so it is also strongly locally
if there is a neighborhood 𝑈 of 𝑥0 such that for every 𝑥 ∈ 𝑈, observable at any 𝑥. But the Hermann-Krener condition fails
𝑥 and 𝑥0 are distinguishable. at 𝑥 = 0.
(b) Let 𝑥 ∈ R, 𝑥Δ = 0, and 𝑦 = 𝑥2 . Then 𝐻 = {𝑥2 }. The
Remark 10. Weak local observability at 𝑥0 is in fact a weak system is robustly locally observable at 0, but it is not strongly
property. It holds, for example, for the system locally observable at this point.
(c) Let 𝑥 ∈ R2 , 𝑥Δ = 0, and 𝑦 = 𝑥12 + 𝑥22 . The system
𝑥Δ = 0, 𝑦 = 𝑥12 + ⋅ ⋅ ⋅ + 𝑥𝑛2 , (15) is weakly locally observable at 0, but it is not robustly locally
observable at this point.
But we have an important, though obvious, global equiv-
at 𝑥0 = 0 ∈ R𝑛 . Since all solutions of 𝑥Δ = 0 are constant,
alence.
time does not influence indistinguishability relation. To
distinguish points we have to use only the output function. Proposition 14. Σ is RLO(𝑥0 ) for every 𝑥0 ∈ R𝑛 if and only if
Clearly, it takes different values at 0 and any other point, so Σ is WLO(𝑥0 ) for every 𝑥0 ∈ R𝑛 . In other words, Σ is RLO if
we can distinguish 𝑥0 from any of its neighbors. Observe that and only if Σ is WLO.
local observability fails at any 𝑥0 ≠0, if 𝑛 ≥ 2.
We say that Σ is strongly locally observable at 𝑥0 (SLO(𝑥0 )) By O𝑥 we will denote the algebra of germs at 𝑥 of analytic
if there is a neighborhood 𝑈 of 𝑥0 such that for every distinct functions on R𝑛 (see Appendix B). Let 𝐽𝑥 be the ideal of O𝑥
𝑥1 , 𝑥2 ∈ 𝑈, 𝑥1 and 𝑥2 are distinguishable. generated by germs at 𝑥 of functions from H that vanish at 𝑥.
4 Abstract and Applied Analysis

For 𝑥 ∈ R𝑛 and for an ideal 𝐽 of O𝑥 , let 𝑍(𝐽) be the germ Now, for a point 𝑥 ∈ R𝑛 , we define a sequence of ideals
at 𝑥 of the zero-set of 𝐽. Since 𝐽 is finitely generated (O𝑥 is in O𝑥 related to the system Σ. Let 𝐼𝑥(0) := (0) and 𝐼𝑥(𝑘+1) :=
Noetherian), 𝑍(𝐽) is well defined. Let 𝐼(𝑍(𝐽)) be the ideal of
√ 𝐷(H𝑥 ∪ 𝐼𝑥(𝑘) ). It is clear that instead of H𝑥 in this definition
R
O𝑥 consisting of all germs of analytic functions that vanish on
𝑍(𝐽). one can take the previously defined ideal 𝐽𝑥 ; that is, 𝐼𝑥(𝑘+1) =
√ 𝐷(𝐽𝑥 ∪ 𝐼𝑥(𝑘) ). This leads to the following generalization of
R

Lemma 15. The following conditions are equivalent:


statement (b) of Theorem 16.
(a) 𝑍(𝐽𝑥0 ) ≠{𝑥0 };
(b) arbitrarily close to 𝑥0 there is 𝑥 such that 𝜑(𝑥0 ) = 𝜑(𝑥) Corollary 17. HK(𝑥0 ) ⇔ 𝐽𝑥0 = 𝑚𝑥0 ⇔ 𝐼𝑥(1)
0
= O𝑥0 .
for every 𝜑 ∈ H.
The ideals 𝐼𝑥(𝑘) will be the main tools in studying robust
Proof. (a) holds if and only if arbitrarily close to 𝑥0 there is local observability. First we prove the basic fact.
𝑥 such that all representatives of germs 𝜑 ∈ 𝐽𝑥0 (all defined
in some neighborhood of 𝑥0 ) are 0 at 𝑥. This is equivalent to Proposition 18. For any 𝑘 ≥ 0, 𝐼𝑥(𝑘) ⊆ 𝐼𝑥(𝑘+1) , and there is 𝑠 ≥ 0
the fact that all functions 𝜑 ∈ H take on the same values at such that 𝐼𝑥(𝑠) = 𝐼𝑥(𝑠+1) .
𝑥0 and this 𝑥, which means precisely (b).
Proof. To prove the first part we proceed by induction. It
To characterize weak local observability we use the
is clear that 𝐼𝑥(0) ⊆ 𝐼𝑥(1) , so assume that 𝐼𝑥(𝑘) ⊆ 𝐼𝑥(𝑘+1) for
concept of real radical (see Appendix B).
some 𝑘 ≥ 0. Then also H𝑥 ∪ 𝐼𝑥(𝑘) ⊆ H𝑥 ∪ 𝐼𝑥(𝑘+1) and
Theorem 16. 𝐷(H𝑥 ∪ 𝐼𝑥(𝑘) ) ⊆ 𝐷(H𝑥 ∪ 𝐼𝑥(𝑘+1) ), so finally √𝐷(H𝑥 ∪ 𝐼𝑥(𝑘) ) ⊆
R

R 𝐽
(a) Σ is WLO(𝑥0 ) if and only if √ 𝑥0 = 𝑚𝑥0 . √ 𝐷(H𝑥 ∪ 𝐼𝑥(𝑘+1) ). This means that 𝐼𝑥(𝑘+1) ⊆ 𝐼𝑥(𝑘+2) . Since the
R

(b) HK(𝑥0 ) if and only if 𝐽𝑥0 = 𝑚𝑥0 . ring O𝑥 is Noetherian the sequence of ideals 𝐼𝑥(𝑘) must stabilize
at some 𝑠.
Proof. (a) Σ is not weakly locally observable at 𝑥0 if and
only if arbitrarily close to 𝑥0 there is 𝑥 such that 𝑥0 and Now we can characterize robust local observability.
𝑥 are indistinguishable. By Lemma 15, the last statement
is equivalent to the condition 𝑍(𝐽𝑥0 ) ≠{𝑥0 }, which in turn Theorem 19. System Σ is RLO(𝑥0 ) if and only if 𝐼𝑥(𝑠)0 = O𝑥0 for
means that 𝐼(𝑍(𝐽𝑥0 )) ≠𝐼({𝑥0 }). But 𝐼({𝑥0 }) = 𝑚𝑥0 , so from some 𝑠 > 0.
Theorem B.1 the last inequality is equivalent to the condition
R 𝐽 The proof of Theorem 19 will rely on several lemmas. They
√ 𝑥0 ≠𝑚𝑥0 . This gives the equivalence of both sides in (a).
appeared in a similar form in [3]. However, there were a few
(b) It is enough to prove the proposition for 𝑥 = 0 in R𝑛 . flaws in the proofs, which are now corrected.
Let G be a family of analytic functions on some open set
Observe that 𝑑H(𝑥) = 𝑑𝐽𝑥 (𝑥) for every 𝑥. Thus, if 𝑈 ⊂ R𝑛 . Denote by 𝑆𝑥 (G) the germ at 𝑥 of the level set of G
𝐽0 = 𝑚0 , then 𝑑H(0) contains all the differentials 𝑑𝑥𝑖 for that passes through 𝑥. Thus
𝑖 = 1, . . . , 𝑛, which are linearly independent.
On the other hand, the condition dim 𝑑H(0) = 𝑛 implies 𝑆𝑥 (G) = {𝑦 ∈ 𝑈 : 𝜑 (𝑥) = 𝜑 (𝑦) for 𝜑 ∈ G}𝑥 . (17)
that in a neighborhood 𝑈 of 0, there are functions 𝜑1 , . . . , 𝜑𝑛
The set-germ 𝑆𝑥 (G) is a germ of analytic set. One of the
whose germs belong to 𝐽0 and the differentials at 0, 𝑑𝜑𝑖 (0)
representatives of 𝑆𝑥 (G) is the analytic set in 𝑈 : {𝑦 ∈ 𝑈 :
are linearly independent. We may write 𝜑𝑖 = ∑𝑗 Φ𝑖𝑗 𝑥𝑗 for
𝜑(𝑦) = 𝜑(𝑥) for 𝜑 ∈ G}.
some analytic functions Φ𝑖𝑗 on 𝑈 (sufficiently small). Then
𝑑𝜑𝑖 (0) = ∑𝑗 Φ𝑖𝑗 (0)𝑑𝑥𝑗 . This means that the matrix Φ = Lemma 20. Let 𝑈 be an open subset of R𝑛 and let 𝑋 be an
(Φ𝑖𝑗 ) is invertible at 0 and then in some neighborhood of 0. analytic set in 𝑈. Consider a family G of analytic functions on
Let Ψ = Φ−1 . Then the germs of elements of Ψ are in O𝑥 𝑈. If for every 𝑥 ∉ 𝑋 : 𝑆𝑥 (G) = {𝑥}, then for every 𝑥 ∈ 𝑋 :
and 𝑑𝑥𝑖 = ∑𝑗 Ψ𝑖𝑗 𝑑𝜑𝑗 which means that 𝜑1 , . . . , 𝜑𝑛 generate 𝑆𝑥 (G) ⊂ 𝑋𝑥 .
𝑚0 .
Proof. Suppose that there is 𝑥 ∈ 𝑋 such that 𝑆𝑥 (G) ⊄ 𝑋𝑥 .
Let 𝐺 be a subset of O𝑥 . By 𝐷(𝐺) we will denote the ideal of This means that for every representatives 𝑆̃𝑥 (G) and 𝑋 ̃𝑥 we
O𝑥 generated by Jacobians 𝜕(𝜑1 , . . . , 𝜑𝑛 )/𝜕(𝑥1 , . . . , 𝑥𝑛 ), where have 𝑆̃𝑥 (G) ⊄ 𝑋 ̃𝑥 . Take 𝑋̃𝑥 := 𝑋 and arbitrarily small
𝜑𝑖 ∈ 𝐺. Observe that these Jacobians are well defined on
neighborhood 𝑉 of 𝑥 in 𝑈. Let 𝑆̃𝑥 (G) be a representative of
germs of functions. If G is a family of real analytic functions
𝑆𝑥 (G) in 𝑉. Then there is 𝑦 ∈ 𝑆̃𝑥 (G) such that 𝑦 ∉ 𝑋.
on an open set 𝑈 in R𝑛 , then similarly we define the Jacobian
Take a sequence (𝑦𝑛 ) of such points converging to 𝑥. We may
ideal 𝐷(G) in O𝑈. Furthermore, there is a simple relation
assume that all these points belong to some (large enough)
between Jacobian ideals for functions and germs of functions.
representative 𝑌 of 𝑆𝑥 (G) and that 𝑌 is an analytic set. Only
If G is a family of analytic functions on 𝑈, and 𝑥0 ∈ 𝑈 then
finite number of points 𝑦𝑛 may be isolated points of 𝑌. This
we have
means that arbitrarily close to 𝑥 there is a point 𝑦𝑛 for which
𝐷(G)𝑥0 = 𝐷 (G𝑥0 ) . (16) 𝑆𝑦𝑛 (G) ≠{𝑦𝑛 }. Thus we get a contradiction.
Abstract and Applied Analysis 5

Lemma 21. Let 𝑈 be an open subset of R𝑛 and let G be a family Lemma 24. Let 𝑈 be an open subset of R𝑛 and let 𝜑1 , . . . , 𝜑𝑘
of analytic functions on 𝑈. For every 𝑥 ∈ 𝑈: if 𝑥 ∉ 𝑍(𝐷(G)), be analytic functions on 𝑈 whose gradients are linearly inde-
then 𝑆𝑥 (G) = {𝑥}. pendent at each point of 𝑈. Let 𝑌 = 𝑍(𝜑1 , . . . , 𝜑𝑘 ), 𝑥0 ∈ 𝑌, and
let G be a family of analytic functions on 𝑈.
Proof. If 𝑥 ∉ 𝑍(𝐷(G)), then there are functions 𝜑1 , . . . , 𝜑𝑛 ∈ If 𝑌 ⊂ 𝑍(𝐷(G ∪ {𝜑1 , . . . , 𝜑𝑘 })), then arbitrarily close to 𝑥0
𝐺 such that there is 𝑥 ∈ 𝑌 such that 𝑆𝑥 (G) ≠{𝑥}.
𝜕 (𝜑1 , . . . , 𝜑𝑛 )
(𝑥) ≠0. (18) Proof. Changing the coordinates we can obtain 𝜑𝑖 = 𝑥𝑖 , 𝑖 =
𝜕 (𝑥1 , . . . , 𝑥𝑛 ) 1, . . . , 𝑘. Let 𝜓1 , . . . , 𝜓𝑛−𝑘 ∈ G. Then for 𝑥 ∈ 𝑌 we have
Thus, the map 𝑥 󳨃 → (𝜑1 (𝑥), . . . , 𝜑𝑛 (𝑥)) is injective in a
neighborhood of 𝑥. This implies that 𝑆𝑥 (G) = {𝑥}. ∇𝜑1 (𝑥)
[ .. ]
[ . ]
Lemma 22. Assume that 𝑋𝑥(𝑘)0 ≠𝑋𝑥(𝑘+1) for some 𝑘 ≥ 0. Then [ ]
0 [ ∇𝜑𝑘 (𝑥) ]
there is a neighborhood 𝑈 of 𝑥0 in R𝑛 and a representative 0 = det [
[ ∇𝜓1 (𝑥) ]
]
̃(𝑘+1) of 𝑋(𝑘+1) in 𝑈 such that for every 𝑥 ∈ 𝑈:
𝑋 [ ]
𝑥0 𝑥0 [ .. ]
[ . ]
̃(𝑘+1) 󳨐⇒ 𝑆𝑥 (H) = {𝑥} .
𝑥∉𝑋 (19) ∇𝜓 (𝑥)
𝑥0 [ 𝑛−𝑘 ]
(21)
Proof. First observe that 𝑍(𝐷(H|𝑈)) is a representative of 𝐼 0
𝑍(𝐷(H𝑥0 )). We proceed by induction. Let 𝑘 = 0 and assume [ ]
= det [( 𝜕𝜓𝑖 (𝑥)) ]
that 𝑋𝑥(1)0 ≠𝑋𝑥(0)0 = R𝑛𝑥0 . Take any neighborhood 𝑈 of 𝑥0 . 𝜕𝑥𝑗 𝑖=1,...,𝑛−𝑘
̃(1) is a [ 𝑗=1,...,𝑛 ]
Since 𝑋𝑥(1)0 = 𝑍(𝐷(H𝑥0 )), then 𝑍(𝐷(H|𝑈)) =: 𝑋 𝑥0
representative of 𝑋𝑥(1)0 in 𝑈. If 𝑥 ∉ 𝑋 ̃(1) , then by Lemma 21, 𝜕𝜓𝑖 (𝑥)
𝑥0 = det ( ) .
𝑆𝑥 (H) = 𝑆𝑥 (H|𝑈) = {𝑥}. 𝜕𝑥𝑗 𝑖=1,...,𝑛−𝑘
Now assume that the statement of the lemma holds for 𝑗=𝑘+1,...,𝑛

𝑘 − 1 ≥ 0. Hence, there is a neighborhood 𝑈 of 𝑥0 and


a representative 𝑋 ̃(𝑘) = 𝑍(𝜑1 , . . . , 𝜑𝑠 ) of 𝑋(𝑘) such that if Note that 𝑌 is an analytic manifold and the last term above
𝑥0 𝑥0
is actually the Jacobian of a map defined on 𝑌. Hence, after
𝑥 ∈ 𝑈 and 𝑥 ∉ 𝑋 ̃(𝑘) , then 𝑆𝑥 (H) = {𝑥}. The functions
𝑥0 restricting to the manifold 𝑌, we get 𝐷(G|𝑌 ) = 0. From
𝜑1 , . . . , 𝜑𝑠 are representatives on 𝑈 of generators of 𝐼𝑥(𝑘) 0
. The Lemma 23, arbitrarily close to 𝑥0 there is 𝑥 ∈ 𝑌 such that
̃
set 𝑋𝑥0 (𝑘+1)
:= 𝑍(𝐷(H|𝑈 ∪ {𝜑1 , . . . , 𝜑𝑠 })) is a representative 𝑆𝑥 (G|𝑌 ) ≠{𝑥}. But 𝑆𝑥 (G|𝑌 ) ⊂ 𝑆𝑥 (G) so also 𝑆𝑥 (G) ≠{𝑥}.
(𝑘+1)
of 𝑋𝑥0 . Clearly 𝑋 ̃(𝑘+1) ⊆ 𝑋 ̃(𝑘) . Take 𝑥 ∈ 𝑈 such that
𝑥0 𝑥0 Lemma 25. Assume that 𝑋𝑥(𝑠)0 = 𝑋𝑥(𝑠+1) for some 𝑠 ≥ 0 and
𝑥 ∉ 𝑋 ̃ (𝑘+1)
. If 𝑥 ∉ ̃ , then 𝑆𝑥 (H) = {𝑥}. Assume then
𝑋 (𝑘)
(𝑠)
0
𝑥0 𝑥0 𝑋𝑥0 ≠0. Then in every neighborhood of 𝑥0 there is 𝑥 such that
that 𝑥 ∈ 𝑋 ̃(𝑘) . From Lemma 20 we get 𝑆𝑥 (H) ⊆ (𝑋 ̃(𝑘) )𝑥 .
𝑥0 𝑥0 𝑆𝑥 (H) ≠{𝑥}.
Then 𝑆𝑥 (H) = 𝑆𝑥 (H) ∩ (𝑋 ̃(𝑘) )𝑥 = 𝑆𝑥 (H|𝑈 ∪ {𝜑1 , . . . , 𝜑𝑠 }).
𝑥0
Since 𝑥 ∉ 𝑍(𝐷(H|𝑈 ∪ {𝜑1 , . . . , 𝜑𝑠 })), then, by Lemma 21, Proof. Let 𝜑1 , . . . , 𝜑𝑘 be representatives of generators of the
𝑆𝑥 (H) = 𝑆𝑥 (H|𝑈 ∪ {𝜑1 , . . . , 𝜑𝑠 }) = {𝑥}. This finishes the ideal 𝐼𝑥(𝑠)0 , defined on some common neighborhood 𝑈 of 𝑥0 .
inductive step of the proof. Then 𝑋 ̃(𝑠) := 𝑍(𝜑1 , . . . , 𝜑𝑘 ) is a representative of 𝑋(𝑠) in 𝑈.
𝑥0 𝑥0
In every neighborhood of 𝑥0 one can find a regular point of
Lemma 23. Let 𝑥0 ∈ 𝑈 ⊂ R𝑛 , where 𝑈 is open, and let G be ̃(𝑠) (see [11, 12]). Let 𝑥 be such a point and
a family of analytic functions on 𝑈. If 𝐷(G) = 0 (zero ideal), the analytic set 𝑋 𝑥0
then arbitrarily close to 𝑥0 there is 𝑥 ∈ 𝑈 such that 𝑆𝑥 (G) ≠{𝑥}. let 𝑉 be a neighborhood of 𝑥 in R𝑛 such that 𝑌 := 𝑉 ∩ 𝑋 ̃(𝑠)
𝑥0
is an analytic manifold. Then 𝑌 = 𝑍(𝜑1|𝑉 , . . . , 𝜑𝑘|𝑉 ). We
Proof. Let may assume that the gradients of 𝜑1|𝑉 , . . . , 𝜑𝑘|𝑉 are linearly
independent on 𝑌. Otherwise, after possible shrinking of
𝜕 (𝜑1 , . . . , 𝜑𝑛 )
𝑠 = max rank (𝑥) . (20) 𝑉, we can remove the functions whose gradients are linear
𝑥∈𝑈 𝜕 (𝑥1 , . . . , 𝑥𝑛 ) combinations of the gradients of other functions. Let 𝑥 ∈ 𝑌.
𝜑𝑖 ∈G
Then 𝑥 ∈ 𝑋 ̃(𝑠+1) := 𝑍(𝐷(H|𝑈 ∪ {𝜑1 , . . . , 𝜑𝑘 })) so that 𝑌 ⊂
𝑥0
Then 𝑠 < 𝑛 and arbitrarily close to 𝑥0 there is 𝑥 ∈
𝑍(𝐷(H|𝑉 ∪ {𝜑1|𝑉 , . . . , 𝜑𝑘|𝑉 })). The statement of the lemma
𝑈 and 𝜑1 , . . . , 𝜑𝑛 ∈ G such that rank(𝜕(𝜑1 , . . . , 𝜑𝑛 )/
follows now from Lemma 24.
𝜕(𝑥1 , . . . , 𝑥𝑛 )) (𝑥) = 𝑠. This rank is preserved in some
neighborhood 𝑉 of 𝑥. Thus, we may assume that gradients
of 𝜑1 , . . . , 𝜑𝑠 are linearly independent at every point of 𝑉 and Proof of Theorem 19. Sufficiency. Assume that 𝐼𝑥(𝑠)0 = O𝑥0 for
span 𝑑G(𝑥) for 𝑥 ∈ 𝑉. Then by Frobenius Theorem, 𝑉 is a
union of integral manifolds of codistribution 𝑑G. The integral some 𝑠 > 0. This means that 𝑋𝑥(𝑠)0 = 0. From Lemma 22 it
manifolds are the level sets of G and have dimension greater follows that nontrivial set-germs 𝑆𝑥 (H) (i.e., different from
than or equals to 1. This means that 𝑆𝑥 (G) ≠{𝑥}. {𝑥}) may be found only in 𝑋 ̃(𝑠) —some representative of 𝑋(𝑠) .
𝑥0 𝑥0
6 Abstract and Applied Analysis

But 𝑋𝑥(𝑠)0 = 0 so in some neighborhood of 𝑥0 the level set- In the discretized system the ordinary derivative is replaced
germs of H must be trivial. This means that Σ is robustly by the delta derivative on the discrete time-scale.
locally observable at 𝑥0 . Thus, (23) is replaced with

̃(𝑠) = 𝑋
Necessity. Assume that 𝑋 ̃(𝑠+1) for some 𝑠 ≥ 0. From 𝑥 (𝑡 + 𝜇T (𝑡)) − 𝑥 (𝑡)
𝑥0 𝑥0 = 𝑓 (𝑥 (𝑡) , 𝑢 (𝑡)) , 𝑦 (𝑡) = ℎ (𝑥 (𝑡))
Lemma 25 it follows that Σ is not robustly locally observable. 𝜇T (𝑡)
(25)

The statements of Corollary 17, Proposition 18, and for 𝑡 ∈ T.


Theorem 19 can be translated into the language of germs Let H denote the observation algebra of the system Σ and
of analytic sets. Let 𝑋𝑥(𝑘) = 𝑍(𝐼𝑥(𝑘) ). Because the ideals 𝐼𝑥(𝑘) HT the observation algebra of the system ΣT . Observe that
are real, we also get 𝐽(𝑋𝑥(𝑘) ) = 𝐼𝑥(𝑘) , so there is one-to-one each generator of H may be approximated by a correspond-
correspondence between the ideals and the set-germs. We ing generator of HT for 𝜇(𝑡) sufficiently small. This follows
have then the following. from the form of the operators Γ𝑔𝑡0 on R and T, which is used
in this procedure.
Corollary 26. Let 𝑥0 ∈ R𝑛 . A natural question is which properties related to observ-
ability are preserved under discretization.
(a) HK(𝑥0 ) if and only if 𝑋𝑥(1)0 = 0.
Proposition 28. If 𝑥1 and 𝑥2 are distinguishable by Σ, then
(b) For every 𝑘 ≥ 0, 𝑋𝑥(𝑘)0 ⊇ 𝑋𝑥(𝑘+1)
0
, and for some 𝑠 ≥ 0, there is 𝜇̂ > 0 such that 𝑥1 and 𝑥2 are distinguishable by ΣT at
𝑋𝑥(𝑠)0 = 𝑋𝑥(𝑠+1)
0
. 𝑡 ∈ T whenever 𝜇T (𝑡) < 𝜇.̂
(c) Σ is RLO(𝑥0 ) if and only if for some 𝑠 ≥ 0, 𝑋𝑥(𝑠)0 = 0. Proof. Suppose that for every 𝜇̂ > 0 there are a time-scale
T and 𝑡 ∈ T with 0 < 𝜇T (𝑡) < 𝜇̂ such that 𝑥1 and 𝑥2
Example 27. Consider the following system: are indistinguishable by ΣT at 𝑡. This means that that there
is a sequence (T𝑖 ) of time-scales and a sequence (𝑡𝑖 ) of real
𝑥1Δ = 𝑥1 𝑢, numbers such that 𝑡𝑖 ∈ T𝑖 and 𝜇T𝑖 (𝑡𝑖 ) → 0 when 𝑖 → ∞
𝑡
𝑥2Δ = 𝑥2 , and for every 𝜑 ∈ HT𝑖𝑖 , 𝜑(𝑥1 ) = 𝜑(𝑥2 ). Every function 𝜓 ∈ H
𝑡
(22) may be approximated by functions from HT𝑖𝑖 ; that is, there is
𝑥3Δ = 0, 𝑡
a sequence of functions (𝜙𝑖 ) such that 𝜑𝑖 ∈ HT𝑖𝑖 and 𝜑𝑖 → 𝜓
on some compact set containing 𝑥1 and 𝑥2 . This implies that
𝑦 = 𝑥12 + 𝑥22 + 𝑥32 ,
𝜓(𝑥1 ) = 𝜓(𝑥2 ), so 𝑥1 and 𝑥2 are indistinguishable by Σ.
and choose 𝑥0 = 0. Then for an arbitrary time-scale we In particular, distinguishability of 𝑥1 and 𝑥2 is preserved
get H = R[𝑥12 , 𝑥22 , 𝑥32 ] and 𝐼𝑥(1)
0
= 𝐽𝑥0 = (𝑥1 𝑥2 𝑥3 ). Thus for quantum discretization, where T = 𝑞N , if 𝑞 is sufficiently
𝑋𝑥(1)0 is the germ of a union of three planes intersecting at close to 1.
We will show now that the Hermann-Krener rank con-
0. In the next step we obtain 𝐼𝑥(2) = √(𝑥
R 2 2 2 2 2 2
1 𝑥2 , 𝑥2 𝑥3 , 𝑥1 𝑥3 ) =
0 dition is preserved under discretization. Let HK(𝑥0 ) denote
(𝑥1 𝑥2 , 𝑥2 𝑥3 , 𝑥1 𝑥3 ). A quick calculation shows that 𝑋𝑥(2)0 is the the Hermann-Krener condition at 𝑥0 for the system Σ and
germ of a union of three lines intersecting at 0. Finally, 𝐼𝑥(3) = HKT (𝑥0 ) for the system ΣT .
0
(4)
𝑚𝑥0 and then 𝐼𝑥0 = O𝑥0 , so Σ is robustly locally observable at Proposition 29. If 𝐻𝐾(𝑥0 ), then there is 𝜇̂ > 0 such that for
𝑥0 . ̂ then
every time-scale T if there is 𝑡 ∈ T with 0 < 𝜇T (𝑡) < 𝜇,
𝐻𝐾T (𝑥0 ).
4. Discretization
Proof. Assume that HK(𝑥0 ) holds. Then there are
As we are going to consider several time-scales, we will denote 𝜓1 , . . . , 𝜓𝑛 ∈ H such that 𝑑𝜓1 (𝑥0 ), . . . , 𝑑𝜓𝑛 (𝑥0 ) are linearly
the graininess function on the time-scale T by 𝜇T . A time- independent. Each 𝜓𝑖 may be approximated by some
scale T is called discrete, if 𝜇T (𝑡) > 0 for all 𝑡 ∈ T. 𝜑𝑖 ∈ H𝑡T for 𝑡 ∈ T and 𝜇T (𝑡) sufficiently small. Observe that
Let Σ be the continuous-time system the functions 𝜑𝑖 depend actually on the parameter 𝜇T (𝑡) and
this dependence is continuous. For 𝜇T (𝑡) sufficiently small
𝑥̇= 𝑓 (𝑥, 𝑢) , 𝑦 = ℎ (𝑥) (23) also 𝑑𝜑1 (𝑥0 ), . . . , 𝑑𝜑𝑛 (𝑥0 ) will be linearly independent. This
means that HKT (𝑥0 ) holds for such T.
and let ΣT be its discretization
The converse of Proposition 29 does not hold.
𝑥Δ = 𝑓 (𝑥, 𝑢) , 𝑦 = ℎ (𝑥) (24)
Example 30. Let Σ be
on a discrete time-scale T. Usually T is equal to 𝑐Z for some
𝑐 > 0, but nonhomogeneous time-scales are allowed as well. 𝑥1̇ = −𝑢, 𝑥2̇ = 3𝑥12 𝑢, 𝑦 = 𝑥13 + 𝑥2 . (26)
Abstract and Applied Analysis 7

The observation algebra H is generated by a single function on the scale of integers and on the quantum scales. Different
𝜓(𝑥1 , 𝑥2 ) = 𝑥13 + 𝑥2 , which means that the Hermann-Krener concepts of local observability for systems on arbitrary time-
condition does not hold at any point. The discretized system scales have been considered. We have established relations
ΣT is given by between these concepts and provided characterizations of
weak and robust local observability with the aid of certain
𝑥1 (𝑡 + 𝜇T (𝑡)) = 𝑥1 (𝑡) − 𝜇T (𝑡) 𝑢 (𝑡) , ideals of the ring of germs of analytic functions and real
radicals of those ideals. Equivalent geometric characteriza-
𝑥2 (𝑡 + 𝜇T (𝑡)) = 𝑥2 (𝑡) + 3𝜇T (𝑡) 𝑥12 (𝑡) 𝑢 (𝑡) , (27) tions have been given. Observation algebras from which the
ideals are obtained and the ideals themselves depend on the
𝑦 (𝑡) = 𝑥1 (𝑡)3 + 𝑥2 (𝑡) . time-scale on which the systems is defined, but once the
ideals are computed, the procedures and the criteria of local
The observation algebra HT contains now the functions observability are the same for all time-scales. This allows for
𝜓1 (𝑥1 , 𝑥2 ) = 𝑥13 + 𝑥2 and 𝜓2 (𝑥1 , 𝑥2 ) = 𝑥1 . The Hermann- unified treatment of observability of systems on arbitrary
Krener condition is then satisfied at all points and for all time-scales.
discrete time-scales T.
The language of time-scales allows for a natural descrip-
Remark 31. Hermann-Krener rank condition is equivalent tion of discretization of continuous-time systems: the ordi-
to the property that the ideal 𝐽𝑥0 is maximal; that is, it is nary derivative is replaced by delta derivative on a discrete
generated by the coordinate functions. One can show that this time-scale T. The paper contains preliminary results on
property is preserved when the ideal 𝐽𝑥0 is replaced with the preservation of properties related to observability under
ideals corresponding to systems ΣT if T contains 𝑡 with 𝜇T (𝑡) discretization. In particular Hermann-Krener rank condition
sufficiently small. In characterizations of weak and robust is preserved. Preservation of other properties, in particular
local observability there appear real radicals of ideals. It is weak and robust local observability, is stated as an open
not clear whether desired properties of the radicals like max- problem. To solve the problem one will have to study limit
imality (for WLO(𝑥0 )) or nonproperness (for RLO(𝑥0 ) are properties of real radicals for rings of germs of analytic
preserved under discretizations. Thus preservation of weak functions. This will be a subject of a future research.
and robust local observability under Euler discretization is
still an open problem. Appendices
We finish this discussion with a positive example. A. Calculus on Time-scales
Example 32. Let Σ be
A time-scale T is an arbitrary nonempty closed subset of the
𝑥1̇ = 𝑥2 𝑢, 𝑥2̇ = −𝑥1 𝑢, 𝑦= 𝑥12 + 𝑥22 . (28) set R of real numbers. In particular R, ℎZ for ℎ > 0 and 𝑞N :=
{𝑞𝑘 , 𝑘 ∈ N} for 𝑞 > 1 are time-scales. We assume that T is
The observation algebra of Σ is generated by 𝜑(𝑥1 , 𝑥2 ) = 𝑥12 + a topological space with the relative topology induced from
𝑥22 . R. If 𝑡0 , 𝑡1 ∈ T, then [𝑡0 , 𝑡1 ]T denotes the intersection of the
ordinary closed interval with T. Similar notation is used for
The discretization gives ΣT open, half-open or infinite intervals.
For 𝑡 ∈ T we define the forward jump operator 𝜎 : T → T
𝑥1Δ = 𝑥2 𝑢, 𝑥2Δ = −𝑥1 𝑢, 𝑦 = 𝑥12 + 𝑥22 , (29) by 𝜎(𝑡) := inf{𝑠 ∈ T : 𝑠 > 𝑡} if 𝑡 ≠ sup T and 𝜎(sup T) = sup T
when sup T is finite; the backward jump operator 𝜌 : T → T
where by 𝜌(𝑡) := sup{𝑠 ∈ T : 𝑠 < 𝑡} if 𝑡 ≠ inf T and 𝜌(inf T) = inf T
when inf T is finite; the forward graininess function 𝜇 : T →
𝑥 (𝑡 + 𝜇T (𝑡)) − 𝑥 (𝑡)
𝑥Δ (𝑡) = . (30) [0, ∞) by 𝜇(𝑡) := 𝜎(𝑡) − 𝑡; the backward graininess function
𝜇T (𝑡) ] : T → [0, ∞) by ](𝑡) := 𝑡 − 𝜌(𝑡).
The observation algebra of ΣT is also generated by 𝜑(𝑥1 , 𝑥2 ) = If 𝜎(𝑡) > 𝑡, then 𝑡 is called right-scattered, while if 𝜌(𝑡) < 𝑡,
it is called left-scattered. If 𝑡 < sup T and 𝜎(𝑡) = 𝑡, then 𝑡 is
𝑥12 + 𝑥22 .
called right-dense. If 𝑡 > inf T and 𝜌(𝑡) = 𝑡, then 𝑡 is left-
Thus, Σ and ΣT are both weakly locally observable at 𝑥 =
dense.
0. They are not weakly locally observable at any other point.
Example 27 describes a positive behavior of robust local The time-scale T is homogeneous, if 𝜇 and ] are constant
observability under discretization. In fact, the calculations are functions. When 𝜇 ≡ 0 and ] ≡ 0, then T = R or T is a
the same for all time-scales. closed interval (in particular a half-line). When 𝜇 is constant
and greater than 0, then T = 𝜇Z + 𝑐, for some 𝑐 ∈ R.
Let T 𝜅 := {𝑡 ∈ T : 𝑡 is nonmaximal or left-dense}. Thus
5. Conclusions 𝜅
T is obtained from T by removing its maximal point if this
We have shown that the methods of real analytic geometry point exists and is left-scattered.
and real algebra developed for continuous time systems may Let 𝑓 : T → R and 𝑡 ∈ T 𝜅 . The delta derivative of 𝑓 at
be used for systems on arbitrary time-scales, in particular 𝑡, denoted by 𝑓Δ (𝑡), is the real number with the property that
8 Abstract and Applied Analysis

given any 𝜀 there is a neighborhood 𝑈 = (𝑡 − 𝛿, 𝑡 + 𝛿)T such analytic functions that vanish on 𝐴. If 𝐼 is an ideal of O𝑥 , then
that 𝑍(𝐼) will denote the zero set-germ of 𝐼 (at 𝑥). Let us recall
󵄨󵄨 󵄨 that 𝑍(𝐼) is defined as the intersection of the set-germs 𝑍(𝜑𝑖 ),
󵄨󵄨(𝑓 (𝜎 (𝑡)) − 𝑓 (𝑠)) − 𝑓Δ (𝑡) (𝜎 (𝑡) − 𝑠)󵄨󵄨󵄨 ≤ 𝜀 |𝜎 (𝑡) − 𝑠| (A.1) 𝑖 = 1, . . . , 𝑘, where 𝜑1 , . . . , 𝜑𝑘 are generators of the ideal 𝐼.
󵄨 󵄨
Since only finite intersections of set-germs are defined, we
for all 𝑠 ∈ 𝑈. If 𝑓Δ (𝑡) exists, then we say that 𝑓 is must use here the property that O𝑥 is Noetherian.
delta differentiable at 𝑡. Moreover, we say that 𝑓 is delta We have a natural duality between ideals and set-germs.
differentiable on T 𝑘 provided 𝑓Δ (𝑡) exists for all 𝑡 ∈ T 𝑘 . If 𝐼1 ⊂ 𝐼2 , then 𝑍(𝐼2 ) ⊂ 𝑍(𝐼1 ).
Let 𝑃 be any commutative ring with a unit and let 𝐼 be an
Example A.1. If T = R, then 𝑓Δ (𝑡) = 𝑓󸀠 (𝑡). If T = ℎZ, then ideal of 𝑃. Then the real radical of 𝐼, denoted by √ R
𝐼, is the
𝑓Δ (𝑡) = (𝑓(𝑡 + ℎ) − 𝑓(𝑡))/ℎ. If T = 𝑞N , then 𝑓Δ (𝑡) = (𝑓(𝑞𝑡) − set of all 𝑎 ∈ 𝑃 for which there is 𝑚 ∈ N, 𝑘 ∈ N ∪ {0} and
𝑓(𝑡))/((𝑞 − 1)𝑡). 𝑏1 , . . . , 𝑏𝑘 ∈ 𝑃 such that
A function 𝑓 : T → R is called rd-continuous provided
it is continuous at right-dense points in T and its left-sided 𝑎2𝑚 + 𝑏12 + ⋅ ⋅ ⋅ + 𝑏𝑘2 ∈ 𝐼. (B.1)
limits exist (finite) at left-dense points in T. If 𝑓 is continuous,
then it is rd-continuous. The real radical is an ideal in 𝑃 and it contains 𝐼. If 𝐼 is a proper
A function 𝑓 : T → R is called regressive, if 1 + ideal of 𝑃, then also √ R
𝐼 is proper. An ideal 𝐼 is called real if
𝜇(𝑡)𝑓(𝑡) ≠0 for all 𝑡 ∈ T. √𝐼 = 𝐼.
R

A function 𝐹 : T → R is called an antiderivative of


𝑓 : T → R provided 𝐹Δ (𝑡) = 𝑓(𝑡) holds for all 𝑡 ∈ T 𝜅 . Let Theorem B.1 (see [16]). Let 𝑥 ∈ R𝑛 . If 𝐼 is an ideal of O𝑥 , then
𝑎, 𝑏 ∈ T. Then the delta integral of 𝑓 on the interval [𝑎, 𝑏)T is
𝐽 (𝑍 (𝐼)) = √𝐼.
R
(B.2)
defined by
𝑏 Theorem B.1 implies that there is a 1 : 1 correspondence
∫ 𝑓 (𝜏) Δ𝜏 := ∫ 𝑓 (𝜏) Δ𝜏 := 𝐹 (𝑏) − 𝐹 (𝑎) . (A.2) between germs of analytic sets at 𝑥 and real ideals of O𝑥 .
𝑎 [𝑎,𝑏)T

Riemann and Lebesgue delta integrals on time-scales Acknowledgment


have been also defined (see, e.g., [13]). It can be shown that
every rd-continuous function has an antiderivative and its This work was supported by the Bialystok University of
Riemann and Lebesgue integrals agree with the delta integral Technology Grant S/WI/2/2011.
defined above.
𝑏 𝑏 References
Example A.2. If T = R, then ∫𝑎 𝑓(𝜏)Δ𝜏 = ∫𝑎 𝑓(𝜏)𝑑𝜏, where
the integral on the right is the usual Riemann integral. If T = [1] R. Hermann and A. J. Krener, “Nonlinear controllability and
𝑏 observability,” IEEE Transactions on Automatic Control, vol. 22,
ℎZ, ℎ > 0, then ∫𝑎 𝑓(𝜏)Δ𝜏 = ∑(𝑏/ℎ)−1
𝑡=(𝑎/ℎ) 𝑓(𝑡ℎ)ℎ for 𝑎 < 𝑏. no. 5, pp. 728–740, 1977.
[2] Z. Bartosiewicz, “Local observability of nonlinear systems,”
B. Basic Real Geometry Systems & Control Letters, vol. 25, no. 4, pp. 295–298, 1995.
[3] Z. Bartosiewicz, “Real analytic geometry and local observabil-
We assume that the reader is familiar with the concepts of ity,” in Differential Geometry and Control (Boulder, CO, 1997),
germs of functions and sets, and with fundamentals of the G. Ferreyra, R. Gardner, H. Hermes, and H. Sussmann, Eds.,
sheaf theory and theory of analytic sets. Necessary definitions vol. 64 of Proc. Sympos. Pure Math., pp. 65–72, American
can be found, for example, in [14, 15]. If 𝐵 is a “global” object (a Mathematical Society, Providence, RI, USA, 1999.
set, a function, or a family of functions), 𝐵𝑥 will always denote [4] F. Albertini and D. D’Alessandro, “Observability and forward-
its germ at the point 𝑥 (but the precise meaning of the germ backward observability of discrete-time nonlinear systems,”
will depend on the meaning of the object). If 𝐶 is a germ, 𝐶 ̃ Mathematics of Control, Signals, and Systems, vol. 15, no. 4, pp.
275–290, 2002.
will denote one of its representatives. By O𝑥 we denote the
algebra of germs of real analytic functions at 𝑥, where 𝑥 ∈ R𝑛 [5] D. Mozyrska and Z. Bartosiewicz, “Algebraic criteria for stable
local ob- servability of analytic systems on R𝑛 ,” in Proceedings of
(𝑛 fixed throughout the paper), and by 𝑚𝑥 the (only) maximal
European Control Conference (ECC ’97), Brussels, Belgium, July
ideal of O𝑥 , consisting of all germs in O𝑥 that vanish at 𝑥. By 1997.
O we denote the sheaf of germs of real analytic functions on
[6] Z. Bartosiewicz, “Observability of linear positive systems on
R𝑛 . time scales,” in Proceedings of the 51st IEEE Conference on
If 𝑈 is an open subset in R𝑛 , then O𝑈 will mean the algebra Decision and Control, pp. 2581–2586, Maui, Hawaii, December
of real analytic functions on 𝑈. If 𝐴 is a subalgebra of O𝑈 and 2012.
𝑥0 ∈ 𝑈, then 𝐴 𝑥0 means the set of germs at 𝑥0 of functions [7] Z. Bartosiewicz and E. Pawluszewicz, “Realizations of nonlinear
from 𝐴. Of course 𝐴 𝑥0 is again an algebra over R. If 𝐼 is an control systems on time scales,” IEEE Transactions on Automatic
ideal of O𝑈, then 𝐼𝑥0 means the ideal of O𝑥0 generated by the Control, vol. 53, no. 2, pp. 571–575, 2008.
germs at 𝑥0 of function from 𝐼. [8] E. Pawłuszewicz, “Observability of nonlinear control systems
Consider a set-germ 𝐴 in R𝑛 (at some point 𝑥). Then on time scales,” International Journal of Systems Science, vol. 43,
𝐽(𝐴) denotes the ideal of O𝑥 consisting of germs (at 𝑥) of real no. 12, pp. 2268–2274, 2012.
Abstract and Applied Analysis 9

[9] H. J. Sussmann, “Existence and uniqueness of minimal realiza-


tions of nonlinear systems,” Mathematical Systems Theory, vol.
10, no. 3, pp. 263–284, 1976/77.
[10] Z. Bartosiewicz and J. Johnson, “Systems on universe spaces,”
Acta Applicandae Mathematicae, vol. 37, no. 3, pp. 233–266,
1994.
[11] R. Narasimhan, Introduction to the Theory of Analytic Spaces,
Lecture Notes in Mathematics, no. 25, Springer, Berlin, Ger-
many, 1966.
[12] H. Cartan, “Variétés analytiques réelles et variétés analytiques
complexes,” Bulletin de la Société Mathématique de France, vol.
85, pp. 77–99, 1957.
[13] G. Sh. Guseinov, “Integration on time scales,” Journal of Math-
ematical Analysis and Applications, vol. 285, no. 1, pp. 107–127,
2003.
[14] R. C. Gunning and H. Rossi, Analytic Functions of Several
Complex Variables, Prentice-Hall, Englewood Cliffs, NJ, USA,
1965.
[15] F. Guaraldo, P. Macrı̀, and A. Tancredi, Topics on Real Analytic
Spaces, Advanced Lectures in Mathematics, Friedr. Vieweg &
Sohn, Braunschweig, Germany, 1986.
[16] J.-J. Risler, “Le théorème des zéros en géométries algébrique
et analytique réelles,” Bulletin de la Société Mathématique de
France, vol. 104, no. 2, pp. 113–127, 1976.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 162418, 7 pages
http://dx.doi.org/10.1155/2013/162418

Research Article
Positive Solutions Using Bifurcation Techniques for
Boundary Value Problems of Fractional Differential Equations

Yansheng Liu
School of Mathematical Sciences, Shandong Normal University, Jinan 250014, China

Correspondence should be addressed to Yansheng Liu; yanshliu@gmail.com

Received 3 April 2013; Accepted 30 September 2013

Academic Editor: Soheil Salahshour

Copyright © 2013 Yansheng Liu. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper is concerned with the existence of positive solutions for a class of boundary value problems of fractional differential
equations with parameter. The main tools used here are bifurcation techniques and topological degree theory. Finally, an example
is worked out to demonstrate the main result.

1. Introduction numbers 𝜆∗ and 𝜆∗∗ with 𝜆∗ < 𝜆∗∗ such that the above
system has at least two positive solutions for 𝜆 ∈ (0, 𝜆∗ ) and
During the last few decades, fractional calculus and fractional no solution for 𝜆 > 𝜆∗∗ under some suitable assumptions
differential equations have been studied extensively since such as the following.
fractional-order models are found to be more adequate
than integer-order models in some real-world problems. In (A1) There exists an interval [𝑎, 𝑏] ⊂ (0, 1) such that
fact, fractional derivatives provide an excellent tool for the lim𝑢 → +∞ 𝑓(𝑠, 𝑢)/𝑢 = +∞ uniformly with respect to
description of memory and hereditary properties of various 𝑠 ∈ [𝑎, 𝑏].
materials and processes. The mathematical modeling of sys-
tems and processes in the fields of physics, chemistry, aerody- In [9], Bai and Lü consider the following nonlinear frac-
namics, electrodynamics of complex medium, polymer rhe- tional differential equation Dirichlet-type boundary value
ology, and so forth involves derivatives of fractional order. problem:
For details and examples, see [1–7] and references therein.
Recently, there have been a few papers which deal with the 𝐷0𝛼+ 𝑢 (𝑡) + 𝑓 (𝑡, 𝑢 (𝑡)) = 0, 𝑡 ∈ (0, 1) ,
boundary value problem for fractional differential equation. (2)
For example, in [8], Tian and Liu investigated the following 𝑢 (0) = 𝑢 (1) = 0,
singular fractional boundary value problem (BVP, for short)
of the form where 1 < 𝛼 ≤ 2 is a real number and 𝐷0𝛼+ is the standard
Riemann-Liouville differentiation. The corresponding Green
𝐶
𝐷0𝛼+ 𝑢 (𝑡) + 𝜆𝑓 (𝑡, 𝑢 (𝑡)) = 0, 0 < 𝑡 < 1, function is derived. By means of some fixed point theorems
on cone, the existence and multiplicity of positive solutions
𝑢(𝑗) (0) = 0, 0 ≤ 𝑗 ≤ 𝑛 − 1, 𝑗 ≠2, (1) for BVP (2) were investigated.
In [10], Jiang and Yuan further investigated BVP (2).
𝑢󸀠󸀠 (1) = 0, Comparing with [9], they deduced some new properties of
the Green function, which extended the results of integer-
where 𝐶𝐷0𝛼+ is Caputo’s fractional derivatives, 𝑛 − 1 < 𝛼 ≤ 𝑛, order Dirichlet boundary value problems. Based on these new
𝑛 ≥ 4, and 𝑓 : (0, 1) × (0, +∞) → [0, +∞) is continuous; that properties and Krasnoselskii fixed point theorem, the exis-
is, 𝑓(𝑡, 𝑢) may be singular at 𝑡 = 0, 1 and 𝑢 = 0. By con- tence and multiplicity of positive solutions for BVP (2) were
structing a special cone, they obtained that there exist positive considered.
2 Abstract and Applied Analysis

In this paper, by using bifurcation techniques, we con- Then there exists a connected component C of T containing
sider the following boundary value problem of fractional dif- [𝑎, 𝑏] × 0 in R × 𝑉, and either
ferential equation:
(i) C is unbounded in R × 𝑉 or
𝐷0𝛼+ 𝑢 (𝑡) + 𝜂𝑓 (𝑡, 𝑢 (𝑡)) = 0, 𝑡 ∈ (0, 1) , (ii) C ∩ [(R \ [𝑎, 𝑏]) × 0] ≠0.
(3)
𝑢 (0) = 𝑢 (1) = 0,
Lemma 2 (Schmitt [16]). Let 𝑉 be a real reflexive Banach
where 1 < 𝛼 ≤ 2, 𝐷0𝛼+is the standard Riemann-Liouville space. Let 𝐺 : R × 𝑉 to 𝑉 be completely continuous, and let
differentiation, 𝑟 > 0 is a given constant, and 𝑓 : [0, 1] × 𝑎, 𝑏 ∈ R (𝑎 < 𝑏) be such that the solution of (4) is, a priori,
R+ → R+ is a given continuous function satisfying some bounded in 𝑉 for 𝜆 = 𝑎 and 𝜆 = 𝑏; that is, there exists an 𝑅 > 0
assumptions that will be specified later. such that
It is remarkable that the method used in references men-
tioned above was fixed point theorems and the same kind of 𝐺 (𝑎, 𝑢) ≠𝑢 ≠𝐺 (𝑏, 𝑢) (7)
conditions was used such that the nonlinearity 𝑓(𝑡, 𝑢) satisfies
for all 𝑢 with ‖𝑢‖ ≥ 𝑅. Furthermore, assume that
superlinear or sublinear condition at 0 and ∞, which is sim-
ilar to (A1). To the best of our knowledge, there is no paper deg (𝐼 − 𝐺 (𝑎, ⋅) , 𝐵𝑅 (0) , 0) ≠deg (𝐼 − 𝐺 (𝑏, ⋅) , 𝐵𝑅 (0) , 0) ,
studying such fractional differential equations using bifurca- (8)
tion ideas. As we know, the bifurcation technique is widely
used in solving BVP of integer-order differential equations for 𝑅 > 0 sufficiently large. Then there exists a closed connected
(see, e.g., [11–13] and references therein). In [14], by virtue set C of solutions of (4) that is unbounded in [𝑎, 𝑏] × 𝑉, and
of bifurcation ideas, the authors studied a kind of BVP of either
differential inclusions. The purpose of present paper is to
fill this gap. By using bifurcation techniques and topological (i) C is unbounded in 𝜆 direction or
degree theory, the existence of positive solutions of BVP (3) is (ii) there exists an interval [𝑐, 𝑑] such that (𝑎, 𝑏)∩(𝑐, 𝑑) = 0
investigated. The main features of present paper are as follows. and C bifurcates from infinity in [𝑐, 𝑑] × 𝑉.
First, the nonlinearity 𝑓(𝑡, 𝑢) is asymptotically linear at 0 and
∞, not super-linear or sub-linear (see the condition (H1) in Lemma 3 (Guo [17]). Let Ω be a bounded open set of real
Section 2 and example in Section 4). Next, the main method Banach space 𝐸, and let 𝐴 : Ω → 𝐸 be completely continuous.
used here is bifurcation techniques and topological degree, If there exists 𝑦0 ∈ 𝐸, 𝑦0 ≠𝜃 such that
not fixed point theorem on cone, which is different from the
references. 𝑥 ∈ 𝜕Ω, 𝜏 ≥ 0 󳨐⇒ 𝑥 − 𝐴𝑥 ≠𝜏𝑦0 , (9)
The paper is organized as described below. At the end of
this section, for completeness, we list some results on bifur- then
cation theory from interval and topological degree of com-
deg (𝐼 − 𝐴, Ω, 𝜃) = 0. (10)
pletely continuous operators. Section 2 contains background
materials and preliminaries. In Section 3, by using bifurcation
techniques and topological degree theory, bifurcation results 2. Background Materials and Preliminaries
from infinity and trivial solution are established. Then the
main results of present paper are given and proved. Finally in For convenience, we present some necessary definitions and
Section 4, an example is worked out to demonstrate the main results on fractional calculus theory (see [6]).
results.
Definition 4. The fractional (arbitrary) order integral of the
Lemma 1 (Schmitt and Thompson [15]). Let 𝑉 be a real reflex- function ℎ ∈ 𝐿1 ([𝑎, 𝑏]) of order 𝛼 ∈ R+ is defined by
ive Banach space. Let 𝐺 : R × 𝑉 to 𝑉 be completely continuous
such that 𝐺(𝜆, 0) = 0, for all 𝜆 ∈ R. Let 𝑎, 𝑏 ∈ R (𝑎 < 𝑏) be 𝛼
𝑡
(𝑡 − 𝑠)𝛼−1
𝐼𝑎+ ℎ (𝑡) = ∫ ℎ (𝑠) 𝑑𝑠, (11)
such that 𝑢 = 0 is an isolated solution of the equation 𝑎 Γ (𝛼)

𝑢 − 𝐺 (𝜆, 𝑢) = 0, 𝑢 ∈ 𝑉, (4) where Γ is the gamma function. When 𝑎 = 0, we write


𝐼𝛼 ℎ(𝑡) = [ℎ ∗ 𝜑𝛼 ](𝑡), where 𝜑𝛼 (𝑡) = 𝑡𝛼−1 /Γ(𝛼) for 𝑡 > 0, and
for 𝜆 = 𝑎 and 𝜆 = 𝑏, where (𝑎, 0), (𝑏, 0) are not bifurcation 𝜑𝛼 (𝑡) = 0 for 𝑡 ≤ 0, and 𝜑𝛼 → 𝛿(𝑡) as 𝛼 → 0, where 𝛿 is the
points of (4). Furthermore, assume that delta function.
deg (𝐼 − 𝐺 (𝑎, ⋅) , 𝐵𝑟 (0) , 0) ≠deg (𝐼 − 𝐺 (𝑏, ⋅) , 𝐵𝑟 (0) , 0) ,
(5) Definition 5. For a function ℎ given on the interval [𝑎, 𝑏],
the 𝛼th Riemann-Liouville fractional-order derivative of ℎ is
where 𝐵𝑟 (0) is an isolating neighborhood of the trivial solution. defined by
Let
𝛼 1 𝑑 𝑛 𝑡
T = {(𝜆, 𝑢) : (𝜆, 𝑢) 𝑖𝑠 𝑎 𝑠𝑜𝑙𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 (4) 𝑤𝑖𝑡ℎ 𝑢 ≠0} (𝐷𝑎+ ℎ) (𝑡) = ( ) ∫ (𝑡 − 𝑠)𝑛−𝛼−1 ℎ (𝑠) 𝑑𝑠, (12)
(6) Γ (𝑛 − 𝛼) 𝑑𝑡 𝑎

∪ ([𝑎, 𝑏] × 0) . where 𝑛 = [𝛼] + 1.


Abstract and Applied Analysis 3

Lemma 6. Let 𝛼 > 0; then, the differential equation Lemma 9 (see [10]). The function 𝐺(𝑡, 𝑠) defined by (18) has
the following properties.
𝐷0𝛼+ 𝑢 (𝑡) = 0. (13)
(i) 𝐺(𝑡, 𝑠) > 0, ∀𝑡, 𝑠 ∈ (0, 1).
has solutions 𝑢(𝑡) = 𝑐1 𝑡𝛼−1 + 𝑐2 𝑡𝛼−2 + ⋅ ⋅ ⋅ + 𝑐𝑛 𝑡𝛼−𝑛 , for some (ii) The function 𝐺∗ (𝑡, 𝑠) =: 𝑡2−𝛼 𝐺(𝑡, 𝑠) has the following
𝑐𝑖 ∈ R, 𝑖 = 0, 1, 2, . . . , 𝑛, where 𝑛 is the smallest integer greater properties:
than or equal to 𝛼.
𝛼−1
𝑡 (1 − 𝑡) 𝑠(1 − 𝑠)𝛼−1 ≤ 𝐺∗ (𝑡, 𝑠)
Notice that 𝐷0𝛼+ 𝐼𝛼 ℎ(𝑡) = ℎ(𝑡) for all ℎ ∈ 𝐶(0, 1) ∩ 𝐿(0, 1). Γ (𝛼)
From Lemma 6, we deduce the following result.
1 (19)
≤ 𝑠(1 − 𝑠)𝛼−1 ,
1
Lemma 7. Assume that 𝑢 ∈ 𝐶(0, 1)∩𝐿 [0, 1] with a derivative Γ (𝛼)
of order 𝑛 that belongs to 𝐶(0, 1) ∩ 𝐿1 [0, 1]. Then ∀𝑡, 𝑠 ∈ [0, 1] .
𝐼0𝛼+ 𝐷0𝛼+ 𝑢 (𝑡) = 𝑢 (𝑡) + 𝑐1 𝑡 𝛼−1 𝛼−2
+ 𝑐2 𝑡 𝛼−𝑛
+ ⋅ ⋅ ⋅ + 𝑐𝑛 𝑡 , (14)
The basic space used in this paper is
for some 𝑐𝑖 ∈ R, 𝑖 = 0, 1, 2, . . . , 𝑛, where 𝑛 is the smallest integer 𝐸 =: {𝑢 ∈ 𝐶 [0, 1] : 𝑢 (0) = 𝑢 (1) = 0} . (20)
greater than or equal to 𝛼.
Obviously, 𝐸 is a Banach space with norm ‖𝑢‖ = max𝑡∈𝐽 |𝑢(𝑡)|
For more detailed results of fractional calculus, we refer (for all 𝑢 ∈ 𝐸).
the reader to [6]. Let
Now let us list the following assumption satisfied
throughout the paper. 𝑄 := {V ∈ 𝐸 : V (𝑡) ≥ (𝛼 − 1) 𝑡 (1 − 𝑡) V (𝑠) ≥ 0, ∀𝑠, 𝑡 ∈ (0, 1)} .
(21)
(H1) There exist two positive numbers 𝑟, 𝑅 with 𝑟 < 𝑅 and
functions 𝑎0 , 𝑎0 , 𝑏∞ , 𝑎0 ∈ 𝐶(𝐽, R+ ) with 𝑎0 (𝑡), 𝑎0 (𝑡), It is easy to see that 𝑄 is a cone of 𝐸. Moreover, from (21), we
𝑏∞ (𝑡), 𝑎0 (𝑡) ≢ 0 in any subinterval of [0, 1] such that have, for all V ∈ 𝑄,

𝑓 (𝑡, 𝑢) ⊂ [𝑎0 (𝑡) (𝑢 − 𝜉1 (𝑡, 𝑢)) , 𝑎0 (𝑡) (𝑢 + 𝜉2 (𝑡, 𝑢))] , V (𝑡) ≥ (𝛼 − 1) 𝑡 (1 − 𝑡) ‖V‖ , ∀𝑡 ∈ [0, 1] . (22)
For the sake of using bifurcation technique to investigate
∀ (𝑡, 𝑢) ∈ 𝐽 × [0, 𝑟] ,
BVP (3), we study the following fractional boundary value
𝑓 (𝑡, 𝑢) ⊂ [𝑏∞ (𝑡) (𝑢 − 𝜁1 (𝑡, 𝑢)) , 𝑏∞ (𝑡) (𝑢 + 𝜁2 (𝑡, 𝑢))] , problem with parameter 𝜆:
𝐷0𝛼+ 𝑦 (𝑡) + 𝜆𝑓 (𝑡, 𝑦 (𝑡)) = 0, 𝑡 ∈ (0, 1) ,
∀ (𝑡, 𝑢) ∈ 𝐽 × [𝑅, +∞) , (23)
(15) 𝑦 (0) = 𝑦 (1) = 0.
A function (𝜆, 𝑢) is said to be a solution of BVP (23) if
where 𝐽 = [0, 1], 𝜉𝑖 , 𝜁𝑖 ∈ 𝐶(𝐽 × R+ ) with 𝜉𝑖 (𝑡, 𝑡𝛼−2 𝑢) = 𝑜(𝑡𝛼−2 𝑢) (𝜆, 𝑢) satisfies (23). In addition, if 𝜆 > 0, 𝑢(𝑡) > 0 for 𝑡 ∈
as 𝑢 → 0 uniformly with respect to 𝑡 ∈ [0, 1] (𝑖 = 1, 2), and (0, 1), then (𝜆, 𝑢) is said to be a positive solution of BVP (23).
𝜁𝑖 (𝑡, 𝑡𝛼−2 𝑢) = 𝑜(𝑡𝛼−2 𝑢) as 𝑢 → +∞ uniformly with respect to Obviously, if 𝜆 > 0, 𝑢 ∈ 𝑄 \ {𝜃} is a solution of BVP (23), then
𝑡 ∈ [0, 1] (𝑖 = 1, 2). by (22), we know that (𝜆, 𝑢) is a positive solution of BVP (23),
To solve BVP (3), we first consider the following linear where 𝜃 denotes the zero element of Banach space 𝐸.
boundary problem of fractional differential equation: Define
𝐷0𝛼+ 𝑢 (𝑡) + 𝑔 (𝑡) = 0, 𝑡 ∈ (0, 1) , 𝑓 (𝑡, 𝑢) , (𝑡, 𝑢) ∈ 𝐽 × R+ ,
(16) 𝑓 (𝑡, 𝑢) = { (24)
𝑢 (0) = 𝑢 (1) = 0, 𝑓 (𝑡, 0) , (𝑡, 𝑢) ∈ 𝐽 × (−∞, 0) .

where 𝑔 ∈ 𝐶[0, 1]. We cite the following two lemmas from Then 𝑓(𝑡, 𝑢) ≥ 0 on 𝐽 × R. Let
references. 1
𝐴 𝜆 V (𝑡) =: 𝜆 ∫ 𝐺∗ (𝑡, 𝑠) 𝑓 (𝑠, 𝑠𝛼−2 V (𝑠)) 𝑑𝑠, ∀V ∈ 𝑄. (25)
Lemma 8 (see [9]). Given 𝑔 ∈ 𝐶[0, 1], then 0

1 By assumption (H1) and using a similar process of the proof of


𝑢 (𝑡) = ∫ 𝐺 (𝑡, 𝑠) 𝑔 (𝑠) d𝑠 (17) Lemma 4.1 in [10], we know that 𝐴 𝜆 : 𝐶[0, 1] → 𝑄 is com-
0 pletely continuous.
is a solution of (16), where From Lemma 8, if V ∈ 𝐶[0, 1] is the the fixed point of
operator 𝐴 𝜆 , then 𝑦(𝑡) = 𝑡𝛼−2 V(𝑡) is the solution of
1 [𝑡 (1 − 𝑠)]𝛼−1 − (𝑡 − 𝑠)𝛼−1 , 0 ≤ 𝑠 ≤ 𝑡 ≤ 1,
𝐺 (𝑡, 𝑠) = { 𝐷0𝛼+ 𝑦 (𝑡) + 𝜆𝑓 (𝑡, 𝑦 (𝑡)) = 0, 𝑡 ∈ (0, 1) ,
Γ (𝛼) [𝑡 (1 − 𝑠)]𝛼−1 , 0 ≤ 𝑡 ≤ 𝑠 ≤ 1. (26)
(18) 𝑦 (0) = 𝑦 (1) = 0.
4 Abstract and Applied Analysis

Let To prove Theorems 12 and 13, we first prove the following


lemmas.
Σ =: {(𝜆, V) ∈ R+ × 𝐶 [0, 1] : V = 𝐴 𝜆 V, V ≠𝜃}, (27)
where 𝜃 is the zero element of 𝐶[0, 1]. From Lemma 9 and the Lemma 14. Let [𝑐, 𝑑] ⊂ R+ be a compact interval with
definitions of 𝑓 and the cone 𝑄, it is easy to see that Σ ⊂ 𝑄. [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] ∩ [𝑐, 𝑑] = 0. Then there exists 𝛿1 ∈ (0, 𝑟) such
Moreover, we have the following conclusion. that
V ≠𝐴 𝜆 V, ∀𝜆 ∈ [𝑐, 𝑑] , ∀V ∈ 𝐸 with 0 < ‖V‖ ≤ 𝛿1 . (32)
Lemma 10. For 𝜆 > 0, if V is a nontrivial fixed point of operator
𝐴 𝜆 , then (𝜆, 𝑡𝛼−2 V(𝑡)) is a positive solution of BVP (23). Fur- Proof. If this is false, then there exist {(𝜇𝑛 , V𝑛 )} ⊂ [𝑐, 𝑑] ×
thermore, (𝜆, 𝑦) is a positive solution of BVP (23) if and only if 𝐶[0, 1] with ‖V𝑛 ‖ → 0 + (𝑛 → +∞) such that V𝑛 = 𝐴 𝜇𝑛 V𝑛 .
(𝜆, 𝑦) is a nontrivial solution of BVP (26). Without loss of generality, assume 𝜇𝑛 → 𝜇 ∈ [𝑐, 𝑑] and
‖V𝑛 ‖ < 𝑟 for all 𝑛. Notice that V𝑛 ∈ 𝑄. By Lemma 10 and (21),
For 𝑎 ∈ 𝐶(𝐽, R+ ) with 𝑎(𝑡) ≢ 0 in any subinterval of 𝐽, we have V𝑛 (𝑡) > 0 in (0, 1). Set 𝑤𝑛 = V𝑛 /‖V𝑛 ‖. Then 𝑤𝑛 =
define the linear operator 𝐿 𝑎 : 𝐶(𝐽) → 𝐶(𝐽) by 𝐴 𝜇𝑛 V𝑛 /‖V𝑛 ‖. From the definition of 𝑓(𝑡, 𝑢), it is easy to see that
1 {𝑤𝑛 } is relatively compact in 𝐶[0, 1]. Taking a subsequence
𝐿 𝑎 𝑢 (𝑡) = ∫ 𝐺∗ (𝑡, 𝑠) 𝑎 (𝑠) 𝑠𝛼−2 𝑢 (𝑠) 𝑑𝑠, (28) and relabeling if necessary, suppose 𝑤𝑛 → 𝑤 in 𝐶[0, 1]. Then
0 ‖𝑤‖ = 1 and 𝑤 ∈ 𝑄.
where 𝐺∗ (𝑡, 𝑠) is defined by Lemma 9. On the other hand, from (H1), we know
From Lemmas 8 and 9 and the well-known Krein-Rut-
man Theorem, one can obtain the following lemma. 𝑓 (𝑡, 𝑢) ∈ [𝑎0 (𝑡) (𝑢 − 𝜉1 (𝑡, 𝑢)) , 𝑎0 (𝑡) (𝑢 + 𝜉2 (𝑡, 𝑢))] ,

Lemma 11. The operator 𝐿 𝑎 defined by (28) is completely con- ∀ (𝑡, 𝑢) ∈ 𝐽 × [0, 𝑟] .
tinuous and has a unique characteristic value 𝜆 1 (𝑎), which is (33)
positive, real, and simple and the corresponding eigenfunction Therefore, by virtue of (25), we know
𝜙(𝑡) is of one sign in (0, 1); that is, 𝜙(𝑡) = 𝜆 1 (𝑎)𝐿 𝑎 𝜙(𝑡) for all
𝑡 ∈ 𝐽. 1
𝑤𝑛 (𝑡) ≤ 𝜇𝑛 ∫ 𝐺∗ (𝑡, 𝑠) 𝑎0 (𝑠) (𝑠𝛼−2 𝑤𝑛 (𝑠)
Notice that the operator 𝐿 𝑎 can be regarded as 𝐿 𝑎 : 0

𝐿2 [0, 1] → 𝐿2 [0, 1]. This together with Lemma 11 guarantees


that 𝜆 1 (𝑎) is also the characteristic value of 𝐿∗𝑎 , where 𝐿∗𝑎 is 𝜉2 (𝑠, 𝑠𝛼−2 V𝑛 (𝑠))
+ 󵄩󵄩 󵄩󵄩 ) 𝑑𝑠,
the conjugate operator of 𝐿 𝑎 . Let 𝜑∗ denote the nonnegative 󵄩󵄩V𝑛 󵄩󵄩
eigenfunction of 𝐿∗𝑎 corresponding to 𝜆 1 (𝑎). Then we have (34)
𝜑∗ (𝑡) = 𝜆 1 (𝑎) 𝐿∗𝑎 𝜑∗ (𝑡) , ∀𝑡 ∈ 𝐽. (29) 1
𝑤𝑛 (𝑡) ≥ 𝜇𝑛 ∫ 𝐺∗ (𝑡, 𝑠) 𝑎0 (𝑠)
0
3. Main Results (35)
𝛼−2
𝜉1 (𝑠, 𝑠𝛼−2 V𝑛 (𝑠))
The main results of present paper are the following two × (𝑠 𝑤𝑛 (𝑠) − 󵄩󵄩 󵄩󵄩 ) 𝑑𝑠.
theorems. 󵄩󵄩V𝑛 󵄩󵄩

Let 𝜓∗ and 𝜓∗ be the positive eigenfunctions of 𝐿∗𝑎0 , 𝐿∗𝑎0 corre-


Theorem 12. Suppose that either
sponding to 𝜆 1 (𝑎0 ) and 𝜆 1 (𝑎0 ), respectively. Then from (34),
(i) 𝜂 ∈ [𝜆 1 (𝑎0 ), 𝜆 1 (𝑏∞ )] or it follows that
(ii) 𝜂 ∈ [𝜆 1 (𝑏∞ ), 𝜆 1 (𝑎0 )]. ⟨𝑤𝑛 , 𝜓∗ ⟩ ≤ 𝜇𝑛 ⟨𝐿 𝑎0 𝑤𝑛 , 𝜓∗ ⟩
Then BVP (3) has at least one positive solution. 1
+ 𝜇𝑛 ∫ 𝜓∗ (𝑡)
Theorem 13. Suppose the following. 0 (36)
1 𝜉2 (𝑠, 𝑠𝛼−2 V𝑛 (𝑠))
(H2) There exist 𝑅 > 0 and ℎ ∈ 𝐿[0, 1] such that ∗
× ∫ 𝐺 (𝑡, 𝑠) 𝑎0 (𝑠) 󵄩󵄩 󵄩󵄩 𝑑𝑠 𝑑𝑡.
0 󵄩󵄩V𝑛 󵄩󵄩
𝑓 (𝑡, 𝑢) ≤ ℎ (𝑡) 𝑢, for 𝑡 ∈ [0, 1] , 𝑢 ∈ [0, 𝑅] ,
Letting 𝑛 → +∞ and using condition (H1), we have
Γ (𝛼) (30)
𝜂< 1
. ⟨V, 𝜓∗ ⟩ ≤ 𝜇 ⟨𝐿 𝑎0 V, 𝜓∗ ⟩ = 𝜇 ⟨V, 𝐿∗𝑎0 𝜓∗ ⟩
∫0 [𝑠 (1 − 𝑠)]𝛼−1 ℎ (𝑠) 𝑑𝑠
𝜓∗ (37)
In addition, if = 𝜇 ⟨V, ⟩,
𝜆 1 (𝑎0 )
𝜂 > max {𝜆 1 (𝑎0 ) , 𝜆 1 (𝑏∞ )} , (31)
which implies 𝜇 ≥ 𝜆 1 (𝑎0 ). Similarly, one can deduce from
then BVP (3) has at least two positive solutions. (35) that 𝜇 ≤ 𝜆 1 (𝑎0 ).
Abstract and Applied Analysis 5

Consequently, 𝜆 1 (𝑎0 ) ≤ 𝜇 ≤ 𝜆 1 (𝑎0 ), which contradicts Let 𝜓∗ be the positive eigenfunction of 𝐿∗𝑎0 corresponding to
𝜇 ∈ [𝑐, 𝑑]. Therefore, there exists 𝛿1 ∈ (0, 𝑟) such that 𝜆 1 (𝑎0 ). Then
V ≠𝐴 𝜆 V, ∀𝜆 ∈ [𝑐, 𝑑] , ∀V ∈ 𝐸 with 0 < ‖V‖ ≤ 𝛿1 . (38)
⟨V𝑛 , 𝜓∗ ⟩ ≥ 𝜆 (1 − 𝜎) ⟨𝐿 𝑎0 V𝑛 , 𝜓∗ ⟩

= 𝜆 (1 − 𝜎) ⟨V𝑛 , 𝐿∗𝑎0 𝜓∗ ⟩
0 (47)
Lemma 15. For 𝜇 ∈ (0, 𝜆 1 (𝑎 )), there exists 𝛿1 ∈ (0, 𝑟) such
that 𝜓∗
= 𝜆 (1 − 𝜎) ⟨V𝑛 , ⟩.
deg (𝐼 − 𝐴 𝜇 , 𝐵𝛿 , 0) = 1, ∀𝛿 ∈ (0, 𝛿1 ] . (39) 𝜆 1 (𝑎0 )
This together with ⟨V𝑛 , 𝜓∗ ⟩ > 0 guarantees that
Proof. Notice that [0, 𝜇] ∩ [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] = 0. From
Lemma 14, there exists 𝛿1 ∈ (0, 𝑟) such that 𝜆 (1 − 𝜎) ≤ 𝜆 1 (𝑎0 ) , (48)
V ≠𝐴 𝜆 V, ∀𝜆 ∈ [0, 𝜇] , ∀V ∈ 𝐶 [0, 1] with 0 < ‖V‖ ≤ 𝛿1 ,
which is a contradiction. Therefore, (44) holds. By Lemma 3,
(40)
for each 𝜆 > 𝜆 1 (𝑎0 ), there exists 𝛿2 > 0 such that
which means
deg (𝐼 − 𝐴 𝜆 , 𝐵𝛿 , 0) = 0, ∀𝛿 ∈ (0, 𝛿2 ] . (49)
V ≠𝜏𝐴 𝜇 V, ∀𝜏 ∈ [0, 1] , ∀V ∈ 𝐶 [0, 1] with 0 < ‖V‖ ≤ 𝛿1 .
(41)
Therefore, by the homotopy invariance of topological Theorem 17. [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] is a bifurcation interval of
degree, we have positive solutions from the trivial solution for BVP (23); that is,
there exists an unbounded component C0 of positive solutions
deg (𝐼 − 𝐴 𝜇 , 𝐵𝛿 , 0) = deg (𝐼, 𝐵𝛿 , 0) = 1, ∀𝛿 ∈ (0, 𝛿1 ] .
of BVP (23), which meets [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] × {0}. Moreover,
(42) there exists no bifurcation interval of positive solutions from the
trivial solution which is disjointed with [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )].

Lemma 16. For 𝜆 > 𝜆 1 (𝑎0 ), there exists 𝛿2 ∈ (0, 𝑟) such that Proof. By virtue of (27) and Lemma 10, we need only to prove
that there exists an unbounded component C0 of Σ, which
deg (𝐼 − 𝐴 𝜆 , 𝐵𝛿 , 0) = 0, ∀𝛿 ∈ (0, 𝛿2 ] . (43) meets [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] × {0}, and there exists no bifurcation
Proof. First we prove that for 𝜆 > 𝜆 1 (𝑎0 ), there exists 𝛿2 ∈ interval of Σ from the trivial solution which is disjointed with
(0, 𝑟) such that [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )].
For fixed 𝑛 ∈ N with 𝜆 1 (𝑎0 ) − (1/𝑛) > 0, by Lemmas 15
V − 𝐴 𝜆 V ≠𝜏𝜑0 , ∀𝜏 ≥ 0, ∀V ∈ 𝐶 [0, 1] with 0 < ‖V‖ ≤ 𝛿2 , and 16 and their proof, there exists 𝑟 > 0 such that all of the
(44) conditions of Lemma 1 are satisfied with 𝐺(𝜆, 𝑢) = 𝐴 𝜆 𝑢,
where 𝜑0 is the positive eigenfunctions of 𝐿 𝑎0 corresponding 𝑎 = 𝜆 1 (𝑎0 ) − (1/𝑛), and 𝑏 = 𝜆 1 (𝑎0 ) + (1/𝑛). This together with
to 𝜆 1 (𝑎0 ). Lemma 10 guarantees that there exists a closed connected set
If this is false, then there exist V𝑛 ∈ 𝐶[0, 1] with ‖V𝑛 ‖ → 0 C𝑛 of Σ containing [𝜆 1 (𝑎0 ) − (1/𝑛), 𝜆 1 (𝑎0 ) + (1/𝑛)] × 0 in
(𝑛 → +∞) and 𝜏𝑛 ≥ 0 such that R+ ×𝐶[0, 1]. From Lemma 14, the case (ii) of Lemma 1 cannot
occur. Thus, C𝑛 bifurcates from [𝜆 1 (𝑎0 )−(1/𝑛), 𝜆 1 (𝑎0 )+1/𝑛]×
V𝑛 − 𝐴 𝜆 V𝑛 = 𝜏𝑛 𝜑0 . (45) 0 and is unbounded in R+ × 𝐶[0, 1]. Moreover, for any closed
By Lemma 10, we have V𝑛 (𝑡) > 0 in (0, 1). From 𝜆 > interval [𝑐, 𝑑] ⊂ [𝜆 1 (𝑎0 ) − 1/𝑛, 𝜆 1 (𝑎0 ) + 1/𝑛] \ [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )],
𝜆 1 (𝑎0 ), there exists 𝜎 > 0 satisfying 𝜆 > (1 − 𝜎)𝜆 1 (𝑎0 ). Then by Lemma 14, there exists 𝛿1 > 0 such that the set {V ∈
condition (H1) guarantees that there exists 𝛿 ∈ (0, 𝑟) such 𝐶[0, 1] : (𝜆, V) ∈ Σ, 0 < ‖V‖ ≤ 𝛿1 , 𝜆 ∈ [𝑐, 𝑑]} = 0.
Therefore, C𝑛 must be bifurcated from [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] × {0},
that |𝜉1 (𝑡, 𝑡𝛼−2 𝑢)| < 𝜎𝑡𝛼−2 𝑢 for 𝑢 ∈ (0, 𝛿). Noticing ‖V𝑛 ‖ →
which implies that C𝑛 can be regarded as C0 . In addition,
0(𝑛 → +∞), there exists 𝑁 > 0 such that ‖V𝑛 ‖ < 𝛿 for 𝑛 > 𝑁.
using Lemma 14 again, there exists no bifurcation interval of
Consequently, by virtue of (25) and (45), for 𝑛 > 𝑁, we know
positive solutions from the trivial solution which is disjointed
V𝑛 (𝑡) = 𝐴 𝜆 V𝑛 (𝑡) + 𝜏𝑛 𝜑0 with [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )].
1 By a process similar to the above, one can obtain the fol-
≥ 𝜆 ∫ 𝐺∗ (𝑡, 𝑠) 𝑎0 (𝑠) lowing conclusions.
0

× (𝑠𝛼−2 V𝑛 (𝑠) − 𝜉1 (𝑠, 𝑠𝛼−2 V𝑛 (𝑠))) 𝑑𝑠 (46) Lemma 18. Let [𝑐, 𝑑] ⊂ R+ be a compact interval with
[𝜆 1 (𝑏∞ ), 𝜆 1 (𝑏∞ )] ∩ [𝑐, 𝑑] = 0. Then there exists 𝑅1 > 𝑅 such
1
that
≥ 𝜆 (1 − 𝜎) ∫ 𝐺∗ (𝑡, 𝑠) 𝑎0 (𝑠) 𝑠𝛼−2 V𝑛 (𝑠) 𝑑𝑠
0
𝑢 ≠𝐴 𝜆 𝑢, ∀𝜆 ∈ [𝑐, 𝑑] , ∀𝑢 ∈ 𝐶 [0, 1] with ‖𝑢‖ ≥ 𝑅1 .
= 𝜆 (1 − 𝜎) 𝐿 𝑎0 V𝑛 (𝑡) . (50)
6 Abstract and Applied Analysis

Lemma 19. For 𝜇 ∈ (0, 𝜆 1 (𝑏∞ )), there exists 𝑅1 > 𝑅 such that In fact, from assumption (H2), it follows that there exists
𝜀 > 0 such that
deg (𝐼 − 𝐴 𝜇 , 𝐵𝑅 , 0) = 1, ∀𝑅 ≥ 𝑅1 . (51)
𝜂+𝜀 1
∫ [𝑠 (1 − 𝑠)]𝛼−1 ℎ (𝑠) 𝑑𝑠 < 1. (55)
Lemma 20. For 𝜆 > 𝜆 1 (𝑏∞ ), there exists 𝑅2 > 𝑅 such that Γ (𝛼) 0

deg (𝐼 − 𝐴 𝜆 , 𝐵𝑅 , 0) = 0, ∀𝑅 ≥ 𝑅2 . (52) If there is a solution (𝜆, V) of V = 𝐴 𝜆 V such that 0 ≤ 𝜆 ≤ 𝜂 + 𝜀


and ‖V‖ = 𝑅, then
Theorem 21. [𝜆 1 (𝑏∞ ), 𝜆 1 (𝑏∞ )] is a bifurcation interval of pos-
itive solutions from infinity for BVP (23), and there exists no 0 ≤ V (𝑡) ≤ ‖V‖ = 𝑅 for 𝑡 ∈ [0, 1] . (56)
bifurcation interval of positive solutions from infinity which is By virtue of (25) and Lemma 9, we have
disjoint with [𝜆 1 (𝑏∞ ), 𝜆 1 (𝑏∞ )]. More precisely, there exists an
unbounded component C∞ of solutions of BVP (23) which 1
meets [𝜆 1 (𝑏∞ ), 𝜆 1 (𝑏∞ )] × ∞ and is unbounded in 𝜆 direction. 𝑅 = ‖V‖ = max 𝜆 ∫ 𝐺∗ (𝑡, 𝑠) 𝑓 (𝑠, 𝑠𝛼−2 V (𝑠)) 𝑑𝑠
𝑡∈𝐽 0

Now we are in position to prove Theorems 12 and 13. 1


≤ (𝜂 + 𝜀) 𝑅 max ∫ 𝐺∗ (𝑡, 𝑠) 𝑠𝛼−2 ℎ (𝑠) 𝑑𝑠 (57)
𝑡∈𝐽 0
Proof of Theorem 12. Obviously the solution of the form (𝜂, 𝑦)
(𝑦 ≠𝜃) of (23) is a positive solution of BVP (3). So by (𝜂 + 𝜀) 𝑅 1
Lemma 10, it is sufficient to show that there is a component ≤ ∫ [𝑠 (1 − 𝑠)]𝛼−1 ℎ (𝑠) 𝑑𝑠 < 𝑅,
Γ (𝛼) 0
C of Σ that crosses the hyperplane {𝜂} × 𝐶[0, 1], where Σ ⊂
R+ × 𝐶[0, 1] is defined by (27). which is a contradiction. Thus, Σ ∩ ([0, 𝜂 + 𝜀] × 𝜕𝐵𝑅 ) = 0.
Next, from Theorem 17, there exist unbounded compo-
Case i (𝜆 1 (𝑎0 ) < 𝜂 < 𝜆 1 (𝑏∞ )). By Theorem 17, there exists an nents C0 of solutions of (23), which meet [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] ×
unbounded component C0 of positive solutions of BVP (23), {0}. By (54), we know C0 ∩([0, 𝜂+𝜀]×𝜕𝐵𝑅 ) = 0. This together
which bifurcates from [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] × {𝜃}. Therefore, there with the fact that C0 is unbounded, 𝜆 1 (𝑎0 ) < 𝜂, and C0 ∩
exists (𝜇𝑛 , V𝑛 ) ∈ C0 such that ({0} × 𝐶[0, 1]) = 0 guarantees that C0 crosses the hyperplane
󵄩 󵄩 {𝜂} × 𝐶[0, 1]. Then BVP (23) has a positive solution V1 with
𝜇𝑛 + 󵄩󵄩󵄩V𝑛 󵄩󵄩󵄩 󳨀→ +∞ as 𝑛 󳨀→ +∞. (53)
(𝜂, V1 ) ∈ C0 and ‖V1 ‖ < 𝑅. By Lemma 10, 𝑡𝛼−2 V1 (𝑡) is a positive
If there exists some 𝑛 ∈ N such that 𝜇𝑛 ≥ 𝜂, the conclusion solution of BVP (3).
follows. Suppose, on the contrary, 𝜇𝑛 < 𝜂 for all 𝑛 ∈ N. Since Similarly, by Theorem 21 and (54), BVP (3) has a positive
(0, 𝜃) is the only solution of (23) with 𝜆 = 0, by Lemmas 14 solution 𝑡𝛼−2 V2 (𝑡) with (𝜂, V2 ) ∈ C∞ and ‖V2 ‖ > 𝑅. The
and 18, we know C0 ∩ ({0} × 𝐶[0, 1]) = 0. Therefore, 𝜇𝑛 ∈ conclusion follows.
(0, 𝜂) for all 𝑛 ∈ N. Taking a subsequence and relabeling if
Immediately, from the proof of Theorem 13, we have the
necessary, suppose 𝜇𝑛 → 𝜇∗ as 𝑛 → +∞. Then 𝜇∗ ∈ [0, 𝜂].
following corollary.
This together with (53) guarantees that ‖V𝑛 ‖ → +∞.
Choose [𝑐, 𝑑] = [0, 𝜆 1 (𝑏∞ ) − (1/𝑚)] for 𝑚 ∈ N. From Corollary 22. Suppose that assumption (H2) holds. In addi-
Lemma 18, it follows that 𝜇∗ > 𝜆 1 (𝑏∞ ) − (1/𝑚) for each 𝑚 ∈ tion, suppose that one of the following two conditions holds:
N, which means 𝜇∗ ≥ 𝜆 1 (𝑏∞ ) > 𝜂. This is a contradiction.
(i) 𝜆 1 (𝑎0 ) < 𝜂;
Case ii (𝜆 1 (𝑏∞ ) < 𝜂 < 𝜆 1 (𝑎0 )). From Theorem 21, there exists (ii) 𝜆 1 (𝑏∞ ) < 𝜂.
an unbounded component C∞ of solutions of (23) which
bifurcates from [𝜆 1 (𝑏∞ ), 𝜆 1 (𝑏∞ )] × ∞ and is unbounded in 𝜆 Then BVP (3) has at least one positive solution.
direction.
If C∞ ∩ (R+ × {0}) = 0, using the fact that C∞ ∩ ({0} × Remark 23. Corollary 22 is different from Theorem 12
𝐶[0, 1]) = 0 and C∞ is unbounded in 𝜆 direction, we know though their results are similar.
that C∞ must cross the hyperplane {𝜂} × 𝐶[0, 1].
If C∞ ∩ (R+ × {0}) ≠0, by C∞ ∩ ({0} × 𝐶[0, 1]) = 0 and 4. An Example
Theorem 17, we know C∞ ∩ (R+ × {0}) ∈ [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] ×
{0}. Therefore, C∞ joins [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] × {0} to [𝜆 1 (𝑏∞ ), Let 𝜌 be the unique characteristic value of 𝐿 1 corresponding
𝜆 1 (𝑏∞ )] × ∞. This together with 𝜆 1 (𝑏∞ ) < 𝜂 < 𝜆 1 (𝑎0 ) guar- to positive eigenfunctions with 𝑎(𝑡) ≡ 𝑡 in (28). From
antees that C∞ crosses the hyperplane {𝜂} × 𝐶[0, 1]. Lemma 11, it follows that 𝜌 exists. Now we are ready to
give the following example.
Proof of Theorem 13. First we show that there exists 𝜀 > 0
such that Example 1. Consider the following boundary value problem
of fractional differential inclusions:
Σ ∩ ([0, 𝜂 + 𝜀] × 𝜕𝐵𝑅 ) = 0, (54)
𝐷01.5
+ 𝑢 (𝑡) + 𝜂𝑓 (𝑡, 𝑢 (𝑡)) = 0, 𝑡 ∈ (0, 1) ,
+
where 𝐵𝑅 = {V ∈ 𝐶[0, 1] : ‖V‖ < 𝑅}, Σ ⊂ R × 𝐶[0, 1] is (58)
defined by (27). 𝑢 (0) = 𝑢 (1) = 0,
Abstract and Applied Analysis 7

where fractional derivative,” Journal of Mathematical Analysis and


1 Applications, vol. 367, no. 1, pp. 260–272, 2010.
𝑓 (𝑡, 𝑢) = 𝜌𝑡𝑢 [ℎ (𝑢) + sin + 𝑡 sin (𝑡𝑢)] , [8] C. Tian and Y. Liu, “Multiple positive solutions for a class of frac-
𝑢
tional singular boundary value problems,” Georgian Academy of
3 1 Sciences, vol. 56, pp. 115–131, 2012.
{
{ , 𝑢 ∈ (0, ] ;
{
{ 2 2 (59) [9] Z. Bai and H. Lü, “Positive solutions for boundary value prob-
{
{ lem of nonlinear fractional differential equation,” Journal of
1
ℎ (𝑢) = {1 + 𝑢, 𝑢 ∈ [ , 3) ;
{
{ 2
Mathematical Analysis and Applications, vol. 311, no. 2, pp. 495–
{
{ 505, 2005.
{
{4, 𝑢 ∈ [3, +∞) . [10] D. Jiang and C. Yuan, “The positive properties of the Green
function for Dirichlet-type boundary value problems of non-
Then BVP (58) has at least one positive solution as 𝜂 ∈ linear fractional differential equations and its application,” Non-
[1/3, 2/5]. linear Analysis. Theory, Methods & Applications, vol. 72, no. 2,
pp. 710–719, 2010.
Proof. BVP (58) can be regarded as the form (3). Let 𝑓(𝑡, 𝑢) = [11] Y. Liu and D. O’Regan, “Bifurcation techniques for Lidstone
0 for 𝑢 = 0; then, 𝑓(𝑡, 𝑢) is continuous. boundary value problems,” Nonlinear Analysis. Theory, Methods
From (59), choose 𝑟 = 1/2, 𝑅 = 3, 𝑎0 (𝑡) = 𝜌𝑡/2, 𝑎0 (𝑡) = & Applications, vol. 68, no. 9, pp. 2801–2812, 2008.
(5/2)𝜌𝑡, 𝑏∞ (𝑡) = 3𝜌𝑡, 𝑏∞ (𝑡) = 5𝜌𝑡, 𝜉1 (𝑡, 𝑢) = −2𝑡𝑢 sin(𝑡𝑢), [12] R. Ma and J. Xu, “Bifurcation from interval and positive
𝜉2 (𝑡, 𝑢) = (2/5)𝑡𝑢 sin(𝑡𝑢), 𝜁1 (𝑡, 𝑢) = −(1/3)𝑢 sin(1/𝑢), solutions of a nonlinear fourth-order boundary value problem,”
𝜁2 (𝑡, 𝑢) = (1/5)𝑢 sin(1/𝑢). Nonlinear Analysis. Theory, Methods & Applications, vol. 72, no.
It is easy to see 𝜉𝑖 (𝑡, 𝑡𝛼−2 𝑢) = 𝑜(𝑡𝛼−2 𝑢) as 𝑢 → 0 and 1, pp. 113–122, 2010.
𝜁𝑖 (𝑡, 𝑡𝛼−2 𝑢) = 𝑜(𝑡𝛼−2 𝑢) as 𝑢 → +∞ both uniformly with [13] J. Xu and R. Ma, “Bifurcation from interval and positive solu-
respect to 𝑡 ∈ [0, 1], (𝑖 = 1, 2). tions for second order periodic boundary value problems,”
Applied Mathematics and Computation, vol. 216, no. 8, pp. 2463–
Therefore, (H1) is satisfied.
2471, 2010.
By the definition of 𝜌, it is easy to see 𝜆 1 (𝑎0 ) = 2/5,
[14] Y. Liu and H. Yu, “Bifurcation of positive solutions for a class of
𝜆 1 (𝑏∞ ) = 1/3.
boundary value problems of fractional differential inclusions,”
As a result, by Theorem 12, BVP (58) has at least one pos- Abstract and Applied Analysis, vol. 2013, Article ID 942831, 8
itive solution as 𝜂 ∈ [1/3, 2/5]. pages, 2013.
[15] K. Schmitt and R. C. Thompson, Nonlinear Analysis and Dif-
Acknowledgments ferential Equations: An Introduction, University of Utah Lecture
Note, Salt Lake City, Utah, USA, 2004.
The author wishes to thank the anonymous referees for their [16] K. Schmitt, “Positive solutions of semilinear elliptic boundary
valuable suggestions. Research is supported by the NNSF of value problems,” in Topological Methods in Differential Equa-
China (11171192), the Graduate Educational Innovation Foun- tions and Inclusions, vol. 472, pp. 447–500, Kluwer Academic
dation of Shandong Province (SDYY1005), and the Natural Publishers, Dordrecht, The Netherlands, 1995.
Science Foundation of Shandong Province (ZR2013AM005). [17] D. Guo, Nonlinear Functional Analysis, Shandong Science and
Technology Press, Jinan, China, 2001, (Chinese).
References
[1] B. Ahmad and J. J. Nieto, “Existence of solutions for nonlocal
boundary value problems of higher-order nonlinear fractional
differential equations,” Abstract and Applied Analysis, vol. 2009,
Article ID 494720, 9 pages, 2009.
[2] K. Diethelm, The Analysis of Fractional Differential Equations:
An Application-Oriented Exposition Using Differential Operators
of Caputo Type, vol. 2004, Springer, Berlin, Germany, 2010.
[3] A. A. Kilbas and J. J. Trujillo, “Differential equations of frac-
tional order: methods, results and problems—I,” Applicable
Analysis, vol. 78, no. 1-2, pp. 153–192, 2001.
[4] A. A. Kilbas and J. J. Trujillo, “Differential equations of frac-
tional order: methods, results and problems—II,” Applicable
Analysis, vol. 81, no. 2, pp. 435–493, 2002.
[5] I. Podlubny, Fractional Differential Equations, vol. 198 of Math-
ematics in Science and Engineering, Academic Press, San Diego,
Calif, USA, 1999.
[6] S. G. Samko, A. A. Kilbas, and O. I. Marichev, Fractional Integral
and Derivatives (Theory and Applications), Gordon and Breach,
Yverdon, Switzerland, 1993.
[7] Z. Wei, Q. Li, and J. Che, “Initial value problems for fractional
differential equations involving Riemann-Liouville sequential
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 412028, 14 pages
http://dx.doi.org/10.1155/2013/412028

Research Article
Approximation of Eigenvalues of Sturm-Liouville Problems by
Using Hermite Interpolation

M. M. Tharwat1,2 and S. M. Al-Harbi3


1
Department of Mathematics, Faculty of Science, King Abdulaziz University, Jeddah, Saudi Arabia
2
Department of Mathematics, Faculty of Science, Beni-Suef University, Beni Suef, Egypt
3
Department of Mathematics, University College, Umm Al-Qura University, P.O. Box 8140, Makkah, Saudi Arabia

Correspondence should be addressed to M. M. Tharwat; zahraa26@yahoo.com

Received 5 August 2013; Accepted 21 September 2013

Academic Editor: Dumitru Baleanu

Copyright © 2013 M. M. Tharwat and S. M. Al-Harbi. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

Eigenvalue problems with eigenparameter appearing in the boundary conditions usually have complicated characteristic
determinant where zeros cannot be explicitly computed. In this paper, we use the derivative sampling theorem “Hermite
interpolations” to compute approximate values of the eigenvalues of Sturm-Liouville problems with eigenvalue parameter in one
or two boundary conditions. We use recently derived estimates for the truncation and amplitude errors to compute error bounds.
Also, using computable error bounds, we obtain eigenvalue enclosures. Also numerical examples, which are given at the end of the
paper, give comparisons with the classical sinc method and explain that the Hermite interpolations method gives remarkably better
results.

1. Introduction where 𝑆𝑛 (𝑡) is the sequences of sinc functions as follows:


The mathematical modeling of many practical problems in
mechanics and other areas of mathematical physics requires { sin (𝜎𝑡 − 𝑛𝜋) 𝑛𝜋
solutions of boundary value problems (see, [1–7]) and frac- { (𝜎𝑡 − 𝑛𝜋) ,
{ 𝑡 ≠ ,
𝜎
𝑆𝑛 (𝑡) := { (2)
tional differential equations (see, [8–13]). It is well known that {
{1, 𝑛𝜋
many topics in mathematical physics require the investigation 𝑡= .
{ 𝜎
of the eigenvalues and eigenfunctions of Sturm-Liouville
type boundary value problems. The literature on computing
eigenvalues of various types of Sturm-Liouville problems is Series (1) converges absolutely and uniformly on R, cf.
little and we refer to [14–17]. [18–21]. Sometimes, series (1) is called the derivative sampling
Let 𝜎 > 0 and let PW2𝜎 be the Paley-Wiener space of all theorem. Our task is to use (1) to compute eigenvalues
2
𝐿 (R), entire functions of exponential type 𝜎. Assume that of Sturm-Liouville problems with eigenvalue parameter in
boundary conditions numerically. This approach is a fully
𝑓(𝑡) ∈ PW2𝜎 ⊂ PW22𝜎 . Then 𝑓(𝑡) can be reconstructed via the
new technique that uses the recently obtained estimates for
Hermite-type sampling series as
the truncation and amplitude errors associated with (1), cf.
[22]. Both types of errors normally appear in numerical tech-
niques that use interpolation procedures. In the following we

𝑛𝜋 2 summarize these estimates. The truncation error associated
𝑓 (𝑡) = ∑ [𝑓 ( ) 𝑆𝑛 (𝑡)
𝑛=−∞ 𝜎 with (1) is defined to be
(1)
𝑛𝜋 sin (𝜎𝑡 − 𝑛𝜋)
󸀠
+𝑓 ( ) 𝑆𝑛 (𝑡)] , 𝑅𝑁 (𝑓) (𝑡) := 𝑓 (𝑡) − 𝑓𝑁 (𝑡) , 𝑁 ∈ Z+ , 𝑡 ∈ R, (3)
𝜎 𝜎
2 Abstract and Applied Analysis

where 𝑓𝑁(𝑡) is the truncated series as follows: 0 < ] ≤ 1, then for 0 < 𝜀 ≤ min{𝜋/𝜎, 𝜎/𝜋, 1/√𝑒}, we have,
[22],
𝑛𝜋 2
𝑓𝑁 (𝑡) = ∑ [𝑓 ( ) 𝑆𝑛 (𝑡) 󵄩󵄩 󵄩
𝜎 󵄩󵄩A (𝜀, 𝑓)󵄩󵄩󵄩∞
|𝑛|≤𝑁
(4)
𝑛𝜋 sin (𝜎𝑡 − 𝑛𝜋) 4𝑒1/4 𝜋
+𝑓󸀠 ( ) 𝑆𝑛 (𝑡)] . ≤ {√3𝑒 (1 + 𝜎) + (( ) 𝐴 + 𝑀𝑓 ) 𝜌 (𝜀)
𝜎 𝜎 𝜎 (] + 1) 𝜎
1
It is proved in [22] that if 𝑓(𝑡) ∈ PW2𝜎 and 𝑓(𝑡) is sufficiently + (𝜎 + 2 + log (2)) 𝑀𝑓 } 𝜀 log ( ) ,
𝜀
smooth in the sense that there exists 𝑘 ∈ Z+ such that 𝑡𝑘 𝑓(𝑡) ∈ (10)
𝐿2 (R), then for 𝑡 ∈ R, |𝑡| < 𝑁𝜋/𝜎, we have
where
󵄨󵄨 󵄨
󵄨󵄨𝑅𝑁 (𝑓) (𝑡)󵄨󵄨󵄨
3𝜎 󵄨󵄨 󵄨 𝜎 ]
𝐴 := (󵄨󵄨𝑓 (0)󵄨󵄨󵄨 + 𝑀𝑓 ( ) ) ,
≤ 𝑇𝑁,𝑘,𝜎 (𝑡) 𝜋 𝜋
(11)
1
𝜉𝑘,𝜎 𝐸𝑘 |sin 𝜎𝑡|2 1 1 𝜌 (𝜀) := 𝛾 + 10 log ( ) ,
:= ( + ) 𝜀
√3(𝑁 + 1)𝑘 (𝑁𝜋 − 𝜎𝑡)3/2 (𝑁𝜋 + 𝜎𝑡)3/2
and 𝛾 := lim𝑛 → ∞ [∑𝑛𝑘=1 1/𝑘 − log 𝑛] ≅ 0.577216 is the Euler-
𝜉𝑘,𝜎 (𝜎𝐸𝑘 + 𝑘𝐸𝑘−1 ) |sin 𝜎𝑡|2 Mascheroni constant.
+
𝜎(𝑁 + 1)𝑘 The classical [23] sampling theorem of Whittaker, Kotel-
nikov, and Shannon (WKS) for 𝑓 ∈ PW2𝜎 is the series
1 1
×( + ), representation as follows:
√𝑁𝜋 − 𝜎𝑡 √𝑁𝜋 + 𝜎𝑡
(5) ∞
𝑛𝜋
𝑓 (𝑡) = ∑ 𝑓 ( ) 𝑆𝑛 (𝑡) , 𝑡 ∈ R, (12)
𝑛=−∞ 𝜎
where the constants 𝐸𝑘 and 𝜉𝑘,𝜎 are given by
where the convergence is absolute and uniform on R and

󵄨󵄨 𝑘 󵄨2 𝜎𝑘+1/2 it is uniform on compact sets of C cf. [23–25]. Series (12),
𝐸𝑘 := √ ∫ 󵄨󵄨𝑡 𝑓 (𝑡)󵄨󵄨󵄨 𝑑𝑡, 𝜉𝑘,𝜎 := . (6) which is of Lagrange interpolation type, has been used to
−∞ 𝜋𝑘+1 √1 − 4−𝑘 compute eigenvalues of second-order eigenvalue problems,
The amplitude error occurs when approximate samples are see for example, [17, 26–29]. The use of (12) in numerical
used instead of the exact ones, which we cannot compute. It analysis is known as the sinc method established by Stenger
is defined to be et al., cf. [30–32]. The aim of this paper is to investigate
the possibilities of using Hermite interpolations rather than
A (𝜀, 𝑓) (𝑡) Lagrange interpolations, to compute the eigenvalues numer-
ically. Notice that, due to Paley-Wiener’s theorem [33] 𝑓 ∈

𝑛𝜋 𝑛𝜋 PW2𝜎 if and only if there is 𝑔(⋅) ∈ 𝐿2 (−𝜎, 𝜎) such that
= ∑ [{𝑓 ( ) − 𝑓̃ ( )} 𝑆𝑛2 (𝑡)
𝑛=−∞ 𝜎 𝜎 𝜎
1
𝑓 (𝑡) = ∫ 𝑔 (𝑥) 𝑒𝑖𝑥𝑡 𝑑𝑥. (13)
𝑛𝜋 ̃󸀠 ( 𝑛𝜋 )} sin (𝜎𝑡 − 𝑛𝜋) 𝑆 (𝑡)] , √2𝜋 −𝜎
+ {𝑓󸀠 ( )−𝑓 𝑛
𝜎 𝜎 𝜎
Therefore, 𝑓󸀠 (𝑡) ∈ PW2𝜎 , that is, 𝑓󸀠 (𝑡) also has an expansion of
𝑡 ∈ R,
the form (12). However, 𝑓󸀠 (𝑡) can also be obtained by term-
(7) by-term differentiation formula of (12) as follows:
̃
where 𝑓(𝑛𝜋/𝜎) and 𝑓 ̃󸀠 (𝑛𝜋/𝜎) are approximate samples of ∞
𝑛𝜋 󸀠
𝑓(𝑛𝜋/𝜎) and 𝑓󸀠 (𝑛𝜋/𝜎), respectively. Let us assume that the 𝑓󸀠 (𝑡) = ∑ 𝑓 ( ) 𝑆𝑛 (𝑡) , (14)
𝑛=−∞ 𝜎
differences 𝜀𝑛 := 𝑓(𝑛𝜋/𝜎) − 𝑓(𝑛𝜋/𝜎), ̃ 𝜀𝑛󸀠 := 𝑓󸀠 (𝑛𝜋/𝜎) −
̃󸀠 (𝑛𝜋/𝜎), and 𝑛 ∈ Z are bounded by a positive number 𝜀,
𝑓 see [23, page 52] for convergence. Thus, the use of Hermite
that is, |𝜀𝑛 |, |𝜀𝑛󸀠 | ≤ 𝜀. If 𝑓(𝑡) ∈ PW2𝜎 satisfies the natural decay interpolations will not cost any additional computational
conditions efforts since the samples 𝑓(𝑛𝜋/𝜎) will be used to compute
both 𝑓(𝑡) and 𝑓󸀠 (𝑡) according to (12) and (14), respectively.
󵄨󵄨 󵄨󵄨 󵄨󵄨󵄨󵄨 𝑛𝜋 󵄨󵄨󵄨󵄨 󵄨󵄨 󸀠 󵄨󵄨 󵄨󵄨󵄨 󸀠 𝑛𝜋 󵄨󵄨󵄨
󵄨󵄨𝜀𝑛 󵄨󵄨 ≤ 󵄨󵄨𝑓 ( )󵄨󵄨 , Now, we consider the following differential equations:
󵄨󵄨𝜀𝑛 󵄨󵄨 ≤ 󵄨󵄨𝑓 ( )󵄨󵄨 , 󵄨 󵄨 󵄨󵄨 (8)
󵄨 𝜎 󵄨 𝜎 󵄨󵄨
𝑀𝑓 ℓ (𝑦) := −𝑦󸀠󸀠 (𝑥, 𝜇) + 𝑞 (𝑥) 𝑦 (𝑥, 𝜇) = 𝜇2 𝑦 (𝑥, 𝜇) ,
󵄨󵄨 󵄨
󵄨󵄨𝑓 (𝑡)󵄨󵄨󵄨 ≤ ]+1 , 𝑡 ∈ R − {0} , (9) (15)
|𝑡| 𝑥 ∈ [0, 1] ,
Abstract and Applied Analysis 3

with the following boundary conditions: These zeros are real and simple. The function Δ(𝜇) is an
entire function of 𝜇. We aim to approximate Δ(𝜇) and hence
𝑎1 𝑦 (0, 𝜇) + 𝑎2 𝑦󸀠 (0, 𝜇) = 𝜇2 (𝑎1󸀠 𝑦 (0, 𝜇) + 𝑎2󸀠 𝑦󸀠 (0, 𝜇)) , (16) its zeros, that is, the eigenvalues by the use of the Hermite
Interpolation. The idea is to split Δ(𝜇) into two parts, one
𝑏1 𝑦 (1, 𝜇) + 𝑏2 𝑦󸀠 (1, 𝜇) = 𝜇2 (𝑏1󸀠 𝑦 (1, 𝜇) + 𝑏2󸀠 𝑦󸀠 (1, 𝜇)) , (17) is known and the other is unknown, but lies in a Paley-
Wiener space. Then we approximate the unknown part to
where 𝜇 is a complex spectral parameter, 𝑞(⋅) is assumed to get the approximate Δ(𝜇) and then compute the approximate
be real valued and continuous on [0, 1], and 𝑎𝑖 , 𝑏𝑖 , 𝑎𝑖󸀠 , 𝑏𝑖󸀠 ∈ R, zeros. Using the method of variation of constants, the solution
𝑖 = 0, 1 satisfying 𝑦(𝑥, 𝜇) satisfies Volterra integral equation as follows:

((𝑎1󸀠 , 𝑎2󸀠 ) = (0, 0) or 𝑎1 𝑎2󸀠 − 𝑎1󸀠 𝑎2 > 0) , 𝑦 (𝑥, 𝜇) = (𝑎2 − 𝑎2󸀠 𝜇2 ) cos 𝜇𝑥
(18) (21)
((𝑏1󸀠 , 𝑏2󸀠 ) = (0, 0) or 𝑏1󸀠 𝑏2 − 𝑏1 𝑏2󸀠 > 0) . sin 𝜇𝑥
− (𝑎1 − 𝑎1󸀠 𝜇2 ) + 𝑇 [𝑦] (𝑥, 𝜇) ,
𝜇
The eigenvalue problem (15)–(17) will be denoted by
where 𝑇 is the Volterra operator defined by
Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 𝑏󸀠 ) when (𝑎1󸀠 , 𝑎2󸀠 ) ≠ (𝑏1󸀠 , 𝑏2󸀠 ). It is a Sturm-
(0, 0) ≠
Liouville problem when the eigenparameter 𝜇 appears lin- 𝑥 sin 𝜇 (𝑥 − 𝑡)
early in both boundary conditions. The classical problem 𝑇 [𝑦] (𝑥, 𝜇) = ∫ 𝑞 (𝑡) 𝑦 (𝑡, 𝜇) 𝑑𝑡. (22)
0 𝜇
when 𝑎1󸀠 = 𝑎2󸀠 = 𝑏1󸀠 = 𝑏2󸀠 = 0, which we denote by Π(𝑞, 𝑎, 𝑏, 0,
0) has a countable set of real and simple eigenvalues with ∞ Differentiating (21), we get
as the only possible limit point, [34, 35]. In [14], the authors
used Hermite-type sampling series (1) to compute the eigen- 𝑦󸀠 (𝑥, 𝜇) = (𝑎2󸀠 𝜇2 − 𝑎2 ) 𝜇 sin 𝜇𝑥
values of problem Π(𝑞, 𝑎, 𝑏, 0, 0) numerically. In [36], see (23)
̃ [𝑦] (𝑥, 𝜇) ,
+ (𝑎1󸀠 𝜇2 − 𝑎1 ) cos 𝜇𝑥 + 𝑇
also [37], Annaby and Tharwat proved that Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 𝑏󸀠 )
has a denumerable set of real and simple eigenvalues with
̃ is the Volterra operator
where 𝑇
∞ as the limit point using techniques similar of those
established in [38–40], where also sampling theorems have 𝑥
been established. Similar results are established in [38] for the ̃ [𝑦] (𝑥, 𝜇) = ∫ cos 𝜇 (𝑥 − 𝑡) 𝑞 (𝑡) 𝑦 (𝑡, 𝜇) 𝑑𝑡.
𝑇 (24)
0
problem when the eigenparameter appears in one condition,
that is, when 𝑎1󸀠 = 𝑎2󸀠 = 0, (𝑏1󸀠 , 𝑏2󸀠 ) ≠ (0, 0) or equivalently Define 𝑓(⋅, 𝜇) and 𝑔(⋅, 𝜇) to be
when (𝑎1󸀠 , 𝑎2󸀠 ) ≠ (0, 0) and 𝑏1󸀠 = 𝑏2󸀠 = 0. These problems will ̃ [𝑦] (𝑥, 𝜇) . (25)
𝑓 (𝑥, 𝜇) := 𝑇 [𝑦] (𝑥, 𝜇) , 𝑔 (𝑥, 𝜇) := 𝑇
be denoted by Π(𝑞, 𝑎, 𝑏, 0, 𝑏󸀠 ), Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 0), respectively.
The aim of the present work is to compute the eigenval- In the following, we will make use of the estimates [44] as
ues of Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 𝑏󸀠 ), Π(𝑞, 𝑎, 𝑏, 0, 𝑏󸀠 ), and Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 0) follows:
numerically by the Hermite interpolations with an error 󵄨󵄨 sin 𝑧 󵄨󵄨
analysis. This method is based on sampling theorem, Hermite |cos 𝑧| ≤ 𝑒|I𝑧| , 󵄨󵄨 󵄨󵄨 ≤ 𝑐0 𝑒|I𝑧| ,
󵄨󵄨 󵄨 (26)
interpolations, but applied to regularized functions. Hence, 󵄨 𝑧 󵄨󵄨 1 + |𝑧|
avoiding any (multiple) integration and keeping the number where 𝑐0 is some constant (we may take 𝑐0 ≃ 1.72). For
of terms in the Cardinal series manageable. It has been convenience, we define the constants by
demonstrated that the method is capable of delivering higher
1
order estimates of the eigenvalues at a very low cost, see [41– 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨
𝜏 := ∫ 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 𝑑𝑡, 𝑐1 := 󵄨󵄨󵄨𝑎2 󵄨󵄨󵄨 + 𝑐0 󵄨󵄨󵄨𝑎1 󵄨󵄨󵄨 ,
43]. In Sections 2 and 3 we derive the Hermite interpolation 0
technique to compute the eigenvalues of Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 𝑏󸀠 ) and 󵄨 󵄨 󵄨 󵄨
Π(𝑞, 𝑎, 𝑏, 0, 𝑏󸀠 ) with error estimates, respectively. The last 𝑐2 := 󵄨󵄨󵄨󵄨𝑎2󸀠 󵄨󵄨󵄨󵄨 + 𝑐0 󵄨󵄨󵄨󵄨𝑎1󸀠 󵄨󵄨󵄨󵄨 , 𝑐3 := 𝑐0 𝜏,
section involves some illustrative examples. 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨
𝑐4 := exp 𝑐3 , 𝑐5 := max {𝑐1 , 𝑐2 , 󵄨󵄨󵄨𝑏1 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑏2 󵄨󵄨󵄨 𝜏, 󵄨󵄨󵄨󵄨𝑏1󸀠 󵄨󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑏2󸀠 󵄨󵄨󵄨󵄨 𝜏} .
2. Treatment of Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 𝑏󸀠 ) (27)
From (21) and (25), we get
In this section, we derive approximate values of the eigenval-
ues of Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 𝑏󸀠 ). Let 𝑦(⋅, 𝜇) denote the solution of (15) 𝑓 (𝑥, 𝜇)
satisfying the following initial conditions:
𝑥 sin 𝜇 (𝑥 − 𝑡)
=∫ 𝑞 (𝑡) [ (𝑎2 − 𝑎2󸀠 𝜇2 ) cos 𝜇𝑡
𝑦 (0, 𝜇) = 𝑎2 − 𝑎2󸀠 𝜇2 , 󸀠
𝑦 (0, 𝜇) = 𝑎1󸀠 𝜇2 − 𝑎1 . (19) 0 𝜇
Thus, 𝑦(⋅, 𝜇) satisfies the boundary condition (16). The eigen- sin 𝜇𝑡 (28)
− (𝑎1 − 𝑎1󸀠 𝜇2 ) ] 𝑑𝑡
values of the problem Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 𝑏󸀠 ) are the zeros of the 𝜇
function as follows:
𝑥 sin 𝜇 (𝑥 − 𝑡)
+∫ 𝑞 (𝑡) 𝑓 (𝑡, 𝜇) 𝑑𝑡.
Δ (𝜇) := (𝑏1󸀠 𝜇2 − 𝑏1 ) 𝑦 (1, 𝜇) + (𝑏2󸀠 𝜇2 󸀠
− 𝑏2 ) 𝑦 (1, 𝜇) . (20) 0 𝜇
4 Abstract and Applied Analysis

Lemma 1. For 0 ≤ 𝑥 ≤ 1, 𝜇 ∈ C, the following estimates hold: Combining (31) and (32), we obtain 0 ≤ 𝑥 ≤ 1, 𝜇 ∈ C,
󵄨󵄨 󵄨
󵄨󵄨𝑓 (𝑥, 𝜇)󵄨󵄨󵄨
1
󵄨 󵄨2 𝑐 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨2
𝑐3 𝑐4 (𝑐1 + 𝑐2 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 ) |I𝜇|𝑥 ≤ 𝑒|I𝜇|𝑥 0󵄨󵄨 󵄨󵄨 ∫ 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 [󵄨󵄨󵄨𝑎2 󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑎2󸀠 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨
𝑓 (𝑥, 𝜇) ≤ 󵄨 󵄨 𝑒 , (29) 1 + 󵄨󵄨𝜇󵄨󵄨 0
1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 (33)
󵄨 󵄨 󵄨 󵄨 󵄨 󵄨2
+ (󵄨󵄨󵄨𝑎1 󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑎1󸀠 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 ) 𝑐0 𝑡] 𝑑𝑡
󵄨 󵄨2
𝜏𝑐3 𝑐4 (𝑐1 + 𝑐2 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 ) |I𝜇|𝑥 𝑥
𝑔 (𝑥, 𝜇) ≤ 󵄨 󵄨 𝑒 . (30) 󵄨 󵄨󵄨 󵄨
1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 + 𝑐0 𝑒|I𝜇|𝑥 ∫ 𝑒−|I𝜇|𝑡 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 󵄨󵄨󵄨𝑓 (𝑡, 𝜇)󵄨󵄨󵄨 𝑑𝑡.
0
Applying Gronwall’s inequality, cf. for example, [34, page 51],
yields 𝜇 ∈ C,
Proof. We divide 𝑓(⋅, 𝜇) into two parts 𝑓1 (⋅, 𝜇) and 𝑓2 (⋅, 𝜇) and
󵄨 󵄨
estimate each of them. Indeed, for 𝑥 ∈ [0, 1] and 𝜇 ∈ C we 𝑒−|I𝜇|𝑥 󵄨󵄨󵄨𝑓 (𝑥, 𝜇)󵄨󵄨󵄨
have 1
𝑐 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨2
≤ [ 0󵄨󵄨 󵄨󵄨 ∫ 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 [󵄨󵄨󵄨𝑎2 󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑎2󸀠 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨
1 + 󵄨󵄨𝜇󵄨󵄨 0
󵄨󵄨 󵄨
󵄨󵄨𝑓1 (𝑥, 𝜇)󵄨󵄨󵄨
󵄨 󵄨 󵄨 󵄨 󵄨 󵄨2
+ (󵄨󵄨󵄨𝑎1 󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑎1󸀠 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 ) 𝑐0 𝑡] 𝑑𝑡]
󵄨󵄨󵄨 𝑥 sin 𝜇 (𝑥 − 𝑡)
= 󵄨󵄨󵄨∫ 𝑞 (𝑡) [ (𝑎2 − 𝑎2󸀠 𝜇2 ) cos 𝜇𝑡
󵄨󵄨 0 𝜇 𝑥
󵄨 󵄨
󵄨󵄨 × exp (𝑐0 ∫ 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 𝑑𝑡)
sin 𝜇𝑡 󵄨 (34)
] 𝑑𝑡󵄨󵄨󵄨
0
− (𝑎1 − 𝑎1󸀠 𝜇2 )
𝜇 󵄨󵄨 1
𝑐0 󵄨 󵄨 󵄨 󵄨 󵄨󵄨 󸀠 󵄨󵄨 󵄨 󵄨2
𝑥 ≤[ 󵄨󵄨 󵄨󵄨 ∫ 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 [󵄨󵄨󵄨𝑎2 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑎2 󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨
󵄨 󵄨 𝑐0 (𝑥 − 𝑡) 1 + 󵄨󵄨𝜇󵄨󵄨 0
≤ 𝑒|I𝜇|𝑥 ∫ 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 󵄨 󵄨
0 1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 (𝑥 − 𝑡)
󵄨 󵄨 󵄨 󵄨 󵄨 󵄨2
+ (󵄨󵄨󵄨𝑎1 󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑎1󸀠 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 ) 𝑐0 𝑡] 𝑑𝑡]
󵄨 󵄨 󵄨 󵄨 󵄨 󵄨2 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨2
× [ 󵄨󵄨󵄨𝑎2 󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑎2󸀠 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 + (󵄨󵄨󵄨𝑎1 󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑎1󸀠 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 )
1
󵄨 󵄨
× exp (𝑐0 ∫ 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 𝑑𝑡) ,
𝑐0 𝑡 0
× 󵄨 󵄨 ] 𝑑𝑡
1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 𝑡 from which we get
󵄨󵄨 󵄨
𝑐0 𝑥 󵄨
𝑥
󵄨 󵄨 󵄨 󵄨󵄨 󸀠 󵄨󵄨 󵄨 󵄨2 󵄨󵄨𝑓 (𝑥, 𝜇)󵄨󵄨󵄨
≤ 𝑒|I𝜇|𝑥 󵄨󵄨 󵄨󵄨 ∫ 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 [󵄨󵄨󵄨𝑎2 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑎2 󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨
1 + 󵄨󵄨𝜇󵄨󵄨 𝑥 0
󵄨󵄨 󵄨󵄨 󵄨󵄨 󸀠 󵄨󵄨 󵄨󵄨 󵄨󵄨2 󵄨󵄨 󵄨󵄨 󵄨󵄨 󸀠 󵄨󵄨 󵄨󵄨 󵄨󵄨2
󵄨 󵄨 󵄨 󵄨 󵄨 󵄨2 |I𝜇|𝑥 [ 𝑐0 [󵄨󵄨𝑎2 󵄨󵄨 + 󵄨󵄨󵄨𝑎2 󵄨󵄨󵄨 󵄨󵄨𝜇󵄨󵄨 + (󵄨󵄨𝑎1 󵄨󵄨 + 󵄨󵄨󵄨𝑎1 󵄨󵄨󵄨 󵄨󵄨𝜇󵄨󵄨 ) 𝑐0 ]
+ (󵄨󵄨󵄨𝑎1 󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑎1󸀠 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 ) 𝑐0 𝑡] 𝑑𝑡 ≤𝑒 [ 󵄨 󵄨
1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨
1
[
𝑐0 󵄨 󵄨 󵄨 󵄨 󵄨󵄨 󸀠 󵄨󵄨 󵄨 󵄨2
≤ 𝑒|I𝜇|𝑥 󵄨󵄨 󵄨󵄨 ∫ 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 [󵄨󵄨󵄨𝑎2 󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑎2 󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 1 1
(35)
1 + 󵄨󵄨𝜇󵄨󵄨 0 󵄨 󵄨 ] 󵄨 󵄨
× ∫ 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 𝑑𝑡] exp (𝑐0 ∫ 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 𝑑𝑡)
󵄨 󵄨 󵄨 󵄨 󵄨 󵄨2 0 0
+ (󵄨󵄨󵄨𝑎1 󵄨󵄨󵄨 + 󵄨󵄨󵄨󵄨𝑎1󸀠 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 ) 𝑐0 𝑡] 𝑑𝑡. ]
(31) 󵄨󵄨 󵄨󵄨2
𝑐3 𝑐4 (𝑐1 + 𝑐2 󵄨󵄨𝜇󵄨󵄨 ) |I𝜇|𝑥
= 󵄨 󵄨 𝑒 .
1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨
Moreover, 0 ≤ 𝑥 ≤ 1, 𝜇 ∈ C, Then from (25) and (29), we obtain the estimate (30).
Now we split Δ(𝜇) into two parts via
󵄨 𝑥 󵄨󵄨 Δ (𝜇) = G (𝜇) + S (𝜇) , (36)
󵄨󵄨 󵄨 󵄨󵄨 sin 𝜇 (𝑥 − 𝑡) 󵄨
󵄨󵄨𝑓2 (𝑥, 𝜇)󵄨󵄨󵄨 = 󵄨󵄨󵄨∫ 𝑞 (𝑡) 𝑓 (𝑡, 𝜇) 𝑑𝑡󵄨󵄨󵄨
󵄨󵄨 0 𝜇 󵄨󵄨 where G(𝜇) is known part
𝑥
𝑐0 (𝑥 − 𝑡) |I𝜇|(𝑥−𝑡) G (𝜇)
≤∫ 󵄨 󵄨 𝑒
0 1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 (𝑥 − 𝑡) sin 𝜇
= (𝑏1󸀠 𝜇2 − 𝑏1 ) [(𝑎2 − 𝑎2󸀠 𝜇2 ) cos 𝜇 − (𝑎1 − 𝑎1󸀠 𝜇2 ) ]
󵄨 󵄨󵄨 󵄨 𝜇
× 󵄨󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 󵄨󵄨󵄨𝑓 (𝑡, 𝜇)󵄨󵄨󵄨 𝑑𝑡
𝑥
󵄨󵄨 󵄨󵄨 󵄨 +(𝑏2󸀠 𝜇2 − 𝑏2 )[(𝑎2󸀠 𝜇2 − 𝑎2 )𝜇 sin 𝜇+(𝑎1󸀠 𝜇2 − 𝑎1 ) cos 𝜇] ,
󵄨󵄨𝑞 (𝑡)󵄨󵄨󵄨 󵄨󵄨󵄨𝑓 (𝑡, 𝜇)󵄨󵄨󵄨 𝑑𝑡.
|I𝜇|𝑥 −|I𝜇|𝑡
≤ 𝑐0 𝑒 ∫ 𝑒 (32) (37)
0
Abstract and Applied Analysis 5

and S(𝜇) is unknown part Proof. Since S(𝜇) is entire, then also F𝜃,𝑚 (𝜇) is entire in 𝜇.
Combining the estimates | sin 𝑧/𝑧| ≤ (𝑐0 /(1 + |𝑧|))𝑒|I𝑧| and
S (𝜇) = (𝑏1󸀠 𝜇2 − 𝑏1 ) 𝑓 (1, 𝜇) + (𝑏2󸀠 𝜇2 − 𝑏2 ) 𝑔 (1, 𝜇) . (38) (39), we obtain
𝑚
Then, from Lemma 1, we have the following lemma. 󵄨󵄨 󵄨 𝑐0
󵄨󵄨F𝜃,𝑚 (𝜇)󵄨󵄨󵄨 ≤ (
|I𝜇|𝑚𝜃
󵄨 󵄨) 𝑒
1 + 𝜃 󵄨󵄨󵄨𝜇󵄨󵄨󵄨
Lemma 2. The function S(𝜇) is entire in 𝜇 and the following (45)
󵄨 󵄨2 2
estimate holds: 𝑐3 𝑐4 𝑐5 (1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 ) |I𝜇|
⋅ 󵄨 󵄨 𝑒 , 𝜇 ∈ C,
󵄨󵄨 󵄨󵄨2 2 1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨
󵄨󵄨 󵄨󵄨 𝑐3 𝑐4 𝑐5 (1 + 󵄨󵄨𝜇󵄨󵄨 ) |I𝜇| (39)
󵄨󵄨S (𝜇)󵄨󵄨 ≤ 󵄨 󵄨 𝑒 .
1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 leading to (42). Therefore, we get

Proof. Since 𝑚 󵄨 󵄨𝑚−4 󵄨 󵄨2 2


󵄨󵄨 𝑚−4 󵄨󵄨 𝑐3 𝑐4 𝑐5 𝑐0 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 (1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 )
󵄨󵄨𝜇 F𝜃,𝑚 (𝜇)󵄨󵄨 ≤ , 𝜇 ∈ R.
󵄨 󵄨 󵄨 󵄨 𝑚+1
S (𝜇) (1 + 𝜃 󵄨󵄨󵄨𝜇󵄨󵄨󵄨)
󵄨 󵄨 󵄨 󵄨2 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨 󵄨2 󵄨 󵄨 󵄨 󵄨 (46)
≤ (󵄨󵄨󵄨󵄨𝑏1󸀠 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑏1 󵄨󵄨󵄨) 󵄨󵄨󵄨𝑓 (1, 𝜇)󵄨󵄨󵄨 + (󵄨󵄨󵄨󵄨𝑏2󸀠 󵄨󵄨󵄨󵄨 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 + 󵄨󵄨󵄨𝑏2 󵄨󵄨󵄨) 󵄨󵄨󵄨𝑔 (1, 𝜇)󵄨󵄨󵄨 ,
(40) That is, 𝜇𝑚−4 F𝜃,𝑚 (𝜇) ∈ 𝐿2 (R). Moreover, we get

then from (29) and (30) we get (39). ∞


󵄨󵄨 𝑚−4 󵄨2
∫ 󵄨󵄨𝜇 F𝜃,𝑚 (𝜇)󵄨󵄨󵄨 𝑑𝜇
−∞
󵄨 󵄨
The analyticity of S(𝜇) and estimate (39) are not adequate
to prove that S(𝜇) lies in a Paley-Wiener space. To solve this ≤ 𝑐32 𝑐42 𝑐52 𝑐02𝑚
(47)
problem, we will multiply S(𝜇) by a regularization factor. Let
󵄨󵄨 󵄨󵄨2𝑚−8 󵄨󵄨󵄨2 4
𝜃 ∈ (0, 1) and let 𝑚 ∈ Z+ , 𝑚 > 4 be fixed. Let F𝜃,𝑚 (𝜇) be the ∞ 󵄨󵄨𝜇󵄨󵄨 (1 + 󵄨󵄨󵄨𝜇󵄨󵄨 ) 2 2 2 2𝑚 2
function ×∫ 󵄨󵄨 󵄨󵄨 2𝑚+2 𝑑𝜇 = 2𝑐3 𝑐4 𝑐5 𝑐0 ]0 .
−∞ (1 + 𝜃 󵄨󵄨𝜇󵄨󵄨)
𝑚
sin 𝜃𝜇
F𝜃,𝑚 (𝜇) := ( ) S (𝜇) , 𝜇 ∈ C. (41)
𝜃𝜇 What we have just proved is that F𝜃,𝑚 (𝜇) belongs to the
Paley-Wiener space PW2𝜎 with 𝜎 = 1 + 𝑚𝜃. Since F𝜃,𝑚 (𝜇) ∈
More specifications on 𝑚, 𝜃 will be given later on. Then we
have the next lemma. PW2𝜎 ⊂ PW22𝜎 , then we can reconstruct the functions F𝜃,𝑚 (𝜇)
via the following sampling formula:
Lemma 3. F𝜃,𝑚 (𝜇) is an entire function of 𝜇 which satisfies

the estimates 𝑛𝜋 2
F𝜃,𝑚 (𝜇) = ∑ [F𝜃,𝑚 ( ) 𝑆𝑛 (𝜇)
𝑛=−∞ 𝜎
𝑚 󵄨󵄨 󵄨󵄨2 2
󵄨󵄨 󵄨󵄨 𝑐3 𝑐4 𝑐5 𝑐0 (1 + 󵄨󵄨𝜇󵄨󵄨 ) |I𝜇|(1+𝑚𝜃)
󵄨󵄨F𝜃,𝑚 (𝜇)󵄨󵄨 ≤ 󵄨 󵄨 𝑚+1 𝑒 . (42) 𝑛𝜋 sin (𝜎𝜇 − 𝑛𝜋)
(1 + 𝜃 󵄨󵄨󵄨𝜇󵄨󵄨󵄨) + F󸀠𝜃,𝑚 ( ) 𝑆𝑛 (𝜇)] .
𝜎 𝜎
(48)
Moreover, 𝜇𝑚−4 F𝜃,𝑚 (𝜇) ∈ 𝐿2 (R) and
Let 𝑁 ∈ Z+ , 𝑁 > 𝑚 and approximate F𝜃,𝑚 (𝜇) by its

󵄨󵄨 𝑚−4 󵄨2 truncated series F𝜃,𝑚,𝑁(𝜇), where
󵄨󵄨𝜇 F𝜃,𝑚 (𝜇)󵄨󵄨󵄨 𝑑𝜇 ≤ √2𝑐3 𝑐4 𝑐5 𝑐0 ]0 ,
𝑚
𝐸𝑚−4 (F𝜃,𝑚 ) = √ ∫
−∞
𝑁
(43) 𝑛𝜋 2
F𝜃,𝑚,𝑁 (𝜇) := ∑ [F𝜃,𝑚 ( ) 𝑆𝑛 (𝜇)
𝑛=−𝑁 𝜎
where
𝑛𝜋 sin (𝜎𝜇 − 𝑛𝜋)
]0 + F󸀠𝜃,𝑚 ( ) 𝑆𝑛 (𝜇)] .
𝜎 𝜎
(49)
:= ( ((𝑚 (2𝑚 − 1) + 4𝜃2 ) Γ [2𝑚 + 2] + 144𝑚 (4𝑚2 − 1) 𝜃4
Since all eigenvalues are real, then from now on we restrict
× (280𝜃4 Γ [2𝑚 − 7] + 20𝜃2 Γ [2𝑚 − 5] + Γ [2𝑚 − 3]))
ourselves to 𝜇 ∈ R. Since 𝜇𝑚−4 F𝜃,𝑚 (𝜇) ∈ 𝐿2 (R), the
−1 1/2 truncation error, cf. (5), is given for |𝜇| < 𝑁𝜋/𝜎 by
× (𝑚 (4𝑚2 − 1) Γ [2𝑚 + 2] 𝜃2𝑚+1 ) ) .
󵄨󵄨 󵄨
(44) 󵄨󵄨F𝜃,𝑚 (𝜇) − F𝜃,𝑚,𝑁 (𝜇)󵄨󵄨󵄨 ≤ 𝑇𝑁,𝑚−4,𝜎 (𝜇) , (50)
6 Abstract and Applied Analysis

where In the following we use the technique of [26], where only


truncation error analysis is considered to determine enclo-
𝑇𝑁,𝑚−4,𝜎 (𝜇) 2
sure intervals for the eigenvalues, see also [41]. Let 𝜇∗ be an
󵄨 󵄨2
𝜉𝑚−4,𝜎 𝐸𝑚−4 󵄨󵄨󵄨sin 𝜎𝜇󵄨󵄨󵄨 1 1 eigenvalue; that is,
:= ( + )
√3(𝑁 + 1)𝑚−4 (𝑁𝜋 − 𝜎𝜇)
3/2
(𝑁𝜋 + 𝜎𝜇)
3/2 −𝑚
sin 𝜃𝜇∗
Δ (𝜇∗ ) = G (𝜇∗ ) + ( ) F𝜃,𝑚 (𝜇∗ ) = 0. (57)
󵄨 󵄨2 𝜃𝜇∗
𝜉𝑚−4,𝜎 (𝜎𝐸𝑚−4 + (𝑚 − 4) 𝐸𝑚−5 ) 󵄨󵄨󵄨sin 𝜎𝜇󵄨󵄨󵄨
+
𝜎(𝑁 + 1)𝑚−4 Then it follows that
−𝑚
1 1 sin 𝜃𝜇∗ ̃ 𝜃,𝑚,𝑁 (𝜇∗ )
×( + ). G (𝜇∗ ) + ( ) F
√𝑁𝜋 − 𝜎𝜇 √𝑁𝜋 + 𝜎𝜇 𝜃𝜇∗
(51) −𝑚 −𝑚
sin 𝜃𝜇∗ ̃ 𝜃,𝑚,𝑁 (𝜇∗ ) − ( sin 𝜃𝜇 )

𝑁 =( ) F F𝜃,𝑚 (𝜇∗ )
The samples {F𝜃,𝑚 (𝑛𝜋/𝜎)}𝑁 and
𝑛=−𝑁
{F󸀠𝜃,𝑚 (𝑛𝜋/𝜎)}𝑛=−𝑁,
in 𝜃𝜇∗ 𝜃𝜇∗
general, are not known explicitly. So we approximate them −𝑚 −𝑚
by solving numerically 8𝑁 + 4 initial value problems sin 𝜃𝜇∗ ̃ 𝜃,𝑚,𝑁 (𝜇∗ ) − ( sin 𝜃𝜇 )

= [( ) F F𝜃,𝑚,𝑁 (𝜇∗ )]
at the nodes {𝑛𝜋/𝜎}𝑁 ̃ 𝑁 𝜃𝜇∗ 𝜃𝜇∗
𝑛=−𝑁 . Let {F𝜃,𝑚 (𝑛𝜋/𝜎)}𝑛=−𝑁 and let
̃ 󸀠 (𝑛𝜋/𝜎)}𝑁
{F be the approximations of the samples of sin 𝜃𝜇∗
−𝑚
sin 𝜃𝜇∗
−𝑚
𝜃,𝑚 𝑛=−𝑁
𝑁 + [( ) F𝜃,𝑚,𝑁 (𝜇∗ ) − ( ) F𝜃,𝑚 (𝜇∗ )]
{F𝜃,𝑚 (𝑛𝜋/𝜎)}𝑁 and {F󸀠𝜃,𝑚 (𝑛𝜋/𝜎)}𝑛=−𝑁,
respectively. Now 𝜃𝜇∗ 𝜃𝜇∗
𝑛=−𝑁
̃
we define F𝜃,𝑚,𝑁(𝜇), which approximates F𝜃,𝑚,𝑁(𝜇) as (58)

𝑁 and so
̃ 𝜃,𝑚,𝑁 (𝜇) := ∑ [F
F ̃ 𝜃,𝑚 ( 𝑛𝜋 ) 𝑆2 (𝜇)
𝜎 𝑛 󵄨󵄨 −𝑚 󵄨󵄨
𝑛=−𝑁 󵄨󵄨 sin 𝜃𝜇∗ ̃ 𝜃,𝑚,𝑁 (𝜇∗ )󵄨󵄨󵄨󵄨
󵄨󵄨G (𝜇∗ ) + ( ) F
󵄨󵄨 𝜃𝜇∗ 󵄨󵄨
𝑛𝜋 sin (𝜎𝜇 − 𝑛𝜋) 󵄨 󵄨
̃󸀠 (
+F ) 𝑆𝑛 (𝜇)] , (59)
𝜃,𝑚
𝜎 𝜎 󵄨󵄨 sin 𝜃𝜇∗ 󵄨󵄨−𝑚
󵄨 󵄨󵄨
≤ 󵄨󵄨󵄨 ∗
󵄨 (𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A (𝜀)) .
󵄨󵄨 𝜃𝜇∗ 󵄨󵄨󵄨
𝑁 > 𝑚.
(52) ̃ 𝜃,𝑚,𝑁(𝜇∗ ) is given and
−𝑚
Since G(𝜇∗ ) + (sin 𝜃𝜇∗ /𝜃𝜇∗ ) F
∗ ∗ −𝑚 ∗
Using standard methods for solving initial problems, we may | sin 𝜃𝜇 /𝜃𝜇 | (𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A(𝜀)) has computable
assume that for |𝑛| < 𝑁, upper bound, we can define an enclosure for 𝜇∗ by solving
󵄨󵄨󵄨 𝑛𝜋 󵄨
̃ 𝜃,𝑚 ( 𝑛𝜋 )󵄨󵄨󵄨󵄨 < 𝜀,
the following system of inequalities:
󵄨󵄨F𝜃,𝑚 ( ) − F
󵄨󵄨 𝜎 𝜎 󵄨󵄨 󵄨󵄨 sin 𝜃𝜇∗ 󵄨󵄨−𝑚
󵄨 󵄨󵄨
󵄨󵄨󵄨 󸀠 󵄨
(53) − 󵄨󵄨󵄨 ∗
󵄨 (𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A (𝜀))
𝑛𝜋
󵄨󵄨F𝜃,𝑚 ( ) − F ̃ 󸀠 ( 𝑛𝜋 )󵄨󵄨󵄨󵄨 < 𝜀, 󵄨󵄨 𝜃𝜇∗ 󵄨󵄨󵄨
󵄨󵄨 𝜎 𝜃,𝑚
𝜎 󵄨󵄨
−𝑚
sin 𝜃𝜇∗ ̃ 𝜃,𝑚,𝑁 (𝜇∗ )
for a sufficiently small 𝜀. From (42) we can see that F𝜃,𝑚 (𝜇) ≤ G (𝜇∗ ) + ( ) F (60)
satisfies the condition (9) when 𝑚 > 4 and therefore 𝜃𝜇∗
whenever 0 < 𝜀 ≤ min{𝜋/𝜎, 𝜎/𝜋, 1/√𝑒} we have 󵄨󵄨 sin 𝜃𝜇∗ 󵄨󵄨−𝑚
󵄨 󵄨󵄨
󵄨󵄨 ̃ 𝜃,𝑚,𝑁 (𝜇)󵄨󵄨󵄨󵄨 ≤ A (𝜀) , 𝜇 ∈ R, ≤ 󵄨󵄨󵄨 ∗
󵄨 (𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A (𝜀)) .
󵄨󵄨F𝜃,𝑚,𝑁 (𝜇) − F (54) 󵄨󵄨 𝜃𝜇∗ 󵄨󵄨󵄨
󵄨 󵄨
where there is a positive constant 𝑀F𝜃,𝑚 for which, cf. (10), Its solution is an interval containing 𝜇∗ , and over which
−𝑚 ̃
the graph G(𝜇∗ ) + (sin 𝜃𝜇∗ /𝜃𝜇∗ ) F ∗
𝜃,𝑚,𝑁 (𝜇 ) is squeezed
2𝑒1/4 √ 𝜋 between the graphs as follows:
A (𝜀) := { 3𝑒 (1 + 𝜎) + ( 𝐴 + 𝑀F𝜃,𝑚 ) 𝜌 (𝜀)
𝜎 𝜎
(55) 󵄨󵄨 sin 𝜃𝜇∗ 󵄨󵄨−𝑚
󵄨 󵄨󵄨
1 −󵄨󵄨󵄨 ∗
󵄨 (𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A (𝜀)) ,
+ (𝜎 + 2 + log (2)) 𝑀F𝜃,𝑚 } 𝜀 log ( ) . 󵄨󵄨 𝜃𝜇∗ 󵄨󵄨󵄨
𝜀 (61)
󵄨󵄨 sin 𝜃𝜇∗ 󵄨󵄨−𝑚
Here 󵄨󵄨 󵄨󵄨 ∗
󵄨󵄨 󵄨 (𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A (𝜀)) .
󵄨󵄨 𝜃𝜇∗ 󵄨󵄨󵄨
3𝜎 󵄨󵄨 󵄨 𝜎
𝐴 := (󵄨F (0)󵄨󵄨 + 𝑀 ) ,
𝜋 󵄨 𝜃,𝑚 󵄨 𝜋 F𝜃,𝑚 Using the fact that
(56)
1
𝜌 (𝜀) := 𝛾 + 10 log ( ) . ̃ 𝜃,𝑚,𝑁 (𝜇) 󳨀→ F𝜃,𝑚 (𝜇)
F (62)
𝜀
Abstract and Applied Analysis 7

uniformly over any compact set and since 𝜇∗ is a simple root, and 𝜃 is chosen sufficiently small for which |𝜃𝜇| < 𝜋.
we obtain the following for large 𝑁 and sufficiently small 𝜀: Therefore, 𝜃, 𝑚 must be chosen so that for |𝜇| < 𝑁𝜋/𝜎
𝜕 sin 𝜃𝜇 −𝑚 ̃ 󵄨󵄨 󵄨󵄨
(G (𝜇) + ( ) F𝜃,𝑚,𝑁 (𝜇)) ≠
0 (63) 𝑚 > 4, 𝜃 ∈ (0, 1) , 󵄨󵄨𝜃𝜇󵄨󵄨 < 𝜋. (69)
𝜕𝜇 𝜃𝜇
Let 𝜇∗ be an eigenvalue and let 𝜇𝑁 be its approximation. Thus,
in a neighborhood of 𝜇∗ . Hence, the graph of Δ(𝜇∗ ) = 0 and Δ̃ 𝑁(𝜇𝑁) = 0. From (68) we have |Δ ̃ 𝑁(𝜇∗ )| ≤
G(𝜇) + (sin 𝜃𝜇/𝜃𝜇)−𝑚 F ̃ 𝜃,𝑚,𝑁(𝜇) intersects the graphs ∗ ∗ −𝑚 ∗
| sin 𝜃𝜇 /𝜃𝜇 | (𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A(𝜀)). Now we estimate the
−| sin 𝜃𝜇/𝜃𝜇| (𝑇𝑁,𝑚−4,𝜎 (𝜇) + A(𝜀)) and | sin 𝜃𝜇/𝜃𝜇|−𝑚
−𝑚
error |𝜇∗ − 𝜇𝑁| for an eigenvalue 𝜇∗ .
(𝑇𝑁,𝑚−4,𝜎 (𝜇) + A(𝜀)) at two points with abscissae
𝑎− (𝜇∗ , 𝑁, 𝜀) ≤ 𝑎+ (𝜇∗ , 𝑁, 𝜀) and the solution of the system of 2
Theorem 5. Let 𝜇∗ be an eigenvalue of Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 𝑏󸀠 ). For
inequalities (60) is the interval sufficient large 𝑁 we have the following estimate:
𝐼𝜀,𝑁 := [𝑎− (𝜇∗ , 𝑁, 𝜀) , 𝑎+ (𝜇∗ , 𝑁, 𝜀)] (64) 󵄨 󵄨−𝑚
󵄨󵄨 ∗ 󵄨 󵄨󵄨 sin 𝜃𝜇𝑁 󵄨󵄨󵄨 𝑇𝑁,𝑚−4,𝜎 (𝜇𝑁) + A (𝜀)
∗ 󵄨󵄨𝜇 − 𝜇𝑁󵄨󵄨󵄨 < 󵄨󵄨󵄨 󵄨 󵄨 󵄨 . (70)
and in particular 𝜇 ∈ 𝐼𝜀,𝑁. Summarizing the above 󵄨󵄨 𝜃𝜇𝑁 󵄨󵄨󵄨 inf 𝜁∈𝐼𝜀,𝑁 󵄨󵄨󵄨Δ󸀠 (𝜁)󵄨󵄨󵄨
discussion, we arrive at the following lemma which is similar
to that of [26]. Proof. Since Δ(𝜇𝑁) − Δ ̃ 𝑁(𝜇𝑁) = Δ(𝜇𝑁) − Δ(𝜇∗ ), then from
2 (68) and after replacing 𝜇 by 𝜇𝑁, we obtain
Lemma 4. For any eigenvalue 𝜇∗ , we can find 𝑁0 ∈ Z+ and
sufficiently small 𝜀 such that 𝜇∗ ∈ 𝐼𝜀,𝑁 for 𝑁 > 𝑁0 . Moreover, 󵄨󵄨 sin 𝜃𝜇 󵄨󵄨−𝑚
󵄨󵄨 ∗ 󵄨 󵄨 𝑁 󵄨󵄨
we get 󵄨󵄨Δ (𝜇𝑁) − Δ (𝜇 )󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨 󵄨 (𝑇𝑁,𝑚−4,𝜎 (𝜇𝑁) + A (𝜀)) .
󵄨󵄨 𝜃𝜇𝑁 󵄨󵄨󵄨
[𝑎− (𝜇∗ , 𝑁, 𝜀) , 𝑎+ (𝜇∗ , 𝑁, 𝜀)] 󳨀→ {𝜇∗ } (71)
(65)
as 𝑁 󳨀→ ∞, 𝜀 󳨀→ 0. Using the mean value theorem yields that for some 𝜁 ∈ 𝐽𝜀,𝑁 :=
󸀠 󸀠 [min(𝜇∗ , 𝜇𝑁), max(𝜇∗ , 𝜇𝑁)],
Proof. Since all eigenvalues of Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 ) are sim-
ple, then for large 𝑁 and sufficiently small 𝜀 we have 󵄨󵄨 ∗ 󵄨
󵄨󵄨(𝜇 − 𝜇𝑁) Δ󸀠 (𝜁)󵄨󵄨󵄨
(𝜕/𝜕𝜇)(G(𝜇) + (sin 𝜃𝜇/𝜃𝜇)−𝑚 F̃ 𝜃,𝑚,𝑁(𝜇)) > 0, in a neighbor- 󵄨 󵄨
hood of 𝜇∗ . Choose 𝑁0 such that 󵄨󵄨󵄨 sin 𝜃𝜇𝑁 󵄨󵄨󵄨−𝑚
≤ 󵄨󵄨󵄨 󵄨󵄨 (𝑇𝑁,𝑚−4,𝜎 (𝜇𝑁) + A (𝜀)) , 𝜁 ∈ 𝐽𝜀,𝑁 ⊂ 𝐼𝜀,𝑁.
sin 𝜃𝜇 −𝑚 ̃ 󵄨󵄨 𝜃𝜇𝑁 󵄨󵄨󵄨
G (𝜇) + ( ) F𝜃,𝑚,𝑁0 (𝜇) (72)
𝜃𝜇
(66)
󵄨󵄨 sin 𝜃𝜇 󵄨󵄨−𝑚 Since the eigenvalues are simple, then for sufficiently large
󵄨 󵄨󵄨
= ±󵄨󵄨󵄨 󵄨 (𝑇𝑁0 ,𝑚−4,𝜎 (𝜇) + A (𝜀)) 𝑁inf 𝜁∈𝐼𝜀,𝑁 |Δ󸀠 (𝜁)| > 0 and we get (70).
󵄨󵄨 𝜃𝜇 󵄨󵄨󵄨
has two distinct solutions which we denote by 𝑎− (𝜇∗ , 𝑁0 , 𝜀) ≤
𝑎+ (𝜇∗ , 𝑁0 , 𝜀). The decay of 𝑇𝑁,𝑚−4,𝜎 (𝜇) → 0 as 𝑁 → ∞ and 3. The Case of Π(𝑞, 𝑎, 𝑏, 0, 𝑏󸀠 )
A(𝜀) → 0 as 𝜀 → 0 will ensure the existence of the solutions This section includes briefly a treatment similarly to
𝑎− (𝜇∗ , 𝑁, 𝜀) and 𝑎+ (𝜇∗ , 𝑁, 𝜀) as 𝑁 → ∞ and 𝜀 → 0. For the that of the previous section for the eigenvalue problem
second point we recall that F ̃ 𝜃,𝑚,𝑁(𝜇) → F𝜃,𝑚 (𝜇) as 𝑁 → Π(𝑞, 𝑎, 𝑏, 0, 𝑏󸀠 ) introduced in Section 1. Notice that condition
∞ and as 𝜀 → 0. Hence, by taking the limit we obtain (18) implies that the analysis of problem Π(𝑞, 𝑎, 𝑏, 0, 𝑏󸀠 ) is
−𝑚 not included in that of Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 𝑏󸀠 ). Let 𝜓(⋅, 𝜇) denote the
sin 𝜃𝜇∗
G (𝑎+ (𝜇∗ , ∞, 0)) + ( ) F𝜃,𝑚 (𝑎+ (𝜇∗ , ∞, 0)) = 0, solution of (15) satisfying the following initial conditions:
𝜃𝜇∗
−𝑚 𝜓 (0, 𝜇) = 𝑎2 , 𝜓󸀠 (0, 𝜇) = −𝑎1 . (73)
sin 𝜃𝜇∗
G (𝑎− (𝜇∗ , ∞, 0)) + ( ) ∗
F𝜃,𝑚 (𝑎− (𝜇 , ∞, 0)) = 0.
𝜃𝜇∗ Thus, 𝜓(⋅, 𝜇) satisfies the boundary condition (16). The
(67) eigenvalues of the problem Π(𝑞, 𝑎, 𝑏, 0, 𝑏󸀠 ) are the zeros of the
function as follows:
That is, Δ(𝑎+ ) = Δ(𝑎− ) = 0. This leads us to conclude that
𝑎+ = 𝑎− = 𝜇∗ , since 𝜇∗ is a simple root. Ω (𝜇) := (𝑏1󸀠 𝜇2 − 𝑏1 ) 𝜓 (1, 𝜇) + (𝑏2󸀠 𝜇2 − 𝑏2 ) 𝜓󸀠 (1, 𝜇) . (74)
̃ 𝑁(𝜇) := G(𝜇) + (sin 𝜃𝜇/𝜃𝜇)−𝑚 F
Let Δ ̃ 𝜃,𝑚,𝑁(𝜇). Then (50)
Recall that Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 𝑏󸀠 ) has denumerable set of real and
and (54) imply
simple eigenvalues, cf. [38]. Using the method of variation
󵄨󵄨 󵄨󵄨−𝑚 of constants, the solution 𝜓(𝑥, 𝜇) satisfies Volterra integral
󵄨󵄨 ̃ 𝑁 (𝜇)󵄨󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨 sin 𝜃𝜇 󵄨󵄨󵄨 (𝑇𝑁,𝑚−4,𝜎 (𝜇) + A (𝜀)) ,
󵄨󵄨Δ (𝜇) − Δ
󵄨 󵄨 󵄨󵄨󵄨 𝜃𝜇 󵄨󵄨󵄨 equation as follows:
(68)
𝑁𝜋 sin 𝜇𝑥
󵄨󵄨󵄨𝜇󵄨󵄨󵄨 < 𝜓 (𝑥, 𝜇) = 𝑎2 cos 𝜇𝑥 − 𝑎1 + 𝑇 [𝜓] (𝑥, 𝜇) , (75)
󵄨 󵄨 𝜎 𝜇
8 Abstract and Applied Analysis

where 𝑇 is the Volterra operator defined in (22). Differentiat- reconstruct the functions R𝜃,𝑚 (𝜇) via the following sampling
ing (75), we get formula:
̃ [𝜓] (𝑥, 𝜇) , R𝜃,𝑚 (𝜇)
𝜓󸀠 (𝑥, 𝜇) = −𝑎2 𝜇 sin 𝜇𝑥 − 𝑎1 cos 𝜇𝑥 + 𝑇 (76)

̃ is the Volterra operator defined in (24). Define 𝑛𝜋 2
where 𝑇 = ∑ [R𝜃,𝑚 ( ) 𝑆𝑛 (𝜇)
𝜎 (86)
ℎ1 (⋅, 𝜇) and ℎ2 (⋅, 𝜇) to be 𝑛=−∞

̃ [𝜓] (𝑥, 𝜇) . 𝑛𝜋 sin (𝜎𝜇 − 𝑛𝜋)


ℎ1 (𝑥, 𝜇) := 𝑇 [𝜓] (𝑥, 𝜇) , ℎ2 (𝑥, 𝜇) := 𝑇 + R󸀠𝜃,𝑚 ( ) 𝑆𝑛 (𝜇)] .
(77) 𝜎 𝜎
Let 𝑁 ∈ Z+ , 𝑁 > 𝑚, and approximate R𝜃,𝑚 (𝜇) by its
As in the preceding section we split Ω(𝜇) into
truncated series R𝜃,𝑚,𝑁(𝜇), where
Ω (𝜇) := K (𝜇) + U (𝜇) , (78) R𝜃,𝑚,𝑁 (𝜇)
where K(𝜇) is the known part 𝑁
𝑛𝜋 2
:= ∑ [R𝜃,𝑚 ( ) 𝑆𝑛 (𝜇)
sin 𝜇 𝜎 (87)
K (𝜇) = (𝑏1󸀠 𝜇2 − 𝑏1 ) [𝑎2 cos 𝜇 − 𝑎1 ] 𝑛=−𝑁
𝜇 (79) 𝑛𝜋 sin (𝜎𝜇 − 𝑛𝜋)
+R󸀠𝜃,𝑚 ( ) 𝑆𝑛 (𝜇)] .
+ (𝑏2󸀠 𝜇2 − 𝑏2 ) [−𝑎2 𝜇 sin 𝜇 − 𝑎1 cos 𝜇] , 𝜎 𝜎

and U(𝜇) is the unknown one Since all eigenvalues are real, then from now on we restrict
ourselves to 𝜇 ∈ R. Since 𝜇𝑚−2 R𝜃,𝑚 (𝜇) ∈ 𝐿2 (R), the
U (𝜇) := (𝑏1󸀠 𝜇2 − 𝑏1 ) ℎ1 (1, 𝜇) + (𝑏2󸀠 𝜇2 − 𝑏2 ) ℎ2 (1, 𝜇) . (80) truncation error, cf. (5), is given for |𝜇| < 𝑁𝜋/𝜎 by
󵄨󵄨󵄨R𝜃,𝑚 (𝜇) − R𝜃,𝑚,𝑁 (𝜇)󵄨󵄨󵄨 ≤ 𝑇𝑁,𝑚−2,𝜎 (𝜇) , (88)
Then, as in the previous section, U(𝜇) is entire in 𝜇 for each 󵄨 󵄨
𝑥 ∈ [0, 1] for which where
󵄨󵄨 󵄨󵄨2 𝑇𝑁,𝑚−2,𝜎 (𝜇)
󵄨󵄨 󵄨󵄨 𝑐1 𝑐3 𝑐4 𝑐6 (1 + 󵄨󵄨𝜇󵄨󵄨 ) |I𝜇|
󵄨󵄨U (𝜇) 󵄨󵄨 ≤ 󵄨 󵄨 𝑒 , 𝜇 ∈ C, (81)
1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 󵄨 󵄨2
𝜉𝑚−2,𝜎 𝐸𝑚−2 󵄨󵄨󵄨sin 𝜎𝜇󵄨󵄨󵄨 1 1
:= ( + )
where 𝑐6 := max{|𝑏1 | + |𝑏2 |𝜏, |𝑏1󸀠 | + |𝑏2󸀠 |𝜏}. √3(𝑁 + 1)𝑚−2 (𝑁𝜋 − 𝜎𝜇)
3/2
(𝑁𝜋 + 𝜎𝜇)
3/2

Let 𝜃 ∈ (0, 1) and let 𝑚 be as in the previous section, but


󵄨 󵄨2
𝑚 > 2. Define R𝑚,𝜃 (𝜇) to be 𝜉𝑚−2,𝜎 (𝜎𝐸𝑚−2 + (𝑚 − 2) 𝐸𝑚−3 ) 󵄨󵄨󵄨sin 𝜎𝜇󵄨󵄨󵄨
+
𝜎(𝑁 + 1)𝑚−2
sin 𝜃𝜇 𝑚
R𝑚,𝜃 (𝜇) = ( ) U (𝜇) , 𝜇 ∈ C. (82)
𝜃𝜇 1 1
×( + ).
√𝑁𝜋 − 𝜎𝜇 √𝑁𝜋 + 𝜎𝜇
Hence,
(89)
󵄨 󵄨2
󵄨󵄨 󵄨 (1 + 󵄨󵄨󵄨𝜇󵄨󵄨󵄨 ) |I𝜇|(1+𝑚𝜃)
𝑐0𝑚 𝑐1 𝑐3 𝑐4 𝑐6 𝑁
󵄨󵄨R𝑚,𝜃 (𝜇)󵄨󵄨󵄨 ≤ 𝑒 , 𝜇 ∈ C, The samples {R𝜃,𝑚 (𝑛𝜋/𝜎)}𝑁 and {R󸀠𝜃,𝑚 (𝑛𝜋/𝜎)}𝑛=−𝑁, in
󵄨 󵄨 𝑚+1 𝑛=−𝑁
(1 + 𝜃 󵄨󵄨󵄨𝜇󵄨󵄨󵄨) general, are not known explicitly. So we approximate them
(83) by solving numerically 4𝑁 + 2 initial value problems
at the nodes {𝑛𝜋/𝜎}𝑁 . Let {R ̃ 𝜃,𝑚 (𝑛𝜋/𝜎)}𝑁 and let
and 𝜇𝑚−2 R𝑚,𝜃 (𝜇) ∈ 𝐿2 (R) with 𝑛=−𝑁 𝑛=−𝑁
̃ 󸀠 (𝑛𝜋/𝜎)}𝑁
{R be the approximations of the samples of
𝜃,𝑚 𝑛=−𝑁
𝑁

󵄨󵄨 𝑚−2 󵄨2 {R𝜃,𝑚 (𝑛𝜋/𝜎)}𝑁 and {R󸀠𝜃,𝑚 (𝑛𝜋/𝜎)}𝑛=−𝑁, respectively. Now
𝐸𝑚−2 (R𝑚,𝜃 ) = √ ∫ 󵄨󵄨𝜇 R𝑚,𝜃 (𝜇)󵄨󵄨󵄨 𝑑𝜇 𝑛=−𝑁
−∞ (84) ̃ 𝜃,𝑚,𝑁(𝜇), which approximates R𝜃,𝑚,𝑁(𝜇)
we define R

≤ √2𝑐0𝑚 𝑐1 𝑐3 𝑐4 𝑐6 𝜔0 , ̃ 𝜃,𝑚,𝑁 (𝜇)


R

where 𝑁
̃ 𝜃,𝑚 ( 𝑛𝜋 2
:= ∑ [R ) 𝑆𝑛 (𝜇)
𝜎
12𝜃2 Γ [2𝑚 − 3] + Γ [2𝑚 − 1] 𝑛=−𝑁
(90)
𝜔0 := √ . (85)
𝜃2𝑚−1 Γ [2 (𝑚 + 1)]
̃ 󸀠 ( 𝑛𝜋 ) sin (𝜎𝜇 − 𝑛𝜋) 𝑆𝑛 (𝜇)] ,
+R 𝜃,𝑚
𝜎 𝜎
Thus, R𝑚,𝜃 (𝜇) belongs to the Paley-Wiener space PW2𝜎 with
𝜎 = 1 + 𝑚𝜃. Since R𝜃,𝑚 (𝜇) ∈ PW2𝜎 ⊂ PW22𝜎 , then we can 𝑁 > 𝑚.
Abstract and Applied Analysis 9

Using standard methods for solving initial problems, we and 𝜃 is chosen sufficiently small for which |𝜃𝜇| < 𝜋.
may assume that for |𝑛| < 𝑁 Therefore, 𝜃, 𝑚 must be chosen so that for |𝜇| < 𝑁𝜋/𝜎
󵄨 󵄨
󵄨󵄨
󵄨󵄨R ( 𝑛𝜋 ) − R
󵄨 𝑚 > 2, 𝜃 ∈ (0, 1) , 󵄨󵄨󵄨𝜃𝜇󵄨󵄨󵄨 < 𝜋.
󵄨󵄨 𝜃,𝑚 ̃ 𝜃,𝑚 ( 𝑛𝜋 )󵄨󵄨󵄨󵄨 < 𝜀, (99)
󵄨 𝜎 𝜎 󵄨󵄨 Let 𝜇∗ be an eigenvalue and 𝜇𝑁 be its approximation. Thus
(91)
󵄨󵄨 󸀠 󵄨 ̃ 𝑁(𝜇𝑁) = 0. From (98) we have |Ω
̃ 𝑁(𝜇∗ )| ≤
󵄨󵄨R ( 𝑛𝜋 ) − R ̃ 󸀠 ( 𝑛𝜋 )󵄨󵄨󵄨󵄨 < 𝜀, Ω(𝜇∗ ) = 0 and Ω
󵄨󵄨 𝜃,𝑚
󵄨 𝜎 𝜃,𝑚
𝜎 󵄨󵄨 ∗ ∗ −𝑚 ∗
| sin 𝜃𝜇 /𝜃𝜇 | (𝑇𝑁,𝑚−2,𝜎 (𝜇 ) + A(𝜀)). Now we estimate the
error |𝜇∗ − 𝜇𝑁| for an eigenvalue 𝜇∗ . Finally we have the
for a sufficiently small 𝜀. From (83) we can see that R𝜃,𝑚 (𝜇) following estimate.
satisfies the condition (9) when 𝑚 > 2 and therefore
whenever 0 < 𝜀 ≤ min{𝜋/𝜎, 𝜎/𝜋, 1/√𝑒} we have 2
Theorem 7. Let 𝜇∗ be an eigenvalue of the problem
󵄨󵄨 ̃ 𝜃,𝑚,𝑁 (𝜇)󵄨󵄨󵄨󵄨 ≤ A (𝜀) ,
󵄨󵄨R𝜃,𝑚,𝑁 (𝜇) − R Π(𝑞, 𝑎, 𝑏, 0, 𝑏󸀠 ). For sufficient large 𝑁 we have the following
󵄨 󵄨 𝜇 ∈ R, (92)
estimate
󵄨 󵄨−𝑚
where there is a positive constant 𝑀R𝜃,𝑚 for which, cf. (10), 󵄨󵄨 ∗ 󵄨 󵄨󵄨 sin 𝜃𝜇𝑁 󵄨󵄨󵄨 𝑇𝑁,𝑚−2,𝜎 (𝜇𝑁) + A (𝜀)
󵄨󵄨𝜇 − 𝜇𝑁󵄨󵄨󵄨 < 󵄨󵄨󵄨 󵄨 󵄨 󵄨 . (100)
and 󵄨󵄨 𝜃𝜇𝑁 󵄨󵄨󵄨 inf 𝜁∈I𝜀,𝑁 󵄨󵄨󵄨Ω󸀠 (𝜁)󵄨󵄨󵄨
2𝑒1/4 √ 𝜋
A (𝜀) := { 3𝑒 (1 + 𝜎) + ( 𝐴 + 𝑀R𝜃,𝑚 ) 𝜌 (𝜀) 4. Numerical Examples
𝜎 𝜎
(93)
1 This section includes two detailed worked examples illus-
+ (𝜎 + 2 + log (2)) 𝑀R𝜃,𝑚 } 𝜀 log ( ) . trating the above technique. Examples 1 and 2 computed in
𝜀
[27, 45] with the classical sinc method, where only truncation
Here error analysis is considered, respectively. It is clearly seen that
3𝜎 󵄨󵄨 our new method (Hermite interpolations) gives remarkably
󵄨 𝜎
𝐴 := (󵄨R (0)󵄨󵄨 + 𝑀 ), better results than in [27, 45], see also [41–43]. We indicate in
𝜋 󵄨 𝜃,𝑚 󵄨 𝜋 R𝜃,𝑚 these examples the effect of the amplitude error in the method
(94)
1 by determining enclosure intervals for different values of
𝜌 (𝜀) := 𝛾 + 10 log ( ) .
𝜀 𝜀. We also indicate the effect of the parameters 𝑚 and 𝜃
by several choices. Each example is exhibited via figures
As in the above section, we have the following lemma. that accurately illustrate the procedure near to some of
2
the approximated eigenvalues. More explanations are given
Lemma 6. For any eigenvalue 𝜇∗ of the problem below. Recall that 𝑎± (𝜇) and 𝑏± (𝜇) are defined by
Π(𝑞, 𝑎, 𝑏, 0, 𝑏󸀠 ), we can find 𝑁0 ∈ Z+ and sufficiently
small 𝜀 such that 𝜇∗ ∈ I𝜀,𝑁 for 𝑁 > 𝑁0 , where 󵄨󵄨 󵄨󵄨−𝑚
̃ 𝑁 (𝜇) ± 󵄨󵄨󵄨 sin 𝜃𝜇 󵄨󵄨󵄨 (𝑇𝑁,𝑚−3,𝜎 (𝜇) + A (𝜀)) ,
𝑎± (𝜇) = Δ
󵄨󵄨󵄨 𝜃𝜇 󵄨󵄨󵄨
I𝜀,𝑁 := [𝑏− (𝜇∗ , 𝑁, 𝜀) , 𝑎𝑏+ (𝜇∗ , 𝑁, 𝜀)] , (95)
󵄨󵄨 󵄨󵄨 𝑁𝜋
󵄨󵄨𝜇󵄨󵄨 < ,
𝑏− , 𝑏+ are the solutions of the inequalities 𝜎
(101)
󵄨󵄨 sin 𝜃𝜇 󵄨󵄨−𝑚
󵄨 󵄨󵄨 󵄨󵄨 󵄨󵄨−𝑚
− 󵄨󵄨󵄨 󵄨 (𝑇𝑁,𝑚−2,𝜎 (𝜇) + A (𝜀))
󵄨󵄨 𝜃𝜇 󵄨󵄨󵄨 ̃ 𝑁 (𝜇) ± 󵄨󵄨󵄨 sin 𝜃𝜇 󵄨󵄨󵄨 (𝑇𝑁,𝑚−2,𝜎 (𝜇) + A (𝜀)) ,
𝑏± (𝜇) = Ω 󵄨󵄨 𝜃𝜇 󵄨󵄨
(96) 󵄨 󵄨
󵄨󵄨󵄨 sin 𝜃𝜇 󵄨󵄨󵄨−𝑚
≤Ω ̃ 𝑁 (𝜇) ≤ 󵄨󵄨 󵄨 󵄨󵄨 󵄨󵄨 𝑁𝜋
󵄨󵄨 𝜃𝜇 󵄨󵄨󵄨 (𝑇𝑁,𝑚−2,𝜎 (𝜇) + A (𝜀)) . 󵄨󵄨𝜇󵄨󵄨 < ,
󵄨 󵄨 𝜎
(102)
Moreover, we get
respectively. Recall also that the enclosure intervals 𝐼𝜀,𝑁 :=
[𝑏− (𝜇∗ , 𝑁, 𝜀) , 𝑏+ (𝜇∗ , 𝑁, 𝜀)] 󳨀→ {𝜇∗ } [𝑎− , 𝑎+ ] and I𝜀,𝑁 := [𝑏− , 𝑏+ ] are determined by solving
(97)
as 𝑁 󳨀→ ∞, 𝜀 󳨀→ 0. 󵄨󵄨 󵄨󵄨 𝑁𝜋
𝑎± (𝜇) = 0, 󵄨󵄨𝜇󵄨󵄨 < , (103)
−𝑚 ̃
𝜎
̃ 𝑁(𝜇) := K(𝜇) + (sin 𝜃𝜇/𝜃𝜇)
Let Ω R𝜃,𝑚,𝑁(𝜇). Then (88)
󵄨󵄨 󵄨󵄨 𝑁𝜋
and (92) imply 𝑏± (𝜇) = 0, 󵄨󵄨𝜇󵄨󵄨 < . (104)
𝜎
󵄨󵄨 ̃ 𝑁 (𝜇)󵄨󵄨󵄨󵄨
󵄨󵄨Ω (𝜇) − Ω respectively. We would like to mention that Mathematica
󵄨 󵄨
has been used to obtain the exact values for the three
󵄨󵄨 sin 𝜃𝜇 󵄨󵄨−𝑚
󵄨 󵄨󵄨 󵄨󵄨 󵄨󵄨 𝑁𝜋 examples where eigenvalues cannot be computed concretely.
≤ 󵄨󵄨󵄨 󵄨󵄨 (𝑇𝑁,𝑚−2,𝜎 (𝜇) + A (𝜀)) , 󵄨󵄨𝜇󵄨󵄨 <
󵄨󵄨 𝜃𝜇 󵄨󵄨 𝜎 Mathematica is also used in rounding the exact eigenvalues,
(98) which are square roots.
10 Abstract and Applied Analysis

0.08 0.06
0.06
0.04
0.04

0.02 0.02

0.00 0.00
−0.02
−0.02
−0.04
−0.04
−0.06
1.030 1.035 1.040 1.045 1.050 1.055 1.060 1.030 1.035 1.040 1.045 1.050 1.055 1.060

b− b−
Ω(𝜇) Ω(𝜇)
b+ b+

Figure 1: The enclosure interval dominating 𝜇1 for 𝑚 = 10, 𝑁 = 15, Figure 2: The enclosure interval dominating 𝜇1 for 𝑚 = 10, 𝑁 = 15,
𝜃 = 1/5, and 𝜀 = 10−5 . 𝜃 = 1/5, and 𝜀 = 10−10 .

Example 1. The boundary value problem [27] 0.4


󸀠󸀠 2
−𝑦 (𝑥, 𝜇) − 𝑦 (𝑥, 𝜇) = 𝜇 𝑦 (𝑥, 𝜇) , 0 ≤ 𝑥 ≤ 1,
0.2
𝑦󸀠 (0, 𝜇) = 0, (105)
0.0
𝑦 (1, 𝜇) + 𝑦󸀠 (1, 𝜇) = 𝜇2 (2𝑦 (1, 𝜇) + 𝑦󸀠 (1, 𝜇)) ,
−0.2
is a special case of the problem treated in the previous section
with 𝑎1 = 0, 𝑎2 = 𝑏1 = 𝑏2 = 𝑏2󸀠 = 1, 𝑏1󸀠 = 2 and 𝑞(𝑥) = −1. The
−0.4
characteristic function is

Ω (𝜇) = (−1 + 2𝜇2 ) cos [√1 + 𝜇2 ] 3.510 3.515 3.520 3.525 3.530
(106)
b−
− (−1 + 𝜇2 ) √1 + 𝜇2 sin [√1 + 𝜇2 ] . Ω(𝜇)
b+
The function K(𝜇) will be
Figure 3: The enclosure interval dominating 𝜇2 for 𝑚 = 10, 𝑁 = 15,
K (𝜇) = (−1 + 2𝜇2 ) cos [𝜇] − 𝜇 (−1 + 𝜇2 ) sin [𝜇] . (107) 𝜃 = 1/5, and 𝜀 = 10−5 .

As is clearly seen, the eigenvalues cannot be computed


explicitly. Tables 1, 2, and 3 indicate the application of our
technique to this problem and the effect of 𝑚, 𝜃 and 𝜀. By exact is a special case of problem Π(𝑞, 𝑎, 𝑏, 𝑎󸀠 , 𝑏󸀠 ), when 𝑞(𝑥) = −1,
we mean the zeros of Ω(𝜇) computed by Mathematica. 𝑎2 = 𝑎1󸀠 = 𝑏1 = 𝑏2󸀠 = 0, and 𝑎1 = 𝑎2󸀠 = 𝑏1󸀠 = 𝑏2 = 1. Here the
Figures 1 and 2 illustrate the enclosure intervals dom- characteristic function is
inating 𝜇1 for 𝑁 = 15, 𝑚 = 10, 𝜃 = 1/5, and
𝜀 = 10−5 and 𝜀 = 10−10 respectively. The middle curve sin √𝜇2 + 1
represents Ω(𝜇), while the upper and lower curves represent Δ (𝜇) = (1 − 𝜇4 ) cos √𝜇2 + 1 − (2𝜇2 + 𝜇4 ) .
the curves of 𝑏+ (𝜇), 𝑏− (𝜇), respectively. We notice that when √𝜇2 + 1
𝜀 = 10−10 , the two curves are almost identical. Similarly, (109)
Figures 3 and 4 illustrate the enclosure intervals dominating
𝜇2 for 𝑁 = 15, 𝑚 = 10, 𝜃 = 1/5, and 𝜀 = 10−5 and 𝜀 = 10−10 After computing G(𝜇), we obtain
respectively.

Example 2. The boundary value problem [45] G (𝜇) = (1 + 𝜇2 ) ((1 − 𝜇2 ) cos 𝜇 − 𝜇 sin 𝜇) . (110)

−𝑦󸀠󸀠 (𝑥, 𝜇) − 𝑦 (𝑥, 𝜇) = 𝜇2 𝑦 (𝑥, 𝜇) 0 ≤ 𝑥 ≤ 1, As is clearly seen, the eigenvalues cannot be computed
(108) explicitly. As in the previous example, Figures 5, 6, 7, and 8
𝑦 (0, 𝜇) = 𝜇2 𝑦󸀠 (0, 𝜇) , 𝑦󸀠 (1, 𝜇) = 𝜇2 𝑦 (1, 𝜇) , illustrate the results of Tables 4, 5, 6, and 7.
Abstract and Applied Analysis 11

Table 1: With 𝑁 = 15, the approximation 𝜇𝑘,𝑁 and the exact solution 𝜇𝑘 for different choices of 𝑚 and 𝜃.

𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
Exact 𝜇𝑘 1.0493258679653497 3.5207214555369464 6.505146961583527 9.578576417519093
𝜇𝑘,𝑁
𝑚=6
𝜃 = 1/9 1.0493258487568435 3.5207214557864277 6.5051469611825405 9.578576417536349
𝜃 = 1/12 1.0493291624256957 3.5207212860735546 6.5051469673141336 9.578576393056661
𝑚 = 10
𝜃 = 1/5 1.0493258679653554 3.5207214555369624 6.505146961583516 9.57857641751921
𝜃 = 1/8 1.049325865990155 3.520721455899672 6.505146961464727 9.578576417559768
Table 2: Absolute error |𝜇𝑘 − 𝜇𝑘,𝑁 |.

𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
𝑚=6
𝜃 = 1/9 1.92085 × 10−8 2.49481 × 10−10 4.00987 × 10−10 1.72555 × 10−11
𝜃 = 1/12 3.29446 × 10−6 1.69463 × 10−7 5.73061 × 10−9 2.44624 × 10−8
𝑚 = 10
𝜃 = 1/5 5.77316 × 10−15 1.59872 × 10−14 1.15463 × 10−14 1.1724 × 10−13
𝜃 = 1/8 1.97519 × 10−9 3.62725 × 10−10 1.188 × 10−10 4.0675 × 10−11

Table 3: For 𝑁 = 15, 𝑚 = 10, and 𝜃 = 1/5, the exact solution 𝜇𝑘 are all inside the interval [𝑏− , 𝑏+ ] for different values of 𝜀.

𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
Exact 𝜇𝑘 1.0493258679653497 3.5207214555369464 6.505146961583527 9.578576417519093
I𝜀,𝑁 , 𝜀 = 10−5 [1.04294069, 1.05557896] [3.51981844, 3.52162396] [6.50375768, 6.50653831] [9.55222305, 9.60712093]
I𝜀,𝑁 , 𝜀 = 10−10 [1.04932561, 1.049326118] [3.52072141, 3.52072149] [6.50514690, 6.50514702] [9.57857529, 9.57857754]
𝐸8 (R𝜃,𝑚 ) = 4.51845 × 108 , 𝐸7 (R𝜃,𝑚 ) = 2.29709 × 105 , ] = 1, 𝑀R𝜃,𝑚 = 4.55609 × 104 .

Table 4: With 𝑁 = 40, the approximation 𝜇𝑘,𝑁 and the exact solution 𝜇𝑘 for different choices of 𝜃.

𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
Exact 𝜇𝑘 0.4828692021748484 1.966318052350425 4.827089429919572 7.919684444168381
𝜇𝑘,𝑁
𝑚=8
𝜃 = 1/32 0.48286920221045176 1.96631805234574 4.827089429919605 7.919684444168366
𝜃 = 1/35 0.4828692337692527 1.966318047624416 4.8270894299720776 7.91968444416245

Table 5: Absolute error |𝜇𝑘 − 𝜇𝑘,𝑁 |.

𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
𝑚=8
𝜃 = 1/32 3.56034 × 10−11 4.68492 × 10−12 3.28626 × 10−14 1.5099 × 10−14
𝜃 = 1/35 3.15944 × 10−8 4.72601 × 10−9 5.25047 × 10−11 5.93126 × 10−12

Table 6: The approximation 𝜇𝑘,𝑁 and the exact solution 𝜇𝑘 for 𝑁 = 40, 𝑚 = 14 and 𝜃 = 1/26.

𝜇𝑘 exact 𝜇𝑘 𝜇𝑘,𝑁 |𝜇𝑘 − 𝜇𝑘,𝑁 |


𝜇1 0.4828692021748484698568637 0.4828692021748484678442680 2.012596 × 10−18
𝜇2 1.966318052350424642326091 1.9663180523504246423320204 5.93 × 10−21
𝜇3 4.8270894299195722717631337 4.8270894299195722717463715 1.6762 × 10−20
𝜇4 7.9196844441683813942255769 7.9196844441683813942260057 4.29 × 10−22

Table 7: For 𝑁 = 40, 𝑚 = 14 and 𝜃 = 1/26, the exact solution 𝜇𝑘 are all inside the interval [𝑎− , 𝑎+ ] for different values of 𝜀.

𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
Exact 𝜇𝑘 0.4828692021748484 1.966318052350425 4.827089429919572 7.919684444168381
I𝜀,𝑁 ,, 𝜀 = 10−5 [0.47918888, 0.48651557] [1.96592879, 1.96670680] [4.82707252, 4.82710633] [7.91968171, 7.919687175]
I𝜀,𝑁 , 𝜀 = 10−10 [0.48284084, 0.48289756] [1.96631794, 1.96631815] [4.82708919, 4.82708966] [7.919684437, 7.919684450]
𝐸10 (F𝜃,𝑚 ) = 2.83057 × 1018 , 𝐸9 (F𝜃,𝑚 ) = 1.12829 × 1014 , ] = 1, 𝑀F𝜃,𝑚 = 1.57716 × 107 .
12 Abstract and Applied Analysis

0.03
0.4
0.02
0.2
0.01

0.0 0.00

−0.01
−0.2
−0.02
−0.4 −0.03

−0.04
3.510 3.515 3.520 3.525 3.530 7.91967 7.91968 7.91968 7.91969

b− a−
Ω(𝜇) Δ(𝜇)
b+ a+

Figure 4: The enclosure interval dominating 𝜇2 for 𝑚 = 10, 𝑁 = 15, Figure 7: The enclosure interval dominating 𝜇4 for 𝑚 = 14, 𝑁 = 40,
𝜃 = 1/5, and 𝜀 = 10−10 . 𝜃 = 1/26, and 𝜀 = 10−5 .

0.15 0.03

0.10 0.02

0.05 0.01

0.00 0.00

−0.05 −0.01

−0.10 −0.02

−0.15 −0.03

0.42 0.44 0.46 0.48 0.50 0.52 0.54 7.91967 7.91968 7.91968 7.91969

a− a−
Δ(𝜇) Δ(𝜇)
a+ a+

Figure 5: The enclosure interval dominating 𝜇1 for 𝑚 = 14, 𝑁 = 40, Figure 8: The enclosure interval dominating 𝜇4 for 𝑚 = 14, 𝑁 = 40,
𝜃 = 1/26, and 𝜀 = 10−5 . 𝜃 = 1/26, and 𝜀 = 10−10 .

Acknowledgment
0.10
This research has been supported by a grant from the Institute
0.05 of Scientific Research at Umm AL-Qura University, Saudi
Arabia.
0.00

−0.05 References
−0.10 [1] E. H. Doha, A. H. Bhrawy, and R. M. Hafez, “A Jacobi-
Jacobi dual-Petrov-Galerkin method for third- and fifth-order
−0.15 differential equations,” Mathematical and Computer Modelling,
vol. 53, no. 9-10, pp. 1820–1832, 2011.
0.42 0.44 0.46 0.48 0.50 0.52 0.54 [2] E. H. Doha, A. H. Bhrawy, and M. A. Saker, “Integrals of
Bernstein polynomials: an application for the solution of high
a−
even-order differential equations,” Applied Mathematics Letters,
Δ(𝜇)
a+
vol. 24, no. 4, pp. 559–565, 2011.
[3] E. Tohidi, A. H. Bhrawy, and K. Erfani, “A collocation method
Figure 6: The enclosure interval dominating 𝜇1 for 𝑚 = 14, 𝑁 = 40, based on Bernoulli operational matrix for numerical solution
𝜃 = 1/26, and 𝜀 = 10−10 . of generalized pantograph equation,” Applied Mathematical
Abstract and Applied Analysis 13

Modelling, vol. 37, no. 6, pp. 4283–4294, 2013. [19] J. R. Higgins, G. Schmeisser, and J. J. Voss, “The sampling
[4] E. H. Doha and A. H. Bhrawy, “Efficient spectral-Galerkin theorem and several equivalent results in analysis,” Journal of
algorithms for direct solution for second-order differential Computational Analysis and Applications, vol. 2, no. 4, pp. 333–
equations using Jacobi polynomials,” Numerical Algorithms, vol. 371, 2000.
42, no. 2, pp. 137–164, 2006. [20] G. Hinsen, “Irregular sampling of bandlimited 𝐿𝑝 -functions,”
[5] A. H. Bhrawy, “A Jacobi-Gauss-Lobatto collocation method Journal of Approximation Theory, vol. 72, no. 3, pp. 346–364,
for solving generalized Fitzhugh-Nagumo equation with time- 1993.
dependent coefficients,” Applied Mathematics and Computation, [21] D. Jagerman and L. Fogel, “Some general aspects of the sampling
vol. 222, pp. 255–264, 2013. theorem,” IRE Transactions on Information Theory, vol. 2, pp.
[6] A. H. Bhrawy, M. A. Abdelkawy, and A. Biswas, “Cnoidal and 139–146, 1956.
snoidal wave solutions to coupled nonlinear wave equations by [22] M. H. Annaby and R. M. Asharabi, “Error analysis associated
the extended Jacobi’s elliptic function method,” Communica- with uniform Hermite interpolations of bandlimited functions,”
tions in Nonlinear Science and Numerical Simulation, vol. 18, no. Journal of the Korean Mathematical Society, vol. 47, no. 6, pp.
4, pp. 915–925, 2013. 1299–1316, 2010.
[7] A. Imani, A. Aminataei, and A. Imani, “Collocation method via [23] J. R. Higgins, Sampling Theory in Fourier and Signal Analysis:
Jacobi polynomials for solving nonlinear ordinary differential Foundations, Oxford University Press, Oxford, UK, 1996.
equations,” International Journal of Mathematics and Mathe- [24] P. L. Butzer, J. R. Higgins, and R. L. Stens, “Sampling theory of
matical Sciences, vol. 2011, Article ID 673085, 11 pages, 2011. signal analysis,” in Development of Mathematics 1950–2000, pp.
[8] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional 193–234, Birkhäuser, Basel, Switzerland, 2000.
Calculus: Models and Numerical Methods, vol. 3 of Series on [25] P. L. Butzer, G. Schmeisser, and R. L. Stens, “An introduction
Complexity, Nonlinearity and Chaos, World Scientific, Hacken- to sampling analysis,” in Non Uniform Sampling: Theory and
sack, NJ, USA, 2012. Practices, F. Marvasti, Ed., pp. 17–121, Kluwer, New York, NY,
[9] D. Baleanu, A. H. Bhrawy, and T. M. Taha, “Two efficient USA, 2001.
generalized Laguerre spectral algorithms for fractional initial [26] A. Boumenir, “Higher approximation of eigenvalues by the
value problems,” Abstract and Applied Analysis, vol. 2013, Article sampling method,” BIT Numerical Mathematics, vol. 40, no. 2,
ID 546502, 10 pages, 2013. pp. 215–225, 2000.
[10] D. Baleanu, A. H. Bhrawy, and T. M. Taha, “A modified gen- [27] A. Boumenir, “The sampling method for Sturm-Liouville prob-
eralized Laguerre spectral methods for fractional differential lems with the eigenvalue parameter in the boundary condition,”
equations on the half line,” Abstract and Applied Analysis, vol. Numerical Functional Analysis and Optimization, vol. 21, no. 1-2,
2013, Article ID 413529, 12 pages, 2013. pp. 67–75, 2000.
[11] F. Ghaemi, R. Yunus, A. Ahmadian, S. Salahshour, M. B. [28] A. Boumenir and B. Chanane, “Eigenvalues of Sturm-Liouville
Suleiman, and S. F. Saleh, “Application of fuzzy fractional kinetic systems using sampling theory,” Applied Analysis, vol. 62, pp.
equations to modelling of the acid hydrolysis reaction,” Abstract 323–334, 1996.
and Applied Analysis, vol. 2013, Article ID 610314, 19 pages, 2013.
[29] M. M. Tharwat, A. H. Bhrawy, and A. Yildirim, “Numerical
[12] A. H. Bhrawy and A. S. Alofi, “The operational matrix of frac- computation of the eigenvalues of a discontinuous Dirac system
tional integration for shifted Chebyshev polynomials,” Applied using the Sinc method with error analysis,” International Journal
Mathematics Letters, vol. 26, no. 1, pp. 25–31, 2013. of Computer Mathematics, vol. 89, no. 15, pp. 2061–2080, 2012.
[13] A. H. Bhrawy and M. M. Al-Shomrani, “A shifted Legendre [30] J. Lund and K. Bowers, Sinc Methods for Quadrature and
spectral method for fractional-order multi-point boundary Differential Equations, SIAM, Philadelphia, Pa, USA, 1992.
value problems,” Advances in Difference Equations, vol. 2012, p.
[31] F. Stenger, “Numerical methods based on Whittaker cardinal, or
8, 2012.
Sinc functions,” SIAM Review, vol. 23, no. 2, pp. 165–224, 1981.
[14] M. H. Annaby and R. M. Asharabi, “Computing eigenvalues of
[32] F. Stenger, Numerical Methods Based on Sinc and Analytic
Sturm-Liouville problems by Hermite interpolations,” Numeri-
Functions, Springer, New York, NY, USA, 1993.
cal Algorithms, vol. 60, no. 3, pp. 355–367, 2012.
[33] R. P. Boas, Entire Functions, Academic Press, New York, NY,
[15] M. M. Tharwat, A. H. Bhrawy, and A. S. Alofi, “Computing
USA, 1954.
eigenvalues of discontinuous Sturm-Liouville problems with
eigenparameter in all boundary conditions using Hermite [34] M. S. P. Eastham, Theory of Ordinary Differential Operators, Van
approximation,” Abstract and Applied Analysis, vol. 2013, Article Nostrand Reinhold, London, UK, 1970.
ID 498457, 14 pages, 2013. [35] B. M. Levitan and I. S. Sargsjan, “Introduction to spectral the-
[16] M. M. Tharwat, A. H. Bhrawy, and A. S. Alofi, “Approximation ory: selfadjoint ordinary differential operators,” in Translation
of eigenvalues of discontinuous Sturm-Liouville problems with of Mthematical Monographs, vol. 39, American Mathematical
eigenparameter in all boundary conditions,” Boundary Value Society, Providence, RI, USA, 1975.
Problems, vol. 2013, p. 132, 2013. [36] M. H. Annaby and M. M. Tharwat, “On sampling theory and
[17] M. M. Tharwat, A. H. Bhrawy, and A. Yildirim, “Numerical eigenvalue problems with an eigenparameter in the boundary
computation of eigenvalues of discontinuous Sturm-Liouville conditions,” SUT Journal of Mathematics, vol. 42, no. 2, pp. 157–
problems with parameter dependent boundary conditions 176, 2006.
using Sinc method,” Numerical Algorithms, vol. 63, no. 1, pp. 27– [37] M. M. Tharwat, “Discontinuous Sturm-Liouville problems and
48, 2013. associated sampling theories,” Abstract and Applied Analysis,
[18] G. R. Grozev and Q. I. Rahman, “Reconstruction of entire vol. 2011, Article ID 610232, 30 pages, 2011.
functions from irregularly spaced sample points,” Canadian [38] C. T. Fulton, “Two-point boundary value problems with
Journal of Mathematics, vol. 48, no. 4, pp. 777–793, 1996. eigenvalue parameter contained in the boundary conditions,”
14 Abstract and Applied Analysis

Proceedings of the Royal Society of Edinburgh A, vol. 77, no. 3-4,


pp. 293–308, 1977.
[39] E. C. Titchmarsh, Eigenfunction Expansions Associated with
Second-Order Differential Equations, Part 1, Clarendon Press,
Oxford, UK, 1962.
[40] J. Walter, “Regular eigenvalue problems with eigenvalue param-
eter in the boundary condition,” Mathematische Zeitschrift, vol.
133, pp. 301–312, 1973.
[41] M. H. Annaby and M. M. Tharwat, “The Hermite interpolation
approach for computing eigenvalues of dirac systems,” Mathe-
matical and Computer Modelling, vol. 57, pp. 2459–2472, 2013.
[42] M. M. Tharwat, “Computing eigenvalues and Hermite inter-
polation for Dirac systems with eigenparameter in boundary
conditions,” Boundary Value Problems, vol. 2013, p. 36, 2013.
[43] M. M. Tharwat and A. H. Bhrawy, “Computation of eigenvalues
of discontinuous Dirac system using Hermite interpolation
technique,” Advances in Difference Equations, vol. 2012, p. 59,
2012.
[44] K. Chadan and P. C. Sabatier, Inverse Problems in Quantum
Scattering Theory, Springer, New York, NY, USA, 2nd edition,
1989.
[45] S. M. Al-Harbi, “Computing of eigenvalues of Sturm-Liouville
problems with eigenparameter dependent boundary condi-
tions,” Iranian Journal of Science and Technology A, vol. 35, no.
4, pp. 293–299, 2011.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 176730, 10 pages
http://dx.doi.org/10.1155/2013/176730

Research Article
Approximate Solutions of Fisher’s Type Equations with
Variable Coefficients

A. H. Bhrawy1,2 and M. A. Alghamdi1


1
Department of Mathematics, Faculty of Science, King Abdulaziz University, Jeddah 21589, Saudi Arabia
2
Department of Mathematics, Faculty of Science, Beni-Suef University, Beni-Suef 62511, Egypt

Correspondence should be addressed to A. H. Bhrawy; alibhrawy@yahoo.co.uk

Received 6 September 2013; Accepted 20 September 2013

Academic Editor: Dumitru Baleanu

Copyright © 2013 A. H. Bhrawy and M. A. Alghamdi. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

The spectral collocation approximations based on Legendre polynomials are used to compute the numerical solution of time-
dependent Fisher’s type problems. The spatial derivatives are collocated at a Legendre-Gauss-Lobatto interpolation nodes. The
proposed method has the advantage of reducing the problem to a system of ordinary differential equations in time. The four-stage
A-stable implicit Runge-Kutta scheme is applied to solve the resulted system of first order in time. Numerical results show that the
Legendre-Gauss-Lobatto collocation method is of high accuracy and is efficient for solving the Fisher’s type equations. Also the
results demonstrate that the proposed method is powerful algorithm for solving the nonlinear partial differential equations.

1. Introduction was firstly introduced by Fisher in [22] to describe the propa-


gation of a mutant gene. Fisher equations have a wide applica-
Spectral methods (see, for instance, [1–5]) are powerful tech- tion in a large number of the chemical kinetics [23], logistic
niques that we use to numerically solve linear and nonlinear population growth [24], flame propagation [25], population
partial differential equations either in their strong or weak in one-dimensional habitual [26], neutron population in
forms. What sets spectral methods apart from others like a nuclear reaction [27], neurophysiology [28], branching
finite difference methods or finite element methods is that to Brownian motion [23], autocatalytic chemical reactions [29],
get a spectral method we approximate the solutions by high and nuclear reactor theory [30].
order orthogonal polynomial expansions. The orthogonal In recent years, many physicists and mathematicians have
polynomial approximations can have very high convergence paid much attention to the Fisher equations due to their
rates, which allow us to use fewer degrees of freedom for a importance in mathematical physics. In [31], Öǧün and Kart
desired level of accuracy. The most common spectral method utilized truncated Painlevè expansions for presenting some
from the strong form of the equations is known as collocation. exact solutions of Fisher and generalized Fisher equations.
In collocation techniques, the partial differential equation Tan et al. [28] proposed the homotopy analysis method to
must be satisfied at a set of grid, or more precisely, collocation find analytical solution of Fisher equations. Gunzburger et
points (see, for instance, [6–10]). Spectral methods also have al. [32] applied the discrete finite element approximation for
become increasingly popular for solving fractional differen- obtaining a numerical solution of the forced Fisher equation.
tial equations [11–21]. Dag et al. [33] discussed and applied the B-spline Galerkin
In this paper, we present an accurate numerical solution method for Fisher’s equation. Bastani and Salkuyeh [34]
based on Legendre-Gauss-Lobatto collocation method for proposed the compact finite difference approach in combi-
Fisher’s type equations. The Fisher equation in the form nation with third-order Runge-Kutta scheme to solve Fisher’s
equation. More recently, Mittal and Jain [35] investigated the
cubic B-spline scheme for solving Fisher’s reaction-diffusion
𝑢𝑡 = 𝐷𝑢𝑥𝑥 + ]𝑢 (1 − 𝑢) (1) problem. However, the fisher equations have been studied in
2 Abstract and Applied Analysis

many other articles by numerous numerical methods such where


as pseudospectral method [36, 37], finite difference method
[38–44], finite element method [45], B-spline algorithm [46], 𝐶𝑞 (𝑘, 𝑖)
and Galerkin method [47, 48].
To increase the numerical solution accuracy, spectral 2𝑞−1 (2𝑖 + 1) Γ [(𝑞 + 𝑘 − 𝑖) /2] Γ [(𝑞 + 𝑘 + 𝑖 + 1) /2]
= .
collocation methods based on orthogonal polynomials are Γ [𝑞] Γ [(2 − 𝑞 + 𝑘 − 𝑖) /2] Γ [(3 − 𝑞 + 𝑘 + 𝑖) /2]
often chosen. Doha et al. [49] proposed and developed a (4)
new numerical algorithm for solving the initial-boundary
system of nonlinear hyperbolic equations based on spec- The analytical form of Legendre polynomial is
tral collocation method; a Chebyshev-Gauss-Radau colloca-
[𝑛/2]
tion method in combination with the implicit Runge-Kutta
scheme are employed to obtain highly accurate approxi- 𝐿 𝑛 (𝑥) = ∑ 𝑐𝑘(𝑛) 𝑥𝑛−2𝑘 , (5)
𝑖=0
mations to this system of nonlinear hyperbolic equations.
In [50], Bhrawy proposed an efficient Jacobi-Gauss-Lobatto
collocation method for approximating the solution of the where 𝑐𝑘(𝑛) = (−1)𝑘 (2𝑛 − 2𝑘)!/2𝑛 (𝑛 − 𝑘)!(𝑛 − 2𝑘)!𝑘!, and
generalized Fitzhugh-Nagumo equation in which the Jacobi- 𝑛
Gauss-Lobatto points are used as collocation nodes for 𝑛 {
{2, even,
spatial derivatives. Moreover, the Jacobi spectral collocation [ ] = {𝑛 − 1 (6)
2 { , odd.
methods are used to solve some problems in mathematical { 2
physics, (see, for instance, [51–53]).
Indeed, there are no results on Legendre-Gauss-Lobatto It is also generating from the following relation:
collocation method for solving nonlinear Fisher-type equa-
tions subject to initial-boundary conditions. Therefore, the 2𝑘 + 3 𝑘+1
𝐿 𝑘+2 (𝑥) = 𝑥𝐿 𝑘+1 (𝑥) − 𝐿 (𝑥) , (7)
objective of this work is to present a numerical algorithm 𝑘+2 𝑘+2 𝑘
for solving such equation based on Legendre-Gauss-Lobatto
pseudospectral method. The spatial derivatives are approx- with 𝐿 0 (𝑥) = 1, 𝐿 1 (𝑥) = 𝑥, and satisfies the orthogonality
imated at these grid points by approximating the deriva- condition
tives of Legendre polynomial that interpolates the solutions. 1
Moreover, we set the boundary conditions in the collocation (𝐿 𝑘 (𝑥), 𝐿 𝑙 (𝑥))𝑤 = ∫ 𝐿 𝑘 (𝑥) 𝐿 𝑙 (𝑥) 𝑤 (𝑥) = ℎ𝑘 𝛿𝑙𝑘 . (8)
method. The problem is then reduced to system of first- −1
order ordinary differential equations in time. The four-
where 𝑤(𝑥) = 1, ℎ𝑘 = 2/(2𝑘 + 1). Let 𝑆𝑁 be the space of all
stage A-stable implicit Runge-Kutta scheme is proposed for
polynomials of degree ≤ 𝑁, then for any 𝜙 ∈ 𝑆2𝑁−1 (0, 𝐿),
treating the this system of equations. Finally, some illustrative
examples are implemented to illustrate the efficiency and 1 𝑁
applicability of the proposed approach. ∫ 𝑤 (𝑥) 𝜙 (𝑥) 𝑑𝑥 = ∑𝜛𝑁,𝑗 𝜙 (𝑥𝑁,𝑗 ) . (9)
The rest of this paper is structured as follows. In the next −1 𝑗=0
section, some properties of Legendre polynomials, which
are required for implementing our algorithm, are presented. Let us define the following discrete inner product and norm:
Section 3 is devoted to the development of Gauss-Lobatto
collocation technique for a general form of Fisher-type equa- 𝑁

tions based on the Legendre polynomials, and in Section 4 the (𝑢, V)𝑤 = ∑ 𝑢 (𝑥𝑁,𝑗 ) V (𝑥𝑁,𝑗 ) 𝜛𝑁,𝑗 , (10)
𝑗=0
proposed method is implemented to obtain some numerical
results for three problems of Fisher-type equations with
where 𝑥𝑁,𝑗 and 𝜛𝑁,𝑗 are the nodes and the corresponding
known exact solutions. Finally, a brief conclusion is provided
weights of the Legendre-Gauss-Lobatto quadrature formula
in Section 5.
on the interval (−1, 1), respectively.

2. Legendre Polynomials
3. Legendre Spectral Collocation Method
The Legendre polynomials 𝐿 𝑘 (𝑥) (𝑘 = 0, 1, . . .,) satisfy the
following Rodrigues’ formula: Because of the pseudospectral method is an efficient and
accurate numerical scheme for solving various problems in
(−1)𝑘 𝑘 𝑘 physical space, including variable coefficient and singularity
𝐿 𝑘 (𝑥) = 𝑘
𝐷 ((1 − 𝑥2 ) ) ; (2)
(see, [54, 55]), we propose this method based on Legendre
2 𝑘!
polynomials for approximating the solution of the nonlinear
we recall also that 𝐿 𝑘 (𝑥) is a polynomial of degree 𝑘, and generalized Burger-Fisher model equation and Fisher model
therefore, the 𝑞th derivative of 𝐿 𝑘 (𝑥) is given by with variable coefficient.
𝑘−𝑞
(𝑞)
𝐿 𝑘 (𝑥) = ∑ 𝐶𝑞 (𝑘, 𝑖) 𝐿 𝑖 (𝑥) , (3) 3.1. (1+1)-Dimensional Generalized Burger-Fisher Equation.
𝑖=0(𝑘+𝑖=even) In this subsection, we derive a Legendre pseudospectral
Abstract and Applied Analysis 3

algorithm to solve numerically the generalized Burger-Fisher Subsequently, the second-order spatial partial derivatives
problem: of 𝑢(𝑥, 𝑡) may be written at the same collocation nodes as

𝑁 𝑁
𝑢𝑡 + ]𝑢𝛿 𝑢𝑥 − 𝑢𝑥𝑥 − 𝛾𝑢 (1 − 𝑢𝛿 ) = 0, (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] , 1 󸀠󸀠
𝑢𝑥𝑥 (𝑥𝑛 , 𝑡) = ∑ ( ∑ 𝐿 𝑗 (𝑥𝑖 ) (𝐿 𝑗 (𝑥𝑛 )) 𝜛𝑁,𝑖 ) 𝑢 (𝑥𝑖 , 𝑡)
(11) 𝑖=0 𝑗=0 ℎ𝑗

𝑁
where 𝐷 = {𝑥 : −1 ≤ 𝑥 ≤ 1}. Subject to
= ∑𝐵𝑛𝑖 𝑢 (𝑥𝑖 , 𝑡)
𝑖=0
𝑢 (𝑥, 𝑡) = 𝑔 (𝑡) , 𝑥 = −1, 1, (12)
𝑁
𝑢 (𝑥, 0) = 𝑓 (𝑥) , 𝑥 ∈ 𝐷. (13) = ∑𝐵𝑛𝑖 𝑢𝑖 ,
𝑖=0
(19)
In the following, we shall derive an efficient algorithm for
the numerical solution of (11)–(13). Let the approximation
where
of 𝑢(𝑥, 𝑡) be given in terms of the Legendre polynomials
expansion: 𝑁
1 󸀠󸀠
𝐵𝑛𝑖 = ∑ 𝐿 𝑗 (𝑥𝑖 ) (𝐿 𝑗 (𝑥𝑛 )) 𝜛𝑁,𝑖 . (20)
𝑁 ℎ
𝑗=0 𝑗
𝑇
𝑢 (𝑥, 𝑡) = ∑ 𝑎𝑗 (𝑡) 𝐿 𝑗 (𝑥) , a = (𝑎0 , 𝑎1 , . . . , 𝑎𝑁) . (14)
𝑗=0 In collocation methods, one specifically seeks the approx-
imate solution such that the problem (11) is satisfied exactly
Making use of relations (8) and (10) gives at the Legendre Gauss-Lobatto set of interpolation points 𝑥𝑛 ;
𝑛 = 1, . . . , 𝑁 − 1. The approximation is exact at the 𝑁 −
𝑁 1 collocation points. Therefore, (11) after using relations (17)–
1 𝑁
𝑢 (𝑥, 𝑡) = ∑ ( ∑𝐿 (𝑥 ) 𝐿 (𝑥) 𝜛𝑁,𝑖 𝑢 (𝑥𝑖 , 𝑡)) (15) (20), can be written as
𝑗=0 ℎ𝑗 𝑖=0 𝑗 𝑖 𝑗
𝑁
𝑢𝑛⋅ (𝑡) + ]𝑢𝑛𝛿 (𝑡) ∑𝐴 𝑛𝑖 𝑢𝑖 (𝑡)
or equivalently 𝑖=0

𝑁 (21)
𝑁 𝑁
1 − ∑𝐵𝑛𝑖 𝑢𝑖 (𝑡) − 𝛾𝑢𝑛 (𝑡) (1 − 𝑢𝑛𝛿 (𝑡)) = 0,
𝑢 (𝑥, 𝑡) = ∑ ( ∑ 𝐿 𝑗 (𝑥𝑖 ) 𝐿 𝑗 (𝑥) 𝜛𝑁,𝑖 ) 𝑢 (𝑥𝑖 , 𝑡) . (16)
𝑖=0 𝑗=0 ℎ𝑗
𝑖=0

𝑛 = 1, . . . , 𝑁 − 1,
The Gauss-Lobatto points were introduced by way
of (9). We then saw that the polynomial approximation where 𝑢𝑛 (𝑡) = 𝑢(𝑥𝑛 , 𝑡) and 𝑢𝑛⋅ (𝑡) = 𝜕𝑢𝑛⋅ (𝑡)/𝜕𝑡.
𝑢(𝑥, 𝑡) can be characterized by (𝑁 + 1) nodal values 𝑢(𝑥𝑖 , 𝑡). Now the two values 𝑢0 (𝑡) and 𝑢𝑁(𝑡) can be determined
The approximation of the spatial partial derivatives of first- from the boundary conditions (12), then (21) can be reformu-
order for 𝑢(𝑥, 𝑡) can be computed at the Legendre Gauss- lated as
Lobatto interpolation nodes as
𝑁−1

𝑁 𝑁
𝑢𝑛⋅ (𝑡) + ]𝑢𝑛𝛿 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡)
1 󸀠 𝑖=1
𝑢𝑥 (𝑥𝑛 , 𝑡) = ∑ ( ∑ 𝐿 𝑗 (𝑥𝑖 ) (𝐿 𝑗 (𝑥𝑛 )) 𝜛𝑁,𝑖 ) 𝑢 (𝑥𝑖 , 𝑡)
𝑖=0 ℎ
𝑗=0 𝑗 𝑁−1
(22)
− ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡) + ]𝑢𝑛𝛿 (𝑡) 𝑑𝑛 (𝑡) − 𝑑̃𝑛 (𝑡)
𝑁
𝑖=1
= ∑𝐴 𝑛𝑖 𝑢 (𝑥𝑖 , 𝑡) (17)
𝑖=0 − 𝛾𝑢𝑛 (𝑡) (1 − 𝑢𝑛𝛿 (𝑡)) = 0, 𝑛 = 1, . . . , 𝑁 − 1,
𝑁
= ∑𝐴 𝑛𝑖 𝑢𝑖 (𝑡) , 𝑛 = 0, 1, . . . , 𝑁, where
𝑖=0
𝑑𝑛 (𝑡) = 𝐴 𝑛0 𝑢0 (𝑡) + 𝐴 𝑛𝑁𝑢𝑁(𝑡),
where (23)
𝑑̃𝑛 (𝑡) = 𝐵𝑛0 𝑢0 (𝑡) + 𝐵𝑛𝑁𝑢𝑁(𝑡).
𝑁
1 󸀠
𝐴 𝑛𝑖 = ∑ 𝐿 𝑗 (𝑥𝑖 ) (𝐿 𝑗 (𝑥𝑛 )) 𝜛𝑁,𝑖 , Approximation (22) automatically satisfies the boundary
𝑗=0 ℎ𝑗 (18) conditions (12), but we need an initial condition for each
of the 𝑢𝑛 (𝑡) to integrate (22) in time. The initial condi-
𝑢𝑖 (𝑡) = 𝑢 (𝑥𝑖 , 𝑡) . tion is usually taken to be the interpolant of the initial
4 Abstract and Applied Analysis

function 𝑓(𝑥); that is 𝑢𝑛 (0) = 𝑓(𝑥𝑛 ). Therefore, the approx- Table 1: Absolute errors for Example 1.
imation of (11)–(13) is reduced to the solution of system of
𝑥 𝑡 𝐸(𝑥, 𝑡) 𝑥 𝑡 𝐸(𝑥, 𝑡)
ordinary differential equations in time. Consider
−1 5.56 × 10−11 −1 8.93 × 10−11
𝑁−1 𝑁−1
−0.5 1.46 × 10−8 −0.5 8.23 × 10−9
𝑢𝑛⋅ (𝑡) + ]𝑢𝑛𝛿 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡) − ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡)
0 0.1 1.95 × 10−8 0 0.2 1.41 × 10−8
𝑖=1 𝑖=1
0.5 1.55 × 10−8 0.5 1.17 × 10−8
+ ]𝑢𝑛𝛿 (𝑡) 𝑑𝑛 (𝑡) − 𝑑̃𝑛 (𝑡) − 𝛾𝑢𝑛 (𝑡) (1 − 𝑢𝑛𝛿 (𝑡)) = 0, (24) 1 5.56 × 10−11 1 8.93 × 10−11

𝑛 = 1, . . . , 𝑁 − 1,
which can be written in the matrix form
𝑢𝑛 (0) = 𝑓 (𝑥𝑛 ) .
𝑈⋅ (𝑡) = 𝐹 (𝑡, 𝑢 (𝑡)) ,
Let us denote (30)
𝑇
𝑈 (0) = 𝑓,
𝑈⋅ (𝑡) = [𝑢1⋅ (𝑡) , 𝑢2⋅ (𝑡) , . . . , 𝑢𝑁−1

(𝑡)] ,
where
𝑇
𝑈 (0) = [𝑢1 (0), 𝑢2 (0), . . . , 𝑢𝑁−1 (0)] ,
𝐹𝑛 (𝑡, 𝑢 (𝑡))
𝑇
𝑓 = [𝑓(𝑥1 ), 𝑓(𝑥2 ), . . . , 𝑓(𝑥𝑁−1 )] , 𝑁−1
= 𝑏 (𝑡) ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡) + 𝑏 (𝑡) 𝑑̃𝑛 (𝑡) + 𝑐𝑢𝑛 (𝑡) (1 − 𝑢𝑛 (𝑡)) ,
𝑇
𝐹 (𝑡, 𝑢 (𝑡)) = [𝐹1 (𝑡, 𝑢 (𝑡)) , 𝐹2 (𝑡, 𝑢 (𝑡)) , . . . , 𝐹𝑁−1 (𝑡, 𝑢 (𝑡))] , 𝑖=1

𝑁−1 𝑁−1 𝑛 = 1, . . . , 𝑁 − 1.
𝐹𝑛 (𝑡, 𝑢 (𝑡)) = − ]𝑢𝑛𝛿 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡) + ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡) (31)
𝑖=1 𝑖=1

− ]𝑢𝑛𝛿 (𝑡) 𝑑𝑛 (𝑡) + 𝑑̃𝑛 (𝑡) + 𝛾𝑢𝑛 (𝑡) (1 − 𝑢𝑛𝛿 (𝑡)) , 4. Numerical Examples

𝑛 = 1, . . . , 𝑁 − 1. In this section, three nonlinear time-dependent Fisher-type


equations on finite interval are implemented to demonstrate
(25)
the accuracy and capability of the proposed algorithm, and all
Then (24) can be written in the matrix form of them were performed on the computer using a program
written in Mathematica 8.0. The absolute errors in the
𝑈⋅ (𝑡) = 𝐹 (𝑡, 𝑢 (𝑡)) given tables are 𝐸(𝑥, 𝑡) = |𝑢(𝑥, 𝑡) − 𝑢̃(𝑥, 𝑡) where 𝑢(𝑥, 𝑡)
(26) and 𝑢̃(𝑥, 𝑡) are the exact and numerical solution at selected
𝑈 (0) = 𝑓.
points (𝑥, 𝑡).
This system of ordinary differential equations can be solved
by using four-stage A-stable implicit Runge-Kutta scheme. Example 1. Consider the nonlinear time-dependent one-
dimensional Fisher-type equations
3.2. (1+1)-Dimensional Fisher Equation with Variable Coef- 𝑢𝑡 = 𝑢𝑥𝑥 + 𝑢 (1 − 𝑢) (𝑢 − 𝛾) , (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] , (32)
ficient. In this subsection, we extend the application of the
Legendre pseudospectral method to solve numerically the where 𝐷 = {𝑥 : −1 < 𝑥 < 1}. Subject to
Fisher equation with variable coefficient,
1+𝛾 𝛾−1 𝛾−1 1+𝛾
𝑢𝑡 − 𝑏 (𝑡) 𝑢𝑥𝑥 − 𝑐𝑢 (1 − 𝑢) = 0, (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] , (27) 𝑢 (1, 𝑡) = − tanh [ (1 − 𝑡)] ,
2 2 2√2 √2
subject to the initial-boundary conditions 1+𝛾 𝛾−1 𝛾−1 1+𝛾
𝑢 (−1, 𝑡) = + tanh [ (1 + 𝑡)] , (33)
𝑢 (𝑥, 0) = 𝑓 (𝑥) , 𝑥 ∈ 𝐷, 2 2 2√2 √2
(28)
𝑢 (𝑥, 𝑡) = 𝑔 (𝑡) . 1+𝛾 𝛾−1 𝛾−1
𝑢 (𝑥, 0) = − tanh [ (𝑥)] , 𝑥 ∈ 𝐷.
2 2 2√2
Proceeding as in the previous subsection we can obtain 𝑢𝑥𝑥
in the same form as (19), and then (27) can be collocated in The exact solution is
the Legendre Gauss-Lobatto points as: 1+𝛾 𝛾−1 𝛾−1 1+𝛾
𝑢 (𝑥, 𝑡) = − tanh [ (𝑥 − 𝑡)] . (34)
𝑁−1 2 2 2√2 √2
𝑢𝑛⋅ (𝑡) − 𝑏 (𝑡) ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡) − 𝑏 (𝑡) 𝑑̃𝑛 (𝑡) − 𝑐𝑢𝑛 (𝑡) (1 − 𝑢𝑛 (𝑡))
𝑖=1 In Table 1, we introduce the absolute errors between the
approximate and exact solutions for problem (32) using the
= 0, 𝑛 = 1, . . . , 𝑁 − 1, proposed method for different values of 𝑥 and 𝑡, with 𝛾 =
(29) 10−2 and 𝑁 = 20.
Abstract and Applied Analysis 5

×10−8
4
0.7
1.0 3 1.0
0.6 E
̃
u 2
0.5
1
0.4
0
−1.0 0.5 −1.0 0.5 t
t
−0.5 −0.5

0.0 0.0
x x
0.5 0.5
0.0 0.0
1.0 1.0
(a) (b)

Figure 1: The result of the L-GL-C method at 𝛾 = 10−2 and 𝑁 = 20. (a) The approximate solution. (b) The absolute error.

0.8 Example 2. Consider the nonlinear time-dependent one-


dimensional generalized Burger-Fisher-type equations
0.7

𝑢𝑡 = 𝑢𝑥𝑥 − ]𝑢𝛿 𝑢𝑥 + 𝛾𝑢 (1 − 𝑢𝛿 ) , (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] ,


u and u
̃

0.6
(35)

0.5
where 𝐷 = {𝑥 : −1 < 𝑥 < 1}. Subject to
0.4
−1.0 −0.5 0.0 0.5 1.0
x 𝑢 (1, 𝑡)
u(x, 0) ̃ (x, 0.5)
u
1 1 ]𝛿
̃ (x, 0)
u u(x, 0.9) = ( − tanh [
u(x, 0.5) ̃ (x, 0.9)
u 2 2 2 (𝛿 + 1)
1/𝛿
Figure 2: The curves of approximate solutions and the exact ] 𝛾 (𝛿 + 1)
solutions of problem (32) at 𝑡 = 0.0, 𝑡 = 0.5, and 𝑡 = 0.9 with 𝛾 = × (1 − ( + ) 𝑡)]) ,
𝛿+1 ]
10−2 and 𝑁 = 20.
𝑢 (−1, 𝑡)
1 1 ]𝛿
= ( + tanh [
−2
In case of 𝛾 = 10 and 𝑁 = 20, the approximate solu- 2 2 2 (𝛿 + 1)
tion and absolute errors of problem (32) are displayed in ] 𝛾 (𝛿 + 1) 1/𝛿
Figures 1(a) and 1(b), respectively. In Figure 2, we plotted the × (1 + ( + ) 𝑡)]) ,
curves of approximate solutions and exact solutions of pro- 𝛿+1 ]
blem (32) for different values of 𝑡 (0.0,0.5 and 0.9) with 𝛾 = 1 1 ]𝛿 1/𝛿
10−2 and 𝑁 = 20. It is clear from this figure that approximate 𝑢 (𝑥, 0) = ( − tanh [ 𝑥]) , 𝑥 ∈ 𝐷.
2 2 2 (𝛿 + 1)
solutions and exact solutions completely coincide for the
chosen values of 𝑡. (36)
6 Abstract and Applied Analysis

×10−10
8

1.0 6 1.0
0.502
E 4
̃
u
0.500 2
0
−1.0 0.5 t −1.0 0.5 t
−0.5 −0.5

0.0 0.0
x x
0.5 0.5
1.0 0.0 1.0 0.0
(a) (b)

Figure 3: The result of the L-GL-C method at ] = 𝛾 = 10−2 , 𝛿 = 1, and 𝑁 = 20. (a) The approximate solution. (b)The absolute error.

0.5030 The absolute errors for problem (35) are listed in Table 2
using the L-GL-C method with ] = 𝛾 = 10−2 , 𝑁 = 20, and
0.5025 various choices of 𝛿.
To illustrate the effectiveness of the Legendre pseudospec-
0.5020
u and u
̃

tral method for problem (35), we displayed in Figures 3(a) and


0.5015 3(b) the approximate solution and the absolute error with ] =
𝛾 = 10−2 , 𝛿 = 1, and 𝑁 = 20. The graph of curves of exact
0.5010 and approximate solutions with different values of 𝑡 (0.0,
0.5, and 0.9) is given in Figure 4. Moreover, the approximate
0.5005 solution and the absolute error with ] = 𝛾 = 10−2 , 𝛿 = 2, and
𝑁 = 20 are displayed in Figures 5(a) and 5(b), respectively.
−1.0 −0.5 0.0 0.5 1.0 The curves of exact and approximate solutions of problem
x (35) with 𝛿 = 2 are displayed in Figure 6 with values of
parameters listed in its caption.
u(x, 0) ̃ (x, 0.5)
u
̃ (x, 0)
u u(x, 0.9)
Example 3. Consider the nonlinear time-dependent one-
u(x, 0.5) ̃ (x, 0.9)
u
dimensional Fisher-type equations with variable coefficient
Figure 4: The curves of approximate solutions and the exact
solutions of problem (35) at 𝑡 = 0.0, 𝑡 = 0.5, and 𝑡 = 0.9 with
] = 𝛾 = 10−2 , 𝛿 = 1, and 𝑁 = 20. 𝑎 𝑎
𝑢𝑡 = − coth ( 𝑡 + 𝑐) 𝑢𝑥𝑥 + 𝑎𝑢 (1 − 𝑢) ,
6𝜇2 6 (38)
(𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] ,
The exact solution of (35) is

𝑢 (𝑥, 𝑡) where 𝐷 = {𝑥 : −1 < 𝑥 < 1}. Subject to


1 1 ]𝛿
= ( − tanh (
2 2 2 (𝛿 + 1)
1 𝑎 𝜇 5𝑎
1/𝛿
𝑢 (1, 𝑡) = coth ( 𝑡 + 𝑐) sech2 ( + 𝑡)
] 𝛾 (𝛿 + 1) 4 6 2 12
× (𝑥 − ( + ) 𝑡))) .
𝛿+1 ] 1 𝜇 5𝑎 1
(37) + tanh ( + 𝑡) + ,
2 2 12 2
Abstract and Applied Analysis 7

×10−8

0.710 1.0 2 1.0


E 1
̃ 0.708
u
0.706 0
−1.0 0.5 t −1.0 0.5
t
−0.5 −0.5
0.0 0.0
x x
0.5 0.5
0.0 0.0
1.0 1.0
(a) (b)

Figure 5: The result of the L-GL-C method at ] = 𝛾 = 10−2 , 𝛿 = 2, and 𝑁 = 20. (a) The approximate solution. (b)The absolute error.

Table 2: Absolute errors for Example 2.

𝛿=1 𝛿=2 𝛿=3


𝑥 𝑡 𝐸(𝑥, 𝑡) 𝑥 𝑡 𝐸(𝑥, 𝑡) 𝑥 𝑡 𝐸(𝑥, 𝑡)
−1 8.51 × 10−11 −1 1.26 × 10−10 −1 6.91 × 10−11
−0.5 1.16 × 10−10 −0.5 2.34 × 10−8 −0.5 1.09 × 10−8
0 0.1 5.36 × 10−12 0 0.1 2.61 × 10−8 0 0.1 1.30 × 10−8
0.5 1.53 × 10−11 0.5 2.33 × 10−8 0.5 1.09 × 10−8
1 8.51 × 10−11 1 1.26 × 10−10 1 6.91 × 10−11
−1 8.71 × 10−11 −1 1.22 × 10−10 −1 1.41 × 10−10
−0.5 7.77 × 10−10 −0.5 8.24 × 10−9 −0.5 4.27 × 10−9
0 0.5 6.49 × 10−10 0 0.5 1.16 × 10−8 0 0.5 5.89 × 10−9
−10 −9
0.5 3.81 × 10 0.5 8.26 × 10 0.5 4.28 × 10−9
1 8.71 × 10−11 1 1.22 × 10−10 1 1.41 × 10−10

1 𝑎 𝜇 5𝑎
𝑢 (−1, 𝑡) = coth ( 𝑡 + 𝑐) sech2 (− + 𝑡) The exact solution of (38) is
4 6 2 12
1 𝑎 𝜇𝑥 5𝑎
1 𝜇 5𝑎 1 𝑢 (𝑥, 𝑡) = coth ( 𝑡 + 𝑐) sech2 ( + 𝑡)
+ tanh (− + 𝑡) + , 4 6 2 12
2 2 12 2 (41)
(39) 1 𝜇𝑥 5𝑎 1
+ tanh ( + 𝑡) + .
2 2 12 2
1 𝜇𝑥
𝑢 (𝑥, 0) = coth (𝑐) sech2 ( ) Table 3 lists the absolute errors for problem (38) using
4 2
(40) the L-GL-C method. From numerical results of this table, it
1 𝜇𝑥 1 can be concluded that the numerical solutions are in excellent
+ tanh ( ) + , 𝑥 ∈ 𝐷.
2 2 2 agreement with the exact solutions.
8 Abstract and Applied Analysis

0.711 [3] C. I. Gheorghiu, Spectral Methods for Differential Problems,


T. Popoviciu Institute of Numerical Analysis,, Cluj-Napoca,
Romaina, 2007.
0.710
[4] E. H. Doha, W. M. Abd-Elhameed, and A. H. Bhrawy, “New
spectral-Galerkin algorithms for direct solution of high even-
u and u
̃

0.709 order differential equations using symmetric generalized Jacobi


polynomials,” Collectanea Mathematica, vol. 64, no. 3, pp. 373–
394, 2013.
0.708 [5] E. H. Doha and A. H. Bhrawy, “An efficient direct solver for mul-
tidimensional elliptic Robin boundary value problems using a
Legendre spectral-Galerkin method,” Computers & Mathema-
0.707 tics with Applications, vol. 64, no. 4, pp. 558–571, 2012.
−1.0 −0.5 0.0 0.5 1.0 [6] E. H. Doha, A. H. Bhrawy, D. Baleanu, and S. S. Ezz-Eldien, “On
x shifted Jacobi spectral approximations for solving fractional
differential equations,” Applied Mathematics and Computation,
u(x, 0) ̃ (x, 0.5)
u vol. 219, no. 15, pp. 8042–8056, 2013.
̃ (x, 0)
u u(x, 0.9)
[7] O. R. Isik and M. Sezer, “Bernstein series solution of a class of
u(x, 0.5) ̃ (x, 0.9)
u
Lane-Emden type equations,” Mathematical Problems in Engi-
Figure 6: The curves of approximate solutions and the exact neering, Article ID 423797, 9 pages, 2013.
solutions of problem (35) at 𝑡 = 0.0, 𝑡 = 0.5, and 𝑡 = 0.9 with [8] E. Tohidi, A. H. Bhrawy, and K. Erfani, “A collocation method
] = 𝛾 = 10−2 , 𝛿 = 2, and 𝑁 = 20. based on Bernoulli operational matrix for numerical solution of
generalized pantograph equation,” Applied Mathematical Mod-
elling, vol. 37, no. 6, pp. 4283–4294, 2013.
Table 3: Absolute errors for Example 3. [9] M. S. Mechee and N. Senu, “Numerical study of fractional dif-
ferential equations of Lane-Emden type by method of colloca-
𝑥 𝑡 𝐸(𝑥, 𝑡) 𝑥 𝑡 𝐸(𝑥, 𝑡)
tion,” Applied Mathematics, vol. 3, pp. 851–856, 2012.
−1 9.81 × 10−10 −1 7.05 × 10−9
[10] A. H. Bhrawy and W. M. Abd-Elhameed, “New algorithm for
−0.5 1.00 × 10−7 −0.5 1.93 × 10−6 the numerical solutions of nonlinear third-order differential
0 0.1 1.45 × 10−7 0 0.5 2.84 × 10−6 equations using Jacobi-Gauss collocation method,” Mathemat-
0.5 1.19 × 10−7 0.5 2.12 × 10−6 ical Problems in Engineering, vol. 2011, Article ID 837218, 14
1 9.81 × 10−10 1 7.06 × 10−9 pages, 2011.
[11] A. Ahmadian, M. Suleiman, S. Salahshour, and D. Baleanu, “A
Jacobi operational matrix for solving a fuzzy linear fractional
5. Conclusion differential equation,” Advances in Difference Equations, vol.
2013, article 104, 29 pages, 2013.
In this paper, based on the Legendre-Gauss-Lobatto pseudos- [12] D. Baleanu, A. H. Bhrawy, and T. M. Taha, “Two efficient gen-
pectral approximation we proposed an efficient numerical eralized Laguerre spectral algorithms for fractional initial value
algorithm to solve nonlinear time-dependent Fisher-type problems,” Abstract and Applied Analysis, vol. 2013, Article ID
equations with constant and variable coefficients. The method 546502, 10 pages, 2013.
is based upon reducing the nonlinear partial differential [13] D. Baleanu, A. H. Bhrawy, and T. M. Taha, “A modified Gen-
equation into a system of first-order ordinary differential eralized Laguerre spectral method for fractional differential
equations in the expansion coefficient of the spectral solution. equations on the half line,” Abstract and Applied Analysis, vol.
2013, Article ID 413529, 12 pages, 2013.
Numerical examples were also provided to illustrate the
[14] A. Ahmadian, M. Suleiman, and S. Salahshour, “An opera-
effectiveness of the derived algorithms. The numerical experi-
tional matrix based on Legendre polynomials for solving fuzzy
ments show that the Legendre pseudospectral approximation fractional-order differential equations,” Abstract and Applied
is simple and accurate with a limited number of collocation Analysis, vol. 2013, Article ID 505903, 29 pages, 2013.
nodes. [15] F. Ghaemi, R. Yunus, A. Ahmadian, S. Salahshourd, M.
Suleiman, and S. F. Saleh, “Application of fuzzy fractional kinetic
Conflict of Interests equations to modelling of the acid hydrolysis reaction,” Abstract
and Applied Analysis, vol. 2013, Article ID 610314, 19 pages, 2013.
The authors declare that there is no conflict of interests [16] M. H. Atabakzadeh, M. H. Akrami, and G. H. Erjaee, “Cheby-
regarding the publication of this paper. shev operational matrix method for solving multi-order frac-
tional ordinary differential equations,” Applied Mathematical
References Modelling, vol. 37, no. 20-21, pp. 8903–8911, 2013.
[17] A. H. Bhrawy and M. M. Al-Shomrani, “A shifted Legendre
[1] D. A. Kopriva, Implementing Spectral Methods for Partial Dif- spectral method for fractional-order multi-point boundary
ferential Equations: Algorithms for Scientists and Engineers, value problems,” Advances in Difference Equations, vol. 2012,
Springer, Berlin, Germany, 2009. article 8, 19 pages, 2012.
[2] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, Spec- [18] X. Ma and C. Huang, “Spectral collocation method for linear
tral Methods: Fundamentals in Single Domains, Springer, Berlin, fractional integro-differential equations,” Applied Mathematical
Germany, 2006. Modelling, 2013.
Abstract and Applied Analysis 9

[19] M. R. Eslahchi, M. Dehghan, and M. Parvizi, “Application of Applied Mathematics and Computation, vol. 216, no. 8, pp. 2472–
the collocation method for solving nonlinear fractional integro- 2478, 2010.
differential equations,” Journal of Computational and Applied [39] X. Y. Chen, Numerical methods for the Burgers-Fisher equation
Mathematics, vol. 257, pp. 105–128, 2013. [M.S. thesis], University of Aeronautics and Astronautics, Nan-
[20] Y. Yangy and Y. Huang, “Spectral-collocation methods for jing, China, 2007.
fractional pantograph delay-integrodifferential equations,” [40] R. E. Mickens and A. B. Gumel, “Construction and analysis of
Advances in Mathematical Physics. In press. a non-standard finite difference scheme for the Burgers-Fisher
[21] A. H. Bhrawy and A. S. Alofi, “The operational matrix of frac- equation,” Journal of Sound and Vibration, vol. 257, no. 4, pp.
tional integration for shifted Chebyshev polynomials,” Applied 791–797, 2002.
Mathematics Letters, vol. 26, no. 1, pp. 25–31, 2013. [41] R. E. Mickens, “A best finite-difference scheme for the Fisher
[22] R. A. Fisher, “The wave of advance of advantageous genes,” equation,” Numerical Methods for Partial Differential Equations,
Annals of Eugenics, vol. 7, pp. 335–369, 1937. vol. 10, no. 5, pp. 581–585, 1994.
[23] A. J. Khattak, “A computational meshless method for the gen- [42] R. E. Mickens, “Relation between the time and space step-
eralized Burger’s-Huxley equation,” Applied Mathematical Mod- sizes in nonstandard finite-difference schemes for the Fisher
elling, vol. 33, no. 9, pp. 3718–3729, 2009. equation,” Numerical Methods for Partial Differential Equations,
[24] N. F. Britton, Reactiondiffusion Equations and Their Applications vol. 15, no. 1, pp. 51–55, 1997.
to Biology, Academic Press, London, UK, 1986. [43] N. Parekh and S. Puri, “A new numerical scheme for the Fisher
[25] D. A. Frank, Diffusion and Heat Exchange in Chemical Kinetics, equation,” Journal of Physics A, vol. 23, no. 21, pp. L1085–L1091,
Princeton University Press, Princeton, NJ, USA, 1955. 1990.
[26] A. Wazwaz, “The extended tanh method for abundant solitary [44] R. Rizwan-Uddin, “Comparison of the nodal integral method
wave solutions of nonlinear wave equations,” Applied Mathe- and nonstandard finite-difference schemes for the Fisher equa-
matics and Computation, vol. 187, no. 2, pp. 1131–1142, 2007. tion,” SIAM Journal on Scientific Computing, vol. 22, no. 6, pp.
[27] W. Malfliet, “Solitary wave solutions of nonlinear wave equa- 1926–1942, 2001.
tions,” American Journal of Physics, vol. 60, no. 7, pp. 650–654, [45] R. Chernma, “Exact and numerical solutions of tiie generalized
1992. fisher equation,” Reports on Mathematkxl Physics, vol. 47, pp.
[28] Y. Tan, H. Xu, and S.-J. Liao, “Explicit series solution of travel- 393–411, 2001.
ling waves with a front of Fisher equation,” Chaos, Solitons and [46] A. Şahin, I. Dag, and B. Saka, “A B-spline algorithm for the
Fractals, vol. 31, no. 2, pp. 462–472, 2007. numerical solution of Fisher’s equation,” Kybernetes, vol. 37, no.
[29] H. N. A. Ismail, K. Raslan, and A. A. A. Rabboh, “Ado- 2, pp. 326–342, 2008.
mian decomposition method for Burger’s-Huxley and Burger’s- [47] J. Roessler and H. Hüssner, “Numerical solution of the 1 +
Fisher equations,” Applied Mathematics and Computation, vol. 2 dimensional Fisher’s equation by finite elements and the
159, no. 1, pp. 291–301, 2004. Galerkin method,” Mathematical and Computer Modelling, vol.
[30] J. Canosa, “Diffusion in nonlinear multiplicative media,” Journal 25, no. 3, pp. 57–67, 1997.
of Mathematical Physics, vol. 10, no. 10, pp. 1862–1868, 1969. [48] J. Roessler and H. Hüssner, “Numerical solution of the 1 +
[31] A. Öǧün and C. Kart, “Exact solutions of Fisher and generalized 2 dimensional Fisher’s equation by finite elements and the
Fisher equations with variable coefficients,” Acta Mathematicae Galerkin method,” Mathematical and Computer Modelling, vol.
Applicatae Sinica, vol. 23, no. 4, pp. 563–568, 2007. 25, no. 3, pp. 57–67, 1997.
[32] M. D. Gunzburger, L. S. Hou, and W. Zhu, “Fully discrete finite [49] E. H. Doha, A. H. Bhrawy, R. M. Hafez, and M. A. Abdelkawy,
element approximations of the forced Fisher equation,” Journal “A Chebyshev-Gauss-Radau scheme for nonlinear hyperbolic
of Mathematical Analysis and Applications, vol. 313, no. 2, pp. system of first order,” Applied Mathematics and Information
419–440, 2006. Science, vol. 8, no. 2, pp. 1–10, 2014.
[33] I. Dag, A. Şahin, and A. Korkmaz, “Numerical investigation [50] A. H. Bhrawy, “A Jacobi-Gauss-Lobatto collocation method
of the solution of Fisher’s equation via the B-spline galerkin for solving generalized Fitzhugh-Nagumo equation with time-
method,” Numerical Methods for Partial Differential Equations, dependent coefficients,” Applied Mathematics and Computation,
vol. 26, no. 6, pp. 1483–1503, 2010. vol. 222, pp. 255–264, 2013.
[34] M. Bastani and D. K. Salkuyeh, “A highly accurate method to [51] E. H. Doha, D. Baleanu, A. H. Bhrawy, and M. A. Abdelkawy, “A
solve Fisher’s equation,” Pramana, vol. 78, no. 3, pp. 335–346, Jacobi collocation method for solving nonlinear Burgers’-type
2012. equations,” Abstract and Applied Analysis, vol. 2013, Article ID
[35] R. C. Mittal and R. K. Jain, “Numerical solutions of nonlinear 760542, 12 pages, 2013.
Fisher’s reaction-diffusion equation with modified cubic B- [52] A. H. Bhrawy, L. M. Assas, and M. A. Alghamdi, “Fast spec-
spline collocation method,” Mathematical Sciences, vol. 7, article tral collocation method for solving nonlinear time-delayed Bur-
12, 2013. gers-type equations with positive power terms,” Abstract and
[36] J. Gazdag and J. Canosa, “Numerical solution of Fisher’s equa- Applied Analysis, vol. 2013, Article ID 741278, 12 pages, 2013.
tion,” Journal of Applied Probability, vol. 11, pp. 445–457, 1974. [53] A. H. Bhrawy, L. M. Assas, and M. A. Alghamdi, “An efficient
[37] T. Zhao, C. Li, Z. Zang, and Y. Wu, “Chebyshev-Legendre spectral collocation algorithm for nonlinear Phi-four equa-
pseudo-spectral method for the generalised Burgers-Fisher tions,” Boundary Value Problems, vol. 2013, article 87, 16 pages,
equation,” Applied Mathematical Modelling, vol. 36, no. 3, pp. 2013.
1046–1056, 2012. [54] A. Saadatmandi, “Bernstein operational matrix of fractional
[38] G. Gürarslan, “Numerical modelling of linear and nonlin- derivatives and its applications,” Applied Mathematical Mod-
ear diffusion equations by compact finite difference method,” elling, 2013.
10 Abstract and Applied Analysis

[55] A. H. Bhrawy and A. S. Alofi, “A Jacobi-Gauss collocation


method for solving nonlinear Lane-Emden type equations,”
Communications in Nonlinear Science and Numerical Simula-
tion, vol. 17, no. 1, pp. 62–70, 2012.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 760542, 12 pages
http://dx.doi.org/10.1155/2013/760542

Research Article
A Jacobi Collocation Method for Solving Nonlinear
Burgers-Type Equations

E. H. Doha,1 D. Baleanu,2,3 A. H. Bhrawy,4,5 and M. A. Abdelkawy5


1
Department of Mathematics, Faculty of Science, Cairo University, Giza 12613, Egypt
2
Department of Chemical and Materials Engineering, Faculty of Engineering, King Abdulaziz University, Jeddah 21589, Saudi Arabia
3
Department of Mathematics and Computer Sciences, Faculty of Arts and Sciences, Cankaya University, Eskiehir Yolu 29 Km,
06810 Ankara, Turkey
4
Department of Mathematics, Faculty of Science, King Abdulaziz University, Jeddah 21589, Saudi Arabia
5
Department of Mathematics, Faculty of Science, Beni-Suef University, Beni-Suef 62511, Egypt

Correspondence should be addressed to M. A. Abdelkawy; melkawy@yahoo.com

Received 24 July 2013; Accepted 15 August 2013

Academic Editor: Soheil Salahshour

Copyright © 2013 E. H. Doha et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We solve three versions of nonlinear time-dependent Burgers-type equations. The Jacobi-Gauss-Lobatto points are used as
collocation nodes for spatial derivatives. This approach has the advantage of obtaining the solution in terms of the Jacobi parameters
𝛼 and 𝛽. In addition, the problem is reduced to the solution of the system of ordinary differential equations (SODEs) in time. This
system may be solved by any standard numerical techniques. Numerical solutions obtained by this method when compared with the
exact solutions reveal that the obtained solutions produce high-accurate results. Numerical results show that the proposed method
is of high accuracy and is efficient to solve the Burgers-type equation. Also the results demonstrate that the proposed method is a
powerful algorithm to solve the nonlinear partial differential equations.

1. Introduction an accurate solution of the Lane-Emden-type equation.


Moreover, Doha et al. [11] developed the shifted Jacobi-Gauss
Spectral methods (see, e.g., [1–3] and the references therein) collocation method to solve nonlinear high-order multipoint
are techniques used in applied mathematics and scientific boundary value problems. To the best of our knowledge, there
computing to numerically solve linear and nonlinear dif- are no results on Jacobi-Gauss-Lobatto collocation method
ferential equations. There are three well-known versions of for solving Burgers-type equations arising in mathematical
spectral methods, namely, Galerkin, tau, and collocation physics. This partially motivated our interest in such method.
methods. Spectral collocation method is characterized by For time-dependent partial differential equations, spec-
the fact of providing highly accurate solutions to nonlinear tral methods have been studied in some articles for several
differential equations [3–6]; also it has become increasingly decades. In [12], Ierley et al. investigated spectral methods to
popular for solving fractional differential equations [7–9]. numerically solve time-dependent class of parabolic partial
Bhrawy et al. [5] proposed a new Bernoulli matrix method differential equations subject to periodic boundary condi-
for solving high-order Fredholm integro-differential equa- tions. Tal-Ezer [13, 14] introduced spectral methods using
tions with piecewise intervals. Saadatmandi and Dehghan polynomial approximation of the evolution operator in the
[10] developed the Sinc-collocation approach for solving Chebyshev Least-Squares sense for time-dependent parabolic
multipoint boundary value problems; in this approach the and hyperbolic equations, respectively. Moreover, Coutsias
computation of solution of such problems is reduced to solve et al. [15] developed spectral integration method to solve
some algebraic equations. Bhrawy and Alofi [4] proposed some time-dependent partial differential equations. Zhang
the spectral-shifted Jacobi-Gauss collocation method to find [16] applied the Fourier spectral scheme in spatial together
2 Abstract and Applied Analysis

with the Legendre spectral method to solve time-dependent the system of (𝑁 + 1) ordinary differential equations (ODEs)
partial differential equations and gave error estimates of the in time. This system can be solved by one of the possible
method. Tang and Ma [17] introduced the Legendre spectral methods of numerical analysis such as the Euler method,
method together with the Fourier approximation in spatial Midpoint method, and the Runge-Kutta method. Finally, the
for time-dependent first-order hyperbolic equations with accuracy of the proposed method is demonstrated by test
periodic boundary conditions. Recently, the author of [18] problems.
proposed an accurate numerical algorithm to solve the gen- The remainder of the paper is organized as follows.
eralized Fitzhugh-Nagumo equation with time-dependent In the next section, we introduce some properties of the
coefficients. Jacobi polynomials. In Section 3, the way of constructing
In [20], Bateman introduced the one-dimensional quasi- the Gauss-Lobatto collocation technique for nonlinear time-
linear parabolic partial differential equation, while Burgers dependent Burgers-type equations is described using the
[21] developed it as mathematical modeling of turbulence, Jacobi polynomials, and in Section 4 the proposed method
and it is referred as one-dimensional Burgers’ equation. is applied to three problems of nonlinear time-dependent
Many authors gave different solutions for Burgers’ equation Burgers-type equations. Finally, some concluding remarks
by using various methods. Kadalbajoo and Awasthi [22] are given in Section 5.
and Gülsu [23] used a finite-difference approach method to
find solutions of one-dimensional Burgers’ equation. Crank-
2. Some Properties of Jacobi Polynomials
Nicolson scheme for Burgers’ equation is developed by Kim,
[24]. Nguyen and Reynen [25, 26], Gardner et al. [27, 28] (𝛼,𝛽)
The standard Jacobi polynomials of degree 𝑘 (𝑃𝑘 (𝑥), 𝑘 =
and Kutluay et al. [29] used methods based on the Petrov- 0, 1, . . .) with the parameters 𝛼 > −1, 𝛽 > −1 are satisfying
Galerkin, Least-Squares finite-elements, and B-spline finite the following relations:
element methods to solve Burgers’ equation. A method
based on collocation of modified cubic B-splines over finite (𝛼,𝛽) (𝛼,𝛽)
𝑃𝑘 (−𝑥) = (−1)𝑘 𝑃𝑘 (𝑥) ,
elements has been investigated by Mittal and Jain in [30].
In this work, we propose a J-GL-C method to numerically
(𝛼,𝛽) (−1)𝑘 Γ (𝑘 + 𝛽 + 1)
solve the following three nonlinear time-dependent Burgers’- 𝑃𝑘 (−1) = , (4)
type equations: 𝑘!Γ (𝛽 + 1)

(1) time-dependent 1D Burgers’ equation: (𝛼,𝛽) Γ (𝑘 + 𝛼 + 1)


𝑃𝑘 (1) = .
𝑘!Γ (𝛼 + 1)
𝑢𝑡 + ]𝑢𝑢𝑥 − 𝜇𝑢𝑥𝑥 = 0; (𝑥, 𝑡) ∈ [𝐴, 𝐵] × [0, 𝑇] , (1)
Let 𝑤(𝛼,𝛽) (𝑥) = (1 − 𝑥)𝛼 (1 + 𝑥)𝛽 ; then we define the weighted
(2) time-dependent 1D generalized Burger-Fisher equa- space 𝐿2𝑤(𝛼,𝛽) as usual, equipped with the following inner
tion: product and norm:

𝑢𝑡 = 𝑢𝑥𝑥 − ]𝑢𝛿 𝑢𝑥 + 𝛾𝑢 (1 − 𝑢𝛿 ) ; (𝑥, 𝑡) ∈ [𝐴, 𝐵] × [0, 𝑇] , 1


(𝑢, V)𝑤(𝛼,𝛽) = ∫ 𝑢 (𝑥) V (𝑥) 𝑤(𝛼,𝛽) (𝑥) 𝑑𝑥,
(2) −1 (5)
(3) time-dependent 1D generalized Burgers-Huxley ‖𝑢‖𝑤(𝛼,𝛽) = (𝑢, 𝑢)1/2
𝑤(𝛼,𝛽)
.
equation:
The set the of Jacobi polynomials forms a complete 𝐿2𝑤(𝛼,𝛽) -
𝛿 𝛿 𝛿
𝑢𝑡 + ]𝑢 𝑢𝑥 − 𝑢𝑥𝑥 − 𝜂𝑢 (1 − 𝑢 ) (𝑢 − 𝛾) = 0; orthogonal system, and
(3) 󵄩󵄩 (𝛼,𝛽) 󵄩󵄩
(𝑥, 𝑡) ∈ [𝐴, 𝐵] × [0, 𝑇] . 󵄩󵄩𝑃 󵄩
󵄩󵄩 𝑘 󵄩󵄩󵄩𝑤(𝛼,𝛽)
(6)
In order to obtain the solution in terms of the Jacobi 2𝛼+𝛽+1 Γ (𝑘 + 𝛼 + 1) Γ (𝑘 + 𝛽 + 1)
parameters 𝛼 and 𝛽, the use of the Jacobi polynomials for = ℎ𝑘 = .
(2𝑘 + 𝛼 + 𝛽 + 1) Γ (𝑘 + 1) Γ (𝑘 + 𝛼 + 𝛽 + 1)
solving differential equations has gained increasing popu-
larity in recent years (see, [31–35]). The main concern of Let 𝑆𝑁(−1, 1) be the set of polynomials of degree at most
this paper is to extend the application of J-GL-C method 𝑁, and due to the property of the standard Jacobi-Gauss
to solve the three nonlinear time-dependent Burgers-type quadrature, it follows that for any 𝜙 ∈ 𝑆2𝑁+1 (−1, 1),
equations. It would be very useful to carry out a systematic
study on J-GL-C method with general indexes (𝛼, 𝛽 > 1 𝑁
(𝛼,𝛽) (𝛼,𝛽)
−1). The nonlinear time-dependent Burgers’-type equation is ∫ 𝑤(𝛼,𝛽) (𝑥) 𝜙 (𝑥) 𝑑𝑥 = ∑𝜛𝑁,𝑗 𝜙 (𝑥𝑁,𝑗 ) , (7)
−1 𝑗=0
collocated only for the space variable at (𝑁 − 1) points, and
for suitable collocation points, we use the (𝑁 − 1) nodes of (𝛼,𝛽) (𝛼,𝛽)
the Jacobi-Gauss-Lobatto interpolation which depends upon where 𝑥𝑁,𝑗 (0 ≤ 𝑗 ≤ 𝑁) and 𝜛𝑁,𝑗 (0 ≤ 𝑗 ≤ 𝑁)
the two general parameters (𝛼, 𝛽 > −1); these equations are the nodes and the corresponding Christoffel numbers of
together with the two-point boundary conditions constitute the Jacobi-Gauss-quadrature formula on the interval (−1, 1),
Abstract and Applied Analysis 3

respectively. Now, we introduce the following discrete inner and accordingly, (14) takes the form
product and norm:
𝑢 (𝑥, 𝑡)
𝑁
(𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
(𝑢, V)𝑤(𝛼,𝛽) = ∑ 𝑢 (𝑥𝑁,𝑗 ) V (𝑥𝑁,𝑗 ) 𝜛𝑁,𝑗 , 𝑁
1 𝑁 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
𝑗=0 (8) = ∑( ∑𝑃 (𝑥𝑁,𝑖 ) 𝜛𝑁,𝑖 𝑢 (𝑥𝑁,𝑖 , 𝑡)) 𝑃𝑗 (𝑥) ,
𝑗=0 ℎ𝑗 𝑖=0 𝑗
‖𝑢‖𝑤(𝛼,𝛽) = (𝑢, 𝑢)1/2
𝑤(𝛼,𝛽)
. (15)

For 𝛼 = 𝛽, one recovers the ultraspherical polynomials or equivalently takes the form
(symmetric Jacobi polynomials) and for 𝛼 = 𝛽 = ∓1/2, 𝛼 =
𝛽 = 0, the Chebyshev of the first and second kinds 𝑢 (𝑥, 𝑡)
and the Legendre polynomials, respectively; and for the
nonsymmetric Jacobi polynomials, the two important special 𝑁 𝑁
1 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
cases 𝛼 = −𝛽 = ±1/2 (the Chebyshev polynomials of the third = ∑ (∑ 𝑃𝑗 (𝑥𝑁,𝑖 ) 𝑃𝑗 (𝑥) 𝜛𝑁,𝑖 ) 𝑢 (𝑥𝑁,𝑖 , 𝑡) .
𝑖=0 ℎ
𝑗=0 𝑗
and fourth kinds) are also recovered.
(16)
3. Jacobi Spectral Collocation Method The spatial partial derivatives with respect to 𝑥 in (9) can be
computed at the J-GL-C points to give
Since the collocation method approximates the differential
equations in physical space, it is very easy to implement and (𝛼,𝛽)
be adaptable to various problems, including variable coeffi- 𝑢𝑥 (𝑥𝑁,𝑛 , 𝑡)
cient and nonlinear differential equations (see, for instance 𝑁 𝑁
[4, 6]). In this section, we develop the J-GL-C method to 1 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) 󸀠 (𝛼,𝛽) (𝛼,𝛽)
= ∑(∑ 𝑃𝑗 (𝑥𝑁,𝑖 ) (𝑃𝑗 (𝑥𝑁,𝑛 )) 𝜛𝑁,𝑖 )𝑢(𝑥𝑁,𝑖 , 𝑡)
numerically solve the Burgers-type equations. ℎ
𝑖=0 𝑗=0 𝑗

𝑁
3.1. (1 + 1)-Dimensional Burgers’ Equation. In 1939, Burg- (𝛼,𝛽)
ers has simplified the Navier-Stokes equation by dropping = ∑𝐴 𝑛𝑖 𝑢 (𝑥𝑁,𝑖 , 𝑡) , 𝑛 = 0, 1, . . . , 𝑁,
𝑖=0
the pressure term to obtain his one-dimensional Burgers’
equation. This equation has many applications in applied (𝛼,𝛽)
𝑢𝑥𝑥 (𝑥𝑁,𝑛 , 𝑡)
mathematics, such as modeling of gas dynamics [36, 37],
modeling of fluid dynamics, turbulence, boundary layer 𝑁 𝑁
1 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) 󸀠󸀠 (𝛼,𝛽) (𝛼,𝛽)
behavior, shock wave formation, and traffic flow [38]. In this = ∑ (∑ 𝑃𝑗 (𝑥𝑁,𝑖 ) (𝑃𝑗 (𝑥𝑁,𝑛 )) 𝜛𝑁,𝑖 )𝑢(𝑥𝑁,𝑖 , 𝑡)
subsection, we derive a J-GL-C method to solve numerically 𝑖=0 ℎ
𝑗=0 𝑗
the (1 + 1)-dimensional Burgers’ model problem:
𝑁
(𝛼,𝛽)
𝑢𝑡 + ]𝑢𝑢𝑥 − 𝜇𝑢𝑥𝑥 = 0; (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] , (9) = ∑𝐵𝑛𝑖 𝑢(𝑥𝑁,𝑖 , 𝑡) ,
𝑖=0
(17)
where
where
𝐷 = {𝑥 : −1 ≤ 𝑥 ≤ 1} , (10)
𝑁
1 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) 󸀠 (𝛼,𝛽)
subject to the boundary conditions 𝐴 𝑛𝑖 = ∑ 𝑃𝑗 (𝑥𝑁,𝑖 ) (𝑃𝑗 (𝑥𝑁,𝑛 )) 𝜛𝑁,𝑖 ,
𝑗=0 ℎ𝑗
𝑢 (−1, 𝑡) = 𝑔1 (𝑡) , 𝑢 (1, 𝑡) = 𝑔2 (𝑡) , 𝑡 ∈ [0, 𝑇] , (11) (18)
𝑁
1 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) 󸀠󸀠 (𝛼,𝛽)
and the initial condition 𝐵𝑛𝑖 = ∑ 𝑃𝑗 (𝑥𝑁,𝑖 ) (𝑃𝑗 (𝑥𝑁,𝑛 )) 𝜛𝑁,𝑖 .
𝑗=0 ℎ𝑗
𝑢 (𝑥, 0) = 𝑓 (𝑥) , 𝑥 ∈ 𝐷. (12)
Making use of (17) and (18) enables one to rewrite (9) in the
Now we assume that form:

𝑁 𝑁 𝑁

𝑢 (𝑥, 𝑡) = ∑𝑎𝑗 (𝑡) 𝑃𝑗


(𝛼,𝛽)
(𝑥) , (13) 𝑢̇𝑛 (𝑡) + ]𝑢𝑛 (𝑡) ∑𝐴 𝑛𝑖 𝑢𝑖 (𝑡) − 𝜇∑𝐵𝑛𝑖 𝑢𝑖 (𝑡) = 0,
𝑗=0 𝑖=0 𝑖=0 (19)
𝑛 = 1, . . . , 𝑁 − 1,
and if we make use of (6)–(8), then we find
where
1 𝑁 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
𝑎𝑗 (𝑡) = ∑𝑃𝑗 (𝑥𝑁,𝑖 ) 𝜛𝑁,𝑖 𝑢 (𝑥𝑁,𝑖 , 𝑡) , (14) (𝛼,𝛽)
ℎ𝑗 𝑖=0 𝑢𝑛 (𝑡) = 𝑢 (𝑥𝑁,𝑛 , 𝑡) . (20)
4 Abstract and Applied Analysis

Using Equation (19) and using the two-point boundary and physics applications, gas dynamic, and traffic flow. The
conditions (11) generate a system of (𝑁 − 1) ODEs in time: Burger-Fisher equation can be written in the following form:
𝑁−1 𝑁−1
𝑢𝑡 = 𝑢𝑥𝑥 − ]𝑢𝑢𝑥 + 𝛾𝑢 (1 − 𝑢) ; (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] , (26)
𝑢̇𝑛 (𝑡) + ]𝑢𝑛 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡) − 𝜇 ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡)
𝑖=1 𝑖=1 (21) where
+]𝑑𝑛 (𝑡) − 𝜇𝑑̃𝑛 (𝑡) = 0, 𝑛 = 1, . . . , 𝑁 − 1,
𝐷 = {𝑥 : −1 < 𝑥 < 1} , (27)
where
subject to the boundary conditions
𝑑𝑛 (𝑡) = 𝐴 𝑛𝑜 𝑔1 (𝑡) + 𝐴 𝑛𝑁𝑔2 (𝑡) ,
(22) 𝑢 (−1, 𝑡) = 𝑔1 (𝑡) , 𝑢 (1, 𝑡) = 𝑔2 (𝑡) , (28)
𝑑̃𝑛 (𝑡) = 𝐵𝑛𝑜 𝑔1 (𝑡) + 𝐵𝑛𝑁𝑔2 (𝑡) .
and the initial condition
Then the problem (9)–(12) transforms to the SODEs:
𝑢 (𝑥, 0) = 𝑓 (𝑥) , 𝑥 ∈ 𝐷. (29)
𝑁−1 𝑁−1
𝑢̇𝑛 (𝑡) + ]𝑢𝑛 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡) − 𝜇 ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡)
𝑖=1 𝑖=1
The same procedure of Section 3.1 can be used to reduce
(23) (26)–(29) to the system of nonlinear differential equations
+ ]𝑑𝑛 (𝑡) − 𝜇𝑑̃𝑛 (𝑡) = 0, 𝑛 = 1, . . . , 𝑁 − 1, in the unknown expansion coefficients of the sought-for
semianalytical solution. This system is solved by using the
(𝛼,𝛽)
𝑢𝑛 (0) = 𝑓 (𝑥𝑁,𝑛 ) , implicit Runge-Kutta method.

which may be written in the following matrix form: 3.3. (1 + 1)-Dimensional Generalized Burgers-Huxley Equa-
tion. The Huxley equation is a nonlinear partial differential
u̇(𝑡) = F (𝑡, 𝑢 (𝑡)) , equation of second order of the form
(24)
u (0) = f,
𝑢𝑡 − 𝑢𝑥𝑥 − 𝑢 (𝑘 − 𝑢) (𝑢 − 1) = 0; 𝑘 ≠
0. (30)
where
It is an evolution equation that describes the nerve propaga-
𝑇
u̇(𝑡) = [𝑢̇1 (𝑡) , 𝑢̇2 (𝑡) , . . . , 𝑢̇𝑁−1 (𝑡)] , tion [47] in biology from which molecular CB properties can
be calculated. It also gives a phenomenological description of
𝑇
f = [𝑓 (𝑥𝑁,1 ) , 𝑓 (𝑥𝑁,2 ) , . . . , 𝑓 (𝑥𝑁,𝑁−1 )] , the behavior of the myosin heads II. In addition to this non-
linear evolution equation, combined forms of this equation
𝑇
F (𝑡, 𝑢 (𝑡)) = [𝐹1 (𝑡, 𝑢 (𝑡)) , 𝐹2 (𝑡, 𝑢 (𝑡)) , . . . , 𝐹𝑁−1 (𝑡, 𝑢 (𝑡))] , and Burgers’ equation will be investigated. It is interesting to
point out that this equation includes the convection term 𝑢𝑥
𝑁−1 𝑁−1
and the dissipation term 𝑢𝑥𝑥 in addition to other terms. In this
𝐹𝑛 (𝑡, 𝑢 (𝑡)) = −]𝑢𝑛 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡) + 𝜇 ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡) subsection, we derive J-GL-C method to solve numerically
𝑖=1 𝑖=1
the (1+1)-dimensional generalized Burgers-Huxley equation:
− ]𝑑𝑛 (𝑡) + 𝜇𝑑̃𝑛 (𝑡) , 𝑛 = 1, . . . , 𝑁 − 1.
(25) 𝑢𝑡 + ]𝑢𝛿 𝑢𝑥 − 𝑢𝑥𝑥 − 𝜂𝑢 (1 − 𝑢𝛿 ) (𝑢𝛿 − 𝛾) = 0,
(31)
The SODEs (24) in time may be solved using any standard (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] ,
technique, like the implicit Runge-Kutta method.
where
3.2. (1 + 1)-Dimensional Burger-Fisher Equation. The Burger-
Fisher equation is a combined form of Fisher and Burgers’ 𝐷 = {𝑥 : −1 < 𝑥 < 1} , (32)
equations. The Fisher equation was firstly introduced by
Fisher in [39] to describe the propagation of a mutant subject to the boundary conditions:
gene. This equation has a wide range of applications in a
large number of the fields of chemical kinetics [40], logistic 𝑢 (−1, 𝑡) = 𝑔1 (𝑡) , 𝑢 (1, 𝑡) = 𝑔2 (𝑡) , (33)
population growth [41], flame propagation [42], population
in one-dimensional habitual [43], neutron population in a and the initial condition:
nuclear reaction [44], neurophysiology [45], autocatalytic
chemical reactions [19], branching the Brownian motion 𝑢 (𝑥, 0) = 𝑓 (𝑥) , 𝑥 ∈ 𝐷. (34)
processes [40], and nuclear reactor theory [46]. Moreover,
the Burger-Fisher equation has a wide range of applications in The same procedure of Sections 3.1 and 3.2 is used to solve
various fields of financial mathematics, applied mathematics numerically (30)–(34).
Abstract and Applied Analysis 5

4. Numerical Results and the initial condition:


To illustrate the effectiveness of the proposed method in the 𝑢 (𝑥, 0)
present paper, three test examples are carried out in this
section. The comparison of the results obtained by various 1/𝛿

[ −]𝛿 + 𝛿√] + 4𝜂 (1 + 𝛿) ]]
2
choices of the Jacobi parameters 𝛼 and 𝛽 reveals that the [𝛾 𝛾
present method is very effective and convenient for all choices = [ + tanh [𝑥𝛾 ]] ,
2 2 4 (𝛿 + 1)
of 𝛼 and 𝛽. We consider the following three examples. [ [ ]]

Example 1. Consider the nonlinear time-dependent one- 𝑥 ∈ [𝐴, 𝐵] .


dimensional generalized Burgers-Huxley equation:
(37)

The exact solution of (35) is


𝑢𝑡 = 𝑢𝑥𝑥 − ]𝑢𝛿 𝑢𝑥 + 𝜂𝑢 (1 − 𝑢𝛿 ) (𝑢𝛿 − 𝛾) ; 𝑢 (𝑥, 𝑡)
(35)
(𝑥, 𝑡) ∈ [𝐴, 𝐵] × [0, 𝑇] ,
[𝛾 𝛾
=[ +
2 2
[
subject to the boundary conditions: 2
[ −]𝛿 + 𝛿√] + 4𝜂 (1 + 𝛿)
× tanh [𝛾
4 (𝛿 + 1)
[
𝑢 (𝐴, 𝑡) 1/𝛿
]𝛾 (1+𝛿−𝛾)(√]2 +4𝜂(1+𝛿)−])
]]
× (𝑥− 𝑡)]] .
[𝛾 𝛾 2(𝛿+1)2
=[ + ]]
2 2 (38)
[
The difference between the measured value of the approx-
[ −]𝛿 + 𝛿√] + 4𝜂 (1 + 𝛿)
2
imate solution and its actual value (absolute error), given by
× tanh [𝛾
4 (𝛿 + 1)
[ 𝐸 (𝑥, 𝑡) = |𝑢 (𝑥, 𝑡) − 𝑢̃ (𝑥, 𝑡)| , (39)

× (𝐴 − (]𝛾 (1+𝛿−𝛾) (√]2 +4𝜂 (1 + 𝛿)−])) where 𝑢(𝑥, 𝑡) and 𝑢̃(𝑥, 𝑡), is the exact solution and the
approximate solution at the point (𝑥, 𝑡), respectively.
1/𝛿 In the cases of 𝛾 = 10−3 , ] = 𝜂 = 𝛿 = 1, and 𝑁 = 4,
−1 ]] Table 1 lists the comparison of absolute errors of problem
× (2(𝛿 + 1)2 ) 𝑡) ]] ,
(35) subject to (36) and (37) using the J-GL-C method for
]] different choices of 𝛼 and 𝛽 with references [19], in the
interval [0, 1]. Moreover in Tables 2 and 3, the absolute errors
𝑢 (𝐵, 𝑡)
of this problem with 𝛼 = 𝛽 = 1/2 and various choices of
𝑥, 𝑡 for 𝛿 = 1 (3), in both intervals [0, 1] and [−1, 1], are
[𝛾 𝛾 given, respectively. In Table 4, maximum absolute errors with
=[ +
2 2 various choices of (𝛼, 𝛽) for both values of 𝛿 = 1, 3 are given
[ where ] = 𝛾 = 𝜂 = 0.001, in both intervals [0, 1] and [−1, 1].
Moreover, the absolute errors of problem (35) are shown in
[ −]𝛿 + 𝛿√] + 4𝜂 (1 + 𝛿)
2 Figures 1, 2, and 3 for 𝛿 = 1, 2, and 3 with values of parameters
× tanh [𝛾 listed in their captions, respectively, while in Figure 4, we
4 (𝛿 + 1)
[ plotted the approximate solution of this problem where 𝛼 = 0,
𝛽 = 1, ] = 𝜂 = 𝛾 = 10−3 , and 𝑁 = 12 for 𝛿 = 1. These
× (𝐵 − (]𝛾 (1+𝛿−𝛾) (√]2 +4𝜂 (1+𝛿)−])) figures demonstrate the good accuracy of this algorithm for
all choices of 𝛼, 𝛽, and 𝑁 and moreover in any interval.
1/𝛿
Example 2. Consider the nonlinear time-dependent one
2 −1 ]]
× (2(𝛿 + 1) ) 𝑡) ]] , dimensional Burgers-type equation:
]]
(36) 𝑢𝑡 + ]𝑢𝑢𝑥 − 𝜇𝑢𝑥𝑥 = 0; (𝑥, 𝑡) ∈ [𝐴, 𝐵] × [0, 𝑇] , (40)
6 Abstract and Applied Analysis

Table 1: Comparison of absolute errors of Example 1 with results from different articles, where 𝑁 = 4, 𝛾 = 10−3 , and ] = 𝜂 = 𝛿 = 1.

Our method for various values of (𝛼, 𝛽) with 𝑁 = 4


(𝑥, 𝑡) ADM [19] (0, 0) (−1/2, −1/2) (1/2, 1/2) (−1/2, 1/2) (0, 1)
(0.1, 0.05) 1.94 × 10−7 2.99 × 10−8 1.28 × 10−9 2.32 × 10−8 1.28 × 10−9 1.34 × 10−8
(0.1, 0.1) 3.87 × 10−7 5.98 × 10−8 2.56 × 10−9 4.63 × 10−8 2.56 × 10−9 2.68 × 10−8
(0.1, 1) 38.75 × 10−7 5.97 × 10−7 2.46 × 10−8 4.63 × 10−7 2.46 × 10−8 2.68 × 10−7
(0.5, 0.05) 1.94 × 10−7 1.02 × 10−8 5.26 × 10−8 1.06 × 10−8 5.26 × 10−8 3.64 × 10−8
(0.5, 0.1) 3.87 × 10−7 2.03 × 10−8 1.05 × 10−7 2.13 × 10−8 1.05 × 10−7 7.28 × 10−8
(0.5, 1) 38.57 × 10−7 2.04 × 10−7 10.54 × 10−7 2.13 × 10−7 10.54 × 10−7 7.28 × 10−7
(0.9, 0.05) 1.94 × 10−7 7.93 × 10−9 5.22 × 10−8 4.49 × 10−9 5.22 × 10−8 3.07 × 10−8
(0.9, 0.1) 3.87 × 10−7 1.59 × 10−8 1.04 × 10−7 8.99 × 10−9 1.04 × 10−7 6.15 × 10−8
(0.9, 1) 38.76 × 10−7 1.59 × 10−7 10.46 × 10−7 9.01 × 10−8 10.46 × 10−7 6.14 × 10−7

Table 2: Absolute errors with 𝛼 = 𝛽 = 1/2, 𝛿 = 1 and various choices of 𝑥, 𝑡 for Example 1.

𝑥 𝑡 𝐴 𝐵 ] 𝛾 𝜂 𝑁 𝐸 𝐴 𝐵 𝐸
0.0 0.1 0 1 0.001 0.001 0.001 12 7.04 × 10−12 −1 1 2.47 × 10−11
0.1 5.61 × 10−11 4.19 × 10−12
0.2 7.29 × 10−11 7.01 × 10−12
0.3 8.26 × 10−12 4.59 × 10−11
0.4 4.37 × 10−11 5.74 × 10−11
0.5 2.47 × 10−11 1.44 × 10−11
0.6 7.01 × 10−12 2.16 × 10−11
0.7 5.74 × 10−11 3.74 × 10−11
0.8 2.15 × 10−11 1.03 × 10−10
0.9 1.03 × 10−10 1.25 × 10−10
1 4.64 × 10−12 4.64 × 10−12
0.0 0.2 0 1 0.001 0.001 0.001 12 1.33 × 10−11 −1 1 4.92 × 10−11
0.1 1.11 × 10−10 8.33 × 10−12
0.2 1.46 × 10−10 1.42 × 10−11
0.3 1.65 × 10−11 9.20 × 10−11
0.4 8.74 × 10−11 1.15 × 10−10
0.5 4.92 × 10−11 2.87 × 10−11
0.6 1.42 × 10−11 4.30 × 10−11
0.7 1.14 × 10−10 7.49 × 10−11
0.8 4.30 × 10−11 2.05 × 10−10
0.9 2.05 × 10−10 2.51 × 10−10
1 1.09 × 10−11 1.09 × 10−11

subject to the boundary conditions If we apply the generalized tanh method [48], then we find
that the analytical solution of (40) is
𝑐 𝑐 𝑐 𝑐 𝑐 𝑐
𝑢 (𝐴, 𝑡) = − tanh [ (𝐴 − 𝑐𝑡)] , 𝑢 (𝑥, 𝑡) = − tanh [ (𝑥 − 𝑐𝑡)] . (43)
] ] 2𝜇 ] ] 2𝜇
(41)
𝑐 𝑐 𝑐 In Table 5, the maximum absolute errors of (40) subject to
𝑢 (𝐵, 𝑡) = − tanh [ (𝐵 − 𝑐𝑡)] ,
] ] 2𝜇 (41) and (42) are introduced using the J-GL-C method, with
various choices of (𝛼, 𝛽) in both intervals [0, 1] and [−1, 1].
and the initial condition Absolute errors between exact and numerical solutions of this
problem are introduced in Table 6 using the J-GL-C method
for 𝛼 = 𝛽 = 1/2 with 𝑁 = 20, and ] = 10, 𝜇 = 0.1 and
𝑐 𝑐 𝑐 𝑐 = 0.1 in both intervals [0, 1] and [−1, 1]. In Figures 5, 6,
𝑢 (𝑥, 0) = − tanh [ 𝑥] , 𝑥 ∈ [𝐴, 𝐵] . (42)
] ] 2𝜇 and 7, we displayed the absolute errors of problem (40) for
Abstract and Applied Analysis 7

Table 3: Absolute errors with 𝛼 = 𝛽 = 1/2, 𝛿 = 3 and various choices of 𝑥, 𝑡 for Example 1.

𝑥 𝑡 𝐴 𝐵 ] 𝛾 𝜂 𝑁 𝐸 𝐴 𝐵 𝐸
0.0 0.1 0 1 0.001 0.001 0.001 12 5.62 × 10−8 −1 1 8.32 × 10−9
0.1 6.50 × 10−6 2.99 × 10−6
0.2 4.67 × 10−6 1.65 × 10−6
0.3 3.08 × 10−6 2.44 × 10−6
0.4 1.65 × 10−6 3.09 × 10−6
0.5 8.33 × 10−9 1.97 × 10−6
0.6 1.65 × 10−6 4.67 × 10−6
0.7 3.09 × 10−6 3.79 × 10−6
0.8 4.67 × 10−6 6.53 × 10−6
0.9 6.53 × 10−6 35.32 × 10−6
1 5.66 × 10−8 5.66 × 10−8
0.0 0.2 0 1 0.001 0.001 0.001 12 2.27 × 10−7 −1 1 2.54 × 10−8
0.1 12.98 × 10−6 5.97 × 10−6
0.2 9.32 × 10−6 3.32 × 10−6
0.3 6.16 × 10−6 4.89 × 10−6
0.4 3.30 × 10−6 6.21 × 10−6
0.5 2.54 × 10−8 3.93 × 10−6
0.6 3.32 × 10−6 9.36 × 10−6
0.7 6.21 × 10−6 7.58 × 10−6
0.8 9.36 × 10−6 13.10 × 10−6
0.9 13.10 × 10−6 70.74 × 10−6
1 2.27 × 10−7 2.27 × 10−7

Table 4: Maximum absolute errors with various choices of (𝛼, 𝛽) for both values of 𝛿 = 1, 3 Example 3.

𝛼 𝛽 𝐴 𝐵 ] 𝛾 𝜂 𝛿 𝑁 𝑀𝐸 𝐴 𝐵 𝑀𝐸

0 0 0 1 0.001 0.001 0.001 1 12 1.39 × 10−9 −1 1 2.55 × 10−9

1/2 1/2 6.38 × 10−10 1.72 × 10−9

−1/2 −1/2 3.83 × 10−9 4.65 × 10−9

−1/2 1/2 5.04 × 10−9 4.58 × 10−9

0 1 2.16 × 10−9 2.42 × 10−9

0 0 0 1 0.001 0.001 0.001 3 12 1.84 × 10−4 −1 1 14.75 × 10−4

1/2 1/2 2.23 × 10−5 4.13 × 10−4

−1/2 −1/2 33.44 × 10−4 81.17 × 10−4

−1/2 1/2 56.96 × 10−4 90.97 × 10−4

0 1 5.24 × 10−4 16.46 × 10−4

different numbers of collation points and different choices Example 3. Consider the nonlinear time-dependent one-
of 𝛼 and 𝛽 in interval [0, 1] with values of parameters being dimensional generalized Burger-Fisher-type equation:
listed in their captions. Moreover, in Figure 8, we see that,
the approximate solution and the exact solution are almost
coincided for different values of 𝑡 (0, 0.5 and 0.9) of problem
(40) where ] = 10, 𝜇 = 0.1, 𝑐 = 0.1, 𝛼 = 𝛽 = −0.5, and 𝑁 = 20 𝑢𝑡 = 𝑢𝑥𝑥 − ]𝑢𝛿 𝑢𝑥 + 𝛾𝑢 (1 − 𝑢𝛿 ) ; (𝑥, 𝑡) ∈ [𝐴, 𝐵] × [0, 𝑇] ,
in interval [−1, 1]. (44)
8 Abstract and Applied Analysis

×10−7 0.00050005
8
10

̃ (x, t)
6 1.0 0.00050000
E 4

u
0.00049995
2
0 −10 5 t
0.0 0.5 t −5
0
x
0.5 5
x 10
0

0.0
1.0 Figure 4: The approximate solution of problem (35) where 𝛼 = 0,
𝛽 = 1, ] = 𝜂 = 𝛾 = 10−3 , and 𝑁 = 12 for 𝛿 = 1.
Figure 1: The absolute error of problem (35) where 𝛼 = 𝛽 = 0,
] = 𝜂 = 𝛾 = 10−3 , and 𝑁 = 4 for 𝛿 = 1.

×10−9
−7 1.5
×10
4 1.0
1
3 1.0 E
5
E 2
1 0
0.0 0.5
0 t
0.5 t
0.2
0.5
0.4 x
x 0.6
0.0
0.8 1.0
0.0
Figure 5: The absolute error of problem (40) where ] = 10, 𝜇 = 0.1,
Figure 2: The absolute error of problem (35) where 𝛼 = 𝛽 = 1/2,
𝑐 = 0.1, −𝛼 = 𝛽 = 1/2, and 𝑁 = 12.
] = 𝜂 = 𝛾 = 10−3 , and 𝑁 = 4 for 𝛿 = 2.

×10−10
×10−6
4
2 3 1.0
1.0
E 1 E 2
1
0 0
0.5 0.0 0.5 t
0.2 t
0.4
0.5
x 0.6 x
0.8
0.0 0.0
1.0 1.0

Figure 3: The absolute error of problem (35) where −𝛼 = 𝛽 = 1/2, Figure 6: The absolute error of problem (40) where ] = 10, 𝜇 = 0.1,
] = 𝜂 = 𝛾 = 10−3 , and 𝑁 = 4 for 𝛿 = 3. 𝑐 = 0.1, 𝛼 = 𝛽 = 1/2, and 𝑁 = 20.
Abstract and Applied Analysis 9

Table 5: Maximum absolute errors with various choices of (𝛼, 𝛽) for Table 6: Absolute errors with 𝛼 = 𝛽 = 1/2 and various choices of
Example 2. 𝑥, 𝑡 for Example 2.

𝑥 𝑡 𝐴 𝐵 𝜇 ] 𝑁 𝐸 𝐴 𝐵 𝐸
𝛼 𝛽 𝐴 𝐵 𝜇 ] 𝑁 𝑀𝐸 𝐴 𝐵 𝑀𝐸
0.0 0.1 0 1 0.1 10 20 1.34 × 10−12 −1 1 9.21 × 10−11
0 0 0 1 0.1 10 4 1.49 × 10−6 −1 1 5.62 × 10−7 0.1 8.71 × 10−11 8.70 × 10−11
−10
0.2 1.07 × 10 8.16 × 10−11
1/2 1/2 2.45 × 10−6 8.25 × 10−7 0.3 1.07 × 10−10
7.61 × 10−11
−1/2 −1/2 6.51 × 10−7 6.51 × 10−7 0.4 1.01 × 10−10 7.01 × 10−11
−11
0.5 9.21 × 10 6.38 × 10−11
−1/2 1/2 3.84 × 10−6 4.67 × 10−7 0.6 8.16 × 10−11
5.70 × 10−11
−11
1/2 −1/2 1.20 × 10−6 1.20 × 10−6 0.7 7.01 × 10 4.88 × 10−11
−11
0.8 5.70 × 10 3.85 × 10−11
0 0 0 1 0.1 10 16 1.43 × 10−9 −1 1 1.06 × 10−9 0.9 3.85 × 10−11
2.30 × 10−11
−12
1 1.34 × 10 1.34 × 10−12
1/2 1/2 1.62 × 10−9 1.18 × 10−9 −12
0.0 0.2 0 1 0.1 10 20 1.77 × 10 −1 1 3.35 × 10−10
−1/2 −1/2 6.70 × 10−10 4.45 × 10−10 0.1 2.71 × 10−10 3.17 × 10−10
−10
−1/2 1/2 6.68 × 10−10 4.49 × 10−10 0.2 3.61 × 10 2.98 × 10−10
−10
0.3 3.78 × 10 2.77 × 10−10
1/2 −1/2 6.67 × 10−10 4.23 × 10−10 0.4 3.64 × 10−10
2.55 × 10−10
−10
0.5 3.34 × 10 2.31 × 10−10
−10
0.6 2.98 × 10 2.03 × 10−10
−10
0.7 2.55 × 10 1.71 × 10−10
−10
0.8 2.02 × 10 1.30 × 10−10
0.9 1.30 × 10−10 7.70 × 10−11
−12
×10−9 1 1.77 × 10 1.77 × 10−12
1.5 1.0
1
E 0.5 and the initial condition
0 1/𝛿
0.5 1 1 ]𝛿
0.0 t 𝑢 (𝑥, 0) = [ − tanh [ (𝑥)]] , 𝑥 ∈ [𝐴, 𝐵] .
2 2 2 (𝛿 + 1)
0.5 (46)
x
0.0 The exact solution of (44) is
1.0

Figure 7: The absolute error of problem (40) where ] = 10, 𝜇 = 0.1, 𝑢 (𝑥, 𝑡)
𝑐 = 0.1, 𝛼 = 𝛽 = −1/2, and 𝑁 = 16. 1/𝛿
1 1 ]𝛿 ] 𝛾 (𝛿+1)
= [ − tanh [ (𝑥−( + ) 𝑡)]] .
2 2 2 (𝛿+1) 𝛿+1 ]
subject to the boundary conditions (47)

𝑢 (𝐴, 𝑡) In Table 7, we listed a comparison of absolute errors


of problem (44) subject to (45) and (46) using the J-GL-
1 1 C method with [19]. Absolute errors between exact and
=[ −
2 2 numerical solutions of (44) subject to (45) and (46) are
1/𝛿 introduced in Table 8 using the J-GL-C method for 𝛼 = 𝛽 = 0
]𝛿 ] 𝛾 (𝛿+1) with 𝑁 = 16, respectively, and ] = 𝛾 = 10−2 . In Figures 9 and
×tanh [ (𝐴−( + ) 𝑡)]] ,
2 (𝛿 + 1) 𝛿+1 ] 10, we displayed the absolute errors of problem (44) where
𝑢 (𝐵, 𝑡) ] = 𝛾 = 10−2 at 𝑁 = 20 and (𝛼 = 𝛽 = 0 and 𝛼 = 𝛽 = −1/2) in
interval [−1, 1], respectively. Moreover, in Figures 11 and 12,
1 1 we see that, in interval [−1, 1], the approximate solution and
=[ − the exact solution are almost coincided for different values
2 2
of 𝑡 (0, 0.5 and 0.9) of problem (44) where ] = 𝛾 = 10−2 at
1/𝛿
]𝛿 ] 𝛾 (𝛿 + 1) 𝑁 = 20 and (𝛼 = 𝛽 = 0 and 𝛼 = 𝛽 = −1/2), respectively.
× tanh [ (𝐵 − ( + ) 𝑡)]] ,
2 (𝛿 + 1) 𝛿+1 ] This asserts that the obtained numerical results are accurate
(45) and can be compared favorably with the analytical solution.
10 Abstract and Applied Analysis

Table 7: Comparison of absolute errors of Example 3 with results from [19], where 𝑁 = 4, 𝛼 = 𝛽 = 0, and various choices of 𝑥, 𝑡.

𝑥 𝑡 𝛾 ] 𝛿 [19] 𝐸 𝑥 𝑡 𝛾 ] 𝛿 [19] 𝐸
0.1 0.005 0.001 0.001 1 9.69 × 10−6 1.85 × 10−6 0.1 0.0005 1 1 2 1.40 × 10−3 3.83 × 10−5
0.001 1.94 × 10−6 3.72 × 10−7 0.0001 2.80 × 10−4 3.88 × 10−5
0.01 1.94 × 10−5 3.72 × 10−6 0.001 2.80 × 10−3 3.76 × 10−5
0.5 0.005 9.69 × 10−6 7.04 × 10−6 0.5 0.0005 1.35 × 10−3 2.32 × 10−5
0.001 1.94 × 10−6 1.41 × 10−6 0.0001 2.69 × 10−4 2.38 × 10−5
0.01 1.94 × 10−5 1.41 × 10−5 0.001 2.69 × 10−3 2.25 × 10−5
0.9 0.005 9.69 × 10−6 3.21 × 10−6 0.9 0.0005 1.28 × 10−3 1.58 × 10−5
0.001 1.94 × 10−6 6.42 × 10−7 0.0001 2.55 × 10−4 1.55 × 10−5
0.01 1.94 × 10−5 6.42 × 10−6 0.001 2.55 × 10−3 1.61 × 10−5

0.0013 Table 8: Absolute errors with 𝛼 = 𝛽 = 0, 𝛿 = 1 and various choices


of 𝑥, 𝑡 for Example 3.
0.0012
𝑥 𝑡 𝛾 ] 𝛿 𝑁 𝐸 𝑥 𝑡 𝛾 ] 𝛿 𝑁 𝐸
0.0 0.1 1 1 1 16 3.26 × 10−9 0.0 0.2 1 1 1 16 3.57 × 10−11
0.0011
3.82 × 10−9 7.75 × 10−11
u and u
̃

0.1 0.1
0.2 4.28 × 10−9 0.2 1.96 × 10−10
0.0010
0.3 4.66 × 10−9 0.3 3.52 × 10−10
0.4 4.93 × 10−9 0.4 5.62 × 10−10
0.009
0.5 5.05 × 10−9 0.5 7.31 × 10−10
0.6 4.93 × 10−9 0.6 7.87 × 10−10
0.008
−0.4 −0.2 0.0 0.2 0.4 0.7 4.49 × 10−9 0.7 8.23 × 10−10
x 0.8 3.61 × 10−9 0.8 8.17 × 10−10
0.9 2.26 × 10−9 0.9 8.05 × 10−10
u(x, 0.0) ̃ (x, 0.5)
u
̃ (x, 0.0)
u u(x, 0.9) 1.0 1.13 × 10−11 1.0 1.09 × 10−11
u(x, 0.5) ̃ (x, 0.9)
u

Figure 8: The approximate and exact solutions for different values


of 𝑡 (0, 0.5 and 0.9) of problem (40) where ] = 10, 𝜇 = 0.1, 𝑐 = 0.1,
𝛼 = 𝛽 = −0.5, and 𝑁 = 20.

×10−10
8
6 1.0
4
E
2
0
−1.0 0.5 t
−0.5
0.0
x
×10−10 0.5
0.0
6 1.0
1.0
4
Figure 10: The absolute error of problem (44) where 𝛼 = 𝛽 = 1/2
E 2
and ] = 𝛾 = 10−2 at 𝑁 = 20.
0
−1.0 0.5
t
−0.5
0.0
5. Conclusion
x
0.5 An efficient and accurate numerical scheme based on the J-
0.0 GL-C spectral method is proposed to solve nonlinear time-
1.0
dependent Burgers-type equations. The problem is reduced
Figure 9: The absolute error of problem (44) where 𝛼 = 𝛽 = 0 and to the solution of a SODEs in the expansion coefficient of
] = 𝛾 = 10−2 at 𝑁 = 20. the solution. Numerical examples were given to demonstrate
Abstract and Applied Analysis 11

0.5030 [3] S. Nguyen and C. Delcarte, “A spectral collocation method to


solve Helmholtz problems with boundary conditions involving
0.5025 mixed tangential and normal derivatives,” Journal of Computa-
tional Physics, vol. 200, no. 1, pp. 34–49, 2004.
0.5020 [4] A. H. Bhrawy and A. S. Alofi, “A Jacobi-Gauss collocation
u and u
̃

method for solving nonlinear Lane-Emden type equations,”


0.5015 Communications in Nonlinear Science and Numerical Simula-
tion, vol. 17, no. 1, pp. 62–70, 2012.
0.5010
[5] A. H. Bhrawy, E. Tohidi, and F. Soleymani, “A new Bernoulli
matrix method for solving high-order linear and nonlinear
0.5005
Fredholm integro-differential equations with piecewise inter-
vals,” Applied Mathematics and Computation, vol. 219, no. 2, pp.
−1.0 −0.5 0.0 0.5 1.0 482–497, 2012.
x
[6] A. H. Bhrawy and M. Alshomrani, “A shifted legendre spectral
u(x, 0.6) ̃ (x, 0.9)
u method for fractional-ordermulti-point boundary value prob-
̃ (x, 0.6)
u u(x, 1.3) lems,” Advances in Difference Equations, vol. 2012, article 8, 2012.
u(x, 0.9) ̃ (x, 1.3)
u [7] E. H. Doha, A. H. Bhrawy, D. Baleanu, and S. S. Ezz-Eldien, “On
shifted Jacobi spectral approximations for solving fractional
Figure 11: The approximate and exact solutions for different values differential equations,” Applied Mathematics and Computation,
of 𝑡 (0, 0.5 and 0.9) of problem (44) where 𝛼 = 𝛽 = 0 and ] = 𝛾 = vol. 219, no. 15, pp. 8042–8056, 2013.
10−2 at 𝑁 = 20.
[8] L. Zhu and Q. Fan, “Solving fractional nonlinear Fredholm
integro-differential equations by the second kind Chebyshev
wavelet,” Communications in Nonlinear Science and Numerical
0.5030
Simulation, vol. 17, no. 6, pp. 2333–2341, 2012.
0.5025 [9] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A new
Jacobi operational matrix: an application for solving fractional
0.5020 differential equations,” Applied Mathematical Modelling, vol. 36,
u and u
̃

no. 10, pp. 4931–4943, 2012.


0.5015 [10] A. Saadatmandi and M. Dehghan, “The use of sinc-collocation
method for solving multi-point boundary value problems,”
0.5010 Communications in Nonlinear Science and Numerical Simula-
tion, vol. 17, no. 2, pp. 593–601, 2012.
0.5005 [11] E. H. Doha, A. H. Bhrawy, and R. M. Hafez, “On shifted Jacobi
spectral method for high-order multi-point boundary value
−1.0 −0.5 0.0 0.5 1.0 problems,” Communications in Nonlinear Science and Numerical
x Simulation, vol. 17, no. 10, pp. 3802–3810, 2012.
[12] G. Ierley, B. Spencer, and R. Worthing, “Spectral methods in
u(x, 0.6) ̃ (x, 0.9)
u
time for a class of parabolic partial differential equations,”
̃ (x, 0.6)
u u(x, 1.3)
Journal of Computational Physics, vol. 102, no. 1, pp. 88–97, 1992.
u(x, 0.9) ̃ (x, 1.3)
u
[13] H. Tal-Ezer, “Spectral methods in time for hyperbolic equa-
Figure 12: The approximate and exact solutions for different values tions,” SIAM Journal on Numerical Analysis, vol. 23, no. 1, pp.
of 𝑡 (0, 0.5 and 0.9) of problem (44) where 𝛼 = 𝛽 = 1/2 and ] = 𝛾 = 11–26, 1986.
10−2 at 𝑁 = 20. [14] H. Tal-Ezer, “Spectral methods in time for parabolic problems,”
SIAM Journal on Numerical Analysis, vol. 26, no. 1, pp. 1–11, 1989.
[15] E. A. Coutsias, T. Hagstrom, J. S. Hesthaven, and D. Tor-
the validity and applicability of the method. The results show res, “Integration preconditioners for differential operators in
spectral tau-methods,” in Proceedings of the 3rd International
that the J-GL-C method is simple and accurate. In fact by
Conference on Spectral and High Order Methods, A. V. Ilin and L.
selecting few collocation points, excellent numerical results
R. Scott, Eds., pp. 21–38, University of Houston, Houston, Tex,
are obtained. USA, 1996, Houston Journal of Mathematics.
[16] F.-y. Zhang, “Spectral and pseudospectral approximations in
References time for parabolic equations,” Journal of Computational Mathe-
matics, vol. 16, no. 2, pp. 107–120, 1998.
[1] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, [17] J.-G. Tang and H.-P. Ma, “A Legendre spectral method in
Spectral Methods: Fundamentals in Single Domains, Springer, time for first-order hyperbolic equations,” Applied Numerical
New York, NY, USA, 2006. Mathematics, vol. 57, no. 1, pp. 1–11, 2007.
[2] E. H. Doha and A. H. Bhrawy, “An efficient direct solver [18] A. H. Bhrawy, “A Jacobi-Gauss-Lobatto collocation method
for multidimensional elliptic Robin boundary value problems for solving generalized Fitzhugh-Nagumo equation with time-
using a Legendre spectral-Galerkin method,” Computers & dependent coefficients,” Applied Mathematics and Computation,
Mathematics with Applications, vol. 64, no. 4, pp. 558–571, 2012. vol. 222, pp. 255–264, 2013.
12 Abstract and Applied Analysis

[19] H. N. A. Ismail, K. Raslan, and A. A. A. Rabboh, “Ado- [36] A. Veksler and Y. Zarmi, “Wave interactions and the analysis of
mian decomposition method for Burger’s-Huxley and Burger’s- the perturbed Burgers equation,” Physica D, vol. 211, no. 1-2, pp.
Fisher equations,” Applied Mathematics and Computation, vol. 57–73, 2005.
159, no. 1, pp. 291–301, 2004. [37] A. Veksler and Y. Zarmi, “Freedom in the expansion and
[20] H. Bateman, “Some recent researchers on the motion of fluids,” obstacles to integrability in multiple-soliton solutions of the
Monthly Weather Review, vol. 43, pp. 163–170, 1915. perturbed KdV equation,” Physica D, vol. 217, no. 1, pp. 77–87,
[21] J. M. Burgers, “A mathematical model illustrating the theory 2006.
of turbulence,” in Advances in Applied Mechanics, pp. 171–199, [38] Z.-Y. Ma, X.-F. Wu, and J.-M. Zhu, “Multisoliton excitations for
Academic Press, New York, NY, USA, 1948. the Kadomtsev-Petviashvili equation and the coupled Burgers
[22] M. K. Kadalbajoo and A. Awasthi, “A numerical method based equation,” Chaos, Solitons and Fractals, vol. 31, no. 3, pp. 648–
on Crank-Nicolson scheme for Burgers’ equation,” Applied 657, 2007.
Mathematics and Computation, vol. 182, no. 2, pp. 1430–1442, [39] R. A. Fisher, “The wave of advance of advantageous genes,”
2006. Annals of Eugenics, vol. 7, pp. 335–369, 1937.
[23] M. Gülsu, “A finite difference approach for solution of Burgers’ [40] A. J. Khattak, “A computational meshless method for the
equation,” Applied Mathematics and Computation, vol. 175, no. generalized Burger’s-Huxley equation,” Applied Mathematical
2, pp. 1245–1255, 2006. Modelling, vol. 33, no. 9, pp. 3718–3729, 2009.
[24] P. Kim, “Invariantization of the Crank-Nicolson method for [41] N. F. Britton, Reaction-Diffusion Equations and their Applica-
Burgers’ equation,” Physica D, vol. 237, no. 2, pp. 243–254, 2008. tions to Biology, Academic Press, London, UK, 1986.
[25] H. Nguyen and J. Reynen, “A space-time least-square finite [42] D. A. Frank, Diffusion and Heat Exchange in Chemical Kinetics,
element scheme for advection-diffusion equations,” Computer Princeton University Press, Princeton, NJ, USA, 1955.
Methods in Applied Mechanics and Engineering, vol. 42, no. 3, [43] A.-M. Wazwaz, “The extended tanh method for abundant
pp. 331–342, 1984. solitary wave solutions of nonlinear wave equations,” Applied
[26] H. Nguyen and J. Reynen, “A space-time finite element approach Mathematics and Computation, vol. 187, no. 2, pp. 1131–1142,
to Burgers’ equation,” in Numerical Methods for Non-Linear 2007.
Problems, C. Taylor, E. Hinton, D. R. J. Owen, and E. Onate, Eds., [44] W. Malfliet, “Solitary wave solutions of nonlinear wave equa-
vol. 2, pp. 718–728, 1984. tions,” American Journal of Physics, vol. 60, no. 7, pp. 650–654,
[27] L. R. T. Gardner, G. A. Gardner, and A. Dogan, “A Petrov- 1992.
Galerkin finite element scheme for Burgers’ equation,” The [45] Y. Tan, H. Xu, and S.-J. Liao, “Explicit series solution of
Arabian Journal for Science and Engineering C, vol. 22, no. 2, pp. travelling waves with a front of Fisher equation,” Chaos, Solitons
99–109, 1997. and Fractals, vol. 31, no. 2, pp. 462–472, 2007.
[28] L. R. T. Gardner, G. A. Gardner, and A. Dogan, in A Least- [46] J. Canosa, “Diffusion in nonlinear multiplicate media,” Journal
Squares Finite Element Scheme For Burgers Equation, Mathe- of Mathematical Physics, vol. 31, pp. 1862–1869, 1969.
matics, University of Wales, Bangor, UK, 1996. [47] A.-M. Wazwaz, “Travelling wave solutions of generalized
[29] S. Kutluay, A. Esen, and I. Dag, “Numerical solutions of forms of Burgers, Burgers-KdV and Burgers-Huxley equations,”
the Burgers’ equation by the least-squares quadratic B-spline Applied Mathematics and Computation, vol. 169, no. 1, pp. 639–
finite element method,” Journal of Computational and Applied 656, 2005.
Mathematics, vol. 167, no. 1, pp. 21–33, 2004. [48] E. Fan and Y. C. Hon, “Generalized tanh method extended
[30] R. C. Mittal and R. K. Jain, “Numerical solutions of nonlinear to special types of nonlinear equations,” Zeitschrift fur Natur-
Burgers’ equation with modified cubic B-splines collocation forschung A, vol. 57, no. 8, pp. 692–700, 2002.
method,” Applied Mathematics and Computation, vol. 218, no.
15, pp. 7839–7855, 2012.
[31] E. H. Doha and A. H. Bhrawy, “Efficient spectral-Galerkin algo-
rithms for direct solution of fourth-order differential equations
using Jacobi polynomials,” Applied Numerical Mathematics, vol.
58, no. 8, pp. 1224–1244, 2008.
[32] E. H. Doha and A. H. Bhrawy, “A Jacobi spectral Galerkin
method for the integrated forms of fourth-order elliptic dif-
ferential equations,” Numerical Methods for Partial Differential
Equations, vol. 25, no. 3, pp. 712–739, 2009.
[33] S. Kazem, “An integral operational matrix based on Jacobi
polynomials for solving fractional-order differential equations,”
Applied Mathematical Modelling, vol. 37, no. 3, pp. 1126–1136,
2013.
[34] A. Ahmadian, M. Suleiman, S. Salahshour, and D. Baleanu, “A
Jacobi operational matrix for solving a fuzzy linear fractional
differential equation,” Advances in Difference Equations, vol.
2013, p. 104, 2013.
[35] E. H. Doha, A. H. Bhrawy, and R. M. Hafez, “A Jacobi-
Jacobi dual-Petrov-Galerkin method for third- and fifth-order
differential equations,” Mathematical and Computer Modelling,
vol. 53, no. 9-10, pp. 1820–1832, 2011.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 316978, 6 pages
http://dx.doi.org/10.1155/2013/316978

Research Article
Mappings for Special Functions on Cantor Sets and
Special Integral Transforms via Local Fractional Operators

Yang Zhao,1,2 Dumitru Baleanu,3,4,5 Mihaela Cristina Baleanu,6


De-Fu Cheng,2 and Xiao-Jun Yang7
1
Electronic and Information Technology Department, Jiangmen Polytechnic, Jiangmen 529090, China
2
College of Instrumentation & Electrical Engineering, Jilin University, Changchun 130061, China
3
Department of Chemical and Materials Engineering, Faculty of Engineering, King Abdulaziz University,
P.O. Box 80204, Jeddah 21589, Saudi Arabia
4
Department of Mathematics and Computer Sciences, Faculty of Arts and Sciences, Cankaya University, 06530 Ankara, Turkey
5
Institute of Space Sciences, Magurele, 077125 Bucharest, Romania
6
Mihail Sadoveanu Theoretical High School, District 2, Street Popa Lazar No. 8, 021586 Bucharest, Romania
7
Department of Mathematics and Mechanics, China University of Mining and Technology, Xuzhou Campus,
Xuzhou, Jiangsu 221008, China

Correspondence should be addressed to De-Fu Cheng; chengdefu@jlu.edu.cn

Received 27 August 2013; Accepted 24 September 2013

Academic Editor: Ali H. Bhrawy

Copyright © 2013 Yang Zhao et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The mappings for some special functions on Cantor sets are investigated. Meanwhile, we apply the local fractional Fourier
series, Fourier transforms, and Laplace transforms to solve three local fractional differential equations, and the corresponding
nondifferentiable solutions were presented.

1. Introduction The continuous-time finance based on the Mittag-Leffler


function was given [20]. In [21], the fractional radial diffusion
Special functions [1] play an important role in mathematical in a cylinder based on the Mittag-Leffler function was
analysis, function analysis physics, and so on. We recall investigated. In [22], the Mittag-Leffler stability theorem for
here some very well examples, the Gamma function [2], fractional nonlinear systems with delay was considered. The
hypergeometric function [3], Bessel functions [4], Whittaker stochastic linear Volterra equations of convolution type based
function [5], G-function [6], q-special functions [7], Fox’s on the Mittag-Leffler function were suggested in [23].
H-functions [8], Mittag-Leffler function [9], and Wright’s Recently, based on the Mittag-Leffler functions on Cantor
function [10]. sets via the fractal measure, the special integral transforms
The Mittag-Leffler function had successfully been applied based on the local fractional calculus theory were suggested
to solve the practical problems [11–15]. For example, the in [24]. In this work, some applications for the local fractional
Mittag-Leffler-type functions in fractional evolution pro- calculus theory are studied in [24–36]. The main aim of this
cesses were suggested [15]. Solutions for fractional reaction- paper is to investigate the mappings for special functions
diffusion equations via Mittag-Leffler-type functions were on Cantor sets and some applications of special integral
discussed [16]. The Mittag-Leffler stability of fractional order transforms to nondifferentiable problems.
nonlinear dynamic systems was presented [17]. Models The paper is organized as follows. In Section 2, the map-
based on Mittag-Leffler functions for anomalous relaxation pings for special functions on Cantor sets are investigated.
in dielectrics were proposed [18]. In [19], the anomalous In Section 3, the special integral transforms within local
relaxation via the Mittag-Leffler functions was reported. fractional calculus and some applications to nondifferentiable
2 Abstract and Applied Analysis

1 4.5
0.9
4
0.8
0.7 3.5

0.6
3
0.5
0.4 2.5

0.3 2
0.2
1.5
0.1
0 1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 1: Graph of 𝑥2𝛼 for 𝛼 = ln 2/ ln 3. Figure 2: Graph of 𝐸𝛼 (𝑥𝛼 ) for 𝛼 = ln 2/ ln 3.

0.7
problems are presented. Finally, in Section 4, the conclusions
are presented. 0.6

0.5
2. Mappings for Special Functions on
Cantor Sets 0.4

In order to give the mappings for special functions on Cantor


0.3
sets, we first recall some basic definitions about the fractal
measure theory [25].
0.2
Let Lebesgue-Cantor staircase function be defined as [25]
0.1
𝐻𝛼 (𝐹 ∩ (0, 𝑥)) = Γ (1 + 𝛼) 0 𝐼(𝛼)
𝑥 1,
(1)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
where 𝐹 is a cantor set, 𝐻𝛼 (⋅) is the 𝛼-dimensional Hausdorff
𝛼
measure, 0 𝐼(𝛼) Figure 3: Graph of sin𝛼 𝑥 for 𝛼 = ln 2/ ln 3.
𝑥 (⋅) is local fractional integral operator [24–31],
and Γ(⋅) is a Gamma function.
Following (1), we obtain
The sine on Cantor sets is defined by [24, 25]
𝐻𝛼 (𝐹 ∩ (0, 𝑥)) = 𝑥𝛼 , (2)

𝑥𝛼(2𝑘+1)
which is a Lebesgue-Cantor staircase function. For its graph, sin𝛼 𝑥𝛼 = ∑ (−1)𝑘 , (5)
𝑘=0
Γ [1 + 𝛼 (2𝑘 + 1)]
please see [28].
In this way, we define some real-valued functions on
Cantor sets as follows [24–26]. and its corresponding graph is depicted in Figure 3.
The Cantor staircase function is defined as [25] The cosine on Cantor sets is [24, 25]


𝑓 (𝑥) = 𝑥2𝛼 , 𝑥2𝛼𝑘
(3) cos𝛼 𝑥𝛼 = ∑ (−1)𝑘 , (6)
𝑘=0
Γ (1 + 2𝛼𝑘)
and its graph is shown in Figure 1.
The Mittag-Leffler functions on Cantor sets are given by with graph in Figure 4.
[24, 25] Hyperbolic sine on Cantor sets is defined by [24, 25]


𝑥𝛼𝑘 ∞
𝑥𝛼(2𝑘+1)
𝐸𝛼 (𝑥𝛼 ) = ∑ , (4) sinh𝛼 𝑥𝛼 = ∑ , (7)
𝑘=0
Γ (1 + 𝑘𝛼) 𝑘=0
Γ [1 + 𝛼 (2𝑘 + 1)]

and we draw the corresponding graph in Figure 2. and we draw its graphs as shown in Figure 5.
Abstract and Applied Analysis 3

1 2.5

0.9

0.8
2
0.7

0.6
1.5
0.5

0.4

1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 4: Graph of cos𝛼 𝑥𝛼 for 𝛼 = ln 2/ ln 3. Figure 6: Graph of cosh𝛼 𝑥𝛼 for 𝛼 = ln 2/ ln 3.

2 3. Special Integral Transforms within Local


1.8 Fractional Calculus
0.6 In this section, we introduce the conceptions of special
1.4 integral transforms within the local fractional calculus con-
1.2
cluding the local fractional Fourier series and Fourier and
Laplace transforms. After that, we present three illustrative
1 examples.
0.8
0.6 3.1. Definitions of Special Integral Transforms within Local
Fractional Calculus. We here present briefly some results
0.4 used in the rest of the paper.
0.2 Let 𝑓(𝑥) ∈ 𝐶𝛼 (−∞, ∞). Local fractional trigonometric
Fourier series of 𝑓(𝑥) is given by [24, 26–28]
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 ∞
𝛼
Figure 5: Graph of sinh𝛼 𝑥 for 𝛼 = ln 2/ ln 3. 𝑓 (𝑥) = 𝑎0 + ∑𝑎𝑘 sin𝛼 (𝑘𝛼 𝜔0𝛼 𝑥𝛼 )
𝑖=1
(12)

+ ∑𝑏𝑘 cos𝛼 (𝑘𝛼 𝜔0𝛼 𝑥𝛼 ) .
Hyperbolic cosine on Cantor sets is defined as [24, 25] 𝑖=1


The local fractional Fourier coefficients read as
𝛼 𝑥2𝛼𝑘
cosh𝛼 𝑥 = ∑ , (8) 1 𝑇
𝑘=0
Γ (1 + 2𝛼𝑘) 𝑎0 = ∫ 𝑓 (𝑥) (𝑑𝑥)𝛼 ,
𝑇𝛼 0
and its graph is shown in Figure 6. 2 𝛼 𝑇
𝑎𝑘 = ( ) ∫ 𝑓 (𝑥) sin𝛼 (𝑘𝛼 𝜔0𝛼 𝑥𝛼 ) (𝑑𝑥)𝛼 , (13)
Following (4)–(8), we have 𝑇 0

𝐸𝛼 (𝑖𝛼 𝑥𝛼 ) = cos𝛼 𝑥𝛼 + 𝑖𝛼 sin𝛼 𝑥𝛼 , 2 𝛼 𝑇


(9) 𝑏𝑘 = ( ) ∫ 𝑓 (𝑥) cos𝛼 (𝑘𝛼 𝜔0𝛼 𝑥𝛼 ) (𝑑𝑥)𝛼 .
𝑇 0

where 𝑖𝛼 is a fractal unit of an imaginary number [24, 26–32]. We notice that the above results are obtained from
If for 𝜀, 𝛿 > 0 and 𝜀, 𝛿 ∈ 𝑅, 𝑓(𝑥) satisfies the condition Pythagorean theorem in the generalized Hilbert space
[24–26] [24, 26–28].
Let 𝑓(𝑥) ∈ 𝐶𝛼 (−∞, ∞). The local fractional Fourier
󵄨󵄨 󵄨
󵄨󵄨𝑓 (𝑥) − 𝑓 (𝑥0 )󵄨󵄨󵄨 < 𝜀 ;
𝛼
(10) transform of 𝑓(𝑥) is suggested by [24, 29–32]
𝐹𝛼 {𝑓 (𝑥)} = 𝑓𝜔𝐹,𝛼 (𝜔)
for 𝑥 ∈ [𝑎, 𝑏] we write it as follows:
1 ∞ (14)
= ∫ 𝐸𝛼 (−𝑖𝛼 𝜔𝛼 𝑥𝛼 ) 𝑓 (𝑥) (𝑑𝑥)𝛼 .
𝑓 (𝑥) ∈ 𝐶𝛼 (𝑎, 𝑏) . (11) Γ (1 + 𝛼) −∞
4 Abstract and Applied Analysis

The inverse formula is expressed as follows [24, 29–32]: Submitting (20)-(21) into (18), we obtain

𝑓 (𝑥) = 𝐹𝛼−1 (𝑓𝜔𝐹,𝛼 (𝜔)) ∞


𝑎 ( ∑ 𝐴 𝑛 𝑛𝛼 cos𝛼 (𝑛𝛼 𝑥𝛼 )
1 ∞ (15) 𝑛=1
= 𝛼 ∫ 𝐸 (𝑖𝛼 𝜔𝛼 𝑥𝛼 ) 𝑓𝜔𝐹,𝛼 (𝜔) (𝑑𝜔)𝛼 .
(2𝜋) −∞ 𝛼 ∞
+ ∑ 𝐵𝑛 𝑛𝛼 sin𝛼 (𝑛𝛼 𝑥𝛼 ))
Let 𝑓(𝑥) ∈ 𝐶𝛼 (−∞, ∞). The local fractional Laplace
𝑘=1
transform of 𝑓(𝑥) is defined as [24, 32, 33]

𝐿 𝛼 {𝑓 (𝑥)} = 𝑓𝑠𝐿,𝛼 (𝑠) + 𝑏 (𝑎0 + ∑ 𝐴 𝑛 sin𝛼 (𝑛𝛼 𝑥𝛼 ) (22)
𝑛=1
1 ∞ (16)
= ∫ 𝐸𝛼 (−𝑠𝛼 𝑥𝛼 ) 𝑓 (𝑥) (𝑑𝑥)𝛼 . ∞
Γ (1 + 𝛼) 0 + ∑ 𝐵𝑛 cos𝛼 (𝑛𝛼 𝑥𝛼 ))
𝑛=1
The inverse formula local fractional Laplace transform of
𝑓(𝑥) is derived as [24, 32, 33] ∞
= ∑ sin𝛼 (𝑛𝛼 𝑥𝛼 ) .
𝑓 (𝑥) = 𝐿−1 𝐿,𝛼
𝛼 {𝑓𝑠 (𝑠)} 𝑛=1

𝛽+𝑖∞ (17) Hence, we get


1
= 𝛼 ∫ 𝐸 (𝑠𝛼 𝑥𝛼 ) 𝑓𝑠𝐿,𝛼 (𝑠) (𝑑𝑠)𝛼 ,
(2𝜋) 𝛽−𝑖∞ 𝛼 𝑎0 𝑏 = 0,
where 𝑓(𝑥) is local fractional continuous, 𝑠𝛼 = 𝛽𝛼 + 𝑖𝛼 ∞𝛼 , 𝑎𝐴 𝑛 𝑛𝛼 + 𝑏𝐵𝑛 = 0, (23)
and Re(𝑠) = 𝛽 > 0.
For more details of special integral transforms via local 𝑎𝐵𝑛 𝑛𝛼 + 𝑏𝐴 𝑛 = 1.
fractional calculus, see [24, 32, 33] and the references therein.
Therefore, we can calculate
3.2. Applications of Local Fractional Fourier Series and Fourier
and Laplace Transforms to the Differential Equation on Cantor 𝑎0 = 0,
Sets. We now present the powerful tool of the methods
𝑏
presented above in three illustrative examples. 𝐴𝑛 = − , (24)
𝑎2 𝑛2𝛼 − 𝑏2
Example 1. Let us begin with the local fractional differential 𝑎𝑛𝛼
equation on Cantor set in the following form: 𝐵𝑛 = .
𝑎2 𝑛2𝛼 − 𝑏2
𝑑𝛼 𝑦
𝑎 + by (𝑥) = 𝑓 (𝑥) , 𝑥 ∈ (−∞, +∞) , (18) In view of (24), we give the solution of (18) as follows:
𝑑𝛼 𝑥

where 𝑎 and 𝑏 are constants and the nondifferentiable 𝑏
𝑦𝑝 (𝑥) = − ∑ sin𝛼 (𝑛𝛼 𝑥𝛼 )
function 𝑓(𝑥) is periodic of period 2𝜋 so that it can be 𝑛=1 𝑎2 𝑛2𝛼 − 𝑏2
expanded in a local fractional Fourier series as follows: (25)

𝑎𝑛𝛼
∞ + ∑ 2 2𝛼 cos𝛼 (𝑛𝛼 𝑥𝛼 ) .
𝑓 (𝑥) = ∑ sin𝛼 (𝑛𝛼 𝑥𝛼 ) . (19) 𝑛=1 𝑎 𝑛 − 𝑏2

𝑛=1
Example 2. We now consider the following differential equa-
Here, we give a particular solution in the following form:
tion on Cantor sets:

𝑦𝑝 (𝑥) = 𝑎0 + ∑ 𝐴 𝑛 sin𝛼 (𝑛𝛼 𝑥𝛼 ) 𝑑2𝛼 𝑥
+ 𝑝𝑥 = 𝑓 (𝑡) , +∞ > 𝑡 > −∞, (26)
𝑛=1 𝑑2𝛼 𝑡
(20)

𝛼 𝛼 subject to the initial value condition
+ ∑ 𝐵𝑛 cos𝛼 (𝑛 𝑥 ) .
𝑛=1
𝑑𝛼 𝑥 󵄨󵄨󵄨󵄨
󵄨 = 0, 𝑥 (0) = 0, (27)
Following (20), we have 𝑑𝛼 𝑡 󵄨󵄨󵄨𝑡=0

𝑦𝑝(𝛼) (𝑥) = ∑ 𝐴 𝑛 𝑛𝛼 cos𝛼 (𝑛𝛼 𝑥𝛼 ) where 𝑝 is constant and 𝑓(𝑡) is the local fractional continuous
𝑛=1 function so that its local fractional Fourier transform exists.
(21) Application of local fractional Fourier transform gives

+ ∑ 𝐵𝑛 𝑛𝛼 sin𝛼 (𝑛𝛼 𝑥𝛼 ) .
𝑛=1 −𝜔2𝛼 𝑥𝜔𝐹,𝛼 (𝜔) + 𝑝𝑥𝜔𝐹,𝛼 (𝜔) = 𝑓𝜔𝐹,𝛼 (𝜔) , (28)
Abstract and Applied Analysis 5

so that References
2𝛼
(−𝜔 + 𝑝) 𝑥𝜔𝐹,𝛼 (𝜔) = 𝑓𝜔𝐹,𝛼 (𝜔) . (29) [1] G. E. Andrews, R. Askey, and R. Roy, Special Functions,
Cambridge University Press, Cambridge, UK, 1999.
From (29), we have
[2] Á. Elbert and A. Laforgia, “On some properties of the gamma
𝑓𝜔𝐹,𝛼 (𝜔) function,” Proceedings of the American Mathematical Society,
𝑥𝜔𝐹,𝛼 (𝜔) = . (30) vol. 128, no. 9, pp. 2667–2673, 2000.
(−𝜔2𝛼 + 𝑝)
[3] N. Virchenko, S. L. Kalla, and A. Al-Zamel, “Some results on a
Therefore, taking the inverse formula of local fractional generalized hypergeometric function,” Integral Transforms and
Fourier transform, we have Special Functions, vol. 12, no. 1, pp. 89–100, 2001.
𝑝−(1/2) 𝑡 [4] J. Choi and P. Agarwal, “Certain unified integrals associated
𝑥 (𝑡) = − ∫ 𝑓 (𝑡 − 𝜏) sin𝛼 (𝑝1/2 𝜏𝛼 ) (𝑑𝜏)𝛼 . (31) with Bessel functions,” Boundary Value Problems, vol. 2013, no.
Γ (1 + 𝛼) −∞ 1, pp. 1–9, 2013.
Example 3. Let us find the solution to the differential equa- [5] P. J. McNamara, “Metaplectic Whittaker functions and crystal
tion on Cantor sets bases,” Duke Mathematical Journal, vol. 156, no. 1, pp. 1–31, 2011.
𝑑2𝛼 𝑥 𝑑𝛼 𝑥 [6] E. W. Barnes, “The theory of the G-function,” Quarterly Journal
+ − 2𝑥 = 𝑓 (𝑡) , 𝑡 > 0, (32) of Mathematics, vol. 31, pp. 264–314, 1899.
𝑑2𝛼 𝑡 𝑑𝛼 𝑡
[7] R. Floreanini and L. Vinet, “𝑈𝑞 (𝑠𝑙(2)) and 𝑞-special functions,”
subject to the initial value condition Contemporary Mathematics, vol. 160, p. 85, 1994.
𝑑𝛼 𝑥 󵄨󵄨󵄨󵄨 [8] H. M. Srivastava, K. C. Gupta, and S. P. Goyal, The H-Functions
󵄨 = 0, 𝑥 (0) = 0, (33)
𝑑𝛼 𝑡 󵄨󵄨󵄨𝑡=0 of One and Two Variables with Applications, South Asian
Publishers, New Delhi, India, 1982.
where 𝑓(𝑡) is the local fractional continuous function so that [9] R. Gorenflo, A. A. Kilbas, and S. V. Rogosin, “On the generalized
its local fractional Laplace transform exists. Mittag-Leffler type functions,” Integral Transforms and Special
Taking the local fractional Laplace transform, from (32), Functions, vol. 7, no. 3-4, pp. 215–224, 1998.
we have [10] R. K. Saxena and M. Saigo, “Certain properties of fractional
(𝑠2𝛼 𝑥𝑠𝐿,𝛼 (𝑠) − 𝑠𝛼 𝑥 (0) − 𝑥(𝛼) (0)) + (𝑠𝛼 𝑥𝑠𝐿,𝛼 (𝑠) − 𝑥 (0)) calculus operators associated with generalized Mittag-Leffler
(34) function,” Fractional Calculus & Applied Analysis, vol. 8, no. 2,
pp. 141–154, 2005.
+ 2𝑥𝑠𝐿,𝛼 (𝑠) = 𝑓𝑠𝐿,𝛼 (𝑠) ,
[11] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional
so that Calculus Models and Numerical Methods, Series on Complexity,
Nonlinearity and Chaos, World Scientific, Singapore, 2012.
𝑓𝑠𝐿,𝛼 (𝑠)
𝑥𝑠𝐿,𝛼 (𝑠) = . (35) [12] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory
𝑠2𝛼 + 𝑠𝛼 − 2 and Applications of Fractional Differential Equations, vol. 204,
When the local fractional convolution of two functions is Elsevier, Amsterdam, The Netherlands, 2006.
given by [24] [13] J. Sabatier, O. P. Agrawal, and J. A. T. Machado, Advances in
𝑡 Fractional Calculus: Theoretical Developments and Applications
1
𝑓1 (𝑡) ∗ 𝑓2 (𝑡) = ∫ 𝑓1 (𝑡 − 𝜏) 𝑓2 (𝜏) (𝑑𝜏)𝛼 (36) in Physics and Engineering, Springer, New York, NY, USA, 2007.
Γ (1 + 𝛼) 0 [14] S. Hu, Y. Q. Chen, and T. S. Qiu, Fractional Processes and
and the local fractional Laplace transform of 𝑓1 (𝑡) ∗ 𝑓2 (𝑡) is Fractional-Order Signal Processing: Techniques and Applications,
[24] Springer, New York, NY, USA, 2012.
𝐿,𝛼 𝐿,𝛼 [15] F. Mainardi and R. Gorenflo, “On Mittag-Leffler-type functions
𝐿 𝛼 {𝑓1 (𝑡) ∗ 𝑓2 (𝑡)} = 𝑓𝑠,1 (𝑠) 𝑓𝑠,2 (𝑠) , (37) in fractional evolution processes,” Journal of Computational and
the inverse formula of the local fractional Laplace transform Applied Mathematics, vol. 118, no. 1-2, pp. 283–299, 2000.
together with the local fractional convolution theorem gives [16] R. K. Saxena, A. M. Mathai, and H. J. Haubold, “Fractional
the solution reaction-diffusion equations,” Astrophysics and Space Science,
𝑡 vol. 305, no. 3, pp. 289–296, 2006.
1
𝑥 (𝑡) = ∫ 𝑓 (𝑡 − 𝜏) (𝐸𝛼 (−2𝜏𝛼 ) + 𝐸𝛼 (𝜏𝛼 )) (𝑑𝜏)𝛼 . [17] Y. Li, Y. Chen, and I. Podlubny, “Mittag-Leffler stability of
Γ (1 + 𝛼) 0 fractional order nonlinear dynamic systems,” Automatica, vol.
(38) 45, no. 8, pp. 1965–1969, 2009.
[18] E. C. de Oliveira, F. Mainardi, and J. Vaz Jr., “Models
4. Conclusions based on Mittag-Leffler functions for anomalous relaxation in
dielectrics,” The European Physical Journal Special Topics, vol.
In this work, we investigated the mappings for special 193, no. 1, pp. 161–171, 2011.
functions on Cantor sets and special integral transforms [19] D. S. F. Crothers, D. Holland, Y. P. Kalmykov, and W. T. Coffey,
via local fractional calculus, namely, the local fractional “The role of Mittag-Leffler functions in anomalous relaxation,”
Fourier series, Fourier transforms, and Laplace transforms, Journal of Molecular Liquids, vol. 114, no. 1-3, pp. 27–34, 2004.
respectively. These transformations were applied successfully [20] E. Scalas, R. Gorenflo, and F. Mainardi, “Fractional calculus and
to solve three local fractional differential equations, and the continuous-time finance,” Physica A, vol. 284, no. 1–4, pp. 376–
nondifferentiable solutions were reported. 384, 2000.
6 Abstract and Applied Analysis

[21] B. N. N. Achar and J. W. Hanneken, “Fractional radial diffusion


in a cylinder,” Journal of Molecular Liquids, vol. 114, no. 1–3, pp.
147–151, 2004.
[22] S. J. Sadati, D. Baleanu, A. Ranjbar, R. Ghaderi, and T. Abdel-
jawad, “Mittag-Leffler stability theorem for fractional nonlinear
systems with delay,” Abstract and Applied Analysis, vol. 2010,
Article ID 108651, 7 pages, 2010.
[23] S. W. J. Welch, R. A. L. Rorrer, and R. G. Duren Jr., “Application
of time-based fractional calculus methods to viscoelastic creep
and stress relaxation of materials,” Mechanics Time-Dependent
Materials, vol. 3, no. 3, pp. 279–303, 1999.
[24] X. J. Yang, Local Fractional Functional Analysis and Its Applica-
tions, Asian Academic Publisher, Hong Kong, China, 2011.
[25] X. J. Yang, Advanced Local Fractional Calculus and Its Applica-
tions, World Science, New York, NY, USA, 2012.
[26] M.-S. Hu, R. P. Agarwal, and X.-J. Yang, “Local fractional
Fourier series with application to wave equation in fractal
vibrating string,” Abstract and Applied Analysis, vol. 2012,
Article ID 567401, 15 pages, 2012.
[27] Y. Zhang, A. Yang, and X.-J. Yang, “1-D heat conduction in a
fractal medium: a solution by the local fractional Fourier series
method,” Thermal Science, vol. 17, no. 3, pp. 953–956, 2013.
[28] Y.-J. Yang, D. Baleanu, and X.-J. Yang, “Analysis of fractal wave
equations by local fractional Fourier series method,” Advances
in Mathematical Physics, vol. 2013, Article ID 632309, 6 pages,
2013.
[29] X. J. Yang, D. Baleanu, and J. A. T. Machado, “Mathematical
aspects of Heisenberg uncertainty principle within local frac-
tional Fourier analysis,” Boundary Value Problems, vol. 2013, no.
1, pp. 131–146, 2013.
[30] A. M. Yang, Y. Z. Zhang, and Y. Long, “The Yang-Fourier
transforms to heat-conduction in a semi-infinite fractal bar,”
Thermal Science, vol. 17, no. 3, pp. 707–7713, 2013.
[31] F. Gao, W. P. Zhong, and X. M. Shen, “Applications of Yang-
Fourier transform to local fractional equations with local
fractional derivative and local fractional integral,” Advanced
Materials Research, vol. 461, pp. 306–310, 2012.
[32] J.-H. He, “Asymptotic methods for solitary solutions and com-
pactons,” Abstract and Applied Analysis, vol. 2012, Article ID
916793, 130 pages, 2012.
[33] C. F. Liu, S. S. Kong, and S. J. Yuan, “Reconstructive schemes
for variational iteration method within Yang-Laplace transform
with application to fractal heat conduction problem,” Thermal
Science, vol. 17, no. 3, pp. 715–721, 2013.
[34] G. S. Chen, “Generalizations of hölder’s and some related
integral inequalities on fractal space,” Journal of Function Spaces
and Applications, vol. 2013, Article ID 198405, 9 pages, 2013.
[35] A. Carpinteri, B. Chiaia, and P. Cornetti, “The elastic problem
for fractal media: basic theory and finite element formulation,”
Computers and Structures, vol. 82, no. 6, pp. 499–508, 2004.
[36] A. K. Golmankhaneh, V. Fazlollahi, and D. Baleanu, “Newto-
nian mechanics on fractals subset of real-line,” Romania Reports
in Physics, vol. 65, pp. 84–93, 2013.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 542839, 9 pages
http://dx.doi.org/10.1155/2013/542839

Research Article
New Wavelets Collocation Method for
Solving Second-Order Multipoint Boundary Value Problems
Using Chebyshev Polynomials of Third and Fourth Kinds

W. M. Abd-Elhameed,1,2 E. H. Doha,2 and Y. H. Youssri2


1
Department of Mathematics, Faculty of Science, King Abdulaziz University, Jeddah, Saudi Arabia
2
Department of Mathematics, Faculty of Science, Cairo University, Giza 12613, Egypt

Correspondence should be addressed to W. M. Abd-Elhameed; walee 9@yahoo.com

Received 7 August 2013; Revised 13 September 2013; Accepted 13 September 2013

Academic Editor: Soheil Salahshour

Copyright © 2013 W. M. Abd-Elhameed et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

This paper is concerned with introducing two wavelets collocation algorithms for solving linear and nonlinear multipoint boundary
value problems. The principal idea for obtaining spectral numerical solutions for such equations is employing third- and fourth-
kind Chebyshev wavelets along with the spectral collocation method to transform the differential equation with its boundary
conditions to a system of linear or nonlinear algebraic equations in the unknown expansion coefficients which can be efficiently
solved. Convergence analysis and some specific numerical examples are discussed to demonstrate the validity and applicability of
the proposed algorithms. The obtained numerical results are comparing favorably with the analytical known solutions.

1. Introduction polynomials for solving the differentiated forms of high-


odd-order boundary value problems with the aid of Petrov-
Spectral methods are one of the principal methods of dis- Galerkin method, while in the two papers [10, 11], the authors
cretization for the numerical solution of differential equa- handled third- and fifth-order differential equations using
tions. The main advantage of these methods lies in their Jacobi tau and Jacobi collocation methods.
accuracy for a given number of unknowns (see, e.g., [1– Multipoint boundary value problems (BVPs) arise in a
4]). For smooth problems in simple geometries, they offer variety of applied mathematics and physics. For instance, the
exponential rates of convergence/spectral accuracy. In con- vibrations of a guy wire of uniform cross-section composed
trast, finite difference and finite-element methods yield only of 𝑁 parts of different densities can be set up as a multipoint
algebraic convergence rates. The three most widely used spec- BVP, as in [12]; also, many problems in the theory of
tral versions are the Galerkin, collocation, and tau methods. elastic stability can be handled by the method of multipoint
Collocation methods [5, 6] have become increasingly popular problems [13]. The existence and multiplicity of solutions of
for solving differential equations, also they are very useful in multipoint boundary value problems have been studied by
providing highly accurate solutions to nonlinear differential many authors; see [14–17] and the references therein. For
equations. two-point BVPs, there are many solution methods such as
Many practical problems arising in numerous branches of orthonormalization, invariant imbedding algorithms, finite
science and engineering require solving high even-order and difference, and collocation methods (see, [18–20]). However,
high odd-order boundary value problems. Legendre poly- there seems to be little discussion about numerical solutions
nomials have been previously used for obtaining numerical of multipoint boundary value problems.
spectral solutions for handling some of these kinds of prob- Second-order multipoint boundary value problems (BVP)
lems (see, e.g., [7, 8]). In [9], the author has constructed some arise in the mathematical modeling of deflection of can-
algorithms by selecting suitable combinations of Legendre tilever beams under concentrated load [21, 22], deformation
2 Abstract and Applied Analysis

of beams and plate deflection theory [23], obstacle problems in some form in any scheme of the numerical solution,
[24], Troesch’s problem relating to the confinement of a and it is well known that other numerical methods do not
plasma column by radiation pressure [25, 26], temperature perform well near singularities. Finally, due to their rapid
distribution of the radiation fin of trapezoidal profile [21, convergence, Chebyshev wavelets collocation method does
27], and a number of other engineering applications. Many not suffer from the common instability problems associated
authors have used numerical and approximate methods with other numerical methods.
to solve second-order BVPs. The details about the related The main aim of this paper is to develop two new spectral
numerical methods can be found in a large number of papers algorithms for solving second-order multipoint BVPs based
(see, for instance, [21, 23, 24, 28]). The Walsh wavelets and on shifted third- and fourth-kind Chebyshev wavelets. The
the semiorthogonal B-spline wavelets are used in [23, 29] method reduces the differential equation with its boundary
to construct some numerical algorithms for the solution of conditions to a system of algebraic equations in the unknown
second-order BVPs with Dirichlet and Neumann boundary expansion coefficients. Large systems of algebraic equations
conditions. Na [21] has found the numerical solution of may lead to greater computational complexity and large
second-, third-, and fourth-order BVPs by converting them storage requirements. However the third- and fourth-kind
into initial value problems and then applying a class of Chebyshev wavelets collocation method reduces drastically
methods like nonlinear shooting, method of reduced physical the computational complexity of solving the resulting alge-
parameters, method of invariant imbedding, and so forth. braic system.
The presented approach in this paper can be applied to both The structure of the paper is as follows. In Section 2,
BVPs and IVPs with a slight modification, but without the we give some relevant properties of Chebyshev polynomi-
transformation of BVPs into IVPs or vice versa. als of third and fourth kinds and their shifted ones. In
Wavelets theory is a relatively new and an emerging area Section 3, the third- and fourth-kind Chebyshev wavelets are
in mathematical research. It has been applied to a wide range constructed. Also, in this section, we ascertain the conver-
of engineering disciplines; particularly, wavelets are very gence of the Chebyshev wavelets series expansion. Two new
successfully used in signal analysis for wave form repre- shifted Chebyshev wavelets collocation methods for solving
sentation and segmentations, time frequency analysis, and second-order linear and nonlinear multipoint boundary
fast algorithms for easy implementation. Wavelets permit value problems are implemented and presented in Section 4.
the accurate representation of a variety of functions and In Section 5, some numerical examples are presented to
operators. Moreover, wavelets establish a connection with fast show the efficiency and the applicability of the presented
numerical algorithms, (see [30, 31]). algorithms. Some concluding remarks are given in Section 6.
The application of Legendre wavelets for solving differen-
tial and integral equations is thoroughly considered by many
2. Some Properties of 𝑉𝑘 (𝑥) and 𝑊𝑘 (𝑥)
authors (see, for instance, [32, 33]). Also, Chebyshev wavelets
are used for solving some fractional and integral equations The Chebyshev polynomials 𝑉𝑘 (𝑥) and 𝑊𝑘 (𝑥) of third and
(see, [34, 35]). fourth kinds are polynomials of degree 𝑘 in 𝑥 defined,
Chebyshev polynomials have become increasingly crucial respectively, by (see [38])
in numerical analysis, from both theoretical and practical
points of view. It is well known that there are four kinds cos (𝑘 + (1/2)) 𝜃 sin (𝑘 + (1/2)) 𝜃
of Chebyshev polynomials, and all of them are special cases 𝑉𝑘 (𝑥) = , 𝑊𝑘 (𝑥) = ,
cos (𝜃/2) sin (𝜃/2)
of the more widest class of Jacobi polynomials. The first (1)
and second kinds are special cases of the symmetric Jacobi
polynomials (i.e., ultraspherical polynomials), while the third where 𝑥 = cos 𝜃; also they can be obtained explicitly as
and fourth kinds are special cases of the nonsymmetric Jacobi (𝛼,𝛽)
two particular cases of Jacobi polynomials 𝑃𝑘 (𝑥) for the
polynomials. In the literature, there is a great concentration two nonsymmetric cases correspond to 𝛽 = −𝛼 = ±1/2.
on the first and second kinds of Chebyshev polynomials 𝑇𝑛 (𝑥) Explicitly, we have
and 𝑈𝑛 (𝑥) and their various uses in numerous applications,
(see, for instance, [36]). However, there are few articles that 2
(2𝑘 𝑘!)
concentrate on the other two types of Chebyshev polynomi- 𝑉𝑘 (𝑥) = 𝑃𝑘(−1/2,1/2) (𝑥) ,
als, namely, third and fourth kinds 𝑉𝑛 (𝑥) and 𝑊𝑛 (𝑥), either (2𝑘)!
from theoretical or practical point of view and their uses (2)
𝑘 2
in various applications (see, e.g., [37]). This motivates our (2 𝑘!)
interest in such polynomials. We therefore intend in this work 𝑊𝑘 (𝑥) = 𝑃𝑘(1/2,−1/2) (𝑥) .
(2𝑘)!
to use them in a marvelous application of multipoint BVPs
arising in physics. It is readily seen that
There are several advantages of using Chebyshev wavelets
approximations based on collocation spectral method. First, 𝑊𝑘 (𝑥) = (−1)𝑘 𝑉𝑘 (−𝑥) . (3)
unlike most numerical techniques, it is now well established
that they are characterized by exponentially decaying errors. Hence, it is sufficient to establish properties and relations for
Second, approximation by wavelets handles singularities in 𝑉𝑛 (𝑥) and then deduce their corresponding properties and
the problem. The effect of any such singularities will appear relations for 𝑊𝑛 (𝑥) (by replacing 𝑥 by −𝑥).
Abstract and Applied Analysis 3

The polynomials 𝑉𝑛 (𝑥) and 𝑊𝑛 (𝑥) are orthogonal on wavelet. When the dilation parameter 𝑎 and the translation
(−1, 1); that is, parameter 𝑏 vary continuously, then we have the following
1
family of continuous wavelets:
∫ 𝜔1 (𝑥) 𝑉𝑘 (𝑥) 𝑉𝑗 (𝑥) 𝑑𝑥
−1 𝑡−𝑏
𝜓𝑎,𝑏 (𝑡) = |𝑎|−1/2 𝜓 ( ), 𝑎, 𝑏 ∈ R, 𝑎 ≠
0. (13)
1 𝑎
= ∫ 𝜔2 (𝑥) 𝑊𝑘 (𝑥) 𝑊𝑗 (𝑥) 𝑑𝑥 (4)
−1 Each of the third- and fourth-kind Chebyshev wavelets
𝜓𝑛𝑚 (𝑡) = 𝜓(𝑘, 𝑛, 𝑚, 𝑡) has four arguments: 𝑘, 𝑛 ∈ N, 𝑚 is
𝜋, 𝑘 = 𝑗, the order of the polynomial 𝑉𝑚∗ (𝑡) or 𝑊𝑚∗ (𝑡), and 𝑡 is the
={
0, 𝑘 ≠
𝑗, normalized time. They are defined explicitly on the interval
[0, 1] as
where
1+𝑥 1−𝑥 𝜓𝑛𝑚 (𝑡)
𝜔1 (𝑥) = √ , 𝜔2 (𝑥) = √ , (5)
1−𝑥 1+𝑥 2 (𝑘+1)/2
{
{ 𝑉∗ (2𝑘 𝑡 − 𝑛) ,
{
{ √𝜋 𝑚
and they may be generated by using the two recurrence {
{
{
{
relations {
{
resp.,
{
{
{ 2(𝑘+1)/2 𝑛 𝑛+1
𝑉𝑘 (𝑥) = 2𝑥𝑉𝑘−1 (𝑥) − 𝑉𝑘−2 (𝑥) , 𝑘 = 2, 3, . . . , (6) ={ 𝑊𝑚∗ (2𝑘 𝑡 − 𝑛) , 𝑡 ∈ [ 𝑘 , 𝑘 ] , 0 ⩽ 𝑚 ⩽ 𝑀,
{
{ √𝜋 2 2
{
{
with the initial values {
{ 0 ⩽ 𝑛 ⩽ 2𝑘 − 1,
{
{
{
{
𝑉0 (𝑥) = 1, 𝑉1 (𝑥) = 2𝑥 − 1, (7) {
{0, otherwise.
{
𝑊𝑘 (𝑥) = 2𝑥 𝑊𝑘−1 (𝑥) − 𝑊𝑘−2 (𝑥) , 𝑘 = 2, 3, . . . , (8) (14)
with the initial values
3.1. Function Approximation. A function 𝑓(𝑡) defined over
𝑊0 (𝑥) = 1, 𝑊1 (𝑥) = 2𝑥 + 1. (9) [0, 1] may be expanded in terms of Chebyshev wavelets as

The shifted Chebyshev polynomials of third and fourth ∞ ∞

kinds are defined on [0, 1], respectively, as 𝑓 (𝑡) = ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝑡) , (15)
𝑛=0 𝑚=0
𝑉𝑛∗ (𝑥) = 𝑉𝑛 (2𝑥 − 1) , 𝑊𝑛∗ (𝑥) = 𝑊𝑛 (2𝑥 − 1) . (10)
where
All results of Chebyshev polynomials of third and fourth 1
kinds can be easily transformed to give the corresponding 𝑐𝑛𝑚 = (𝑓 (𝑡) , 𝜓𝑛𝑚 (𝑡))𝜔∗ = ∫ 𝜔𝑖∗ 𝑓 (𝑡) 𝜓𝑛𝑚 (𝑡) 𝑑𝑡, (16)
𝑖
results for their shifted ones. 0
The orthogonality relations of 𝑉𝑛∗ (𝑡) and 𝑊𝑛∗ (𝑡) on [0, 1]
are given by and the weights 𝑤𝑖∗ , 𝑖 = 1, 2, are given in (12).
Assume that 𝑓(𝑡) can be approximated in terms of
1 Chebyshev wavelets as
∫ 𝑤1∗ 𝑉𝑚∗ (𝑡) 𝑉𝑛∗ (𝑡) 𝑑𝑡
0
2𝑘 −1 𝑀
1 𝑓 (𝑡) ≃ ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝑡) . (17)
= ∫ 𝑤2∗ 𝑊𝑚∗ (𝑡) 𝑊𝑛∗ (𝑡) 𝑑𝑡 (11) 𝑛=0 𝑚=0
0

𝜋
{ , 𝑚 = 𝑛, 3.2. Convergence Analysis. In this section, we state and prove
= {2 a theorem to ascertain that the third- and fourth-kind Cheby-
0, 𝑚 ≠
𝑛, shev wavelets expansion of a function 𝑓(𝑡), with bounded
{
second derivative, converges uniformly to 𝑓(𝑡).
where
𝑡 1−𝑡 Theorem 1. Assume that a function 𝑓(𝑡) ∈ 𝐿2𝜔∗ [0, 1], 𝜔1∗ =
𝜔1∗ =√ 𝜔2∗ =√
1
, . (12)
√𝑡/(1 − 𝑡) with |𝑓󸀠󸀠 (𝑡)| ⩽ 𝐿, can be expanded as an infinite
1−𝑡 𝑡
series of third-kind Chebyshev wavelets; then this series con-
3. Shifted Third- and Fourth-Kind verges uniformly to 𝑓(𝑡). Explicitly, the expansion coefficients
in (16) satisfy the following inequality:
Chebyshev Wavelets
Wavelets constitute of a family of functions constructed from 󵄨󵄨 󵄨󵄨 2√2𝜋 𝐿𝑚2
󵄨󵄨𝑐𝑛𝑚 󵄨󵄨 < , ∀𝑛 ⩾ 0, 𝑚 > 1. (18)
dilation and translation of single function called the mother (𝑛 + 1)5/2 (𝑚4 − 1)
4 Abstract and Applied Analysis

Proof. From (16), it follows that Finally, since 𝑛 ⩽ 2𝑘 − 1, we have

2(𝑘+1)/2 (𝑛+1)/2
𝑘
󵄨󵄨 󵄨󵄨 2√2𝜋 𝐿𝑚2
𝑐𝑛𝑚 = ∫ 𝑓 (𝑡) 𝑉𝑚∗ (2𝑘 𝑡 − 𝑛) 𝜔1∗ (2𝑘 𝑡 − 𝑛) 𝑑𝑡. 󵄨󵄨𝑐𝑛𝑚 󵄨󵄨 < . (24)
√𝜋 𝑛/2𝑘 (𝑛 + 1)5/2 (𝑚4 − 1)
(19)
Remark 2. The estimation in (18) is also valid for the coef-
If we make use of the substitution 2𝑘 𝑡 − 𝑛 = cos 𝜃 in (19), then ficients of fourth-kind Chebyshev wavelets expansion. The
we get proof is similar to the proof of Theorem 1.

2(−𝑘+1)/2 𝜋 cos 𝜃 + 𝑛
𝑐𝑛𝑚 = ∫ 𝑓( ) 4. Solution of Multipoint BVPs
√𝜋 0 2𝑘
In this section, we present two Chebyshev wavelets col-
cos (𝑚 + (1/2)) 𝜃 √ 1 + cos 𝜃 location methods, namely, third-kind Chebyshev wavelets
× sin 𝜃𝑑𝜃
cos (𝜃/2) 1 − cos 𝜃 collocation method (3CWCM) and fourth-kind Chebyshev
wavelets collocation method (4CWCM), to numerically solve
2(−𝑘+3)/2 𝜋 cos 𝜃 + 𝑛 1 𝜃 the following multipoint boundary value problem (BVP):
= ∫ 𝑓( ) cos (𝑚 + ) 𝜃 cos ( ) 𝑑𝜃
√𝜋 0 2 𝑘 2 2
𝑎 (𝑥) 𝑦󸀠󸀠 (𝑥) + 𝑏 (𝑥) 𝑦󸀠 (𝑥) + 𝑐 (𝑥) 𝑦 (𝑥)
(−𝑘+1)/2 𝜋
2 cos 𝜃 + 𝑛 (25)
= ∫ 𝑓( )
√𝜋 0 2𝑘 + 𝑓 (𝑥, 𝑦󸀠 ) + 𝑔 (𝑥, 𝑦) = 0, 0 ⩽ 𝑥 ⩽ 1,

× [cos (𝑚 + 1) 𝜃 + cos 𝑚𝜃] 𝑑𝜃, 𝑚0

(20) 𝛼0 𝑦 (0) + 𝛼1 𝑦󸀠 (0) = ∑𝜆 𝑖 𝑦 (𝜉𝑖 ) + 𝛿0 ,


𝑖=1
(26)
which in turn, and after performing integration by parts two 𝑚1
󸀠
times, yields 𝛽0 𝑦 (1) + 𝛽1 𝑦 (1) = ∑𝜇𝑖 𝑦 (𝜂𝑖 ) + 𝛿1 ,
𝑖=1
𝜋
1 cos 𝜃 + 𝑛
𝑐𝑛𝑚 = 5𝑘/2 ∫ 𝑓󸀠󸀠 ( ) 𝛾𝑚 (𝜃) 𝑑𝜃, (21) where 𝑎(𝑥), 𝑏(𝑥), and 𝑐(𝑥) are piecewise continuous on [0, 1];
2 √ 2𝜋 0 2𝑘
𝑎(0)𝑎(1) may equal zero; 0 < 𝜉𝑖 ; 𝜂𝑖 < 1; 𝛼𝑖 , 𝛽𝑖 , 𝜆 𝑖 , 𝜇𝑖 , and 𝛿𝑖
where are constants such that (𝛼02 + 𝛼12 )(𝛽02 + 𝛽12 ) ≠
0; 𝑓 is a nonlinear
function of 𝑦󸀠 , and 𝑔 is a nonlinear function in 𝑦.
1 sin 𝑚𝜃 sin (𝑚 + 2) 𝜃 Consider an approximate solution to (25) and (26) which
𝛾𝑚 (𝜃) = ( − )
𝑚+1 𝑚 𝑚+2 is given in terms of Chebyshev wavelets as
(22)
1 sin (𝑚 − 1) 𝜃 sin (𝑚 + 1) 𝜃 2𝑘 −1 𝑀
+ ( − ).
𝑚 𝑚−1 𝑚+1 𝑦𝑘,𝑀 (𝑥) = ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝑥) ; (27)
𝑛=0 𝑚=0
Now, we have
then the substitution of (27) into (25) enables one to write the
󵄨 󵄨󵄨
󵄨󵄨 󵄨󵄨 󵄨󵄨󵄨
𝜋
1 cos 𝜃 + 𝑛 󵄨󵄨 residual of (25) in the form
󵄨󵄨𝑐𝑛𝑚 󵄨󵄨 = 󵄨󵄨 5𝑘/2 ∫ 𝑓󸀠󸀠 ( ) 𝛾𝑚 (𝜃) 𝑑𝜃 󵄨󵄨
󵄨󵄨 2 √2𝜋 0 2𝑘 󵄨󵄨
2𝑘 −1 𝑀
1 󵄨󵄨 𝜋 󵄨󵄨 󸀠󸀠
󵄨󵄨 󸀠󸀠 cos 𝜃 + 𝑛 󵄨󵄨 𝑅 (𝑥) = ∑ ∑ 𝑐𝑛𝑚 𝑎 (𝑥) 𝜓𝑛𝑚 (𝑥)
= 5𝑘/2 󵄨󵄨∫ 𝑓 ( ) 𝛾𝑚 (𝜃) 𝑑𝜃 󵄨󵄨
2 √2𝜋 󵄨󵄨 0 2𝑘 󵄨󵄨 𝑛=0 𝑚=2

𝜋 2𝑘 −1 𝑀
𝐿 󵄨󵄨 󵄨
⩽ ∫ 󵄨𝛾𝑚 (𝜃)󵄨󵄨󵄨 𝑑𝜃 󸀠
+ ∑ ∑ 𝑐𝑛𝑚 𝑏 (𝑥) 𝜓𝑛𝑚 (𝑥)
25𝑘/2 √2𝜋 0 󵄨 𝑛=0 𝑚=1
𝐿√𝜋 1 1 1 2𝑘 −1 𝑀
⩽ [ ( + ) (23)
2(5𝑘+1)/2 𝑚+1 𝑚 𝑚+2 + ∑ ∑ 𝑐𝑛𝑚 𝑐 (𝑥) 𝜓𝑛𝑚 (𝑥) (28)
1 1 1 𝑛=0 𝑚=0
+ ( + )]
𝑚 𝑚−1 𝑚+1 2𝑘 −1 𝑀
󸀠
+ 𝑓 (𝑥, ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝑥))
𝐿√2𝜋 1 1
= 5𝑘/2 [ 2 + ] 𝑛=0 𝑚=1
2 𝑚 + 2𝑚 𝑚2 − 1
2𝑘 −1 𝑀
2𝐿√2𝜋 𝑚2 + 𝑔 (𝑥, ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝑥)) .
< 5𝑘/2 ( 4 ).
2 𝑚 −1 𝑛=0 𝑚=0
Abstract and Applied Analysis 5

Now, the application of the typical collocation method (see, Example 2. Consider the second-order linear BVP (see, [39,
e.g., [5]) gives 40]):

𝑦󸀠󸀠 = sinh 𝑥 − 2, 0 < 𝑥 < 1,


𝑅 (𝑥𝑖 ) = 0, 𝑖 = 1, 2, . . . , 2𝑘 (𝑀 + 1) − 2, (29)
3 (32)
𝑦󸀠 (0) = 0, 𝑦 (1) = 3𝑦 ( ) .
5
where 𝑥𝑖 are the first (2𝑘 (𝑀 + 1) − 2) roots of 𝑉2∗𝑘 (𝑀+1) (𝑥) or
𝑊2∗𝑘 (𝑀+1) (𝑥). Moreover, the use of the boundary conditions The exact solution of problem (32) is given by
(26) gives 1 3 11
𝑦 (𝑥) = (sinh 1 − 3 sinh ) + sinh 𝑥 − 𝑥2 − 𝑥 + .
2 5 25
2𝑘 −1 𝑀 (33)
𝛼0 ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (0)
𝑛=0 𝑚=0 In Table 2, the maximum absolute error 𝐸 is listed for
𝑘 = 1 and various values of 𝑀, while in Table 3, we
2𝑘 −1 𝑀 give a comparison between the best errors resulted from
󸀠
+ 𝛼1 ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (0) the application of various methods for Example 2, while in
𝑛=0 𝑚=1
Figure 1, we give a comparison between the exact solution of
𝑘
𝑚0 2 −1 𝑀 (32) with three approximate solutions.
= ∑ ∑ ∑ 𝜆 𝑖 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝜉𝑖 ) + 𝛿0 , (30)
𝑖=1 𝑛=0 𝑚=0 Example 3. Consider the second-order singular nonlinear
BVP (see [40, 41]):
2𝑘 −1 𝑀 2𝑘 −1 𝑀
󸀠
𝛽0 ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (1) + 𝛽1 ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (1) 𝑥 (1 − 𝑥) 𝑦󸀠󸀠 + 6𝑦󸀠 + 2𝑦 + 𝑦2
𝑛=0 𝑚=0 𝑛=0 𝑚=1

𝑚1 2𝑘 −1 𝑀
= 6 cosh 𝑥 + (2 + 𝑥 − 𝑥2 + sinh 𝑥) sinh 𝑥,
= ∑ ∑ ∑ 𝜇𝑖 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝜂𝑖 ) + 𝛿1 . 0 < 𝑥 < 1,
𝑖=1 𝑛=0 𝑚=0 (34)
2 2
𝑦 (0) + 𝑦 ( ) = sinh ( ) ,
Equations (29) and (30) generate 2𝑘 (𝑀 + 1) equations in 3 3
the unknown expansion coefficients, 𝑐𝑛𝑚 , which can be 1 4 1 4
solved with the aid of the well-known Newton’s iterative 𝑦 (1) + 𝑦 ( ) = sinh ( ) + sinh 1,
2 5 2 5
method. Consequently, we get the desired approximate solu-
tion 𝑦𝑘,𝑀(𝑥) given by (27). with the exact solution 𝑦(𝑥) = sinh 𝑥. In Table 4, the maxi-
mum absolute error 𝐸 is listed for 𝑘 = 0 and various values
of 𝑀, while in Table 5 we give a comparison between the
5. Numerical Examples best errors resulted from the application of various methods
for Example 3. This table shows that our two algorithms are
In this section, the presented algorithms in Section 4 are
more accurate if compared with the two methods developed
applied to solve both of linear and nonlinear multipoint
in [40, 41].
BVPs. Some examples are considered to illustrate the effi-
ciency and applicability of the two proposed algorithms. Example 4. Consider the second-order nonlinear BVP (see
[42]):
Example 1. Consider the second-order nonlinear BVP (see
[6, 28]): 𝑦󸀠󸀠 + (1 + 𝑥 + 𝑥3 ) 𝑦2 = 𝑓 (𝑥) , 0 < 𝑥 < 1,

3 2 2 1 2 1 7
𝑦󸀠󸀠 + 𝑦 + (𝑦󸀠 ) + 1 = 0, 0 < 𝑥 < 1, 𝑦 (0) = 𝑦 ( ) + 𝑦 ( ) − 0.0286634, (35)
8 1089 6 9 3 9
(31)
1 1 2 1 7
𝑦 (0) = 0, 𝑦 ( ) = 𝑦 (1) . 𝑦 (1) = 𝑦 ( ) + 𝑦 ( ) − 0.0401287,
3 5 9 2 9
where
The two proposed methods are applied to the problem for
the case corresponding to 𝑘 = 0 and 𝑀 = 8. The numerical 1
𝑓 (𝑥) = [ − 6 cos (𝑥 − 𝑥2 ) + sin (𝑥 − 𝑥2 )
solutions are shown in Table 1. Due to nonavailability of the 9
exact solution, we compare our results with Haar wavelets
method [6], ADM solution [28] and ODEs Solver from × (−3(1 − 2𝑥)2 + (1 + 𝑥 + 𝑥3 ) sin (𝑥 − 𝑥2 ))] .
Mathematica which is carried out by using Runge-Kutta
method. This comparison is also shown in Table 1. (36)
6 Abstract and Applied Analysis

0.070 0.5
0.050
0.4
0.030
0.3
ln|y(x)|

0.020

y(x)
0.015
0.2
0.010
0.1

0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
x x

Exact k = 0, M = 3 Exact k = 1, M = 2
k = 0, M = 4 k = 0, M = 2 k = 1, M = 3 k = 1, M = 1

Figure 1: Different solutions of Example 2. Figure 2: Different solutions of Example 5.

The exact solution of (35) is given by 𝑦(𝑥) = (1/3) sin(𝑥 − to obtain an approximate solution of 𝑦(𝑥). If we make use of
𝑥2 ). In Table 6, the maximum absolute error 𝐸 is listed for (27), then the approximate solution 𝑦0,3 (𝑥) can be expanded
𝑘 = 2 and various values of 𝑀, and in Table 7 we give a in terms of third-kind Chebyshev wavelets as
comparison between best errors resulted from the application
of various methods for Example 4. This table shows that 2 2
𝑦0,3 (𝑥) = 𝑐0,0 √ + 𝑐 √ (4𝑥 − 3)
our two algorithms are more accurate if compared with the 𝜋 0,1 𝜋
method developed in [42]. (42)
2
+ 𝑐0,2 √ (16𝑥2 − 20𝑥 + 5) .
Example 5. Consider the second-order singular linear BVP: 𝜋
𝑦󸀠󸀠 + 𝑓 (𝑥) 𝑦 = 𝑔 (𝑥) , 0 < 𝑥 < 1, If we set

1 2
𝑦 (0) + 16𝑦 ( ) = 3√4 𝑒, V𝑖 = √ 𝑐 , 𝑖 = 0, 1, 2, (43)
4 (37) 𝜋 0,𝑖
3 then (42) reduces to the form
𝑦 (1) + 16𝑦 ( ) = 3√𝑒3 ,
4

4
𝑦0,3 (𝑥) = V0 + V1 (4𝑥 − 3) + V2 (16𝑥2 − 20𝑥 + 5) . (44)
where
1 If we substitute (44) into (39), then the residual of (39) is
{
{3𝑥, 0 ⩽ 𝑥 ⩽ ,
𝑓 (𝑥) = { 2 (38) given by
{2𝑥, 1 < 𝑥 ⩽ 1,
2
{ 2 𝑅 (𝑥) = 2V2 + [V1 + V2 (8𝑥 − 5)]
and 𝑔(𝑥) is chosen such that the exact solution of (37) is
𝑦(𝑥) = 𝑥(1 − 𝑥)𝑒𝑥 . In Table 8, the maximum absolute error 𝐸 − 4 [V0 + V1 (4𝑥 − 3) (45)
is listed for 𝑘 = 1 and various values of 𝑀, while in Figure 2,
we give a comparison between the exact solution of (37) with + V2 (16𝑥2 − 20𝑥 + 5)] − 2.
three approximate solutions.
We enforce the residual to vanish at the first root of 𝑉3∗ (𝑥) =
Example 6. Consider the following nonlinear second-order 64𝑥3 −112𝑥2 +56𝑥−7, namely, at 𝑥1 = 0.18825509907063323,
BVP: to get
2
𝑦󸀠󸀠 + (𝑦󸀠 ) − 64𝑦 = 32, 0 < 𝑥 < 1, (39) 6.10388V22 + 0.5V12 − 3.49396V2 V1 − 2.60388V2
(46)
1 + 4.49396V1 − 2V0 = 1.
𝑦 (0) + 𝑦 ( ) = 1, (40)
4
Furthermore, the use of the boundary conditions (40) and
1 (41) yields
4𝑦 ( ) − 𝑦 (1) = 0, (41)
2
2V0 − 5V1 + 6V2 = 1,
with the exact solution 𝑦(𝑥) = 16𝑥2 . We solve (39) using (47)
3CWCM for the case corresponding to 𝑘 = 0 and 𝑀 = 3, 3V0 − 5V1 − 5V2 = 0.
Abstract and Applied Analysis 7

Table 1: Comparison between different solutions for Example 1.

𝑥 3CWCM 4CWCM Haar method [6] ADM method [28] ODEs solver from Mathematica
0.1 0.06560 0.06560 0.06561 0.0656 0.06560
0.3 0.16587 0.16587 0.16588 0.1658 0.16587
0.5 0.22369 0.22369 0.22369 0.2236 0.22369
0.7 0.23820 0.23820 0.23821 0.2382 0.23820
0.9 0.20920 0.20920 0.20910 0.2092 0.20920

Table 2: The maximum absolute error 𝐸 for Example 2.

𝑘 𝑀 10 11 12 13 14 15
3CWCM 5.173 ⋅ 10−9 7.184 ⋅ 10−11 4.418 ⋅ 10−12 5.018 ⋅ 10−14 2.442 ⋅ 10−15 2.220 ⋅ 10−16
0
4CWCM 3.324 ⋅ 10−10 2.533 ⋅ 10−11 2.811 ⋅ 10−12 1.665 ⋅ 10−14 1.110 ⋅ 10−15 2.220 ⋅ 10−16
𝑀 4 5 6 7 8 9
3CWCM 4.644 ⋅ 10−6 1.001 ⋅ 10−8 1.840 ⋅ 10−9 2.547 ⋅ 10−10 6.247 ⋅ 10−11 5.681 ⋅ 10−12
1
4CWCM 2.15 ⋅ 10−6 5.247 ⋅ 10−9 7.548 ⋅ 10−10 1.004 ⋅ 10−10 2.154 ⋅ 10−11 4.257 ⋅ 10−12

Table 3: The best errors for Example 2.

Method in [39] Method in [40] 3CWCM 4CWCM


9.00 ⋅ 10−6 4.00 ⋅ 10−5 2.22 ⋅ 10−16 2.22 ⋅ 10−16

Table 4: The maximum absolute error 𝐸 for Example 3.

𝑀 8 9 10 11 12 13
3CWCM 4.444 ⋅ 10−10 2.759 ⋅ 10−11 4.389 ⋅ 10−13 1.643 ⋅ 10−14 4.441 ⋅ 10−16 2.220 ⋅ 10−16
4CWCM 4.478 ⋅ 10−9 2.179 ⋅ 10−10 1.527 ⋅ 10−12 2.975 ⋅ 10−14 4.441 ⋅ 10−16 2.220 ⋅ 10−16

Table 5: The best errors for Example 3.

Method in [40] Method in [41] 3CWCM 4CWCM


3.00 ⋅ 10−8 6.60 ⋅ 10−7 2.22 ⋅ 10−16 2.22 ⋅ 10−16

Table 6: The maximum absolute error 𝐸 for Example 4.

𝑀 4 5 6 7 8 9
3CWCM 2.881 ⋅ 10−3 3.441 ⋅ 10−3 1.212 ⋅ 10−5 1.933 ⋅ 10−5 1.990 ⋅ 10−6 3.010 ⋅ 10−8
4CWCM 1.241 ⋅ 10−3 8.542 ⋅ 10−4 7.526 ⋅ 10−6 1.002 ⋅ 10−6 6.321 ⋅ 10−7 2.354 ⋅ 10−9

Table 7: The best errors for Example 4.

Method in [42] 3CWCM 4CWCM


8.00 ⋅ 10−6 3.010 ⋅ 10−8 2.35 ⋅ 10−9

Table 8: The maximum absolute error 𝐸 for Example 5.

𝑀 7 8 9 10 11 12 13 14
3CWCM 7.1 ⋅ 10−6 3.2 ⋅ 10−7 1.3 ⋅ 10−8 7.4 ⋅ 10−10 1.3 ⋅ 10−11 3.5 ⋅ 10−13 9.3 ⋅ 10−15 8.1 ⋅ 10−16
4CWCM 2.8 ⋅ 10−5 1.5 ⋅ 10−6 6.3 ⋅ 10−8 2.4 ⋅ 10−9 7.8 ⋅ 10−11 2.3 ⋅ 10−12 6.3 ⋅ 10−14 1.7 ⋅ 10−15
8 Abstract and Applied Analysis

The solution of the nonlinear system of (46) and (47) gives [6] Siraj-ul-Islam, I. Aziz, and B. Šarler, “The numerical solution of
second-order boundary-value problems by collocation method
V0 = 10, V1 = 5, V2 = 1, (48) with the Haar wavelets,” Mathematical and Computer Modelling,
vol. 52, no. 9-10, pp. 1577–1590, 2010.
and consequently [7] A. H. Bhrawy, A. S. Alofi, and S. I. El-Soubhy, “An exten-
sion of the Legendre-Galerkin method for solving sixth-order
𝑦0,3 (𝑥) = 10 + 5 (4𝑥 − 3) + (16𝑥2 − 20𝑥 + 5) = 16𝑥2 , (49) differential equations with variable polynomial coefficients,”
Mathematical Problems in Engineering, vol. 2012, Article ID
896575, 13 pages, 2012.
which is the exact solution. [8] E. H. Doha and W. M. Abd-Elhameed, “Efficient solutions of
multidimensional sixth-order boundary value problems using
Remark 3. It is worth noting here that the obtained numerical symmetric generalized Jacobi-Galerkin method,” Abstract and
results in the previous solved six examples are very accurate, Applied Analysis, vol. 2012, Article ID 749370, 19 pages, 2012.
although the number of retained modes in the spectral [9] W. M. Abd-Elhameed, “Efficient spectral Legendre dual-Petrov-
expansion is very few, and again the numerical results are Galerkin algorithms for the direct solution of (2𝑛 + 1)th-
comparing favorably with the known analytical solutions. order linear differential equations,” Journal of the Egyptian
Mathematical Society, vol. 17, no. 2, pp. 189–211, 2009.
6. Concluding Remarks [10] A. H. Bhrawy and W. M. Abd-Elhameed, “New algorithm for
the numerical solutions of nonlinear third-order differential
In this paper, two algorithms for obtaining numerical spectral equations using Jacobi-Gauss collocation method,” Mathemat-
wavelets solutions for second-order multipoint linear and ical Problems in Engineering, vol. 2011, Article ID 837218, 14
nonlinear boundary value problems are analyzed and dis- pages, 2011.
cussed. Chebyshev polynomials of third and fourth kinds [11] A. H. Bhrawy, A. S. Alofi, and S. I. El-Soubhy, “Spectral shifted
are used. One of the advantages of the developed algorithms Jacobi tau and collocation methods for solving fifth-order
is their availability for application on singular boundary boundary value problems,” Abstract and Applied Analysis, vol.
2011, Article ID 823273, 14 pages, 2011.
value problems. Another advantage is that high accurate
approximate solutions are achieved using a few number of [12] M. Moshiinsky, “Sobre los problemas de condiciones a la
terms of the approximate expansion. The obtained numerical frontiera en una dimension de caracteristicas discontinuas,”
Boletin de La Sociedad Matematica Mexicana, vol. 7, article 125,
results are comparing favorably with the analytical ones.
1950.
[13] S. P. Timoshenko, Theory of Elastic Stability, McGraw-Hill Book,
Acknowledgment New York, NY, USA, 2nd edition, 1961.
[14] R. P. Agarwal and I. Kiguradze, “On multi-point boundary value
The authors would like to thank the referee for his valuable problems for linear ordinary differential equations with singu-
comments and suggestions which improved the paper in its larities,” Journal of Mathematical Analysis and Applications, vol.
present form. 297, no. 1, pp. 131–151, 2004.
[15] Z. Du, “Solvability of functional differential equations with
References multi-point boundary value problems at resonance,” Computers
& Mathematics with Applications, vol. 55, no. 11, pp. 2653–2661,
[1] E. H. Doha, W. M. Abd-Elhameed, and Y. H. Youssri, “Efficient 2008.
spectral-Petrov-Galerkin methods for the integrated forms [16] W. Feng and J. R. L. Webb, “Solvability of 𝑚-point boundary
of third- and fifth-order elliptic differential equations using value problems with nonlinear growth,” Journal of Mathematical
general parameters generalized Jacobi polynomials,” Applied Analysis and Applications, vol. 212, no. 2, pp. 467–480, 1997.
Mathematics and Computation, vol. 218, no. 15, pp. 7727–7740, [17] H. B. Thompson and C. Tisdell, “Three-point boundary value
2012. problems for second-order, ordinary, differential equations,”
[2] W. M. Abd-Elhameed, E. H. Doha, and Y. H. Youssri, “Efficient Mathematical and Computer Modelling, vol. 34, no. 3-4, pp. 311–
spectral-Petrov-Galerkin methods for third- and fifth-order 318, 2001.
differential equations using general parameters generalized [18] M. R. Scott and H. A. Watts, “SUPORT—a computer code for
Jacobi polynomials,” Quaestiones Mathematicae, vol. 36, no. 1, two-point boundary-value problems via orthonormalization,”
pp. 15–38, 2013. Sandia Labs Report 75-0198, Sandia Laboratories, Albuquerque,
[3] E. H. Doha, W. M. Abd-Elhameed, and A. H. Bhrawy, “New NM, USA, 1975.
spectral-Galerkin algorithms for direct solution of high even- [19] M. R. Scott and H. A. Watts, “Computational solution of linear
order differential equations using symmetric generalized Jacobi two-point boundary value problems via orthonormalization,”
polynomials,” Collectanea Mathematica, vol. 64, no. 3, pp. 373– SIAM Journal on Numerical Analysis, vol. 14, no. 1, pp. 40–70,
394, 2013. 1977.
[4] E. H. Doha, A. H. Bhrawy, and R. M. Hafez, “On shifted Jacobi [20] M. R. Scott and W. H. Vandevender, “A comparison of several
spectral method for high-order multi-point boundary value invariant imbedding algorithms for the solution of two-point
problems,” Communications in Nonlinear Science and Numerical boundary-value problems,” Applied Mathematics and Compu-
Simulation, vol. 17, no. 10, pp. 3802–3810, 2012. tation, vol. 1, no. 3, pp. 187–218, 1975.
[5] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, [21] T. Y. Na, Computational Methods in Engineering Boundary Value
Spectral Methods, Springer, Berlin, Germany, 2006. Problems, vol. 145, Academic Press, New York, NY, USA, 1979.
Abstract and Applied Analysis 9

[22] K. E. Bisshopp and D. C. Drucker, “Large deflection of cantilever [40] F. Geng, “Solving singular second order three-point bound-
beams,” Quarterly of Applied Mathematics, vol. 3, pp. 272–275, ary value problems using reproducing kernel Hilbert space
1945. method,” Applied Mathematics and Computation, vol. 215, no.
[23] W. Glabisz, “The use of Walsh-wavelet packets in linear bound- 6, pp. 2095–2102, 2009.
ary value problems,” Computers & Structures, vol. 82, no. 2-3, [41] F. Z. Geng, “A numerical algorithm for nonlinear multi-
pp. 131–141, 2004. point boundary value problems,” Journal of Computational and
[24] Siraj-ul-Islam, M. A. Noor, I. A. Tirmizi, and M. A. Khan, Applied Mathematics, vol. 236, no. 7, pp. 1789–1794, 2012.
“Quadratic non-polynomial spline approach to the solution of [42] A. Saadatmandi and M. Dehghan, “The use of sinc-collocation
a system of second-order boundary-value problems,” Applied method for solving multi-point boundary value problems,”
Mathematics and Computation, vol. 179, no. 1, pp. 153–160, 2006. Communications in Nonlinear Science and Numerical Simula-
[25] S. Robert and J. Shipman, “Solution of Troesch’s two-point tion, vol. 17, no. 2, pp. 593–601, 2012.
boundary value problems by shooting techniques,” Journal of
Computational Physics, vol. 10, pp. 232–241, 1972.
[26] E. Wiebel, Confinement of a Plasma Column by Radiation
Pressure in the Plasma in a Magnetic Field, Stanford University
Press, Stanfor, Calif, USA, 1958.
[27] H. H. Keller and E. S. Holdrege, “Radiation heat transfer for
annular fins of trapezoidal profile,” International Journal of High
Performance Computing Applications, vol. 92, pp. 113–116, 1970.
[28] M. Tatari and M. Dehgan, “The use of the Adomian decomposi-
tion method for solving multipoint boundary value problems,”
Physica Scripta, vol. 73, pp. 672–676, 2006.
[29] M. Lakestani and M. Dehghan, “The solution of a second-order
nonlinear differential equation with Neumann boundary con-
ditions using semi-orthogonal B-spline wavelets,” International
Journal of Computer Mathematics, vol. 83, no. 8-9, pp. 685–694,
2006.
[30] A. Constantmldes, Applied Numerical Methods with Personal
Computers, McGraw-Hill, New York, NY, USA , 1987.
[31] D. E. Newland, An Introduction to Random Vibrations, Spectral
and Wavelet Analysis, Longman Scientific and Technical, New
York, NY, USA, 1993.
[32] M. Razzaghi and S. Yousefi, “Legendre wavelets method for the
solution of nonlinear problems in the calculus of variations,”
Mathematical and Computer Modelling, vol. 34, no. 1-2, pp. 45–
54, 2001.
[33] M. Razzaghi and S. Yousefi, “Legendre wavelets method for
constrained optimal control problems,” Mathematical Methods
in the Applied Sciences, vol. 25, no. 7, pp. 529–539, 2002.
[34] E. Babolian and F. Fattahzadeh, “Numerical solution of differ-
ential equations by using Chebyshev wavelet operational matrix
of integration,” Applied Mathematics and Computation, vol. 188,
no. 1, pp. 417–426, 2007.
[35] L. Zhu and Q. Fan, “Solving fractional nonlinear Fredholm
integro-differential equations by the second kind Chebyshev
wavelet,” Communications in Nonlinear Science and Numerical
Simulation, vol. 17, no. 6, pp. 2333–2341, 2012.
[36] E. H. Doha, W. M. Abd- Elhameed, and Y. H. Youssri, “Second
kind Chebyshev operational matrix algorithm for solving dif-
ferential equations of Lane-Emden type,” New Astronomy, vol.
23-24, pp. 113–117, 2013.
[37] E. H. Doha, W. M. Abd-Elhameed, and M. A. Bassuony, “New
algorithms for solving high even-order differential equations
using third and fourth Chebyshev-Galerkin methods,” Journal
of Computational Physics, vol. 236, pp. 563–579, 2013.
[38] J. C. Mason and D. C. Handscomb, Chebyshev Polynomials,
Chapman & Hall, New York, NY, USA, 2003.
[39] Y. Lin, J. Niu, and M. Cui, “A numerical solution to nonlinear
second order three-point boundary value problems in the
reproducing kernel space,” Applied Mathematics and Computa-
tion, vol. 218, no. 14, pp. 7362–7368, 2012.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 513808, 9 pages
http://dx.doi.org/10.1155/2013/513808

Research Article
Numerical Solution of a Class of Functional-Differential
Equations Using Jacobi Pseudospectral Method

A. H. Bhrawy,1,2 M. A. Alghamdi,1 and D. Baleanu3,4,5


1
Department of Mathematics, Faculty of Science, King Abdulaziz University, Jeddah 21589, Saudi Arabia
2
Department of Mathematics, Faculty of Science, Beni-Suef University, Beni-Suef 62511, Egypt
3
Department of Chemical and Materials Engineering, Faculty of Engineering, King Abdulaziz University, Jeddah 21589, Saudi Arabia
4
Department of Mathematics and Computer Sciences, Cankaya University, Eskisehir Yolu 29.km, 06810 Ankara, Turkey
5
Institute of Space Sciences, P.O. Box MG-23, 76900 Magurele-Bucharest, Romania

Correspondence should be addressed to A. H. Bhrawy; alibhrawy@yahoo.co.uk

Received 24 August 2013; Accepted 18 September 2013

Academic Editor: Soheil Salahshour

Copyright © 2013 A. H. Bhrawy et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The shifted Jacobi-Gauss-Lobatto pseudospectral (SJGLP) method is applied to neutral functional-differential equations (NFDEs)
with proportional delays. The proposed approximation is based on shifted Jacobi collocation approximation with the nodes of
Gauss-Lobatto quadrature. The shifted Legendre-Gauss-Lobatto Pseudo-spectral and Chebyshev-Gauss-Lobatto Pseudo-spectral
methods can be obtained as special cases of the underlying method. Moreover, the SJGLP method is extended to numerically
approximate the nonlinear high-order NFDE with proportional delay. Some examples are displayed for implicit and explicit forms
of NFDEs to demonstrate the computation accuracy of the proposed method. We also compare the performance of the method with
variational iteration method, one-leg 𝜃-method, continuous Runge-Kutta method, and reproducing kernel Hilbert space method.

1. Introduction method to solve the pantograph type equation. The one-leg 𝜃


approach was implemented to solve nonlinear NFDEs in [14].
In the last four decades, spectral method has become increas- Meanwhile, Trif [15] proposed a direct solution technique
ingly popular and been successfully applied in solving all by using the Chebyshev Tau operational matrix method for
types of differential equations owing to its high order of solving the pantograph type equation. More recently, Bhrawy
accuracy (see, for instance, [1–5]). Recently there has been et al. [16] proposed the Legendre pseudospectral algorithm
a growing interest in applying spectral methods for the with studying the error analysis and stability of the proposed
numerical solution of fractional differential equations and algorithm for a class of DDEs. Sun and Zhang [17] proposed
delay differential equations (DDEs). Moreover, the principal a compact difference method for solving nonlinear partial
difficulty in studying NFDEs with proportional delays lies DDE. Cordero and Escalante [18] extended the application
in their special transcendental nature. Thus we propose of segmented tau approach for solving a class of NFDEs,
an efficient technique to solve such differential equations meanwhile, the history-valued NFDEs were approximated in
numerically based on shifted Jacobi polynomials. [19]. Işik et al. [20] introduced a new Bernstein collocation
In the last two decades, some numerical approaches for method for the numerical solution of DDEs of pantograph-
treating several types of DDEs were presented in [6–12]. In type with retarded case. The existence of solutions of neutral
this direction, Zhao et al. [13] studied the stability of Runge- functional-differential equations with proportional delays
Kutta approach and applied it for solving the NFDEs with had been discussed in [21–24].
proportional delays. Tohidi et al. [12] derived the operational The aim of this paper is to develop a direct solution
matrix of Bernoulli polynomial and used it with collocation technique to approximate the linear high-order NFDEs with
2 Abstract and Applied Analysis

proportional delays using the shifted Jacobi polynomials on 𝐿2𝑤(𝜃,𝜗) with respect to the weight function are defined as
the interval [0, 𝐿]; we dedicate the shifted Jacobi-Gauss- follows:
Lobatto pseudospectral (SJGLP) method to find the approx-
1
imate solution 𝑢𝑁(𝑥). Approximate semianalytical solution (𝑢, V)𝑤(𝜃,𝜗) = ∫ 𝑢 (𝑥) V (𝑥) 𝑤(𝜃,𝜗) (𝑥) 𝑑𝑥,
with high accuracy can be obtained by selecting a limited −1 (4)
number of Gauss-Lobatto collocation points for the linear
and nonlinear high-order NFDEs with proportional delays. ‖𝑢‖𝑤(𝜃,𝜗) = (𝑢, 𝑢)1/2
𝑤(𝜃,𝜗)
.
For suitable collocation points we use the (𝑁 − 𝑚 + 1) nodes
of the shifted Jacobi-Gauss-Lobatto interpolation on (0, 𝐿) in The set of Jacobi polynomials forms a complete 𝐿2𝑤(𝜃,𝜗) -
which the nodes are distinct and lie between 0 and 𝐿, and orthogonal system, and
the two endpoints of the domain (0 and 𝐿) are used as the
󵄩󵄩 (𝜃,𝜗) 󵄩󵄩
first and last collocation points, respectively. The algorithms 󵄩󵄩𝐽𝑘 󵄩󵄩 (𝜃,𝜗) := ℎ𝑘
given in [16] can be obtained as special cases from the 󵄩 󵄩𝑤
proposed algorithms. Finally, the accuracy of the proposed 2𝜃+𝜗+1 Γ (𝑘 + 𝜃 + 1) Γ (𝑘 + 𝜗 + 1)
method is showed by test problems. From the results, these = .
(2𝑘 + 𝜃 + 𝜗 + 1) Γ (𝑘 + 1) Γ (𝑘 + 𝜃 + 𝜗 + 1)
algorithms are extremely efficient and accurate for solving (5)
NFDEs.
The paper is arranged in the following way. In the (𝛼,𝛽) (𝛼,𝛽)
next section, some basic properties of Jacobi polynomials Let us denote 𝑃𝐿,𝑘 (𝑥) = 𝐽𝑘 ((2𝑥/𝐿) − 1), 𝐿 > 0. By the
which are required in the present paper are given, and shifted Jacobi polynomial of degree 𝑘 and by using (1) and (2);
in Section 3, the way of constructing the pseudospectral then we deduce that
technique for NFDEs with proportional delays is described
(𝛼,𝛽) Γ (𝑘 + 𝛽 + 1)
using the shifted Jacobi polynomials. In Section 4, we 𝑃𝐿,𝑘 (0) = (−1)𝑘 , (6)
investigate the shifted Jacobi-Gauss-Lobatto pseudo-spectral Γ (𝛽 + 1) 𝑘!
(SJGLP) method for solving nonlinear high-order NFDEs
with proportional delays. Some numerical results exhibit- (𝛼,𝛽)
(−1)𝑘−𝑞 Γ (𝑘 + 𝛽 + 1) (𝑘 + 𝛼 + 𝛽 + 1)𝑞
ing the accuracy of the proposed algorithm are given in 𝐷𝑞 𝑃𝐿,𝑘 (0) = , (7)
𝐿𝑞 Γ (𝑘 − 𝑞 + 1) Γ (𝑞 + 𝛽 + 1)
Section 5. Finally, a conclusion is given in Section 6.
(𝛼,𝛽) Γ (𝑚 + 𝑘 + 𝛼 + 𝛽 + 1) (𝛼+𝑚,𝛽+𝑚)
𝐷𝑚 𝑃𝐿,𝑘 (𝑥) = 𝑃 (𝑥) . (8)
2. Preliminaries 𝐿𝑚 Γ (𝑘 + 𝛼 + 𝛽 + 1) 𝐿,𝑘−𝑚

In this section, we briefly recall some properties of the Jacobi The following inner product and norm
polynomials (𝐽𝑘(𝜃,𝜗) (𝑥), 𝑘 = 0, 1, . . ., 𝜃 > −1, 𝜗 > −1), which
𝐿
satisfy the following relations: (𝛼,𝛽)
(𝑢, V)𝑤(𝛼,𝛽) = ∫ 𝑢 (𝑥) V (𝑥) 𝑤𝐿 (𝑥) 𝑑𝑥,
𝐿 0
(9)
𝐽𝑘(𝜃,𝜗) (−𝑥) = (−1)𝑘 𝐽𝑘(𝜃,𝜗) (𝑥) ,
‖V‖𝑤(𝛼,𝛽) = (V, V)1/2(𝛼,𝛽) ,
𝐿 𝑤𝐿
(−1)𝑘 Γ (𝑘 + 𝜗 + 1)
𝐽𝑘(𝜃,𝜗) (−1) = , (1)
𝑘!Γ (𝜗 + 1) are defined on the weighted space 𝐿2𝑤(𝛼,𝛽) (0, 𝐿) subject to the
𝐿
Γ (𝑘 + 𝜃 + 1) (𝛼,𝛽) (𝛼,𝛽)
𝐽𝑘(𝜃,𝜗) (1) = . weight function 𝑤𝐿 (𝑥) = (𝐿 − 𝑥)𝛼 𝑥𝛽 . Moreover, 𝑃𝐿,𝑘 (𝑥)
𝑘!Γ (𝜃 + 1) forms a complete 𝐿2𝑤(𝛼,𝛽) (0, 𝐿)-orthogonal system.
𝐿

The 𝑞th derivative of Jacobi polynomials of degree 𝑘 can be According to (5), we get
given by
󵄩󵄩 (𝛼,𝛽) 󵄩󵄩2 𝐿 𝛼+𝛽+1 (𝛼,𝛽) (𝛼,𝛽)
󵄩󵄩𝑃 󵄩
Γ (𝑗 + 𝜃 + 𝜗 + 𝑞 + 1) (𝜃+𝑞,𝜗+𝑞) 󵄩󵄩 𝐿,𝑘 󵄩󵄩󵄩𝑤(𝛼,𝛽) = ( 2 ) ℎ𝑘 := ℎ𝐿,𝑘 . (10)
𝐷(𝑞) 𝐽𝑘(𝜃,𝜗)
𝐿
(𝑥) = 𝑞 𝐽 (𝑥) . (2)
2 Γ (𝑗 + 𝜃 + 𝜗 + 1) 𝑘−𝑞
3. Linear High-Order NFDE with
These polynomials are the only polynomials arising as eigen- Proportional Delay
functions of the following singular Sturm-Liouville equation:
In this section, we shall investigate solutions to NFDEs with
2 󸀠󸀠 󸀠 proportional delays of the form
(1 − 𝑥 ) 𝜙 (𝑥) + [𝜗 − 𝜃 + (𝜃 + 𝜗 + 2) 𝑥] 𝜙 (𝑥)
(3)
(𝑚)
+ 𝑛 (𝑛 + 𝜃 + 𝜗 + 1) 𝜙 (𝑥) = 0. (𝑢(𝑥) + 𝑎 (𝑥) 𝑢 (𝛾𝑚 𝑥))
𝑚−1 (11)
Let 𝑤(𝜃,𝜗) (𝑥) = (1 − 𝑥)𝜃 (1 + 𝑥)𝜗 ; then we define the weighted = 𝛽𝑢 (𝑥) + ∑ 𝑏𝑛 (𝑥) 𝑢(𝑛) (𝛾𝑛 𝑥) + 𝑓 (𝑥) , 𝑥 ≥ 0,
space 𝐿2𝑤(𝜃,𝜗) as usual. The inner product and the norm of 𝑛=0
Abstract and Applied Analysis 3

(𝛼,𝛽) (𝛼,𝛽)
with the initial conditions where 𝑥𝐿,𝑁,𝑗 and 𝜛𝐿,𝑁,𝑗 are the nodes and the corresponding
weights of the shifted Jacobi-Gauss-Lobatto quadrature for-
𝑚−1
mula on the interval (0, 𝐿), respectively. Obviously,
∑ 𝑐𝑖𝑛 𝑢(𝑛) (0) = 𝜆 𝑖 , 𝑖 = 0, 1, . . . , 𝑚 − 1, (12)
𝑛=0

where 𝑎 and 𝑏𝑛 (𝑛 = 0, 1, . . . , 𝑚 − 1) are given functions, (𝑢, V)𝑤(𝛼,𝛽) ,𝑁 = (𝑢, V)𝑤(𝛼,𝛽) , ∀𝑢, V ∈ 𝑆2𝑁−1 . (16)
𝐿 𝐿
meanwhile, 𝛽, 𝛾𝑛 , 𝑐𝑖𝑛 , and 𝜆 𝑖 are constants with 0 < 𝛾𝑛 < 1
(𝑛 = 0, 1, . . . , 𝑚). This class of equations plays an important
role in modeling phenomena of the real world.
In the pseudo-spectral methods [25–28], one needs to The shifted Jacobi-Gauss-Lobatto pseudo-spectral meth-
exactly satisfy the differential equation at specified collo- od for solving (28) and (29) is to seek 𝑢𝑁(𝑥) ∈ 𝑆𝑁(0, 𝐿), such
cation points in the domain of solution. Generally, the that
distribution of the collocation nodes can be freely chosen,
but an accurate approximations are obtained by selecting the
collocation nodes as the zeros of the orthogonal polynomials. (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝑚)
(𝑢 (𝑥𝐿,𝑁−𝑚,𝑘 ) + 𝑎 (𝑥𝐿,𝑁−𝑚,𝑘 ) 𝑢 (𝛾𝑚 𝑥𝐿,𝑁−𝑚,𝑘 ))
For shifted Jacobi polynomials, two commonly used quadra-
ture and collocation nodes, namely, (i) shifted Jacobi-Gauss (𝛼,𝛽)
= 𝛽𝑢 (𝑥𝐿,𝑁−𝑚,𝑘 )
nodes (in the interior of the domain) and (ii) shifted Jacobi-
Gauss-Lobatto nodes (in the interior and at the two endpoints 𝑚−1
(𝛼,𝛽) (𝛼,𝛽)
of the domain). + ∑ 𝑏𝑛 (𝑥𝐿,𝑁−𝑚,𝑘 ) 𝑢(𝑛) (𝛾𝑛 𝑥𝐿,𝑁−𝑚,𝑘 )
Now, we will present the shifted Jacobi-Gauss-Lobatto 𝑛=0
(𝛼,𝛽) (17)
type quadratures. Let 𝑥𝑁,𝑗 , 0 ⩽ 𝑗 ⩽ 𝑁, be the nodes of the (𝛼,𝛽)
(𝛼,𝛽) + 𝑓 (𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
Jacobi-Gauss-Lobatto interpolation on (−1, 1), and let 𝜛𝑁,𝑗 ,
0 ⩽ 𝑗 ⩽ 𝑁, be the corresponding weights. Throughout 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
this paper, we assume that 𝑥𝐿,𝑁,𝑗 , 0 ⩽ 𝑗 ⩽ 𝑁, stands for
𝑚−1
the nodes of the shifted Jacobi-Gauss-Lobatto interpolation
(𝛼,𝛽) ∑ 𝑐𝑖𝑛 𝑢(𝑛) (0) = 𝜆 𝑖 , 𝑖 = 0, 1, . . . , 𝑚 − 1,
on the interval (0, 𝐿). Thus 𝑥𝐿,𝑁,𝑗 , 0 ⩽ 𝑗 ⩽ 𝑁, and their 𝑛=0
(𝛼,𝛽) (𝛼,𝛽)
corresponding weights are 𝜛𝐿,𝑁,𝑗 = (𝐿/2)𝛼+𝛽+1 𝜛𝑁,𝑗 , 0 ⩽ 𝑗 ⩽
𝑁. Let 𝑆𝑁(0, 𝐿) be the set of all polynomials of degree ≤ 𝑁.
One gets for any 𝜙 ∈ 𝑆2𝑁−1 (0, 𝐿), (𝛼,𝛽)
where the 𝑥𝐿,𝑁−𝑚,𝑘 , 𝑘 = 1, 2, . . . , 𝑁 − 𝑚 − 1, are distinct and
(𝛼,𝛽) (𝛼,𝛽)
𝐿 lie between 0 and 𝐿, 𝑥𝐿,𝑁−𝑚,0 = 0, and 𝑥𝐿,𝑁−𝑚,𝑁−𝑚 = 𝐿.
𝛼 𝛽
∫ (𝐿 − 𝑥) 𝑥 𝜙 (𝑥) 𝑑𝑥 For simplicity in presentation and without loss of generality,
0 assume that 𝑎(𝑥) ≡ 1. We now derive the collocation
𝐿 𝛼+𝛽+1 1 𝐿 algorithm for solving (28) and (29). To do this, consider the
=( ) ∫ (1 − 𝑥)𝛼 (1 + 𝑥)𝛽 𝜙 ( (𝑥 + 1)) 𝑑𝑥 solution is approximated by a truncated Jacobi expansion
2 −1 2
(13)
𝐿 𝛼+𝛽+1 𝑁 (𝛼,𝛽) 𝐿 (𝛼,𝛽)
=( ) ∑𝜛𝑁,𝑗 𝜙 ( (𝑥𝑁,𝑗 + 1)) 𝑁
2 𝑗=0 2 (𝛼,𝛽) 𝑇
𝑢𝑁 (𝑥) = ∑ 𝑎ℎ 𝑃𝐿,ℎ (𝑥) , a = (𝑎0 , 𝑎1 , . . . , 𝑎𝑁) . (18)
𝑁 ℎ=0
(𝛼,𝛽) (𝛼,𝛽)
= ∑𝜛𝐿,𝑁,𝑗 𝜙 (𝑥𝐿,𝑁,𝑗 ) .
𝑗=0
We first approximate 𝑢(𝑥) and 𝑢𝑛 (𝑥) as (32). By substitut-
Let us first introduce some basic notations that will be ing these approximation in (28), we get
used in the sequel. We set

(𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)


𝑆𝑁 (0, 𝐿) = span {𝑃𝐿,0 (𝑥) , 𝑃𝐿,1 (𝑥) , . . . , 𝑃𝐿,𝑁 (𝑥)} . (14) 𝑁 𝑁
(𝛼,𝛽) (𝛼,𝛽)
∑ 𝑎ℎ 𝐷(𝑚) 𝑃𝐿,ℎ (𝑥) + ∑ 𝑎ℎ 𝐷(𝑚) 𝑃𝐿,ℎ (𝛾𝑚 𝑥)
ℎ=0 ℎ=0
The discrete inner product and norm are defined by
𝑁
(𝛼,𝛽)
𝑁 = 𝛽 ∑ 𝑎ℎ 𝑃𝐿,ℎ (𝑥) (19)
(𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
(𝑢, V)𝑤(𝛼,𝛽) ,𝑁 = ∑ 𝑢 (𝑥𝐿,𝑁,𝑗 ) V (𝑥𝐿,𝑁,𝑗 ) 𝜛𝐿,𝑁,𝑗 , ℎ=0
𝐿
𝑗=0 (15) 𝑚−1 𝑁
(𝛼,𝛽)
+ ∑ ∑ 𝑏𝑛 (𝑥) 𝑎ℎ 𝐷(𝑛) 𝑃𝐿,ℎ (𝛾𝑛 𝑥) + 𝑓 (𝑥) .
‖𝑢‖𝑤(𝛼,𝛽) ,𝑁 = √(𝑢, 𝑢)𝑤(𝛼,𝛽) ,𝑁,
𝐿 𝐿 𝑛=0 ℎ=0
4 Abstract and Applied Analysis

Then, by virtue of (8), we deduce that Thus (22) with relation (23) can be written as a linear
algebraic system. To do this, let us consider
𝑁
Γ (ℎ + 𝛼 + 𝛽 + 𝑚 + 1) (𝛼+𝑚,𝛽+𝑚) 𝑇
∑ 𝑎ℎ ( 𝑃 (𝑥) a = (𝑎0 , 𝑎1 , . . . , 𝑎𝑁) ,
ℎ=0
𝐿𝑛 Γ (ℎ + 𝛼 + 𝛽 + 1) 𝐿,ℎ−𝑚
𝑓𝑘 = 𝑓 (𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, (24)
Γ (ℎ + 𝛼 + 𝛽 + 𝑚 + 1) (𝛼+𝑚,𝛽+𝑚)
+ 𝛾𝑚𝑚 𝑚 𝑃 (𝛾𝑚 𝑥)) 𝑇
𝐿 Γ (ℎ + 𝛼 + 𝛽 + 1) 𝐿,ℎ−𝑚 f = (𝑓0 , 𝑓1 , . . . , 𝑓𝑁−𝑚 , 𝜆 0 , . . . , 𝜆 𝑚−1 ) .
𝑁
(𝛼,𝛽) The matrix system associated with (22) and (23) becomes
= 𝛽 ∑ 𝑎ℎ 𝑃𝐿,ℎ (𝑥)
ℎ=0
𝑚−1
𝑚−1 𝑁
Γ (ℎ + 𝛼 + 𝛽 + 𝑛 + 1) (𝐴 + 𝛾𝑚𝑚 𝐵 + 𝛽𝐶 + ∑ 𝛾𝑛𝑛 𝐷𝑛 + 𝐸) a = f, (25)
+ ∑ ∑ 𝛾𝑛(𝑛) 𝑏𝑛 (𝑥) 𝑎ℎ 𝑛=0
𝑛=0 ℎ=0 𝐿𝑛 Γ (ℎ + 𝛼 + 𝛽 + 1)
(𝛼+𝑛,𝛽+𝑛) where the matrices 𝐴, 𝐵, 𝐶, 𝐷𝑖 , 𝑖 = 1, 2, . . . , 𝑚 − 1, and
× 𝑃𝐿,ℎ−𝑛 (𝛾𝑛 𝑥) + 𝑓 (𝑥) . 𝐸 are given explicitly in the following. If we denote that
(20) 𝐴 = (𝑎𝑘𝑗 )0<𝑘,𝑗<𝑁, 𝐵 = (𝑏𝑘𝑗 )0<𝑘,𝑗<𝑁, 𝐶 = (𝑐𝑘𝑗 )0<𝑘,𝑗<𝑁, 𝐷𝑛 =
𝑛
(𝑑𝑘𝑗 )0<𝑘,𝑗<𝑁, 𝑛 = 1, 2, . . . , 𝑚 − 1, and 𝐸 = (𝑒𝑘𝑗 )0<𝑘,𝑗<𝑁, then the
𝑛
Also, by substituting (32) in (29), we obtain elements 𝑎𝑘𝑗 , 𝑏𝑘𝑗 , 𝑐𝑘𝑗 , and 𝑑𝑘𝑗 are given by

𝑚−1 𝑁
(𝛼,𝛽)
𝑎𝑘𝑗
∑ ∑ 𝑐𝑖𝑛 𝑎ℎ 𝐷(𝑛) 𝑃𝐿,ℎ (0) = 𝜆 𝑖 . (21)
𝑛=0 ℎ=0
{ Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1)
{
{
{
{ 𝐿𝑛 Γ (𝑗 + 𝛼 + 𝛽 + 1)
{
{
To find the solution 𝑢𝑁(𝑥), we first collocate (20) at the { (𝛼+𝑚,𝛽+𝑚)
{ ×𝑃𝐿,𝑗−𝑚 (𝛼,𝛽)
(𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
(𝑁 − 𝑚 + 1) shifted Jacobi roots that yields ={
{
{ 𝑗 = 0, 1, . . . , 𝑁,
{
{
{
{ 0, 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁,
𝑁 {
{
Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1) (𝛼+𝑚,𝛽+𝑚) (𝛼,𝛽) 𝑗 = 0, 1, . . . , 𝑁,
∑𝑎𝑗 ( 𝑃 (𝑥𝐿,𝑁−𝑚,𝑘 ) {
𝑗=0 𝐿𝑛 Γ (𝑗 + 𝛼 + 𝛽 + 1) 𝐿,𝑗−𝑚
𝑏𝑘𝑗
Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1)
+ 𝛾𝑚𝑚 𝑚 Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1)
𝐿 Γ (𝑗 + 𝛼 + 𝛽 + 1) {
{
{
{ 𝑚 (𝑗 + 𝛼 + 𝛽 + 1)
{ 𝐿 Γ(𝛼+𝑚,𝛽+𝑚)
{
{
(𝛼+𝑚,𝛽+𝑚) (𝛼,𝛽) {
{ ×𝑃𝐿,𝑗−𝑚 (𝛼,𝛽)
(𝛾𝑚 𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
× 𝑃𝐿,𝑗−𝑚 (𝛾𝑚 𝑥𝐿,𝑁−𝑚,𝑘 ) ) ={
{
{ 𝑗 = 0, 1, . . . , 𝑁,
{
{
{
{ 0, 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁,
𝑁
(𝛼,𝛽) (𝛼,𝛽)
{
{
= 𝛽∑ 𝑎𝑗 𝑃𝐿,𝑗 (𝑥𝐿,𝑁−𝑚,𝑘 ) { 𝑗 = 0, 1, . . . , 𝑁,
𝑗=0
𝑐𝑘𝑗
𝑚−1 𝑁
(𝛼,𝛽)
+ ∑ ∑𝛾𝑛(𝑛) 𝑏𝑛 (𝑥𝐿,𝑁−𝑚,𝑘 ) 𝑎𝑗 (𝛼,𝛽)
{ −𝑃
𝑛=0 𝑗=0 { 𝐿,𝑗 (𝛼,𝛽)
= { × (𝑥𝐿,𝑁−𝑚,𝑘 ), 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑗 = 0, 1, . . . , 𝑁,
Γ (𝑗 + 𝛼 + 𝛽 + 𝑛 + 1) (𝛼+𝑛,𝛽+𝑛) {
× 𝑃
(𝛼,𝛽)
(𝛾𝑛 𝑥𝐿,𝑁−𝑚,𝑘 ) {0, 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁, 𝑗 = 0, 1, . . . , 𝑁,
𝐿𝑛 Γ (𝑗 + 𝛼 + 𝛽 + 1) 𝐿,𝑗−𝑛
𝑛
𝑑𝑘𝑗
(𝛼,𝛽)
+ 𝑓 (𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚.
(𝛼,𝛽)
(22) { −𝑏𝑛 (𝑥𝐿,𝑁−𝑚,𝑘 )
{
{
{
{ Γ (𝑗 + 𝛼 + 𝛽 + 𝑛 + 1)
{ ×
{
{
{ 𝐿𝑛 Γ (𝑗 + 𝛼 + 𝛽 + 1)
Next, (21), after using (7), can be written as {
{ (𝛼+𝑛,𝛽+𝑛) (𝛼,𝛽)
= { ×𝑃𝐿,𝑗−𝑛 (𝛾𝑛 𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
{
{
𝑚−1 𝑁
Γ (𝑗 + 𝛽 + 1) (𝑗 + 𝛼 + 𝛽 + 1)𝑛 {
{ 𝑗 = 0, 1, . . . , 𝑁,
{
{
∑ ∑ (−1)𝑗−𝑛 𝑐𝑖𝑛 𝑎𝑗 = 𝜆𝑖, {
{ 0, 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁,
𝐿𝑛 Γ (𝑗 − 𝑛 + 1) Γ (𝑛 + 𝛽 + 1) {
{
𝑛=0 𝑗=0 (23)
{ 𝑗 = 0, 1, . . . , 𝑁.
𝑖 = 0, 1, . . . , 𝑚 − 1. (26)
Abstract and Applied Analysis 5

Moreover, the elements of the matrix corresponding to the (𝛼,𝛽) (𝛼,𝛽)


= 𝐺 (𝑥𝐿,𝑁−𝑚,𝑘 , 𝑢 (𝑥𝐿,𝑁−𝑚,𝑘 ) ,
mixed initial conditions are given by
(𝛼,𝛽) 𝑑 (𝛼,𝛽)
𝑒𝑘𝑗 𝑢 (𝛾0 𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑢 (𝛾1 𝑥𝐿,𝑁−𝑚,𝑘 ) , . . . ,
𝑑𝑥

0, 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑑𝑚 (𝛼,𝛽)
{
{ 𝑢 (𝛾𝑚 𝑥𝐿,𝑁−𝑚,𝑘 )) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚.
{
{ 𝑑𝑥𝑚
{
{ 𝑗 = 0, 1, . . . , 𝑁,
{
{
{
{ 𝑚−1 (30)
{
{ 𝑗−𝑛
{ ∑ (−1)
{
= { 𝑛=0
{
{ ×𝑐𝑘−𝑁−𝑚+1,𝑛 Now, we approximate the numerical solution as a trun-
{
{
{
{ Γ(𝑗+𝛽+1)(𝑗+𝛼+𝛽+1)𝑛 cated series expansion of shifted Jacobi polynomial in the
{
{ × 𝑛 , 𝑘 = 𝑁−𝑚+1, . . . , 𝑁,
{
{ form
{
{ 𝐿 Γ(𝑗−𝑛+1) Γ(𝑛+𝛽+1)
{
{ 𝑗 = 0, 1, . . . , 𝑁.
(27) 𝑁
(𝛼,𝛽)
𝑢𝑁 (𝑥) = ∑𝑎𝑗 𝑃𝐿,𝑗 (𝑥) . (31)
𝑗=0
In the case of 𝑎(𝑥) ≠0, 𝑏𝑛 (𝑥) ≠
0, 𝑛 = 0, 1, . . . , 𝑚 − 1, and
𝛽 ≠ 0, the linear system (25) can be solved by forming explic-
itly the LU factorization; that is, 𝐴+𝛾𝑚𝑚 𝐵+𝛽𝐶+∑𝑚−1 𝑛
𝑛=0 𝛾𝑛 𝐷𝑛 +
Accordingly, (30) can be written as
𝐸 = LU. The expense of calculating LU factorization is 𝑂(𝑁3 )
operations, and the expense of solving the linear system (25), 𝑁
𝑑𝑚 (𝛼,𝛽) (𝛼,𝛽)
provided that the factorization is known, is 𝑂(𝑁2 ). ∑ 𝑎𝑗 𝑃 (𝑥𝐿,𝑁−𝑚,𝑘 )
𝑗=0 𝑑𝑥𝑚 𝐿,𝑗

4. Nonlinear High-Order NFDE with 𝑁


(𝛼,𝛽) (𝛼,𝛽)
Proportional Delay = 𝐺 (𝑥, ∑𝑎𝑗 𝑃𝐿,𝑗 (𝑥𝐿,𝑁−𝑚,𝑘 ) ,
𝑗=0
In this section, we investigate the shifted Jacobi-Gauss-
𝑁
Lobatto pseudospectral method to numerically approximate (𝛼,𝛽) (𝛼,𝛽)
∑ 𝑎𝑗 𝑃𝐿,𝑗 (𝛾0 𝑥𝐿,𝑁−𝑚,𝑘 ) ,
the nonlinear high-order NFDE with proportional delay; 𝑗=0 (32)
namely,
𝑁
𝑑 (𝛼,𝛽) (𝛼,𝛽)
𝑚 ∑ 𝑎𝑗 𝑃𝐿,𝑗 (𝛾1 𝑥𝐿,𝑁−𝑚,𝑘 ) , . . . ,
𝑑 𝑢 (𝑥) 𝑗=0 𝑑𝑥
𝑑𝑥𝑚
𝑁
𝑑𝑢 (𝛾1 𝑥) 𝑑𝑚 𝑢 (𝛾𝑚 𝑥) 𝑑𝑚 (𝛼,𝛽) (𝛼,𝛽)
= 𝐺 (𝑥, 𝑢 (𝑥) , 𝑢 (𝛾0 𝑥) , ,..., ) , (28) ∑𝑎𝑗 𝑃 (𝛾𝑚 𝑥𝐿,𝑁−𝑚,𝑘 )) ,
𝑑𝑥 𝑑𝑥𝑚 𝑗=0 𝑑𝑥𝑚 𝐿,𝑗

0 ≤ 𝑥 ≤ 𝐿, 𝑘 = 0, 1, . . . , 𝑁 − 𝑚.

subject to Next, making use of relation (8), thus the high-order


derivatives of the proposed solution with proportional delays,
𝑚−1 (𝑑𝑢(𝛾1 𝑥)/𝑑𝑥), . . . , (𝑑𝑚 𝑢(𝛾𝑚 𝑥)/𝑑𝑥𝑚 ), can be expressed explic-
∑ 𝑐𝑖𝑛 𝑢(𝑛) (0) = 𝜆 𝑖 , 𝑖 = 0, 1, . . . , 𝑚 − 1, (29) itly in terms of the shifted Jacobi polynomials and the
𝑛=0 expansion coefficients 𝑎𝑗 ; consequently

where 𝛾𝑛 , 𝑐𝑖𝑛 , and 𝜆 𝑖 are constants with 0 < 𝛾𝑛 < 1 (𝑛 = 0,


𝑁
1, . . . , 𝑚), meanwhile, 𝐺 is nonlinear in general. This equation Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1) (𝛼+𝑚,𝛽+𝑚) (𝛼,𝛽)
∑ 𝑎𝑗 𝑃 (𝑥𝐿,𝑁−𝑚,𝑘 )
is a generalized form of the proportional delay differential 𝑗=0 𝐿𝑛 Γ (𝑗 + 𝛼 + 𝛽 + 1) 𝐿,𝑗−𝑚
equations given in [14, 29, 30] and plays an important role
in modeling phenomena of the real world. 𝑁 𝑁
(𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
The shifted Jacobi-Gauss-Lobatto pseudospectral approx- = 𝐺 (𝑥, ∑𝑎𝑗 𝑃𝐿,𝑗 (𝑥𝐿,𝑁−𝑚,𝑘 ) , ∑ 𝑎𝑗 𝑃𝐿,𝑗 (𝛾0 𝑥𝐿,𝑁−𝑚,𝑘 ) ,
imation for (28) is to find 𝑢𝑁(𝑥) ∈ 𝑆𝑁(0, 𝐿) such that 𝑗=0 𝑗=0

𝑁
Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1)
𝑑𝑚 (𝛼,𝛽) ∑ 𝑎𝑗 𝛾1 (
𝑢 (𝑥𝐿,𝑁−𝑚,𝑘 ) 𝐿𝑛 Γ (𝑗 + 𝛼 + 𝛽 + 1)
𝑑𝑥𝑚 𝑗=0
6 Abstract and Applied Analysis

(𝛼+𝑚,𝛽+𝑚) (𝛼,𝛽)
× 𝑃𝐿,𝑗−𝑚 (𝛾1 𝑥𝐿,𝑁−𝑚,𝑘 ) ) , . . . , 0.35

0.30
𝑁
Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1) 0.25
∑𝑎𝑗 𝛾𝑚𝑚 ( 𝑛
𝑗=0 𝐿 Γ (𝑗 + 𝛼 + 𝛽 + 1) 0.20

u(x)
0.15
(𝛼+𝑚,𝛽+𝑚) (𝛼,𝛽)
× 𝑃𝐿,𝑗−𝑚 (𝛾𝑚 𝑥𝐿,𝑁−𝑚,𝑘 ) ) ) , 0.10

0.05
𝑘 = 0, 1, . . . , 𝑁 − 𝑚.
0.00
(33) 0 2 4 6 8 10
x

Exact solution
Finally, to find the unknown expansion coefficients 𝑎𝑗 , 𝑗 = Approximate method at N = 20
0, 1, . . . , 𝑁, we implement any iteration technique to solve a
Figure 1: Comparison of the approximate solution with the exact
system of 𝑁 + 1 nonlinear algebraic equations resulting from solution for 𝛼 = −1/2, 𝛽 = 1/2 at 𝑁 = 28 for Example 1.
the combination of 𝑁 − 𝑚 + 1 nonlinear algebraic equations
resulting from (33) and 𝑚 linear algebraic equations resulting
from (29). ×10−12
4.0

3.0
5. Numerical Results
2.0
In this section, we will carry out three test examples to study
1.0
Error

the validity and effectiveness of the proposed method and


also show that high accurate solutions are achieved using a 0
few number of the Jacobi Gauss-Lobatto points. Moreover, −1.0
comparisons with other methods reveal that the present
method is accurate and convenient. All the numerical com- −2.0
putations have been performed by the symbolic computation −3.0
software Mathematica 8.0. 0 2 4 6 8 10

Example 1. Consider the first-order NFDE with proportional x


delay considered in [31] Figure 2: The error between the approximate solution and the exact
solution in the interval [0, 10] for 𝛼 = −1/2, 𝛽 = 1/2 at 𝑁 = 28 for
Example 1.
𝑢󸀠 (𝑥) = − 𝑢 (𝑥) + 0.1𝑢 (0.8𝑥) + 0.5𝑢󸀠 (0.8𝑥)

+ (0.32𝑥 − 0.5) 𝑒−0.8𝑥 + 𝑒−𝑥 , 𝑥 ≥ 0, (34) Example 2. Let us consider the second-order NFDE with
proportional delay
𝑢 (0) = 0,
𝑥 1 𝑥 𝜋
𝑢󸀠󸀠 (𝑥) = 𝑢󸀠 ( ) − 𝑥𝑢󸀠󸀠 ( ) −
2 2 2 2
which has the exact solution 𝑥𝑒−𝑥 . 𝜋𝑥 𝜋𝑥
× (2 cos ( ) + 𝜋𝑥 sin ( )
2 2 (35)
Table 1 lists the absolute error using Jacobi pseudospectral
method for three choices of 𝛼, 𝛽 at 𝑁 = 20 in the interval + 2𝜋 sin (𝜋𝑥) ) , 𝑥 ∈ [0, 6] ,
[0, 1]. We compare the errors obtained by the proposed
method with variational iteration (VI) method [32], the one- 𝑢 (0) = 0, 𝑢󸀠 (0) = 𝜋,
leg 𝜃 method [14, 33] with 𝜃 = 0.8, and RKHSM method [31].
The graph of analytical solution and approximate solution in which enjoys exact solution 𝑢(𝑥) = sin(𝜋𝑥).
long interval [0, 10] for 𝑁 = 28 and 𝛼 = −1/2, 𝛽 = 1/2
is displayed in Figure 1 to make it easier to compare with In Table 2, we introduce the absolute error using the
analytical solution. Moreover, in this case the graph of the proposed method at 𝑁 = 24 with various choices of 𝛼 and
error is given in Figure 2. Consequently, we conclude that the 𝛽. The resulting graph of (35) for the presented method in
approximate solution by Jacobi pseudospectral method agree the case of 𝛼 = 𝛽 = 1 at 𝑁 = 24 and the analytic solution are
very well with the exact solution. shown in Figure 3.
Abstract and Applied Analysis 7

Table 1: Absolute errors using SJGLP method with various choices of 𝑁, 𝛼, and 𝛽 for Example 1.

SJGLP method In [32]


𝑥 𝛼 𝛽 In [31] In [33]
𝑁 = 20 𝑛=5 𝑛=6
1/2 1/2 1.58 ⋅ 10−17 1.42 ⋅ 10−4 2.63 ⋅ 10−3 1.30 ⋅ 10−3 4.65 ⋅ 10−3
0.1
0 0 2.23 ⋅ 10−17
1/2 1/2 2.90 ⋅ 10−18 1.17 ⋅ 10−4 4.36 ⋅ 10−3 2.14 ⋅ 10−3 1.45 ⋅ 10−2
0.2
0 0 4.16 ⋅ 10−17
1/2 1/2 9.21 ⋅ 10−18 9.45 ⋅ 10−4 5.40 ⋅ 10−3 2.63 ⋅ 10−3 2.57 ⋅ 10−2
0.3
0 0 4.65 ⋅ 10−17
1/2 1/2 7.09 ⋅ 10−17 7.59 ⋅ 10−4 5.89 ⋅ 10−3 2.84 ⋅ 10−3 3.60 ⋅ 10−2
0.4
0 0 3.52 ⋅ 10−17
1/2 1/2 1.94 ⋅ 10−17 6.03 ⋅ 10−4 5.96 ⋅ 10−3 2.83 ⋅ 10−3 4.43 ⋅ 10−2
0.5
0 0 4.85 ⋅ 10−17
1/2 1/2 1.68 ⋅ 10−17 4.73 ⋅ 10−4 5.71 ⋅ 10−3 2.67 ⋅ 10−3 5.03 ⋅ 10−2
0.6
0 0 5.83 ⋅ 10−17
1/2 1/2 1.22 ⋅ 10−18 3.64 ⋅ 10−4 5.23 ⋅ 10−3 2.39 ⋅ 10−3 5.37 ⋅ 10−2
0.7
0 0 1.62 ⋅ 10−17
1/2 1/2 2.42 ⋅ 10−17 2.75 ⋅ 10−4 4.59 ⋅ 10−3 2.04 ⋅ 10−3 5.47 ⋅ 10−2
0.8
0 0 6.88 ⋅ 10−17
1/2 1/2 3.93 ⋅ 10−17 2.03 ⋅ 10−4 3.84 ⋅ 10−3 1.64 ⋅ 10−3 5.35 ⋅ 10−2
0.9
0 0 5.46 ⋅ 10−18
1/2 1/2 1.33 ⋅ 10−17 1.43 ⋅ 10−4 3.04 ⋅ 10−3 1.22 ⋅ 10−3 5.03 ⋅ 10−2
1.0
0 0 1.25 ⋅ 10−16

Table 2: Absolute errors using SJGLP method with various choices where
of 𝛼 and 𝛽 for Example 2.
1 𝑥/12
𝑥 𝛼 = 𝛽 = −1/2 𝛼 = 𝛽 = 1/2 𝛼=𝛽=0 𝑓 (𝑥) = 𝑒 ( − 13752 sin (2𝑥) − 54𝑒7𝑥/12 cos2 (8𝑥)
54
0.0 4.13 ⋅ 10−16 3.92 ⋅ 10−16 1.11 ⋅ 10−16
0.1 6.76 ⋅ 10−10 1.29 ⋅ 10−9 3.57 ⋅ 10−10 + 1727 cos (2𝑥) + 6𝑒𝑥/36
0.2 5.01 ⋅ 10−9 3.70 ⋅ 10−9 5.86 ⋅ 10−10
8𝑥 8𝑥
0.3 3.24 ⋅ 10−9 9.52 ⋅ 10−9 1.71 ⋅ 10−9 × (48 sin ( ) + 575 cos ( ))
3 3
0.4 4.14 ⋅ 10−8 5.17 ⋅ 10−7 2.16 ⋅ 10−9
0.5 2.61 ⋅ 10−8 2.88 ⋅ 10−8 2.27 ⋅ 10−9 − 18𝑒𝑥/12 (cos (4𝑥) − 24 sin (4𝑥))
1.0 3.72 ⋅ 10−8 1.01 ⋅ 10−8 2.13 ⋅ 10−8
2.0 7.32 ⋅ 10−8 3.89 ⋅ 10−9 3.59 ⋅ 10−8 + 2𝑒𝑥/4 (13752 sin (8𝑥) −1727 cos (8𝑥)) ).
3.0 2.19 ⋅ 10−9 2.84 ⋅ 10−8 1.85 ⋅ 10−7
(37)
4.0 3.52 ⋅ 10−7 7.78 ⋅ 10−7 3.01 ⋅ 10−7
5.0 3.73 ⋅ 10−6 1.24 ⋅ 10−6 1.80 ⋅ 10−6
6.0 3.42 ⋅ 10−6 2.97 ⋅ 10−6 8.74 ⋅ 10−6
The exact solution of the problem is 𝑢(𝑥) = 𝑒𝑥/3 cos(8𝑥).

Table 3 lists the maximum absolute errors using SJGLP


method in the interval [0, 1] at 𝛼 = −𝛽 = 0.5 and different
Example 3. Consider the third-order NFDE with propor- choices of 𝛼 and 𝛽.
tional delays In case of Chebyshev polynomials of the second kind
(𝛼 = 𝛽 = 1/2), the graph of exact solution and approximate
𝑥 solution for 𝑁 = 24 is plotted in Figure 4 in the interval [0, 3].
𝑢󸀠󸀠󸀠 (𝑥) = (𝑢 (𝑥))2 + 𝑢󸀠 ( )
2
𝑥 1 𝑥 6. Conclusion
+ 𝑢󸀠󸀠 ( ) + 𝑢󸀠󸀠󸀠 ( ) + 𝑓 (𝑥) , 𝑥 ≥ 0, (36)
3 2 4 In this paper, we have demonstrated the feasibility of SJGLP
1 575 for solving linear NFDEs with proportional delays. We also
𝑢 (0) = 1, 𝑢󸀠 (0) = , 𝑢󸀠󸀠 (0) = − , have discussed the resulting linear system. Moreover, we have
3 9
8 Abstract and Applied Analysis

1.0 In the future work, we address the Jacobi pseudospectral


approximation for the solution of linear and nonlinear delay
partial differential equations in two and three dimensions
0.5
(see, e.g., [17, 34]).

0.0 References
u(x)

[1] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang,


−0.5 Spectral Methods in Fluid Dynamics, Springer Series in Com-
putational Physics, Springer, New York, NY, USA, 1988.
[2] L. N. Trefethen, Spectral Methods in MATLAB, vol. 10 of
−1.0 Software, Environments, and Tools, SIAM, Philadelphia, Pa,
0 1 2 3 4 5 6
USA, 2000.
x
[3] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A new
Exact solution Jacobi operational matrix: an application for solving fractional
Approximate method at N = 20 differential equations,” Applied Mathematical Modelling, vol. 36,
no. 10, pp. 4931–4943, 2012.
Figure 3: Graph of exact solution and approximate solution for 𝛼 = [4] A. H. Bhrawy and M. Al-shomrani, “A shifted Legendre spectral
𝛽 = 1 and 𝑁 = 24 for Example 2. method for fractional order multi-point boundary value prob-
lems,” Advances in Difference Equations, vol. 2012, article 8, 8
pages, 2012.
2 [5] A. H. Bhrawy, M. A. Alghamdi, and T. M. Taha, “A new
modified generalized Laguerre operational matrix of fractional
integration for solving fractional differential equations on the
1
half line,” Advances in Difference Equations, vol. 2012, article 179,
12 pages, 2012.
u(x)

0 [6] A. Bellen and M. Zennaro, Numerical Methods for Delay


Differential Equations, Numerical Mathematics and Scientific
−1 Computation, Oxford University Press, New York, NY, USA,
2003.
−2 [7] M. Z. Liu and D. Li, “Properties of analytic solution and
numerical solution of multi-pantograph equation,” Applied
Mathematics and Computation, vol. 155, no. 3, pp. 853–871,
0.0 0.5 1.0 1.5 2.0 2.5 3.0
2004.
x
[8] G. A. Bocharov and F. A. Rihan, “Numerical modelling in
Exact solution biosciences using delay differential equations,” Journal of Com-
SJGLP solution at N = 24 putational and Applied Mathematics, vol. 125, no. 1-2, pp. 183–
199, 2000.
Figure 4: Graph of exact solution and approximate solution for 𝛼 = [9] M. A. Jafari and A. Aminataei, “Method of successive approx-
𝛽 = 1/2 and 𝑁 = 24 for Example 3. imations for solving the multi-pantograph delay equations,”
General Mathematics Notes, vol. 8, pp. 23–28, 2012.
[10] S. Yüzbaşı, “An efficient algorithm for solving multi-pantograph
Table 3: Maximum absolute errors using SJGLP method with equation systems,” Computers & Mathematics with Applications,
various choices of 𝛼, 𝛽, and 𝑁 for Example 3. vol. 64, no. 4, pp. 589–603, 2012.
[11] S. Yüzbaşi, N. Şahin, and M. Sezer, “A Bessel collocation method
𝑡 𝛼 = −𝛽 = 0.5 𝛼 = −𝛽 = −0.5 𝛼 = 𝛽 = −0.5
for numerical solution of generalized pantograph equations,”
12 1.74 ⋅ 10−2 3.86 ⋅ 10−3 5.07 ⋅ 10−3 Numerical Methods for Partial Differential Equations, vol. 28, no.
16 1.40 ⋅ 10−6 2.65 ⋅ 10−6 3.54 ⋅ 10−6 4, pp. 1105–1123, 2012.
20 4.38 ⋅ 10−9 6.96 ⋅ 10−10 9.40 ⋅ 10−10 [12] E. Tohidi, A. H. Bhrawy, and K. Erfani, “A collocation method
24 8.25 ⋅ 10−13 1.63 ⋅ 10−12 8.76 ⋅ 10−13 based on Bernoulli operational matrix for numerical solution
28 4.35 ⋅ 10−13 3.44 ⋅ 10−13 3.22 ⋅ 10−13 of generalized pantograph equation,” Applied Mathematical
Modelling, vol. 37, no. 6, pp. 4283–4294, 2013.
[13] J. J. Zhao, Y. Xu, H. X. Wang, and M. Z. Liu, “Stability of a class
of Runge-Kutta methods for a family of pantograph equations
implemented the SJGLP method to numerically approximate of neutral type,” Applied Mathematics and Computation, vol. 181,
the nonlinear high-order NFDE with proportional delay. no. 2, pp. 1170–1181, 2006.
All the given examples reveal that the results of SJGLP [14] W.-S. Wang and S.-F. Li, “On the one-leg 𝜃-methods for solving
method are in excellent agreement with the analytical solu- nonlinear neutral functional differential equations,” Applied
tions. It is concluded from the aforementioned tables and Mathematics and Computation, vol. 193, no. 1, pp. 285–301, 2007.
figures that SJGLP method is an accurate and efficient method [15] D. Trif, “Direct operatorial tau method for pantograph-type
to solve NFDEs when compared with those generated by equations,” Applied Mathematics and Computation, vol. 219, no.
some other methods. 4, pp. 2194–2203, 2012.
Abstract and Applied Analysis 9

[16] A. H. Bhrawy, L. M. Assas, E. Tohidi, and M. A. Alghamdi, [33] W. Wang, T. Qin, and S. Li, “Stability of one-leg 𝜃-methods
“Legendre-Gauss collocation method for neutral functional- for nonlinear neutral differential equations with proportional
differential equations with proportional delays,” Advances in delay,” Applied Mathematics and Computation, vol. 213, no. 1, pp.
Differential Equations, vol. 2013, article 63, 16 pages, 2013. 177–183, 2009.
[17] Z.-Z. Sun and Z.-B. Zhang, “A linearized compact difference [34] Z. Jackiewicz and B. Zubik-Kowal, “Spectral collocation and
scheme for a class of nonlinear delay partial differential equa- waveform relaxation methods for nonlinear delay partial differ-
tions,” Applied Mathematical Modelling, vol. 37, no. 3, pp. 742– ential equations,” Applied Numerical Mathematics, vol. 56, no.
752, 2013. 3-4, pp. 433–443, 2006.
[18] L. F. Cordero and R. Escalante, “Segmented tau approximation
for test neutral functional differential equations,” Applied Math-
ematics and Computation, vol. 187, no. 2, pp. 725–740, 2007.
[19] M. Muslim, “Approximation of solutions to history-valued
neutral functional differential equations,” Computers & Math-
ematics with Applications, vol. 51, no. 3-4, pp. 537–550, 2006.
[20] O. R. Işik, Z. Güney, and M. Sezer, “Bernstein series solutions of
pantograph equations using polynomial interpolation,” Journal
of Difference Equations and Applications, vol. 18, no. 3, pp. 357–
374, 2012.
[21] E. K. Ifantis, “An existence theory for functional-differential
equations and functional-differential systems,” Journal of Dif-
ferential Equations, vol. 29, no. 1, pp. 86–104, 1978.
[22] A. Iserles and Y. Liu, “On neutral functional-differential equa-
tions with proportional delays,” Journal of Mathematical Analy-
sis and Applications, vol. 207, no. 1, pp. 73–95, 1997.
[23] C. He, X. Lv, and J. Niu, “A new method based on the RKHSM
for solving systems of nonlinear IDDEs with proportional
delays,” Abstract and Applied Analysis, vol. 2013, Article ID
541935, 13 pages, 2013.
[24] E. Ishiwata, “On the attainable order of collocation methods for
the neutral functional-differential equations with proportional
delays,” Computing, vol. 64, no. 3, pp. 207–222, 2000.
[25] A. H. Bhrawy and A. S. Alofi, “A Jacobi-Gauss collocation
method for solving nonlinear Lane-Emden type equations,”
Communications in Nonlinear Science and Numerical Simula-
tion, vol. 17, no. 1, pp. 62–70, 2012.
[26] A. H. Bhrawy and M. A. Alghamdi, “A shifted Jacobi-Gauss-
Lobatto collocation method for solving nonlinear fractional
Langevin equation involving two fractional orders in different
intervals,” Boundary Value Problems, vol. 2012, article 62, 13
pages, 2012.
[27] M. Maleki, I. Hashim, M. Tavassoli Kajani, and S. Abbas-
bandy, “An adaptive pseudospectral method for fractional order
boundary value problems,” Abstract and Applied Analysis, vol.
2012, Article ID 381708, 19 pages, 2012.
[28] A. H. Bhrawy, “A Jacobi-Gauss-Lobatto collocation method
for solving generalized Fitzhugh-Nagumo equation with time-
dependent coefficients,” Applied Mathematics and Computation,
vol. 222, pp. 255–264, 2013.
[29] W. Wang, Y. Zhang, and S. Li, “Stability of continuous Runge-
Kutta-type methods for nonlinear neutral delay-differential
equations,” Applied Mathematical Modelling, vol. 33, no. 8, pp.
3319–3329, 2009.
[30] J. X. Kuang and Y. H. Cong, Stability of Numerical Methods
for Delay Differential Equations, Science Press, Beijing, China,
2005.
[31] X. Lv and Y. Gao, “The RKHSM for solving neutral functional-
differential equations with proportional delays,” Mathematical
Methods in the Applied Sciences, vol. 36, no. 6, pp. 642–649, 2013.
[32] X. Chen and L. Wang, “The variational iteration method for
solving a neutral functional-differential equation with propor-
tional delays,” Computers & Mathematics with Applications, vol.
59, no. 8, pp. 2696–2702, 2010.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 828764, 8 pages
http://dx.doi.org/10.1155/2013/828764

Research Article
Numerical Solution of a Kind of Fractional Parabolic Equations
via Two Difference Schemes

Abdon Atangana1 and Dumitru Baleanu2,3,4


1
Institute for Groundwater Studies, Faculty of Natural and Agricultural Sciences, University of the Free State,
Bloemfontein 9300, South Africa
2
Department of Chemical and Materials Engineering, Faculty of Engineering, King Abdulaziz University, P.O. Box 80204,
Jeddah 21589, Saudi Arabia
3
Department of Mathematics and Computer Sciences, Cankaya University, Faculty of Art and Sciences, Balgat, 06530 Ankara, Turkey
4
Institute of Space Sciences, P.O. Box MG-23, Magurele, 76900 Bucharest, Romania

Correspondence should be addressed to Abdon Atangana; abdonatangana@yahoo.fr

Received 5 August 2013; Accepted 23 August 2013

Academic Editor: Soheil Salahshour

Copyright © 2013 A. Atangana and D. Baleanu. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

A kind of parabolic equation was extended to the concept of fractional calculus. The resulting equation is, however, difficult to
handle analytically. Therefore, we presented the numerical solution via the explicit and the implicit schemes. We presented together
the stability and convergence of this time-fractional parabolic equation with two difference schemes. The explicit and the implicit
schemes in this case are stable under some conditions.

1. Introduction the study of the stability and convergence of the numerical


method [5, 6].
A parabolic partial differential equation is a type of second- The purpose of this work is to study the stability and
order partial differential equations (PDEs), describing a wide the convergence of the numerical scheme of the parabolic
family of problems in science including heat diffusion and equation of the following form:
ocean acoustic propagation, in physical or mathematical
systems with a time variable, which behave essentially like 𝛼
𝜕𝑡 V (𝑥, 𝑡) + 𝐶0𝐷𝑡 V (𝑥, 𝑡) − 𝑎 (𝑥, 𝑡) 𝜕𝑥𝑥
2
V (𝑥, 𝑡) + 𝛿V (𝑥, 𝑡)
heat diffusing through a solid [1–4]. This mathematical model
is a simplified description of physical reality expressed in = 𝑓 (𝑥, 𝑡) , V (0, 𝑥) = 0,
mathematical terms. Thus, the investigation of the exact or (1)
approximate solution helps us to understand the means of 0 ≤ 𝑥 ≤ 1, 0 < 𝛼 ≤ 1,
these mathematical models. In most cases, it is difficult, or
infeasible, to find the analytical solution or good numerical V (𝑥, 0) = V (0, 𝑡) , V𝑥 (𝑡, 0) = V𝑡 (𝑥, 0) ,
solution of the problems. Numerical solutions or approximate
where 𝑓(𝑥, 𝑡) is sufficiently given smooth function and
analytical solutions become necessary. Numerical methods
𝑎(𝑥, 𝑡) ≥ 0. Here, 𝛿 is a sufficiently large positive constant.
typically yield approximate solutions to the governing equa-
tion through the discretization of space and time and can
relax the rigid idealized conditions of analytical models or 2. Useful Tools for the Fractional Calculus
lumped-parameter models. They can, therefore, be more real-
istic and flexible for simulating field conditions. Within the Definition 1 (see [7–17]). A real function 𝑓(𝑥), 𝑥 > 0, is said
discredited problem domain, the variable internal properties, to be in the space 𝐶𝜇 , 𝜇 ∈ R, if there exists a real number
boundaries, and stresses of the system are approximated. One 𝑝 > 𝜇, such that 𝑓(𝑥) = 𝑥𝑝 ℎ(𝑥), where ℎ(𝑥) ∈ 𝐶[0, ∞), and
of the most important aspects of this numerical method is it is said to be in space 𝐶𝜇𝑚 if 𝑓(𝑚) ∈ 𝐶𝜇 , 𝑚 ∈ N.
2 Abstract and Applied Analysis

Definition 2 (see [7–16]). The Riemann-Liouville fractional derivative of order 𝛼 for 𝑓 in respect to the 𝑥𝑖 the function as
integral operator of order 𝛼 ≥ 0, of a function 𝑓 ∈ 𝐶𝜇 , 𝜇 ≥ −1, follows:
is defined as
𝑥𝑖
1 󵄨
𝜕𝑥𝑖 𝑓 (𝑥𝑗 )󵄨󵄨󵄨󵄨𝑥 =𝑡 𝑑𝑡,
𝑚−𝛼−1 𝑚
𝑥 𝑎𝜕𝛼x 𝑓 = ∫ (𝑥𝑖 − 𝑡) (8)
𝛼 1 Γ (𝑚 − 𝛼) 𝑎
∫ (𝑥 − 𝑡)𝛼−1 𝑓 (𝑡) 𝑑𝑡,
𝑗
𝐽 𝑓 (𝑥) = 𝛼 > 0, 𝑥 > 0,
Γ (𝛼) 0 (2)
where 𝜕𝑥𝑚𝑖 is the usual partial derivative of integer order 𝑚.
0
𝐽 𝑓 (𝑥) = 𝑓 (𝑥) .
3. Examination of the Numerical Solution via
Properties of the operator can be found in [7–15]; we only
mention the following: Difference Schemes
This section is devoted to the discussion underpinning the
for 𝑓 ∈ 𝐶𝜇 , 𝜇 ≥ −1, 𝛼, 𝛽 ≥ 0, 𝛾 > −1, numerical simulation of the solution above (1) via the explicit
scheme [19–23] and the implicit scheme [23–30]. However,
𝐽𝛼 𝐽𝛽 𝑓 (𝑥) = 𝐽𝛼+𝛽 𝑓 (𝑥) , 𝐽𝛼 𝐽𝛽 𝑓 (𝑥) = 𝐽𝛽 𝐽𝛼 𝑓 (𝑥) , before we present the numerical schemes, we must assume
(3) that (1) has a unique and sufficiently smooth solution [23]. In
𝛾 Γ (𝛾 + 1) 𝛼+𝛾 addition, to present the numerical schemes, we let 𝑥𝑙 = 𝑙ℎ,
𝐽𝛼 𝑥 = 𝑥 . 0 ≤ 𝑙 ≤ 𝑀, 𝑀ℎ = 𝐿, 𝑡𝑘 = 𝑘𝜏, 0 ≤ 𝑘 ≤ 𝑁, and 𝑁𝜏 = 𝑇; ℎ
Γ (𝛼 + 𝛾 + 1)
is the space step size, and 𝑀 and 𝑁 are grid points. We will
Definition 3. The Caputo fractional-order derivative is given start with the implicit scheme.
as follows [7–10]:
3.1. Implicit Scheme for the Main Problem. It is important to
𝑥 𝑑 𝑓 (𝑡)𝑛 recall that the finite difference approximation for the second-
𝐶 𝛼 1
0 𝐷𝑥 (𝑓 (𝑥)) = ∫ (𝑥 − 𝑡)𝑛−𝛼−1 𝑑𝑡, order spatial derivative is known as follows [23]:
Γ (𝑛 − 𝛼) 0 𝑑𝑡𝑛 (4)
𝑛 − 1 ≤ 𝛼 ≤ 𝑛. 𝜕2 V (𝑥𝑙 , 𝑡𝑘+1 ) V (𝑥𝑙+1 , 𝑡𝑘+1 ) − 2V (𝑥𝑙 , 𝑡𝑘+1 ) + V (𝑥𝑙−1 , 𝑡𝑘+1 )
=
𝜕𝑥2 ℎ2
Definition 4. The Riemann-Liouville fractional-order deriva- + 𝑂 (ℎ2 ) .
tive is given as follows [8–16]: (9)

1 𝑑𝑛 𝑥
𝐷𝑥𝛼 (𝑓 (𝑥)) = ∫ (𝑥 − 𝑡)𝑛−𝛼−1 𝑓 (𝑡) 𝑑𝑡, The discretization of the Caputo-type time-fractional-order
Γ (𝑛 − 𝛼) 𝑑𝑥𝑛 0 (5) derivative can be presented as follows:
𝑛 − 1 ≤ 𝛼 ≤ 𝑛. 𝜕𝛼 V (𝑥𝑙 , 𝑡𝑘+1 )
𝜕𝑡𝛼
Definition 5. The Jumarie fractional-order derivative is given
as follows [16]: 𝜏−𝛼
= (V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
Γ (2 − 𝛼)
1 𝑑𝑛 𝑥
𝐷𝑥𝛼 (𝑓 (𝑥)) = ∫ (𝑥 − 𝑡)𝑛−𝛼−1 {𝑓 (𝑡) − 𝑓 (0)} 𝑑𝑡, 𝑘
Γ (𝑛 − 𝛼) 𝑑𝑥𝑛 0 + ∑ [V (𝑥𝑙 , 𝑡𝑘+1−𝑗 ) − V (𝑥𝑙 , 𝑡𝑘−𝑗 )] (10)
𝑗=1
𝑛 − 1 ≤ 𝛼 ≤ 𝑛.
(6) 1−𝛼 1−𝛼
× [(𝑗 + 1) − (𝑗) ]) ,

Lemma 6. If 𝑚 − 1 < 𝛼 ≤ 𝑚, 𝑚 ∈ N, and 𝑓 ∈ 𝐶𝜇𝑚 , 𝜇 ≥ −1, 𝜕V (𝑥𝑙 , 𝑡𝑘+1 ) V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
= + 𝑂 (Δ𝑡) .
then 𝜕𝑡 Δ𝑡

𝐷𝛼 𝐽𝛼 𝑓 (𝑥) = 𝑓 (𝑥) , Now, substituting (9) and (10) into (1), we obtained the follo-
wing expression:
𝑚−1
𝑥𝑘
𝐽𝛼 𝐷0𝛼 𝑓 (𝑥) = 𝑓 (𝑥) − ∑ 𝑓(𝑘) (0+ ) , (7) V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 ) 𝜏−𝛼
𝑘=0
𝑘! +
Δ𝑡 Γ (2 − 𝛼)
𝑥 > 0.
× (V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
Definition 7 (partial derivatives of fractional order [7, 8, 11, 𝑘
18]). Assume now that 𝑓(x) is a function of 𝑛 variables 𝑥𝑖 , + ∑ [V (𝑥𝑙 , 𝑡𝑘+1−𝑗 ) − V (𝑥𝑙 , 𝑡𝑘−𝑗 )]
𝑖 = 1, . . . , 𝑛, also of class 𝐶 on 𝐷 ∈ R𝑛 . We define partial 𝑗=1
Abstract and Applied Analysis 3

1−𝛼 1−𝛼
× [(𝑗 + 1) − (𝑗) ]) It is important to inform that if 𝑘 = 0, then the term of the
sum of the right-hand side automatically vanished. Then, (13)
V (𝑥𝑙+1 , 𝑡𝑘+1 ) − 2V (𝑥𝑙 , 𝑡𝑘+1 ) + V (𝑥𝑙−1 , 𝑡𝑘+1 ) can be divided as follows:
− 𝑎𝑙𝑘 ( )
ℎ2
𝜏−𝛼 𝑎1
+ 𝛿V (𝑥𝑙 , 𝑡𝑘 ) = 𝑓 (𝑥𝑙 , 𝑡𝑘 ) . (Δ𝑡 + + 2 𝑙2 ) V𝑙1 − 𝑎𝑙1 V𝑙+1
1
− 𝑎𝑙1 V𝑙−1
1
Γ (2 − 𝛼) ℎ
(11) (14)
𝜏−𝛼
For ease, let = (Δ𝑡 + + 𝛿) V𝑙0 + 𝑓𝑙0 , 𝑘 = 0,
Γ (2 − 𝛼)
V𝑙𝑘 = V (𝑥𝑙 , 𝑡𝑘 ) , 𝑓𝑙𝑘 = 𝑓 (𝑥𝑙 , 𝑡𝑘 ) ,
(12) 𝜏−𝛼 𝑎𝑘
1−𝛼 1−𝛼 (Δ𝑡 + + 2 𝑙2 ) V𝑙𝑘+1 − 𝑎𝑙𝑘 V𝑙+1
𝑘+1
− 𝑎𝑙𝑘 V𝑙−1
𝑘+1
𝑎 (𝑥𝑙 , 𝑡𝑘 ) = 𝑎𝑙𝑘 , (𝑗 + 1) − (𝑗) = 𝑏𝑗 . Γ (2 − 𝛼) ℎ

Then, (11) can be rephrased as follows: 𝜏−𝛼


= (Δ𝑡 + + 𝛿) V𝑙𝑘 (15)
𝑎𝑘 Γ (2 − 𝛼)
𝜏−𝛼
(Δ𝑡 + + 2 𝑙2 ) V𝑙𝑘+1
Γ (2 − 𝛼) ℎ 𝑘
𝜏−𝛼 𝑘−𝑗
−𝛼
− ∑ [V𝑙 ] 𝑑𝑗 + 𝑏𝑘+1 V𝑗0 + 𝑓𝑙𝑘 ,
𝜏 Γ (2 − 𝛼) 𝑗=1
= (Δ𝑡 + + 𝛿) V𝑙𝑘 + 𝑎𝑙𝑘 V𝑙+1
𝑘+1
(13)
Γ (2 − 𝛼)
𝑘 where 𝑑𝑗 = 𝑏𝑗+1 − 𝑏𝑗 .
𝜏−𝛼 𝑘−𝑗+1 𝑘−𝑗
+ 𝑎𝑙𝑘 V𝑙−1
𝑘+1
− ∑ [V𝑙 − V𝑙 ] 𝑏𝑗 + 𝑓𝑙𝑘 . The above equation can be written in matrix form as
Γ (2 − 𝛼) 𝑗=1 follows:

𝜏−𝛼 𝑎1 V1𝑘+1
Δ𝑡 + + 2 𝑙2 −𝑎1𝑘 0 0 0 ⋅⋅⋅ 0
Γ (2 − 𝛼) ℎ
𝑎1 V2𝑘+1 )
( 𝜏−𝛼 d 0 0 )(
( −𝑎2𝑘 Δ𝑡 + + 2 22 )( )
( Γ (2 − 𝛼) ℎ d 0 0 )( )
( 0 )( .. )
( d 𝜏−𝛼 𝑎 1 )( . )
( 𝑘
+ 2 𝑚−2 𝑘 )( )
( .. −𝑎𝑚−2 Δ𝑡 +
Γ (2 − 𝛼) ℎ2
−𝑎𝑚−2 )(
(
)
)
. 𝑘+1
V𝑚−1
𝜏−𝛼 𝑎1
0 0 𝑘
0 0 −𝑎𝑚−1 Δ𝑡 + + 2 𝑙2 𝑘+1
( Γ (2 − 𝛼) ℎ ) ( V𝑚−1 )
(16)

𝑎1𝑘
𝑓1𝑘 − V (0, 𝑡𝑘 )
ℎ2
( 𝑓2𝑘 )
( .. )
( )
=( . ).
( 𝑘
𝑓𝑚−2 )
( )
𝑘
𝑘 𝑎𝑚−1
𝑓𝑚−1 − V (0, 𝑡𝑘 )
( ℎ2 )

3.1.1. Stability of the Implicit Difference Scheme. In this To analyze the stability, we exploit the Fourier method
subsection, we present the stability analysis of the implicit [22], and the expression of 𝜁𝑘 can be defined as follows:
difference scheme for solving the time-fractional parabolic
equation (1). To accomplish this, we let 𝜁𝑙𝑘 = V𝑙𝑘 − 𝑉𝑙𝑘 , with ℎ ℎ
{
{ 𝜁𝑙𝑘 , if 𝑥𝑙 − < 𝑥 ≤ 𝑥𝑙 + ,
{
{ 2 2
𝑉𝑙𝑘 being the approximate solution of the main problem at
𝜁𝑘 (𝑥) = { 𝑙 = 1, 2, . . . , 𝑀 − 1, (17)
the point (𝑥𝑙 , 𝑡𝑘 ), 𝑘 = 1, . . . , 𝑁; 𝑙 = 1, . . . , 𝑀; in addition, 𝜁𝑘 is {
{
{ ℎ
the transpose of the matrix [𝜁1𝑘 , 𝜁2𝑘 , . . . , 𝜁𝑀 𝑘
]. 0, if 𝐿 − < 𝑥 ≤ 𝐿.
{ 2
4 Abstract and Applied Analysis

Then, the function 𝜁𝑘 (𝑥) can be expressed in Fourier series as And for 𝑘 ≥ 1, we have
follows:
𝜏−𝛼 𝜌ℎ 𝑎𝑘
𝑚=∞
2𝑖𝜋𝑚𝑘 (Δ𝑡 + + 2sin2 ( ) 𝑙2 ) 𝛿𝑘+1
𝜁𝑘 (𝑥) = ∑ 𝛿𝑚 (𝑚) exp [ ], Γ (2 − 𝛼) 2 ℎ
𝑚=−∞ 𝐿
(18) 𝜏−𝛼 𝜌ℎ
𝐿 = (Δ𝑡 + + 2sin2 ( ) 𝛿 − 𝑑1 ) 𝛿𝑘 (25)
1 2𝑖𝜋𝑚𝑥 Γ (2 − 𝛼) 2
𝛿𝑘 (𝑥) = ∫ 𝜌𝑘 (𝑥) exp [ ] 𝑑𝑥.
𝐿 0 𝐿
𝜏−𝛼 𝑘−1
It was proven in [22, 29, 30] that − ∑ [𝛿 ] 𝑑 + 𝑏 𝛿 .
Γ (2 − 𝛼) 𝑗=1 𝑘−𝑗 𝑗 𝑘+1 0
󵄩󵄩 2 󵄩󵄩2 𝑚=∞ 󵄩󵄩
󵄩󵄩𝜌 󵄩󵄩 = ∑ 󵄩󵄩𝛿𝑘 (𝑚)󵄩󵄩󵄩󵄩2 . (19) Observe that 𝛿 ≤ 𝑎𝑙0 /ℎ2 + 𝑑1 then, the following is obtained:
󵄩 󵄩2
𝑚=−∞

Let us now examine the stability of the implicit scheme of the (Δ𝑡 + 𝜏−𝛼 /Γ (2 − 𝛼) + 2sin2 (𝜌ℎ/2) 𝛿 − 𝑑1 )
𝛿1 = 𝛿0 . (26)
main problem (Δ𝑡 + 𝜏−𝛼 /Γ (2 − 𝛼) + 2sin2 (𝜌ℎ/V) (𝑎𝑙0 /ℎ2 ))

𝜏−𝛼 𝑎𝑘 Applying the absolute value on both sides of (20), we have


(Δ𝑡 + + 2 𝑙2 ) 𝜁𝑙𝑘+1 󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨󵄨
Γ (2 − 𝛼) ℎ 󵄨󵄨𝛿1 󵄨󵄨 ≤ 󵄨󵄨𝛿0 󵄨󵄨 . (27)
𝜏−𝛼 Now, let us assume that (18) is true for all 2 ≤ 𝑚 ≤ 𝑘; then,
= (Δ𝑡 + + 𝛿) 𝜁𝑙𝑘 + 𝑎𝑙𝑘 𝜁𝑙+1
𝑘+1
+ 𝑎𝑙𝑘 𝜁𝑙−1
𝑘+1
Γ (2 − 𝛼) (20) 𝜏−𝛼 𝜌ℎ
𝑘 𝛿𝑘+1 = ( (Δ𝑡 + + 2sin2 ( ) 𝛿 − 𝑑1 ) 𝛿𝑘
𝜏−𝛼 𝑘−𝑗 Γ (2 − 𝛼) 2
− ∑ [𝜁 ] 𝑑𝑗 + 𝑏𝑘+1 𝜁𝑗0 + 𝑓𝑙𝑘 ,
Γ (2 − 𝛼) 𝑗=1 𝑙
𝜏−𝛼 𝑘−1 (28)
𝑘 = 1, . . . , 𝑀. − ∑ [𝛿 ] 𝑑 + 𝑏 𝛿 )
Γ (2 − 𝛼) 𝑗=1 𝑘−𝑗 𝑗 𝑘+1 0
We next assume that
−1
𝜏−𝛼 𝜌ℎ 𝑎𝑘
𝜁𝑙𝑘 = 𝛿𝑘 exp [𝑖𝜌𝑙𝑘] (21) × (Δ𝑡 + + 2sin2 ( ) 𝑙2 ) .
Γ (2 − 𝛼) 2 ℎ
with 𝜌 being the real space wave number and 𝑖 = √−1. Then, Now, applying the absolute value on both sides of (28), and
by replacing (21) into (20), we obtain the following: making further use of the inequality theorem, we arrived at
the following:
𝜏−𝛼 𝜌ℎ 𝑎𝑘
(Δ𝑡 + + 2sin2 ( ) 𝑙2 ) 𝛿𝑘+1
Γ (2 − 𝛼) 2 ℎ 󵄨󵄨 𝜏−𝛼 𝜌ℎ 󵄨󵄨
󵄨󵄨 󵄨 󵄨 󵄨󵄨 󵄨
󵄨󵄨𝛿𝑘+1 󵄨󵄨󵄨 ≤ ( 󵄨󵄨󵄨Δ𝑡 + + 2sin2 ( ) 𝛿 − 𝑑1 󵄨󵄨󵄨 󵄨󵄨󵄨𝛿𝑘 󵄨󵄨󵄨
󵄨󵄨 Γ (2 − 𝛼) 2 󵄨󵄨
𝜏−𝛼 𝜌ℎ
= (Δ𝑡 + + 2sin2 ( ) 𝛿 − 𝑑1 ) 𝛿𝑘
Γ (2 − 𝛼) 2 (22)
𝜏−𝛼 𝑘−1 󵄨󵄨 󵄨 󵄨 󵄨
−𝛼 𝑘−1 + ∑ 󵄨󵄨𝛿 󵄨󵄨󵄨 𝑑 + 𝑏 󵄨󵄨𝛿 󵄨󵄨) (29)

𝜏
∑ [𝛿 ] 𝑑 + 𝑏 𝛿 + 𝑓𝑙𝑘 , Γ (2 − 𝛼) 𝑗=1 󵄨 𝑘−𝑗 󵄨 𝑗 𝑘+1 󵄨 0 󵄨
Γ (2 − 𝛼) 𝑗=1 𝑘−𝑗 𝑗 𝑘+1 0
󵄨󵄨 𝑘 󵄨󵄨 −1
󵄨󵄨
󵄨 𝜏−𝛼 2 𝜌ℎ 𝑎𝑙 󵄨󵄨󵄨
𝑘 = 0, . . . , 𝑀 − 1. × (󵄨󵄨Δ𝑡 + + 2sin ( ) 2 󵄨󵄨) .
󵄨󵄨 Γ (2 − 𝛼) 2 ℎ 󵄨󵄨󵄨
󵄨
Lemma 8. Assume that 𝛿𝑘 (𝑘 = 0, . . . , 𝑀 − 1) verified (14) and
Making use of the induction hypothesis and factorizing |𝛿0 |,
that for all (𝑙, 𝑘), 𝛿 ≤ 𝑎𝑙𝑘 /ℎ2 + 𝑑1 (𝑙 = 1, . . . , 𝑁; 𝑘 = 1, . . . , 𝑀); we obtain
then, the following inequality is satisfied:
󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨 𝜏−𝛼 𝜌ℎ
󵄨󵄨𝛿𝑘 󵄨󵄨 ≤ 󵄨󵄨𝛿0 󵄨󵄨 , 𝑘 = 1, 2, . . . , 𝑀. (23) 󵄨󵄨𝛿𝑘+1 󵄨󵄨󵄨 ≤ [(Δ𝑡 + + 2sin2 ( ) 𝛿 − 𝑑1
Γ (2 − 𝛼) 2
Proof. To prove this lemma, we make use of the recursive [
method on the natural number 𝑛 from (3) and (15), we have
𝜏−𝛼 𝑘−1
that, for all 𝑙 = 1, . . . , 𝑁 and for 𝑘 = 0, + ∑𝑑 + 𝑏 )
Γ (2 − 𝛼) 𝑗=1 𝑗 𝑘+1
𝜏−𝛼 𝜌ℎ 𝑎0
(Δ𝑡 + + 2sin2 ( ) 𝑙2 ) 𝛿1 󵄨󵄨 𝑘 󵄨󵄨 −1
Γ (2 − 𝛼) 2 ℎ 󵄨󵄨 𝜏−𝛼 2 𝜌ℎ 𝑎𝑙 󵄨󵄨󵄨 󵄨 󵄨
(24) 󵄨
×(󵄨󵄨Δ𝑡 + + 2sin ( ) 2 󵄨󵄨) ] 󵄨󵄨󵄨𝛿0 󵄨󵄨󵄨 .
󵄨󵄨 Γ (2 − 𝛼) 2 ℎ 󵄨󵄨
𝜏−𝛼 𝜌ℎ 󵄨 󵄨 ]
= (Δ𝑡 + + 2sin2 ( ) 𝛿 − 𝑑1 ) 𝛿0 .
Γ (2 − 𝛼) 2 (30)
Abstract and Applied Analysis 5

But we have that Here, 𝑇𝑙𝑘+1 is the truncate or the remainder term of the
approximation and has the following expression:
𝑘−1
∑ 𝑑𝑗 = 1 − 𝑏𝑘+1 , 0 ≤ 𝑑𝑗 ≤ 1. (31) 𝜏−𝛼 𝑘−1 𝑘−𝑗
𝑗=1 𝑇𝑙𝑘+1 = V (𝑥𝑙 , 𝑡𝑘+1 ) + ∑𝛽 𝑏
Γ (2 − 𝛼) 𝑗=1 𝑙 𝑗
Therefore,
+ 𝑎 (𝑥𝑙 , 𝑡𝑘 ) [V (𝑥𝑙+1 , 𝑡𝑘+1 )
−𝛼 2
󵄨󵄨 󵄨 Δ𝑡 + 𝜏 /Γ (2 − 𝛼) + 2sin (𝜌ℎ/2) 𝛿 − 𝑑1 󵄨󵄨 󵄨󵄨 −2V (𝑥𝑙 , 𝑡𝑘+1 ) + V (𝑥𝑙−1 , 𝑡𝑘+1 )]
󵄨󵄨𝛿𝑘+1 󵄨󵄨󵄨 ≤ [ 󵄨󵄨 󵄨 ] 󵄨𝛿0 󵄨 .
󵄨󵄨Δ𝑡 + 𝜏−𝛼 /Γ (2 − 𝛼) + 2sin2 (𝜌ℎ/2) (𝑎𝑙𝑘 /ℎ2 )󵄨󵄨󵄨 󵄨 󵄨
(32) + (V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )) − 𝑓 (𝑥𝑙 , 𝑡𝑘 ) + 𝛿V (𝑥𝑙 , 𝑡𝑘 ) .
(36)
Thus, It follows from (3), (4), and (5) that,
󵄨󵄨 󵄨 󵄨 󵄨 𝜕𝛼 V (𝑥𝑙 , 𝑡𝑘+1 )
󵄨󵄨𝛿𝑘+1 󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨𝛿0 󵄨󵄨󵄨 . (33)
+ 𝐷1 𝜏
𝜕𝑡𝛼
𝜏−𝛼
=
Theorem 9. The implicit difference scheme for the time- Γ (2 − 𝛼)
fractional parabolic equation (1) is stable providing that, for
all (𝑙, 𝑘), 𝛿 ≤ 𝑎𝑙𝑘 /ℎ2 + 𝑑1 (𝑙 = 1, . . . , 𝑁; 𝑘 = 1, . . . , 𝑀). × (V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
Proof. From (19) and Lemma 8, we obtain
𝑘−1
󵄩󵄩 2 󵄩󵄩 󵄩 󵄩 (37)
󵄩󵄩𝜁 󵄩󵄩 ≤ 󵄩󵄩󵄩𝜁0 󵄩󵄩󵄩 , (34) + ∑ V (𝑥𝑙 , 𝑡𝑘−𝑗 ) 𝑑𝑗 + 𝑑𝑘+1 V (𝑥𝑙 , 𝑡0 )) ,
󵄩 󵄩2 󵄩 󵄩2
𝑗=1

and this proves that the implicit difference scheme for the
𝜕2 V (𝑥𝑙 , 𝑡𝑘+1 )
time-fractional parabolic equation (1) is stable. + ℎ2 𝐷2
𝜕𝑥2
Remark 10. It is observed that, from Theorem 9, the statement V (𝑥𝑙+1 , 𝑡𝑘+1 ) − 2V (𝑥𝑙 , 𝑡𝑘+1 ) + V (𝑥𝑙−1 , 𝑡𝑘+1 )
of stability of implicit difference scheme for time-fractional = ,
parabolic equation (1) depends on the evolution of the ℎ2
function 𝑎(𝑥, 𝑡). It follows that the stability condition can 𝜕V (𝑥𝑙 , 𝑡𝑘+1 ) V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
change in time advancement and space position. + Δ𝑡𝐷3 = .
𝜕𝑡 Δ𝑡
Thus, from (36) and (37), we have the following:
3.1.2. Convergence Analysis of the Implicit Difference Scheme.
Assuming that V(𝑥𝑙 , 𝑡𝑘 ) (𝑙 = 0, . . . , 𝑀 − 1; 𝑘 = 0, . . . , 𝑁 − 1) is 𝑇𝑙𝑘+1 ≤ 𝐷 (𝜏1+𝛼 + ℎ2 𝜏𝛼 + Δ𝑡𝜏𝛼 ) , (38)
the exact solution of (1) at the point (𝑥𝑙 , 𝑡𝑘 ), then, by defining,
𝛽𝑙𝑘 = V(𝑥𝑙 , 𝑡𝑘 ) − V𝑙𝑘 , and 𝛽𝑘 is the transpose of the matrix where 𝐷1 , 𝐷2 , 𝐷3 , and 𝐷 are constant. The interested reader
(𝛽1𝑘 , 𝛽2𝑘 , . . . , 𝛽𝑁
𝑘
); here, 𝛽0 is neglected because of being equal can find the error analysis of the chosen fractional derivative
to zero. Therefore, we have the following relation for the (the Caputo fractional derivative) in [29, 30].
implicit difference scheme for the time-fractional parabolic −1
equation (1) and from (13): Lemma 11. One has that ‖𝛽𝑘 ‖∞ ≤ 𝐷(𝑏𝑘+1 ) [𝜏1+𝛼 + 𝜏𝛼 ℎ2 +
Δ𝑡𝜏𝛼 ] is true for all 𝑘 = 0, 1, . . . , 𝑀 − 1.
𝜏−𝛼 𝑎0
(Δ𝑡 + + 2 𝑙2 ) 𝛽𝑙1 − 𝑎𝑙0 𝛽𝑙+1
1
Proof. Again we employ the induction method to achieve
Γ (2 − 𝛼) ℎ this, so that, for 𝑘 = 0,
− 𝑎𝑙0 𝛽𝑙−1
1
− 𝑓𝑙0 = 𝑇𝑙1 , for 𝑘 = 0, 𝜏−𝛼 𝑎0 󵄨 󵄨 󵄨 1 󵄨󵄨 󵄨 1 󵄨󵄨 󵄨󵄨 1 󵄨󵄨
(Δ𝑡 + + 2 𝑙2 ) 󵄨󵄨󵄨󵄨𝛽𝑙1 󵄨󵄨󵄨󵄨 − 𝑎𝑙0 󵄨󵄨󵄨󵄨𝛽𝑙+1 󵄨󵄨 − 𝑎𝑙0 󵄨󵄨󵄨𝛽𝑙−1
󵄨 󵄨
󵄨󵄨 ≥ 󵄨󵄨𝛽𝑙 󵄨󵄨 ,
󵄨 󵄨 󵄨
Γ (2 − 𝛼) ℎ
𝜏−𝛼 𝑎𝑘 𝜏−𝛼
(Δ𝑡 + + 2 𝑙2 ) 𝛽𝑙𝑘+1 − (Δ𝑡 + + 𝛿) 𝛽𝑙𝑘 (39)
Γ (2 − 𝛼) ℎ Γ (2 − 𝛼)
󵄨 󵄨󵄨
󵄨󵄨 1 󵄨󵄨 󵄨󵄨󵄨 𝜏−𝛼 𝑎𝑙0 1 0 1 0 1 󵄨󵄨
− 𝑎𝑙𝑘 𝛽𝑙+1
𝑘+1
− 𝑎𝑙𝑘 𝛽𝑙−1
𝑘+1
− 𝑓𝑙𝑘 󵄨󵄨𝑇𝑙 󵄨󵄨 = 󵄨󵄨(Δ𝑡 + + 2 ) 𝛽 − 𝑎 𝛽 − 𝑎 𝛽
𝑙 𝑙−1 󵄨󵄨
󵄨
󵄨 󵄨 󵄨󵄨 Γ (2 − 𝛼) ℎ2 𝑙 𝑙 𝑙+1
󵄨󵄨
󵄨
𝜏−𝛼 𝑘−1 𝑘−𝑗 󵄨 󵄨
≥ 󵄨󵄨󵄨󵄨𝛽𝑙1 󵄨󵄨󵄨󵄨 ,
(40)
=− ∑ 𝛽 𝑏 + 𝑇𝑙𝑘 , for 𝑘 ≥ 1.
Γ (2 − 𝛼) 𝑗=1 𝑙 𝑗
−1 󵄨 󵄨
(35) 𝐷(𝑏1 ) (𝜏1+𝛼 + ℎ2 𝜏𝛼 + Δ𝑡𝜏𝛼 ) ≥ 󵄨󵄨󵄨󵄨𝑇𝑙1 󵄨󵄨󵄨󵄨 .
6 Abstract and Applied Analysis

Now, assuming that, for all 𝑗 = 0, . . . , 𝑀 − 2, ‖𝛽𝑗 ‖∞ ≤ The discretization of the Caputo-type time-fractional-order
−1
𝐷(𝑏𝑗+1 ) [𝜏1+𝛼 + 𝜏𝛼 ℎ2 + Δ𝑡𝜏𝛼 ], then, derivative can be presented as follows:

𝜏−𝛼 𝑎𝑘 󵄨 󵄨 𝜕𝛼 V (𝑥𝑙 , 𝑡𝑘+1 )


(Δ𝑡 + + 2 𝑙2 ) 󵄨󵄨󵄨󵄨𝛽𝑙𝑘+1 󵄨󵄨󵄨󵄨
Γ (2 − 𝛼) ℎ 𝜕𝑡𝛼
𝜏−𝛼 󵄨 󵄨 󵄨 𝑘+1 󵄨󵄨 󵄨 𝑘+1 󵄨󵄨 𝜏−𝛼
− (Δ𝑡 + + 𝛿) 󵄨󵄨󵄨󵄨𝛽𝑙𝑘 󵄨󵄨󵄨󵄨 − 𝑎𝑙𝑘 󵄨󵄨󵄨󵄨𝛽𝑙+1 󵄨󵄨 − 𝑎𝑙𝑘 󵄨󵄨󵄨𝛽𝑙−1
󵄨 󵄨
󵄨󵄨
󵄨
= (V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
Γ (2 − 𝛼) Γ (2 − 𝛼)
󵄨 󵄨
≥ 󵄨󵄨󵄨󵄨𝛽𝑙𝑘+1 󵄨󵄨󵄨󵄨 , 𝑘
(45)
+ ∑ [V (𝑥𝑙 , 𝑡𝑘+1−𝑗 ) − V (𝑥𝑙 , 𝑡𝑘−𝑗 )]
󵄨󵄨 𝑗=1
󵄨󵄨
󵄨󵄨(Δ𝑡 + 𝜏
−𝛼
𝑎𝑙𝑘
󵄨󵄨 + 2 2
) 𝛽𝑙𝑘+1 1−𝛼 1−𝛼
󵄨󵄨 Γ (2 − 𝛼) ℎ × [(𝑗 + 1) − (𝑗) ]) ,
−𝛼 󵄨
𝜏 𝑘+1 󵄨󵄨󵄨 𝜕V (𝑥𝑙 , 𝑡𝑘 ) V (𝑥𝑙+1 , 𝑡𝑘 ) − V (𝑥𝑙 , 𝑡𝑘 )
− (Δ𝑡 + + 𝛿) 𝛽𝑙𝑘 − 𝑎𝑙𝑘 𝛽𝑙+1
𝑘+1
− 𝑎𝑙𝑘 𝛽𝑙−1 󵄨󵄨 = + 𝑂 (Δ𝑡) .
Γ (2 − 𝛼) 󵄨󵄨 𝜕𝑡 Δ𝑡
󵄨󵄨 𝑘+1 󵄨󵄨
≥ 󵄨󵄨󵄨𝛽𝑙 󵄨󵄨󵄨 ,
Now, substituting (44) and (45) into (1), we obtained the
󵄨󵄨 󵄨󵄨 following expression:
󵄨󵄨 𝜏−𝛼 𝑘−1 𝑘−𝑗 𝑘+1 󵄨󵄨
󵄨 󵄨 󵄨
󵄨󵄨 ∑ 𝛽𝑙 𝑏𝑗 + 𝑇𝑙 󵄨󵄨󵄨 ≥ 󵄨󵄨󵄨󵄨𝛽𝑙𝑘+1 󵄨󵄨󵄨󵄨 ,
󵄨󵄨
󵄨󵄨 Γ (2 − 𝛼) 𝑗=1 󵄨󵄨
󵄨 󵄨 V (𝑥𝑙+1 , 𝑡𝑘 ) − V (𝑥𝑙 , 𝑡𝑘 )
𝜏−𝛼 𝑘−1 󵄨󵄨󵄨 𝑘−𝑗 󵄨󵄨󵄨 󵄨 󵄨 Δ𝑡
∑ 󵄨𝛽 󵄨 𝑏 + 𝑇𝑙𝑘+1 ≥ 󵄨󵄨󵄨󵄨𝛽𝑙𝑘+1 󵄨󵄨󵄨󵄨 .
Γ (2 − 𝛼) 𝑗=1 󵄨󵄨 𝑙 󵄨󵄨 𝑗 𝜏−𝛼
+ (V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
(41) Γ (2 − 𝛼)
𝑘
Making use of the induction hypothesis, we obtain the
+ ∑ [V (𝑥𝑙 , 𝑡𝑘+1−𝑗 ) − V (𝑥𝑙 , 𝑡𝑘−𝑗 )]
following: 𝑗=1 (46)

𝜏−𝛼 𝑘−1󵄩󵄩󵄩 𝑘−𝑗 󵄩󵄩󵄩 󵄨 󵄨 × [(𝑗 + 1)


1−𝛼
− (𝑗)
1−𝛼
])
∑ 󵄩𝛽 󵄩 𝑏 + [𝜏1+𝛼 + 𝜏𝛼 ℎ2 + Δ𝑡𝜏𝛼 ] ≥ 󵄨󵄨󵄨󵄨𝛽𝑙𝑘+1 󵄨󵄨󵄨󵄨 ,
Γ (2 − 𝛼) 𝑗=1󵄩󵄩 𝑙 󵄩󵄩∞ 𝑗
V (𝑥𝑙+1 , 𝑡𝑘 ) − 2V (𝑥𝑙 , 𝑡𝑘 ) + V (𝑥𝑙−1 , 𝑡𝑘 )
−1 − 𝑎𝑙𝑘 ( )
0
(𝑏 + 𝑏
𝑘+1
−𝑏 𝑘+1
) (𝑏 𝑘+1
) × 𝐷 [𝜏 1+𝛼 𝛼 2
+ 𝜏 ℎ + Δ𝑡𝜏 ] 𝛼 ℎ2

󵄨 󵄨 + 𝛿V (𝑥𝑙 , 𝑡𝑘 ) = 𝑓 (𝑥𝑙 , 𝑡𝑘 ) .
≥ 󵄨󵄨󵄨󵄨𝛽𝑙𝑘+1 󵄨󵄨󵄨󵄨 ,
−1 󵄨 󵄨 For simplicity, let
(𝑏𝑘+1 ) × 𝐷 [𝜏1+𝛼 + 𝜏𝛼 ℎ2 + Δ𝑡𝜏𝛼 ] ≥ 󵄨󵄨󵄨󵄨𝛽𝑙𝑘+1 󵄨󵄨󵄨󵄨 .
(42)
V𝑙𝑘 = V (𝑥𝑙 , 𝑡𝑘 ) , 𝑓𝑙𝑘 = 𝑓 (𝑥𝑙 , 𝑡𝑘 ) ,
This completes the proof. (47)
1−𝛼 1−𝛼
𝑎 (𝑥𝑙 , 𝑡𝑘 ) = 𝑎𝑙𝑘 , (𝑗 + 1) − (𝑗) = 𝑏𝑗 .
Theorem 12. The implicit difference scheme of the time-
fractional parabolic equation (1) is convergent, and there exists
Then, (37) can be rewritten as follows
a constant 𝐷 such that,
󵄨󵄨 󵄨
󵄨󵄨V (𝑥𝑙 , 𝑡𝑘 ) − V𝑙𝑘 󵄨󵄨󵄨 ≤ 𝐷 (𝜏1+𝛼 + 𝜏𝛼 ℎ2 + Δ𝑡𝜏𝛼 ) , V𝑙𝑘+1 = V𝑙𝑘 [𝐵1 + 1 − 𝑟𝑙𝑘 − 𝛿] + V𝑙+1
𝑘
[𝑟𝑙𝑘 −𝐵1 ]
󵄨 󵄨
(43)
𝑓𝑜𝑟 (𝑙 = 0, . . . , 𝑁 − 1; 𝑘 = 0, . . . , 𝑀 − 1) . 𝑘 (48)
𝑘+1−𝑗 𝑘−𝑗
+ 𝑟𝑙𝑘 V𝑙−1
𝑘
+ 𝐵1 𝑓𝑙𝑘 − ∑ [V𝑙 − V𝑙 ] 𝑏𝑗 ,
𝑗=1
3.2. Explicit Difference Scheme for Time-Fractional Parabolic
Equation. It is important to recall that the finite-difference
approximation for the second-order spatial derivative is where 𝐵1 = Γ(2 − 𝛼)𝜏𝛼 and 𝑟𝑙𝑘 = (Γ(2 − 𝛼)𝜏𝛼 /ℎ2 )𝑎𝑙𝑘 .
known as follows:
𝜕2 V (𝑥𝑙 , 𝑡𝑘 ) V (𝑥𝑙+1 , 𝑡𝑘 ) − 2V (𝑥𝑙 , 𝑡𝑘 ) + V (𝑥𝑙−1 , 𝑡𝑘 ) 3.2.1. Stability of the Explicit Difference Scheme of the Time-
= + 𝑂 (ℎ2 ) .
𝜕𝑥2 ℎ2 Fractional Parabolic Equation. Following the discussion pre-
(44) sented earlier for the analysis of the implicit scheme, we
Abstract and Applied Analysis 7

obtain the following roundoff error equation from explicit Now, making use of the induction hypothesis, we arrive at the
scheme equation (1): following:

𝛽𝑙𝑘+1 = 𝛽𝑙𝑘 [𝐵1 + 1 − 𝑟𝑙𝑘 − 𝛿] + 𝛽𝑙+1


𝑘
[𝑟𝑙𝑘 −𝐵1 ] 󵄨󵄨 󵄨󵄨
󵄨󵄨 󵄨󵄨󵄨󵄨 󵄨󵄨 󵄨 󵄨
𝑘−1
󵄨󵄨 2 𝜎ℎ 𝑘 󵄨󵄨 󵄨
𝛿 ≤
󵄨󵄨 𝑘+1 󵄨󵄨 󵄨󵄨 1 [𝐵 + 1 − sin ( ) 𝑟𝑙 − 𝛿] + ∑ 𝑑𝑗 + 𝑑𝑘+1 󵄨󵄨󵄨 󵄨󵄨𝛿0 󵄨󵄨 .
𝑘 󵄨󵄨 2 󵄨󵄨
𝑘−𝑗 󵄨 𝑗=1 󵄨
+ 𝑟𝑙𝑘 𝛽𝑙−1
𝑘
+ 𝐵1 𝑓𝑙𝑘 − ∑ [𝛽𝑙 ] 𝑑𝑗 − 𝑑𝑘+1 𝛽0 , (49) (58)
𝑗=1

for 𝑘 = 1, . . . , 𝑀 − 1, And this produces, since ∑𝑘−1


𝑗=1 𝑑𝑗 + 𝑑𝑘+1 = 1,

󵄨 󵄨󵄨
where 󵄨󵄨 󵄨 󵄨󵄨 2 𝜎ℎ 󵄨󵄨 󵄨 󵄨 󵄨
󵄨󵄨𝛿𝑘+1 󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨[𝐵1 + 1 − sin ( ) 𝑟𝑙 − 𝛿]󵄨󵄨󵄨 󵄨󵄨󵄨𝛿0 󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨𝛿0 󵄨󵄨󵄨 .
𝑘
(59)
󵄨󵄨 2 󵄨󵄨
𝑑𝑗 = 𝑏𝑗+1 − 𝑏𝑗 . (50)
This completes the proof.
Again, we suppose that 𝛽𝑙𝑘 in (39) can be given in exponential
form as follows: Theorem 14. The explicit difference scheme of time-fractional
parabolic equation (1) is stable under the condition that, for all
𝛽𝑙𝑘 = 𝛿𝑘 𝑒(−𝑖𝜎𝑙𝑘) . (51)
(𝑙 = 0, 1, . . . , 𝑁 − 1; 𝑘 = 0, . . . , 𝑀 − 1), 𝑟𝑙𝑘 ≤ 1 − 𝛿 + 𝐵1 and
Then, replacing this form into (49), we arrive at the following 𝑟𝑙𝑘 ≥ 𝐵1 − 𝛿.
expression:
Proof. It is straightforward from (11) and Lemma 13, that,
𝜎ℎ
𝛿𝑘+1 = 𝛿𝑘 [𝐵1 + 1 − sin ( ) 𝑟𝑙𝑘 − 𝛿]
2
󵄩󵄩 2 󵄩󵄩 󵄩 󵄩
2 󵄩󵄩𝜁 󵄩󵄩 ≤ 󵄩󵄩󵄩𝜁0 󵄩󵄩󵄩 (60)
󵄩 󵄩2 󵄩 󵄩2
𝑘−1
(52)
− ∑ 𝛿𝑘−𝑗 𝑑𝑗 + 𝛿0 𝑑𝑘+1 , for 𝑘 = 0, . . . , 𝑀 − 1. and this proves that the explicit difference scheme for the
𝑗=1 time-fractional parabolic equation (3) is stable.

Lemma 13. Suppose that 𝛿𝑘 is the solution of (40) for all (𝑙 =


3.2.2. Convergence Analysis of the Explicit Scheme
0, . . . , 𝑁−1; 𝑘 = 0, . . . , 𝑀−1), 𝑟𝑙𝑘 ≤ 1−𝛿+𝐵1 , and 𝑟𝑙𝑘 ≥ 𝐵1 −𝛿,
then, the following inequality holds: Theorem 15. The explicit scheme is convergent if, for all (𝑙 =
󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨󵄨 0, 1, . . . , 𝑁−1; 𝑘 = 0, . . . , 𝑀−1), 𝑟𝑙𝑘 ≤ 1−𝛿+𝐵1 and 𝑟𝑙𝑘 ≥ 𝐵1 −𝛿
󵄨󵄨𝛿𝑘 󵄨󵄨 ≤ 󵄨󵄨𝛿0 󵄨󵄨 , 𝑓𝑜𝑟 𝑘 ≥ 1. (53)
and there exists a constant 𝐷 such that
Proof. To prove this, we make use of the recurrence technique 󵄨󵄨 𝑘 󵄨
on the natural number 𝑘, and we will examine first the case 󵄨󵄨V𝑙 − V (𝑥𝑙 , 𝑡𝑘 )󵄨󵄨󵄨 ≤ 𝐷 (𝜏 + ℎ2 + Δ𝑡) . (61)
󵄨 󵄨
for 𝑘 = 1. Therefore, if 𝑘 = 1, we have that
Using the same method in implicit difference, the above
𝜎ℎ
𝛿1 = 𝛿0 [𝐵1 + 1 − sin ( ) 𝑟𝑙0 − 𝛿] .
2
(54) theorem can be proven.
2
Since 𝑟𝑙0 ≤ 1 − 𝛿 + 𝐵1 and 𝑟𝑙0 ≥ 𝐵1 − 𝛿, then, we have 4. Conclusions
󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨󵄨
󵄨󵄨𝛿1 󵄨󵄨 ≤ 󵄨󵄨𝛿0 󵄨󵄨 . (55) Partial differential equations are sometimes very difficult to
be solved analytically. It is, therefore, sometimes easy to solve
We next assume that, for all 𝑗 = 1, . . . , 𝑘, |𝛿𝑘 | ≤ |𝛿0 |; then them via numerical techniques. In this paper, we solve the
󵄨󵄨󵄨 time-fractional parabolic equation (1) via the implicit and
󵄨󵄨 󵄨󵄨 󵄨󵄨󵄨 2 𝜎ℎ 𝑘 explicit difference schemes. We study the stability and the
󵄨󵄨𝛿𝑘+1 󵄨󵄨 = 󵄨󵄨𝛿𝑘 [𝐵1 + 1 − sin ( ) 𝑟𝑙 − 𝛿]
󵄨󵄨 2 convergence of implicit and explicit difference schemes.
󵄨󵄨
(56)
󵄨󵄨
𝑘−1 󵄨󵄨 References
− ∑ 𝛿𝑘−𝑗 𝑑𝑗 + 𝛿0 𝑑𝑘+1 󵄨󵄨󵄨󵄨 .
󵄨󵄨
𝑗=1 󵄨 [1] A. Ashyralyev, F. Dal, and Z. Pinar, “On the numerical solution
of fractional hyperbolic partial differential equations,” Mathe-
Now, making use of the triangular inequality, we arrive at the matical Problems in Engineering, vol. 2009, Article ID 730465,
following inequality: 2009.
󵄨 󵄨󵄨
󵄨󵄨 󵄨 󵄨󵄨 2 𝜎ℎ 󵄨 [2] A. Ashyralyev, “A note on fractional derivatives and fractional
󵄨󵄨𝛿𝑘+1 󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨𝛿𝑘 [𝐵1 + 1 − sin ( ) 𝑟𝑙 − 𝛿]󵄨󵄨󵄨
𝑘
powers of operators,” Journal of Mathematical Analysis and
󵄨󵄨 2 󵄨󵄨
Applications, vol. 357, no. 1, pp. 232–236, 2009.
𝑘−1
(57) [3] I. Podlubny and A. M. A. El-Sayed, On Two Definitions of
󵄨󵄨 󵄨󵄨 󵄨 󵄨
+ ∑ 󵄨󵄨󵄨𝛿𝑘−𝑗 󵄨󵄨󵄨 𝑑𝑗 + 󵄨󵄨󵄨𝛿0 󵄨󵄨󵄨 𝑑𝑘+1 . Fractional Calculus, Slovak Academy of Science, Institute of
𝑗=1 Experimental Physics, 1996.
8 Abstract and Applied Analysis

[4] A. Yakar and M. E. Koksal, “Existence results for solutions diffusion equation,” Journal of Computational Physics, vol. 213,
of nonlinear fractional differential equations,” Abstract and no. 1, pp. 205–213, 2006.
Applied Analysis, vol. 2012, Article ID 267108, 12 pages, 2012. [22] C. M. Chen, F. Liu, I. Turner, and V. Anh, “A Fourier method
[5] A. Atangana and J. F. Botha, “Generalized groundwater flow for the fractional diffusion equation describing sub-diffusion,”
equation using the concept of variable order derivative,” Bound- Journal of Computational Physics, vol. 227, no. 2, pp. 886–897,
ary Value Problems, vol. 2013, article 53, 2013. 2007.
[6] A. Atangana and S. C. Oukouomi Noutchie, “Stability and con- [23] S. B. Yuste and L. Acedo, “An explicit finite difference method
vergence of a tme-fractional variable order Hantush equation and a new von Neumann-type stability analysis for fractional
for a deformable aquifer,” Abstract and Applied Analysis, vol. diffusion equations,” SIAM Journal on Numerical Analysis, vol.
2013, Article ID 691060, 8 pages, 2013. 42, no. 5, pp. 1862–1874, 2005.
[7] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional [24] I. Podlubny, A. Chechkin, T. Skovranek, Y. Chen, and B. M.
Calculus Models and Numerical Methods Series on Complexity, Vinagre Jara, “Matrix approach to discrete fractional calculus
Nonlinearity and Chaos, World Scientific, 2012. II: partial fractional differential equations,” Journal of Compu-
[8] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory tational Physics, vol. 228, no. 8, pp. 3137–3153, 2009.
and Applications of Fractional Differential Equations, vol. 204 [25] E. Hanert, “On the numerical solution of space-time fractional
of North-Holland Mathematics Studies, Elsevier Science B.V., diffusion models,” Computers and Fluids, vol. 46, no. 1, pp. 33–
Amsterdam, The Netherlands, 2006. 39, 2011.
[9] K. B. Oldham and J. Spanier, The Fractional Calculus, vol. 111 of [26] P. Zhuang, F. Liu, V. Anh, and I. Turner, “Numerical methods for
Mathematics in Science and Engineering, Academic Press, New the variable-order fractional advection-diffusion equation with
York, NY, USA, 1974. a nonlinear source term,” SIAM Journal on Numerical Analysis,
[10] B. Ross, Ed., Fractional Calculus and Its Applications, vol. 457 of vol. 47, no. 3, pp. 1760–1781, 2009.
Lecture Notes in Mathematics, Springer, Berlin, Germany, 1975. [27] R. Lin, F. Liu, V. Anh, and I. Turner, “Stability and conver-
[11] Y. Luchko and R. Gorenflo, “The initial value problem for some gence of a new explicit finite-difference approximation for the
fractional differential equations with the Caputo derivative,” variable-order nonlinear fractional diffusion equation,” Applied
Preprint Series A08–98, Freic Universitat Berlin, 1998, Fachbre- Mathematics and Computation, vol. 212, no. 2, pp. 435–445,
ich Mathematik and Informatik. 2009.
[12] A. Atangana and E. Alabaraoye, “Solving system of fractional [28] J. Crank and P. Nicolson, “A practical method for numerical
partial differential equations arisen in the model of HIV evaluation of solutions of partial differential equations of the
infection of CD4+ cells and attractor one-dimensional Keller- heat conduction type,” Proceedings of the Cambridge Philosoph-
Segel equation,” Advances in Difference Equations, vol. 2013, ical Society, vol. 43, no. 1, pp. 50–67, 1947.
article 94, 2013. [29] K. Diethelm, N. J. Ford, and A. D. Freed, “Detailed error analysis
[13] I. Podlubny, Fractional Differential Equations, vol. 198 of Math- for a fractional Adams method,” Numerical Algorithms, vol. 36,
ematics in Science and Engineering, Academic Press, San Diego, no. 1, pp. 31–52, 2004.
Calif, USA, 1999. [30] C. P. Li and C. X. Tao, “On the fractional Adams method,”
[14] A. Atangana and A. Secer, “A Note on fractional order deriva- Computers and Mathematics with Applications, vol. 58, no. 8, pp.
tives and table of fractional derivatives of some special func- 1573–1588, 2009.
tions,” Abstract and Applied Analysis, vol. 2013, Article ID
279681, 8 pages, 2013.
[15] K. S. Miller and B. Ross, An Introduction to the Fractional
Calculus and Fractional Differential Equations, John Wiley &
Sons, New York, NY, USA, 1993.
[16] A. Atangana and A. Kilicman, “Analytical solutions of the space-
time fractional derivative of advection dispersion equation,”
Mathematical Problems in Engineering, vol. 2013, Article ID
853127, 9 pages, 2013.
[17] X.-J. Yang, D. Baleanu, and J. A. Tenreiro Machado, “Systems of
navier-stokes equations on cantor sets,” Mathematical Problems
in Engineering, vol. 2013, Article ID 769724, 8 pages, 2013.
[18] F. Mainardi, “Fractional calculus: some basic problems in
continuum and statistical mechanics,” in Fractals and Fractional
Calculus in Continuum Mechanics, vol. 378 of CISM Courses and
Lectures, pp. 291–348, Springer, Vienna, Austria, 1997.
[19] M. M. Meerschaert and C. Tadjeran, “Finite difference approx-
imations for fractional advection-dispersion flow equations,”
Journal of Computational and Applied Mathematics, vol. 172, no.
1, pp. 65–77, 2004.
[20] Y. Zhang, “A finite difference method for fractional partial
differential equation,” Applied Mathematics and Computation,
vol. 215, no. 2, pp. 524–529, 2009.
[21] C. Tadjeran, M. M. Meerschaert, and H. P. Scheffler, “A second-
order accurate numerical approximation for the fractional
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 176180, 8 pages
http://dx.doi.org/10.1155/2013/176180

Research Article
Existence Results for a Class of Fractional Differential Equations
with Periodic Boundary Value Conditions and with Delay

Hadi Karami,1 Azizollah Babakhani,1 and Dumitru Baleanu2,3


1
Department of Mathematics, Faculty of Basic Science, Babol University of Technology, Babol 47148-71167, Iran
2
Department of Mathematics and Computer Science, Cankaya University, Turkey
3
Institute of Space Sciences, P.O. Box, MG-23, 76900 Magurele-Bucharest, Romania

Correspondence should be addressed to Azizollah Babakhani; babakhani@nit.ac.ir

Received 11 June 2013; Accepted 15 August 2013

Academic Editor: Ali H. Bhrawy

Copyright © 2013 Hadi Karami et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We discuss the existence and uniqueness of solution for two types of fractional order ordinary and delay differential equations.
Fixed point theorems are the main tool used here to establish the existence and uniqueness results. First we use Banach contraction
principle to prove the uniqueness of solution and then Krasnoselskii’s fixed point theorem to show the existence of the solution
under certain conditions in a Banach space.

1. Introduction applications of fractional calculus amount to replacing the


time derivative in a given evolution equation by a derivative
In mathematics delay differential equations are a type of of fractional order. The results of several studies clearly stated
differential equation in which the derivative of unknown that the fractional derivatives seem to arise generally and
function at a certain time is given in terms of the values of universally from important mathematical reasons. Recently,
the function at previous times. interesting attempts have been made to give physical meaning
While physical events such as acceleration and deceler- to the initial conditions for fractional differential equations
ation take little time compared to the times needed to travel with Riemann-Liouville fractional derivatives which could
most distances, times involved in biological processes such as be found in [8, 9, 36, 37].
gestation and maturation can be substantial when compared Recently Benchoohra et al. [28] studied existence of
to the data-collection times in most population studies. solutions for a class of fractional differential equations with
Therefore, it is often imperative to explicitly incorporate infinite delay; namely,
these process times into mathematical models of population
dynamics. These process times are often called delay times, 𝐷𝛼 𝑢 (𝑡) = 𝑢 (𝑡) + 𝑓 (𝑡, 𝑢𝑡 ) , for each 𝑡 ∈ [0, 𝑏] , 0 < 𝛼 < 1,
and the models that incorporate such delay times are referred
as delay differential equation models [1, 2]. 𝑢 (𝑡) = 𝜙 (𝑡) , 𝑡 ∈ [−∞, 0] ,
Recently theory of fractional differential equations (1)
attracted many scientists and mathematicians to work on
them [3–12]. For the existence of solutions for fractional where 𝐷𝛼 is the standard Riemann-Liouville fractional
differential equations, one can see [13–30] and references derivative and 𝑓 satisfies some assumptions.
therein. The results have been obtained by using fixed First, in this paper we consider nonlinear delayed frac-
point theorems like Picard’s, Schauder fixed-point theorem, tional differential equations:
and Banach contraction mapping principle. About the 𝐷𝛼 𝑢 (𝑡) = 𝑢 (𝑡) + 𝑓 (𝑡, 𝑢𝑡 ) , 𝑡 ∈ (0, 1) , 0 < 𝛼 < 1,
development of existence theorems for fractional functional (2)
differential equations, many publications exist [31–35]. Many 𝑢 (𝑡) = 𝜙 (𝑡) , 𝑡 ∈ [−𝜏, 0] ,
2 Abstract and Applied Analysis

associated with boundary conditions We denote 𝐷𝑎𝛼+ 𝑢(𝑡) as 𝐷𝑎𝛼 𝑦(𝑡) and 𝐷𝑎−𝛼+ 𝑢(𝑡) as 𝐷𝑎−𝛼 𝑢(𝑡).
Further 𝐷0𝛼+ 𝑢(𝑡) and 𝐷0−𝛼+ 𝑢(𝑡) are referred as 𝐷𝛼 𝑢(𝑡) and
𝑢 (0) = lim+ 𝑡1−𝛼 𝑢 (𝑡) = 𝑢 (1) , (3) 𝐷−𝛼 𝑢(𝑡), respectively.
𝑡→0

where 𝐷𝛼 is the standard Riemann-Liouville fractional Definition 2. A two-parameter function of the Mittag-Leffler
derivative and 𝑓 is a continuous function. Here 𝑢𝑡 (⋅) repre- type is defined by
sents the properitoneal state from time −𝜏 up to time 𝑡 which
is defined by 𝑢𝑡 (𝜃) = 𝑢(𝑡 + 𝜃), −𝜏 ≤ 𝜃 ≤ 0. We proved ∞
𝑧𝑘
the uniqueness of existence solutions for (2) with periodic 𝐸𝛼,𝛽 (𝑧) = ∑ , (𝛼 > 0, 𝛽 > 0) . (10)
𝑘=0
Γ (𝛼𝑘 + 𝛽)
boundary condition (3) under some further conditions.
For investigating to establish an existence theorem, we
also consider a class of nonlinear delayed fractional differen- Definition 3. The beta function is usually defined by
tial equations of the form
1
𝐷𝛼 𝑢 (𝑡) = 𝑢 (𝑡) + 𝑓 (𝑡, 𝑢 (𝑡) , 𝑢 (𝑡 − 𝜏)) , 𝑡 ∈ (0, 1) , 𝐵 (𝑧, 𝑤) = ∫ 𝜏𝑧−1 (1 − 𝜏)𝑤−1 𝑑𝜏,
(4) 0 (11)
𝑢 (𝑡) = 𝜙 (𝑡) , 𝑡 ∈ [−𝜏, 0] , 0 < 𝛼 < 1,
(Re (𝑧) > 0, Re (𝑤) > 0) .
with periodic boundary condition
and we have also the following expression for the beta
𝑢 (0) = lim+ 𝑡1−𝛼 𝑢 (𝑡) = 𝑐 = 𝑢 (1) , function:
𝑡→0
(5)
󵄨
𝐷−(1−𝛼) 𝑢(𝑡)󵄨󵄨󵄨󵄨𝑡=0 = 𝑐Γ (𝛼) . Γ (𝑧) Γ (𝑤)
𝐵 (𝑧, 𝑤) = . (12)
Γ (𝑧 + 𝑤)
The paper has been organized as follows. In Section 2 we
give basic definitions and preliminary. Unique solution of Theorem 4 (Arzela-Ascoli’s theorem). A subset of 𝐶[𝑎, 𝑏] is
(2)-(3) under some conditions is proved in Section 3. The compact if and only if it is closed, bounded, and equicontinuous.
existence solution of (4)-(5) under some assumptions has
been presented in Section 4. Theorem 5 (Banach’s fixed point theorem). Consider a metric
space 𝑋 = (𝑋, 𝑑), where 𝑋 ≠0. Suppose that 𝑋 is complete and
2. Preliminaries 𝑇 : 𝑋 → 𝑋 is a contraction on 𝑋. Then 𝑇 has precisely one
fixed point.
For the convenience of the readers, we firstly present the
necessary definitions from the fractional calculus theory and Theorem 6 (Krasnoselskii’s fixed point theorem). Let 𝐵 be a
functional analysis. These definitions and results can be found nonempty closed convex subset of a Banach space (𝑋, ‖ ⋅ ‖).
in the literature [3, 7, 38]. Suppose that 𝑇1 , and 𝑇2 map 𝐵 into 𝑋 such that
Let 𝐶[0, 1] be the Banach space of all continuous real
functions defined on [0, 1] with the norm (1) for any 𝑥, 𝑦 ∈ 𝐵 we have 𝑇1 𝑥 + 𝑇2 𝑦 ∈ 𝐵,
‖𝑢‖ =: max {|𝑢 (𝑡)| : 𝑡 ∈ [0, 1]} . (6) (2) 𝑇1 is a contraction,
Let 𝐶𝑟 [0, 1], 𝑟 ≥ 0, be the space of all functions 𝑓 such that (3) 𝑇2 is continuous and 𝑇2 (𝐵) is contained in a compact
𝑡𝑟 𝑢(𝑡) ∈ 𝐶[0, 1] which is a Banach space when endowed with set.
the norm
Then there exists 𝑧 ∈ 𝐵 such that z = 𝑇1 𝑧 + 𝑇2 𝑧.
‖𝑢‖𝑟 =: max {𝑡𝑟 |𝑢 (𝑡)| : 𝑡 ∈ [0, 1]} . (7)
Definition 1. For a function 𝑢 defined on an interval [𝑎, 𝑏], the 3. Uniqueness of Solution
Riemann-Liouville fractional integral of 𝑢 of order 𝛼 > 0 is
defined by In this section we prove (2) with boundary condition (3) and
𝑡 another condition on 𝑓 has a unique solution. Before proving,
1
𝐷𝑎−𝛼+ 𝑢 (𝑡) = ∫ (𝑡 − 𝑠)𝛼−1 𝑢 (𝑠) 𝑑𝑠, 𝑡 > 𝑎, (8) we need to introduce some notations that will be provided in
Γ (𝛼) 𝑎 the following.
and Riemann-Liouville fractional derivative of 𝑢(𝑡) of order Let Ω = {𝑢 : [−𝜏, 1] → R, 𝑢 ∈ 𝐶1−𝛼 [0, 1]}. Consider the
𝛼 > 0 defined by operator 𝑁 : Ω → Ω defined by

𝑑𝑛 𝜙 (𝑡) , 𝑡 ∈ [−𝜏, 0] ,
𝐷𝑎𝛼+ 𝑢 (𝑡) = {𝐷𝑎−(𝛼−𝑛) 𝑢 (𝑡)} , 𝑛 − 1 < 𝛼 ≤ 𝑛, (9) {
𝑑𝑡𝑛
+
{
{
𝑁𝑢 (𝑡) = { 1 (13)
provided that the right-hand side of the pervious equation is {
{
pointwise defined on (𝑎, +∞). ∫ 𝐺 (𝑡, 𝑠) 𝑓 (𝑠, 𝑢𝑠 ) 𝑑𝑠, 𝑡 ∈ (0, 1) ,
{ 0 1,𝛼
Abstract and Applied Analysis 3

where Proof. We prove that 𝑇 : 𝐶0 → 𝐶0 is a contraction map. For


each 𝑡 ∈ [0, 1] and for 𝑧1 , 𝑧2 ∈ 𝐶0 we have
{ Γ (𝛼) 𝐸𝛼,𝛼 (𝑡𝛼 ) 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) 𝑡𝛼−1 (1 − 𝑠)𝛼−1
{
{ 󵄨 󵄨
{
{
{ 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) 𝑡1−𝛼 󵄨󵄨󵄨𝑇𝑧1 (𝑡) − 𝑇𝑧2 (𝑡)󵄨󵄨󵄨
{
{ 𝛼−1 𝛼
{ +(𝑡 − 𝑠) 𝐸𝛼,𝛼 ((𝑡 − 𝑠) ) ,
{ 1
𝐺1,𝛼 (𝑡, 𝑠) = { 0 ≤ 𝑠 ≤ 𝑡 ≤ 1, 󵄨 󵄨
{ ≤ 𝑡1−𝛼 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨
{
{ Γ (𝛼) 𝐸𝛼,𝛼 (𝑡𝛼 ) 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) 𝑡𝛼−1 (1 − 𝑠)𝛼−1 0
{
{
{
{ , 󵄨 󵄨
(21)
{
{ 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) × 󵄨󵄨󵄨𝑓 (𝑠, 𝑧1𝑠 + 𝑥𝑠 ) − 𝑓 (𝑠, 𝑧2𝑠 + 𝑥𝑠 )󵄨󵄨󵄨 𝑑𝑠,
{ 0 ≤ 𝑡 ≤ 𝑠 ≤ 1, 1
(14) 󵄨 󵄨󵄨 󵄨
≤ 𝐾𝑡1−𝛼 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨 󵄨󵄨󵄨𝑧1𝑠 − 𝑧2𝑠 󵄨󵄨󵄨 𝑑𝑠;
0
is given in Section 4. Let 𝑥(⋅) : [−𝜏, 1] → R be the function
defined by using the definition of ‖ ⋅ ‖1−𝛼 we get

󵄩󵄩 󵄩
0, 𝑡 ∈ (0, 1) , 󵄩󵄩𝑇𝑧1 (𝑡) − 𝑇𝑧2 (𝑡)󵄩󵄩󵄩1−𝛼
𝑥 (𝑡) = { (15)
𝜙 (𝑡) , 𝑡 ∈ [−𝜏, 0] . 1
󵄨 󵄨
≤ 𝐾 max {𝑡1−𝛼 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨 𝑠𝛼−1 𝑑𝑠} (22)
𝑡∈[0,1] 0
1−𝛼
For each 𝑧 ∈ 𝐶1−𝛼 ([0, 1], R) with lim𝑡 → 0+ 𝑡 𝑧(𝑡) = 𝑐 we 󵄩 󵄩
denote 𝑧 the function defined by × 󵄩󵄩󵄩𝑧1 − 𝑧2 󵄩󵄩󵄩1−𝛼 .

Moreover,
𝑧 (𝑡) , 𝑡 ∈ (0, 1) ,
𝑧 (𝑡) = { (16)
0, 𝑡 ∈ [−𝜏, 0] . 1
󵄨 󵄨
𝑡1−𝛼 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨 𝑠𝛼−1 𝑑𝑠
0
If 𝑢(⋅) satisfies the integral equation, (23)
Γ (𝛼) 𝐸𝛼,𝛼 (1) Γ(𝛼)2
≤ ( 󵄨󵄨 󵄨󵄨 + 1) 𝐸𝛼,𝛼 (1) Γ .
1 󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨 (2𝛼)
𝑢 (𝑡) = ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑓 (𝑠, 𝑢𝑠 ) 𝑑𝑠, (17)
0
Indeed we have
we can decompose 𝑢(⋅) as 𝑢(𝑡) = 𝑧(𝑡) + 𝑥(𝑡), 0 < 𝑡 < 1, which
implies 𝑢𝑡 = 𝑧𝑡 + 𝑥𝑡 for every 0 < 𝑡 < 1, and function 𝑧(⋅) 1
󵄨 󵄨
satisfies 𝑡1−𝛼 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨 𝑠𝛼−1 𝑑𝑠
0
󵄨 󵄨
1 Γ (𝛼) 󵄨󵄨󵄨𝐸𝛼,𝛼 (𝑡𝛼 )󵄨󵄨󵄨
𝑧 (𝑡) = ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑓 (𝑠, 𝑧𝑠 + 𝑥𝑠 ) 𝑑𝑠. (18) ≤ 󵄨󵄨 󵄨
0 󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨
∞ 1
1
Set 𝐶0 = {𝑧 ∈ (𝐶1−𝛼 [0, 1], R) : lim𝑡 → 0+ 𝑡1−𝛼 𝑧(𝑡) = 𝑐}. 𝐶0 ×∑ ∫ (1 − 𝑠)𝛼−1 𝑠𝛼−1 𝑑𝑠
is Banach space with the norm ‖ ⋅ ‖𝑟 . Let 𝑇 : 𝐶0 → 𝐶0 be 𝑖=0 Γ (𝛼𝑖 + 𝛼) 0
defined by ∞
1
+ 𝑡𝛼−1 ∑ (24)
1 𝑖=0 Γ (𝛼𝑖 + 𝛼)
𝑇𝑧 (𝑡) = ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑓 (𝑠, 𝑧𝑠 + 𝑥𝑠 ) 𝑑𝑠, 0 < 𝑡 < 1. (19) 󵄨 󵄨
0 𝑡 Γ (𝛼) 󵄨󵄨󵄨𝐸𝛼,𝛼 (𝑡𝛼 )󵄨󵄨󵄨
× ∫ (𝑡 − 𝑠)𝛼−1 𝑠𝛼−1 𝑑𝑠 󵄨󵄨 󵄨
0 󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨
Note that as operator 𝑁 has a fixed point, equivalently 𝑇 has
a fixed point and so instead we try to prove that 𝑇 has a fixed 1

point. × 𝐸𝛼,𝛼 (1) ∫ (1 − 𝑠)𝛼−1 𝑠𝛼−1 𝑑𝑠


0
𝑡
Theorem 7. Assume that there exists a constant 𝐾 > 0 such
that |𝑓(𝑡, 𝑢) − 𝑓(𝑡, V)| ≤ 𝐾|𝑢 − V| for each 𝑡 ∈ [0, 1] and all + 𝑡1−𝛼 𝐸𝛼,𝛼 (1) ∫ (𝑡 − 𝑠)𝛼−1 𝑠𝛼−1 𝑑𝑠.
0
𝑢, V ∈ R. Then the problem (2)-(3) has a unique solution in
𝐶0 ⊆ 𝐶1−𝛼 [0, 1] provided that Note that

Γ (𝛼) 𝐸𝛼,𝛼 (1) (Γ (𝛼))2 𝑡


Γ(𝛼)2
𝐾 ( 󵄨󵄨 󵄨 + 1) 𝐸 (1) < 1. (20) ∫ 𝑠𝛼−1 (𝑡 − 𝑠)𝛼−1 𝑑𝑠 = 𝑡2𝛼−1 . (25)
󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨
𝛼,𝛼
Γ (2𝛼) 0 Γ (2𝛼)
4 Abstract and Applied Analysis

Hence we have with


1
󵄨 󵄨 [𝐷𝛼−1 𝑦(𝑡)]𝑡=0 = 𝑏1 = 𝑐Γ (𝛼) ,
𝑡1−𝛼 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨 𝑠𝛼−1 𝑑𝑠 (32)
0
󵄨 󵄨
Γ (𝛼) 󵄨󵄨󵄨𝐸𝛼,𝛼 (𝑡𝛼 )󵄨󵄨󵄨 Γ(𝛼)2 where lim𝑡 → 0+ 𝑡1−𝛼 𝑦(𝑡) = 𝑐. Laplace transform of (31) yields
≤ 󵄨󵄨 󵄨󵄨 𝐸𝛼,𝛼 (1) Γ
󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨 (2𝛼)
(26) 𝑠𝛼 𝑌 (𝑠) − 𝑌 (𝑠) = 𝐻 (𝑠) + 𝑏1 (33)
2
1−𝛼 2𝛼−1 Γ(𝛼)
+𝑡 𝐸𝛼,𝛼 (1) 𝑡
Γ (2𝛼) from which
󵄨 󵄨
Γ (𝛼) 󵄨󵄨󵄨𝐸𝛼,𝛼 (𝑡𝛼 )󵄨󵄨󵄨 𝛼 (Γ (𝛼))2 𝐻 (𝑠) 𝑏1
= ( 󵄨󵄨 󵄨 + 𝑡 ) 𝐸 (1) . 𝑌 (𝑠) = + (34)
󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨 𝑠𝛼 − 1 𝑠𝛼 − 1
𝛼,𝛼
Γ (2𝛼)
Using (22) and (23) we get and the inverse Laplace transform gives the solution

󵄩󵄩 󵄩 Γ (𝛼) 𝐸𝛼,𝛼 (1) 𝑦 (𝑡) = 𝑏1 𝑡𝛼−1 𝐸𝛼,𝛼 (𝑡𝛼 )


󵄩󵄩𝑇𝑧1 − 𝑇𝑧2 󵄩󵄩󵄩1−𝛼 ≤ 𝐾 ( 󵄨󵄨 󵄨 + 1)
󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨 (35)
(27) 𝑡
2
Γ(𝛼) 󵄩󵄩 + ∫ (𝑡 − 𝑠)𝛼−1 𝐸𝛼,𝛼 (𝑡 − 𝑠)𝛼 ℎ (𝑠) 𝑑𝑠.
󵄩
× 𝐸𝛼,𝛼 (1) 󵄩𝑧 − 𝑧2 󵄩󵄩󵄩1−𝛼 . 0
Γ (2𝛼) 󵄩 1
Hence we have
This completes the proof.
𝑦 (𝑡) = 𝑐Γ (𝛼) 𝑡𝛼−1 𝐸𝛼,𝛼 (𝑡𝛼 )
4. Existence of Solution 𝑡 (36)
+ ∫ (𝑡 − 𝑠)𝛼−1 𝐸𝛼,𝛼 (𝑡 − 𝑠)𝛼 ℎ (𝑠) 𝑑𝑠.
In this section, by using Krasnoselskii’s theorem, we discuss 0
the existence solution of (4) under some assumptions on
𝑓 and further conditions. Before proving this theorem, we Therefore,
prove the following lemma which will be used in the next
theorem. 𝑦 (1) = 𝑐Γ (𝛼) + 𝐸𝛼,𝛼 (1)
1 (37)
Lemma 8. Consider the following nonlinear fractional differ- + ∫ (1 − 𝑠)𝛼−1 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) ℎ (𝑠) 𝑑𝑠,
ential equation of the form 0
𝛼
𝐷 𝑢 (𝑡) = 𝑢 (𝑡) + ℎ (𝑡) , 0 < 𝑡 < 1, 0 < 𝛼 < 1, (28) which leads to
with periodic boundary conditions 𝑐 (1 − Γ (𝛼) 𝐸𝛼,𝛼 (1))
1−𝛼
𝑢 (0) = lim+ 𝑡 𝑢 (𝑡) = 𝑐 = 𝑢 (1) , 1 (38)
𝑡→0
(29) = ∫ (1 − 𝑠)𝛼−1 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) ℎ (𝑠) 𝑑𝑠;
󵄨 0
𝐷−(1−𝛼) 𝑢(𝑡)󵄨󵄨󵄨󵄨𝑡=0 = 𝑐Γ (𝛼) ,
since Γ(𝛼)𝐸𝛼,𝛼 (1) ≠1 we have
where ℎ is a continuous function. Then the periodic boundary
value problem (28)-(29) is equivalent to an integral equation 1 1
1
given by 𝑢(𝑡) = ∫0 𝐺1,𝛼 (𝑡, 𝑠)ℎ(𝑠)𝑑𝑠 ∈ 𝐶1−𝛼 [0, 1], where 𝑐= ∫ (1 − 𝑠)𝛼−1 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) ℎ (𝑠) 𝑑𝑠.
1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) 0
(39)
{ Γ (𝛼) 𝐸𝛼,𝛼 (𝑡𝛼 ) 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) 𝑡𝛼−1 (1 − 𝑠)𝛼−1
{
{
{
{ 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)
{
{ Then the solution of the problem (28)-(29) is given by
{ 𝛼−1 𝛼
{ +(𝑡 − 𝑠) 𝐸𝛼,𝛼 ((𝑡 − 𝑠) ) ,
{
𝐺1,𝛼 (𝑡, 𝑠) = { 0 ≤ 𝑠 ≤ 𝑡 ≤ 1, Γ (𝛼)
{
{ 𝑦 (𝑡) = 𝑡𝛼−1 𝐸𝛼,𝛼 (𝑡𝛼 )
{
{ Γ (𝛼) 𝐸𝛼,𝛼 (𝑡𝛼 ) 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) 𝑡𝛼−1 (1 − 𝑠)𝛼−1 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)
{
{ ,
{
{ 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)
{ 1
{ 0 ≤ 𝑡 ≤ 𝑠 ≤ 1. × ∫ (1 − 𝑠)𝛼−1 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) ℎ (𝑠) 𝑑𝑠 (40)
(30) 0
𝑡
Proof. We consider the following fractional differential equa- + ∫ (𝑡 − 𝑠)𝛼−1 𝐸𝛼,𝛼 ((𝑡 − 𝑠)𝛼 ) ℎ (𝑠) 𝑑𝑠.
tion: 0

𝛼
𝐷 𝑦 (𝑡) − 𝑦 (𝑡) = ℎ (𝑡) , (0 < 𝑡 < 1, 0 < 𝛼 < 1) , (31) This completes the proof.
Abstract and Applied Analysis 5

Now we prove our main result using Lemma 8 and two ≤ (𝑀1 ‖𝑢‖1−𝛼 + 𝑀2 ‖𝑢‖21−𝛼 )
more assumptions which follow next. 󵄨 󵄨
Γ (𝛼) 󵄨󵄨󵄨𝐸𝛼,𝛼 (𝑡𝛼 )󵄨󵄨󵄨
(H1) We assume that 𝑓(𝑡, 𝑢, V) can be written as 𝑓1 (𝑡, 𝑢) + × ( 󵄨󵄨 󵄨 𝐸𝛼,𝛼+1 (1)
󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨
𝑓2 (𝑡, 𝑢, V), where 𝑓1 , 𝑓2 are Lipschitz continuous.
Moreover assume that the function 𝑓1 , 𝑓2 satisfies the
following relations: + 𝑡𝐸𝛼,𝛼+1 (𝑡𝛼 ) ) ≤ 𝑟.

󵄨󵄨 󵄨 (45)
󵄨󵄨𝑓1 (𝑡, 𝑢 (𝑡))󵄨󵄨󵄨 ≤ 𝑀1 ‖𝑢 (𝑡)‖1−𝛼 ,
󵄨󵄨 󵄨 (41) (ii) We will prove that 𝑇1 is a contraction:
󵄨󵄨𝑓2 (𝑡, 𝑢 (𝑡) , V (𝑡))󵄨󵄨󵄨 ≤ 𝑀2 ‖𝑢 (𝑡)‖1−𝛼 ‖V(𝑡)‖1−𝛼 .
󵄨󵄨 󵄨󵄨
󵄨󵄨 𝑇1 𝑢 − 𝑇1 V󵄨󵄨󵄨󵄨󵄨󵄨1−𝛼
󵄨 󵄨
(H2) Let Ω = 𝐶1−𝛼 ([−𝜏, 1], R) denote collection of the = 𝑡1−𝛼 󵄨󵄨󵄨𝑇1 𝑢 (𝑡) − 𝑇1 V (𝑡)󵄨󵄨󵄨 ,
space of all function 𝑢 such that 𝑡1−𝛼 𝑢(𝑡) ∈ 𝐶[0, 1].
1
Define the set 𝐵 = {𝑢 ∈ Ω : 𝑡1−𝛼 |𝑢| ≤ 𝑟}, where 𝑟 󵄨 󵄨
≤ 𝑡1−𝛼 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨
satisfies 0
󵄨 󵄨
(𝑀1 ‖𝑢‖1−𝛼 + 𝑀2 ‖𝑢‖21−𝛼 ) × 󵄨󵄨󵄨𝑓1 (𝑠, 𝑢 (𝑠)) − 𝑓1 (𝑠, V (𝑠))󵄨󵄨󵄨 𝑑𝑠,
1
󵄨󵄨 󵄨 󵄨 󵄨 󵄨 󵄨
󵄨Γ (𝛼) 󵄨󵄨󵄨𝐸𝛼,𝛼 (𝑡𝛼 ) 󵄨󵄨󵄨 ≤ 𝐿 𝑓1 𝑡1−𝛼 ∫ 󵄨󵄨󵄨 𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨 𝑢 − V󵄨󵄨󵄨 𝑑𝑠,
× ( 󵄨󵄨 󵄨 𝛼
󵄨 𝐸𝛼,𝛼+1 (1) + 𝑡𝐸𝛼,𝛼+1 (𝑡 )) ≤ 𝑟.
(46)
󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨
0
1
(42) 󵄨 󵄨 󵄨
≤ 𝐿 𝑓1 𝑡1−𝛼 ∫ 󵄨󵄨󵄨󵄨 𝐺1,𝛼 (𝑡, 𝑠) 𝑠𝛼−1 𝑠1−𝛼 󵄨󵄨󵄨󵄨 𝑢 − V󵄨󵄨󵄨󵄨 𝑑𝑠
0

Furthermore we assume that Γ (𝛼) 𝐸𝛼,𝛼 (1)


≤ 𝐿 𝑓1 ( 󵄨󵄨 󵄨 + 1)
󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨
Γ (𝛼) 𝐸𝛼,𝛼 (1) (Γ (𝛼))2
𝐿 𝑓1 ( 󵄨󵄨 󵄨 + 1) 𝐸 (1) < 1, (43)
󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨
𝛼,𝛼
Γ (2𝛼) (Γ (𝛼))2
× 𝐸𝛼,𝛼 (1) ‖𝑢 − V‖1−𝛼 .
Γ (2𝛼)
where 𝐿 𝑓1 is a Lipschitz constant of 𝑓.
Then 𝑇1 is a contraction.
Theorem 9. If the assumptions (H1) and (H2) satisfied, then (iii) Finally we prove that 𝑇2 is continuous and 𝑇2 (𝐵) is
the problem (4) with periodic boundary value condition (5) has contained in a compact set. To prove the continuity of 𝑇2 let
at least one solution. us consider a sequence 𝑢𝑛 converging to 𝑢. Taking the norm
of 𝑇2 𝑢𝑛 (𝑡) − 𝑇2 𝑢(𝑡) we have
Proof. (i) Note that by Lemma 8, (4)-(5) is equivalent to 󵄩󵄩 󵄩
󵄩󵄩𝑇2 𝑢𝑛 (𝑡) − 𝑇2 𝑢 (𝑡)󵄩󵄩󵄩1−𝛼
integral equation (17). Define 𝑇1 , 𝑇2 : Ω → Ω by
󵄨 󵄨
= 𝑡1−𝛼 󵄨󵄨󵄨𝑇2 𝑢𝑛 (𝑡) − 𝑇2 𝑢 (𝑡)󵄨󵄨󵄨
1
𝑇1 𝑢 (𝑡) = ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑓1 (𝑠, 𝑢 (𝑠)) 𝑑𝑠, 1
󵄨 󵄨
0
(44) ≤ 𝑡1−𝛼 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨
1 0
𝑇2 𝑢 (𝑡) = ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠. 󵄨
0
× 󵄨󵄨󵄨𝑓2 (𝑠, 𝑢𝑛 (𝑠) , 𝑢𝑛 (𝑠 − 𝜏))
󵄨
For 𝑢 ∈ 𝐵 we have, −𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏))󵄨󵄨󵄨 𝑑𝑠
1 (47)
󵄨 󵄨 󵄨 󵄨
𝑡1−𝛼 󵄨󵄨󵄨𝑇1 𝑢 (𝑡) + 𝑇2 𝑢 (𝑡)󵄨󵄨󵄨 ≤ 𝑡1−𝛼 𝐿 𝑓2 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨
0

󵄨
1
󵄨 󵄨 󵄨
≤ 𝑡1−𝛼 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨 × (󵄨󵄨󵄨𝑢𝑛 (𝑠) − 𝑢 (𝑠)󵄨󵄨󵄨
0
󵄨 󵄨
󵄨 + 󵄨󵄨󵄨𝑢𝑛 (𝑠 − 𝜏) − 𝑢 (𝑠 − 𝜏)󵄨󵄨󵄨) 𝑑𝑠
× 󵄨󵄨󵄨𝑓1 (𝑠, 𝑢 (𝑠))
1
󵄨 󵄨 󵄨
+𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏))󵄨󵄨󵄨 𝑑𝑠 ≤ 2𝐿 𝑓2 𝑡1−𝛼 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨 𝑠𝛼−1
0

≤ (𝑀1 ‖𝑢‖1−𝛼 + 𝑀2 ‖𝑢‖21−𝛼 ) 𝑡1−𝛼 󵄩 󵄩


× 𝑑𝑠󵄩󵄩󵄩𝑢𝑛 − 𝑢󵄩󵄩󵄩1−𝛼 .
1 Hence whenever 𝑢𝑛 → 𝑢 we have 𝑇2 𝑢𝑛 → 𝑇2 𝑢. This proves
󵄨 󵄨
× ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨 𝑑𝑠 the continuity of 𝑇2 .
0
6 Abstract and Applied Analysis

1
On the other hand for 0 ≤ 𝑡1 < 𝑡2 ≤ 1 and 𝑢 ∈ 𝐵 we have × ∫ 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) (1 − 𝑠)𝛼−1
𝑡2
× 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠
󵄨󵄨 1−𝛼 󵄨 Γ (𝛼) 𝐸𝛼,𝛼 (𝑡2𝛼 )
󵄨󵄨𝑡1 𝑇2 𝑢 (𝑡1 ) − 𝑡21−𝛼 𝑇2 𝑢 (𝑡2 )󵄨󵄨󵄨
󵄨 󵄨 −
󵄨󵄨 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)
1
󵄨
= 󵄨󵄨󵄨󵄨𝑡11−𝛼 ∫ 𝐺1,𝛼 (𝑡1 , 𝑠) 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠 1
󵄨󵄨 0 × ∫ 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼−1 ) (1 − 𝑠)𝛼−1
𝑡2
1 󵄨󵄨 󵄨󵄨
󵄨 󵄨
− 𝑡21−𝛼 ∫ 𝐺1,𝛼 (𝑡2 , 𝑠) 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠󵄨󵄨󵄨󵄨 ×𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠󵄨󵄨󵄨󵄨 .
0 󵄨󵄨 󵄨󵄨
󵄨󵄨 Γ (𝛼) 𝐸 (𝑡𝛼 ) (48)
󵄨
≤ 󵄨󵄨󵄨󵄨
𝛼,𝛼 1
󵄨󵄨 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) Then we have
󵄨󵄨 1−𝛼 󵄨
𝑡1 󵄨󵄨𝑡1 𝑇2 𝑢 (𝑡1 ) − 𝑡21−𝛼 𝑇2 𝑢 (𝑡2 )󵄨󵄨󵄨
󵄨 󵄨
× ∫ 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) (1 − 𝑠)𝛼−1
0 ≤ 𝑀2 ‖𝑢‖21−𝛼
× 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠
Γ (𝛼) 󵄨󵄨 𝛼 󵄨
󵄨𝐸 (𝑡 ) − 𝐸𝛼,𝛼 (𝑡2 )󵄨󵄨󵄨
𝛼
×(
𝑡1
𝛼 𝛼−1 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) 󵄨 𝛼,𝛼 1
+ 𝑡11−𝛼 ∫ 𝐸𝛼,𝛼 ((𝑡1 − 𝑠) ) (𝑡1 − 𝑠)
0 𝑡
1
󵄨 󵄨
× 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠 × ∫ 󵄨󵄨󵄨𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 )󵄨󵄨󵄨 (1 − 𝑠)𝛼−1 𝑑𝑠
0

Γ (𝛼) 𝐸𝛼,𝛼 (𝑡2𝛼 ) 𝑡1


󵄨
+ ∫ 󵄨󵄨󵄨󵄨𝑡11−𝛼 𝐸𝛼,𝛼 ((𝑡1 − 𝑠) ) (𝑡1 − 𝑠)
𝛼 𝛼−1

1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) 0
𝛼 󵄨
𝐸𝛼,𝛼 ((𝑡2 − 𝑠) )󵄨󵄨󵄨󵄨 𝑑𝑠
𝛼−1
𝑡1 −𝑡21−𝛼 (𝑡2 − 𝑠)
× ∫ 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) (1 − 𝑠)𝛼−1
0
Γ (𝛼) 󵄨 𝛼 󵄨 (49)
󵄨 󵄨󵄨𝐸𝛼,𝛼 (𝑡1 ) − 𝐸𝛼,𝛼 (𝑡2 )󵄨󵄨󵄨
𝛼
+ 󵄨󵄨
× 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨 󵄨
𝑡1 𝑡
− 𝑡21−𝛼 ∫ 𝐸𝛼,𝛼 ((𝑡2 − 𝑠) ) (𝑡2 − 𝑠)
𝛼 𝛼−1 2
󵄨 󵄨
× ∫ 󵄨󵄨󵄨𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 )󵄨󵄨󵄨 (1 − 𝑠)𝛼−1 𝑑𝑠
0 𝑡 1

× 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑡2


󵄨 𝛼 󵄨
− 𝑡21−𝛼 ∫ 󵄨󵄨󵄨󵄨𝐸𝛼,𝛼 ((𝑡2 − 𝑠) )󵄨󵄨󵄨󵄨 (𝑡2 − 𝑠) 𝑑𝑠
𝛼−1

Γ (𝛼) 𝐸𝛼,𝛼 (𝑡1𝛼 ) 𝑡 1


+
1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) Γ (𝛼)
+ 󵄨󵄨 󵄨
𝑡2 󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨
× ∫ 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) (1 − 𝑠)𝛼−1 󵄨 󵄨 󵄨
𝑡1 × 󵄨󵄨󵄨𝐸𝛼,𝛼 (𝑡1𝛼 ) − 󵄨󵄨󵄨𝐸𝛼,𝛼 (𝑡2𝛼 ) 󵄨󵄨󵄨
1
󵄨 󵄨
× 𝑓 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠 × ∫ 󵄨󵄨󵄨𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 )󵄨󵄨󵄨 (1 − 𝑠)𝛼−1 𝑑𝑠) .
𝑡 2
Γ (𝛼) 𝐸𝛼,𝛼 (𝑡2𝛼 )
− Hence we deduce that if |𝑡1 − 𝑡2 | → 0 then |𝑡11−𝛼 𝑇2 (𝑡2 ) −
1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) 𝑡21−𝛼 𝑇2 (𝑡2 )| → 0.
𝑡2 Then {𝑡1−𝛼 𝑇2 𝑢 : 𝑢 ∈ 𝐵} is equicontinuous. Moreover we
× ∫ 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) (1 − 𝑠)𝛼−1 show that {𝑡1−𝛼 𝑇2 𝑢 : 𝑢 ∈ 𝐵} is a bounded set in 𝐶[0, 1]. Indeed
𝑡1
we have
× 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠 󵄨 󵄨
𝑡1−𝛼 󵄨󵄨󵄨𝑇2 𝑢 (𝑡)󵄨󵄨󵄨
𝑡2 1
󵄨 󵄨
− 𝑡21−𝛼 ∫ 𝐸𝛼,𝛼 (𝑡2 − 𝑠) (𝑡2 − 𝑠)
𝛼 𝛼−1 ≤ 𝑀2 ‖𝑢‖21−𝛼 𝑡1−𝛼 ∫ 󵄨󵄨󵄨𝐺1,𝛼 (𝑡, 𝑠)󵄨󵄨󵄨 𝑑𝑠
𝑡1 0
≤ 𝑀2 ‖𝑢‖21−𝛼
× 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠 󵄨 󵄨
Γ (𝛼) 󵄨󵄨󵄨𝐸𝛼,𝛼 (𝑡𝛼 )󵄨󵄨󵄨 𝛼
× ( 󵄨󵄨 󵄨 𝐸𝛼,𝛼+1 (1) + 𝑡𝐸𝛼,𝛼+1 (𝑡 )) .
Γ (𝛼) 𝐸𝛼,𝛼 (𝑡1𝛼 ) 󵄨󵄨1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)󵄨󵄨󵄨
+
1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) (50)
Abstract and Applied Analysis 7

Then by Arzela-Ascoli’s theorem we conclude that 𝑇2 is [13] A. El-Sayed, “Fractional order differential equation,” Kyungpook
compact. By using Krasnoselskii’s theorem there exists 𝑧 ∈ 𝐵 Mathematical Journal, vol. 28, no. 2, pp. 18–22, 1988.
such that 𝑧 is a fixed point of 𝑇. This completes the proof. [14] A. El-Sayed, “On the fractional differential equations,” Applied
Mathematics and Computation, vol. 49, no. 2-3, pp. 205–213,
1992.
5. Conclusions
[15] A. El-Sayed, “Nonlinear functional-differential equations of
We considered two types of nonlinear delay fractional dif- arbitrary orders,” Nonlinear Analysis. Theory, Methods & Appli-
ferential equations (FDE) with periodic boundary conditions cations. An International Multidisciplinary Journal. Series A:
involving Remann-Liouville fractional derivative possessing Theory and Methods, vol. 33, no. 2, pp. 181–186, 1998.
with a lower terminal at 0. In order to obtain the results in [16] R. P. Agarwal, Y. Zhou, and Y. He, “Existence of fractional
this paper, we have shown the existence and the uniqueness of neutral functional differential equations,” Computers & Math-
solution for a class of nonlinear delayed FDE by Banach con- ematics with Applications, vol. 59, no. 3, pp. 1095–1100, 2010.
traction principle. Then using Krasnoselskii’s fixed point the- [17] V. Daftardar-Gejji and A. Babakhani, “Analysis of a system
orem we established an existence theorem for a different type of fractional differential equations,” Journal of Mathematical
Analysis and Applications, vol. 293, no. 2, pp. 511–522, 2004.
of the equation that we have proven its uniqueness theorem.
[18] Y. Liu, “Existence of solutions for impulsive differential models
on half lines involving Caputo fractional derivatives,” Commu-
Acknowledgment nications in Nonlinear Science and Numerical Simulation, vol. 18,
no. 10, pp. 2604–2625, 2013.
The authors would like thank to the anonymous referees for
[19] D. Delbosco and L. Rodino, “Existence and uniqueness for a
their valuable comments and suggestions. nonlinear fractional differential equation,” Journal of Mathe-
matical Analysis and Applications, vol. 204, no. 2, pp. 609–625,
References 1996.
[20] J. Deng and L. Ma, “Existence and uniqueness of solutions
[1] W. Boyce and R. Diprima, Elementary Differential Equations of initial value problems for nonlinear fractional differential
and Boundary Value Problems, John Wiley & Sons, New York, equations,” Applied Mathematics Letters, vol. 23, no. 6, pp. 676–
NY, USA, 1997. 680, 2010.
[2] F. Moon and M. Johnson, Dynamics and Chaos in Manufactur- [21] K. Diethelm and N. J. Ford, “Analysis of fractional differential
ing Processes, chapter 1, John Wiley & Sons, New York, NY, USA, equations,” Journal of Mathematical Analysis and Applications,
1998. vol. 265, no. 2, pp. 229–248, 2002.
[3] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory and [22] A. El-Sayed, A. El-Mesiry, and H. El-Saka, “On the fractional-
Application of Fractional Differential Equations, Elsevier Science order logistic equation,” Applied Mathematics Letters, vol. 20,
B.V., Amsterdam, The Netherlands, 2006. no. 7, pp. 817–823, 2007.
[4] S. Das, Functional Fractional Calculus for System Identification [23] K. M. Furati and N.-E. Tatar, “Long time behavior for a
and Controls, Springer, Berlin, Germany, 2008. nonlinear fractional model,” Journal of Mathematical Analysis
[5] R. Hilfer, Ed., Applications of Fractional Calculus in Physics, and Applications, vol. 332, no. 1, pp. 441–454, 2007.
World Scientific, River Edge, NJ, USA, 2000. [24] S. B. Hadid, “Local and global existence theorems on differential
[6] N. Heymans and I. Podlubny, “Physical interpretation of initial equations of non-integer order,” Journal of Fractional Calculus,
conditions for fractional differential equations with Riemann- vol. 7, pp. 101–105, 1995.
Liouville fractional derivatives,” Rheologica Acta, vol. 45, no. 5, [25] S. Hadid, B. Masaedeh, and S. Momani, “On the existence of
pp. 765–771, 2006. maximal and minimal solutions of differential equations of
[7] I. Podlubny, Fractional Differential Equations, Academic Press, non-integer order,” Journal of Fractional Calculus, vol. 9, pp. 41–
London, UK, 1999. 44, 1996.
[8] I. Podlubny, “Geometric and physical interpretation of frac- [26] R. W. Ibrahim and S. Momani, “On the existence and unique-
tional integration and fractional differentiation,” Fractional ness of solutions of a class of fractional differential equations,”
Calculus & Applied Analysis, vol. 5, no. 4, pp. 367–386, 2002, Journal of Mathematical Analysis and Applications, vol. 334, no.
Dedicated to the 60th anniversary of Prof. Francesco Mainardi. 1, pp. 1–10, 2007.
[9] I. Podlubny, I. Petráš, B. M. Vinagre, P. O’Leary, and L. Dorčák, [27] V. Lakshmikantham and A. S. Vatsala, “Basic theory of frac-
“Analogue realizations of fractional-order controllers. Frac- tional differential equations,” Nonlinear Analysis. Theory, Meth-
tional order calculus and its applications,” Nonlinear Dynamics, ods & Applications. An International Multidisciplinary Journal.
vol. 29, no. 1–4, pp. 281–296, 2002. Series A: Theory and Methods, vol. 69, no. 8, pp. 2677–2682,
[10] J. Sabatier, O. P. Agrawal, and J. A. T. Machado, Advances in 2008.
Fractional Calculus: Theoretical Developments and Applications [28] M. Benchohra, J. Henderson, S. K. Ntouyas, and A. Ouahab,
in Physics and Engineering, Springer, New York, NY, USA, 2007. “Existence results for fractional order functional differential
[11] J. T. Machado, V. Kiryakova, and F. Mainardi, “Recent history of equations with infinite delay,” Journal of Mathematical Analysis
fractional calculus,” Communications in Nonlinear Science and and Applications, vol. 338, no. 2, pp. 1340–1350, 2008.
Numerical Simulation, vol. 16, no. 3, pp. 1140–1153, 2011. [29] M. Belmekki, J. J. Nieto, and R. Rodrı́guez-López, “Existence
[12] J. T. Machado, A. M. Galhano, and J. J. Trujillo, “On devel- of periodic solution for a nonlinear fractional differential equa-
opment of fractional calculus during the last fifty years,” in tion,” Boundary Value Problems, vol. 2009, Article ID 324561, 18
Scientometrics, Springer, New York, NY, USA, 2013. pages, 2009.
8 Abstract and Applied Analysis

[30] S. Abbas, “Existence of solutions to fractional order ordinary


and delay differential equations and applications,” Electronic
Journal of Differential Equations, vol. 2011, no. 09, pp. 1–11, 2011.
[31] V. Lakshmikantham and A. S. Vatsala, “Theory of fractional
differential inequalities and applications,” Communications in
Applied Analysis, vol. 11, no. 3-4, pp. 395–402, 2007.
[32] C. Yu and G. Gao, “Existence of fractional differential equa-
tions,” Journal of Mathematical Analysis and Applications, vol.
310, no. 1, pp. 26–29, 2005.
[33] S. Abbas, “Pseudo almost automorphic solutions of fractional
order neutral differential equation,” Semigroup Forum, vol. 81,
no. 3, pp. 393–404, 2010.
[34] R. P. Agarwal, M. Benchohra, and S. Hamani, “Boundary
value problems for fractional differential equations,” Georgian
Mathematical Journal, vol. 16, no. 3, pp. 401–411, 2009.
[35] J. Banas and T. Zajcac, “A new approach to the theory of
functional integral equations of fractional order,” Journal of
Mathematical Analysis and Applications, vol. 375, no. 2, pp. 375–
387, 2011.
[36] C. Giannantoni, “The problem of the initial conditions and their
physical meaning in linear differential equations of fractional
order,” Applied Mathematics and Computation, vol. 141, no. 1, pp.
87–102, 2003.
[37] N. Heymans and I. Podlubny, “Physical interpretation of initial-
conditions for fractional differential equations with Riemann-
Liouville fractional derivatives,” Rheologica Acta, vol. 45, no. 5,
pp. 765–771, 2006.
[38] E. Kreyszig, Introductory Functional Analysis with Applications,
John Wiley & Sons, New York, NY, USA, 1978.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 610314, 19 pages
http://dx.doi.org/10.1155/2013/610314

Research Article
Application of Fuzzy Fractional Kinetic Equations to
Modelling of the Acid Hydrolysis Reaction

Ferial Ghaemi,1 Robiah Yunus,1 Ali Ahmadian,1,2,3 Soheil Salahshour,4


Mohamed Suleiman,2 and Shanti Faridah Saleh5
1
Institute of Advanced Technology, Universiti Putra Malaysia (UPM), 43400 Serdang, Selangor, Malaysia
2
Institute for Mathematical Research, Universiti Putra Malaysia (UPM), 43400 Serdang, Selangor, Malaysia
3
Department of Mathematics, Universiti Putra Malaysia (UPM), 43400 Serdang, Selangor, Malaysia
4
Young Researchers and Elite Club, Mobarakeh Branch, Islamic Azad University, P.O. Box 9189945, Mobarakeh, Iran
5
Department of Chemical and Environmental Engineering, Faculty of Engineering, Universiti Putra Malaysia (UPM),
43400 Serdang, Selangor, Malaysia

Correspondence should be addressed to Ali Ahmadian; ahmadian.hosseini@gmail.com

Received 20 May 2013; Revised 24 June 2013; Accepted 27 June 2013

Academic Editor: Ali H. Bhrawy

Copyright © 2013 Ferial Ghaemi et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

In view of the usefulness and a great importance of the kinetic equation in specific chemical engineering problems, we discuss
the numerical solution of a simple fuzzy fractional kinetic equation applied for the hemicelluloses hydrolysis reaction. The fuzzy
approximate solution is derived based on the Legendre polynomials to the fuzzy fractional equation calculus. Moreover, the
complete error analysis is explained based on the application of fuzzy Caputo fractional derivative. The main advantage of the
present method is its superior accuracy which is obtained by using a limited number of Legendre polynomials. The method is
computationally interesting, and the numerical results demonstrate the effectiveness and validity of the method for solving fuzzy
fractional differential equations.

1. Introduction the hemicelluloses hydrolysis reaction catalysed by sulphuric


acid. We apply a family of orthogonal functions to derive a
A rigorous kinetic study of the acid hydrolysis of ligno- numerical method for solving this type of kinetic equation of
cellulosic materials is a complicated issue due to several the fuzzy Caputo fractional order.
factors: (1) the structure of the whole cells should be protected
against the attacks of chemicals, (2) lignin hydrophobicity 1.1. Acid Hydrolyzing Reaction. First of all, in 1945, Faith [5]
creates an arduous access of protons to the raw material, (3) investigated the dilute acid hydrolysis process by using wood
interaction with other components, (4) presence of strong material in a 0.5 sulphuric acid solution. Neureiter et al.
bonds in the raw material, and (5) variable exposition of [6] found that the temperature has an important role in the
hemicelluloses surface to the chemical attack along the formation of sugar degradation products. For hydrolyzing
reaction [1]. Various researchers adapted Saeman’s kinetics to the hemicellulose and cellulose, the temperature ranges are
describe the hydrolysis of hemicellulose and the formation of from 100∘ C to 140∘ C. The acidity (concentration of acid) of
byproducts at high temperatures [2, 3]. These models begin the system is another parameter that affects the hydrolysis
with the work of Saeman for the hydrolysis of wood using process. For the hydrolyzing of acid, both penetration into
sulfuric acid [4]. the biomass and dispersion in the reactor can influence not
In the present work, our focus is to find the approximate only the general reaction but also the reactor performance.
solution of the fuzzy fractional model of the kinetics of The sulphuric acid diffusivity depends on the nature of
2 Abstract and Applied Analysis

lignocellulosic substances. It is revealed that the diffusivity of matrix for fractional derivatives was introduced by [53] and
sulfuric acid in agricultural residues is extraordinary higher applied with spectral methods for numerical solution of
than in hard wood [7]. It is with this motivation to investigate multiterm linear and nonlinear fractional differential equa-
the effect of sulfuric acid concentration on both of hydrolysis tions subject to initial conditions which were developed by
of oil palm empty fruit bunch (OPEFB) fiber hydrolyzing and Kazem et al. [52] who applied fractional Legendre orthogonal
analyzing the appropriate kinetic model. The OPEFB biomass functions for solving these types of equations. Subsequently,
includes cellulose, hemicellulose, and lignin. In this study, to Doha et al. [55] introduced a new efficient Chebyshev spec-
obtain an in depth understanding to the hydrolysis reaction tral algorithms for solving linear and nonlinear multiterm
and an optimum reaction condition for the process, the fractional orders differential equations. Thereafter, Bhrawy et
reaction kinetics at a complete set of reaction conditions were al. [60] propose a method to approximate multiterm frac-
investigated. The respective rate constants were determined tional differential equations with variable coefficients using
as functions of temperature and sulfuric acid concentration a quadrature shifted Legendre tau approach. Consequently,
by a nonlinear regression analysis. this way has been followed by several authors [54, 57, 59, 61].
On the other hand, the study of fuzzy differential equa-
1.2. Fractional Kinetic Equation Model and Numerical Meth- tions is rapidly developing as a new field of fuzzy mathe-
ods. It is now well established that fractional kinetic equa- matics. The fuzzy differential equations have been studied by
tions represent an appropriate model to describe physical several authors [62–70]. Lately, Agarwal et al. [71] proposed
phenomena such as diffusion in porous media with fractal the concept of solutions for fractional differential equations
geometry, kinematics in viscoelastic media, relaxation pro- with uncertainty. They have considered Riemann-Liouville’s
cesses in complex systems (including viscoelastic materials, differentiability with a fuzzy initial condition to solve FFDEs.
glassy materials, synthetic polymers, and biopolymers), prop- Afterward, several authors have studied the existence and
agation of seismic waves, and the rate of change of chemical uniqueness of the solution of the fuzzy fractional differential
composition of a star [8–18]. equations (FFDEs) under different types of fuzzy fractional
These equations are obtained from the classical kinetic differentiability [72–75]; nevertheless, it is a little bit surpris-
equation by replacing the first- or second-order derivative by ing that few papers reported numerical methods for solving
fractional derivative [19–30]. In the nonstochastic situation, FFDEs [75–78].
fractional kinetic equations have been studied by Kochubeı̆ The scope of this paper is to derive an explicit formula
[31], Saichev and Zaslavsky [17], Zaslavsky [18], Haubold and for fuzzy fractional-order derivative of shifted fractional Leg-
Mathai [32], and Saxena et al. [33, 34].
endre polynomials of any degree in terms of shifted fractional
In the few years, the analytical and numerical meth-
Legendre polynomials themselves, in the fuzzy Caputo sense
ods for solving fractional differential equations (FDEs)
with order (0 < V ≤ 1). Also, we are concerned with the direct
have attracted much more consideration of mathematicians.
solution technique for solving the fuzzy fractional kinetic
Although some studies have been exploited to solve FDEs
equation (FFKE) which is extracted by using the kinetics data
analytically [35–38], most of them do not have an exact
of the acid hydrolysis reaction subject to nonhomogeneous
analytical solution. Therefore, numerical and approximation
initial conditions.
techniques have been used for solving these equations. Some
In this paper, the fuzzy fractional derivative of the
of the most common methods are homotopy perturbation
proposed FFKE is approximated based on the shifted frac-
method [39], Haar wavelet method [40], Spline collocation
tional Legendre polynomials presented in [52], and then the
method [41], fractional difference method (FDM) [42], power
operational matrix of the fractional Caputo derivative of
series method [43], Adomian decomposition method (ADM)
[44], Kronecker convolution product [45], spectral methods order (0 < V ≤ 1) is specified, and we apply the shifted
[46], He’s variational iteration method [47], and homotopy fractional Legendre spectral tau (SCT) method to construct
analysis method [48]. the spectral solution for such problem. To the best of the
It is somewhat extraordinary that recently the orthogonal authors knowledge, such approach has not been employed for
functions received remarkable utilizing for the fractional- solving linear fractional kinetic equations under uncertainty.
order differential equations [49, 50]. Much efforts have been The paper is organized as follows. In Section 2, we
made to develop accurate algorithms using tau and collo- introduce some necessary definitions of fuzzy sets, fractional
cation based on operational matrices of some orthogonal calculus theory, kinetic equations, relevant properties of Leg-
polynomials such as block pulse functions [51], Legendre endre polynomials, and some of the main properties of fuzzy
polynomials [52, 53], Chebyshev polynomials [54–56], Jacobi fractional derivatives. Section 3 is devoted to presentation
polynomials [57, 58], and Laguerre polynomials [59]. The of the governing fractional kinetic equation. In Section 4,
main characteristic behind the approach using this technique the proposed method is explained for numerical solution
is that it reduces FDEs to those of solving a system of algebraic of the FFDEs. In Section 5, the derived FFKE is solved
equations thus notably simplifying the problem. based on the different values of constant coefficients in the
Furthermore, spectral methods have been found very equation by applying the presented technique, and the error
robust tools for solving many types of fractional differential of the approximate solution is depicted to demonstrate the
equations which have inspired many authors to apply them effectiveness of the method. Finally, some conclusions are
for these kinds of equations. The shifted Legendre operational drawn.
Abstract and Applied Analysis 3

2. Preliminaries It is easy to see that 𝐷 is a metric in RF and has the


following properties (see [80, 83]):
In this section, some basic definitions and notations related
to the fuzzy sets and functions, Legendre polynomials, and (i) 𝐷(𝑢 ⊕ 𝑤, V ⊕ 𝑤) = 𝐷(𝑢, V), for all 𝑢, V, 𝑤 ∈ RF ,
fractional kinetic equations are presented which will be used (ii) 𝐷(𝑘 ⊙ 𝑢, 𝑘 ⊙ V) = |𝑘|𝐷(𝑢, V), for all 𝑘 ∈ R, 𝑢, V ∈ RF ,
throughout the paper.
(iii) 𝐷(𝑢⊕V, 𝑤⊕𝑒) ≤ 𝐷(𝑢, 𝑤)+𝐷(V, 𝑒), for all 𝑢, V, 𝑤 ∈ RF ,
2.1. Basic Definitions of Fuzzy Sets. The basic definitions (iv) 𝐷(𝑢 + V, 0) ≤ 𝐷(𝑢, 0) + 𝐷(V, 0), for all 𝑢, V ∈ RF ,
presented in this section are given in [79–82]. (v) (RF , 𝐷) is a complete metric space.

Definition 1. Let 𝑢 be a fuzzy set in R. 𝑢 is called a fuzzy Definition 4 (see [84]). Let 𝑓 and 𝑔 be the two fuzzy-number-
number if valued functions on the interval [𝑎, 𝑏], that is, 𝑓, 𝑔 : [𝑎, 𝑏] →
RF . The uniform distance between fuzzy-number-valued
(i) 𝑢 is normal: there exists 𝑥0 ∈ R such that 𝑢(𝑥0 ) = 1, functions is defined by
(ii) 𝑢 is convex: for all 𝑥, 𝑦 ∈ R and 0 ≤ 𝜆 ≤ 1, it holds
that 𝐷∗ (𝑓, 𝑔) := sup 𝐷 (𝑓 (𝑥) , 𝑔 (𝑥)) . (9)
𝑥∈[𝑎,𝑏]
𝑢 (𝜆𝑥 + (1 − 𝜆) 𝑦) ≥ min {𝑢 (𝑥) , 𝑢 (𝑦)} , (1)
Remark 5 (see [83]). Let 𝑓 : [𝑎, 𝑏] → RF be fuzzy contin-
(iii) 𝑢 is upper semicontinuous: for any 𝑥0 ∈ R, it holds uous. Then, from property (iv) of Hausdorff distance, we can
that define
󵄨 󵄨 󵄨 󵄨
𝑢 (𝑥0 ) ≥ lim± 𝑢 (𝑥) , (2) 𝐷 (𝑓 (𝑥) , 0̃) = sup max {󵄨󵄨󵄨𝑓1𝑟 (𝑥)󵄨󵄨󵄨 , 󵄨󵄨󵄨𝑓2𝑟 (𝑥)󵄨󵄨󵄨} , ∀𝑥 ∈ [𝑎, 𝑏] .
𝑥 → 𝑥0 𝑟∈[0,1]
(10)
(iv) [𝑢]0 = supp(𝑢) is a compact subset of R.
Definition 6 (see [85]). Let 𝑥, 𝑦 ∈ RF . If there exists 𝑧 ∈ RF
In this paper, the set of all fuzzy numbers is denoted by such that 𝑥 = 𝑦 ⊕ 𝑧, then 𝑧 is called the H-difference of 𝑥 and
RF . 𝑦, and it is denoted by 𝑥 ⊖ 𝑦.
Definition 2. Let 𝑢 ∈ RF and 𝑟 ∈ [0, 1]. The 𝑟-cut of 𝑢 is In this paper, the sign “⊖” always stands for H-difference,
the crisp set [𝑢]𝑟 that contains all elements with membership and note that 𝑥 ⊕ 𝑦 ≠𝑥 + (−𝑦). Also, throughout the
degree in 𝑢 greater than or equal to 𝑟, that is, paper, the Hukuhara-diference and generalized Hukuhara-
differentiability are assumed to be existed.
[𝑢]𝑟 = {𝑥 ∈ R | 𝑢 (𝑥) ≥ 𝑟} . (3)
For a fuzzy number 𝑢, its 𝑟-cuts are closed intervals in R, and Theorem 7 (see [69]). Let 𝐹 : (𝑎, 𝑏) → RF be a function,
we denote them by and denote [𝐹(𝑡)]𝑟 = [𝑓𝑟 (𝑡), 𝑔𝑟 (𝑡)], for each 𝑟 ∈ [0, 1]. Then,

[𝑢]𝑟 = [𝑢1𝑟 , 𝑢2𝑟 ] . (4) (1) if 𝐹 is (1)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are
differentiable functions and
According to Zadeh’s extension principle, the operation 𝑟
of addition on RF is defined as follows: [𝐹󸀠 (𝑡)] = [𝑓𝑟󸀠 (𝑡) , 𝑔𝑟󸀠 (𝑡)] , (11)
(𝑢 + V) (𝑥) = sup min {𝑢 (𝑦) , V (𝑥 − 𝑦)} , 𝑥 ∈ R,
𝑦∈R
(5) (2) if 𝐹 is (2)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are
differentiable functions and
and scalar multiplication of a fuzzy number is given by
𝑟
𝑥 [𝐹󸀠 (𝑡)] = [𝑔𝑟󸀠 (𝑡) , 𝑓𝑟󸀠 (𝑡)] . (12)
{𝑢 ( ) , 𝑘 > 0,
(𝑘 ⊙ 𝑢) (𝑥) = { 𝑘 (6)
̃ Definition 8 (see [86]). Consider the 𝑛 × 𝑛 linear system of
{0, 𝑘 = 0,
equations
where 0̃ ∈ RF .
𝑎11 𝑥1 + 𝑎12 𝑥2 + ⋅ ⋅ ⋅ + 𝑎1𝑛 𝑥𝑛 = 𝑦1 ,
Definition 3 (see [80]). The distance 𝐷(𝑢, V) between two 𝑎21 𝑥1 + 𝑎22 𝑥2 + ⋅ ⋅ ⋅ + 𝑎2𝑛 𝑥𝑛 = 𝑦2 ,
fuzzy numbers 𝑢 and V is defined as
(13)
..
𝐷 (𝑢, V) = sup 𝑑H ([𝑢]𝑟 , [V]𝑟 ) , (7) .
𝑟∈[0,1]
𝑎𝑛1 𝑥1 + 𝑎𝑛2 𝑥2 + ⋅ ⋅ ⋅ + 𝑎𝑛𝑛 𝑥𝑛 = 𝑦𝑛 .
where
󵄨 󵄨 󵄨 󵄨
𝑑H ([𝑢]𝑟 , [V]𝑟 ) = max {󵄨󵄨󵄨𝑢1𝑟 − V1𝑟 󵄨󵄨󵄨 , 󵄨󵄨󵄨𝑢2𝑟 − V2𝑟 󵄨󵄨󵄨} (8) The matrix form of the earlier equations is

is the Hausdorff distance between [𝑢]𝑟 and [V]𝑟 . 𝐴𝑋 = 𝑌, (14)


4 Abstract and Applied Analysis

where the coefficient matrix 𝐴 = (𝑎𝑖𝑗 ), 1 ≤ 𝑖, 𝑗 ≤ 𝑛 is a crisp 2.2. Fractional Legendre Polynomials. The shifted Legendre
𝑛 × 𝑛 matrix and 𝑦𝑖 ∈ RF , 1 ≤ 𝑖 ≤ 𝑛. This system is called a polynomials are generated from the three-term recurrence
fuzzy linear system (FLS). relation:

Definition 9 (see [86]). A fuzzy number vector (𝑥1 , 𝑥2 , . . . , (2𝑖 + 1) (2𝑥 − 1) 𝑖


𝐿 𝑖+1 (𝑥) = 𝐿 𝑖 (𝑥) − 𝐿 (𝑥) ,
𝑥𝑛 )𝑡 given by 𝑥𝑖 = (𝑥𝑖 𝑟− , 𝑥𝑖 𝑟+ ), 1 ≤ 𝑖 ≤ 𝑛, 0 ≤ 𝑟 ≤ 1 is called a 𝑖+1 𝑖 + 1 𝑖−1
solution of the fuzzy linear system (13) if 𝑖 = 1, 2, . . . , (20)
𝑟
𝑛 𝑛
𝑟 𝐿 0 (𝑥) = 1, 𝐿 1 (𝑥) = 2𝑥 − 1.
( ∑ 𝑎𝑖𝑗 𝑥𝑗 ) = ∑(𝑎𝑖𝑗 𝑥𝑗 )− = 𝑦𝑖 𝑟− ,
𝑗=1 − 𝑗=1 The analytic form of the shifted Legendre polynomial 𝐿 𝑛 (𝑥)
𝑟
(15) of degree 𝑛 is given by
𝑛 𝑛
𝑟
( ∑ 𝑎𝑖𝑗 𝑥𝑗 ) = ∑(𝑎𝑖𝑗 𝑥𝑗 )+ = 𝑦𝑖 𝑟+ . 𝑛
(𝑛 + 𝑖)! 𝑥𝑖 𝑛
𝐿 𝑛 (𝑥) = ∑(−1)𝑛+𝑖 = ∑ 𝑒 𝑥𝑖 , (21)
𝑗=1 + 𝑗=1
𝑖=0 (𝑛 − 𝑖)! (𝑖!)2 𝑖=0 𝑖,𝑛
If for a particular 𝑘, 𝑎𝑘𝑗 > 0, 1 ≤ 𝑗 ≤ 𝑛, we simply get in which
𝑛 𝑛 (𝑛 + 𝑖)!
∑ 𝑎𝑘𝑗 𝑥𝑗 𝑟− = 𝑦𝑘 𝑟− , ∑ 𝑎𝑘𝑗 𝑥𝑗 𝑟+ = 𝑦𝑘 𝑟+ . 𝑒𝑖,𝑛 = (−1)𝑛+𝑖 , (22)
(16) (𝑛 − 𝑖)!(𝑖!)2
𝑗=1 𝑗=1
where
To solve fuzzy linear systems, see [87].
𝐿 𝑛 (0) = (−1)𝑛 , 𝐿 𝑛 (1) = 1. (23)
In this part, a brief approach to the fractional calculus
based on the crisp concept is provided. For more details, see The fractional-order Legendre functions (FLFs) can be
[19, 25, 30]. defined by introducing the change of variable 𝑡 = 𝑥𝛼 and
𝛼 > 0 on shifted Legendre polynomials. Let the FLFs 𝐿 𝑖 (𝑥𝛼 )
Definition 10. The Caputo type fractional derivative of order be denoted by FL𝛼𝑖 (𝑥) [52]. The fractional-order Legendre
V > 0 of a function 𝑓 : (0, ∞) → R is defined by functions are a particular solution of the normalized eigen-
functions of the singular Sturm-Liouville problem
𝑥
𝑐 1
𝐷V 𝑓 (𝑥) = ∫ (𝑥 − 𝑡)𝑛−V−1 𝑓 (𝑡) 𝑑𝑡. (17) 󸀠
Γ (𝑛 − V) 0 ((𝑥 − 𝑥1+𝛼 ) FL󸀠𝛼 2
𝑖 ) + 𝛼 𝑖 (𝑖 + 1) 𝑥
𝛼−1
FL𝛼𝑖 (𝑥) = 0,
(24)
One of the most important advantages of using a Caputo 𝑥 ∈ (0, 1) .
type fractional derivative is that the Caputo derivative of a
constant is zero, which means that this kind of derivative can Then, FL𝛼𝑖 (𝑥) by using (20) can be obtained as follows
be used to model the rate of change.
(2𝑖 + 1) (2𝑥𝛼 − 1) 𝛼 𝑖
For the Caputo derivative, we have FL𝛼𝑖+1 (𝑥) = FL𝑖 (𝑥) − FL𝛼 (𝑥) ,
𝑖+1 𝑖 + 1 𝑖−1
𝑐
𝐷V 𝐶 = 0, (𝐶 is a constant) ,
𝑖 = 1, 2, . . . ,
𝑐 V 𝛽
𝐷𝑥 FL𝛼0 (𝑥) = 1, FL𝛼1 (𝑥) = 2𝑥𝛼 − 1.
(25)
{0,
{ for 𝛽 ∈ N0 , 𝛽 < ⌈V⌉ ,
= { Γ (𝛽 + 1) 𝛽−V The analytic form of FL𝛼𝑖 (𝑥) of degree 𝑖𝛼 given by
{ 𝑥 , for 𝛽 ∈ N0 , 𝛽 ≥ ⌈V⌉ or 𝛽 ∉ N, 𝛽 > ⌊V⌋.
{ Γ(𝛽+1−V) 𝑖
(18)
FL𝛼𝑖 (𝑥) = ∑𝑏𝑠,𝑖 𝑥𝑠𝛼 , (26)
𝑠=0
The ceiling function ⌈V⌉ is used to denote the smallest integer
greater than or equal to V and floor function ⌊V⌋ to denote the where
largest integer less than or equal to V. Also, N = {1, 2, . . .} and
(𝑖 + 𝑠)!
N0 = {0, 1, 2, . . .}. 𝑏𝑠,𝑖 = (−1)𝑖+𝑠 . (27)
(𝑖 − 𝑠)!(𝑠!)2
Definition 11 (see [30]). Similar to the differential equation of
integer order, the Caputo’s fractional differentiation is a linear Note that FL𝛼𝑖 (0) = (−1)𝑖 and FL𝛼𝑖 (1) = 1. The FLFs are
operation, that is, orthogonal with respect to the weight function 𝑤(𝑥) = 𝑥𝛼−1
in the interval (0, 1] with the orthogonality property
𝑐
𝐷V (𝜆𝑓 (𝑥) + 𝜇𝑔 (𝑥)) = 𝜆𝑐 𝐷V 𝑓 (𝑥) + 𝜇𝑐 𝐷V 𝑔 (𝑥) , (19) 1
1
∫ FL𝛼𝑛 (𝑥) FL𝛼𝑚 (𝑥) 𝑤 (𝑥) 𝑑𝑥 = 𝛿 . (28)
where 𝜆 and 𝜇 are constants. 0 (2𝑛 + 1) 𝛼 𝑛𝑚
Abstract and Applied Analysis 5

A function 𝑓(𝑥), square integrable in (0, 1], may be expressed The production and destruction of species is described by
in terms of FLFs as kinetic equations governing the change of the number density
𝑁𝑖 of species 𝑖 over time [32–34], that is,

𝑓 (𝑥) = ∑𝑎𝑖 FL𝛼𝑖 (𝑥) , (29) 𝑑
𝑁 (𝑡) = 𝑎𝑁𝑖 (𝑡) , (34)
𝑖=0
𝑑𝑡 𝑖
where the coefficients 𝑎𝑗 are given by where 𝑎 is a constant. The destruction rate of the particles of
type 𝑖 is given by
1
𝑎𝑗 = 𝛼 (2𝑖 + 1) ∫ 𝑓 (𝑥) FL𝛼𝑖 (𝑥) 𝑤 (𝑥) 𝑑𝑥, 𝑑
𝑁 (𝑡) = −𝑏𝑁𝑖 (𝑡) , (35)
0 (30) 𝑑𝑡 𝑖
𝑗 = 0, 1, 2, . . . . where 𝑏 is a constant. Then, the residual effect can be taken as

In practice, only the first (𝑚)-terms FLFs are considered. 𝑑


𝑁 (𝑡) = −𝑐𝑁𝑖 (𝑡) , (36)
Then, we have 𝑑𝑡 𝑖

𝑚−1
where 𝑐 = 𝑏 − 𝑎. The solution of (36) can be seen as a simple
𝑓 (𝑥) ≃ 𝑓𝑚 (𝑥) = ∑ 𝑎𝑖 FL𝛼𝑖 𝑇
(𝑥) = 𝐴 Φ (𝑥) , (31) function
𝑖=0
𝑁𝑖 (𝑡) = 𝑁0 𝑒−𝑐𝑡 . (37)
with A fractional production-destruction equation can be
𝑇
obtained from a standard production-destruction equation
𝐴 = [𝑎0 , 𝑎1 , . . . , 𝑎𝑚−1 ] , by considering a fractional integral in place of a classical
(32) integral. A fractional production-destruction equation can
𝑇
Φ (𝑥) = [FL𝛼0 (𝑥) , FL𝛼1 (𝑥) , . . . , FL𝛼𝑚−1 (𝑥)] . be obtained, after dropping 𝑖, as

Theorem 12 (see [52]). Let the function 𝐷𝑘𝛼 𝑓(𝑥) ∈ 𝐶(0, 1] for 𝑁 (𝑡) − 𝑁0 = −𝑐V 0 𝐷𝑡−V 𝑁 (𝑡) , V > 0, (38)
𝑘 = 0, 1, . . . , 𝑚, (2𝑚 + 1) ≤ 𝛼 and P𝑚 = Span{FL𝛼𝑖 (𝑥)}𝑚−1
𝑖=0 . If where 0 𝐷𝑡−V represents the Riemann-Liouville fractional inte-
𝑓𝑚 = 𝐴𝑇 Φ(𝑥) is the best approximation to 𝑓 from P𝑚 , then the gral.
error bound is presented as follows: It is worth to note here the alternative approach of
Mainardi [93], for solving the fractional relaxation-oscillation
󵄩󵄩 󵄩 𝑀𝛼 1 equation, that is, the fractional production-destruction equa-
󵄩󵄩𝑓 (𝑥) − 𝑓𝑚 (𝑥)󵄩󵄩󵄩𝑤 ≤ √ , (33) tion with the Caputo fractional-order derivative.
Γ (𝑚𝛼 + 1) (2𝑚 + 1) 𝛼
As it can be seen, the fractional kinetic equation is
obtained by replacing the standard integral with the frac-
where 𝑀𝛼 ≥ |𝐷𝑚𝛼 𝑓(𝑥)|, 𝑥 ∈ (0, 1].
tional Riemann-Liouville integral. In the present research,
The later theorem proved that the approximate function firstly, we replace the Caputo fractional-order derivative with
based on the fractional Legendre polynomials converges to the classical derivative in the kinetic equation, and then the
the function 𝑓. derived fractional kinetic equation in the sense of the fuzzy
setting is solved by the proposed technique.
2.3. Fractional Kinetic Equations. Chemical kinetics as a
2.4. Fuzzy Fractional Differentiability. In the fractional lit-
science began in the middle of the 19th century, when
erature, Mittag-Leffler function plays an important role in
Wilhelmy [88] was apparently the first to recognize that the
the theory of fractional calculus/fuzzy fractional calculus and
rate at which a chemical reaction proceeds follows definite
fractional differential equations/fuzzy fractional differential
laws, and although his work paved the way for the law of
equations which is defined by [19, 30, 94]:
mass action of Waage and Guldberg [89], it attracted little
attention until it was taken up by Ostwald towards the end of ∞
𝑧𝑘
the century, as discussed by Laidler [90]. Wilhelmy realized 𝐸𝛼,𝛽 (𝑧) = ∑ , 𝛼 > 0, 𝛽 > 0. (39)
that chemical rates depended on the concentrations of the 𝑘=0
Γ (𝛼𝑘 + 𝛽)
reactants [91]. Also, it is useful to review some basic definitions and
Here, we provide a brief definition to illustrate the theorems related to the FFDEs. For more details see [72, 74,
chemical kinetic process. One can find more details in [91, 75]. Firstly, we present some notations which are used later in
92]. the paper.
Definition 13. Chemical kinetics is the study of the rate at (i) 𝐿R𝑝F (𝑎, 𝑏), 1 ≤ 𝑝 ≤ ∞ is the set of all fuzzy-valued
which a chemical process occurs. Besides information about measurable functions 𝑓 on [𝑎, 𝑏], where ‖𝑓‖𝑝 =
the speed at which reactions occur, kinetics also sheds light on 1 1/𝑝
the reaction mechanism (exactly how the reaction occurs). (∫0 (𝑑(𝑓(𝑡), 0))𝑝 𝑑𝑡) .
6 Abstract and Applied Analysis

(ii) 𝐶RF [𝑎, 𝑏] is a space of fuzzy-valued functions which For the sake of simplicity, we say that the fuzzy-valued
are continuous on [𝑎, 𝑏]. function 𝑓 𝑐 [(1) − V] is differentiable if it is differentiable as
(iii) 𝐶𝑛RF [𝑎, 𝑏] indicates the set of all fuzzy-valued func- in Definition 16 case (i), and 𝑓 is 𝑐 [(2) − V] differentiable if it
tions which are continuous up to order 𝑛. is differentiable as in Definition 16 case (ii), and so on for the
other cases.
(iv) 𝐴𝐶RF [𝑎, 𝑏] denotes the set of all fuzzy-valued func-
tions which are absolutely continuous. Theorem 17 (see [74]). Let 0 < V ≤ 1 and 𝑓 ∈ 𝐴𝐶RF [𝑎, 𝑏];
then the fuzzy Caputo fractional derivative exists almost
Note that one can find easily these notations in the crisp everywhere on (𝑎, 𝑏), and for all 0 ≤ 𝑟 ≤ 1, one has
context in [30] and references therein.
V
Definition 14 (see [74]). Let 𝑓 ∈ 𝐶RF [𝑎, 𝑏] ∩ 𝐿RF [𝑎, 𝑏]. (𝑐 𝐷𝑎+ 𝑓) (𝑥; 𝑟)
The Riemann-Liouville integral of fuzzy-valued function 𝑓 is 𝑥 𝑓󸀠 (𝑡) 𝑑𝑡𝑟 𝑥 𝑓󸀠 (𝑡) 𝑑𝑡 𝑟
defined as 1 − 1 +
=[ ∫ V , ∫ ]
𝑥 𝑓
Γ (1 − V) 𝑎 (𝑥 − 𝑡) Γ (1 − V) 𝑎 (𝑥 − 𝑡)V
V 1 (𝑡) 𝑑𝑡
(RL 𝐼𝑎+ 𝑓) (𝑥) = ∫ , 𝑥 > 𝑎, 0 < V ≤ 1. 1−V
Γ (V) 𝑎 (𝑥 − 𝑡)1−V = [(𝐼𝑎+ 𝐷𝑓−𝑟 ) (𝑥) , (𝐼𝑎+
1−V
𝐷𝑓+𝑟 ) (𝑥)] ,
(40) (46)

Definition 15 (see [74]). Let 𝑓 ∈ 𝐶RF [𝑎, 𝑏] ∩ 𝐿RF [𝑎, 𝑏]. Then when 𝑓 is (1) differentiable, and
𝑓 is said to be Caputo’s H-differentiable at 𝑥 when
V
𝑐 V RL 𝛽 (𝑐 𝐷𝑎+ 𝑓) (𝑥; 𝑟)
(i) ( 𝐷𝑎+ 𝑓) (𝑥) =( 𝐷𝑎+ [𝑓 (𝑡) ⊖ 𝑓 (𝑎)]) (𝑥) ,
𝑟
𝑥 𝑓󸀠 (𝑡) 𝑑𝑡 𝑥 𝑓󸀠 (𝑡) 𝑑𝑡 𝑟
(41) 1 + 1 −
𝑐 V RL 𝛽 =[ ∫ , ∫ ]
(ii) ( 𝐷𝑎+ 𝑓) (𝑥) =( 𝐷𝑎+ [−𝑓 (𝑎) ⊖ (−𝑓 (𝑡))]) (𝑥) . Γ (1 − V) 𝑎 (𝑥 − 𝑡)V Γ (1 − V) 𝑎 (𝑥 − 𝑡)V
1−V
Definition 16 (see [74]). Let 𝑓 : 𝐿RF [𝑎,𝑏] ∩ 𝐶RF [𝑎, 𝑏] and 𝑥0 ∈ = [(𝐼𝑎+ 𝐷𝑓+𝑟 ) (𝑥) , (𝐼𝑎+
1−V
𝐷𝑓−𝑟 ) (𝑥)] ,
𝑥
(𝑎, 𝑏) and Φ(𝑥) = (1/Γ(1 − V)) ∫𝑎 (𝑓(𝑡)/(𝑥 − 𝑡)V )𝑑𝑡. We say that (47)
𝑓(𝑥) is fuzzy Caputo fractional differentiable of order 0 < V ≤
1 at 𝑥0 , if there exists an element (𝑐 𝐷𝑎+
V
𝑓)(𝑥0 ) ∈ 𝐶𝐸 [𝑎, 𝑏] such when 𝑓 is (2) differentiable.
that for all 0 ≤ 𝑟 ≤ 1, ℎ > 0,

Φ (𝑥0 + ℎ) ⊖ Φ (𝑥0 ) 3. The Governing Fractional Kinetic Equation


V
(i) (𝑐 𝐷𝑎+ 𝑓) (𝑥0 ) = lim+
ℎ→0 ℎ In this section, the kinetic data experimented based on
(42) the hydrolysis of OPEFB fiber at 120∘ C under low acid
Φ (𝑥0 ) ⊖ Φ (𝑥0 − ℎ)
= lim+ , concentration conditions [95] is modelled by means of the
ℎ→0 ℎ fractional calculus.
or In what follows are the detailed equation for the classic
kinetic model that will be used to analyze the experimental
V Φ (𝑥0 ) ⊖ Φ (𝑥0 + ℎ) data presented by [95].
(ii) (𝑐 𝐷𝑎+ 𝑓) (𝑥0 ) = lim+
ℎ→0 −ℎ
(43)
Φ (𝑥0 − ℎ) ⊖ Φ (𝑥0 ) 3.1. Model Development. By the addition of water to the
= lim+ , xylan, the hydrolysis occurs and xylose is produced
ℎ→0 −ℎ
or 𝑘1 𝑘2
Hemicellulose xylan xylose Degradation products.
𝑐 V Φ (𝑥0 + ℎ) ⊖ Φ (𝑥0 )
(iii) ( 𝐷𝑎+ 𝑓) (𝑥0 ) = lim+
ℎ→0 ℎ
(44) H2 O 3H2 O
Φ (𝑥0 − ℎ) ⊖ Φ (𝑥0 )
= lim+ , (48)
ℎ→0 −ℎ
or In 1945, Saeman [4] worked on dilute sulfuric acid
that was catalyzed hemicellulose hydrolysis. The reaction of
V Φ (𝑥0 ) ⊖ Φ (𝑥0 + ℎ) hemicellulose hydrolysis was modelled as a consecutive first-
(iv) (𝑐 𝐷𝑎+ 𝑓) (𝑥0 ) = lim+
ℎ→0 −ℎ order reaction with two steps. Firstly, xylan is hydrolyzed
(45) to xylose (𝑘1 ), and secondly the furfural was produced by
Φ (𝑥0 ) ⊖ Φ (𝑥0 − ℎ) degradation of xylose (𝑘2 ) when it is released from the solid
= lim+ .
ℎ→0 ℎ matrix and entered the acidic solution. This classic kinetic
Abstract and Applied Analysis 7

Table 1: 𝑘1 and 𝑘2 values calculated at 120∘ C and acid concentrations 9


of 2%–6%. 8
2% H2 SO4 4% H2 SO4 6% H2 SO4 7
𝑘1 0.012 0.014 0.05
6
𝑘2 0.001 0.001 0.05

Xylose (g/L)
5
4
model that will be utilized to analyze this experimental data 3
is revealed by the equation as follows:
2
xylan 󳨃
󳨀→ xylose 󳨃
󳨀→ Decomposed products 1
(49) 0
𝑘1 𝑘2
𝐴 󳨀→ 𝐵 󳨀→ 𝐶, 0 50 100 150 200
Time (min)
where 𝐴 = xylan; 𝐵 = xylose; 𝐶 = Decomposition products; Xcal
𝑘1 = sugar release rate; 𝑘2 = sugar decomposition rate. Xexp
Material balance for components “𝐴” and “𝐵” for the first-
order kinetics gives Figure 1: Comparison between calculated and experimental value
of xylose in hydrolysis of OPEFB with 2% H2 SO4 at 120∘ C [95].
𝑑𝐶𝐴 (𝑡)
− = 𝑘1 𝐶𝐴 (𝑡) (50)
𝑑𝑡 9

in which the initial concentration at 𝑡 = 0 is presented by 8


𝐶𝐴 = 𝐶𝐴 0 . Also, we have the same way for material 𝐵: 7
6
𝑑𝐶𝐵 (𝑡)
Xylose (g/L)

− = 𝑘1 𝐶𝐴 (𝑡) − 𝑘2 𝐶𝐵 (𝑡) (51)


𝑑𝑡 5
4
in which the initial concentration at 𝑡 = 0 is presented by
𝐶𝐵 = 𝐶𝐵0 . Equation (51) can be integrated, and using the 3
given boundary condition results in 2
1
𝐶𝐴 (𝑡) = 𝐶𝐴 0 exp (−𝑘1 𝑡) . (52)
0
0 50 100 150 200
Substituting (51) into (52) gives
Time (min)
𝑑𝐶𝐵 (𝑡) Xcal
+ 𝑘2 𝐶𝐵 (𝑡) = 𝑘1 𝐶𝐴 0 exp (−𝑘1 𝑡) . (53)
𝑑𝑡 Xexp

Figure 2: Comparison between calculated and experimental value


In this part, we apply the fractional Caputo-type deriva-
of xylose in hydrolysis of OPEFB with 4% H2 SO4 at 120∘ C [95].
tive in (53) instead of the classical ones. So, we have
𝑐
𝐷V 𝐶𝐵 (𝑡) + 𝑘2 𝐶𝐵 (𝑡) = 𝑘1 𝐶𝐴 0 exp (−𝑘1 𝑡) . (54)
4. Shifted Fractional Legendre Spectral
3.2. Determination of Reaction Rate Constants. Determina- Tau Method
tions of the rate constants (𝑘1 , 𝑘2 ) depend on the order of
The main objective of this section is to apply the shifted
reaction. By using experimental data, the correct order would
fractional Legendre spectral tau (SFLT) method to construct
be calculated by which function of rate equation best fit
the spectral solution for the fractional linear kinetic equation
the linear demand. When the order is established, the rate
presented in Section 3, in the fuzzy concept. We now derive
constants are determined from the slope of the linear plot
the operational matrix for the fuzzy fractional calculus and
(Table 1) by using Figures 1, 2, and 3 and (55).
formulate the problem in terms of the Caputo fractional
For a first-order reaction, the rate can be determined from
derivative.
the rate of formation of xylose which can be explained by
Please notice that one can easily see why we choose such
equation
orthogonal function for this contribution in [52, 53] and
𝐶𝐴 references therein.
ln = −𝑘1 𝑡. (55) Now, definitions of fuzzy approximation function are
𝐶𝐴 0 established based on the application of fractional Legendre
8 Abstract and Applied Analysis

9 were the fuzzy shifted fractional Legendre coefficient vector


8 𝐴𝑇𝑚 and shifted fractional Legendre vector Φ𝑚 (𝑥) are defined
as
7
6 𝐴𝑇𝑚 = [𝑎0 , 𝑎1 , . . . , 𝑎𝑚−1 ] ,
(60)
Xylose (g/L)

5 Φ𝑚 (𝑥) = [FL𝛼0 (𝑥) , FL𝛼1 (𝑥) , . . . , FL𝛼𝑚−1 (𝑥)] .


4
Definition 20 (see [97]). A fuzzy-valued polynomial 𝑝̃∗ ∈
3
̃ is the best approximation to fuzzy function 𝑓 on 𝜒 =
∏ 𝑁
2 {𝑥0 , 𝑥1 , 𝑥2 , . . . , 𝑥𝑁}, if
1
0 max 𝐷 (𝑝̃∗ (𝑥𝑖 ) , 𝑓𝑖 ) = min { max 𝐷 (𝑝̃ (𝑥𝑖 ) , 𝑓𝑖 )} ,
𝑖=0,1,2,...,𝑁 ̃ ∏
𝑝∈ ̃ 𝑖=0,1,2,...,𝑁
0 50 100 150 200 𝑁

Time (min) (61)


Xcal ̃ is the set of all fuzzy-valued polynomials.
in which ∏
Xexp 𝑁

Figure 3: Comparison between calculated and experimental value The problem is addressed to the best shifted fractional
of xylose in hydrolysis of OPEFB with 6% H2 SO4 at 120∘ C [95]. Legendre approximation, as we use fractional Legendre’s
nodes.

Theorem 21. The best approximation of a fuzzy function based


on the fractional Legendre nodes exists and is unique.
polynomials. For more details about the fuzzy approximation Proof. The proof is an immediate result of Theorem 4.2.1 in
functions, see [96–99]. [97].
The following theorem defines an upper error bound for
4.1. Approximation of Fuzzy Function the fuzzy approximation function based on the fractional
Legendre polynomials. Concerning this theorem, one can
Definition 18. For 𝑦 ∈ 𝐿R𝑝F (0, 1) ∩ 𝐶RF (0, 1) and fractional
find that the fuzzy approximate function based on the FLFs is
Legendre polynomial FL𝛼𝑖 (𝑥) a real-valued function over convergent to the desired fuzzy function.
(0, 1), the fuzzy function is approximated by
Theorem 22. Let one consider that the function 𝑦(𝑥) : [𝑥0 , 1]

∗ → 𝐿R𝑝F [0, 1]∩𝐶RF [0, 1] is continuously fuzzy differentiable for
𝑦 (𝑥) = ∑ 𝑎𝑗 ⊙FL𝛼𝑖 (𝑥) , 𝑥 ∈ (0, 1) , (56) RF
𝑗=0 𝑥0 > 0, 𝑦(𝑥) ∈ 𝐶𝑚 [𝑥0 , 1] and the fractional-order derivative
is 0 < V ≤ 1 and P𝑚 = Span{FL𝛼0 (𝑥), FL𝛼1 (𝑥), . . . , FL𝛼𝑚−1 (𝑥)}.
where the fuzzy coefficients 𝑎𝑗 are obtained by If 𝑦𝑚 = 𝐴𝑇 Φ(𝑥) is the best fuzzy approximation to 𝑦(𝑥) from
P𝑚 , then the error bound is presented as follows:
1
𝑎𝑗 = 𝛼 (2𝑗 + 1) ⊙ ∫ 𝑦 (𝑥) ⊙ FL𝛼𝑖 (𝑥) ⊙ 𝑤 (𝑥) 𝑑𝑥, 𝑀V 1
(57) 𝐷∗ (𝑦 (𝑥) , 𝑦𝑚 (𝑥)) ≤ √ , (62)
0 Γ (𝑚V + 1) (2𝑚 + 1) V

in which FL𝛼𝑖 (𝑥) is as the same in (26), and ∑∗ means addition where 𝑀V ≥ sup𝑟∈[0,1] max𝑥∈[𝑥0 ,1] {|𝐷V 𝑦1𝑟 (𝑥)|, |𝐷V 𝑦2𝑟 (𝑥)|} and
with respect to ⊕ in RF . 𝑦𝑟 (𝑥) = [𝑦1𝑟 (𝑥), 𝑦2𝑟 (𝑥)].

Remark 19. Practically, only the first (𝑚)-terms shifted frac- Proof. Considering Definition 4 for which we know that
tional Legendre polynomials are considered. So, we have
𝐷∗ (𝑦 (𝑥) , 𝑦𝑚 (𝑥))
𝑚−1
𝑦 (𝑥) ≃ 𝑦𝑚 (𝑥) = ∑ ∗ 𝑎𝑗 ⊙ FL𝛼𝑖 (𝑥) = 𝐴𝑇𝑚 ⊙ Φ𝑚 (𝑥) ; (58) = sup 𝐷 (𝑦 (𝑥) , 𝑦𝑚 (𝑥))
𝑥∈[𝑥0 ,1]
𝑗=0
󵄨 󵄨 󵄨 󵄨
= sup sup max {󵄨󵄨󵄨󵄨𝑦1𝑟 (𝑥) − 𝑦𝑚,1
𝑟
(𝑥)󵄨󵄨󵄨󵄨 , 󵄨󵄨󵄨󵄨𝑦2𝑟 (𝑥) − 𝑦𝑚,2
𝑟
(𝑥)󵄨󵄨󵄨󵄨}
hence 𝑥∈[𝑥 ,1] 𝑟∈[0,1]
0

󵄩 󵄩 󵄩 󵄩
𝑚−1 = sup max {󵄩󵄩󵄩󵄩𝑦1𝑟 (𝑥) − 𝑦𝑚,1
𝑟
(𝑥)󵄩󵄩󵄩󵄩𝑤 , 󵄩󵄩󵄩󵄩𝑦2𝑟 (𝑥) − 𝑦𝑚,2
𝑟
(𝑥)󵄩󵄩󵄩󵄩𝑤 } .
𝑟 𝑟 ∗ 𝑟
𝑦 (𝑥) ≃ 𝑦𝑚 (𝑥) = ∑ 𝑎𝑗 ⊙ FL𝛼𝑖 (𝑥) , (59) 𝑟∈[0,1]
𝑗=0 (63)
Abstract and Applied Analysis 9

Using Theorem 12, we have 10−10

𝐷∗ (𝑦 (𝑥) , 𝑦𝑚 (𝑥))
󵄨󵄨 V 󵄨
󵄨󵄨𝐷 𝑦1 (𝑥)󵄨󵄨󵄨 1
≤ sup max { √ ,

Absolute error
𝑟∈[0,1] Γ (𝑚V + 1) (2𝑚 + 1) 𝛼
󵄨󵄨 V 󵄨 10−11
󵄨󵄨𝐷 𝑦2 (𝑥)󵄨󵄨󵄨 1 (64)
√ }
Γ (𝑚V + 1) (2𝑚 + 1) 𝛼

𝑀V 1
≤ √ ,
Γ (𝑚V + 1) (2𝑚 + 1) V
10−12
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
in which 𝑀 ≥ sup𝑟∈[0,1] max𝑥∈[𝑥0 ,1] {|𝐷V 𝑦1𝑟 (𝑥)|, |𝐷V 𝑦2𝑟 (𝑥)|}
V
r-cuts
and 𝑦𝑟 (𝑥) = [𝑦1𝑟 (𝑥), 𝑦2𝑟 (𝑥)].
 = 𝛼 = 0.5  = 𝛼 = 0.85
 = 𝛼 = 0.75  = 𝛼 = 0.95
4.2. Operational Matrix of FLFs
Figure 4: The absolute error for different values V = 𝛼 with 𝑚 = 12,
Lemma 23. The fuzzy Caputo fractional derivative of order 𝑘1 = 0.012, and 𝑘2 = 0.001.
0 < V ≤ 1 over the shifted fractional legendre functions can
be described in the form of
10−8
𝑖
𝑐 Γ (𝑘𝛼 + 1) 𝑘𝛼−V
𝐷V FL𝛼𝑖 (𝑥) = ∑ 𝑏𝑘,𝑖
󸀠
𝑥 , (65)
𝑘=0
Γ (𝑘𝛼 − V + 1)
󸀠 󸀠
where 𝑏𝑘,𝑖 = 0 if 𝑏𝑘,𝑖 = (Γ(𝑘𝛼 + 1)/Γ(𝑘𝛼 − V + 1))𝑏𝑘𝑗 . 10−9
Absolute error

Proof. Taking into account the analytic form of the FLFs


explained in Section 2.2 and (26) and the fuzzy fractional
Caputo definition, we have 10−10
𝑖
𝑐
𝐷V FL𝛼𝑖 (𝑥) = ∑ 𝑏𝑘,𝑖 𝑐 𝐷𝛼 ⊙ 𝑥𝑘𝛼 . (66)
𝑘=0
10−11
Now, by utilizing the properties of the Caputo fractional 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
derivative explained in Section 2.1, the lemma can be r-cuts
proved. m=5 m = 12
m=9 m = 15
The fuzzy Caputo operational matrix based on the FLFs
is stated the same as the crisp concept. For more details, see Figure 5: The absolute error for different values 𝑚 with V = 𝛼 =
[52, 76]. So, we have 0.75, 𝑘1 = 0.012, and 𝑘2 = 0.001.
𝑐 (V)
𝐷V Φ (𝑥) ≃ 𝑐 𝐷 Φ (𝑥) , (67)

where 𝑐 𝐷(V) is the 𝑚×𝑚 operational matrix of fuzzy fractional in which 𝜃𝑖,𝑗,𝑘 are acquired by
V
Caputo’s derivative of FLFs and 𝑐 𝐷 Φ(𝑥) ∈ 𝐶RF [𝑎, 𝑏]. So,
using (67) and Lemma 23, we extend the operational matrix 𝜃𝑖,𝑗,𝑘
of Caputo fractional-order derivative of FLFs in the sense of
the fuzzy setting as follows. = 𝛼 (2𝑗 + 1)
𝑗
Theorem 24. Let one assume that Φ(𝑥) is the FLF vector. 𝐷𝑖,𝑗 (V) (−1)𝑖+𝑗+𝑘+𝑙 (𝑖 + 𝑘)! (𝑙 + 𝑗)!Γ (𝑘𝛼 + 1)
×∑ ,
is the 𝑚-square operational matrix of fuzzy fractional Caputo’s 𝑙=0 (𝑖 − 𝑘)!(𝑘!)2 Γ(𝑘𝛼−V+1)(𝑗 − 𝑙)!(𝑙!)2 (𝑘𝛼 + 𝑙𝛼 + 𝛼 − V)
(V)
derivative of order 0 < V ≤ 1. Then, the elements of 𝐷𝑖,𝑗 are
𝑗 = 0, 1, . . . , 𝑚 − 1.
achieved as
(69)
𝑖
(V)
𝐷𝑖,𝑗 = ∑ 𝜃𝑖,𝑗,𝑘 , 𝑖 = 0, 1, . . . , 𝑚 − 1, (68)
𝑘=⌈𝛼⌉
Note that in 𝐷(V) , the first ⌈V⌉ rows are all zero.
10 Abstract and Applied Analysis

10−4 problem to a system of fuzzy algebraic equations. The large


systems of algebraic equations may lead to greater computa-
tional complexity and large storage requirements. However,
10−5 the operational matrix for the FLFs is structurally spare.
This reduces the computational complexity of the resulting
Absolute error

10−6 algebraic system.


Fractional kinetic differential equation was derived in the
deterministic case in Section 3. Now, in order to investigate
10−7 the mentioned problem in a real case, we use the fuzzy initial
value 𝐶𝐵0 , the fuzzy-valued function 𝑔(𝑥), and the concept of
10−8 Caputo’s H-differentiability for fractional derivative of 𝐶𝐵 (𝑥),
(𝑐 𝐷0+
V
𝐶𝐵 )(𝑥) and generalized H-differentiability [67] for first-
order derivative of 𝐶𝐵 (𝑥).
10−9 Let us consider the fuzzy version of the linear fractional
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cuts
kinetic differential equation in the general form:

𝛼 = 0.95 V
𝛼 = 0.5 (𝑐 𝐷0+ 𝐶𝐵 ) (𝑥) = 𝐶𝐵 (𝑥) ⊕ 𝑔 (𝑥) , 0 < V ≤ 1,
𝛼 = 0.75 𝛼=1 (71)
Figure 6: The absolute error for different values 𝛼 with V = 0.75,
𝐶𝐵 (0) = 𝐶𝐵0 ∈ RF ,
𝑘1 = 0.012, 𝑚 = 7, and 𝑘2 = 0.001.
where 𝐶𝐵 (𝑥) : 𝐿RF ∩ 𝐶RF is a continuous fuzzy-valued func-
V
tion, 𝑐 𝐷0+ (𝑥) denotes the fuzzy Caputo fractional derivative
Proof. If we consider the fractional derivative in the crisp of order V, and 𝑔(𝑥) : [0, 1] → RF .
context, then the proof of this theorem is obtained from Firstly, we state the unknown fuzzy functions 𝐶𝐵 (𝑥),
Theorem 2 in [52]. In the same way, if we consider the (𝑐 𝐷0V+ 𝐶𝐵 )(𝑥) and known fuzzy function 𝑔(𝑥) in terms of the
fuzzy Caputo fractional derivative, then analogously to the FLFs as follows:
demonstration of Theorem 1 in [52] and by using Lemma 23,
𝑚−1
we can prove the relation (69). 𝑟
𝐶𝐵𝑟 (𝑥) ≃ 𝐶𝐵 𝑟𝑚 (𝑥) = ∑ ∗ 𝑎𝑗 ⊙ FL𝛼𝑗 (𝑥) = 𝐴𝑇𝑚 ⊙ Φ𝑟𝑚 (𝑥) ,
Also, we can exhibit the operational matrix 𝐷(V) in the 𝑗=0

alternative form as (72)


𝑚−1
𝐷(V) V 𝑟 V
(𝑐 𝐷0+ 𝐶𝐵 ) (𝑥) ≃ ∑ ∗ 𝑎𝑗 ⊙ 𝑐 𝐷 FL𝛼𝑗 (𝑥)
0 0 0 ... 0 𝑗=0 (73)
.. .. .. ..
. . . ... . = 𝐴𝑇𝑚 ⊙ 𝐷(V) Φ𝑟𝑚 (𝑥) ,
( 0 0 0 ... 0 )
( ⌈V⌉ ⌈V⌉ ⌈V⌉ ⌈V⌉ ) 𝑚−1
( ∑𝜃 ∑ 𝜃⌈V⌉,1,𝑘 ∑ 𝜃⌈V⌉,2,𝑘 . . . ∑ 𝜃⌈V⌉,𝑚−1,𝑘 ) 𝑟
( ⌈V⌉,0,𝑘 ) 𝑟
𝑔 (𝑥) ≃ 𝑔𝑚 (𝑥) = ∑ ∗ 𝑔𝑗 ⊙ FL𝛼𝑗 (𝑥) = 𝐺𝑚
𝑇
⊙ Φ𝑟𝑚 (𝑥) . (74)
( 𝑘=⌈V⌉ )
( 𝑘=⌈V⌉ 𝑘=⌈V⌉ 𝑘=⌈V⌉
) 𝑗=0
( .. .. .. .. )
=(
( 𝑖 . . . ... . ).
)
( 𝑖 𝑖 𝑖 ) The following theorem provides a suitable way to reach
( ∑ 𝜃𝑖,0,𝑘 ∑ 𝜃𝑖,1,𝑘 ∑ 𝜃𝑖,2,𝑘 . . . ∑ 𝜃𝑖,𝑚−1,𝑘 )
( ) the fuzzy unknown coefficient of the fuzzy approximate
( 𝑘=⌈V⌉ 𝑘=⌈V⌉ 𝑘=⌈V⌉ 𝑘=⌈V⌉ )
( ) function 𝐶𝐵 𝑟𝑚 (𝑥), by means of the fuzzy residual function
( .. .. .. .. )
. . . ... . (𝑅𝑚 (𝑥)) of the problem stated.
𝑚−1 𝑚−1 𝑚−1 𝑚−1
∑ 𝜃𝑚−1,0,𝑘 ∑ 𝜃𝑚−1,0,𝑘 ∑ 𝜃𝑚−1,0,𝑘 . . . ∑ 𝜃𝑚−1,0,𝑘
(𝑘=⌈V⌉ 𝑘=⌈V⌉ 𝑘=⌈V⌉ 𝑘=⌈V⌉ ) Theorem 26 (see [76]). Let 𝐶𝐵 (𝑥) ∈ 𝐶RF [0, 1] and 0 < V ≤ 1,
(70) then the fuzzy residual function can be expressed by

Remark 25. If V = 1, then Theorem 24 gives the operational (V) (𝑟) (𝑟)
matrix for the shifted Legendre polynomials function as [(𝑐 𝐷 𝐶𝐵𝑚 ) (𝑥)] = [𝑅𝑚 (𝑥) ⊕ 𝑔𝑚 (𝑥) ⊕ 𝐶𝐵𝑚 (𝑥)] ,
described in Section 2.2.
𝑟 ∈ [0, 1] .
(75)
4.3. The Application of the FLFs for Solving Linear FFKE. In
this section, we are concerned with providing a numerical
solution to linear FFKE which was described in Section 3. We Let ⟨⋅, ⋅⟩RF denote the fuzzy inner product over 𝑋RF =
approximate the fuzzy function 𝑦(𝑥) by means of the FLFs 𝐿2RF ([0, 1]).As in a typical tau method (see [100, 101]), we
and then substitute in the FFKE to derive the approximate make (𝑚 − 1) linear fuzzy equations from the following inner
solution. The method reduces the fuzzy fractional initial product as
Abstract and Applied Analysis 11

 = 𝛼 = 0.5  = 𝛼 = 0.75

Approximate solution

Approximate solution
2 2

1 1

0 0
1 1
1 1
0.5 0.5
r-cu 0.5 r-cu 0.5
ts ts
0 0 x 0 0 x

 = 𝛼 = 0.85  = 𝛼 = 0.95

Approximate solution
2 2
Approximate solution

1 1

0 0
1 1
1 1
0.5 0.5 0.5
r-cu r-cu 0.5
ts 0 0 x ts 0 0 x

Figure 7: The fuzzy approximate solution for different values of fractional orders V = 𝛼 with 𝑚 = 8, 𝑘1 = 0.012, and 𝑘2 = 0.001.

10−9 10−8

10−10

10−11 10−9
Absolute error
Absolute error

10−12

10−13 10−10

10−14

10−15 10−11
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cuts r-cuts

 = 𝛼 = 0.5  = 𝛼 = 0.85 m=5 m = 12


 = 𝛼 = 0.75  = 𝛼 = 0.95 m=9 m = 15

Figure 8: The absolute error for different values V = 𝛼 with 𝑚 = 8, Figure 9: The absolute error for different values 𝑚 with V = 𝛼 =
𝑘1 = 0.014, and 𝑘2 = 0.001. 0.85, 𝑘1 = 0.014, and 𝑘2 = 0.001.

⟨𝑅𝑚 (𝑥; 𝑟) , FL𝛼𝑖 (𝑥; 𝑟)⟩RF = 0̃, 𝑖 = 0, 1, . . . , 𝑚 − 2, 𝑟 ∈ [0, 1] , From (76), we make (𝑚 − 1) fuzzy linear algebraic
(76) equations which are as follows in the expanded form:

1 𝑚−2
where ⟨𝑅𝑚 (𝑥; 𝑟), FL𝛼𝑖 (𝑥; 𝑟)⟩RF = [(FR) ∫0 𝑅𝑚 (𝑥) ⊙ (𝑥), (𝑟)
(𝑟)
∑ ∗ 𝑎𝑗 ⊙ {⟨𝐷(V) FL𝛼𝑗 (𝑥) , FL𝛼𝑖 (𝑥)⟩ − ⟨FL𝛼𝑗 (𝑥) , FL𝛼𝑖 (𝑥)⟩}
FL𝛼𝑖 (𝑥) ⊙ 𝑤(𝑥)𝑑𝑥] . 𝑗=0
12 Abstract and Applied Analysis

10−5 the exact solution reveals that our method is efficient and
convenient.
Now, we consider (54) with the fuzzy initial condition.
10−6
It means that the initial concentration of xylose is a fuzzy
number. So, we have
Absolute error

10−7
Example 27. Consider the following FFDE:

10−8 𝑐 V
𝐷0+ 𝐶𝐵 (𝑡) ⊕ 𝑘2 𝐶𝐵 (𝑡) = 𝑘1 𝐶𝐴 0 exp (−𝑘1 𝑡) , 0 < V ≤ 1,

10−9 𝐶𝐵 (0; 𝑟) = [𝐶𝐵𝑟 01 , 𝐶𝐵𝑟 02 ] , 0 < 𝑟 ≤ 1,


(81)
−10
10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cuts
in which 𝐶𝐵 (𝑡) : 𝐿RF [0, 1] ∩ 𝐶RF [0, 1] is a continuous
fuzzy function, 𝑐 𝐷0V+ denotes the fuzzy Caputo fractional
𝛼 = 0.5 𝛼 = 0.85 derivative of order V, and 𝑘1 , 𝑘2 are constant coefficients
𝛼 = 0.75 𝛼=1 specified from Table 1.
Figure 10: The absolute error for different values 𝛼 with V = 0.85,
Regarding the definition of the Caputo fuzzy fractional
𝑘1 = 0.014, 𝑚 = 10, and 𝑘2 = 0.001.
differentiability for 𝑐 [1 − V] differentiability and Theorem 17,
we can state (81) in a parametric form as follows:
𝑚−2
(𝑟)
= ∑ ∗ 𝑔𝑗 ⊙ ⟨FL𝛼𝑗 (𝑥) , FL𝛼𝑖 (𝑥)⟩ , V
(𝑐 𝐷0+ 𝐶𝐵1 ) (𝑡; 𝑟) + 𝑘2 𝐶𝐵1 (𝑡; 𝑟) = 𝑘1 𝐶𝐴 0 exp (−𝑘1 𝑡) ,
𝑗=0
0 < V ≤ 1,
(77)
𝐶𝐵 (0; 𝑟) = 𝐶𝐵𝑟 01 , 0 < 𝑟 ≤ 1,
for 𝑖 = 0, 1, . . . , 𝑚 − 2. Also, the fuzzy coefficients 𝑔𝑗 are (82)
V
defined as (𝑐 𝐷0+ 𝐶𝐵2 ) (𝑡; 𝑟) + 𝑘2 𝐶𝐵2 (𝑡; 𝑟) = 𝑘1 𝐶𝐴 0 exp (−𝑘1 𝑡) ,
𝑚−1
𝑟
𝑔 (𝑥) ≃ 𝑔𝑚
𝑟
(𝑥) = ∑ ∗ 𝑔𝑗 ⊙ FL𝛼𝑖 (𝑥) = 𝐺𝑚
𝑇
⊙ Φ𝑚 , 0 < V ≤ 1,
(78)
𝑗=0
𝐶𝐵 (0; 𝑟) = 𝐶𝐵𝑟 02 , 0 < 𝑟 ≤ 1,
𝑇
where 𝐺 = [𝑔0 , 𝑔1 , . . . , 𝑔𝑚−1 ] is gained as
where 𝐶𝐵 (0; 𝑟) = [0.5 + 0.5𝑟, 1.5 − 0.5𝑟]. The exact solution of
1
(81) can be gained using (82) as
𝑔𝑖 = (2𝑖 + 1) 𝛼 ∫ FL𝛼𝑖 (𝑥) ⊙ 𝑔 (𝑥) ⊙ 𝑤 (𝑥) 𝑑𝑥,
0 (79)
𝐶𝐵𝑚 1 (𝑡; 𝑟)
𝑖 = 0, 1, . . . , 𝑚 − 1.
= (0.5 + 0.5𝑟) 𝐸V,1 [−𝑘2 𝑡V ]
Consequently, putting (72) in the initial condition of the
problem (71), we have 𝑡
+ ∫ (𝑡 − 𝜏)V−1 𝐸V,V [−𝑘2 (𝑡 − 𝜏)V ] 𝑘1 𝐶𝐴 0 exp (−𝑘1 𝑡) 𝑑𝜏,
𝑚−1 0
(𝑟)
𝐶𝐵 (0) = ∑ ∗ 𝑎𝑗 ⊙ FL𝛼𝑖 (0) = 𝐶𝐵0 . (80) 0 < V ≤ 1,
𝑗=0
𝐶𝐵𝑚 2 (𝑡; 𝑟)
Considering the aforementioned equation with the fuzzy
algebraic equations system (77), (𝑚)-fuzzy linear algebraic = (1.5 − 0.5𝑟) 𝐸V,1 [−𝑘2 𝑡V ]
equations are produced. Solving this resultant fuzzy system
𝑡
based on the method presented in [86], the unknown fuzzy
+ ∫ (𝑡 − 𝜏)V−1 𝐸V,V [−𝑘2 (𝑡 − 𝜏)V ] 𝑘1 𝐶𝐴 0 exp (−𝑘1 𝑡) 𝑑𝜏,
coefficients 𝑎𝑗 for 𝑗 = 0, . . . , 𝑚 − 1 will be obtained. 0

0 < 𝑟 ≤ 1.
5. Numerical Results (83)
In order to illustrate the effectiveness and accuracy of the
proposed method, we carry it out for solving the fractional Employing the proposed method described in Section 4,
kinetic equation which was derived in Section 3 in the sense we can derive the operational matrix of the FLFs based on
of the fuzzy concept. Comparison of the results obtained with the fuzzy fractional derivative of the Caputo type and the
Abstract and Applied Analysis 13

 = 𝛼 = 0.5  = 𝛼 = 0.75

Approximate solution
Approximate solution 2 2

1 1

0 0
1 1
1 1
0.5 0.5
r-cu 0.5 r-cu 0.5
ts x ts
0 0 0 0 x

 = 𝛼 = 0.85  = 𝛼 = 0.95
Approximate solution

Approximate solution
2 2

1 1

0 0
1 1
1 1
0.5 0.5
r-cu 0.5 r-cu 0.5
ts
ts 0 0 x 0 0 x

Figure 11: The fuzzy approximate solution for different values of fractional orders V = 𝛼 with 𝑚 = 12, 𝑘1 = 0.014, and 𝑘2 = 0.001.

10−6 10−8

10−8

10−9
Absolute error
Absolute error

−10
10

10−12
10−10

10−14

10−16 10−11
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cuts r-cuts

 = 𝛼 = 0.5  = 𝛼 = 0.95 m=5 m = 12


 = 𝛼 = 0.75 =𝛼=1 m=9 m = 15

Figure 12: The absolute error for different values V = 𝛼 with 𝑚 = 6, Figure 13: The absolute error for different values 𝑚 with V = 𝛼 =
𝑘1 = 0.05, and 𝑘2 = 0.05. 0.95, 𝑘1 = 0.05, and 𝑘2 = 0.05.

approximate fuzzy function of the solution by means of the


FLFs. So, we can express (81) as
where the matrix of the unknown coefficients of the fuzzy
𝐴 1 𝑇𝑚 [𝐷(V) + 𝐼] Φ (𝑥) = 𝐺1 𝑇𝑚 Φ (𝑥) , approximate solution is 𝐴𝑇𝑚 = [𝐴 1𝑚 , 𝐴 2𝑚 ]𝑇 and the right-
(84) 𝑇
hand side coefficients matrix is 𝐺𝑚 = [𝐺1𝑚 , 𝐺2𝑚 ]𝑇 , where the
𝐴 2 𝑇𝑚 [𝐷(V) + 𝐼] Φ (𝑥) = 𝐺2 𝑇𝑚 Φ (𝑥) , elements of the vector 𝐺𝑇 are achieved from (79).
14 Abstract and Applied Analysis

10−5 where our aim is to find the unknown fuzzy coefficients


{𝑎𝑗 }2𝑗=0 . The operational matrix of the FLFs is as
10−6
0 0 0
10−7 𝜃1,0,1 𝜃1,1,1 𝜃1,2,1
𝐷(0.75) = ( )
Absolute error

2 2 2
−8 ∑ 𝜃1,0,𝑘 ∑ 𝜃1,1,𝑘 ∑ 𝜃1,2,𝑘
10
𝑘=1 𝑘=1 𝑘=1 (88)
10−9 0 0 0
= ( 1.8381 0 0) .
10−10 −1.1751 4.3392 0

The FLFs for 𝑚 = 2 are as follows:


10−11
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cuts { 1,
{
FL0.75
𝑖 (𝑥) = {−1 + 2𝑥(3/4) , (89)
𝛼 = 0.5 𝛼 = 0.95 { (3/4)
𝛼 = 0.75 𝛼=1 {1 − 6𝑥 + 6𝑥(3/2) .
𝛼 = 0.85
Here, we assume that 𝑟-cut = 1, so the right-hand side
Figure 14: The absolute error for different values 𝛼 with V = 0.95,
coefficients vector in terms of the lower and upper parametric
𝑘1 = 0.014, 𝑚 = 8, and 𝑘2 = 0.05.
form of the fuzzy number is as follows:

0.019
Using Theorem 26, we can derive the fuzzy residual 𝐺12 = 𝐺22 = (0.0017) . (90)
function for (81). Afterward, employing the inner product 1.000
explained in Section 4.3 and (81), we can generate fuzzy linear
algebraic system with (𝑚−1) equations which is in the matrix Putting (88) and (90) in (85), we reach a fuzzy linear algebraic
form as follows: equations system. Unknown coefficients vectors 𝐴 12 and 𝐴 22
are obtained by solving this system the values for 𝑟-cut = 1 are
𝐴 1 𝑇𝑚 [𝐷(V) + 𝐼] = 𝐺1 𝑇𝑚 , as
(85)
𝐴 2 𝑇𝑚 [𝐷(V) + 𝐼] = 𝐺2 𝑇𝑚 . 1.0060
𝐴 12 = 𝐴 22 = (0.0059) , (91)
0.0000
Additionally, we approximate fuzzy initial conditions (81) by
using (80) to produce the last equation which is needed to and from the earlier vector, we can attain the approximate
complete our fuzzy algebraic linear equations system. So, we fuzzy function as
have
𝐶𝐵2 1 = 𝐶𝐵2 2 = 1 + 0.0120𝑥(3/4) − 1.1301 × 10−4 𝑥(3/2) . (92)
𝐶𝐵𝑟 01 ≃ 𝐴 1 𝑇𝑚 Φ𝑚 (𝑥) = (0.5 + 0.5𝑟) ,
(86) Now, in the following results, we will reach the approxi-
𝐶𝐵𝑟 02 ≃ 𝐴 2 𝑇𝑚 Φ𝑚 (𝑥) = (1.5 − 0.5𝑟) . mate fuzzy solutions for (81) regarding different values of 𝑘1
and 𝑘2 from Table 1. Actually, we achieve the concentration
From (85) and (86), (𝑚)-fuzzy linear equations are produced of xylose after a specific time from the stated fuzzy fractional
which on can solve it easily using Definitions 8 and 9 and kinetic equation (81).
the method presented in [86] to find the unknown fuzzy In Table 2, we present the approximate solution and the
coefficients, {𝑎𝑗 }𝑚−1
𝑗=0 , of the fuzzy approximate solution. absolute error with 𝑘1 = 0.012, 𝑚 = 7, and 𝑘2 = 0.001
We solved the problem by applying the technique at 𝑇 = 1. As expected, the absolute error demonstrates the
described in Section 4. With 𝑚 = 3, V = 𝛼 = 0.75 and spectral accuracy of the proposed method. The absolute error
from Table 1, we assume that 𝑘1 = 0.012, 𝑘2 = 0.001. We for V = 𝛼 = 0.5, 075, 0.85, 0.95 is plotted in Figure 4. As we
approximate the solution in terms of the lower and upper can see, the absolute error increases gradually but remains
approximate fuzzy functions as in the interval [10−10 , 10−12 ]. This error can be explained
by the computer round-off errors that prevent any further
2 accuracy improvements. This error can be explained by the
𝑇
𝐶𝐵2 1 (𝑥; 𝑟) = ∑𝑎𝑗 𝑟1 FL0.75
𝑗 (𝑥) = [𝑎0 𝑟1 𝑎1 𝑟1 𝑎2 𝑟1 ] Φ (𝑥) , computer round-off errors that prevent any further accuracy
𝑗=0
improvements. Also, Figure 5 shows the absolute error for
(87)
2 different number of the FLFs which is clear that decreasing
𝑇
𝐶𝐵2 2 (𝑥; 𝑟) = ∑𝑎𝑗 𝑟2 FL0.75
𝑗 (𝑥) = [𝑎0 𝑟2 𝑎1 𝑟2 𝑎2 𝑟2 ] Φ (𝑥) , the the absolute error decreases with the increasing number
𝑗=0 of functions occurring. Moreover, in Figure 6, the absolute
Abstract and Applied Analysis 15

 = 𝛼 = 0.5  = 𝛼 = 0.75

Approximate solution

Approximate solution
2 1.5

1 1

0 0.5
1 1
1 0.5 1
0.5
r-cu 0. 5 r-cu 0.5
ts x ts
0 0 0 0 x

 = 𝛼 = 0.95 =𝛼=1
Approximate solution

2 1.5

Approximate solution
1 1

0 0.5
1 1
1 1
0.5 r-cu0.5
r-cu 0.5
ts
0.5
ts 0 x 0 x
0 0

Figure 15: The fuzzy approximate solution for different values of fractional orders V = 𝛼 with 𝑚 = 10, 𝑘1 = 0.05, and 𝑘2 = 0.05.

Table 2: The result of the proposed method for Example 27 with V = 𝛼 = 0.75, 𝑚 = 7, 𝑘1 = 0.012, and 𝑘2 = 0.001.

𝑟 𝐶𝐵7 1 𝐶𝐵1 (1; 𝑟) Error 𝐶𝐵7 2 𝐶𝐵2 (1; 𝑟) Error


0 0.512415005475 0.512415004378 0.109731956854𝑒 − 8 1.511327692084 1.511327690987 0.109732001263𝑒 − 8
0.1 0.562360639806 0.562360638709 0.109731967956𝑒 − 8 1.461382057754 1.461382056656 0.109731956854𝑒 − 8
0.2 0.612306274136 0.612306273039 0.109731967956𝑒 − 8 1.411436423423 1.411436422326 0.109731956854𝑒 − 8
0.3 0.662251908467 0.662251907369 0.109731967956𝑒 − 8 1.361490789093 1.361490787995 0.109731979058𝑒 − 8
0.4 0.712197542797 0.712197541700 0.109731967956𝑒 − 8 1.311545154762 1.311545153665 0.109731979058𝑒 − 8
0.5 0.762143177128 0.762143176030 0.109731979058𝑒 − 8 1.261599520432 1.261599519335 0.109731956854𝑒 − 8
0.6 0.812088811458 0.812088810361 0.109731990161𝑒 − 8 1.211653886101 1.211653885004 0.109731956854𝑒 − 8
0.7 0.862034445788 0.862034444691 0.109731956854𝑒 − 8 1.161708251771 1.161708250674 0.109731979058𝑒 − 8
0.8 0.911980080119 0.911980079022 0.109731967956𝑒 − 8 1.111762617441 1.111762616343 0.109731979058𝑒 − 8
0.9 0.961925714449 0.961925713352 0.109731967956𝑒 − 8 1.061816983110 1.061816982013 0.109731979058𝑒 − 8
1 1.011871348780 1.011871347682 0.109731934649𝑒 − 8 1.011871348780 1.011871347682 0.109731934649𝑒 − 8

error obtained by our method for V = 0.75 and 𝑚 = 7. As considerably, but the method could achieve a suitable error
one can see, the increase in the amount of 𝛼 cannot reduce in the interval [10−9 , 10−14 ]. The absolute error for V = 0.85
the absolute error noticeably. The approximated solutions are and the various choices of 𝑚 and 𝛼 by the presented method
evaluated for V = 𝛼 = 0.5, 0.75, 0.85, 0.95 and 𝑚 = 8. The are shown in Figures 9 and 10, respectively, to make it easier to
results of the numerical simulations are plotted in Figure 7. compare with the analytic solution. We show the approximate
The approximate solutions for V = 𝛼 = 0.85, 𝑚 = 9, 𝑘1 = solution in Figure 11 for V = 𝛼 = 0.5, 0.75, 0.85, 0.95 with
0.014, and 𝑘2 = 0.001 are obtained in Table 3. The obtained 𝑚 = 12.
numerical results are in very good agreement with the exact Table 4 exhibits the error of analytical and numerical
solutions. In Figure 8, the absolute errors for the problem solutions. Numerical results demonstrate the good accuracy
of V = 𝛼 = 0.5, 0.75, 0.85, 0.95 by the proposed method of the proposed method. The absolute error is evaluated
with 𝑚 = 8 are given. It is obvious that the improvement of for various choices of 𝛼 in Figure 12. We see that as V, 𝛼
the value of the V and 𝛼 could not affect the absolute error approaches 1, the solution of the fuzzy fractional kinetic
16 Abstract and Applied Analysis

Table 3: The result of the proposed method for Example 27 with V = 𝛼 = 0.85, 𝑚 = 9, 𝑘1 = 0.014, and 𝑘2 = 0.001.

𝑟 𝐶𝐵9 1 𝐶𝐵1 (1; 𝑟) Error 𝐶𝐵9 2 𝐶𝐵2 (1; 𝑟) Error


0 0.514156300808 0.514156300733 0.753808127029𝑒 − 10 1.513099430793 1.513099430717 0.753803686137𝑒 − 10
0.1 0.564103457307 0.564103457232 0.753809237252𝑒 − 10 1.463152274293 1.463152274218 0.753808127029𝑒 − 10
0.2 0.614050613806 0.614050613731 0.753809237252𝑒 − 10 1.413205117794 1.413205117719 0.753805906583𝑒 − 10
0.3 0.663997770306 0.663997770230 0.753807016806𝑒 − 10 1.363257961295 1.363257961220 0.753805906583𝑒 − 10
0.4 0.713944926805 0.713944926730 0.753809237252𝑒 − 10 1.313310804796 1.313310804720 0.753805906583𝑒 − 10
0.5 0.763892083304 0.763892083229 0.753807016806𝑒 − 10 1.263363648296 1.263363648221 0.753808127029𝑒 − 10
0.6 0.813839239803 0.813839239728 0.753808127029𝑒 − 10 1.213416491797 1.213416491722 0.753805906583𝑒 − 10
0.7 0.863786396303 0.863786396227 0.753809237252𝑒 − 10 1.163469335298 1.163469335223 0.753805906583𝑒 − 10
0.8 0.913733552802 0.913733552726 0.753808127029𝑒 − 10 1.113522178799 1.113522178723 0.753805906583𝑒 − 10
0.9 0.963680709301 0.963680709226 0.753805906583𝑒 − 10 1.063575022300 1.063575022224 0.753810347475𝑒 − 10
1 1.013627865800 1.013627865725 0.753808127029𝑒 − 10 1.013627865800 1.013627865725 0.753808127029𝑒 − 10

Table 4: The result of the proposed method for Example 27 with V = 𝛼 = 0.95, 𝑚 = 11, 𝑘1 = 0.05, and 𝑘2 = 0.05.

𝑟 𝐶𝐵11 1 𝐶𝐵1 (1; 𝑟) Error 𝐶𝐵11 2 𝐶𝐵2 (1; 𝑟) Error


0 0.523578117777 0.523578117838 0.615489881283𝑒 − 10 1.473894867831 1.473894867893 0.614888140404𝑒 − 10
0.1 0.571093955279 0.571093955341 0.615463235931𝑒 − 10 1.426379030328 1.426379030390 0.614917006203𝑒 − 10
0.2 0.713641467788 0.618609792844 0.615431039463𝑒 − 10 1.331347355323 1.378863192887 0.614950312893𝑒 − 10
0.3 0.761157305290 0.666125630346 0.615399953218𝑒 − 10 1.283831517820 1.331347355384 0.614972517354𝑒 − 10
0.4 0.713944926805 0.713641467849 0.615371087420𝑒 − 10 1.236315680317 1.283831517882 0.615005824045𝑒 − 10
0.5 0.763892083304 0.761157305352 0.615342221621𝑒 − 10 1.188799842815 1.236315680379 0.615039130735𝑒 − 10
0.6 0.808673142793 0.808673142854 0.615311135376𝑒 − 10 1.213416491797 1.188799842876 0.615067996534𝑒 − 10
0.7 0.856188980296 0.856188980357 0.615278938909𝑒 − 10 1.141284005312 1.141284005374 0.615099082779𝑒 − 10
0.8 0.903704817798 0.903704817860 0.615252293556𝑒 − 10 1.093768167809 1.093768167871 0.615132389469𝑒 − 10
0.9 0.951220655301 0.951220655363 0.615220097088𝑒 − 10 1.046252330307 1.046252330368 0.615156814376𝑒 − 10
1 0.998736492804 0.998736492865 0.615190121067𝑒 − 10 0.998736492804 0.998736492865 0.615190121067𝑒 − 10

equation approaches that of the integer order fuzzy kinetic differential equations like fuzzy fractional oscillation differ-
equations. In Figure 13, again we see that the method could ential equations of the distributed order.
reach a high accuracy for the approximate solution, but the
increase in the number of the FLFS could not affect the
References
absolute error considerably. Also, in Figure 14, the method
could hit the better accuracy when the values of the fractional [1] A. Rodrı́guez-Chong, J. A. Ramı́rez, G. Garrote, and M.
derivative and fractional order of the Legendre functions Vázquez, “Hydrolysis of sugar cane bagasse using nitric acid:
were the same. In Figure 15, we show the fuzzy approximate a kinetic assessment,” Journal of Food Engineering, vol. 61, no. 2,
solutions which were evaluated for various values of V = 𝛼. pp. 143–152, 2004.
[2] Y. Lu and N. S. Mosier, “Kinetic modeling analysis of maleic
acid-catalyzed hemicellulose hydrolysis in corn stover,” Biotech-
nology and Bioengineering, vol. 101, no. 6, pp. 1170–1181, 2008.
6. Conclusion [3] Y. Sun, X. Lu, S. Zhang, R. Zhang, and X. Wang, “Kinetic study
for Fe(NO3 )3 catalyzed hemicellulose hydrolysis of different
In this paper, a spectral method has been adapted for numer- corn stover silages,” Bioresource Technology, vol. 102, no. 3, pp.
ical solution of fuzzy fractional kinetic equation of order 0 < 2936–2942, 2011.
V ≤ 1. A specific family of the orthogonal polynomials which [4] J. F. Saeman, “Kinetics of wood Saccharification-Hydrolysis of
is so-called FLFs was used as the approximation basis. The cellulose and decomposition of sugars in dilute acid at high
proposed method is characterized by its simplicity, efficiency, temperature,” Industrial and Engineering Chemistry, vol. 37, pp.
and high accuracy. The accuracy and validity of the presented 43–52, 1945.
method were demonstrated through numerical simulation [5] W. Faith, “Development of the Scholler Process in The United
for the derived FFKE. Using the proposed method, we States,” Industrial & Engineering Chemistry, vol. 37, pp. 9–11,
could reach the suitable approximation of the amount of the 1945.
concentration value of xylose after a determined time that is [6] M. Neureiter, H. Danner, C. Thomasser, B. Saidi, and R.
important to analyze the kinetic data in the chemical process. Braun, “Dilute-acid hydrohlysis of sugarcane bagasse at varying
A direction for further research would be to extend the conditions,” Applied Biochemistry and Biotechnology A, vol. 98–
presented method to the other types of the fuzzy fractional 100, pp. 49–58, 2002.
Abstract and Applied Analysis 17

[7] J. S. Kim, Y. Y. Lee, and R. W. Torget, “Cellulose hydrolysis under [25] K. Diethelm, The Analysis of Fractional Differential Equations.
extremely low sulfuric acid and high-temperature conditions,” An Application-Oriented Exposition Using Differential Operators
Applied Biochemistry and Biotechnology A, pp. 91–93, 331–340, of Caputo Type, vol. 2004 of Lecture Notes in Mathematics,
2001. Springer, Berlin, Germany, 2010.
[8] M. Caputo, “Linear model of dissipation whose Q is almost [26] R. Gorenflo and F. Mainardi, “Fractional calculus: integral
frequency independent, II, Geophys,” Journal of the Royal and differential equations of fractional order,” in Fractals and
Astronomical Society, vol. 13, pp. 529–539, 1967. Fractional Calculus in Continuum Mechanics (Udine, 1996), A.
[9] D. D. Clayton, Principles of Stellar Evolution and Nucleosynthe- Carpinteri and F. Mainardi, Eds., vol. 378 of CISM Courses and
sis, The University of Chicago Press, Chicago, Ill, USA, 2nd Lectures, pp. 223–276, Springer, Vienna, Austria, 1997.
edition, 1983. [27] A. Kadem and D. Baleanu, “Fractional radiative transfer
[10] W. G. Glöckle and T. F. Nonnenmacher, “Fox function represen- equation within Chebyshev spectral approach,” Computers &
tation of non-Debye relaxation processes,” Journal of Statistical Mathematics with Applications, vol. 59, no. 5, pp. 1865–1873,
Physics, vol. 71, no. 3-4, pp. 741–757, 1993. 2010.
[11] R. Hilfer, “Fractional time evolution,” in Applications of Frac- [28] K. S. Miller and B. Ross, An Introduction to the Fractional Calcu-
tional Calculus in Physics, pp. 87–130, World Scientific Publish- lus and Fractional Differential Equations, A Wiley-Interscience
ing, River Edge, NJ, USA, 2000. Publication, John Wiley & Sons, New York, NY, USA, 1993.
[12] V. L. Kobelev, E. P. Romanov, L. Y. Kobelev, and Y. L. Kobelev, [29] K. B. Oldham and J. Spanier, The Fractional Calculus, Academic
“Relaxational and diffusive processes in fractal space,” Izvestiya Press, New York, NY, USA, 1974.
Akademii Nauk. Seriya Fizicheskaya, vol. 62, no. 12, pp. 2401– [30] I. Podlubny, Fractional Differential Equations, vol. 198 of Math-
2408, 1998. ematics in Science and Engineering, Academic Press, San Diego,
[13] F. Mainardi, “Fractional diffusive waves in viscoelastic solids,” Calif, USA, 1999.
in Nonlinear Waves in Solids, J. L. Wegner and F. R. Norwood, [31] A. N. Kochubeı̆, “The Cauchy problem for evolution equations
Eds., pp. 93–97, ASME, New York, NY, USA, 1995. of fractional order,” Journal of Differential Equations, vol. 25, pp.
[14] F. Mainardi, “Fractal calculus: some basic problems in contin- 967–974, 1989.
uum and statistical mechanics,” in Fractals and Fractional Cal-
[32] H. J. Haubold and A. M. Mathai, “The fractional kinetic
culus in Continuum Mechanics, A. Carpinteri and F. Mainardi,
equation and thermonuclear functions,” Astrophysics and Space
Eds., pp. 291–348, Springer, Wien, Austria, 1997.
Science, vol. 273, no. 1–4, pp. 53–63, 2000.
[15] F. Mainardi and M. Tomirotti, “Seismic pulse propagation
[33] R. K. Saxena, A. M. Mathai, and H. J. Haubold, “On fractional
with constant Q and stable probability distributions,” Annali di
kinetic equations,” Astrophysics and Space Science, vol. 282, no.
Geofisica, vol. 40, no. 5, pp. 1311–1325, 1997.
1, pp. 281–287, 2002.
[16] R. Metzler, W. G. Glöckle, and T. F. Nonnenmacher, “Fractional
model equation for anomalous diffusion,” Physica A, vol. 211, no. [34] R. K. Saxena, A. M. Mathai, and H. J. Haubold, “Unified
1, pp. 13–24, 1994. fractional kinetic equation and a fractional diffusion equation,”
Astrophysics and Space Science, vol. 290, no. 3-4, pp. 299–310,
[17] A. I. Saichev and G. M. Zaslavsky, “Fractional kinetic equations:
2004.
solutions and applications,” Chaos, vol. 7, no. 4, pp. 753–764,
1997. [35] A. Atangana and A. Kilicman, “Analytical solutions of the space-
[18] G. M. Zaslavsky, “Fractional kinetics of Hamiltonian chaotic time fractional derivative of advection dispersion equation,”
systems,” in Applications of Fractional Calculus in Physics, pp. Mathematical Problems in Engineering, vol. 2013, Article ID
203–239, World Scientific Publishing, River Edge, NJ, USA, 853127, 9 pages, 2013.
2000. [36] A. M. A. El-Sayed, I. L. El-Kalla, and E. A. A. Ziada, “Analytical
[19] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional and numerical solutions of multi-term nonlinear fractional
Calculus. Models and Numerical Methods, vol. 3 of Series on orders differential equations,” Applied Numerical Mathematics,
Complexity, Nonlinearity and Chaos, World Scientific Publish- vol. 60, no. 8, pp. 788–797, 2010.
ing, Hackensack, NJ, USA, 2012. [37] H. Jiang, F. Liu, I. Turner, and K. Burrage, “Analytical solu-
[20] D. Baleanu, Z. B. Güvenc, and J. A. Tenreiro Machado, New tions for the multi-term time-space Caputo-Riesz fractional
Trends in Nanotechnology and Fractional Calculus Applications, advection-diffusion equations on a finite domain,” Journal of
Springer, New York, NY, USA, 2010. Mathematical Analysis and Applications, vol. 389, no. 2, pp. 1117–
[21] D. Baleanu and J. I. Trujillo, “A new method of finding 1127, 2012.
the fractional Euler-Lagrange and Hamilton equations within [38] Z. Odibat, S. Momani, and H. Xu, “A reliable algorithm of
Caputo fractional derivatives,” Communications in Nonlinear homotopy analysis method for solving nonlinear fractional
Science and Numerical Simulation, vol. 15, no. 5, pp. 1111–1115, differential equations,” Applied Mathematical Modelling, vol. 34,
2010. no. 3, pp. 593–600, 2010.
[22] P. L. Butzer and U. Westphal, “An introduction to fractional [39] N. H. Sweilam and M. M. Khader, “Exact solutions of some
calculus,” in Applications of Fractional Calculus in Physics, pp. coupled nonlinear partial differential equations using the
1–85, World Scientific Publishing, River Edge, NJ, USA, 2000. homotopy perturbation method,” Computers & Mathematics
[23] M. M. Djrbashian, Harmonic Analysis and Boundary Value with Applications, vol. 58, no. 11-12, pp. 2134–2141, 2009.
Problems in the Complex Domain, vol. 65 of Operator Theory: [40] M. ur Rehman and R. A. Khan, “A numerical method for solving
Advances and Applications, Birkhäuser, Basel, Switzerland, 1993. boundary value problems for fractional differential equations,”
[24] K. Diethelm, N. J. Ford, A. D. Freed, and Yu. Luchko, “Algo- Applied Mathematical Modelling, vol. 36, no. 3, pp. 894–907,
rithms for the fractional calculus: a selection of numerical 2012.
methods,” Computer Methods in Applied Mechanics and Engi- [41] A. Pedas and E. Tamme, “Spline collocation methods for
neering, vol. 194, no. 6–8, pp. 743–773, 2005. linear multi-term fractional differential equations,” Journal of
18 Abstract and Applied Analysis

Computational and Applied Mathematics, vol. 236, no. 2, pp. Applied Mathematical Modelling, vol. 37, no. 3, pp. 1126–1136,
167–176, 2011. 2013.
[42] S. Momani and Z. Odibat, “Numerical comparison of methods [59] A. H. Bhrawy, M. M. Alghamdi, and T. M. Taha, “A new
for solving linear differential equations of fractional order,” modified generalized Laguerre operational matrix of fractional
Chaos, Solitons & Fractals, vol. 31, no. 5, pp. 1248–1255, 2007. integration for solving fractional differential equations on the
[43] Z. M. Odibat and N. T. Shawagfeh, “Generalized Taylor’s half line,” Advances in Difference Equations, vol. 2012, article 179,
formula,” Applied Mathematics and Computation, vol. 186, no. 2012.
1, pp. 286–293, 2007. [60] A. H. Bhrawy, A. S. Alofi, and S. S. Ezz-Eldien, “A quadrature
[44] N. T. Shawagfeh, “Analytical approximate solutions for nonlin- tau method for fractional differential equations with variable
ear fractional differential equations,” Applied Mathematics and coefficients,” Applied Mathematics Letters, vol. 24, no. 12, pp.
Computation, vol. 131, no. 2-3, pp. 517–529, 2002. 2146–2152, 2011.
[45] A. Kilicman and Z. A. A. Al Zhour, “Kronecker operational [61] S. Esmaeili and M. Shamsi, “A pseudo-spectral scheme for the
matrices for fractional calculus and some applications,” Applied approximate solution of a family of fractional differential equa-
Mathematics and Computation, vol. 187, no. 1, pp. 250–265, 2007. tions,” Communications in Nonlinear Science and Numerical
[46] A. Kadem, Y. Luchko, and D. Baleanu, “Spectral method Simulation, vol. 16, no. 9, pp. 3646–3654, 2011.
for solution of the fractional transport equation,” Reports on [62] S. Abbasbandy, T. A. Viranloo, Ó. López-Pouso, and J. J.
Mathematical Physics, vol. 66, no. 1, pp. 103–115, 2010. Nieto, “Numerical methods for fuzzy differential inclusions,”
[47] S. Abbasbandy, “An approximation solution of a nonlinear Computers & Mathematics with Applications, vol. 48, no. 10-11,
equation with Riemann-Liouville’s fractional derivatives by He’s pp. 1633–1641, 2004.
variational iteration method,” Journal of Computational and [63] S. Abbasbandy, E. Babolian, and M. Allame, “Numerical solu-
Applied Mathematics, vol. 207, no. 1, pp. 53–58, 2007. tion of fuzzy max-min systems,” Applied Mathematics and
[48] S. Abbasbandy and A. Shirzadi, “Homotopy analysis method for Computation, vol. 174, no. 2, pp. 1321–1328, 2006.
multiple solutions of the fractional Sturm-Liouville problems,” [64] T. Allahviranloo, N. Ahmady, and E. Ahmady, “Numerical
Numerical Algorithms, vol. 54, no. 4, pp. 521–532, 2010. solution of fuzzy differential equations by predictor-corrector
[49] C. Hwang and Y. P. Shih, “Laguerre operational matrices for method,” Information Sciences, vol. 177, no. 7, pp. 1633–1647,
fractional calculus and applications,” International Journal of 2007.
Control, vol. 34, no. 3, pp. 577–584, 1981. [65] T. Allahviranloo, S. Abbasbandy, S. Salahshour, and A.
[50] G. Maione, “A digital, noninteger order, differentiator using Hakimzadeh, “A new method for solving fuzzy linear differen-
laguerre orthogonal sequences,” International Journal of Intel- tial equations,” Computing, vol. 92, no. 2, pp. 181–197, 2011.
ligent Systems, vol. 11, pp. 77–81, 2006. [66] E. Babolian, H. Sadeghi Goghary, and S. Abbasbandy, “Numer-
[51] Y. Li and N. Sun, “Numerical solution of fractional differential ical solution of linear Fredholm fuzzy integral equations of the
equations using the generalized block pulse operational matrix,” second kind by Adomian method,” Applied Mathematics and
Computers & Mathematics with Applications, vol. 62, no. 3, pp. Computation, vol. 161, no. 3, pp. 733–744, 2005.
1046–1054, 2011. [67] B. Bede and S. G. Gal, “Generalizations of the differentiability
[52] S. Kazem, S. Abbasbandy, and S. Kumar, “Fractional-order of fuzzy-number-valued functions with applications to fuzzy
Legendre functions for solving fractional order differential differential equations,” Fuzzy Sets and Systems, vol. 151, no. 3,
equations,” Applied Mathematical Modelling, vol. 37, pp. 5498– pp. 581–599, 2005.
5510, 2013. [68] A. Bencsik, B. Bede, J. Tar, and J. Fodor, “Fuzzy differential
[53] A. Saadatmandi and M. Dehghan, “A new operational matrix equations in modeling hydraulic differential servo cylinders,”
for solving fractional-order differential equations,” Computers in Proceedings of the 3rd Romanian-Hungarian Joint Symposium
& Mathematics with Applications, vol. 59, no. 3, pp. 1326–1336, on Applied Computational Intellidence (SACI ’06), Timisoara,
2010. Romania, 2006.
[54] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A Chebyshev [69] Y. Chalco-Cano and H. Román-Flores, “On new solutions of
spectral method based on operational matrix for initial and fuzzy differential equations,” Chaos, Solitons and Fractals, vol.
boundary value problems of fractional order,” Computers & 38, no. 1, pp. 112–119, 2008.
Mathematics with Applications, vol. 62, no. 5, pp. 2364–2373, [70] S. Salahshour and T. Allahviranloo, “Applications of fuzzy
2011. Laplace transforms,” Soft Computing, pp. 1–14, 2012.
[55] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “Efficient [71] R. P. Agarwal, V. Lakshmikantham, and J. J. Nieto, “On the
Chebyshev spectral methods for solving multi-term fractional concept of solution for fractional differential equations with
orders differential equations,” Applied Mathematical Modelling, uncertainty,” Nonlinear Analysis. Theory, Methods & Applica-
vol. 35, no. 12, pp. 5662–5672, 2011. tions, vol. 72, no. 6, pp. 2859–2862, 2010.
[56] M. M. Khader, T. S. El Danaf, and A. S. Hendy, “A computational [72] T. Allahviranloo, S. Salahshour, and S. Abbasbandy, “Explicit
matrix method for solving systems of high order fractional solutions of fractional differential equations with uncertainty,”
differential equations,” Applied Mathematical Modelling, vol. 37, Soft Computing, vol. 16, no. 2, pp. 297–302, 2012.
pp. 4035–4050, 2013. [73] T. Allahviranloo, Z. Gouyandeh, and A. Armand, “Fuzzy frac-
[57] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A new tional differential equations under generalized fuzzy Caputo
Jacobi operational matrix: an application for solving fractional derivative,” Journal of Intelligent and Fuzzy Systems, 2013.
differential equations,” Applied Mathematical Modelling, vol. 36, [74] S. Salahshour, T. Allahviranloo, S. Abbasbandy, and D. Baleanu,
no. 10, pp. 4931–4943, 2012. “Existence and uniqueness results for fractional differential
[58] S. Kazem, “An integral operational matrix based on Jacobi equations with uncertainty,” Advances in Difference Equations,
polynomials for solving fractional-order differential equations,” vol. 2012, article 112, 2012.
Abstract and Applied Analysis 19

[75] S. Salahshour, T. Allahviranloo, and S. Abbasbandy, “Solving reactor,” in Proceedings of International Chemical, Biological and
fuzzy fractional differential equations by fuzzy Laplace trans- Environmental Engineering (IPCBEE ’12), vol. 38, IACSIT Press,
forms,” Communications in Nonlinear Science and Numerical 2012.
Simulation, vol. 17, no. 3, pp. 1372–1381, 2012. [96] S. Abbasbandy and M. Amirfakhrian, “A new approach to
[76] A. Ahmadian, M. Suleiman, S. Salahshour, and D. Baleanu, “A universal approximation of fuzzy functions on a discrete set
Jacobi operational matrix for solving a fuzzy linear fractional of points,” Applied Mathematical Modelling, vol. 30, no. 12, pp.
differential equation,” Advances in Difference Equations, vol. 1525–1534, 2006.
2013, article 104, 2013. [97] S. Abbasbandy and M. Amirfakhrian, “Best approximation of
[77] A. Ahmadian, M. Suleiman, and S. Salahshour, “An opera- fuzzy functions,” Journal of Nonlinear Studies, vol. 14, pp. 88–
tional matrix based on legendre polynomials for solving fuzzy 103, 2007.
fractional-order differential equations,” Abstract and Applied [98] G. A. Anastassiou, “Fuzzy approximation by fuzzy convolution
Analysis, vol. 2013, Article ID 505903, 29 pages, 2013. type operators,” Computers & Mathematics with Applications,
[78] M. Mazandarani and A. V. Kamyad, “Modified fractional Euler vol. 48, no. 9, pp. 1369–1386, 2004.
method for solving fuzzy fractional initial value problem,” Com- [99] H. Huang and C. Wu, “Approximation of fuzzy functions by
munications in Nonlinear Science and Numerical Simulation, vol. regular fuzzy neural networks,” Fuzzy Sets and Systems, vol. 177,
18, no. 1, pp. 12–21, 2013. pp. 60–79, 2011.
[79] P. Diamond and P. Kloeden, Metric Spaces of Fuzzy Sets. Theory [100] M. Dehghan and A. Saadatmandi, “A tau method for the one-
and Applications, World Scientific Publishing, River Edge, NJ, dimensional parabolic inverse problem subject to temperature
USA, 1994. overspecification,” Computers & Mathematics with Applications,
[80] D. Dubois and H. Prade, “Towards fuzzy differential calculus vol. 52, no. 6-7, pp. 933–940, 2006.
part 3: differentiation,” Fuzzy Sets and Systems, vol. 8, no. 3, pp. [101] J. H. Freilich and E. L. Ortiz, “Numerical solution of systems of
225–233, 1982. ordinary differential equations with the Tau method: an error
[81] R. Goetschel, Jr. and W. Voxman, “Elementary fuzzy calculus,” analysis,” Mathematics of Computation, vol. 39, no. 160, pp. 467–
Fuzzy Sets and Systems, vol. 18, no. 1, pp. 31–43, 1986. 479, 1982.
[82] H.-J. Zimmermann, Fuzzy Set Theory—And Its Applications,
Kluwer Academic Publishers, Boston, Mass, USA, 2nd edition,
1992.
[83] G. A. Anastassiou and S. G. Gal, “On a fuzzy trigonometric
approximation theorem of Weierstrass-type,” Journal of Fuzzy
Mathematics, vol. 9, no. 3, pp. 701–708, 2001.
[84] G. A. Anastassiou, Fuzzy Mathematics: Approximation Theory,
vol. 251 of Studies in Fuzziness and Soft Computing, Springer,
Berlin, Germany, 2010.
[85] O. Kaleva, “Fuzzy differential equations,” Fuzzy Sets and Sys-
tems, vol. 24, no. 3, pp. 301–317, 1987.
[86] T. Allahviranloo and M. Afshar Kermani, “Solution of a fuzzy
system of linear equation,” Applied Mathematics and Computa-
tion, vol. 175, no. 1, pp. 519–531, 2006.
[87] M. Friedman, M. Ming, and A. Kandel, “Fuzzy linear systems,”
Fuzzy Sets and Systems, vol. 96, no. 2, pp. 201–209, 1998.
[88] L. F. Wilhelmy, “ber das Gesetz, nach welchem die Einwirkung
der Suren auf Rohrzucker stattfindet,” Poggendorff ’s Annalen
der Physik und Chemie, vol. 81, pp. 413–433, 499–526, 1850.
[89] P. Waage and C. M. Guldberg, “Studier over affiniteten forhan-
dlinger,” VIdenskabs-Selskabet I ChrIstIana, vol. 35–40, pp. 111–
120, 1864.
[90] K. J. Laidler, The World of Physical Chemistry, Oxford University
Press, Oxford, UK, 1993.
[91] A. Cornish-Bowden, Fundamentals of Enzyme Kinetics, John
Wiley & Sons, New York, NY, USA, 4th edition, 2012.
[92] M. Robson Wright, An Introduction to Chemical Kinetics, John
Wiley & Sons, New York, NY, USA, 2004.
[93] F. Mainardi, “Fractional relaxation-oscillation and fractional
diffusion-wave phenomena,” Chaos, Solitons and Fractals, vol.
7, no. 9, pp. 1461–1477, 1996.
[94] P. Humbert and R. P. Agarwal, “Sur la fonction de Mittag-Leffler
et quelques-unes de ses généralisations,” Bulletin des Sciences
Mathématiques. 2e Série, vol. 77, pp. 180–185, 1953.
[95] S. Faridah Salleh, R. Yunus, M. Farid Atan, and D. R. Awg
Biak, “Kinetic studies on acid hydrolysis of OPEFB in a batch
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 413529, 12 pages
http://dx.doi.org/10.1155/2013/413529

Research Article
A Modified Generalized Laguerre Spectral Method for
Fractional Differential Equations on the Half Line

D. Baleanu,1,2,3 A. H. Bhrawy,4,5 and T. M. Taha5


1
Department of Mathematics and Computer Sciences, Faculty of Arts and Sciences, Cankaya University,
Eskisehir Yolu 29.Km, 06810 Ankara, Turkey
2
Department of Chemical and Materials Engineering, Faculty of Engineering, King Abdulaziz University,
Jeddah 21589, Saudi Arabia
3
Institute of Space Sciences, Magurele-Bucharest, Romania
4
Department of Mathematics, Faculty of Science, King Abdulaziz University, Jeddah 21589, Saudi Arabia
5
Department of Mathematics, Faculty of Science, Beni-Suef University, Beni-Suef 62511, Egypt

Correspondence should be addressed to T. M. Taha; taha m taha@yahoo.com

Received 21 May 2013; Accepted 7 July 2013

Academic Editor: Soheil Salahshour

Copyright © 2013 D. Baleanu et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper deals with modified generalized Laguerre spectral tau and collocation methods for solving linear and nonlinear
multiterm fractional differential equations (FDEs) on the half line. A new formula expressing the Caputo fractional derivatives
of modified generalized Laguerre polynomials of any degree and for any fractional order in terms of the modified generalized
Laguerre polynomials themselves is derived. An efficient direct solver technique is proposed for solving the linear multiterm FDEs
with constant coefficients on the half line using a modified generalized Laguerre tau method. The spatial approximation with its
(𝛼,𝛽)
Caputo fractional derivatives is based on modified generalized Laguerre polynomials 𝐿 𝑖 (𝑥) with 𝑥 ∈ Λ = (0, ∞), 𝛼 > −1, and
𝛽 > 0, and 𝑖 is the polynomial degree. We implement and develop the modified generalized Laguerre collocation method based on
the modified generalized Laguerre-Gauss points which is used as collocation nodes for solving nonlinear multiterm FDEs on the
half line.

1. Introduction generalization can be expressed by the Bell polynomials, see


[13–15].
Due to the fact that fractional calculus has extensive appli- Because of the difficulty in obtaining exact solutions to
cations in chemistry, engineering, neuron modeling, and many fractional differential equations, there is a large number
biological sciences, the use of fractional calculus has attracted of papers dealing with the numerical solution of fractional
great interest from the mathematical science research com- differential equation [16–21]. The usual spectral methods
munity (e.g., see [1–5]). based on Legendre, Chebyshev, and Jacobi polynomials
As is well known, spectral methods use orthogonal poly- are only available for bounded domains for the numerical
nomials as the basis functions and so usually provide accurate solution of FDEs [22, 23]. In [22, 24, 25], the authors discussed
numerical results, see [6–9] and the references therein. In the operational matrices of Caputo fractional derivatives of
particular, we could employ tau and pseudospectral modified shifted Chebyshev and Jacobi polynomials with spectral tech-
generalized Laguerre approximations and interpolation to niques to provide numerical approximations to the solution
solve multiterm fractional differential equations on semi- of linear and nonlinear multiterm FDEs on finite intervals.
infinite interval. However, some authors developed the modi- Ahmadian et al. [26] applied the Jacobi operational matrix
fied generalized Laguerre spectral method for the half line for to present an efficient algorithm for solving a class of fuzzy
ordinary, partial, and delay differential equations, see [8, 10– FDEs. Li et al. [27] employed spectral approximations to
12]. The Laguerre polynomials of fractional orders and its compute the fractional integral and the Caputo derivative;
2 Abstract and Applied Analysis

they also presented pseudospectral approximation for a class equipped with the following inner product and norm:
of FDEs. Yüzbaşı [28] proposed Bessel collocation method
for the approximate solution of the Bagley-Torvik equation. (𝑢, V)𝑤(𝛼,𝛽) = ∫ 𝑢 (𝑥) V (𝑥) 𝑤(𝛼,𝛽) (𝑥) 𝑑𝑥,
Atabakzadeh et al. [29] extended the application of the shifted Λ
(2)
Chebyshev operational matrix for obtaining an analytical
approximation solution for linear and nonlinear multiorder ‖V‖𝑤(𝛼,𝛽) = (V, V)1/2
𝑤(𝛼,𝛽)
.
FDE. Moreover, the authors of [30] extended the application
(𝛼,𝛽)
of the Legendre operational matrix for treating the fractional Next, let 𝐿 𝑖 (𝑥) be the modified generalized Laguerre
order fuzzy differential equations. Recently, Bhrawy et al. [31, polynomial of degree 𝑖 for 𝛼 > −1 and 𝛽 > 0, that is defined
32] proposed the operational matrices of Riemann-Liouville by
fractional integration of Chebyshev and modified generalized
Laguerre polynomials which are employed together with (𝛼,𝛽) 1 −𝛼 𝛽𝑥 𝑖 𝑖+𝛼 −𝛽𝑥
𝐿𝑖 (𝑥) = 𝑥 𝑒 𝜕𝑥 (𝑥 𝑒 ) , 𝑖 = 1, 2, . . . . (3)
spectral tau method for solving linear FDEs on finite and 𝑖!
semi-infinite intervals, respectively.
The fundamental aim of this paper is to extend the According to (2.3)-(2.4) of [11] for 𝛼 > −1 and 𝛽 > 0, we have
application of spectral tau and collocation methods based
(𝛼,𝛽) (𝛼+1,𝛽)
on modified generalized Laguerre polynomials to solve the 𝜕𝑥 𝐿 𝑖 (𝑥) = −𝛽𝐿 𝑖−1 (𝑥) , (4)
linear and nonlinear multiterm fractional initial value prob-
(𝛼,𝛽) 1 (𝛼,𝛽)
lems on the half line. We propose the modified generalized 𝐿 𝑖+1 (𝑥) = [(2𝑖 + 𝛼 + 1 − 𝛽𝑥) 𝐿 𝑖 (𝑥)
Laguerre spectral tau (MGLT) approximation for obtaining 𝑖+1
a direct solution technique to solve linear multiterm FDE (𝛼,𝛽)
− (𝑖 + 𝛼) 𝐿 𝑖−1 (𝑥)] , 𝑖 = 1, 2, . . . ,
on the half line. This method requires the Caputo fractional
derivatives of any fractional order of the modified generalized (5)
Laguerre polynomials of any degree which is already stated
(𝛼,𝛽) (𝛼,𝛽)
and proved. where 𝐿 0 (𝑥) = 1 and 𝐿 1 (𝑥) = −𝛽𝑥 + Γ(𝛼 + 2)/Γ(𝛼 + 1).
The modified generalized Laguerre spectral collocation The set of modified generalized Laguerre polynomials is
(MGLC) approximation, which is more reliable, is employed the 𝐿2𝑤(𝛼,𝛽) (Λ)-orthogonal system, namely,
to obtain approximate solution of nonlinear multiterm FDE

on the half line with leading fractional differential operator (𝛼,𝛽) (𝛼,𝛽)
of order ] (𝑚 − 1 < ] ≤ 𝑚) and 𝑚 initial conditions.
∫ 𝐿𝑗 (𝑥) 𝐿 𝑘 (𝑥) 𝑤(𝛼,𝛽) (𝑥) 𝑑𝑥 = ℎ𝑘 𝛿𝑗𝑘 , (6)
0
To be more precise, in such approximation, the nonlinear
FDE is collocated only at (𝑁 − 𝑚 + 1) zeros of modified where 𝛿𝑗𝑘 is the Kronecker function and ℎ𝑘 = Γ(𝑘 + 𝛼 +
generalized Laguerre polynomials. Therefore, the problem 1)/𝛽𝛼+1 𝑘!.
reduces to a system of (𝑁 + 1) nonlinear algebraic equations The modified generalized Laguerre polynomials of degree
which may be solved by any standard technique to find the 𝑖 on the interval Λ, are given by
spectral solution 𝑢𝑁(𝑥). To the best of the our knowledge,
𝑖
the proposed algorithms have not been introduced for the
(𝛼,𝛽) Γ (𝑖 + 𝛼 + 1) 𝛽𝑘
numerical solution of multiterm FDEs on the half line. 𝐿𝑖 (𝑥) = ∑ (−1)𝑘 𝑥𝑘 ,
𝑘=0
Γ (𝑘 + 𝛼 + 1) (𝑖 − 𝑘)!𝑘! (7)
The plan of the paper is as follows. In the next section, we
introduce basic properties of modified generalized Laguerre
𝑖 = 0, 1, . . . ,
polynomials. In Section 3, the Caputo fractional derivative of
the modified generalized Laguerre polynomials is proved. In (𝛼,𝛽)
Section 4, we develop modified generalized Laguerre tau and where 𝐿 𝑖 (0) = Γ(𝑖 + 𝛼 + 1)/Γ(𝛼 + 1)Γ(𝑖 + 1).
collocation spectral methods for solving multiterm FDEs. In The special value
Section 5, several numerical examples are implemented. Also,
(𝛼,𝛽) (−1)𝑞 𝛽𝑞 Γ (𝑖 + 𝛼 + 1)
a conclusion is given in Section 6. 𝐷𝑞 𝐿 𝑖 (0) = , 𝑖 ⩾ 𝑞, (8)
(𝑖 − 𝑞)!Γ (𝑞 + 𝛼 + 1)

2. Properties of Modified Generalized will be of important use later.


Laguerre Polynomials A function 𝑢(𝑥) ∈ 𝐿2𝑤(𝛼,𝛽) (Λ) may be expressed in terms
of modified generalized Laguerre polynomials as
We recall below some relevant properties of the modified
generalized Laguerre polynomials (see [8, 33, 34]). ∞
(𝛼,𝛽)
Now, let Λ = (0, ∞) and 𝑤(𝛼,𝛽) (𝑥) = 𝑥𝛼 𝑒−𝛽𝑥 be a weight 𝑢 (𝑥) = ∑ 𝑎𝑗 𝐿 𝑗 (𝑥) ,
𝑗=0
function on Λ in the usual sense. Define
1 ∞ (𝛼,𝛽) (9)
𝑎𝑗 = ∫ 𝑢 (𝑥) 𝐿 𝑗 (𝑥) 𝑤(𝛼,𝛽) (𝑥) 𝑑𝑥,
𝐿2𝑤(𝛼,𝛽) (Λ) = {V | V is measurable on Λ ℎ𝑗 0
(1)
and ‖V‖𝑤(𝛼,𝛽) < ∞} , 𝑗 = 0, 1, 2, . . . .
Abstract and Applied Analysis 3

In particular applications, the modified generalized Theorem 2. The Caputo fractional derivative of order ] for
Laguerre polynomials up to degree 𝑁 + 1 are considered. modified generalized Laguerre polynomials is given by
Then, we have
𝑁
(𝛼,𝛽) (𝛼,𝛽)
𝑁 𝐷] 𝐿 𝑖 (𝑥) = ∑ Ψ] (𝑖, 𝑗) 𝐿 𝑗 (𝑥) , 𝑖 = ⌈]⌉ , . . . , 𝑁,
(𝛼,𝛽)
𝑢𝑁 (𝑥) = ∑𝑎𝑗 𝐿 𝑗 (𝑥) . (10) 𝑗=0
𝑗=0 (16)

We now turn the modified generalized Laguerre-Gausstype where


quadratures, including modified generalized Laguerre-Gauss
(𝛼,𝛽) Ψ] (𝑖, 𝑗)
and Gauss-Radau interpolations [11, 34, 35]. Let 𝜉𝐺,𝑖,𝑗 and
(𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) 𝑗
𝜉𝑅,𝑖,𝑗 , 0 ≤ 𝑗 ≤ 𝑖, be the zeros of 𝐿 𝑖+1 (𝑥) and 𝑥𝜕𝑥 𝐿 𝑖+1 (𝑥),
𝑖
(−1)𝑘+ℓ 𝛽] 𝑗!Γ (𝑖 + 𝛼 + 1) Γ (𝑘 − ] + 𝛼 + ℓ + 1)
= ∑ ∑ .
respectively. They are arranged in ascending order. Denote 𝑘=⌈]⌉ ℓ=0 (𝑖−𝑘)!
(𝑗 − ℓ)! ℓ! Γ (𝑘 − ]+1) Γ (𝑘+𝛼+1) Γ (𝛼+ℓ+1)
(𝛼,𝛽)
𝜔𝑍,𝑖,𝑗 , (0 ≤ 𝑗 ≤ 𝑖), 𝑍 = 𝐺, 𝑅, the corresponding Christoffel (17)
numbers such that
Proof. With the aid of (7), the ]-order derivative of modified
𝑖 generalized Laguerre polynomials is
(𝛼,𝛽) (𝛼,𝛽)
∫ 𝜙 (𝑥) 𝑤(𝛼,𝛽) (𝑥) 𝑑𝑥 = ∑ 𝜙 (𝜉𝑍,𝑖,𝑗 ) 𝜔𝑍,𝑖,𝑗 . (11)
Λ 𝑗=0 (𝛼,𝛽)
𝐷] 𝐿 𝑖 (𝑥)
𝑖
For the modified generalized Laguerre-Gauss weights, we 𝛽𝑘 Γ (𝑖 + 𝛼 + 1)
have = ∑ (−1)𝑘 𝐷] 𝑥 𝑘
𝑘=0
(𝑖 − 𝑘)! 𝑘! Γ (𝑘 + 𝛼 + 1)
(𝛼,𝛽) Γ (𝑖 + 𝛼 + 2) 1
𝜔𝐺,𝑖,𝑗 = 𝛼 , 0 ≤ 𝑗 ≤ 𝑖. 𝑖
𝛽𝑘 Γ (𝑖 + 𝛼 + 1)
𝛽 Γ (𝑖 + 2) 𝜉 (𝛼,𝛽) 2
(𝛼,𝛽)
[𝜕𝑥 𝐿
(𝛼,𝛽)
(𝜉 )] = ∑ (−1)𝑘 𝑥𝑘−] ,
𝐺,𝑖,𝑗 𝑖+1 𝐺,𝑖,𝑗
𝑘=⌈]⌉
(𝑖 − 𝑘)! Γ (𝑘 − ] + 1) Γ (𝑘 + 𝛼 + 1)
(12)
𝑖 = ⌈]⌉ , . . . , 𝑁.
For the Gauss-Radau weights, we have (18)
(𝛼,𝛽)
𝜔𝑅,𝑖,𝑗 The approximation of 𝑥𝑘−] by 𝑁 + 1 terms of modified
generalized Laguerre series yields
{ (𝛼 + 1) Γ2 (𝛼 + 1) Γ (𝑖 + 1)
{
{ , 𝑗 = 0,
{
{ 𝛽𝛼+1 Γ (𝑖 + 𝛼 + 2) 𝑁
(𝛼,𝛽)
={ 𝑥𝑘−] = ∑𝑏𝑗 𝐿 𝑗 (𝑥) , (19)
{ Γ (𝑖 + 𝛼 + 1)
{ 1 𝑗=0
{
{ 𝛽𝛼 Γ (𝑖 + 2) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
, 1 ≤ 𝑗 ≤ 𝑖,
{ 𝐿 𝑖+1 (𝜉𝑅,𝑖,𝑗 ) 𝜕 𝑥 𝐿 𝑖 (𝜉𝑅,𝑖,𝑗 )
where 𝑏𝑗 is given by
(13)
𝑗
𝛽−𝑘+] 𝑗! Γ (𝑘 − ] + 𝛼 + ℓ + 1)
Note that the earlier two types of quadratures have close 𝑏𝑗 = ∑ (−1)ℓ , (20)
relations, ℓ=0 (𝑗 − ℓ)! (ℓ)!Γ (ℓ + 𝛼 + 1)

(𝛼,𝛽) (𝛼+1,𝛽) (𝛼,𝛽) (𝛼,𝛽) −1 (𝛼+1,𝛽) Thanks to (18)–(20), we can write


𝜉𝑅,𝑖,𝑗 = 𝜉𝐺,𝑖−1,𝑗−1 , 𝜔𝑅,𝑖,𝑗 = (𝜉𝑅,𝑖,𝑗 ) 𝜔𝐺,𝑖−1,𝑗−1 ,
(14)
𝑁
1 ≤ 𝑗 ≤ 𝑖. 𝐷] 𝐿 𝑖
(𝛼,𝛽)
(𝑥) = ∑ Ψ] (𝑖, 𝑗) 𝐿 𝑗
(𝛼,𝛽)
(𝑥) , 𝑖 = ⌈]⌉ , . . . , 𝑁,
𝑗=0
(𝛼,𝛽) (21)
3. The Fractional Derivatives of 𝐿 𝑖 (𝑥)
In this section, we prove the following theorem for expressing where Ψ] (𝑖, 𝑗) = ∑𝑖𝑘=⌈]⌉ 𝜃𝑖𝑗𝑘 , and
explicitly the fractional order derivatives of the modified
generalized Laguerre polynomials in terms of the modified 𝜃𝑖𝑗𝑘
generalized Laguerre polynomials themselves.
𝑗
(−1)𝑘+ℓ 𝛽] 𝑗! Γ (𝑖 + 𝛼 + 1) Γ (𝑘 − ] + 𝛼 + ℓ + 1)
Lemma 1. Let
(𝛼,𝛽)
𝐿 𝑖 (𝑥) be a modified generalized Laguerre =∑ .
ℓ=0 (𝑖 − 𝑘)! (𝑗 − ℓ)! ℓ! Γ (𝑘 − ] + 1) Γ (𝑘 + 𝛼 +1) Γ (𝛼+ℓ+1)
polynomial; then (22)
(𝛼,𝛽)
𝐷] 𝐿 𝑖 (𝑥) = 0, 𝑖 = 0, 1, . . . , ⌈]⌉ − 1, ] > 0. (15)
4 Abstract and Applied Analysis

4. Application of the Fractional Derivative of Let us denote


Modified Generalized Laguerre Polynomials
𝑁
In this section, we consider spectral tau and collocation (𝛼,𝛽) 𝑇
𝑢𝑁 (𝑥) = ∑𝑎𝑗 𝐿 𝑗 (𝑥) , a = (𝑎0 , 𝑎1 , . . . , 𝑎𝑁) ,
methods based on the fractional derivative of modified 𝑗=0
generalized Laguerre polynomials to solve numerically the
linear and nonlinear multiterm FDEs on the half line. (𝛼,𝛽) (27)
𝑔𝑘 = (𝑔, 𝐿 𝑘 (𝑥))𝑤(𝛼,𝛽) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,

4.1. Tau Method for Linear Multiterm FDEs. We apply the 𝑇


g = (𝑔0 , 𝑔1 , . . . , 𝑔𝑁−𝑚 , 𝑑0 , . . . , 𝑑𝑚−1 ) ,
modified generalized Laguerre tau (MGLT) method for
then (26) can be written as
tackling the linear multiterm FDEs on the half line
𝑘
𝑁
𝐷] 𝑢 (𝑥) = ∑ 𝛾𝑗 𝐷𝜁𝑗 𝑢 (𝑥) + 𝛾𝑘+1 𝑢 (𝑥) + 𝑔 (𝑥) , in Λ, ∑𝑎𝑗 [(𝐷] 𝐿 𝑗
(𝛼,𝛽) (𝛼,𝛽)
(𝑥) , 𝐿 𝑘 (𝑥))
𝑗=1 𝑤(𝛼,𝛽)
𝑗=0
(23)
𝑟−1
(𝛼,𝛽) (𝛼,𝛽)
subject to initial conditions + ∑𝛾𝑖 (𝐷𝜁𝑖 𝐿 𝑗 (𝑥) , 𝐿 𝑘 (𝑥))
𝑤(𝛼,𝛽)
𝑖=1
𝑢(𝑖) (0) = 𝑑𝑖 , 𝑖 = 0, . . . , 𝑚 − 1, (24) (𝛼,𝛽) (𝛼,𝛽)
+ 𝛾𝑟 (𝐿 𝑗 (𝑥) , 𝐿 𝑘 (𝑥)) ]
𝑤(𝛼,𝛽)
where 𝛾𝑗 (𝑗 = 1, . . . , 𝑘 + 1) are constant coefficients, 𝑚 − 1 <
(𝛼,𝛽)
] ≤ 𝑚, 0 < 𝜁1 < 𝜁2 < ⋅ ⋅ ⋅ < 𝜁𝑘 < ]. While 𝐷] 𝑢(𝑥) ≡ 𝑢(]) (𝑥) is = (𝑔, 𝐿 𝑘 (𝑥))𝑤(𝛼,𝛽) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
the ]-order fractional derivative of 𝑢(𝑥), and 𝑔(𝑥) is a source
𝑁
function. (𝛼,𝛽)(𝑘−𝑁+𝑚−1)
∑𝑎𝑗 𝐿 𝑗 (0)
The modified generalized Laguerre polynomials
𝑗=0
(𝛼,𝛽)
{𝐿 𝑖 (𝑥) : 𝑖 ≥ 0} form a complete orthogonal system
in weighted space; 𝐿2𝑤(𝛼,𝛽) (Λ). Hence, we define = 𝑑(𝑘−𝑁+𝑚−1) , 𝑘 = 𝑁 − 𝑚 + 1, 𝑁 − 𝑚 + 2, . . . , 𝑁.
(28)
(𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
𝑆𝑁 (Λ) = span{𝐿 0 (𝑥) , 𝐿 1 (𝑥) , . . . , 𝐿 𝑁 (𝑥)} , (25)

then the standard modified generalized Laguerre tau approx- Let us also denote
imation to (23) is to find 𝑢𝑁 ∈ 𝑆𝑁(Λ) such that
𝑟−1 𝐴 = (𝑎𝑘𝑗 )0<𝑘,𝑗<𝑁,
(𝛼,𝛽) (𝛼,𝛽)
(𝐷] 𝑢𝑁, 𝐿 𝑘 (𝑥))𝑤(𝛼,𝛽) + ∑𝛾𝑖 (𝐷𝜁𝑖 𝑢𝑁, 𝐿 𝑘 (𝑥))𝑤(𝛼,𝛽)
𝑖=1 𝐵𝑠 = (𝑏𝑘𝑗
𝑠
) , (29)
0<𝑘,𝑗<𝑁;𝑠=1,2,...,𝑟−1
(𝛼,𝛽)
+ 𝛾𝑟 (𝑢𝑁, 𝐿 𝑘 (𝑥))𝑤(𝛼,𝛽) (26)
(𝛼,𝛽) 𝐶 = (𝑐𝑘𝑗 )0<𝑘,𝑗<𝑁,
= ((𝑔, 𝐿 𝑘 (𝑥))𝑤(𝛼,𝛽) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
(𝑖)
𝑢𝑁 (0) = 𝑑𝑖 , 𝑖 = 0, 1, . . . , 𝑚 − 1. where

(𝛼,𝛽) (𝛼,𝛽)
(𝐷] 𝐿 𝑗 (𝑥) , 𝐿 𝑘 (𝑥)) (𝛼,𝛽) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑗 = 0, 1, . . . , 𝑁,
𝑎𝑘𝑗 = { 𝑘−𝑁+𝑚−1 (𝛼,𝛽)
𝑤
𝐷 𝐿 𝑗 (0) , 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁, 𝑗 = 0, 1, . . . , 𝑁,
(𝛼,𝛽) (𝛼,𝛽)
𝑠 (𝐷𝜁𝑠 𝐿 𝑗 (𝑥) , 𝐿 𝑘 (𝑥)) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑗 = 0, 1, . . . , 𝑁, 𝑠 = 1, 2, . . . , 𝑟 − 1,
𝑏𝑘𝑗 ={ 𝑤(𝛼,𝛽) (30)
0, otherwise,
(𝛼,𝛽) (𝛼,𝛽)
(𝐿 (𝑥), 𝐿 𝑘 (𝑥)) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑗 = 0, 1, . . . , 𝑁,
𝑐𝑘𝑗 = { 𝑗 𝑤(𝛼,𝛽)
0, otherwise,

then by virtue of the orthogonality relation of modified and (16) and after some manipulation, we may deduce that
𝑠
generalized Laguerre polynomials (6) and making use of (8) the elements of 𝑎𝑘𝑗 , 𝑏𝑘𝑗 , 𝑠 = 1, 2, . . . , 𝑟 − 1, and 𝑐𝑘𝑗 are given by
Abstract and Applied Analysis 5

{ℎ𝑘 Ψ] (𝑗, 𝑘) ,
{ 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑗 = 𝑚, 𝑚 + 1, . . . , 𝑁,
𝑎𝑘𝑗 = { (−1)𝑘−𝑁+𝑚−1 𝛽(𝑘−𝑁+𝑚−1) Γ (𝑖 + 𝛼 + 1)!
{ , 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁, 𝑗 = 0, 1, . . . , 𝑁,
{ (𝑖 − 𝑘 + 𝑁 − 𝑚 + 1)! Γ (𝑘 − 𝑁 + 𝑚 + 𝛼 + 2)

𝑠 ℎ Ψ (𝑗, 𝑘) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑗 = ⌈𝜁𝑠 ⌉ , ⌈𝜁𝑠 ⌉ + 1, . . . , 𝑁, 𝑠 = 1, 2, . . . , 𝑟 − 1, (31)


𝑏𝑘𝑗 = { 𝑘 𝜁𝑠
0, otherwise,

ℎ , 𝑘 = 𝑗 = 0, 1, . . . , 𝑁 − 𝑚,
𝑐𝑘𝑗 = { 𝑘
0, otherwise.

Therefore, (28) can be written in the following matrix form: generalized Laguerre roots 𝑥𝑁−𝑚+1,𝑘 , which immediately
yields
𝑟−1
(𝐴 + ∑𝛾𝑖 𝐵𝑖 + 𝛾𝑟 𝐶) a = g. (32) 𝑁
(𝛼,𝛽)
𝑖=1 ∑𝑎𝑗 𝐷] 𝐿 𝑗 (𝑥𝑁−𝑚+1,𝑘 )
𝑗=0
4.2. Collocation Method for Nonlinear Multiterm FDEs. It is 𝑁
(𝛼,𝛽)
known that the collocation method approximates differential = 𝐹 (𝑥𝑁−𝑚+1,𝑘, ∑𝑎𝑗 𝐷𝛽1 𝐿 𝑗 (𝑥𝑁−𝑚+1,𝑘 ) , . . . ,
equations in physical space, so it is easy to apply to various 𝑗=0
problems, including variable coefficient and nonlinear dif- 𝑁
(𝛼,𝛽)
ferential equations (see, for instance, [36]). In this section, ∑ 𝑎𝑗 𝐷𝛽𝑘 𝐿 𝑗 (𝑥𝑁−𝑚+1,𝑘 )) ,
an algorithm for solving fractional differential equation is 𝑗=0
proposed based on modified generalized Laguerre spectral
𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
collocation (MGLC) method in a half line. Here, we study the
(37)
following multiterm FDE:
with (33) written in the form
𝐷] 𝑢 (𝑥) = 𝐹 (𝑥, 𝑢 (𝑥) , 𝐷𝛽1 𝑢 (𝑥) , . . . , 𝐷𝛽𝑘 𝑢 (𝑥)) , in Λ,
𝑁
(33) (𝛼,𝛽)
∑𝑎𝑗 𝐿 𝑗 (0) = 𝑑𝑖 , 𝑖 = 0, 1, . . . , 𝑚 − 1. (38)
with initial conditions 𝑗=0

Finally, the problem is reduced to solve a system of (𝑁 +


𝑢(𝑖) (0) = 𝑑𝑖 , 𝑖 = 0, . . . , 𝑚 − 1, (34)
1) nonlinear algebraic equations, (37)-(38), in the unknown
where the fractional order derivatives are taken to be of the expansion coefficients {𝑎𝑗 ; 𝑗 = 0, 1, . . . , 𝑁}, which can be
Caputo type, 𝑚 − 1 < ] ≤ 𝑚, 0 < 𝛽1 < 𝛽2 < ⋅ ⋅ ⋅ < 𝛽𝑘 < ], and solved by Newton’s iteration method.
𝐹 is a nonlinear function.
We now derive an efficient algorithm for solving non- 5. Numerical Results
linear multiterm FDEs (33)-(34). We expand the numerical
approximation in terms of modified generalized Laguerre We report in this section some numerical results obtained
polynomials with the proposed algorithms in the previous section. Several
test examples are implemented to ensure the effectiveness of
𝑁
(𝛼,𝛽) the proposed methods for fractional differential equations in
𝑢𝑁 (𝑥) = ∑𝑎𝑗 𝐿 𝑗 (𝑥) . (35) the half line.
𝑗=0

By virtue of (16), the fractional derivatives 𝐷] 𝑢(𝑥), Example 3. This example concerns a class of Bagley-Torvik
equation given by the initial value problem
𝐷𝛽1 𝑢(𝑥), . . . , 𝐷𝛽𝑘 𝑢(𝑥) can be expressed explicitly in terms of
the expansion coefficients 𝑎𝑗 . 8
The modified generalized Laguerre spectral collocation 𝐷2 𝑢 (𝑥) + 𝐷3/2 𝑢 (𝑥) + 𝑢 (𝑥) = 𝑥3 + 6𝑥 + 𝑥1.5 ,
Γ (0.5)
method for solving approximately (33), (34) is to find 𝑢𝑁(𝑥) ∈
𝑆𝑁(Λ) such that 𝑢 (0) = 0, 𝑢󸀠 (0) = 0, 𝑥 ∈ [0, 10] .
(39)
𝐷] 𝑢𝑁 (𝑥) = 𝐹 (𝑥, 𝑢𝑁 (𝑥) , 𝐷𝛽1 𝑢𝑁 (𝑥) , . . . , 𝐷𝛽𝑘 𝑢𝑁 (𝑥)) ,
(36) The solution is 𝑢(𝑥) = 𝑥3 .

is collocated at 𝑥𝑁−𝑚+1,𝑘 , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚. In other This equation arises in the treatment of the motion
words, we have to collocate (36) at the (𝑁 − 𝑚 + 1) modified of a large thin plate immersed in a Newtonian fluid.
6 Abstract and Applied Analysis

5 5

4 4

−Log 10 error
−Log 10 error

3 3

2 2

1 1

0 0
20 30 40 50 60 70 80 90 20 30 40 50 60 70 80 90
N N

𝛼 = 1 and 𝛽 = 3 𝛼 = 1 and 𝛽 = 2 𝛼 = 1 and 𝛽 = 3 𝛼 = 1 and 𝛽 = 2


𝛼 = 0 and 𝛽 = 1 𝛼 = 5 and 𝛽 = 4 𝛼 = 0 and 𝛽 = 1 𝛼 = 5 and 𝛽 = 4

Figure 1: Maximum absolute error by using MGLT (Gauss) with the Figure 2: Maximum absolute error by using MGLT (Radau) with
various choices of 𝑁. the various choices of 𝑁.

For the numerical solution of the Bagley-Torvik equation, see 4000


[37–41]. The solution of this problem is obtained by applying
the modified generalized Laguerre tau method and the right- 3000
hand side is treated by modified generalized Laguerre-Gauss
(G) and modified generalized Laguerre-Gauss-Radau (GR)
uN (x)

2000
quadratures with general parameters 𝛼 and 𝛽. The maximum
absolute errors obtained by MGLT method for various
choices of 𝑁, 𝛼, and 𝛽 are given in Figures 1 and 2. Moreover, 1000
the approximate solutions obtained by the proposed method
for 𝛼 = 0, 𝛽 = 2, and two choices of 𝑁 are shown in Figure 3 0
0 5 10 15
to make it easier to compare with the analytic solution.
x
Example 4. Consider the equation Exact solution
uN (x) at N = 20
𝐷2 𝑢 (𝑥) + 𝐷] 𝑢 (𝑥) + 𝑢 (𝑥) = 𝑔 (𝑥) , uN (x) at N = 40
(40)
𝑢 (0) = 0, 𝑢󸀠 (0) = 0, 0 < ] < 1, 𝑥 ∈ [0, 10] , Figure 3: Comparing the exact and approximate solutions at 𝑁 =
20, 40, with 𝛼 = 0 and 𝛽 = 2.
where
1
𝑔 (𝑥) = 𝑥4 − 𝑥3 + 12𝑥2 − 3𝑥
2
(41) The solution of this problem is obtained by applying the
3 24 technique described in Section 4. The maximum absolute
− 𝑥3−] + 𝑥4−] ,
Γ (4 − ]) Γ (5 − ]) error for 𝛾 = {0.1 and 0.01} and various choices of 𝑁, 𝛼, 𝛽,
and ] are shown in Tables 2 and 3; moreover, the right-hand
and the exact solution is given by 𝑢(𝑥) = 𝑥4 − (1/2)𝑥3 (see side is treated by modified generalized Laguerre Gauss and
Table 1). modified generalized Laguerre-Gauss-Radau quadratures.

Example 5. We next consider the following: Example 6. We consider the equation

𝐷2 𝑢 (𝑥) + 𝐷] 𝑢 (𝑥) + 𝑢 (𝑥) = 𝑔 (𝑥) ,


(42) 𝑎𝐷] 𝑢 (𝑥) + 𝑏𝐷]2 𝑢 (𝑥) + 𝑐𝐷]1 𝑢 (𝑥) + 𝑒𝑢 (𝑥) = 𝑔 (𝑥) ,
𝑢 (0) = 1, 𝑢󸀠 (0) = 0, 𝑥 ∈ [0, 10] ,
0 < ]1 < 1, 1 < ]2 < 2, 3 < ] < 4, 𝑥 ∈ [0, 5] ,
where
2𝑏 2𝑐
1 𝑥 𝑔 (𝑥) = 𝑥2−]2 + 𝑥2−]1
𝑔 (𝑥) = (1 − 𝛾 ) cos (𝛾𝑥) + 2
∫ (𝑥 − 𝑡)−]−1 𝑢 (𝑡) 𝑑𝑡, Γ (3 − ]2 ) Γ (3 − ]1 )
Γ (−]) 0
𝑐
(43) − 𝑥1−]1 + 𝑒 (𝑥2 − 𝑥) ,
Γ (2 − ]1 )
and the exact solution is given by 𝑢(𝑥) = cos(𝛾𝑥). (44)
Abstract and Applied Analysis 7

Table 1: Maximum absolute error, with various choices of 𝑁, for Example 4 in which 𝑥 ∈ [0, 10].

𝑁 𝛼 𝛽 ] MGLT(G) MGLT(GR) ] MGLT(G) MGLT(GR) ] MGLT(G) MGLT(GR)


10 1.32 ⋅ 10−2 2.24 ⋅ 10−2 8.48 ⋅ 10−2 1.41 ⋅ 10−1 6.54 ⋅ 10−3 1.07 ⋅ 10−2
30 1.35 ⋅ 10−4 1.78 ⋅ 10−4 1.37 ⋅ 10−3 1.82 ⋅ 10−3 1.84 ⋅ 10−4 2.53 ⋅ 10−4
0 2 0.1 0.5 0.99
50 1.96 ⋅ 10−5 2.47 ⋅ 10−5 2.45 ⋅ 10−4 3.14 ⋅ 10−4 4.30 ⋅ 10−5 5.71 ⋅ 10−5
70 3.41 ⋅ 10−6 4.20 ⋅ 10−6 4.86 ⋅ 10−5 6.10 ⋅ 10−5 7.03 ⋅ 10−6 9.16 ⋅ 10−6
10 4.54 ⋅ 10−3 1.67 ⋅ 10−2 2.36 ⋅ 10−2 8.98 ⋅ 10−2 1.43 ⋅ 10−3 5.85 ⋅ 10−3
30 2.86 ⋅ 10−6 8.06 ⋅ 10−6 2.73 ⋅ 10−5 8.33 ⋅ 10−5 2.96 ⋅ 10−6 1.03 ⋅ 10−5
4 3 0.1 0.5 0.99
50 8.61 ⋅ 10−8 2.37 ⋅ 10−7 1.22 ⋅ 10−6 3.54 ⋅ 10−6 1.72 ⋅ 10−7 5.72 ⋅ 10−7
70 1.81 ⋅ 10−8 5.12 ⋅ 10−8 2.10 ⋅ 10−7 6.53 ⋅ 10−7 3.30 ⋅ 10−8 1.19 ⋅ 10−7
10 4.92 ⋅ 10−3 2.27 ⋅ 10−2 2.77 ⋅ 10−2 1.33 ⋅ 10−1 8.80 ⋅ 10−3 1.67 ⋅ 10−3
30 2.83 ⋅ 10−6 1.01 ⋅ 10−5 2.71 ⋅ 10−5 1.06 ⋅ 10−4 3.00 ⋅ 10−6 1.35 ⋅ 10−5
6 4 0.1 0.5 0.99
50 7.88 ⋅ 10−8 2.73 ⋅ 10−7 5.48 ⋅ 10−7 2.19 ⋅ 10−6 6.94 ⋅ 10−8 3.25 ⋅ 10−7
70 1.49 ⋅ 10−8 5.30 ⋅ 10−8 1.74 ⋅ 10−7 6.87 ⋅ 10−7 2.72 ⋅ 10−8 1.26 ⋅ 10−7

Table 2: Maximum absolute error for 𝛾 = 1/10, with various choices of 𝑁, for Example 5 in which 𝑥 ∈ [0, 10].

𝑁 𝛼 𝛽 ] MGLT(G) MGLT(GR) ] MGLT(G) MGLT(GR) ] MGLT(G) MGLT(GR)


10 7.97 ⋅ 10−4 2.47 ⋅ 10−3 7.28 ⋅ 10−5 1.40 ⋅ 10−4 1.10 ⋅ 10−4 1.89 ⋅ 10−4
30 1.38 ⋅ 10−4 3.81 ⋅ 10−4 7.11 ⋅ 10−6 1.21 ⋅ 10−5 7.36 ⋅ 10−6 1.11 ⋅ 10−5
0 1 1.5 0.9 0.5
50 4.02 ⋅ 10−5 1.09 ⋅ 10−4 1.51 ⋅ 10−6 2.49 ⋅ 10−6 1.29 ⋅ 10−6 1.88 ⋅ 10−6
70 2.51 ⋅ 10−5 6.64 ⋅ 10−5 3.79 ⋅ 10−7 6.15 ⋅ 10−7 5.86 ⋅ 10−7 8.41 ⋅ 10−7
10 5.39 ⋅ 10−4 4.44 ⋅ 10−3 6.29 ⋅ 10−5 2.78 ⋅ 10−4 9.84 ⋅ 10−5 3.54 ⋅ 10−4
30 1.33 ⋅ 10−5 9.33 ⋅ 10−5 8.65 ⋅ 10−7 3.20 ⋅ 10−6 9.31 ⋅ 10−7 2.77 ⋅ 10−6
2 2 1.5 0.9 0.5
50 4.26 ⋅ 10−6 3.00 ⋅ 10−5 1.99 ⋅ 10−7 7.37 ⋅ 10−7 1.71 ⋅ 10−7 5.06 ⋅ 10−7
70 1.78 ⋅ 10−6 1.22 ⋅ 10−5 7.56 ⋅ 10−8 2.72 ⋅ 10−7 5.28 ⋅ 10−8 1.52 ⋅ 10−7
10 1.19 ⋅ 10−4 1.87 ⋅ 10−3 1.67 ⋅ 10−5 1.34 ⋅ 10−4 2.93 ⋅ 10−5 1.86 ⋅ 10−4
30 2.25 ⋅ 10−6 3.48 ⋅ 10−5 1.58 ⋅ 10−7 1.09 ⋅ 10−6 1.76 ⋅ 10−7 1.02 ⋅ 10−6
5 3 1.5 0.9 0.5
50 1.28 ⋅ 10−6 1.92 ⋅ 10−5 5.85 ⋅ 10−8 3.33 ⋅ 10−7 5.97 ⋅ 10−8 3.33 ⋅ 10−7
70 8.52 ⋅ 10−7 1.26 ⋅ 10−5 3.21 ⋅ 10−8 2.30 ⋅ 10−7 2.81 ⋅ 10−8 1.54 ⋅ 10−7

Table 3: Maximum absolute error for 𝛾 = 1/100, with various choices of 𝑁, for Example 5 in which 𝑥 ∈ [0, 10].

𝑁 𝛼 𝛽 ] MGLT(G) MGLT(GR) ] MGLT(G) MGLT(GR) ] MGLT(G) MGLT(GR)


10 7.97 ⋅ 10−6 2.47 ⋅ 10−5 7.28 ⋅ 10−7 1.40 ⋅ 10−6 1.10 ⋅ 10−6 1.89 ⋅ 10−6
30 1.38 ⋅ 10−6 3.81 ⋅ 10−6 7.11 ⋅ 10−8 1.21 ⋅ 10−7 7.39 ⋅ 10−8 1.11 ⋅ 10−7
0 1 1.5 0.9 0.5
50 4.02 ⋅ 10−7 1.09 ⋅ 10−6 1.51 ⋅ 10−8 2.49 ⋅ 10−8 1.29 ⋅ 10−8 1.88 ⋅ 10−8
70 2.51 ⋅ 10−7 6.64 ⋅ 10−7 3.79 ⋅ 10−9 6.15 ⋅ 10−9 5.86 ⋅ 10−9 8.41 ⋅ 10−9
10 5.39 ⋅ 10−6 4.44 ⋅ 10−5 6.29 ⋅ 10−7 2.78 ⋅ 10−6 9.85 ⋅ 10−7 3.54 ⋅ 10−6
30 1.33 ⋅ 10−7 9.33 ⋅ 10−7 8.65 ⋅ 10−9 3.20 ⋅ 10−8 9.31 ⋅ 10−9 2.77 ⋅ 10−8
2 2 1.5 0.9 0.5
50 4.26 ⋅ 10−8 3.00 ⋅ 10−7 2.00 ⋅ 10−9 7.37 ⋅ 10−9 1.71 ⋅ 10−9 5.06 ⋅ 10−9
70 1.78 ⋅ 10−8 1.22 ⋅ 10−7 7.56 ⋅ 10−10 2.72 ⋅ 10−9 5.28 ⋅ 10−10 1.52 ⋅ 10−9
10 1.19 ⋅ 10−6 1.87 ⋅ 10−5 1.67 ⋅ 10−7 1.34 ⋅ 10−6 2.94 ⋅ 10−7 1.87 ⋅ 10−6
30 2.25 ⋅ 10−8 3.48 ⋅ 10−7 1.58 ⋅ 10−9 1.09 ⋅ 10−8 1.76 ⋅ 10−9 1.02 ⋅ 10−8
5 3 1.5 0.9 0.5
50 1.28 ⋅ 10−8 1.92 ⋅ 10−7 5.85 ⋅ 10−10 4.27 ⋅ 10−9 5.97 ⋅ 10−10 3.33 ⋅ 10−9
70 8.52 ⋅ 10−9 1.26 ⋅ 10−7 3.21 ⋅ 10−10 2.30 ⋅ 10−9 2.81 ⋅ 10−10 1.53 ⋅ 10−9

subject to (I) Regarding 𝑎 = 1, 𝑏 = 1, 𝑐 = 1, 𝑒 = 1, ]1 = 0.77,


]2 = 1.44, and ] = 3.91, Figures 4 and 5 display
𝑢 (0) = 0, 𝑢󸀠 (0) = −1, the maximum absolute errors using MGLT method
(45) with the treatment of right-hand side by modi-
𝑢󸀠󸀠 (0) = 2, 𝑢󸀠󸀠󸀠 (0) = 0. fied generalized Laguerre-Gauss and Gauss-Radau,
respectively.
The analytic solution of this problem is 𝑢(𝑥) = 𝑥2 − 𝑥, we (II) Regarding 𝑎 = 1, 𝑏 = 1, 𝑐 = 0.5, 𝑒 = 0.5, ]1 = √2/20,
study two different cases of 𝑎, 𝑏, 𝑐, 𝑒, 𝑘, ]1 , ]2 , and ]. ]2 = √2, and ] = √11, Figures 6 and 7 display
8 Abstract and Applied Analysis

5.5
4
5
4.5 3

−Log 10 error
−Log 10 error

4
2
3.5
3 1
2.5
2 0
10 20 30 40 50
10 20 30 40 50
N
N

𝛼 = 1 and 𝛽 = 4 𝛼 = 1 and 𝛽 = 4
𝛼 = 3 and 𝛽 = 3 𝛼 = 3 and 𝛽 = 3
𝛼 = 3 and 𝛽 = 6
𝛼 = 3 and 𝛽 = 6

Figure 4: Maximum absolute error (case I) by using MGLT (Gauss) Figure 6: Maximum absolute error (case II) by using MGLT
with the various choices of 𝑁. (Gauss) with various choices of 𝑁.

3.5 3.5

3 3

2.5 2.5
−Log 10 error
−Log 10 error

2 2

1.5 1.5

1 1

0.5 0.5

0 0
10 20 30 40 50
10 20 30 40 50
N
N
𝛼 = 1 and 𝛽 = 4
𝛼 = 1 and 𝛽 = 4
𝛼 = 3 and 𝛽 = 3 𝛼 = 3 and 𝛽 = 3
𝛼 = 3 and 𝛽 = 6 𝛼 = 3 and 𝛽 = 6

Figure 5: Maximum absolute error (case I) by using MGLT (Radau) Figure 7: Maximum absolute error (case II) by using MGLT
with the various choices of 𝑁. (Radau) with various choices of 𝑁.

the maximum absolute errors using MGLT method 339Γ (3)


− )
with the treatment of right-hand side by modi- 250Γ (3 − ])
fied generalized Laguerre-Gauss and Gauss-Radau,
29𝑥4 76𝑥3 339𝑥2 27𝑥
respectively. + 𝑐 (𝑥5 − + − + ),
10 25 250 125
(47)
Example 7. Consider the Bagley-Torvik equation with
boundary conditions
and the exact solution is given by 𝑢(𝑥) = 𝑥5 − (29𝑥4 /10) +
𝑎𝐷2 𝑢 (𝑥) + 𝑏𝐷] 𝑢 (𝑥) + 𝑐𝑢 (𝑥) = 𝑔 (𝑥) , (76𝑥3 /25) − (339𝑥2 /250) + (27𝑥/125).
(46)
𝑢 (0) = 0, 𝑢 (1) = 0, 𝑥 ∈ (0, 1) , Bagley-Torvik equation involving fractional derivative of
order 3/2 arises in the modeling of the motion of a rigid
where plate in a Newtonian fluid and a gas in a fluid. We solve this
equation with two-point boundary conditions where 𝑎 = 1,
174𝑥2 456𝑥 339 𝑏 = 8/17, and 𝑐 = 13/51 by using modified generalized
𝑔 (𝑥) = 𝑎 (20𝑥3 − + − )
5 5 125 Laguerre-Gauss and Gauss-Radau quadratures with general
parameters 𝛼 and 𝛽. The maximum absolute errors obtained
120𝑥5−] 696𝑥4−] 456𝑥3−] by MGLT method for various choices of 𝑁, 𝛼, and 𝛽 are
+ 𝑏( − +
Γ (6 − ]) 10Γ (5 − ]) 25Γ (4 − ]) given in Table 4. Moreover, the numerical and exact solutions
Abstract and Applied Analysis 9

Table 4: Maximum absolute error for ] = 3/2, with various choices of 𝑁, for Example 5 in which 𝑥 ∈ (0, 1).

𝑁 𝛼 𝛽 MGLT(G) MGLT(GR) 𝛼 𝛽 MGLT(G) MGLT(GR)


8 1.69 ⋅ 10−3 3.14 ⋅ 10−3 8.98 ⋅ 10−5 8.21 ⋅ 10−4
16 9.97 ⋅ 10−4 2.35 ⋅ 10−3 1.98 ⋅ 10−5 1.91 ⋅ 10−4
0 1 3 4
32 1.57 ⋅ 10−4 3.50 ⋅ 10−4 8.50 ⋅ 10−6 7.89 ⋅ 10−5
64 3.52 ⋅ 10−5 9.97 ⋅ 10−5 3.81 ⋅ 10−6 3.72 ⋅ 10−5
8 6.47 ⋅ 10−5 7.07 ⋅ 10−4 8.61 ⋅ 10−5 1.02 ⋅ 10−3
16 1.53 ⋅ 10−5 1.76 ⋅ 10−4 2.07 ⋅ 10−5 2.63 ⋅ 10−4
4 5 5 5
32 6.40 ⋅ 10−6 7.31 ⋅ 10−5 6.17 ⋅ 10−6 8.06 ⋅ 10−5
64 3.02 ⋅ 10−6 3.57 ⋅ 10−5 3.19 ⋅ 10−6 4.37 ⋅ 10−5

0.01 0.01
0.008 0.008
uN (x)

uN (x)
0.006 0.006
0.004 0.004
0.002 0.002
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x

Exact solution Exact solution


uN (x) at N = 8 uN (x) at N = 8
uN (x) at N = 64 uN (x) at N = 64

Figure 8: Comparing the exact and approximate solutions by using Figure 9: Comparing the exact and approximate solutions by using
MGLT (Gauss) at 𝑁 = 8, 64, with 𝛼 = 2 and 𝛽 = 4. MGLT (Radau) at 𝑁 = 8, 64, with 𝛼 = 2 and 𝛽 = 4.

obtained by the proposed method for 𝛼 = 2, 𝛽 = 4, and two 1


choices of 𝑁 are shown in Figures 8-9 to make it easier to
compare with the analytic solution. 0.995
uN (x)

Example 8. We next consider the following nonlinear initial


0.99
value problem:

𝐷2 𝑢 (𝑥) + 𝐷3/4 𝑢 (𝑥) + 𝑢2 (𝑥) = 𝑔 (𝑥) , 0.985


(48) 0 2 4 6 8 10 12 14
𝑢 (0) = 1, 𝑢󸀠 (0) = 0, 𝑥 ∈ (0, 10) , x

where Exact solution uN (x) at N = 15


uN (x) at N = 5 uN (x) at N = 20
𝑔 (𝑥) = cos2 (𝛾𝑥) − 𝛾2 cos (𝛾𝑥) uN (x) at N = 10

1 𝑥 (49) Figure 10: Comparing the exact and approximate solutions at 𝑁 =


+ ∫ (𝑥 − 𝑡)(−3/4)−1 cos (𝛾𝑡) 𝑑𝑡 5, 10, 15, and 20, where ] = 3/4, 𝛼 = 2, and 𝛽 = 2.
Γ (−3/4) 0

and the exact solution is given by 𝑢(𝑥) = cos(𝛾𝑥).


Example 9. Consider the following nonlinear initial value
The solution of this problem is obtained by applying problem:
the modified generalized Laguerre collocation method. The
approximate solution obtained by the proposed method for 2
𝑥]+1
𝛾 = 0.01, 𝛼 = 2, 𝛽 = 2, and two choices of 𝑁 is shown 𝐷] 𝑢 (𝑥) + 𝑢2 (𝑥) = 𝑥 + ( ), 0 < ] ≤ 2, (50)
in Figure 10 to make it easier to compare with the analytic Γ (] + 2)
solution. Moreover, the absolute error for 𝛾 = 0.01, 𝛼 = 2,
𝛽 = 2, and 𝑁 = 20 is given in Figure 11. whose exact solution is given by 𝑢(𝑥) = (1/Γ(] + 2))𝑥]+1 .
10 Abstract and Applied Analysis

Table 5: Maximum absolute error for ] = 3/2, with various choices of 𝑁, for Example 9 in which 𝑥 ∈ (0, 1).

𝑁 𝛼 𝛽 MGLC 𝛼 𝛽 MGLC
3 2.66 ⋅ 10−2 2.61 ⋅ 10−2
6 2.27 ⋅ 10−3 3.04 ⋅ 10−3
0 1 2 3
9 1.30 ⋅ 10−3 1.36 ⋅ 10−3
12 5.68 ⋅ 10−4 7.52 ⋅ 10−4
3 1.39 ⋅ 10−2 1.07 ⋅ 10−2
6 1.87 ⋅ 10−3 1.86 ⋅ 10−3
2 4 3 6
9 8.47 ⋅ 10−4 6.94 ⋅ 10−4
12 4.02 ⋅ 10−4 3.63 ⋅ 10−4

×10−6 10

Exact and approximate solutions


3.5
3 8
The absolute error

2.5
6
2
1.5 4

1
2
0.5
0
0
0 2 4 6 8 10 12 14 0 1 2 3 4
x x

Figure 11: The absolute error for 𝛾 = 1/100, ] = 3/4, 𝛼 = 2, and uN (x),  = 1.2 uN (x),  = 1.6
𝛽 = 2 at 𝑁 = 20. u(x),  = 1.2 u(x),  = 1.6
uN (x),  = 1.4 uN (x),  = 1.8
u(x),  = 1.4 u(x),  = 1.8

Figure 13: Comparing the exact and approximate solutions at 𝑁 =


Exact and approximate solutions

120 16, 𝛼 = 0, 𝛽 = 1, and ] = 1.2, 1.4, 1.6, and 1.8.


100

80
The maximum absolute errors at ] = 1.5 for various choices
60 of 𝛼, 𝛽, and 𝑁 in the interval [0, 1] are shown in Table 5.
Moreover, Figures 14 and 15 display a comparison between
40 the curves of exact solutions and the approximate solutions
20 at 𝛼 = 0 and 𝛽 = 1 of proposed problem subject to 𝑢(0) = 0
for the four different fractional orders ] = 0.2, 0.4, 0.6, and
0 0.8 in case of 𝑁 = 12 and 𝑁 = 16, respectively. From all
0 2 4 6 8 10
the Figures 12–15, it can be seen that the numerical solutions
x are in complete agreement with the exact solutions for all
uN (x),  = 1.2 uN (x),  = 1.6 values of ]. Also, from the numerical results implemented
u(x),  = 1.2 u(x),  = 1.6 in this example, the classical Laguerre polynomial (𝛼 = 0,
uN (x),  = 1.4 uN (x),  = 1.8 𝛽 = 1), which is used most frequently in practice, is not the
u(x),  = 1.4 u(x),  = 1.8 best one, especially when we are approximating the solution
of fractional differential equations.
Figure 12: Comparing the exact and approximate solutions at 𝑁 =
12, 𝛼 = 0, 𝛽 = 1, and ] = 1.2, 1.4, 1.6, and 1.8.
6. Conclusions
In this paper, we have proposed two efficient spectral meth-
Comparison between the curves of exact solutions and ods based on modified generalized Laguerre polynomials for
the approximate solutions at 𝛼 = 0 and 𝛽 = 1 of proposed tackling linear and nonlinear FDEs on the half line. In these
problem subject to 𝑢(0) = 𝑢󸀠 (0) = 0 for the four different methods, the problem is reduced to the solution of a system
fractional orders ] = 1.2, 1.4, 1.6, and 1.8 in case of 𝑁 = 12 of algebraic equations in the expansion coefficient of the
and 𝑁 = 16 are shown in Figures 12 and 13, respectively. solution. Numerical examples were given to demonstrate the
Abstract and Applied Analysis 11

[3] M. Ortigueira, “Introduction to fraction linear systems. Part 1:


Exact and approximate solutions

continuous-time case,” in IEE Proceedings Vision, Image, Signal


30 Processing, vol. 147, pp. 6–70, 2000.
[4] K. Oldham and J. Spanier, The Fractional Calculus, Academic
Press, New York, NY, USA, 1974.
20 [5] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory
and Applications of Fractional Differential Equations, vol. 204,
Elsevier, Amsterdam, The Netherlands, 2006.
10 [6] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang,
Spectral Methods Fundamentals in Single Domains, Springer,
Berlin, Germany, 2006.
0
0 2 4 6 8 10 [7] E. H. Doha, A. H. Bhrawy, D. Baleanu, and S. S. Ezz-Eldien, “On
shifted Jacobi spectral approximations for solving fractional
x
differential equations,” Applied Mathematics and Computation,
uN (x),  = 0.2 uN (x),  = 0.6 vol. 219, no. 15, pp. 8042–8056, 2013.
u(x),  = 0.2 u(x),  = 0.6 [8] J.-P. Yan and B.-Y. Guo, “A collocation method for initial value
uN (x),  = 0.4 uN (x),  = 0.8 problems of second-order ODEs by using Laguerre functions,”
u(x),  = 0.4 u(x),  = 0.8 Numerical Mathematics. Theory, Methods and Applications, vol.
4, no. 2, pp. 283–295, 2011.
Figure 14: Comparing the exact and approximate solutions at 𝑁 = [9] E. H. Doha and A. H. Bhrawy, “An efficient direct solver
12, 𝛼 = 0, 𝛽 = 1, and ] = 0.2, 0.4, 0.6, and 0.8. for multidimensional elliptic Robin boundary value problems
using a Legendre spectral-Galerkin method,” Computers &
Mathematics with Applications, vol. 64, no. 4, pp. 558–571, 2012.
7 [10] D. Funaro, “Estimates of Laguerre spectral projectors in Sobolev
Exact and approximate solutions

spaces,” in Orthogonal Polynomials and Their Applications,


6 C. Brezinski, L. Gori, and A. Ronveaux, Eds., pp. 263–266,
5 Scientific Publishing, Singapore, 1991.
[11] B.-Y. Guo, L.-L. Wang, and Z.-Q. Wang, “Generalized Laguerre
4 interpolation and pseudospectral method for unbounded
3 domains,” SIAM Journal on Numerical Analysis, vol. 43, no. 6,
pp. 2567–2589, 2006.
2
[12] M. Gülsu, B. Gürbüz, Y. Öztürk, and M. Sezer, “Laguerre poly-
1 nomial approach for solving linear delay difference equations,”
Applied Mathematics and Computation, vol. 217, no. 15, pp.
0
6765–6776, 2011.
0 1 2 3 4
[13] R. Schimming and S. Z. Rida, “Noncommutative Bell polyno-
x
mials,” International Journal of Algebra and Computation, vol.
uN (x),  = 0.2 uN (x),  = 0.6 6, no. 5, pp. 635–644, 1996.
u(x),  = 0.2 u(x),  = 0.6 [14] A. M. A. El-Sayed and S. Z. Rida, “Bell polynomials of arbitrary
uN (x),  = 0.4 uN (x),  = 0.8 (fractional) orders+,” Applied Mathematics and Computation,
u(x),  = 0.4 u(x),  = 0.8 vol. 106, no. 1, pp. 51–62, 1999.
[15] R. A. Van Gorder, “Computation of certain infinite series of the
Figure 15: Comparing the exact and approximate solutions at 𝑁 = form Σ𝑓(𝑛)𝑛𝑘 for arbitrary real-valued 𝑘,” Applied Mathematics
16, 𝛼 = 0, 𝛽 = 1, and ] = 0.2, 0.4, 0.6, and 0.8. and Computation, vol. 215, no. 3, pp. 1209–1216, 2009.
[16] B. Shahriyar, F. Ismail, S. Aghabeigi, A. Ahmadian, and S.
Salahshour, “An eigenvalueeigenvector method for solving a
system of fractional differential equations with uncertainty,”
validity and applicability of the method. The computational Mathematical Problems in Engineering, vol. 2013, Article ID
results show that the proposed methods can be effectively 579761, 11 pages, 2013.
used in numerical solution of time-dependent fractional [17] A. H. Bhrawy, M. M. Tharwat, and Y. Yildirim, “A new formula
partial differential equation and other problems on the half for fractional integrals of Chebyshev polynomials: application
line. for solving multi-term fractional differential equations,” Applied
Mathematical Modelling, vol. 37, no. 6, pp. 4245–4252, 2012.
[18] Y.-L. Jiang and X.-L. Ding, “Waveform relaxation methods for
References fractional differential equations with the Caputo derivatives,”
Journal of Computational and Applied Mathematics, vol. 238, pp.
[1] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional 51–67, 2013.
Calculus Models and Numerical Methods, Complexity, Nonlin-
[19] S. Kazem, S. Abbasbandy, and S. Kumar, “Fractional-order
earity and Chaos, World Scientific, Singapore, 2012.
Legendre functions for solving fractional-order differential
[2] K. Miller and B. Ross, An Introduction to the Fractional Calaulus equations,” Applied Mathematical Modelling. Simulation and
and Fractional Differential Equations, John Wiley & Sons, New Computation for Engineering and Environmental Systems, vol.
York, NY, USA, 1993. 37, no. 7, pp. 5498–5510, 2013.
12 Abstract and Applied Analysis

[20] J. Song, F. Yin, X. Cao, and F. Lu, “Fractional variational Communications in Nonlinear Science and Numerical Simula-
iteration method versus Adomian’s decomposition method in tion, vol. 17, no. 1, pp. 62–70, 2012.
some fractional partial differential equations,” Journal of Applied [37] K. Diethelm and N. J. Ford, “Numerical solution of the Bagley-
Mathematics, vol. 2013, Article ID 392567, 10 pages, 2013. Torvik equation,” BIT. Numerical Mathematics, vol. 42, no. 3, pp.
[21] X. Zhang, L. Liu, Y. Wu, and Y. Lu, “The iterative solutions of 490–507, 2002.
nonlinear fractional differential equations,” Applied Mathemat- [38] S. S. Ray and R. K. Bera, “Analytical solution of the Bagley
ics and Computation, vol. 219, no. 9, pp. 4680–4691, 2013. Torvik equation by Adomian decomposition method,” Applied
[22] A. H. Bhrawy, A. S. Alofi, and S. S. Ezz-Eldien, “A quadrature Mathematics and Computation, vol. 168, no. 1, pp. 398–410, 2005.
tau method for fractional differential equations with variable [39] Y. Çenesiz, Y. Keskin, and A. Kurnaz, “The solution of the
coefficients,” Applied Mathematics Letters, vol. 24, no. 12, pp. Bagley-Torvik equation with the generalized Taylor collocation
2146–2152, 2011. method,” Journal of the Franklin Institute, vol. 347, no. 2, pp. 452–
[23] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “Efficient 466, 2010.
Chebyshev spectral methods for solving multi-term fractional [40] M. A. Z. Raja, J. A. Khan, and I. M. Qureshi, “Solution of
orders differential equations,” Applied Mathematical Modelling, fractional order system of Bagley-Torvik equation using evolu-
vol. 35, no. 12, pp. 5662–5672, 2011. tionary computational intelligence,” Mathematical Problems in
[24] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A Chebyshev Engineering, vol. 2011, Article ID 675075, 18 pages, 2011.
spectral method based on operational matrix for initial and [41] S. S. Ray, “On Haar wavelet operational matrix of general order
boundary value problems of fractional order,” Computers & and its application for the numerical solution of fractional
Mathematics with Applications, vol. 62, no. 5, pp. 2364–2373, Bagley Torvik equation,” Applied Mathematics and Computa-
2011. tion, vol. 218, no. 9, pp. 5239–5248, 2012.
[25] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A new
Jacobi operational matrix: an application for solving fractional
differential equations,” Applied Mathematical Modelling, vol. 36,
no. 10, pp. 4931–4943, 2012.
[26] A. Ahmadian, M. Suleiman, S. Salahshour, and D. Baleanu, “A
Jacobi operational matrix for solving a fuzzy linear fractional
differential equation,” Advances in Difference Equations, 2013,
article 104.
[27] C. Li, F. Zeng, and F. Liu, “Spectral approximations to the
fractional integral and derivative,” Fractional Calculus and
Applied Analysis, vol. 15, no. 3, pp. 383–406, 2012.
[28] S. Yüzbaşı, “Numerical solution of the Bagley-Torvik equation
by the Bessel collocation method,” Mathematical Methods in the
Applied Sciences, vol. 36, no. 3, pp. 300–312, 2013.
[29] M. H. Atabakzadeh, M. H. Akrami, and G. H. Erjaee, “Cheby-
shev operational matrix method for solving multi-order frac-
tional ordinary differential equations,” Applied Mathematical
Modelling, 2013.
[30] A. Ahmadian, M. Suleiman, and S. Salahshour, “An opera-
tional matrix based on Legendre polynomials for solving fuzzy
fractional-order differential equations,” Abstract and Applied
Analysis, vol. 2013, Article ID 505903, 29 pages, 2013.
[31] A. H. Bhrawy and A. S. Alofi, “The operational matrix of frac-
tional integration for shifted Chebyshev polynomials,” Applied
Mathematics Letters, vol. 26, no. 1, pp. 25–31, 2013.
[32] A. H. Bhrawy, M. Alghamdi, and T. M. Taha, “A new modified
generalized Laguerre operational matrix of fractional integra-
tion for solving fractional differential equations on the half line,”
Advances in Difference Equations, vol. 2012, p. 179, 2012.
[33] B.-Y. Guo and X.-Y. Zhang, “A new generalized Laguerre
spectral approximation and its applications,” Journal of Compu-
tational and Applied Mathematics, vol. 181, no. 2, pp. 342–363,
2005.
[34] G. Szegő, Orthogonal Polynomials, vol. 23, American Mathe-
matical Society. Colloquium Publications, 1985.
[35] B.-Y. Guo and C. Zhang, “The spectral method for high order
problems with proper simulations of asymptotic behaviors at
infinity,” Journal of Computational and Applied Mathematics,
vol. 237, no. 1, pp. 269–294, 2013.
[36] A. H. Bhrawy and A. S. Alofi, “A Jacobi-Gauss collocation
method for solving nonlinear Lane-Emden type equations,”
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 816803, 11 pages
http://dx.doi.org/10.1155/2013/816803

Research Article
Numerical Modeling of Fractional-Order Biological Systems

Fathalla A. Rihan1,2
1
Department of Mathematical Sciences, College of Science, UAE University, Al Ain 15551, UAE
2
Department of Mathematics, Faculty of Science, Helwan University, Cairo 11795, Egypt

Correspondence should be addressed to Fathalla A. Rihan; frihan@uaeu.ac.ae

Received 17 May 2013; Accepted 23 June 2013

Academic Editor: Ali H. Bhrawy

Copyright © 2013 Fathalla A. Rihan. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We provide a class of fractional-order differential models of biological systems with memory, such as dynamics of tumor-immune
system and dynamics of HIV infection of CD4+ T cells. Stability and nonstability conditions for disease-free equilibrium and
positive equilibria are obtained in terms of a threshold parameter R0 (minimum infection parameter) for each model. We provide
unconditionally stable method, using the Caputo fractional derivative of order 𝛼 and implicit Euler’s approximation, to find a
numerical solution of the resulting systems. The numerical simulations confirm the advantages of the numerical technique and
using fractional-order differential models in biological systems over the differential equations with integer order. The results may
give insight to infectious disease specialists.

1. Introduction biological systems. Also, they are closely related to fractals


[16], which are abundant in biological systems. It has been
Mathematical models, using ordinary differential equations deduced in [3] that the membranes of cells of biological
with integer order, have been proven valuable in understand- organism have fractional-order electrical conductance and
ing the dynamics of biological systems. However, the behav- then are classified in groups of noninteger-order models.
ior of most biological systems has memory or aftereffects. The Fractional derivatives embody essential features of cell
modelling of these systems by fractional-order differential rheological behavior and have enjoyed greatest success in
equations has more advantages than classical integer-order the field of rheology [17]. In this paper, we propose systems
mathematical modeling, in which such effects are neglected. of FODEs for modeling the interactions of tumor-immune
Accordingly, the subject of fractional calculus (i.e., calculus system (see Section 2) and HIV infection of CD4+ T cells
of integral and derivatives of arbitrary order) has gained with immune system (see Section 3).
popularity and importance, mainly due to its demonstrated However, analytical and closed solutions of these types of
applications in numerous diverse and widespread fields of fractional equations cannot generally be obtained. As a con-
science and engineering. For example, fractional calculus has sequence, approximate and numerical techniques are playing
been successfully applied to system biology [1–5], physics important role in identifying the solution behavior of such
[6–9], chemistry and biochemistry [10], hydrology [11, 12], fractional equations and exploring their applications [18–21].
medicine [13, 14], and finance [15]. In some situations, the The Adomian decomposition method [22, 23], extrapolation
fractional-order differential equations (FODEs) models seem method [24], multistep method [25], monotone iterative
more consistent with the real phenomena than the integer- technique [26], and predictor-corrector approach [18, 27]
order models. This is due to the fact that fractional deriva- are proposed to provide numerical solutions for linear and
tives and integrals enable the description of the memory nonlinear FODEs. The monograph of [28] investigates the
and hereditary properties inherent in various materials and interconnection between fractional differential equations and
processes. Hence there is a growing need to study and use the classical differential equations of integer-order derivatives. In
fractional-order differential and integral equations. this paper, we provide a numerical approach for the resulting
Fractional-order differential equations are naturally system using implicit Euler’s scheme (see Section 4). We also
related to systems with memory which exists in most study the stability and convergence of the proposed method.
2 Abstract and Applied Analysis

We first give the definition of fractional-order integration Remark 3.


and fractional-order differentiation [29, 30]. There are several
approaches to the generalization of the notion of differenti- (i) Suppose that 𝑓(𝑡) ∈ 𝐶[𝑎, 𝑏] and 𝐷∗𝛼 𝑓(𝑡) ∈ 𝐶(𝑎, 𝑏] for
ation to fractional orders, for example, Riemann-Liouville, 0 < 𝛼 ≤ 1, and then we have
Caputo, and Generalized Functions approaches. Let 𝐿1 = 1
𝐿1 [𝑎, 𝑏] be the class of Lebesgue integrable functions on [𝑎, 𝑏], 𝑓 (𝑡) = 𝑓 (𝑎) + 𝐷𝛼 𝑓 (𝜉) (𝑡 − 𝑎)𝛼 ,
𝑎 < 𝑏 < ∞. Γ (𝛼) ∗ (4)
with 𝑎 < 𝜉 < 𝑡 ∀𝑡 ∈ (𝑎, 𝑏] .
Definition 1. The fractional integral (or the Riemann-
Liouville integral) of order 𝛽 ∈ R+ of the function 𝑓(𝑡), 𝑡 > 0
(𝑓 : R+ → R) is defined by (ii) If (i) holds and 𝐷∗𝛼 𝑓(𝑡) ≥ 0 for all 𝑡 ∈ [𝑎, 𝑏], then 𝑓(𝑡)
is nondecreasing for each 𝑡 ∈ [𝑎, 𝑏]. If 𝐷∗𝛼 𝑓(𝑡) ≤ 0
𝑡 for all 𝑡 ∈ [𝑎, 𝑏], then 𝑓(𝑡) is nonincreasing for each
1
𝐼𝑎𝛽 𝑓 (𝑡) = ∫ (𝑡 − 𝑠)𝛽−1 𝑓 (𝑠) 𝑑𝑠, 𝑡 > 0. (1) 𝑡 ∈ [𝑎, 𝑏].
Γ (𝛽) 𝑎

The fractional derivative of order 𝛼 ∈ (𝑛 − 1, 𝑛) of 𝑓(𝑡) is We next provide a class of fractional-order differential
defined by two (nonequivalent) ways: models to describe the dynamics of tumour-immune system
interactions.
(i) Riemann-Liouville fractional derivative: take frac-
tional integral of order (𝑛 − 𝛼), and then take 𝑛th 2. Fractional Model of Tumor-Immune System
derivative as follows:
Immune system is one of the most fascinating schemes from
𝑑𝑛
𝐷∗𝛼 𝑓 (𝑡) = 𝐷∗𝑛 𝐼𝑎𝑛−𝛼 𝑓 (𝑡) , 𝐷∗𝑛 = 𝑛, 𝑛 = 1, 2, . . . . the point of view of biology and mathematics. The immune
𝑑𝑡 system is complex, intricate, and interesting. It is known to be
(2) multifunctional and multipathway, so most immune effectors
do more than one job. Also each function of the immune
(ii) Caputo-fractional derivative: take 𝑛th derivative, and system is typically done by more than one effector, which
then take a fractional integral of order (𝑛 − 𝛼) makes it more robust. The reason of using FODEs is that they
are naturally related to systems with memory which exists in
𝐷𝛼 𝑓 (𝑡) = 𝐼𝑎𝑛−𝛼 𝐷∗𝑛 𝑓 (𝑡) , 𝑛 = 1, 2, . . . . (3) tumor-immune interactions.
Ordinary and delay differential equations have long been
We notice that the definition of time-fractional derivative used in modeling cancer phenomena [33–37], but fractional-
of a function 𝑓(𝑡) at 𝑡 = 𝑡𝑛 involves an integration and order differential equations have short history in modeling
calculating time-fractional derivative that requires all the past such systems with memory. The authors in [1] used a system
history, that is, all the values of 𝑓(𝑡) from 𝑡 = 0 to 𝑡 = of fractional-order differential equations in modeling cancer-
𝑡𝑛 . For the concept of fractional derivative, we will adopt immune system interaction. The model includes two immune
Caputo’s definition which is a modification of the Riemann- effectors: 𝐸1 (𝑡), 𝐸2 (𝑡) (such as cytotoxic T cells and natural
Liouville definition and has the advantage of dealing properly killer cells), interacting with the cancer cells, 𝑇(𝑡), with a
with initial value problems. The following remark addresses Holling function of type III. (Holling type III describes a
some of the main properties of the fractional derivatives and situation in which the number of prey consumed per predator
integrals (see [8, 29–31]). initially rises slowly as the density of prey increases but then
levels off with further increase in prey density. In other words
Remark 2. Let 𝛽, 𝛾 ∈ R+ and 𝛼 ∈ (0, 1). Then the response of predators to prey is depressed at low prey
density, then levels off with further increase in prey density.)
(i) if 𝐼𝑎𝛽 : 𝐿1 → 𝐿1 and 𝑓(𝑡) ∈ 𝐿1 , then 𝐼𝑎𝛽 𝐼𝑎𝛾 𝑓(𝑡) = The model takes the form
𝐼𝑎𝛽+𝛾 𝑓(𝑡);
𝐷𝛼 𝑇 = 𝑎𝑇 − 𝑟1 𝑇𝐸1 − 𝑟2 𝑇𝐸2 ,
(ii) lim𝛽 → 𝑛 𝐼𝑎𝛽 𝑓(𝑥) = 𝐼𝑎𝑛 𝑓(𝑡) uniformly on [𝑎, 𝑏], 𝑛 = 1, 2,
𝑡 𝑇2 𝐸1
3, . . ., where 𝐼𝑎1 𝑓(𝑡) = ∫0 𝑓(𝑠)𝑑𝑠; 𝐷𝛼 𝐸1 = −𝑑1 𝐸1 + , 0 < 𝛼 ≤ 1,
𝑇2 + 𝑘1 (5)
(iii) lim𝛽 → 0 𝐼𝑎𝛽 𝑓(𝑡) = 𝑓(𝑡) weakly;
𝑇2 𝐸2
(iv) if 𝑓(𝑡) is absolutely continuous on [𝑎, 𝑏], then 𝐷𝛼 𝐸2 = −𝑑2 𝐸2 + ,
lim𝛼 → 1 𝐷∗𝛼 𝑓(𝑡) = 𝑑𝑓(𝑡)/𝑑𝑡; 𝑇2 + 𝑘1

(v) thus 𝐷∗𝛼 𝑓(𝑡) = (𝑑/𝑑𝑡)𝐼∗1−𝛼 𝑓(𝑡) (Riemann-Liouville where 𝑇 ≡ 𝑇(𝑡), 𝐸1 ≡ 𝐸1 (𝑡), 𝐸2 ≡ 𝐸2 (𝑡), and 𝑎, 𝑟1 , 𝑟2 , 𝑑1 ,
sense) and 𝐷𝛼 𝑓(𝑡) = 𝐼∗1−𝛼 (𝑑/𝑑𝑡)𝑓(𝑡) (Caputo sense). 𝑑2 , 𝑘1 , and 𝑘2 are positive constants. The interaction terms
in the second and third equations of model (5) satisfy the
The generalized mean value theorem and another property crossreactivity property of the immune system. It has been
are defined in the following remark [32]. assumed that (𝑑1 𝑘1 /(1 − 𝑑1 )) ≪ (𝑑2 𝑘2 /(1 − 𝑑2 )) to avoid
Abstract and Applied Analysis 3

the nonbiological interior solution where both immune To ease the analysis and stability of the steady states with
effectors coexist. The equilibrium points of the system (5) are meaningful parameters and minimize sensitivity (or robust-
ness) of the model, we nondimensionalize the bilinear system
(8) by taking the rescaling
𝑑1 𝑘1 𝑎
E0 = (0, 0, 0) , E1 = (√ , , 0) ,
(1 − 𝑑1 ) 𝑟1
(6) 𝐸 𝑇 𝑝1 𝑇0
𝑥= , 𝑦= , 𝜔= ,
𝑑𝑘 𝑎 𝐸0 𝑇0 𝑡𝑠 𝐸0
E2 = (√ 2 2 , 0, ) .
(1 − 𝑑2 ) 𝑟2 𝑝2 𝑠1 𝑝4
𝜃= , 𝜎= , 𝑎= , (9)
𝑡𝑠 𝑡𝑠 𝐸0 𝑡𝑠
The first equilibrium E0 is the nave, the second E1 is the
memory, and the third E2 is endemic according to the value 𝑝6 𝐸0
𝑏 = 𝑝5 𝑇0 , 1= , 𝜏 = 𝑡𝑠 𝑡.
of the tumor size. Stability analysis shows that the nave state is 𝑡𝑠
unstable. However, the memory state is locally asymptotically
stable if 𝑑1 < 𝑑2 and 𝑑1 < 1. While the endemic state is
locally asymptotically stable if 𝑑2 < 𝑑1 and 𝑑2 < 1, there Therefore, after the previous substitution into (8) and replac-
is bifurcation at 𝑑1 = 1. The stability of the memory state ing 𝜏 by 𝑡, the model becomes
depends on the value of one parameter, namely, the immune
effector death rate. 𝐷𝛼1 𝑥 = 𝜎 + 𝜔𝑥𝑦 − 𝜃𝑥,
Now we modify model (5) to include three populations (10)
of the activated immune-system cells, 𝐸(𝑡); the tumor cells, 𝐷𝛼2 𝑦 = 𝑎𝑦 (1 − 𝑏𝑦) − 𝑥𝑦.
𝑇(𝑡); and the concentration of IL-2 in the single tumor-site
compartment, 𝐼𝐿 (𝑡). We consider the classic bilinear model
that includes Holling function of type I and external effector 2.1. Equilibria and Local Stability of Model (10). The steady
cells 𝑠1 and input of IL-2, 𝑠2 . (holling Type I is a linear states of the reduced model (10) are again the intersection of
relationship, where the predator is able to keep up with the null clines 𝐷𝛼1 𝑥 = 0, 𝐷𝛼2 𝑦 = 0. If 𝑦 = 0, the tumor-
increasing density of prey by eating them in direct proportion free equilibrium is at E0 = (𝑥, 𝑦) = (𝜎/𝜃, 0). This steady state
to their abundance in the environment. If they eat 10% of the is always exist, since 𝜎/𝜃 > 0. From the analysis, it is easy
prey at low density, they continue to eat 10% of them at high to prove that the tumor-free equilibrium E0 = (𝜎/𝜃, 0) of the
densities.) The interactions of the three populations are then model (10) is asymptotically stable if threshold parameter (the
governed by the fractional-order differential model: minimum tumor-clearance parameter) R0 = 𝑎𝜃/𝜎 < 1 and
unstable if R0 > 1.
𝐷𝛼1 𝐸 = 𝑠1 + 𝑝1 𝐸𝑇 − 𝑝2 𝐸 + 𝑝3 𝐸𝐼𝐿 , However, if 𝑦 ≠0, the steady states are obtained by solving

𝐷𝛼2 𝑇 = 𝑝4 𝑇 (1 − 𝑝5 𝑇) − 𝑝6 𝐸𝑇, 0 < 𝛼𝑖 ≤ 1, 𝑖 = 1, 2, 3,


𝜔𝑎𝑏𝑦2 − 𝑎 (𝜔 + 𝜃𝑏) 𝑦 + 𝑎𝜃 − 𝜎 = 0. (11)
𝐷𝛼3 𝐼𝐿 = 𝑠2 + 𝑝7 𝐸𝑇 − 𝑝8 𝐼𝐿 ,
(7) In this case, we have two endemic equilibria, E1 and E2 :

with initial conditions: 𝐸(0) = 𝐸0 , 𝑇(0) = 𝑇0 , and 𝐼𝐿 (0) =


𝐼𝐿 0 . The parameter 𝑝1 is the antigenicity rate of the tumor −𝑎 (𝑏𝜃 − 𝜔) − √Δ
E1 = (𝑥1 , 𝑦1 ) , where 𝑥1 = ,
(immune response to the appearance of the tumor), and 𝑠1 2𝜔
is the external source of the effector cells, with rate of death
𝑝2 . The parameter 𝑝4 incorporates both multiplication and 𝑎 (𝑏𝜃 + 𝜔) + √Δ
𝑦1 = ,
death of tumor cells. The maximal carrying capacity of the 2𝑎𝑏𝜔
(12)
biological environment for tumor cell is 𝑝5−1 ; 𝑝3 is considered −𝑎 (𝑏𝜃 − 𝜔) + √Δ
as the cooperation rate of effector cells with interleukin-2 E2 = (𝑥2 , 𝑦2 ) , where 𝑥2 = ,
parameter; 𝑝6 is the rate of tumor cells; 𝑝7 is the competition 2𝜔
rate between the effector cells and the tumor cells. External 𝑎 (𝑏𝜃 + 𝜔) − √Δ
input of IL-2 into the system is 𝑠2 , and the rate loss parameter 𝑦2 =
2𝑎𝑏𝜔
of IL-2 cells is 𝑝8 .
In the absence of immunotherapy with IL-2, we have
with Δ = 𝑎2 (𝑏𝜃 − 𝜔)2 + 4𝜎𝜔𝑎𝑏 > 0. The Jacobian matrix of
𝐷𝛼1 𝐸 = 𝑠1 + 𝑝1 𝐸𝑇 − 𝑝2 𝐸, the system (10) at the endemic equilibrium E1 is

0 < 𝛼𝑖 ≤ 1, 𝑖 = 1, 2, (8)
𝜔𝑦 − 𝜃 𝜔𝑥1
𝐽 (E1 ) = ( 1 ). (13)
𝐷𝛼2 𝑇 = 𝑝4 𝑇 (1 − 𝑝5 𝑇) − 𝑝6 𝐸𝑇. −𝑦1 𝑎 − 2𝑎𝑏𝑦1 − 𝑥1
4 Abstract and Applied Analysis

Proposition 4. Assume that the endemic equilibrium E1 the healthy CD4+ T-cells is (1 − ((𝐻 + 𝐼)/𝐻max )), and the
exists and has nonnegative coordinates. If R0 = 𝑎𝜃/𝜎 < 1, proliferation of infected cells is neglected. The parameter 𝑠 is
then tr(𝐽(E1 )) > 0 and E1 is unstable. the source of CD4+ T cells from precursors, 𝜇𝐻 is the natural
death rate of CD4+ T cells (𝜇𝐻𝐻max > 𝑠, cf. [39, page 85]),
Proof. Since 𝑟 is their growth rate (thus, 𝑟 > 𝜇𝐻 in general), and 𝐻max is
their carrying capacity. The parameter 𝑘1 represents the rate
𝜔2 − 𝜔 (𝑎𝑏 + 𝑏𝜃) − 𝑎𝑏2 𝜃 of infection of T cells with free virus. 𝑘1󸀠 is the rate at which
tr (𝐽 (E1 )) = infected cells become actively infected. 𝜇𝐼 is a blanket death
2𝑏𝜔
term for infected cells to reflect the assumption that we do not
𝜔 − 𝑎𝑏 √ 2 initially know whether the cells die naturally or by bursting.
+ 𝑎 (𝑏𝜃 + 𝜔)2 − 4𝑎𝑏𝜔 (𝑎𝜃 − 𝜎),
2𝑎𝑏𝜔 In addition, 𝜇𝑏 is the lytic death rate for infected cells. Since
(14) 𝑀 viral particles are released by each lysing cell, this term is
multiplied by the parameter 𝑀 to represent the source for free
then inequality tr(𝐽(E1 )) > 0 is true if virus (assuming a one-time initial infection). Finally, 𝜇𝑉 is the
loss rate of virus. The initial conditions for infection by free
𝑎 [𝜔2 − 𝜔 (𝑎𝑏 + 𝑏𝜃) − 𝑎𝑏2 𝜃] virus are 𝐻(0) = 𝐻0 , 𝐼(0) = 𝐼0 , and 𝑉(0) = 𝑉0 .
(15)
Theorem 6. According to Remark 3 and the fact that 𝑠 ≥ 0,
> (𝑎𝑏 − 𝜔) √𝑎2 (𝑏𝜃 + 𝜔)2 − 4𝑎𝑏𝜔 (𝑎𝜃 − 𝜎). then the system (16) has a unique solution (𝐻, 𝐼, 𝑉)𝑇 which
remains in R3+ and bounded by 𝐻max ; [11].
Therefore, when 𝑎𝜃 < 𝜎, we have 𝜔2 −𝜔𝑏(𝑎+𝜃)−𝑎𝜃𝑏2 > 0, and
hence both sides of the inequality are positive. Therefore if the 3.1. Equilibria and Local Stability of Model (16). To evaluate
equilibrium point E1 exists and has nonnegative coordinates, the equilibrium points of system (16), we put 𝐷𝛼1 𝐻(𝑡) =
then tr(𝐽(E1 )) > 0 and the point (E1 ) is unstable, whenever 𝐷𝛼2 𝐼(𝑡) = 𝐷𝛼3 𝑉(𝑡) = 0. Then the infection-free equilibrium
R0 = 𝑎𝜃/𝜎 < 1. point (uninfected steady state) is E∗0 = (𝐻0 , 0, 0), and
endemic equilibrium point (infected steady state) is E∗+ =
Similarly, we arrive at the following proposition. (𝐻∗ , 𝐼∗ , 𝑉∗ ), where

Proposition 5. If the point E2 exists and has nonnegative


𝜇𝑉𝜇𝐼 𝑘1󸀠 𝐻∗ 𝑉∗
coordinates, then it is asymptotically stable. 𝐻∗ = , 𝐼∗ = ,
𝑘1󸀠 𝑀𝜇𝑏 − 𝑘1 𝜇𝐼 𝜇𝐼
We next examine a fractional-order differential model for (17)
2
HIV infection of CD4+ T cells. 𝜇𝐼 [(𝑠 + (𝑟 − 𝜇𝐻) 𝐻∗ ) 𝐻max − 𝑟𝐻∗ ]
𝑉∗ = .
𝐻∗ [𝑘1󸀠 𝑟𝐻∗ − 𝑘1 𝜇𝐼 𝐻max ]
3. Fractional Model of HIV Infection of
CD4+ T Cells The Jacobian matrix 𝐽(E∗0 ) for system (16) evaluated at the
uninfected steady state E∗0 is then given by
As mentioned before, many mathematical models have
been developed to describe the immunological response to
infection with human immunodeficiency virus (HIV), using 𝐽 (E∗0 )
ordinary differential models (see, e.g., [38, 39]) and delay
differential models [40, 41]. In this section, we extend the −𝜇𝐻 + 𝑟 − 2𝑟𝐻0 −𝑟𝐻0
−𝑘1 𝐻0
analysis and replace the original system of Perelson et al. [39] 𝐻max 𝐻max (18)
by an equivalent fractional-order differential model of HIV =( 0 −𝜇𝐼 𝑘1󸀠 𝐻0 ).
infection of CD4+ T cells that consists of three equations:
0 𝑀𝜇𝑏 − (𝑘1 𝐻0 + 𝜇𝑉)
𝐻+𝐼
𝐷𝛼1 𝐻 = 𝑠 − 𝜇𝐻𝐻 + 𝑟𝐻 (1 − ) − 𝑘1 𝑉𝐻,
𝐻max
Let us introduce the following definition and assumption to
𝐷𝛼2 𝐼 = 𝑘1󸀠 𝑉𝐻 − 𝜇𝐼 𝐼, ease the analysis.
(16)
𝛼3 Definition 7. The threshold parameter R∗0 (the minimum
𝐷 𝑉 = 𝑀𝜇𝑏 𝐼 − 𝑘1 𝑉𝐻 − 𝜇𝑉𝑉,
infection-free parameter) is the parameter that has the
0.5 < 𝛼𝑖 ≤ 1, 𝑖 = 1, 2, 3. property that if R∗0 < 1, then the endemic infected state does
not exist, while if R∗0 > 1, the endemic infected state persists,
Here 𝐻 = 𝐻(𝑡) represents the concentration of healthy CD4+ where
T cells at time 𝑡, 𝐼 = 𝐼(𝑡) represents the concentration of
infected CD4+ T cells, and 𝑉 = 𝑉(𝑡) is the concentration 𝑘1󸀠 𝜇𝑏 𝑀𝐻0
R∗0 = . (19)
of free HIV at time 𝑡. In this system, the logistic growth of 𝜇𝐼 (𝜇𝑉 + 𝑘1 𝐻0 )
Abstract and Applied Analysis 5

We also assume that Then the characteristic equation of the linearized system is
2−1
{𝑟 − 𝜇𝐻 + [(𝑟 − 𝜇𝐻) + 4𝑟𝑠𝐻max ]
1/2
} 𝑃 (𝜆) = 𝜆3 + 𝑎1 𝜆2 + 𝑎2 𝜆 + 𝑎3 = 0, (25)
𝐻0 = . (20)
−1
2𝑟𝐻max 𝑎1 = 𝜇𝐼 + 𝜇𝑉 + 𝑘1 𝐻∗ + 𝐿∗ ,

𝑎2 = 𝐿∗ (𝜇𝐼 + 𝜇𝑉 + 𝑘1 𝐻∗ ) + 𝜇𝐼 (𝜇𝑉 + 𝑘1 𝐻∗ )
The uninfected steady state is asymptotically stable if all of the
eigenvalues 𝜆 of the Jacobian matrix 𝐽(E∗0 ), given by (18), have 𝑟𝑉∗
negative real parts. The characteristic equation det(𝐽(E0 ) − − 𝑘12 𝐻∗ 𝑉∗ − 𝑘1󸀠 𝐻∗ (𝑀𝜇𝑏 − ),
𝐻max (26)
𝐼) = 0 becomes

𝑟𝜇𝑉 𝑉
2𝑟𝐻0 𝑎3 = 𝑘1󸀠 𝐻∗ [𝑘1 𝑀𝜇𝑏 𝑉∗ − − 𝐿∗ 𝑀𝜇𝑏 ]
(𝜆 + 𝜇𝐻 − 𝑟 + ) (𝜆2 + 𝐵𝜆 + 𝐶) = 0, (21) 𝐻max
𝐻max
+ 𝐿∗ 𝜇𝐼 (𝜇𝑉 + 𝑘1 𝐻∗ ) − 𝜇𝐼 𝑘12 𝐻∗ 𝑉∗ .
where 𝐵 = 𝜇𝐼 + 𝑘𝐼 𝐻0 + 𝜇𝑉 and 𝐶 = 𝜇𝐼 (𝑘1 𝐻0 + 𝜇𝑉 ) − 𝑘1󸀠 𝜇𝑏 𝑀𝐻0 .
Hence, the three roots of the characteristic equation (21) are The infected steady state E∗+ is asymptotically stable if all of
the eigenvalues have negative real parts. This occurs if and
𝑟𝐻0 2 only if Routh-Hurwitz conditions are satisfied; that is, 𝑎1 > 0,
𝜆 1 = −𝜇𝐻 + 𝑟 − ≡ −√(𝑟 − 𝜇𝐻) + 4𝑟𝑠𝐻max
−1 < 0,
𝑎3 > 0, and 𝑎1 𝑎2 > 𝑎3 .
𝐻max
(22)
1
𝜆 2,3 = [−𝐵 ± √𝐵2 − 4𝐶] . 4. Implicit Euler’s Scheme for FODEs
2
Since most of the fractional-order differential equations do
Proposition 8. If R∗0 ≡ (𝑘1󸀠 𝜇𝑏 𝑀𝐻0 )/𝜇𝐼 (𝜇𝑉 + 𝑘1 𝐻0 ) < 1, then not have exact analytic solutions, approximation and numer-
𝐶 > 0 and the three roots of the characteristic equation (21) will ical techniques must be used. Several numerical methods
have negative real parts. have been proposed to solve the fractional-order differential
equations [18, 42, 43]. In addition, most of resulting biological
Corollary 9. In case of uninfected steady state E∗0 , one has systems are stiff. (One definition of the stiffness is that the
three cases. global accuracy of the numerical solution is determined by
stability rather than local error, and implicit methods are
(i) If R∗0 < 0, the uninfected state is asymptotically more appropriate for it.) The stiffness often appears due to
stable and the infected steady state E+ does not exist the differences in speed between the fastest and slowest
(unphysical). components of the solutions and stability constraints. In
(ii) If R∗0 = 1, then 𝐶 = 0, and from (21) implies that one addition, the state variables of these types of models are
eigenvalue 𝜆 = 0 and the remaining two eigenvalues very sensitive to small perturbations (or changes) in the
have negative real parts. The uninfected and infected parameters occur in the model. Therefore, efficient use of a
steady states collide, and there is a transcritical bifur- reliable numerical method that based in general on implicit
cation. formulae for dealing with stiff problems is necessary.
Consider biological models in the form of a system of
(iii) If R∗0 > 1, then 𝐶 < 0, and thus at least one eigenvalue FODEs of the form
will be positive real root. Thus, the uninfected state
E0 is unstable, and the endemically infected state E∗+ 𝐷𝛼 𝑋 (𝑡) = 𝐹 (𝑡, 𝑋 (𝑡) , P) , 𝑡 ∈ [0, 𝑇] , 0 < 𝛼 ≤ 1,
emerges. (27)
𝑋 (0) = 𝑋0 .
To study the local stability of the positive infected steady
states E∗+ for R∗0 > 1, we consider the linearized system of Here 𝑋(𝑡) = [𝑥1 (𝑡), 𝑥2 (𝑡), . . . , 𝑥𝑛 (𝑡)]𝑇 , P is the set of param-
(16) at E∗+ . The Jacobian matrix at E∗+ becomes eters appear in the model, and 𝐹(𝑡, 𝑋(𝑡)) satisfies the Lipschitz
condition
−𝑟𝐻∗ ‖𝐹 (𝑡, 𝑋 (𝑡) , P) − 𝐹 (𝑡, 𝑌 (𝑡) , P)‖ ≤ 𝐾 ‖𝑋 (𝑡) − 𝑌 (𝑡)‖ ,
−𝐿∗ −𝑘1 𝐻∗
𝐻max
󸀠 ∗
𝐾 > 0,
𝐽 (E∗+ ) = (𝑘1 𝑉 −𝜇𝐼 𝑘1󸀠 𝐻∗ ). (23) (28)
𝑘1 𝑉∗ 𝑀𝜇𝑏 − (𝑘1 𝐻∗ + 𝜇𝑉 ) where 𝑌(𝑡) is the solution of the perturbed system.
( )
Theorem 10. Problem (27) has a unique solution provided that
Here the Lipschitz condition (28) is satisfied and 𝐾𝑇𝛼 /Γ(𝛼 + 1) < 1.
𝑟 (2𝐻∗ + 𝐼∗ ) Proof. Using the definitions of Section 1, we can apply a
𝐿∗ = − [𝜇𝐻 − 𝑟 + 𝑘1 𝑉∗ + ]. (24)
𝐻max fractional integral operator to the differential equation (27)
6 Abstract and Applied Analysis

and incorporate the initial conditions, thus converting the 1


=
equation into the equivalent equation (1 − 𝛼) Γ (1 − 𝛼)
𝑛 𝑗 𝑗−1
1 𝑡 𝑥𝑖 − 𝑥𝑖
𝑋 (𝑡) = 𝑋 (0) + ∫ (𝑡 − 𝑠)𝛼−1 𝐹 (𝑠, 𝑋 (𝑠) , P) 𝑑𝑠 (29) × ∑ {[ + 𝑂 (ℎ)]
Γ (𝛼) 0 𝑗=1 ℎ

which also is a Volterra equation of the second kind. Define


1−𝛼 1−𝛼
the operator L, such that × [(𝑛 − 𝑗 + 1) − (𝑛 − 𝑗) ] } ℎ1−𝛼
𝑡
1
L𝑋 (𝑡) = 𝑋 (0) + ∫ (𝑡 − 𝑠)𝛼−1 𝐹 (𝑠, 𝑋 (𝑠) , P) 𝑑𝑠. 1 1
Γ (𝛼) 0 =
(30) (1 − 𝛼) Γ (1 − 𝛼) ℎ𝛼
𝑛
Then, we have 𝑗 𝑗−1 1−𝛼 1−𝛼
× ∑ [𝑥𝑖 − 𝑥𝑖 ] [(𝑛 − 𝑗 + 1) − (𝑛 − 𝑗) ]
𝑗=1
‖L𝑋 (𝑡) − L𝑌 (𝑡)‖
1
1 𝑡 +
≤ ∫ (𝑡 − 𝑠)𝛼−1 ‖𝐹 (𝑠, 𝑋 (𝑠) , P) − 𝐹 (𝑠, 𝑌 (𝑠) , P)‖ 𝑑𝑠 (1 − 𝛼) Γ (1 − 𝛼)
Γ (𝛼) 0
𝑛
𝑡 𝑗 𝑗−1 1−𝛼 1−𝛼

𝐾
∫ (𝑡 − 𝑠)𝛼−1 sup |𝑋 (𝑠) − 𝑌 (𝑠)| 𝑑𝑠 × ∑ [𝑥𝑖 − 𝑥𝑖 ] [(𝑛 − 𝑗 + 1) − (𝑛 − 𝑗) ] 𝑂 (ℎ2−𝛼 ) .
Γ (𝛼) 0 𝑠∈[0,𝑇] 𝑗=1
(33)
𝑡
𝐾
≤ ‖𝑋 − 𝑌‖ ∫ 𝑠𝛼−1 𝑑𝑠
Γ (𝛼) 0 Setting
𝐾 1 1
≤ ‖𝑋 − 𝑌‖ 𝑇𝛼 . G (𝛼, ℎ) = , 𝜔𝑗𝛼 = 𝑗1−𝛼 − (𝑗 − 1)
1−𝛼
,
Γ (𝛼) (1 − 𝛼) Γ (1 − 𝛼) ℎ𝛼
(31)
(where 𝜔1𝛼 = 1) ,
𝛼
Thus for 𝐾𝑇 /Γ(𝛼 + 1) < 1, we have (34)

‖L𝑋 (𝑡) − L𝑌 (𝑡)‖ ≤ ‖𝑋 − 𝑌‖ . (32) then the first-order approximation method for the compu-
tation of Caputo’s fractional derivative is then given by the
This implies that our problem has a unique solution. expression

However, converted Volterra integral equation (29) is 𝑛


𝑛−𝑗+1 𝑛−𝑗
with a weakly singular kernel, such that a regularization is 𝐷∗𝛼 𝑥𝑖 (𝑡𝑛 ) = G (𝛼, ℎ) ∑ 𝜔𝑗𝛼 (𝑥𝑖 − 𝑥𝑖 ) + 𝑂 (ℎ) . (35)
not necessary any more. It seems that there exists only a very 𝑗=1
small number of software packages for nonlinear Volterra
equations. In our case the kernel may not be continuous, From the analysis and numerical approximation, we also
and therefore the classical numerical algorithms for the arrive at the following proposition.
integral part of (29) are unable to handle the solution of
(27). Therefore, we implement the implicit Euler’s scheme to Proposition 11. The presence of a fractional differential order
approximate the fractional-order derivative. in a differential equation can lead to a notable increase in
Given model (27) and mesh points T = {𝑡0 , 𝑡1 , . . . , 𝑡𝑁}, the complexity of the observed behaviour, and the solution is
such that 𝑡0 = 0 and 𝑡𝑁 = 𝑇. Then a discrete approximation continuously depends on all the previous states.
to the fractional derivative can be obtained by a simple
quadrature formula, using the Caputo fractional derivative 4.1. Stability and Convergence. We here prove that the
(3) of order 𝛼, 0 < 𝛼 ≤ 1, and using implicit Euler’s fractional-order implicit difference approximation (35) is
approximation as follows (see [44]): unconditionally stable. It follows then that the numerical
solution converges to the exact solution as ℎ → 0.
𝐷∗𝛼 𝑥𝑖 (𝑡𝑛 ) In order to study the stability of the numerical method,
let us consider a test problem of linear scaler fractional
𝑡
1 𝑑𝑥𝑖 (𝑠) −𝛼 differential equation
= ∫ (𝑡𝑛 − 𝑠) 𝑑𝑠
Γ (1 − 𝛼) 0 𝑑𝑠
𝑛 𝑗ℎ 𝑗 𝑗−1 𝐷∗𝛼 𝑢 (𝑡) = 𝜌0 𝑢 (𝑡) + 𝜌1 , 𝑈 (0) = 𝑈0 , (36)
1 𝑥 − 𝑥𝑖
≈ ∑∫ [ 𝑖 + 𝑂 (ℎ)] (𝑛ℎ − 𝑠)−𝛼 𝑑𝑠
Γ (1 − 𝛼) 𝑗=1 (𝑗−1)ℎ ℎ such that 0 < 𝛼 ≤ 1, and 𝜌0 < 0, 𝜌1 > 0 are constants.
Abstract and Applied Analysis 7

1.5 2.5
𝛼 = 0.75 𝛼 = 0.75

2
E2 (t)

T(t), E1 (t), E2 (t)


1
T(t), E1 (t), E2 (t)

1.5

E1 (t)
1
0.5 T(t) T(t)
0.5

E1 (t) E2 (t)
0 0
0 20 40 60 80 100 0 20 40 60 80 100
Time
Time

1.8 1.6
𝛼 = 0.95 𝛼 = 0.75
1.6 1.4
1.4
1.2
Effector cells E1 (t) E2 (t)

1.2 E2 (t)
T(t), E1 (t), E2 (t)

1
1
0.8
0.8
0.6
0.6
T(t) 0.4
0.4

0.2 0.2
E1 (t)
0 0
0 20 40 60 80 100 0 0.5 1 1.5 2 2.5
Time Tumor

Figure 1: Numerical simulations of the FODEs model (5) when 𝛼 = 0.75 and 𝛼 = 0.95 with (𝑎 = 𝑟1 = 𝑟2 = 1; 𝑑1 = 0.3; 𝑑2 = 0.7; 𝑘1 = 0.3;
𝑘2 = 0.7) and when 𝑑1 = 0.7; 𝑑2 = 0.3. The system converges to the stable steady states E1 , E2 . The fractional derivative damps the oscillation
behavior.

Theorem 12. The fully implicit numerical approximation (35), Since (1 − (𝜌0 /G𝛼,ℎ )) ≥ 1 for all G𝛼,ℎ , then
to test problem (36) for all 𝑡 ≥ 0, is consistent and uncon-
ditionally stable. 𝜁1 ≤ 𝜁0 , (39)
𝑛
Proof. We assume that the approximate solution of (36) is of
𝜁𝑛 ≤ 𝜁𝑛−1 + ∑𝜔𝑗(𝛼) (𝜁𝑛−𝑗 − 𝜁𝑛−𝑗+1 ) , 𝑛 ≥ 2.
the form 𝑢(𝑡𝑛 ) ≈ 𝑈𝑛 ≡ 𝜁𝑛 , and then (36) can be reduced to (40)
𝑗=2

𝜌0 Thus, for 𝑛 = 2, the previous inequality implies


(1 − )𝜁
G𝛼,ℎ 𝑛
𝜁2 ≤ 𝜁1 + 𝜔2(𝛼) (𝜁0 − 𝜁1 ) . (41)
(37)
𝑛
𝜌1
= 𝜁𝑛−1 + ∑𝜔𝑗(𝛼) (𝜁𝑛−𝑗 − 𝜁𝑛−𝑗+1 ) + , 𝑛≥2 Using the relation (39) and the positivity of the coefficients
𝑗=2 G𝛼,ℎ 𝜔2 , we get

𝜁2 ≤ 𝜁1 . (42)
or
Repeating the process, we have from (40)
𝜁𝑛−1 + ∑𝑛𝑗=2 𝜔𝑗(𝛼) (𝜁𝑛−𝑗 − 𝜁𝑛−𝑗+1 ) + 𝜌1 /G𝛼,ℎ
𝜁𝑛 = , 𝑛 ≥ 2. 𝑛
(1 − (𝜌0 /G𝛼,ℎ )) 𝜁𝑛 ≤ 𝜁𝑛−1 + ∑𝜔𝑗(𝛼) (𝜁𝑛−𝑗 − 𝜁𝑛−𝑗+1 ) ≤ 𝜁𝑛−1 , (43)
(38) 𝑗=2
8 Abstract and Applied Analysis

4 12

3.5
10
3

Tumor cells, T(t)


8
Effector cells, E(t)

2.5

2 6

1.5
4
1
2
0.5

0
0 50 100 150 0 50 100 150
Time Time
𝛼=1 𝛼=1
𝛼 = 0.9 𝛼 = 0.9
𝛼 = 0.7 𝛼 = 0.7
12

10

8
Tumor cells, T(t)

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Effector cells, E(t)

𝛼=1
𝛼 = 0.9
𝛼 = 0.7

Figure 2: Numerical simulations of the FODEs model (10) when 𝑠 = 0.1181, 𝜔 = 0.1184, 𝜃 = 0.1747, 𝑟 = 0.636, and 𝑏 = 0.002. 𝛼1 = 𝛼2 =
1, 0.9, 0.7. The endemic state E2 is locally asymptotically stable when R0 = 𝑎𝜃/𝜎 > 1. The fractional derivative damps the oscillation behavior.

since each term in the summation is negative. Thus 𝜁𝑛 ≤ equilibrium points (for infection-free and endemic cases) are
𝜁𝑛−1 ≤ 𝜁𝑛−2 ≤ ⋅ ⋅ ⋅ ≤ 𝜁0 . With the assumption that 𝜁𝑛 = |𝑈𝑛 | ≤ the same in both integer-order (when 𝛼 = 1) and fractional-
𝜁0 = |𝑈0 |, which entails ‖𝑈𝑛 ‖ ≤ ‖𝑈0 ‖, we have stability. order (𝛼 < 1) models. We notice that, in the endemic steady
states, the fractional-order derivative damps the oscillation
Of course this numerical technique can be used both for behavior. From the graphs, we can see that FODEs have rich
linear and for nonlinear problems, and it may be extended dynamics and are better descriptors of biological systems
to multiterm FODEs. For more details about stability and than traditional integer-order models.
convergence of the fractional Euler method, we refer to [19, The numerical simulations for the HIV FODE model (16)
45]. (with parameter values given in the captions) are displayed in
Figure 4. We note that the solution of the model, with various
4.2. Numerical Simulations. We employed the implicit Euler’s values of 𝛼, continuously depends on the time-fractional
scheme (35) to solve the resulting biological systems of derivative but arrives to the equilibrium points. The displayed
FODEs (5), (10), and (16). Interesting numerical simulations solutions, in the figure, confirm that the fractional order of
of the fractional tumor-immune models (5) and (10), with the derivative plays the role of time delay in the system. We
step size ℎ = 0.05 and 0.5 < 𝛼 ≤ 1 and parameters values should also note that although the equilibrium points are
given in the captions, are displayed in Figures 1, 2, and 3. The the same for both integer-order and fractional-order models,
Abstract and Applied Analysis 9

0.75 0.35
0.7
0.3
0.65
0.6 0.25
Effector cells, E(t)

Tumor cells, T(t)


0.55 0.2
0.5
0.15
0.45
0.4 0.1
0.35
0.05
0.3
0.25 0
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
Time Time

𝛼=1 𝛼=1
𝛼 = 0.9 𝛼 = 0.9
𝛼 = 0.8 𝛼 = 0.8
0.35

0.3

0.25
Tumor cells, T(t)

0.2

0.15

0.1

0.05

0
0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75
Effector cells, E(t)

𝛼=1
𝛼 = 0.9
𝛼 = 0.8

Figure 3: Numerical simulations of the FODEs model (10) when 𝑠 = 0.1181, 𝜔 = 0.1184, 𝜃 = 0.3747, 𝑟 = 1.636, and 𝑏 = 0.002. 𝛼1 = 𝛼2 =
1, 0.9, 0.7. The infection-free steady state E0 is locally asymptotically stable when R0 = 𝑎𝜃/𝜎 < 1.

the solution of the fractional order model tends to the fixed We provided unconditionally stable numerical technique,
point over a longer period of time. using the Caputo fractional derivative of order 𝛼 and implicit
Euler’s approximation, for the resulting system. The numer-
5. Conclusions ical technique is suitable for stiff problems. The solution
of the system at any time 𝑡∗ is continuously depends on
In fact, fractional-order differential equations are generaliza- all the previous states at 𝑡 ≤ 𝑡∗ . We have seen that the
tions of integer-order differential equations. Using fractional- presence of the fractional differential order leads to a notable
order differential equations can help us to reduce the errors increase in the complexity of the observed behaviour and play
arising from the neglected parameters in modeling biological the role of time lag (or delay term) in ordinary differential
systems with memory and systems distributed parameters. In model. We have obtained stability conditions for disease-
this paper, we presented a class of fractional-order differential free equilibrium and nonstability conditions for positive
models of biological systems with memory to model the equilibria. We should also mention that one of the basic
interaction of immune system with tumor cells and with reasons of using fractional-order differential equations is that
HIV infection of CD4+ T cells. The models possess non- fractional-order differential equations are, at least, as stable
negative solutions, as desired in any population dynamics. as their integer-order counterpart. In addition, the presence
We obtained the threshold parameter R0 that represents the of a fractional differential order in a differential equation can
minimum tumor-clearance parameter or minimum infection lead to a notable increase in the complexity of the observed
free for each model. behaviour, and the solution continuously depends on all
10 Abstract and Applied Analysis

1200

Infected CD4+ T-cell density


T-cell population size 700
1000
Uninfected CD4+

𝛼=1 600
800 500 𝛼=1
600 400
𝛼 = 0.9 300
400
𝛼 = 0.7 200 𝛼 = 0.9
𝛼 = 0.7
200 100
0 0
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
Time Time
×104
Initial density of HIV RNA

10
×104
8 10
𝛼=1 𝛼=1

V(t)
6
5
4 𝛼 = 0.9
𝛼 = 0.9 0
2 𝛼 = 0.7 800 𝛼 = 0.7
600 1000 1200
400 600 800
0 200 200 400
0 20 40 60 80 100 120 140 160 180 200 I(t) 0 0
H(t)
Time

Figure 4: Numerical simulations of the FODEs model (16) when 𝑠 = 20 day−1 mm−3 , 𝜇𝐻 = 0.02 day−1 , 𝜇𝐼 = 0.26 day−1 , 𝜇𝑏 = 0.24 day−1 ,
𝜇𝑉 = 2.4 day−1 , 𝑘1 = 2.4 × 10−5 mm3 day−1 , 𝑘1󸀠 = 2 × 10−5 mm3 day−1 , 𝑟 = 0.3 day−1 , 𝐻max = 1500 mm−3 , and 𝑀 = 1400. The trajectories of
the system approach to the steady state E∗+ .

the previous states. The analysis can be extended to other [8] R. Hilfer, Ed., Applications of Fractional Calculus in Physics,
models related to the immune response systems. World Scientific, River Edge, NJ, USA, 2000.
[9] G. M. Zaslavsky, “Chaos, fractional kinetics, and anomalous
transport,” Physics Reports, vol. 371, no. 6, pp. 461–580, 2002.
Acknowledgments [10] S. B. Yuste, L. Acedo, and K. Lindenberg, “Reaction front in an
A + B → C reaction-subdiffusion process,” Physical Review E,
The work is funded by the UAE University, NRF Project no. vol. 69, no. 3, Article ID 036126, pp. 1–36126, 2004.
NRF-7-20886. The author is grateful to Professor Changpin [11] W. Lin, “Global existence theory and chaos control of fractional
Li’s valuable suggestions and referees’ constructive com- differential equations,” Journal of Mathematical Analysis and
ments. Applications, vol. 332, no. 1, pp. 709–726, 2007.
[12] H. Sheng, Y. Q. Chen, and T. S. Qiu, Fractional Processes and
Fractional-Order Signal Processing, Springer, New York, NY,
References USA, 2012.
[13] K. Assaleh and W. M. Ahmad, “Modeling of speech signals
[1] E. Ahmed, A. Hashish, and F. A. Rihan, “On fractional order using fractional calculus,” in Proceedings of the 9th International
cancer model,” Journal of Fractional Calculus and Applied Symposium on Signal Processing and its Applications (ISSPA ’07),
Analysis, vol. 3, no. 2, pp. 1–6, 2012. Sharjah, United Arab Emirates, February 2007.
[2] A. A. M. Arafa, S. Z. Rida, and M. Khalil, “Fractional modeling [14] Y. Ferdi, “Some applications of fractional order calculus to
dynamics of HIV and CD4+ T-cells during primary infection,” design digital filters for biomedical signal processing,” Journal
Nonlinear Biomedical Physics, vol. 6, no. 1, article 1, 2012. of Mechanics in Medicine and Biology, vol. 12, no. 2, Article ID
[3] K. S. Cole, “Electric conductance of biological systems,” Cold 12400088, 13 pages, 2012.
Spring Harbor Symposia on Quantitative Biology, pp. 107–116, [15] W.-C. Chen, “Nonlinear dynamics and chaos in a fractional-
1993. order financial system,” Chaos, Solitons and Fractals, vol. 36, no.
[4] A. M. A. El-Sayed, A. E. M. El-Mesiry, and H. A. A. El-Saka, “On 5, pp. 1305–1314, 2008.
[16] A. Rocco and B. J. West, “Fractional calculus and the evolution
the fractional-order logistic equation,” Applied Mathematics
of fractal phenomena,” Physica A, vol. 265, no. 3, pp. 535–546,
Letters, vol. 20, no. 7, pp. 817–823, 2007.
1999.
[5] H. Xu, “Analytical approximations for a population growth [17] V. D. Djordjević, J. Jarić, B. Fabry, J. J. Fredberg, and D.
model with fractional order,” Communications in Nonlinear Stamenović, “Fractional derivatives embody essential features
Science and Numerical Simulation, vol. 14, no. 5, pp. 1978–1983, of cell rheological behavior,” Annals of Biomedical Engineering,
2009. vol. 31, no. 6, pp. 692–699, 2003.
[6] L. Debnath, “Recent applications of fractional calculus to [18] K. Diethelm, N. J. Ford, and A. D. Freed, “A predictor-corrector
science and engineering,” International Journal of Mathematics approach for the numerical solution of fractional differential
and Mathematical Sciences, no. 54, pp. 3413–3442, 2003. equations,” Nonlinear Dynamics, vol. 29, no. 1–4, pp. 3–22, 2002.
[7] A. M. A. El-Sayed, “Nonlinear functional-differential equations [19] C. Li and F. Zeng, “The finite difference methods for fractional
of arbitrary orders,” Nonlinear Analysis. Theory, Methods & ordinary differential equations,” Numerical Functional Analysis
Applications, vol. 33, no. 2, pp. 181–186, 1998. and Optimization, vol. 34, no. 2, pp. 149–179, 2013.
Abstract and Applied Analysis 11

[20] J. C. Trigeassou, N. Maamri, J. Sabatier, and A. Oustaloup, “State [39] A. S. Perelson, D. E. Kirschner, and R. De Boer, “Dynamics of
variables and transients of fractional order differential systems,” HIV infection of CD4+ T cells,” Mathematical Biosciences, vol.
Computers & Mathematics with Applications, vol. 64, no. 10, pp. 114, no. 1, pp. 81–125, 1993.
3117–3140, 2012. [40] G. A. Bocharov and F. A. Rihan, “Numerical modelling in
[21] X. Zhang and Y. Han, “Existence and uniqueness of positive biosciences using delay differential equations,” Journal of Com-
solutions for higher order nonlocal fractional differential equa- putational and Applied Mathematics, vol. 125, no. 1-2, pp. 183–
tions,” Applied Mathematics Letters, vol. 25, no. 3, pp. 555–560, 199, 2000.
2012. [41] R. V. Culshaw and S. Ruan, “A delay-differential equation model
[22] A. M. A. El-Sayed and M. Gaber, “The Adomian decomposition of HIV infection of CD4+ T-cells,” Mathematical Biosciences,
method for solving partial differential equations of fractal order vol. 165, no. 1, pp. 27–39, 2000.
in finite domains,” Physics Letters A, vol. 359, no. 3, pp. 175–182, [42] R. Anguelov and J. M.-S. Lubuma, “Nonstandard finite dif-
2006. ference method by nonlocal approximation,” Mathematics and
[23] H. Jafari and V. Daftardar-Gejji, “Solving a system of nonlinear Computers in Simulation, vol. 61, no. 3-6, pp. 465–475, 2003.
fractional differential equations using Adomian decomposi- [43] C. Li and F. Zeng, “Finite difference methods for fractional
tion,” Journal of Computational and Applied Mathematics, vol. differential equations,” International Journal of Bifurcation and
196, no. 2, pp. 644–651, 2006. Chaos in Applied Sciences and Engineering, vol. 22, no. 4, 28
[24] K. Diethelm and G. Walz, “Numerical solution of fractional pages, 2012.
order differential equations by extrapolation,” Numerical Algo- [44] F. A. Rihan, “Computational methods for delay parabolic
rithms, vol. 16, no. 3-4, pp. 231–253, 1997. and time-fractional partial differential equations,” Numerical
[25] L. Galeone and R. Garrappa, “On multistep methods for differ- Methods for Partial Differential Equations, vol. 26, no. 6, pp.
ential equations of fractional order,” Mediterranean Journal of 1556–1571, 2010.
Mathematics, vol. 3, no. 3-4, pp. 565–580, 2006. [45] Z. Odibat and S. Momani, “Numerical methods for nonlinear
[26] G. Wang, R. P. Agarwal, and A. Cabada, “Existence results partial differential equations of fractional order,” Applied Math-
and the monotone iterative technique for systems of nonlinear ematical Modelling, vol. 32, no. 1, pp. 28–39, 2008.
fractional differential equations,” Applied Mathematics Letters,
vol. 25, no. 6, pp. 1019–1024, 2012.
[27] N. J. Ford and A. C. Simpson, “The numerical solution of frac-
tional differential equations: speed versus accuracy,” Numerical
Algorithms, vol. 26, no. 4, pp. 333–346, 2001.
[28] V. Lakshmikantham, S. Leela, and J. Vasundhara, Theory of
Fractional Dynamic Systems, Cambridge Academic Publishers,
Cambridge. UK.
[29] I. Podlubny, Fractional Differential Equations, vol. 198 of Math-
ematics in Science and Engineering, Academic Press, San Diego,
Calif, USA, 1999.
[30] S. G. Samko, A. A. Kilbas, and O. I. Marichev, Fractional
integrals and derivatives, Gordon and Breach Science, Yverdon,
Switzerland, 1993.
[31] C. Li and Y. Ma, “Fractional dynamical system and its lineariza-
tion theorem,” Nonlinear Dynamics, vol. 71, no. 4, pp. 621–633,
2013.
[32] Z. M. Odibat and N. T. Shawagfeh, “Generalized Taylor’s
formula,” Applied Mathematics and Computation, vol. 186, no.
1, pp. 286–293, 2007.
[33] N. Bellomo, A. Bellouquid, J. Nieto, and J. Soler, “Multiscale
biological tissue models and flux-limited chemotaxis for mul-
ticellular growing systems,” Mathematical Models & Methods in
Applied Sciences, vol. 20, no. 7, pp. 1179–1207, 2010.
[34] A. Gökdoğan, A. Yildirim, and M. Merdan, “Solving a fractional
order model of HIV infection of CD+ T cells,” Mathematical and
Computer Modelling, vol. 54, no. 9-10, pp. 2132–2138, 2011.
[35] D. Kirschner and J. C. Panetta, “Modeling immunotherapy
of the tumor—immune interaction,” Journal of Mathematical
Biology, vol. 37, no. 3, pp. 235–252, 1998.
[36] F. A. Rihan, M. Safan, M. A. Abdeen, and D. Abdel Rahman,
“Qualitative and computational analysis of a mathematical
model for tumor-immune interactions,” Journal of Applied
Mathematics, vol. 2012, Article ID 475720, 19 pages, 2012.
[37] R. Yafia, “Hopf bifurcation in differential equations with delay
for tumor-immune system competition model,” SIAM Journal
on Applied Mathematics, vol. 67, no. 6, pp. 1693–1703, 2007.
[38] K. A. Pawelek, S. Liu, F. Pahlevani, and L. Rong, “A model of
HIV-1 infection with two time delays: mathematical analysis
and comparison with patient data,” Mathematical Biosciences,
vol. 235, no. 1, pp. 98–109, 2012.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 740568, 11 pages
http://dx.doi.org/10.1155/2013/740568

Research Article
Oscillation Criteria for Fourth-Order Nonlinear Dynamic
Equations on Time Scales

Xin Wu,1 Taixiang Sun,1 Hongjian Xi,2 and Changhong Chen1


1
College of Mathematics and Information Science, Guangxi University, Nanning, Guangxi 530004, China
2
Department of Mathematics, Guangxi College of Finance and Economics, Nanning, Guangxi 530003, China

Correspondence should be addressed to Taixiang Sun; stx1963@163.com

Received 6 May 2013; Revised 20 June 2013; Accepted 21 June 2013

Academic Editor: Delfim F. M. Torres

Copyright © 2013 Xin Wu et al. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We establish some new oscillation criteria for nonlinear dynamic equation of the form (𝑎(𝑡)(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ )Δ + 𝑞(𝑡)𝑓(𝑥(𝜎(𝑡))) =
0 on an arbitrary time scale T with sup T = ∞, where 𝑎(𝑡), 𝑏(𝑡), 𝑐(𝑡) are positive rd-continuous functions. An example illustrating
the importance of our result is included.

1. Introduction the oscillation and asymptotic behavior of solutions of some


dynamic equations on time scales, and we refer the reader to
A time scale T is an arbitrary nonempty closed set of the paper [3–8] and the references cited therein.
real numbers R with the topology and ordering inherited Recently, Hassan in [9] studied the third-order dynamic
from R. The theory of time scales, which has recently received equation
a lot of attention, was introduced by Hilger in his Ph.D thesis
[1] in order to unify continuous and discrete analysis. The
Δ 𝛾 Δ
cases when a time scale T is equal to R or the set of all (𝑎 (𝑡) {[𝑟 (𝑡) 𝑥Δ (𝑡)] } ) + 𝑓 (𝑡, 𝑥 (𝜏 (𝑡))) = 0, (1)
integers Z represent the classical theories of differential and
difference equations. Many results concerning differential
equations carry over quite easily to corresponding results on a time scale T, where 𝛾 ≥ 1 is the quotient of odd positive
for difference equations, while other results seem to be integers, 𝑎 and 𝑟 are positive rd-continuous functions on T,
completely different from their continuous counterparts. The and the so-called delay function 𝜏 : T → T satisfies 𝜏(𝑡) ≤
study of dynamic equations on time scales reveals such 𝑡 for 𝑡 ∈ T and lim𝑡 → ∞ 𝜏(𝑡) = ∞ and 𝑓 ∈ 𝐶(T ×
discrepancies and helps avoid proving results twice once for R, R) and obtained some oscillation criteria, which improved
differential equations and once again for difference equations. and extended the results that have been established in [10–12].
The general is to prove a result for a dynamic equation where Li et al. in [13] also discussed the oscillation of (1),
the domain of the unknown function is a time scale T. In where 𝛾 > 0 is the quotient of odd positive integers, 𝑓 ∈
this way results not only related to the set of real numbers 𝐶(T × R, R) is assumed to satisfy 𝑢𝑓(𝑡, 𝑢) > 0 for 𝑢 ≠0, and
or set of integers but those pertaining to more general time there exists a positive rd-continuous function 𝑝 on T such
scales are obtained. Therefore, not only can the theory of that 𝑓(𝑡, 𝑢)/𝑢𝛾 ≥ 𝑝(𝑡) for 𝑢 ≠0. They established some new
dynamic equations unify the theories of differential equations sufficient conditions for the oscillation of (1).
and difference equations, but also extends these classical Wang and Xu in [14] extended the Hille and Nehari
cases to cases “in between,” for example, to the so-called 𝑞- oscillation theorems to the third-order dynamic equation
difference equations when T = {1, 𝑞, 𝑞2 , . . . , 𝑞𝑛 , . . .}, which
has important applications in quantum theory (see [2]). In the Δ 𝛾 Δ
last years there has been much research activity concerning (𝑟2 (𝑡) ((𝑟1 (𝑡) 𝑥Δ (𝑡)) ) ) + 𝑞 (𝑡) 𝑓 (𝑥 (𝑡)) = 0, (2)
2 Abstract and Applied Analysis

on a time scale T, where 𝛾 ≥ 1 is a ratio of odd positive on an arbitrary time scale T with sup T = ∞, where 𝑝, 𝑞 ∈

integers and the functions 𝑟𝑖 (𝑡) (𝑖 = 1, 2), 𝑞(𝑡) are positive 𝐶rd (T, (0, ∞)) with ∫𝑡 (1/𝑝(𝑠))Δ𝑠 < ∞ and there exists a
real-valued rd-continuous functions defined on T. 0
positive constant 𝐿 such that 𝑓(𝑦)/𝑦 ≥ 𝐿 for all 𝑦 ≠0; they
Erbe et al. in [15] were concerned with the oscillation of give a new oscillation result of (8).
the third-order nonlinear functional dynamic equation Motivated by the previous studies, in this paper, we will
Δ 𝛾 Δ study the oscillation criteria of the following fourth-order
(𝑎 (𝑡) [(𝑟 (𝑡) 𝑥Δ (𝑡)) ] ) + 𝑓 (𝑡, 𝑥 (𝑔 (𝑡))) = 0, (3) nonlinear dynamic equation:
on a time scale T, where 𝛾 is the quotient of odd positive Δ
Δ Δ
integers, 𝑎 and 𝑟 are positive rd-continuous functions on T, (𝑎 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ) ) + 𝑞 (𝑡) 𝑓 (𝑥 (𝜎 (𝑡))) = 0,
and 𝑔 : T → T satisfies lim𝑡 → ∞ 𝑔(𝑡) = ∞ and 𝑓 ∈
𝐶(T × R, R). The authors obtain some new oscillation criteria
𝑡 ∈ [𝑡0 , ∞)T ,
and extend many known results for oscillation of third-order
dynamic equations. (9)
Qi and Yu in [16] obtained some oscillation criteria for the
fourth-order nonlinear delay dynamic equation where T is a time scale with sup T = ∞ and 𝑡0 ∈ T is a
4
constant and [𝑡0 , ∞)T = [𝑡0 , ∞) ⋂ T. Throughout this paper,
𝑥Δ (𝑡) + 𝑝 (𝑡) 𝑥𝛾 (𝜏 (𝑡)) = 0, (4) we assume that the following conditions are satisfied:
on a time scale T, where 𝛾 is the ratio of odd positive
integers, 𝑝 is a positive real-valued rd-continuous function (H 1 ) 𝑎, 𝑏, 𝑐, 𝑞 ∈ 𝐶rd ([𝑡0 , ∞)T , (0, ∞)), 𝑏Δ (𝑡) ≥ 0 and 𝑐Δ (𝑡)
defined on T, 𝜏 ∈ 𝐶rd (T, T), 𝜏(𝑡) ≤ 𝑡, and lim𝑡 → ∞ 𝜏(𝑡) = ∞. ≥ 0.
Grace et al. in [17] were concerned with the oscillation of ∞ ∞ ∞
(H 2 ) ∫𝑡 (1/𝑎(𝑠))Δ𝑠 < ∫𝑡 (1/𝑏(𝑠))Δ𝑠 = ∫𝑡 (1/𝑐(𝑠))Δ𝑠 =
the fourth-order nonlinear dynamic equation ∞.
0 0 0

4
𝑥Δ (𝑡) + 𝑞 (𝑡) 𝑥𝜆 (𝑡) = 0, (5) (H 3 ) 𝑓 ∈ 𝐶(T, R) and there exists a positive constant 𝑀
on a time scale T, where 𝜆 is the ratio of odd positive such that for any 𝑢 ≠0, 𝑓(𝑢)/𝑢 ≥ 𝑀.
integers, 𝑞 is a positive real-valued rd-continuous function
defined on T. They reduce the problem of the oscillation of all By a solution of (9), we mean a nontrivial real-valued
1
solutions of (5) to the problem of oscillation of two second- function 𝑥 ∈ 𝐶rd ([𝑇𝑥 , ∞)T ) with 𝑇𝑥 ≥ 𝑡0 , which has the
order dynamic equations and give some conditions ensuring property that 𝑎(𝑡)(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ ∈ 𝐶rd 1
([𝑇𝑥 , ∞)T ) and
1
that all bounded solutions of (5) are oscillatory. satisfies (9) on [𝑇𝑥 , ∞)T , where 𝐶rd is the space of differ-
Grace et al. in [18] establish some new criteria for the entiable functions whose derivative is rd-continuous. The
oscillation of fourth-order nonlinear dynamic equations solutions vanishing in some neighborhood of infinity will be
2 Δ2 excluded from our consideration. A solution 𝑥(𝑡) of (9) is
(𝑎𝑥Δ ) (𝑡) + 𝑓 (𝑡, 𝑥𝜎 (𝑡)) = 0, 𝑡 ≥ 𝑡0 , (6) said to be oscillatory if it is neither eventually positive nor
eventually negative; otherwise it is called nonoscillatory.
where 𝑎 is a positive real-valued rd-continuous function

satisfying that ∫𝑡 (𝜎(𝑠)/𝑎(𝑠))Δ𝑠 < ∞, 𝑓 : [𝑡0 , ∞) × R →
0 2. Some Auxiliary Lemmas
R is continuous satisfying sgn 𝑓(𝑡, 𝑥) = sgn 𝑥 and 𝑓(𝑡, 𝑥) ≤
𝑓(𝑡, 𝑦) for 𝑥 ≤ 𝑦 and 𝑡 ≥ 𝑡0 . They also investigate the case We shall employ the following lemmas.
of strongly superlinear and the case of strongly sublinear
equations subject to various conditions. Lemma 1. Assume that 𝑥(𝑡) is an eventually positive solution
Agarwal et al. in [19] were concerned with oscillatory of (9). Then there exists 𝑡1 ∈ [𝑡0 , ∞)T sufficiently large, such
behavior of a fourth-order half-linear delay dynamic equa- that, for 𝑡 ∈ [𝑡1 , ∞)T , one of the following cases holds:
tion with damping
Δ
Δ3
𝛾 Δ
Δ3
𝛾
𝛾 (1) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) < 0, (𝑐(𝑡)𝑥Δ (𝑡)) > 0,
(𝑟(𝑥 ) ) (𝑡) + 𝑝 (𝑡) (𝑥 ) (𝑡) + 𝑞 (𝑡) 𝑥 (𝜏 (𝑡)) = 0, (7) Δ Δ Δ
(𝑏(𝑡)(𝑐(𝑡)𝑥 (𝑡)) ) < 0,
on a time scale T with sup T = ∞, where 𝜆 is the ratio
of odd positive integers, 𝑟, 𝑝, 𝑞 are positive real-valued rd- (2) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ > 0,
Δ Δ Δ
continuous functions defined on T, 𝑟(𝑡) − 𝜇(𝑡)𝑝(𝑡) ≠0, 𝜏 ∈ (𝑏(𝑡)(𝑐(𝑡)𝑥 (𝑡)) ) < 0,
𝐶rd (T, T), 𝜏(𝑡) ≤ 𝑡, and 𝜏(𝑡) → ∞ as 𝑡 → ∞. They
establish some new oscillation criteria of (7). (3) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ > 0,
Δ Δ Δ
Zhang et al. in [20] were concerned with the oscillation of (𝑏(𝑡)(𝑐(𝑡)𝑥 (𝑡)) ) > 0,
a fourth-order nonlinear dynamic equation
Δ (4) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ < 0,
Δ3 Δ Δ
(𝑝𝑥 ) (𝑡) + 𝑞 (𝑡) 𝑓 (𝑥 (𝜎 (𝑡))) = 0, (8) Δ
(𝑏(𝑡)(𝑐(𝑡)𝑥 (𝑡)) ) > 0.
Abstract and Applied Analysis 3

Proof. Let 𝑥(𝑡) be an eventually positive solution of (9). Then Lemma 2 (see [12]). Assume that there exists 𝑇 ∈ T such
there is a 𝑡1 ≥ 𝑡0 , sufficiently large, such that, 𝑥(𝑡) > 0 for 𝑡 ≥ that 𝑈 satisfies
𝑡1 . By (9) we have
Δ
Δ Δ
(𝑎 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ) ) = −𝑞 (𝑡) 𝑓 (𝑥 (𝜎 (𝑡))) 𝑈 (𝑡) > 0, 𝑈Δ (𝑡) > 0, 𝑈ΔΔ (𝑡) > 0,
(15)
≤ −𝑀𝑞 (𝑡) 𝑥 (𝜎 (𝑡)) < 0, 𝑈ΔΔΔ (𝑡) ≤ 0, for 𝑡 ∈ [𝑇, ∞)T .
(10)

which implies that 𝑎(𝑡)(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ is decreasing and Then


one of the following two cases holds.
(a) (𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ > 0 for 𝑡 ≥ 𝑡1 . 𝑡𝑈 (𝑡)
lim inf ≥ 1, (16)
(b) There is a 𝑡2 ≥ 𝑡1 such that (𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ < 0 𝑡→∞ ℎ2 (𝑡, 𝑡0 ) 𝑈Δ (𝑡)
for 𝑡 ∈ [𝑡2 , ∞)T .
If case (a) holds, then 𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ is strictly increas- 𝑡
ing on [𝑡1 , ∞)T and there exist the following two subcases. where ℎ2 (𝑡, 𝑡0 ) = ∫𝑡 (𝜏 − 𝑡0 )Δ𝜏.
0
Δ Δ
(𝑎1 ) (𝑐(𝑡)𝑥 (𝑡)) < 0 for 𝑡 ≥ 𝑡1 .
(𝑎2 ) There exists a 𝑡3 ≥ 𝑡2 such that (𝑐(𝑡)𝑥Δ (𝑡))Δ > 0 for 3. The Main Result
𝑡 ∈ [𝑡3 , ∞)T .
Now we state and prove our main result.
If subcase (𝑎1 ) holds, then we claim 𝑥Δ (𝑡) > 0. If not,
there exists a 𝑡4 ≥ 𝑡3 such that 𝑐(𝑡)𝑥Δ (𝑡) ≤ 𝑐(𝑡4 )𝑥Δ (𝑡4 ) < Theorem 3. Assume that one of the following conditions holds:
0 for 𝑡 ≥ 𝑡4 . Thus, we get
𝑡
1 ∞
𝑥 (𝑡) ≤ 𝑥 (𝑡4 ) + 𝑐 (𝑡4 ) 𝑥Δ (𝑡4 ) ∫ Δ𝑠 󳨀→ −∞, (11) 𝑄 (𝑠)
𝑡4 𝑐 (𝑠) ∫ Δ𝑠 = ∞, (17)
𝑡0 𝑏 (𝑠)
which contradicts 𝑥(𝑡) > 0 eventually. Therefore, we obtain ∞
1 ∞
𝑄 (𝑠)
case (4). ∫ ∫ Δ𝑠Δ𝑢 = ∞, (18)
If subcase (𝑎2 ) holds, then let 𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ ≥ 𝑡0 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
Δ Δ
𝑏(𝑡3 )(𝑐(𝑡3 )𝑥 (𝑡3 )) > 0 we get 𝑡 ∞ ∞
1 𝑄 (𝑠)
Δ lim sup ∫ [𝑀𝑞 (V) ∫ ∫ Δ𝑠Δ𝑢
𝜎(V) 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
Δ Δ Δ
𝑐 (𝑡) 𝑥 (𝑡) ≥ 𝑐 (𝑡3 ) 𝑥 (𝑡3 ) + 𝑏 (𝑡3 ) (𝑐 (𝑡3 ) 𝑥 (𝑡3 )) 𝑡→∞ 𝑡0
[
𝑡 (12)
1 ∞
∫V (𝑄 (𝑠) /𝑏 (𝑠)) Δ𝑠
×∫ Δ𝑠 󳨀→ ∞. ]
𝑡3 𝑏 (𝑠) − ∞ ∞
4𝑐 (V) ∫𝜎(V) (1/𝑐 (𝑢)) ∫𝑢 (𝑄 (𝑠) /𝑏 (𝑠)) Δ𝑠Δ𝑢
]
Therefore, we obtain case (3).
If case (b) holds, then we claim (𝑐(𝑡)𝑥Δ (𝑡))Δ > 0 for 𝑡 ≥ × ΔV = ∞.
𝑡2 . If not, there exists a 𝑡5 ≥ 𝑡2 such that 𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ ≤ (19)
𝑏(𝑡5 )(𝑐(𝑡5 )𝑥Δ (𝑡5 ))Δ < 0 for 𝑡 ≥ 𝑡5 . Integrating this inequality
from 𝑡5 to 𝑡, we get
1
Δ If there exist two positive functions 𝛼, 𝛽 ∈ 𝐶rd ([𝑇0 , ∞)T ,
𝑐 (𝑡) 𝑥Δ (𝑡) ≤ 𝑐 (𝑡5 ) 𝑥Δ (𝑡5 ) + 𝑏 (𝑡5 ) (𝑐 (𝑡5 ) 𝑥Δ (𝑡5 )) (0, +∞)) such that for all sufficiently large 𝑡1 ∈ [𝑡0 , ∞)T , and
𝑡 (13) 𝑡4 > 𝑡3 > 𝑡2 > 𝑡1 , and some constant 𝑑 ∈ (0, 1),
1
×∫ Δ𝑠 󳨀→ −∞.
𝑡5 𝑏 (𝑠)
𝑡
Then, there exists a 𝑡6 ≥ 𝑡5 such that 𝑐(𝑡)𝑥Δ (𝑡) ≤ −𝑀 < lim sup ∫ [(𝑑𝑀𝑞 (𝑠) 𝑄 (𝜎 (𝑠)) ℎ2 (𝜎 (𝑠) , 𝑡0 )
0 for 𝑡 ≥ 𝑡6 . Integrating this inequality from 𝑡6 to 𝑡, we get 𝑡→∞ 𝑡0

𝑡 𝜎(𝑠)
1 1
𝑥 (𝑡) ≤ 𝑥 (𝑡6 ) − 𝑀 ∫ Δ𝑠 󳨀→ −∞, (14) ×∫ Δ𝑢 × (𝜎 (𝑠) 𝑐 (𝑠))−1 ) (20)
𝑡6 𝑐 (𝑠) 𝑡1 𝑏 (𝑢)
which contradicts 𝑥(𝑡) > 0 eventually. The proof is com- 1
− ] Δ𝑠 = ∞,
pleted. 4𝑄 (𝜎 (𝑠)) 𝑎 (𝑠)
4 Abstract and Applied Analysis

𝑡 which implies that 𝑎(𝑡)(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ is decreasing on


[
lim sup ∫ [𝑀𝑞 (𝑠) 𝛼𝜎 (𝑠) [𝑡1 , ∞)T , and so
𝑡→∞ 𝑡4
[ Δ Δ Δ Δ
V
𝑎 (𝑠) (𝑏 (𝑠) (𝑐 (𝑠) 𝑥Δ (𝑠)) ) ≤ 𝑎 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) )
𝜎(𝑠) 𝑧 ∫𝑡 (1/𝑎 (𝑢)) Δ𝑢
×∫ (∫ ( 1
ΔV) for 𝑠 ≥ 𝑡 ≥ 𝑡0 .
𝑡3 𝑡2 𝑏 (V)
(25)

Dividing the previous inequality by 𝑎(𝑠) and integrating the


× (𝑐 (𝑧))−1 ) Δ𝑧 resulting inequality from 𝑡 to 𝑙, we get
(21)
Δ Δ
𝜎(𝑠) −1 𝑏 (𝑙) (𝑐 (𝑙) 𝑥Δ (𝑙)) ≤ 𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡))
1
× (∫ Δ𝑧)
𝑡1 𝑎 (𝑧) Δ Δ
+ 𝑎 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ) (26)
Δ 2 𝜎(𝑠)
[(𝛼 (𝑠))+ ] 𝑎 (𝑠) ∫𝑡 (1/𝑎 (𝑧)) Δ𝑧 ] 𝑙
− 1
] 1
𝑠 ×∫ Δ𝑠.
4𝛼𝜎 (𝑠) ∫𝑡 (1/𝑎 (𝑧)) Δ𝑧 𝑡 𝑎 (𝑠)
1
]
Let 𝑙 → ∞, we obtain
× Δ𝑠 = ∞,
Δ Δ Δ
𝑡 𝜎∫ 𝑞 (V) ΔV ∞
∞ 𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ≥ −𝑎 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ) 𝑄 (𝑡) .
𝑀𝛽 (𝑠)
lim sup ∫ [ ∫ 𝑢 Δ𝑢 (27)
𝑡→∞ 𝑡0 𝑏 (𝑠) 𝑠 𝑎 (𝑢)
[
Hence, there exists a constant 𝑚 > 0 such that
Δ 2 𝜎(𝑠)
[(𝛽 (𝑠))+ ] 𝑐 (𝑠) ∫𝑡 (1/𝑐 (𝑢)) Δ𝑢 ] (22) Δ
− 𝑠
1
] 𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ≥ 𝑚𝑄 (𝑡) . (28)
4𝛽𝜎 (𝑠) ∫𝑡 (1/𝑐 (𝑢)) Δ𝑢
1
] Integrating (28) from 𝑡0 to 𝑡, we get
× Δ𝑠 = ∞, 𝑡
𝑄 (𝑠)
𝑐 (𝑡) 𝑥Δ (𝑡) − 𝑐 (𝑡0 ) 𝑥Δ (𝑡0 ) ≥ 𝑚 ∫ Δ𝑠, (29)
where 𝑡0 𝑏 (𝑠)

1
𝑄 (𝑡) := ∫ Δ𝑠, which implies that
𝑡 𝑎 (𝑠) (23)
𝑡
𝑄 (𝑠) 𝑐 (𝑡 ) 𝑥Δ (𝑡0 )
𝑓+ (𝑡) := max {0, 𝑓 (𝑡)} . ∫ Δ𝑠 ≤ − 0 , (30)
𝑡0 𝑏 (𝑠) 𝑚
Then, every solution 𝑥(𝑡) of (9) is oscillatory.
which contradicts assumption (17).
Proof. Assume that (9) has a nonoscillatory solution 𝑥(𝑡) on Integrating (28) from 𝑡 to ∞, we get
[𝑡0 , ∞)T . Then, without loss of generality, there is a 𝑡1 ≥ 𝑡0 , ∞
𝑄 (𝑠)
sufficiently large, such that 𝑥(𝑡) > 0 for 𝑡 ≥ 𝑡1 . By Lemma 1, −𝑐 (𝑡) 𝑥Δ (𝑡) ≥ 𝑚 ∫ Δ𝑠. (31)
there exist the following four possible cases: 𝑡 𝑏 (𝑠)

(1) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) < 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ > 0, Integrating the previous inequality from 𝑡0 to 𝑡 gives
(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ < 0, 𝑡 ∞
1 𝑄 (𝑠)
(2) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > Δ
0, (𝑐(𝑡)𝑥 (𝑡)) Δ
> 0, 𝑥 (𝑡0 ) − 𝑥 (𝑡) ≥ 𝑚 ∫ ∫ Δ𝑠Δ𝑢, (32)
𝑡0 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ < 0,
(3) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ > 0, which implies
(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ > 0, 𝑡 ∞
1 𝑄 (𝑠) 𝑥 (𝑡0 )
(4) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ < 0, ∫ ∫ Δ𝑠Δ𝑢 ≤ , (33)
𝑡0 𝑐 (𝑢) 𝑢 𝑏 (𝑠) 𝑚
(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ > 0.
If case (1) holds, then which contradicts assumption (18).
Let 𝐴(𝑡) = 𝑎(𝑡)(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ . Integrating (27) from
Δ
Δ Δ 𝑡 to ∞ gives
(𝑎 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ) ) = −𝑞 (𝑡) 𝑓 (𝑥 (𝜎 (𝑡)))
∞ ∞
𝐴 (𝑠) 𝑄 (𝑠) 𝑄 (𝑠)
−𝑐 (𝑡) 𝑥Δ (𝑡) ≥ ∫ − Δ𝑠 ≥ −𝐴 (𝑡) ∫ Δ𝑠.
≤ −𝑀𝑞 (𝑡) 𝑥 (𝜎 (𝑡)) < 0, 𝑡 𝑏 (𝑠) 𝑡 𝑏 (𝑠)
(24) (34)
Abstract and Applied Analysis 5

Integrating (34) from 𝑡 to ∞, we get Integrating (41) from 𝑡1 to 𝑡 gives


∞ ∞
1 𝑄 (𝑠)

𝑅 (𝑡) ∫ ∫ Δ𝑠Δ𝑢
𝐴 (𝑢) ∞ 𝑄 (𝑠) 𝑡 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
𝑥 (𝑡) ≥ − ∫ ∫ Δ𝑠Δ𝑢
𝑡 𝑐 (𝑢) 𝑢 𝑏 (𝑠) ∞
1 ∞
𝑄 (𝑠)
∞ ∞
(35) − 𝑅 (𝑡1 ) ∫ ∫ Δ𝑠Δ𝑢
1 𝑄 (𝑠) 𝑡1 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
≥ −𝐴 (𝑡) ∫ ∫ Δ𝑠Δ𝑢.
𝑡 𝑐 (𝑢) 𝑢 𝑏 (𝑠) 𝑡 ∞ ∞
1 𝑄 (𝑠)
+ ∫ 𝑀𝑞 (V) ∫ ∫ Δ𝑠Δ𝑢ΔV
𝑡1 𝜎(V) 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
Set 𝑡 ∞
𝑅 (V) 𝑄 (𝑠)
≤ −∫ ∫ Δ𝑠ΔV
𝑡1 𝑐 (V) V 𝑏 (𝑠)
𝐴 (𝑡)
𝑅 (𝑡) = for 𝑡 ∈ [𝑡1 , ∞)T . (36)
𝑥 (𝑡) 𝑡
𝑅2 (V) ∞ 𝑄 (𝑠) ∞
1
−∫ ∫ Δ𝑠 ∫
𝑡1 𝑐 (V) V 𝑏 (𝑠) 𝜎(V) 𝑐 (𝑢)
Then, 𝑅(𝑡) < 0 for 𝑡 ∈ [𝑡1 , ∞)T and ∞
𝑄 (𝑠)
×∫ Δ𝑠Δ𝑢ΔV
𝑢 𝑏 (𝑠)
Δ Δ Δ ∞
𝐴 (𝑡) 𝐴 (𝑡) 𝑥 (𝑡) 𝐴 (𝑡) 𝑥 (𝑡) 𝑡 ∫V (𝑄 (𝑠) /𝑏 (𝑠)) Δ𝑠
𝑅Δ (𝑡) = 𝜎
− 𝜎
≤ −𝑀𝑞 (𝑡) − . ≤∫ ΔV,
𝑥 (𝑡) 𝑥 (𝑡) 𝑥 (𝑡) 𝑥 (𝑡) 𝑥𝜎 (𝑡) ∞ ∞
𝑡1 4𝑐 (V) ∫𝜎(V) (1/𝑐 (𝑢)) ∫𝑢 (𝑄 (𝑠) /𝑏 (𝑠)) Δ𝑠Δ𝑢
(37)
(42)
which implies
By (34), we get
𝑡 ∞ ∞
1 𝑄 (𝑠)
∫ [𝑀𝑞 (V) ∫ ∫ Δ𝑠Δ𝑢
𝐴2 (𝑡) ∞
𝑄 (𝑠) 𝑡1 𝜎(V) 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
𝑅Δ (𝑡) ≤ −𝑀𝑞 (𝑡) − 𝜎
∫ Δ𝑠 [
𝑐 (𝑡) 𝑥 (𝑡) 𝑥 (𝑡) 𝑡 𝑏 (𝑠)

(38) ∫V (𝑄 (𝑠) /𝑏 (𝑠)) Δ𝑠
2
𝐴 (𝑡) ∞
𝑄 (𝑠) − ∞ ∞
] ΔV
≤ −𝑀𝑞 (𝑡) − ∫ Δ𝑠. 4𝑐 (V) ∫𝜎(V) (1/𝑐 (𝑢)) ∫𝑢 (𝑄 (𝑠) /𝑏 (𝑠)) Δ𝑠Δ𝑢
𝑐 (𝑡) 𝑥2 (𝑡) 𝑡 𝑏 (𝑠) ]
∞ ∞
1 𝑄 (𝑠)
≤ 𝑅 (𝑡1 ) ∫ ∫ Δ𝑠Δ𝑢
Combining (36) with (38) gives 𝑡1 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
∞ ∞
1 𝑄 (𝑠)
2
− 𝑅 (𝑡) ∫ ∫ Δ𝑠Δ𝑢
𝑅 (𝑡) ∞ 𝑄 (𝑠) 𝑡 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
𝑅Δ (𝑡) ≤ −𝑀𝑞 (𝑡) − ∫ Δ𝑠. (39)
𝑐 (𝑡) 𝑡 𝑏 (𝑠) ∞
1 ∞
𝑄 (𝑠)
≤ 𝑅 (𝑡1 ) ∫ ∫ Δ𝑠Δ𝑢 + 1.
𝑡1 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
In view of (35), we get (43)
Which contradicts assumption (19).

1 ∞
𝑄 (𝑠) If case (2) holds, then set
𝑅 (𝑡) ∫ ∫ Δ𝑠Δ𝑢 ≥ −1. (40)
𝑡 𝑐 (𝑢) 𝑢 𝑏 (𝑠) 𝐴 (𝑡)
𝑅 (𝑡) = Δ
for 𝑡 ∈ [𝑡1 , ∞)T , (44)
𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡))
From (39), we obtain and 𝑅(𝑡) < 0 for 𝑡 ∈ [𝑡1 , ∞)T and
𝐴Δ (𝑡)

1 ∞
𝑄 (𝑠) 𝑅Δ (𝑡) ≤ Δ
𝑅Δ (𝑡) ∫ ∫ Δ𝑠Δ𝑢 (𝑏(𝑐𝑥Δ ) ) (𝜎 (𝑡))
𝜎(𝑡) 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
∞ ∞ Δ Δ
1 𝑄 (𝑠) 𝐴 (𝑡) (𝑏(𝑐𝑥Δ ) ) (𝑡)
≤ −𝑀𝑞 (𝑡) ∫ ∫ Δ𝑠Δ𝑢 − (45)
𝜎(𝑡) 𝑐 (𝑢) 𝑢 𝑏 (𝑠) Δ Δ
(𝑏(𝑐𝑥Δ ) ) (𝑡) (𝑏 (𝑐𝑥Δ ) ) (𝜎 (𝑡))
𝑅2 (𝑡) ∞ 𝑄 (𝑠) ∞
1 ∞
𝑄 (𝑠)
− ∫ Δ𝑠 ∫ ∫ Δ𝑠Δ𝑢. 𝑞 (𝑡) 𝑥 (𝜎 (𝑡)) 𝑅2 (𝑡)
𝑐 (𝑡) 𝑡 𝑏 (𝑠) 𝜎(𝑡) 𝑐 (𝑢) 𝑢 𝑏 (𝑠) ≤ −𝑀 − .
Δ
𝑏 (𝜎 (𝑡)) (𝑐𝑥Δ ) (𝜎 (𝑡)) 𝑎 (𝑡)
(41)
6 Abstract and Applied Analysis

𝑡
On the other hand, let 𝑈(𝑡) = ∫𝑡 𝑢(𝑠)Δ𝑠 for 𝑡 ∈ [𝑡1 , ∞)T , Combining (45) with (54) gives
1
Δ
where 𝑢(𝑡) = 𝑐(𝑡)𝑥 (𝑡); it is easy to check that 𝑈(𝑡) > 0, 𝜎(𝑡)
𝑀𝑑𝑞 (𝑡) ℎ2 (𝜎 (𝑡) , 𝑡0 ) ∫𝑡 (1/𝑏 (𝑠)) Δ𝑠 𝑅2 (𝑡)
𝑈Δ (𝑡) > 0, 𝑈ΔΔ (𝑡) > 0. In view of Δ
𝑅 (𝑡) ≤ − 2
− .
𝜎 (𝑡) 𝑐 (𝜎 (𝑡)) 𝑎 (𝑡)
Δ
(𝑏 (𝑡) 𝑢Δ (𝑡)) = 𝑏Δ (𝑡) 𝑢Δ (𝑡) + 𝑏𝜎 (𝑡) 𝑢ΔΔ (𝑡) < 0, (46) (55)

by (𝐻1 ), we get By (27) we get


𝑅 (𝑡) 𝑄 (𝑡) ≥ −1. (56)
𝑈ΔΔΔ (𝑡) = 𝑢ΔΔ (𝑡) < 0. (47)
Multiplying both sides of (55) with 𝑡 replaced by 𝑠, by 𝑄𝜎 (𝑠),
Therefore, by Lemma 2, for any 𝑑 ∈ (0, 1), there exists 𝑡𝑑 ∈ and integrating with respect to 𝑠 from 𝑡2 to 𝑡 (𝑡 ≥ 𝑡2 ), one
[𝑡1 , ∞)T such that gets
𝑡𝑈 (𝑡) 𝑡
≥ 𝑑 for 𝑡 ∈ [𝑡𝑑 , ∞)T . (48) ∫ 𝑅Δ (𝑠) 𝑄𝜎 (𝑠) Δ𝑠
ℎ2 (𝑡, 𝑡0 ) 𝑈Δ (𝑡) 𝑡2
𝜎(𝑠)
Then, we see that 𝑡 𝑀𝑑𝑞 (𝑠) ℎ2 (𝜎 (𝑠) , 𝑡0 ) ∫𝑡 (1/𝑏 (𝑢)) Δ𝑢
≤ −∫ 2
(57)
𝑡
∫𝑡 𝑐 (𝑠) 𝑥Δ (𝑠) Δ𝑠 𝑡2 𝜎 (𝑠) 𝑐 (𝜎 (𝑠))
𝑑ℎ (𝑡, 𝑡0 ) 𝜎
1
≥ 2 . (49) × 𝑄 (𝑠) Δ𝑠
𝑐 (𝑡) 𝑥Δ (𝑡) 𝑡 𝑡
𝑅2 (𝑠) 𝜎
Since −∫ 𝑄 (𝑠) Δ𝑠.
𝑡2 𝑎 (𝑠)
𝑡 𝑡
∫ 𝑢 (𝑠) Δ𝑠 = ∫ 𝑐 (𝑠) 𝑥Δ (𝑠) Δ𝑠 Thus,
𝑡1 𝑡1
𝑅 (𝑡) 𝑄 (𝑡) ≤ 𝑅 (𝑡2 ) 𝑄 (𝑡2 )
𝑡
Δ 𝜎 𝜎(𝑠)
= 𝑐 (𝑡) 𝑥 (𝑡) − 𝑐 (𝑡1 ) 𝑥 (𝑡1 ) − ∫ 𝑐 (𝑠) 𝑥 (𝑠) Δ𝑠, 𝑡 𝑀𝑑𝑞 (𝑠) ℎ2 (𝜎 (𝑠) , 𝑡0 ) ∫𝑡 (1/𝑏 (𝑢)) Δ𝑢
𝑡1 −∫ 2

(50) 𝑡2 𝜎 (𝑠) 𝑐 (𝜎 (𝑠))

we get × 𝑄𝜎 (𝑠) Δ𝑠
𝑡 𝑡
𝑅 (𝑠) 𝑅2 (𝑠) 𝜎
𝑐 (𝑡) 𝑥 (𝑡) ≥ ∫ 𝑐 (𝑠) 𝑥Δ (𝑠) Δ𝑠. (51) −∫ [ + 𝑄 (𝑠)] Δ𝑠
𝑡1 𝑡2 𝑎 (𝑠) 𝑎 (𝑠)

In view of (49), we obtain that for all 𝑡 ∈ [𝑡𝑑 , ∞)T , ≤ 𝑅 (𝑡2 ) 𝑄 (𝑡2 )
𝜎(𝑠)
[ 𝑀𝑑𝑞 (𝑠) ℎ2 (𝜎 (𝑠) , 𝑡0 ) ∫𝑡2 (1/𝑏 (𝑢)) Δ𝑢
𝑡 𝑡
𝑥 (𝑡) 𝑐 (𝑡) 𝑥 (𝑡) ∫𝑡 𝑐 (𝑠) 𝑥Δ (𝑠) Δ𝑠 𝑑ℎ2 (𝑡, 𝑡0 ) (52) −∫ [
= ≥ 1
≥ . 𝑡2 𝜎 (𝑠) 𝑐 (𝜎 (𝑠))
𝑥Δ (𝑡) 𝑐 (𝑡) 𝑥Δ (𝑡) 𝑐 (𝑡) 𝑥Δ (𝑡) 𝑡
[
On the other hand, there exists 𝑡2 ≥ 𝑡𝑑 such that for any 𝑡 ∈
[𝑡2 , ∞)T , 1 ]
× 𝑄𝜎 (𝑠) − ] Δ𝑠,
4𝑄 (𝜎 (𝑠)) 𝑎 (𝑠)
𝑐 (𝑡) 𝑥Δ (𝑡) = 𝑐 (𝑡2 ) 𝑥Δ (𝑡2 ) ]
(58)
Δ Δ
𝑡 𝑏(𝑠) (𝑐 (𝑠) 𝑥 (𝑠))
+∫ Δ𝑠 which implies that
𝑏 (𝑠) (53)
𝑡2
𝜎(𝑠)
[ 𝑀𝑑𝑞 (𝑠) ℎ2 (𝜎 (𝑠) , 𝑡0 ) ∫𝑡2 (1/𝑏 (𝑢)) Δ𝑢
𝑡 𝑡
Δ 1 Δ
∫ [
≥ 𝑏(𝑡) (𝑐 (𝑡) 𝑥 (𝑡)) ∫ Δ𝑠. 𝜎 (𝑠) 𝑐 (𝜎 (𝑠))
𝑡2 𝑏 (𝑠) 𝑡2
[
It follows from (52) and (53) that (59)
𝜎 1 ]
𝑑ℎ2 (𝑡, 𝑡0 ) Δ × 𝑄 (𝑠) − ] Δ𝑠
𝑥 (𝑡) ≥ 𝑥 (𝑡) 4𝑄 (𝜎 (𝑠)) 𝑎 (𝑠)
𝑡 ]
𝑡 (54)
𝑑ℎ2 (𝑡, 𝑡0 ) ∫𝑡 (1/𝑏 (𝑠)) Δ𝑠 Δ
≤ 𝑅 (𝑡2 ) 𝑄 (𝑡2 ) − 𝑅 (𝑡) 𝑄 (𝑡) ≤ 𝑅 (𝑡2 ) 𝑄 (𝑡2 ) + 1,
Δ
≥ 2
𝑏(𝑡) (𝑐 (𝑡) 𝑥 (𝑡)) .
𝑡𝑐 (𝑡) which contradicts assumption (20).
Abstract and Applied Analysis 7

If case (3) holds, then since Write


Δ 𝑡 Δ Δ
𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ≥ ∫ (𝑏 (𝑠) (𝑐 (𝑠) 𝑥Δ (𝑠)) ) Δ𝑠 𝐴 (𝑡)
𝑡1 𝑅 (𝑡) = 𝛼 (𝑡) Δ
for 𝑡 ∈ [𝑡1 , ∞)T . (66)
(60) 𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡))
𝑡
1
≥ 𝐴 (𝑡) ∫ Δ𝑠,
𝑡1 𝑎 (𝑠)
Thus, 𝑅(𝑡) > 0 and for any 𝑡 ∈ [𝑡1 , ∞)T ,
we have
Δ Δ 𝐴 (𝑡)
𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) 𝑅Δ (𝑡) = 𝛼Δ (𝑡) Δ
( 𝑡 ) ≤ 0. (61) 𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡))
∫𝑡 (1/𝑎 (𝑠)) Δ𝑠
1 Δ
𝜎 𝐴 (𝑡)
Hence, there exists 𝑡2 ∈ [𝑡1 , ∞)T such that + 𝛼 (𝑡) ( Δ
)
𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡))
𝑐 (𝑡) 𝑥Δ (𝑡)
𝛼Δ (𝑡)
Δ
= 𝑅 (𝑡) + 𝛼𝜎 (𝑡)
𝑡 𝑏 (𝑠) (𝑐 (𝑠) 𝑥Δ (𝑠)) 𝛼 (𝑡)
Δ
= 𝑐 (𝑡2 ) 𝑥 (𝑡2 ) + ∫ 𝑠 Δ
𝑡2 ∫𝑡 (1/𝑎 (𝑢)) Δ𝑢 × (𝐴Δ (𝑡) 𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡))
1

𝑠 (62)
∫𝑡 (1/𝑎 (𝑢)) Δ𝑢 Δ Δ
× 1
Δ𝑠 −𝐴 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ) )
𝑏 (𝑠)
Δ 𝑠 Δ Δ −1
𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) 𝑡 ∫𝑡 (1/𝑎 (𝑢)) Δ𝑢 × (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) (𝑏(𝑐𝑥Δ ) ) (𝜎 (𝑡))) (67)
≥ 𝑡 ∫ 1
Δ𝑠,
∫𝑡 (1/𝑎 (𝑠)) Δ𝑠 𝑡2 𝑏 (𝑠)
1 (𝛼Δ (𝑡))+ 𝐴Δ (𝑡)
≤ 𝑅 (𝑡) + 𝛼𝜎 (𝑡) Δ
which implies that 𝛼 (𝑡) (𝑏 (𝑐𝑥Δ ) ) (𝜎 (𝑡))
Δ
𝑐 (𝑡) 𝑥Δ (𝑡) Δ Δ
( ) ≤ 0. (63) 𝐴 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) )
𝑡 𝑠
∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (𝑠)) Δ𝑠 − 𝛼𝜎 (𝑡) Δ Δ
2 1 𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) (𝑏 (𝑐𝑥Δ ) ) (𝜎 (𝑡))
Hence, there exists 𝑡3 ∈ [𝑡2 , ∞)T such that
(𝛼Δ (𝑡))+ 𝐴Δ (𝑡)
𝑥 (𝑡) = 𝑥 (𝑡3 ) = 𝑅 (𝑡) + 𝛼𝜎 (𝑡) Δ
𝛼 (𝑡) (𝑏 (𝑐𝑥Δ ) ) (𝜎 (𝑡))
𝑡
𝑐 (𝑠) 𝑥Δ (𝑠) Δ
+∫ (𝑏(𝑐𝑥Δ ) (𝑡))
𝑡3
𝑠 V
∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV 𝜎 𝑅2 (𝑡)
− 𝛼 (𝑡) .
2 1
𝑎 (𝑡) 𝛼2 (𝑡) (𝑏(𝑐𝑥Δ )Δ ) (𝜎 (𝑡))
𝑠 V
∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV
× 2 1
Δ𝑠
𝑐 (𝑠) (64) By (61) and (65), we get
𝑐 (𝑡) 𝑥Δ (𝑡)
≥ 𝑡 𝑠 𝑥𝜎 (𝑡)
∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (𝑠)) Δ𝑠 𝑅Δ (𝑡) ≤ −𝑀𝑞 (𝑡) 𝛼𝜎 (𝑡) Δ
2 1
(𝑏(𝑐𝑥Δ ) ) (𝜎 (𝑡))
𝑠 V
𝑡 ∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV
×∫ 2 1
Δ𝑠. (𝛼Δ (𝑡))+
𝑡3 𝑐 (𝑠) + 𝑅 (𝑡)
𝛼 (𝑡)
Combining (62) with (64) gives Δ
𝑅2 (𝑡) (𝑏(𝑐𝑥Δ ) ) (𝑡)
𝜎
𝑡 𝑠 V
∫𝑡 (∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV/𝑐 (𝑠)) Δ𝑠 − 𝛼 (𝑡)
𝑎 (𝑡) 𝛼2 (𝑡) (𝑏(𝑐𝑥Δ )Δ ) (𝜎 (𝑡))
𝑥 (𝑡) ≥ 3 2 1
𝑡
∫𝑡 (1/𝑎 (𝑠)) Δ𝑠 (65)
1

Δ ≤ − 𝑀𝑞 (𝑡) 𝛼𝜎 (𝑡)
× 𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) .
8 Abstract and Applied Analysis

V
𝜎(𝑡) 𝑠 ∫𝑡 (1/𝑎 (𝑢)) Δ𝑢 If case (4) holds, then
×∫ (∫ ( 1
ΔV)
𝑡3 𝑡2 𝑏 (V) (𝐴 (𝑡))Δ ≤ 𝑀𝑞 (𝑡) 𝑥 (𝜎 (𝑡)) < 0. (70)

Integrating the previous inequality from 𝑡 to 𝑧, we get


×(𝑐 (𝑠))−1 ) Δ𝑠
𝑧
𝐴 (𝑧) − 𝐴 (𝑡) ≤ ∫ 𝑀𝑞 (𝑠) 𝑥 (𝜎 (𝑠)) Δ𝑠. (71)
−1 𝑡
𝜎(𝑡)
1
× (∫ Δ𝑠) Letting 𝑧 → ∞ in this inequality, we obtain
𝑡1 𝑎 (𝑠)

(𝛼Δ (𝑡))+ 𝑅2 (𝑡) Δ Δ Δ 𝑀 ∫𝑡 𝑞 (𝑠) Δ𝑠
+ 𝑅 (𝑡) − 𝛼𝜎 (𝑡) −(𝑏 (𝑡) (𝑐 (𝑡) 𝑥 (𝑡)) ) + 𝑥 (𝜎 (𝑡)) ≤ 0. (72)
𝛼 (𝑡) 𝑎 (𝑡) 𝛼2 (𝑡) 𝑎 (𝑡)
𝑡
∫𝑡 (1/𝑎 (𝑠)) Δ𝑠 Integrating the previous inequality from 𝑡 to 𝑧, we get
𝜎
× 𝜎(𝑡)
1
≤ −𝑀𝑞 (𝑡) 𝛼 (𝑡) Δ Δ
∫𝑡 (1/𝑎 (𝑠)) Δ𝑠 𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) − 𝑏 (𝑧) (𝑐 (𝑧) 𝑥Δ (𝑧))
1

V ∞ (73)
𝜎(𝑡) 𝑠 ∫𝑡 (1/𝑎 (𝑢)) Δ𝑢 𝑧 ∫𝑠 𝑞 (𝑢) Δ𝑢
×∫ (∫ ( 1
ΔV) + 𝑀𝑥 (𝜎 (𝑡)) ∫ Δ𝑠 ≤ 0.
𝑡3 𝑡2 𝑏 (V) 𝑡 𝑎 (𝑠)
Letting 𝑧 → ∞ in this inequality, we obtain
−1
× (𝑐 (𝑠)) ) Δ𝑠 ∞

∫𝑠 𝑞 (𝑢) Δ𝑢
Δ Δ
𝑏 (𝑡) (𝑐 (𝑡) 𝑥 (𝑡)) + 𝑀𝑥 (𝜎 (𝑡)) ∫ Δ𝑠 ≤ 0.
𝑡 𝑎 (𝑠)
𝜎(𝑡) −1
1 (74)
× (∫ Δ𝑠)
𝑡1 𝑎 (𝑠) Now we set
2 𝜎(𝑡)
[(𝛼Δ (𝑡))+ ] 𝑎 (𝑡) ∫𝑡 (1/𝑎 (𝑠)) Δ𝑠 𝑐 (𝑡) 𝑥Δ (𝑡)
+ 1
. 𝑅 (𝑡) = 𝛽 (𝑡) for 𝑡 ∈ [𝑡1 , ∞)T . (75)
𝑡 𝑥 (𝑡)
4𝛼𝜎 (𝑡) ∫𝑡 (1/𝑎 (𝑠)) Δ𝑠
1

(68) Thus, 𝑅(𝑡) > 0 and for any 𝑡 ∈ [𝑡1 , ∞)T ,

Integrating the last inequality from 𝑡4 (𝑡4 ∈ [𝑡3 , ∞)T ) to 𝑡, 𝑐 (𝑡) 𝑥Δ (𝑡)
𝑅Δ (𝑡) = 𝛽Δ (𝑡)
we get 𝑥 (𝑡)
Δ
𝜎
(𝑐 (𝑡) 𝑥Δ (𝑡)) 𝑥 (𝑡) − 𝑐 (𝑡) 𝑥Δ (𝑡) 𝑥Δ (𝑡)
𝑡
[ +𝛽 (𝑡)
∫ [𝑀𝑞 (𝑠) 𝛼𝜎 (𝑠) 𝑥 (𝑡) 𝑥 (𝜎 (𝑡))
𝑡4 (76)
[ Δ
Δ
𝛽 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡))
V ≤ 𝑅 (𝑡) + 𝛽𝜎 (𝑡)
𝜎(𝑠) 𝑧 ∫𝑡 (1/𝑎 (𝑢)) Δ𝑢 𝛽 (𝑡) 𝑥 (𝜎 (𝑡))
×∫ (∫ ( 1
ΔV)
𝑡3 𝑡2 𝑏 (V) 𝛽𝜎 (𝑡) 𝑥 (𝑡) 2
− 𝑅 (𝑡) .
𝛽2 (𝑡) 𝑐 (𝑡) 𝑥𝜎 (𝑡)
× (𝑐 (𝑧))−1 ) Δ𝑧 Since
(69)
𝑡
𝑐 (𝑠) 𝑥Δ (𝑠)
𝜎(𝑠)
1
−1 𝑥 (𝑡) ≥ 𝑥 (𝑡) − 𝑥 (𝑡1 ) ≥ ∫ Δ𝑠
× (∫ Δ𝑧) 𝑡1 𝑐 (𝑠)
𝑡1 𝑎 (𝑧) 𝑡
(77)
Δ 1
2 𝜎(𝑠) ≥ 𝑐 (𝑡) 𝑥 (𝑡) ∫ Δ𝑠,
[(𝛼Δ (𝑠))+ ] 𝑎 (𝑠) ∫𝑡 (1/𝑎 (𝑧)) Δ𝑧 ] 𝑡1 𝑐 (𝑠)
− 𝑠 ] Δ𝑠 1

4𝛼𝜎 (𝑠) ∫𝑡 (1/𝑎 (𝑧)) Δ𝑧 we get


1
]
Δ
≤ 𝑅 (𝑡4 ) , 𝑥 (𝑡)
( 𝑡 ) ≤ 0. (78)
∫𝑡 (1/𝑐 (𝑠)) Δ𝑠
which contradicts assumption (21). 1
Abstract and Applied Analysis 9

Hence, by (74) and (78), we get So 𝑀 = 1. It is easy to calculate that


∞ ∞
1 1 2
∫ Δ𝑠 = ∫ 2 Δ𝑠 = < ∞,
∞ 𝑡0 𝑎 (𝑠) 𝑡0 𝑠 𝑡0
Δ 𝑀𝛽𝜎 (𝑡) ∞ ∫𝑠 𝑞 (𝑢) Δ𝑢
𝑅 (𝑡) ≤ − ∫ Δ𝑠 ∞
1 ∞
1
𝑏 (𝑡) 𝑡 𝑎 (𝑠) ∫ Δ𝑠 = ∫ 1/5 Δ𝑠 = ∞,
𝑡0 𝑏 (𝑠) 𝑡0 𝑠
(𝛽Δ (𝑡))+ (83)
∞ ∞
+ 𝑅 (𝑡) 1 1
𝛽 (𝑡) ∫ Δ𝑠 = ∫ 3/5 Δ𝑠 = ∞,
𝑡0 𝑐 (𝑠) 𝑡0 𝑠
𝑡
𝜎
𝛽 (𝑡) ∫𝑡 (1/𝑐 (𝑠)) Δ𝑠 ∞
1 ∞
1 2
− 1
𝑅2 (𝑡) 𝑄 (𝑡) = ∫ Δ𝑠 = ∫ 2 Δ𝑠 = .
𝛽2 (𝑡) 𝑐 (𝑡) ∫𝜎(𝑡) (1/𝑐 (𝑠)) Δ𝑠 (79) 𝑡 𝑎 (𝑠) 𝑡 𝑠 𝑡
𝑡1
It is obvious that
∞ ∞ ∞
𝑀𝛽𝜎 (𝑡) ∞ ∫𝑠 𝑞 (𝑢) Δ𝑢 𝑄 (𝑠) 2 2 1
≤− ∫ Δ𝑠 ∫ Δ𝑠 = ∫ 6/5 Δ𝑠 = < ∞. (84)
𝑏 (𝑡) 𝑡 𝑎 (𝑠) 𝑡0 𝑏 (𝑠) 𝑡0 𝑠 1 − 1/21/5 𝑡01/5
2 𝜎(𝑡) Therefore, we get
𝑐 (𝑡) [(𝛽Δ (𝑡))+ ] ∫𝑡 (1/𝑐 (𝑠)) Δ𝑠
+ 1
. ∞ ∞ ∞
𝑡 1 𝑄 (𝑠) 2 1
4𝛽𝜎 (𝑡) ∫𝑡 (1/𝑐 (𝑠)) Δ𝑠 ∫ ∫ Δ𝑠Δ𝑢 = 1/5
∫ 4/5
Δ𝑢 = ∞.
1 𝑡0 𝑐 (𝑢) 𝑢 𝑏 (𝑠) 1 − 1/2 𝑡0 𝑢
(85)

Integrating the previous inequality from 𝑡1 to 𝑡, we get Then, condition (18) holds. By Lemma 2, we get
𝑡 (𝑡 − 𝑡0 ) (𝑡 − 2𝑡0 )
ℎ2 (𝑡, 𝑡0 ) = ∫ (𝜏 − 𝑡0 ) Δ𝜏 = ,
∞ 𝑡0 3
𝑡 𝜎 ∞ ∫ 𝑞 (V) ΔV (86)
[ 𝑀𝛽 (𝑠)
∫ [ ∫ 𝑢 Δ𝑢 (2𝑡 − 𝑡0 ) (2𝑡 − 2𝑡0 )
𝑡1 𝑏 (𝑠) 𝑠 𝑎 (𝑢) ℎ2 (𝜎 (𝑡) , 𝑡0 ) = ℎ2 (2𝑡, 𝑡0 ) = ≥ 𝑡2 ,
[ 3
2 𝜎(𝑠)
while 𝑡 sufficiently large. Let 𝛼(𝑡) = 1, 𝛽(𝑡) = 𝑡. We have that
𝑐 (𝑠) [(𝛽Δ (𝑠))+ ] ∫𝑡 (1/𝑐 (𝑢)) Δ𝑢 ] (80) if 󰜚 ≥ 1/2𝑑, then
− 𝑠
1
] Δ𝑠 𝑡
4𝛽𝜎 (𝑠) ∫𝑡 (1/𝑐 (𝑢)) Δ𝑢
1
] lim sup ∫ [ (𝑑𝑀𝑞 (𝑠) 𝑄 (𝜎 (𝑠)) ℎ2 (𝜎 (𝑠) , 𝑡0 )
𝑡→∞ 𝑡0
≤ 𝑅 (𝑡1 ) , 𝜎(𝑠)
×∫ (1/𝑏 (𝑢)) Δ𝑢)
𝑡1
which contradicts assumption (22). The proof is completed.
× (𝜎 (𝑠) 𝑐 (𝑠))−1
1 (87)
− ] Δ𝑠
4. Example 4𝑄 (𝜎 (𝑠)) 𝑎 (𝑠)

Finally, we give an example to illustrate our main result. 𝑡 𝑑 (󰜚/𝑠6/5 ) (1/𝑠) 𝑠2 𝑠4/5 1
≥ lim sup ∫ [ − ] Δ𝑠
𝑡→∞ 𝑡0 2𝑠𝑠3/5 4𝑠
Example 1. Consider the fourth-order nonlinear dynamic
equation 𝑑󰜚 1 𝑡
1
≥( − ) lim sup ∫ Δ𝑠 = ∞.
2 4 𝑡 → ∞ 𝑡0 𝑠
Δ Since
Δ Δ 󰜚
(𝑡2 (𝑡1/5 (𝑡3/5 𝑥Δ (𝑡)) ) ) + 𝑓 (2𝑡) = 0, 𝑡 ∈ 2Z , (81) 𝑧 V
𝑡6/5 𝜎(𝑠) ∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV
∫ ( 2 1
) Δ𝑧
𝑡3 𝑐 (𝑧)
where 󰜚 > 0 is a constant, and
1/5 𝑧
1 𝜎(𝑠) ∫𝑡2 (1/V ) ΔV (88)
≥ ∫ Δ𝑧
𝑡1 𝑡3 𝑐 (𝑧)
𝑎 (𝑡) = 𝑡2 , 𝑏 (𝑡) = 𝑡1/5 , 𝑐 (𝑡) = 𝑡3/5 ,
󰜚 (82) 1 𝜎(𝑠) 𝑧4/5 1
𝑞 (𝑡) = , 𝑓 (𝑢) = 𝑢 ln (3 + 𝑢 ) . 2 ≥ ∫ Δ𝑧 ≥ 𝑠6/5 ,
𝑡6/5 𝑡1 𝑡3 𝑐 (𝑧) 𝑡1
10 Abstract and Applied Analysis

we get References
[1] S. Hilger, “Analysis on measure chains—a unified approach to
𝑡 continuous and discrete calculus,” Results in Mathematics, vol.
[
lim sup ∫ [𝑀𝑞 (𝑠) 𝛼𝜎 (𝑠) 18, no. 1-2, pp. 18–56, 1990.
𝑡→∞ 𝑡4
[2] V. Kac and P. Cheung, Quantum Calculus, Universitext, 2002.
[
[3] M. Bohner and A. Peterson, Dynamic Equations on Time Scales:
𝑧 V
𝜎(𝑠) ∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV An Introduction with Applications, Birkhäuser, Boston, Mass,
×∫ ( 2 1
) Δ𝑧 USA, 2001.
𝑡3 𝑐 (𝑧) [4] M. Bohner and A. Peterson, Advances in Dynamic Equations on
Time Scales, Birkhäuser, Boston, Mass, USA, 2003.
𝜎(𝑠) −1
1 [5] T. Sun, H. Xi, and X. Peng, “Asymptotic behavior of solutions
× (∫ ( ) Δ𝑧) of higher-order dynamic equations on time scales,” Advances in
𝑡1 𝑎 (𝑧)
Difference Equations, vol. 2011, Article ID 237219, 14 pages, 2011.
2 𝜎(𝑠) [6] T. Sun, H. Xi, X. Peng, and W. Yu, “Nonoscillatory solutions
[(𝛼Δ (𝑠))+ ] 𝑎 (𝑠) ∫𝑡 (1/𝑎 (𝑧)) Δ𝑧 ]
for higher-order neutral dynamic equations on time scales,”
− 𝑠
1
] Δ𝑠
4𝛼𝜎 (𝑠) ∫𝑡 (1/𝑎 (𝑧)) Δ𝑧 Abstract and Applied Analysis, vol. 2010, Article ID 428963, 16
1
] pages, 2010.
6/5 6/5 [7] T. Sun, H. Xi, and W. Yu, “Asymptotic behaviors of higher order
𝑡 (󰜚/𝑠6/5 ) (1/𝑡1 ) 2 𝑠 nonlinear dynamic equations on time scales,” Journal of Applied
≥ lim sup ∫ [ ] Δ𝑠
Mathematics and Computing, vol. 37, no. 1-2, pp. 177–192, 2011.
𝑡→∞ 𝑡4 1/𝑡1
[ ] [8] T. Sun, W. Yu, and H. Xi, “Oscillatory behavior and comparison
𝑡 for higher order nonlinear dynamic equations on time scales,”
= 26/5 󰜚 lim sup ∫ Δ𝑠 = ∞. Journal of Applied Mathematics & Informatics, vol. 30, no. 1-2,
𝑡→∞ 𝑡4 pp. 289–304, 2012.
(89) [9] T. S. Hassan, “Oscillation of third order nonlinear delay
dynamic equations on time scales,” Mathematical and Computer
Since Modelling, vol. 49, no. 7-8, pp. 1573–1586, 2009.
[10] L. Erbe, A. Peterson, and S. H. Saker, “Asymptotic behavior of

∞ ∫𝑢 𝑞 (V) ΔV 1 solutions of a third-order nonlinear dynamic equation on time
∫ Δ𝑢 ≥ 󰜚 , (90) scales,” Journal of Computational and Applied Mathematics, vol.
𝑠 𝑎 (𝑢) 𝑠6/5 181, no. 1, pp. 92–102, 2005.
𝜎(𝑠) [11] L. Erbe, A. Peterson, and S. H. Saker, “Oscillation and asymp-
∫𝑡 (1/𝑐 (𝑢)) Δ𝑢 totic behavior of a third-order nonlinear dynamic equation,”
1
𝑠 ≤ 2, (91) The Canadian Applied Mathematics Quarterly, vol. 14, no. 2, pp.
∫𝑡 (1/𝑐 (𝑢)) Δ𝑢
1 129–147, 2006.
[12] L. Erbe, A. Peterson, and S. H. Saker, “Hille and Nehari
we obtain type criteria for third-order dynamic equations,” Journal of
Mathematical Analysis and Applications, vol. 329, no. 1, pp. 112–
∞ 131, 2007.
𝑡 𝜎 ∞ ∫ 𝑞 (V) ΔV
[ 𝑀𝛽 (𝑠)
lim sup ∫ [ ∫ 𝑢 Δ𝑢 [13] T. Li, Z. Han, S. Sun, and Y. Zhao, “Oscillation results for
𝑡→∞ 𝑡0 𝑏 (𝑠) 𝑠 𝑎 (𝑢) third order nonlinear delay dynamic equations on time scales,”
[ Bulletin of the Malaysian Mathematical Sciences Society, vol. 34,
2 𝜎(𝑠)
no. 3, pp. 639–648, 2011.
[(𝛽Δ (𝑠))+ ] 𝑐 (𝑠) ∫𝑡 (1/𝑐 (𝑢)) Δ𝑢 ] (92) [14] Y. Wang and Z. Xu, “Asymptotic properties of solutions of cer-
− 𝑠
1
] Δ𝑠 tain third-order dynamic equations,” Journal of Computational
𝜎
4𝛽 (𝑠) ∫𝑡 (1/𝑐 (𝑢)) Δ𝑢 and Applied Mathematics, vol. 236, no. 9, pp. 2354–2366, 2012.
1
]
[15] L. Erbe, T. S. Hassan, and A. Peterson, “Oscillation of third
𝑡
1 1 order nonlinear functional dynamic equations on time scales,”
≥ (2󰜚 − ) lim sup ∫ 2/5 Δ𝑠 = ∞.
4 𝑡 → ∞ 𝑡0 𝑠 Differential Equations and Dynamical Systems, vol. 18, no. 1-2,
pp. 199–227, 2010.
Hence, conditions (18), (20), (21), and (22) of Theorem 3 are [16] Y. Qi and J. Yu, “Oscillation criteria for fourth-order nonlinear
satisfied. By Theorem 3, we see that every solution 𝑥(𝑡) of delay dynamic equations,” Electronic Journal of Differential
Equations, vol. 2013, no. 79, pp. 1–17, 2013.
(81) is oscillatory if 󰜚 ≥ 1/2𝑑.
[17] S. R. Grace, M. Bohner, and S. Sun, “Oscillation of fourth-
order dynamic equations,” Hacettepe Journal of Mathematics
Acknowledgments and Statistics, vol. 39, no. 4, pp. 545–553, 2010.
[18] S. R. Grace, S. Sun, and Y. Wang, “On the oscillation of fourth
This project is supported by NNSF of China (11261005) order strongly superlinear and strongly sublinear dynamic
and NSF of Guangxi (2011GXNSFA018135, 2012GXNSFD equations,” Journal of Applied Mathematics and Computing,
A276040) and SF of ED of Guangxi (2013ZD061). 2013.
Abstract and Applied Analysis 11

[19] R. P. Agarwal, M. Bohner, T. Li, and C. Zhang, “Oscillation


theorems for fourth-order half-linear delay dynamic equations
with damping,” Mediterranean Journal of Mathematics, 2013.
[20] C. Zhang, T. Li, R. P. Agarwal, and M. Bohner, “Oscillation
results for fourth-order nonlinear dynamic equations,” Applied
Mathematics Letters, vol. 25, no. 12, pp. 2058–2065, 2012.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 461970, 7 pages
http://dx.doi.org/10.1155/2013/461970

Research Article
The Bernstein Operational Matrices for Solving
the Fractional Quadratic Riccati Differential Equations with
the Riemann-Liouville Derivative

Dumitru Baleanu,1,2,3 Mohsen Alipour,4 and Hossein Jafari5


1
Department of Mathematics, Cankaya University, Ogretmenler Cad. 14, Balgat, 06530 Ankara, Turkey
2
Institute of Space Sciences, P.O. Box MG 23, Magurele, 077125 Bucharest, Romania
3
Department of Chemical and Materials Engineering, Faculty of Engineering, King Abdulaziz University, P.O. Box 80204,
Jeddah 21589, Saudi Arabia
4
Faculty of Basic Science, Babol University of Technology, P.O. Box 47148-71167, Babol, Iran
5
Department of Mathematics, University of Mazandaran, P.O. Box 47416-95447, Babolsar, Iran

Correspondence should be addressed to Mohsen Alipour; m.alipour2323@gmail.com

Received 6 April 2013; Accepted 22 May 2013

Academic Editor: Ali H. Bhrawy

Copyright © 2013 Dumitru Baleanu et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

We obtain the approximate analytical solution for the fractional quadratic Riccati differential equation with the Riemann-Liouville
derivative by using the Bernstein polynomials (BPs) operational matrices. In this method, we use the operational matrix for
fractional integration in the Riemann-Liouville sense. Then by using this matrix and operational matrix of product, we reduce
the problem to a system of algebraic equations that can be solved easily. The efficiency and accuracy of the proposed method are
illustrated by several examples.

1. Introduction where 𝑦𝑎𝑘 (𝑘 = 0, 1, . . . , ⌈𝛼⌉ − 1) are constants and 𝑎(𝑡), 𝑏(𝑡),


and 𝑓(𝑡) are known functions. We can see [4] to guarantee
The Riccati differential equation is named after the Italian the existence and uniqueness of the solution of (1) with initial
Nobleman Count Jacopo Francesco Riccati (1676–1754). The conditions (2).
book of Reid [1] contains the fundamental theories of the The general response expression (1) contains a parameter
Riccati equation, with applications to random processes, 𝛼, the order of the fractional derivative that can be varied to
optimal control, and diffusion problems. Moreover, it is well obtain various responses. In the case that 𝛼 is integer, then (1)
known that the one-dimensional static Schrödinger equation is reduced to the classical Riccati differential equation.
is closely related to a Riccati differential equation [2]. Solitary This problem to develop the analytical and numerical
wave solution of a nonlinear partial differential equation can method to solve the Riccati differential equation with stan-
be represented as a polynomial in two elementary functions dard derivative, the Caputo fractional derivative and the
satisfying a projective Riccati equation [3]. Riemann-Liouville fractional derivative, has attracted much
In this paper, we are dealing with the fractional quadratic attention and has been studied by many authors [5–18].
Riccati differential equation as follows: The aim of this work is using the Bernstein polynomials
𝛼 for solving the problem (1) and (2). We notice that the
𝑎 𝐷𝑡 𝑦 (𝑡) = 𝑓 (𝑡) + 𝑏 (𝑡) 𝑦 (𝑡) + 𝑎 (𝑡) 𝑦2 (𝑡) , 𝑎 < 𝑡 ≤ 𝑏, (1)
problem presented [18] was in the Caputo sense but in our
subject to the initial conditions work, the problem is with the Riemann-Liouville derivative;
therefore we considered a more general space of functions.
𝑦(𝑘) (𝑎) = 𝑦𝑎𝑘 , 𝑘 = 0, 1, . . . , ⌈𝛼⌉ − 1, (2) Also, in [18], the authors used the polynomials in the form of
2 Abstract and Applied Analysis

𝛼
𝐵𝑖,𝑚 (𝑥) = ( 𝑚𝑖 ) 𝑥𝛼𝑖 (1 − 𝑥𝛼 )𝑚−𝑖 (𝑖 = 0, 1, . . . , 𝑚) that is different Definition 2 (see [4, 22–24]). Let 𝛼 ≥ 0 and 𝑛 = ⌈𝛼⌉; the
from the standard Bernstein polynomials. So, the operational operator 𝑎 𝐷𝑡𝛼 , defined by
matrices in this work are different from those in [18].
The organization of this paper is as follows. In Section 2, 𝛼 1 𝑑 𝑛 𝑡
𝑎 𝐷𝑡 𝑓 (𝑡) = ( ) ∫ (𝑡 − 𝑥)𝑛−𝛼−1 𝑓 (𝑥) 𝑑𝑥,
the Bernstein polynomials are introduced. Some basic defi- Γ (𝑛 − 𝛼) 𝑑𝑡 𝑎
nitions and properties of the fractional calculus and also the (7)
BPs operational matrix for the Riemann-Liouville fractional 𝑎 ≤ 𝑡 ≤ 𝑏,
integration are presented in Section 3. In Section 4, by BPs 0
operational matrices, we solve the fractional quadratic Riccati 𝑎 𝐷𝑡 𝑓 (𝑡) = 𝑓 (𝑡) ,
differential equation. In Section 5, we discuss the convergence
of the proposed method. In Section 6, several examples are is called the Riemann-Liouville fractional derivative operator
considered to evaluate the power and effectiveness of the of order 𝛼.
presented method. Some conclusions are summarized in the
Definition 3 (see [4, 22–24]). Let 𝛼 ≥ 0, 𝑛 = ⌈𝛼⌉, and
last section. 𝑐
𝑑𝑛 𝑓(𝑥)/𝑑𝑥𝑛 ∈ 𝐿 1 [𝑎, 𝑏]. The operator 𝑎 𝐷𝑡𝛼 , defined by

𝑐 𝛼 1 𝑡 𝑑𝑛 𝑓 (𝑥)
2. The Bernstein Polynomials and 𝑎 𝐷𝑡 𝑓 (𝑡) = ∫ (𝑡 − 𝑥)𝑛−𝛼−1 𝑑𝑥,
Their Properties Γ (𝑛 − 𝛼) 𝑎 𝑑𝑥𝑛
𝑎 ≤ 𝑡 ≤ 𝑏, (8)
On the interval [0, 1] we define the Bernstein polynomials
(BPs) of mth degree as follows [19]: 𝑐 0
𝑎 𝐷𝑡 𝑓 (𝑡) = 𝑓 (𝑡) ,
𝑚
𝐵𝑖,𝑚 (𝑥) = ( ) 𝑥𝑖 (1 − 𝑥)𝑚−𝑖 , 𝑖 = 0, 1, . . . , 𝑚. (3) is called the Caputo fractional derivative operator of order 𝛼.
𝑖
Set {𝐵0,𝑚 (𝑥), 𝐵1,𝑚 (𝑥), . . . , 𝐵𝑚,𝑚 (𝑥)} in the Hilbert space Lemma 4. If 𝛼 ≥ 0, 𝑛 = ⌈𝛼⌉, and 𝑎 ≤ 𝑡 ≤ 𝑏, then
𝐿2 [0, 1] is a complete basis. We can write Φ𝑚 (𝑥) = 𝐴𝑇𝑚 (𝑥), 𝑐
𝑇 (1) 𝑎 𝐷𝑡𝛼 𝑎 𝐼𝑡𝛼 𝑓 (𝑡) = 𝑓 (𝑡) , (9)
where 𝐴 is a matrix upper triangular, 𝑇𝑚 (𝑥) = [1, 𝑥, . . . , 𝑥𝑚 ] ,
𝑇
and Φ𝑚 (𝑥) = [𝐵0,𝑚 (𝑥), 𝐵1,𝑚 (𝑥), . . . , 𝐵𝑚,𝑚 (𝑥)] [20]. 𝑛−1
𝑓(𝑘) (𝑎)
𝑐
As a result, any polynomial of degree 𝑚 can be expanded (2) 𝑎 𝐼𝑡𝛼 𝑎 𝐷𝑡𝑎 𝑓 (𝑡) = 𝑓 (𝑡) − ∑ (𝑥 − 𝑎)𝑘 , (10)
in terms of linear combination of 𝐵𝑖,𝑚 (𝑥) (𝑖 = 0, 1, . . . , 𝑚) as 𝑘=0
𝑘!
given below: 𝑛−1
𝑐 𝑓(𝑘) (𝑎)
𝑚
(3) 𝑎 𝐷𝑡𝛼 𝑓 (𝑡) = 𝑎 𝐷𝑡𝛼 𝑓 (𝑡) − ∑ (𝑥 − 𝑎)𝑘−𝛼 .
𝑃 (𝑥) = ∑𝑐𝑖 𝐵𝑖,𝑚 (𝑥) = 𝑐𝑇 Φ𝑚 (𝑥) , (4)
𝑘=0
Γ (𝑘 − 𝛼 + 1)
𝑖=0 (11)
where
1 −1 1 Proof. See [22–24].
𝑐 = (∫ Φ (𝑥) Φ(𝑥)𝑇 𝑑𝑥) (∫ 𝑃 (𝑥) Φ (𝑥) 𝑑𝑥) . (5)
0 0 Theorem 5. One can get BPs operational matrix 𝐹𝛼 from
The approximation of functions within the Bernstein polyno- order (𝑚 + 1) × (𝑚 + 1) for the Riemann-Liouville fractional
mials and convergence analysis can be found in [20, 21]. integral as
𝑡
𝛼 1
𝑎 𝐼𝑡 Φ𝑚 (𝑡) = ∫ (𝑡 − 𝑥)𝛼−1 Φ𝑚 (𝑥) 𝑑𝑥 ≈ 𝐹𝛼 Φ𝑚 (𝑡) .
3. BPs Operational Matrix for the Γ (𝛼) 𝑎
(12)
Riemann-Liouville Fractional Integration
In this section, firstly, we give some basic definitions and Proof. See [21].
properties of the fractional calculus which are used further
in this paper.
4. BPs for Solving the Fractional Quadratic
𝛼 Riccati Differential Equation
Definition 1 (see [4, 22–24]). Let 𝛼 ≥ 0; the operator 𝑎 𝐼𝑡 ,
defined on 𝐿 1 [𝑎, 𝑏] by
Firstly, we use the initial conditions to reduce a given initial-
𝑡
𝛼 1 value problem to a problem with zero initial conditions.
𝑎 𝐼𝑡 𝑓 (𝑡) = ∫ (𝑡 − 𝑥)𝛼−1 𝑓 (𝑥) 𝑑𝑥, 𝑎 ≤ 𝑡 ≤ 𝑏, So, we define
Γ (𝛼) 𝑎 (6)
0 𝑦 (𝑡) = 𝑦̂ (𝑡) + 𝑧 (𝑡) ,
𝑎 𝐼𝑡 𝑓 (𝑡) = 𝑓 (𝑡) , (13)

is called the Riemann-Liouville fractional integral operator of ̂ is some known function that satisfied the initial
where 𝑦(𝑡)
order 𝛼. conditions (2) and 𝑧(𝑡) is a new unknown function.
Abstract and Applied Analysis 3

Substituting (13) in (1) and (2), we have an initial-value Therefore we can reduce (22) by (23)–(25) as
problem as follows:
𝐶𝑇 Φ𝑚 (𝑡) = 𝐹𝑇 Φ𝑚 (𝑡) + Φ𝑚 (𝑡)𝑇 𝐵̂ 𝐶𝛼 + Φ𝑚 (𝑡)𝑇 𝐴
̂𝐶̃𝛼 𝐶𝛼 .
𝛼
𝑎 𝐷𝑡 𝑧 (𝑡) = 𝑓̃ (𝑡) + ̃𝑏 (𝑡) 𝑧 (𝑡) + 𝑎̃ (𝑡) 𝑧2 (𝑡) , 𝑎 < 𝑡 ≤ 𝑏, (26)
(14)
Finally, we obtain the following nonlinear system of algebraic
subject to the initial conditions equation:
𝑧(𝑘) (𝑎) = 0, 𝑘 = 0, 1, . . . , ⌈𝛼⌉ − 1. (15) 𝐶 = 𝐹 + 𝐵̂ 𝐶𝛼 + 𝐴
̂𝐶̃𝛼 𝐶𝛼 , (27)
On the other hand, by (11) in Lemma 4 we can write such that by solving this system we can obtain the vector 𝐶.
𝑐 𝛼 𝛼 Then, we can get
𝑎 𝐷𝑡 𝑧 (𝑡) = 𝑎 𝐷𝑡 𝑧 (𝑡) . (16)

̃ 𝑎̃(𝑡), ̃𝑏(𝑡), 𝑦 (𝑡) ≈ 𝑦̂ (𝑡) + 𝐶𝑇 𝐹𝛼 Φ𝑚 (𝑡) . (28)


Also, by using Lemma 3.3 in [20] the inputs 𝑓(𝑡),
𝛼
and 𝑎 𝐷𝑡 𝑧(𝑡) can be approximated as follows:

𝑓̃ (𝑡) ≈ 𝐹𝑇 Φ𝑚 (𝑡) , (17) 5. Convergence Analysis


𝛼 In this section, we investigate the convergence analysis for the
𝑎 𝐷𝑡 𝑧 (𝑡) ≈ 𝐶𝑇 Φ𝑚 (𝑡) , (18) method presented in Section 4.
The problem (14) changes to the following problems
𝑎̃ (𝑡) ≈ 𝐴𝑇 Φ𝑚 (𝑡) , (19)
̃𝑏 (𝑡) ≈ 𝐵𝑇 Φ (𝑡) ,
𝛼
𝑎 𝐷𝑡 𝑧 (𝑡) = 𝑓̃ (𝑡) + ̃𝑏 (𝑡) ( 𝑎 𝐼𝑡𝛼 𝑎 𝐷𝑡𝛼 𝑧 (𝑡))
𝑚 (20)
(29)
2
where 𝐹, 𝐴, and 𝐵 are known (𝑚 + 1) column vectors and 𝐶 + 𝑎̃ (𝑡) ( 𝑎 𝐼𝑡𝛼 𝑎 𝐷𝑡𝛼 𝑧 (𝑡)) , 𝑎 < 𝑡 ≤ 𝑏,
is an unknown (𝑚 + 1) column vector.
since
From (10), (15), (16), (18), and (12), we have
𝑛−1 (𝑘)
𝑧 (𝑡) 𝛼 𝑐 𝛼 𝑧 (𝑎)
𝑧 (𝑡) = 𝑎 𝐼𝑡 𝑎 𝐷𝑡 𝑧 (𝑡) + ∑ (𝑥 − 𝑎)𝑘
(10)
𝑘=0
𝑘!
𝑛−1 (𝑘) (30)
𝑐 𝑧 (𝑎) 𝑐
= 𝐼𝑡𝛼 𝑎 𝐷𝑡𝛼 𝑧 (𝑡) + ∑ (𝑥 − 𝑎)𝑘 = 𝑎 𝐼𝑡𝛼 𝑎 𝐷𝑡𝛼 𝑧 (𝑡) 𝑐
= 𝐼𝑡𝛼 𝑎 𝐷𝑡𝛼 𝑧 (𝑡) = 𝐼𝑡𝛼 𝑎 𝐷𝑡𝛼 𝑧 (𝑡) .
(10) 𝑎 𝑘! (15)
𝑘=0 (15) 𝑎 (16) 𝑎

= 𝐼𝑡𝛼 𝑎 𝐷𝑡𝛼 𝑧 (𝑡) ≈ 𝐼𝑡𝛼 (𝐶𝑇 Φ𝑚 (𝑡)) By taking 𝑢(𝑡) = 𝑎 𝐷𝑡𝛼 𝑧(𝑡) we obtain the following fractional
(16) 𝑎 (18) 𝑎
integral equation:
= 𝐶𝑇 𝑎 𝐼𝑡𝛼 Φ𝑚 (𝑡) ≈ 𝐶𝑇 𝐹𝛼 Φ𝑚 (𝑡) 2
(12) 𝑢 (𝑡) = 𝑓̃ (𝑡) + ̃𝑏 (𝑡) 𝑎 𝐼𝑡𝛼 𝑢 (𝑡) + 𝑎̃ (𝑡) ( 𝑎 𝐼𝑡𝛼 𝑢 (𝑡)) ,
(31)
= 𝐶𝛼𝑇 Φ𝑚 (𝑡) , 𝑎 < 𝑡 ≤ 𝑏.
(21)
If we use the approximation 𝑢(𝑡) ≈ 𝐶𝑇 Φ𝑚 (𝑡), then the
where 𝐶𝛼𝑇 = 𝐶𝑇 𝐹𝛼 . problem (31) from space 𝐶1 [0, 1] reduces to the following
Now, by substituting (17)–(21) into (14), we obtain problem in space 𝑆𝑚 = Span{𝐵0,𝑚 (𝑡), 𝐵1,𝑚 (𝑡), . . . , 𝐵𝑚,𝑚 (𝑡)}:

𝐶𝑇 Φ𝑚 (𝑡) = 𝐹𝑇 Φ𝑚 (𝑡) + 𝐵𝑇 Φ𝑚 (𝑡) Φ𝑚 (𝑡)𝑇 𝐹𝛼𝑇 𝐶 + 𝐴𝑇 Φ𝑚 (𝑡) 𝐶𝑇 Φ𝑚 (𝑡) = 𝑓̃ (𝑡) + ̃𝑏 (𝑡) 𝑎 𝐼𝑡𝛼 (𝐶𝑇 Φ𝑚 (𝑡))
(32)
× (𝐶𝑇 𝐹𝛼 Φ𝑚 (𝑡) Φ𝑚 (𝑡)𝑇 𝐹𝛼𝑇 𝐶) 2
+ 𝑎̃ (𝑡) ( 𝑎 𝐼𝑡𝛼 (𝐶𝑇 Φ𝑚 (𝑡))) , 𝑎 < 𝑡 ≤ 𝑏.
= 𝐹𝑇 Φ𝑚 (𝑡) + 𝐵𝑇 Φ𝑚 (𝑡) Φ𝑚 (𝑡)𝑇 𝐶𝛼 + 𝐴𝑇 Φ𝑚 (𝑡)
Now, similar to Theorem 6.1 in [21], we propose the next
× (𝐶𝛼𝑇 Φ𝑚 𝑇
(𝑡) Φ𝑚 (𝑡) 𝐶𝛼 ) . theorem.
(22) Theorem 6. Suppose that 𝑢∗ (𝑡) ∈ 𝐶1 [0, 1] is the exact solution
of (31) and 𝜇𝑚 = 𝐽[𝑢𝑚 ] = Min𝑢∈𝑆𝑚 𝐽[𝑢], where
Then, from Lemma 3.5 in [20] we have
󵄩 2󵄩
̃𝛼 ,
𝐶𝛼𝑇 Φ𝑚 (𝑥) Φ𝑚 (𝑥)𝑇 ≈ Φ𝑚 (𝑥)𝑇 𝐶 (23) 𝐽 [𝑢] = 󵄩󵄩󵄩󵄩𝑢 (𝑡) − 𝑓̃ (𝑡) − ̃𝑏 (𝑡) 𝑎 𝐼𝑡𝛼 𝑢 (𝑡) − 𝑎̃ (𝑡) ( 𝑎 𝐼𝑡𝛼 𝑢 (𝑡)) 󵄩󵄩󵄩󵄩 .
(33)
̂
𝐴𝑇 Φ𝑚 (𝑥) Φ𝑚 (𝑥)𝑇 ≈ Φ𝑚 (𝑥)𝑇 𝐴, (24)
𝑇 𝑇 𝑇̂
Then one has 𝜇𝑚 → 0 as 𝑚 → ∞ (i.e., 𝑢𝑚 (𝑡) → 𝑢∗ (𝑡) as
𝐵 Φ𝑚 (𝑥) Φ𝑚 (𝑥) ≈ Φ𝑚 (𝑥) 𝐵. (25) 𝑚 → ∞).
4 Abstract and Applied Analysis

Proof. By substituting (33) instead of (41) in [21], we can use ×10−9


the proof of Theorem 6.1 in [21].
2.5
∗ 1
Theorem 7. Suppose that 𝑢 (𝑡) ∈ 𝐶 [0, 1] is the exact solution
2

Absolute error y(t)


of (31) and 𝑢𝑚 (𝑡) ∈ 𝑆𝑚 is the obtained solution of (26). Then
one has 𝑢𝑚 (𝑡) → 𝑢∗ (𝑡) as 𝑚 → ∞. 1.5

Proof. Substituting (17), (19), (20), and (12) in (32) we have


1
𝑇 𝑇
𝐶 Φ𝑚 (𝑡) = (𝐹 Φ𝑀 (𝑡) + 𝑒𝑓𝑀
̃) 0.5
𝑇
+ (𝐵 Φ𝑀 (𝑡) + 𝑒̃𝑏𝑀) (𝐶𝑇 (𝐹𝛼 Φ𝑀 (𝑡) + 𝐸𝐼𝑀)) 0
0 0.2 0.4 0.6 0.8 1
2
+ (𝐴𝑇 Φ𝑀 (𝑡) + 𝑒𝑎𝑀 𝑇 𝑀
̃ ) (𝐶 (𝐹𝛼 Φ𝑀 (𝑡) + 𝐸𝐼 )) . t
(34)
Figure 1: Behavior of the absolute error function in Example 1 for
𝛼 = 1 and 𝑚 = 10.
From Lemmas 2.3 and 4.1 in [21], we have 𝑒𝑓𝑀 𝑀 𝑀 𝑀
̃ , 𝑒̃𝑏 , 𝑒𝑎̃ , 𝐸𝐼

0 as 𝑀 → ∞. So we can observe that as 𝑀 increases, (34)
gets close to (32). Now, by taking 𝑀 = 𝑚 we propose the
following problem that gets close to (32) as 𝑚 increases: 0.7
𝛼 = 0.7
𝑇 𝑇 𝑇 𝑇 0.6 𝛼 = 0.8
𝐶 Φ𝑚 (𝑡) = 𝐹 Φ𝑚 (𝑡) + 𝐵 Φ𝑚 (𝑡) Φ𝑚 (𝑡) 𝐶𝛼
0.5 𝛼 = 0.9
(35)
𝑇
+𝐴 Φ𝑚 (𝑡) (𝐶𝛼𝑇 Φ𝑚 𝑇
(𝑡) Φ𝑚 (𝑡) 𝐶𝛼 ) . 0.4
𝛼=1
y(t)

Then by (23) and (25), (35) reduces to the following equation: 0.3
0.2
𝐶𝑇 Φ𝑚 (𝑡) = 𝐹𝑇 Φ𝑚 (𝑡) + (Φ𝑀(𝑡)𝑇 𝐵̂ + 𝐸̂𝐵𝑀) 𝐶𝛼 Exact for
(36) 0.1 𝛼=1
̃𝛼 + 𝐸̃𝑀 ) 𝐶𝛼 ) .
+ 𝐴𝑇 Φ𝑚 (𝑡) ((Φ𝑀(𝑡)𝑇 𝐶 0
𝐶𝛼
0 0.2 0.4 0.6 0.8 1
Equation (36) gets to (35) as 𝑀 → ∞, because from t
Lemma 3.1 in [21] 𝐸̂𝐵𝑀, 𝐸̃𝐶𝑀𝛼 → 0 as 𝑀 → ∞. Then by
Figure 2: Behavior of 𝑦10 (𝑡) in Example 1 for different 𝛼 and exact
deleting 𝐸̂𝐵𝑀, 𝐸̃𝐶𝑀𝛼 , taking 𝑀 = 𝑚, and using (24) in (36), we solution for 𝛼 = 1.
have
̂ 𝛼
𝐶𝑇 Φ𝑚 (𝑡) = 𝐹𝑇 Φ𝑚 (𝑡) + Φ𝑚 (𝑡)𝑇 𝐵𝐶
(37) subject to the initial condition as 𝑦(0) = 0. The exact solution
̂ + 𝐸̂𝑀) 𝐶
+ (Φ𝑀(𝑡)𝑇 𝐴 ̃𝛼 𝐶𝛼 ,
𝐴 of the equation for 𝛼 = 1 is given as
where, from Lemma 3.1 in [21], 𝐸̂𝐴 𝑀
→ 0 as 𝑀 → ∞. 𝑒2𝑡 − 1
Now, by taking 𝑀 = 𝑚 and deleting 𝐸̂𝐴 𝑀 𝑦 (𝑡) = . (39)
in (37), we get 𝑒2𝑡 + 1
(26). Obviously, if 𝑢̃𝑚 (𝑡) is solution of (32), then we have
𝑢̃𝑚 − 𝑢𝑚 → 0 as 𝑚 → ∞. Numerical results compared to [14] are given in Table 1 and
On the other hand, from Theorem 6 we obtained 𝑢̃𝑚 → also Figure 1 shows the absolute error for our method for 𝛼 =
𝑢∗ as 𝑚 → ∞. Therefore we can write 𝑢𝑚 → 𝑢∗ as 𝑚 → ∞ 1 and Figure 2 shows behavior 𝑦10 (𝑡) for different values of 𝛼.
and the proof is complete.
Example 2. Consider the following quadratic Riccati differ-
ential equation of fractional order [14]
𝛼
6. Illustrative Numerical Examples 0 𝐷𝑡 𝑦 (𝑡) = 2𝑦 (𝑡) − 𝑦(𝑡)2 + 1, 0 < 𝑡 ≤ 1, (40)

In this section, we apply our method with 𝑚 = 10 (BPs of subject to the initial condition as 𝑦(0) = 0. The exact solution
degree 𝑚 = 10) to solve the following examples. We define of the equation for 𝛼 = 1 is given as
𝑦𝑚 (𝑡) and 𝑦(𝑡) for the approximate solution and the exact
1 √2 − 1
solution, respectively. 𝑦 (𝑡) = 1 + √2 tanh (√2𝑡 + log ( )) . (41)
2 √2 + 1
Example 1. Consider the nonlinear Riccati differential equa-
tion [14]: Numerical results compared to [14] are given in Table 2 and
also, Figure 3 shows the absolute error for our method for 𝛼 =
𝛼
0 𝐷𝑡 𝑦 (𝑡) = −𝑦(𝑡)2 + 1, 0 < 𝑡 ≤ 1, (38) 1 and Figure 4 shows behavior 𝑦10 (𝑡) for different values of 𝛼.
Abstract and Applied Analysis 5

Table 1: Numerical results for 𝛼 = 1 and 𝑚 = 10 in Example 1 with Table 4: Numerical results for 𝛼 = 2.5 in Example 3 with
comparison to exact solution and [14]. comparison to [25, 26].

𝑡 Exact Present method Reference [14] 𝑡 Present method ADM [25] FDTM [26]
𝑚 = 10 𝑚 = 10
0.1 0.099668 0.099668 0.099668 0.1 0.000952 0.000952 0.000952
0.2 0.197375 0.197375 0.197375 0.2 0.005383 0.005383 0.005383
0.3 0.291313 0.291313 0.291313 0.3 0.014833 0.014833 0.014833
0.4 0.379949 0.379949 0.379944 0.4 0.030450 0.030450 0.030450
0.5 0.462117 0.462117 0.462078 0.5 0.053197 0.053197 0.053197
0.6 0.537050 0.537050 0.536857 0.6 0.083925 0.083925 0.083925
0.7 0.604368 0.604368 0.603631 0.7 0.123412 0.123412 0.123412
0.8 0.664037 0.664037 0.661706 0.8 0.172391 0.172391 0.172391
0.9 0.716298 0.716298 0.709919 0.9 0.231574 0.231574 0.231574
1 0.761594 0.761594 0.746032 1 0.301676 0.301676 0.301676

×10−7
Table 2: Numerical results for 𝛼 = 1 and 𝑚 = 10 in Example 2 with
comparison to exact solution and [14]. 1

𝑡 Present method
Exact Reference [14] 0.8
𝑚 = 10 Absolute error y(t)
0.1 0.110295 0.110295 0.110294 0.6
0.2 0.241977 0.241977 0.241965
0.3 0.395105 0.395105 0.395106 0.4
0.4 0.567812 0.567812 0.568115
0.2
0.5 0.756014 0.756014 0.757564
0.6 0.953566 0.953566 0.958259 0
0.7 1.152949 1.152949 1.163459 0 0.2 0.4 0.6 0.8 1
0.8 1.346364 1.365240 t
1.346364
0.9 1.526911 1.526911 1.554960 Figure 3: Behavior of the absolute error function in Example 2 for
1 1.689499 1.689499 1.723810 𝛼 = 1 and 𝑚 = 10.

Table 3: Numerical results for 𝛼 = 1.5 in Example 3 with with the initial conditions
comparison to [25–27].
𝑦(𝑘) (0) = 0, 𝑘 = 0, 1, . . . , 𝑛 − 1. (43)
𝑡 Present method ADM [25] FDTM [26] BPFs [27]
𝑚 = 10 This problem has been studied by using ADM [25], FDTM
0.1 0.023779 0.023790 0.023790 0.023800 [26], and BPFs [27]. Our results with 𝛼 = 1.5, 𝛼 = 2.5 are
0.2 0.067336 0.067330 0.067330 0.067335 compared to [25–27] in Tables 3 and 4. Therefore, we see that
0.3 0.123896 0.123896 0.123900
our method is very effective and obtained solutions that are
0.123886
in good agreement with the results in [25–27]. Also, Figure 5
0.4 0.191373 0.191362 0.191362 0.191368
shows behavior 𝑦𝑚 (𝑡) for different values of 𝛼.
0.5 0.268851 0.268856 0.268856 0.268862
0.6 0.356235 0.356238 0.356238 0.356244
0.7 0.453958 0.453950 0.453950 0.453956 7. Conclusion
0.8 0.562999 0.563007 0.563007 0.563014 In this paper, we proposed a numerical method for solving
0.9 0.685066 0.685056 0.685056 0.685067 the fractional quadratic Riccati differential equations by
1 0.822540 0.822511 0.822509 0.822525 the operational matrices of the Bernstein polynomials. We
applied operational matrix for fractional integration in the
Riemann-Liouville sense. Then by using this matrix and
operational matrix of product, we reduced the fractional
Example 3. Consider the nonlinear fractional differential
quadratic Riccati differential equation to a system of algebraic
equation:
equations that can be solved easily. Finally, examples have
been simulated to demonstrate the high performance of the
𝛼
0 𝐷𝑡 𝑦 (𝑡) = 𝑦2 (𝑡) + 1, 0 < 𝑡 ≤ 1, 𝑛 − 1 < 𝛼 ≤ 𝑛, (42) proposed method. We saw that the results were in good
6 Abstract and Applied Analysis

[7] S. Abbasbandy, “A new application of He’s variational iteration


𝛼 = 0.7 method for quadratic Riccati differential equation by using
1.5 𝛼 = 0.8 Adomian’s polynomials,” Journal of Computational and Applied
𝛼 = 0.9 Mathematics, vol. 207, no. 1, pp. 59–63, 2007.
𝛼=1 [8] Y. Tan and S. Abbasbandy, “Homotopy analysis method for
1
y(t)

quadratic Riccati differential equation,” Communications in


Nonlinear Science and Numerical Simulation, vol. 13, no. 3, pp.
539–546, 2008.
0.5
[9] F. Geng, Y. Lin, and M. Cui, “A piecewise variational iteration
Exact for method for Riccati differential equations,” Computers & Mathe-
𝛼=1
matics with Applications, vol. 58, no. 11-12, pp. 2518–2522, 2009.
0
0 0.2 0.4 0.6 0.8 1
[10] F. Z. Geng and X. M. Li, “A new method for Riccati differential
equations based on reproducing kernel and quasilinearization
t
methods,” Abstract and Applied Analysis, vol. 2012, Article ID
Figure 4: Behavior of 𝑦10 (𝑡) in Example 2 for different 𝛼 and exact 603748, 8 pages, 2012.
solution for 𝛼 = 1. [11] N. A. Khan, A. Ara, and M. Jamil, “An efficient approach for
solving the Riccati equation with fractional orders,” Computers
& Mathematics with Applications, vol. 61, no. 9, pp. 2683–2689,
0.8 2011.
𝛼 = 1.5
[12] S. Momani and N. Shawagfeh, “Decomposition method for
0.6 solving fractional Riccati differential equations,” Applied Math-
ematics and Computation, vol. 182, no. 2, pp. 1083–1092, 2006.
[13] J. Cang, Y. Tan, H. Xu, and S.-J. Liao, “Series solutions of
y(t)

0.4 non-linear Riccati differential equations with fractional order,”


𝛼 = 2.5 Chaos, Solitons and Fractals, vol. 40, no. 1, pp. 1–9, 2009.
0.2 [14] Z. Odibat and S. Momani, “Modified homotopy perturbation
𝛼 = 3.5 𝛼 = 4.5 method: application to quadratic Riccati differential equation
of fractional order,” Chaos, Solitons and Fractals, vol. 36, no. 1,
0 pp. 167–174, 2008.
0 0.2 0.4 0.6 0.8 1 [15] C. Li and Y. Wang, “Numerical algorithm based on Adomian
t decomposition for fractional differential equations,” Computers
& Mathematics with Applications, vol. 57, no. 10, pp. 1672–1681,
Figure 5: Plot of 𝑦10 (𝑡) for different 𝛼 in Example 3. 2009.
[16] H. Jafari and H. Tajadodi, “He’s variational iteration method for
solving fractional Riccati differential equation,” International
agreement with the analytical solutions and the solutions in Journal of Differential Equations, vol. 2010, Article ID 764738,
the open literatures. Also, we observed that the solutions 8 pages, 2010.
approach to classical solutions as the order of the fractional [17] M. Merdan, “On the solutions fractional Riccati differential
derivatives approaches 1, for fixed 𝑚. equation with modified Riemann-Liouville derivative,” Inter-
national Journal of Differential Equations, vol. 2012, Article ID
346089, 17 pages, 2012.
References [18] S. Yüzbaşı, “Numerical solutions of fractional Riccati type
[1] W. T. Reid, Riccati Differential Equations, vol. 86, Academic differential equations by means of the Bernstein polynomials,”
Press, New York, NY, USA, 1972, Mathematics in Science and Applied Mathematics and Computation, vol. 219, no. 11, pp.
Engineering. 6328–6343, 2013.
[2] V. V. Kravchenko, Applied Pseudoanalytic Function Theory, [19] E. W. Cheney, Introduction to Approximation Theory, AMS
Frontiers in Mathematics, Birkhäuser, Basel, Switzerland, 2009. Chelsea, Providence, RI, USA, 2nd edition, 1998.
[3] R. Conte and M. Musette, “Link between solitary waves and [20] M. Alipour, D. Rostamy, and D. Baleanu, “Solving multi-
projective Riccati equations,” Journal of Physics A, vol. 25, no. dimensional FOCPs with inequality constraint by BPs opera-
21, pp. 5609–5623, 1992. tional matrices,” Journal of Vibration and Control, 2012.
[4] K. Diethelm, The Analysis of Fractional Differential Equations: [21] D. Rostamy, M. Alipour, H. Jafari, and D. Baleanu, “Solving
An Application-Oriented Exposition Using Differential Operators multi-term orders fractional differential equations by opera-
of Caputo Type, vol. 2004 of Lecture Notes in Mathematics, tional matrices of BPs with convergence analysis,” Romanian
Springer, Berlin, Germany, 2010. Reports in Physics, vol. 65, no. 2, pp. 334–349, 2013.
[5] S. Abbasbandy, “Iterated He’s homotopy perturbation method [22] I. Podlubny, Fractional Differential Equations, vol. 198, Aca-
for quadratic Riccati differential equation,” Applied Mathemat- demic Press, New York, NY, USA, 1999.
ics and Computation, vol. 175, no. 1, pp. 581–589, 2006. [23] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory and
[6] S. Abbasbandy, “Homotopy perturbation method for quadratic Applications of Fractional Differential Equations, Elsevier, San
Riccati differential equation and comparison with Adomian’s Diego, Calif, USA, 2006.
decomposition method,” Applied Mathematics and Computa- [24] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional
tion, vol. 172, no. 1, pp. 485–490, 2006. Calculus Models and Numerical Methods, vol. 3 of Series on
Abstract and Applied Analysis 7

Complexity, Nonlinearity and Chaos, World Scientific, Hacken-


sack, NJ, USA, 2012.
[25] N. T. Shawagfeh, “Analytical approximate solutions for nonlin-
ear fractional differential equations,” Applied Mathematics and
Computation, vol. 131, no. 2-3, pp. 517–529, 2002.
[26] A. Arikoglu and I. Ozkol, “Solution of fractional differential
equations by using differential transform method,” Chaos,
Solitons and Fractals, vol. 34, no. 5, pp. 1473–1481, 2007.
[27] Y. Li and N. Sun, “Numerical solution of fractional differential
equations using the generalized block pulse operational matrix,”
Computers & Mathematics with Applications, vol. 62, no. 3, pp.
1046–1054, 2011.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 739464, 7 pages
http://dx.doi.org/10.1155/2013/739464

Research Article
Fractional Dynamics of Genetic Algorithms Using
Hexagonal Space Tessellation

J. A. Tenreiro Machado
Department of Electrical Engineering, Institute of Engineering, Polytechnic of Porto, Rua Dr. António Bernardino de Almeida 431,
4200-072 Porto, Portugal

Correspondence should be addressed to J. A. Tenreiro Machado; jtm@isep.ipp.pt

Received 15 April 2013; Accepted 7 June 2013

Academic Editor: Dumitru Baleanu

Copyright © 2013 J. A. Tenreiro Machado. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

The paper formulates a genetic algorithm that evolves two types of objects in a plane. The fitness function promotes a relationship
between the objects that is optimal when some kind of interface between them occurs. Furthermore, the algorithm adopts an
hexagonal tessellation of the two-dimensional space for promoting an efficient method of the neighbour modelling. The genetic
algorithm produces special patterns with resemblances to those revealed in percolation phenomena or in the symbiosis found in
lichens. Besides the analysis of the spacial layout, a modelling of the time evolution is performed by adopting a distance measure and
the modelling in the Fourier domain in the perspective of fractional calculus. The results reveal a consistent, and easy to interpret,
set of model parameters for distinct operating conditions.

1. Introduction as Fourier, Euler, Laplace, Liouville, and Riemann devoted


efforts to the development of the theory of FC, but the
This paper analyzes the fractional order dynamics during the field remained primarily of pure mathematics. Things started
search for the optimal solution in a plane with an hexagonal to change in the beginning of the twentieth century with
tessellation by means of a genetic algorithm. These three dis- the work of Olivier Heaviside in operational calculus and
tinct scientific topics are recognized to be efficient approaches electromagnetism [6]. Nevertheless, only in the last two
in particular areas, namely, in the problems of modelling decades FC witnessed a substantial progress in the application
including long-range memory effects, space representation to physics, engineering, and biology [7–11].
using geometric shapes with no overlaps and no gaps, and Genetic algorithms (GAs) are a computer heuristic that
robust optimization in cases where standard techniques do mimics the process of evolution and belong to the class
not yield adequate solutions. This paper integrates the three of the so-called evolutionary algorithms [12–16]. Genetic
methodologies in the analysis of a complex evolutionary algorithms (GAs) were invented in the 60s by John Holland
optimization for producing solutions somehow resembling and developed by him and his colleagues. GAs are inspired
the percolation phenomenon, in the inorganic world, or, in the genetic structure and behaviour of chromosomes
alternatively, the lichens, in the scope of living beings. within a population of individuals (the solutions) having
Fractional calculus (FC) is a branch of mathematical in mind the ideas that (i) individuals compete between
analysis that generalizes the operations of differentiation themselves, (ii) most successful individuals tend to produce
and integration from integer up to real or complex orders more offspring, (iii) the genetic information of “good” indi-
[1–5]. The concept emerged in September 30, 1695, when viduals disseminates in the population and tends to produce
Guillaume de l’Hôpital wrote to Gottfried Leibniz a letter offspring that are better than their parents, and (iv) successive
asking him about the meaning of 𝑑1/2 𝑦/𝑑𝑥1/2 , to which generations become more suited to their living environ-
Leibniz replied “an apparent paradox, from which one-day ment. GAs implement an intelligent exploration of the space
useful consequences will be drawn.” Important minds such of solutions by exploiting historical information to direct
2 Abstract and Applied Analysis

the search into the region of better performance. During the Bearing these ideas in mind this paper is organized
last decades a growing amount of successful application to as follows. Section 2 formulates the main algorithms and
real-world problems demonstrated that GAs are a powerful methods. Section 3 presents the experiments and analyzes the
and robust optimisation technique. results. Finally, Section 4 draws the main conclusions.
The hexagonal tessellation is a regular tiling of the
Euclidean plane, in which each vertex meets three hexagons 2. Main Algorithms and Methods
[17, 18]. There are two other regular tessellations of the plane,
namely, the triangular and the square tilings. Nevertheless, In this section are introduced briefly some aspects of FC and
the hexagonal tessellation constitutes the best way to divide Laplace transform and the computational implementation of
a given surface into regions of equal areas, while having GA.
the least total perimeter. This forms the so-called “hon-
eycomb conjecture” that dates back to the ancient Greek 2.1. Fractional Calculus. The most used definitions of a
mathematician Pappus of Alexandria (c. 290–c. 350) and fractional derivative of order 𝛼 are the Riemann-Liouville,
was proven in 1999 by Hales [19]. We find this structure Grünwald-Letnikov, and Caputo formulations [29]. Frac-
in nature, such as crystals or honeycombs, built by honey tional derivatives capture the history of past events, contrary
bees, and in man-made structures [20, 21], or even as art in to integer derivatives that are merely “local” operators. This
the famous Maurits Escher woodcuts and lithographs [22]. property has been recognized both in natural and man-made
Many other examples can be mentioned such as graphene and phenomena, where modelling becomes simpler using FC
superbenzene, substances with atoms arranged in a regular rather than building complicated integer order expressions.
hexagonal structure [23], or pineapples [24], a fruit with a Using the Laplace transform, for zero initial conditions,
rough skin having a hexagonal pattern of nodules. we have the expression
The three scientific concepts are put together for sim-
ulating and modelling an evolutionary process in a two- L {𝐷𝑡𝛼 𝑓 (𝑡)} = 𝑠𝛼 L {𝑓 (𝑡)} , (1)
dimensional space. First, it is considered a plane where some
kind of process evolves. The plane is discretized by means where 𝑠 and L denote the Laplace variable and operator,
of a regular hexagonal pattern, and the evolution consists of respectively.
the optimization using a standard GA. Second, the evolution In the scope of FC it is also important to mention the
of the GA population is described using a fractional order Mittag-Leffler function 𝐸𝛼 (𝑡) defined as [30–33]
model that approximates the numerical results. For that

purpose, the best individual in each generation of the GA 𝑡𝑘
𝐸𝛼 (𝑡) = ∑ , 𝛼 ∈ C, Re (𝛼) > 0. (2)
population is analysed in the viewpoint of fitness function,
𝑘=0
Γ (𝛼𝑘 + 1)
compared with the previous case, and the result is converted
into the Fourier domain. An important aspect is also the The Mittag-Leffler function is a generalization of the
fitness function that measures the “performance” of each GA exponential and the power laws. The first occurs in phenom-
individual. In the two-dimensional space are considered two ena governed by integer dynamics, and the second emerges in
distinct types of objects, and it is assumed that a “good per- fractional dynamics. In particular, when 𝛼 = 1 yields 𝐸1 (𝑡) =
formance” corresponds to a spacial arrangement exhibiting 𝑒𝑡 , while, for large values of 𝑡, the asymptotic behaviour of the
some type of interface between them. By other words, it is ML leads to 𝐸𝛼 (−𝑡) ≈ (1/Γ(1 − 𝛼))(1/𝑡), 𝛼 ≠1, 0 < 𝛼 < 2.
assumed that some kind of cooperation, or synergy, exists Applying the Laplace transform
between the two objects, such that they should coexist close
to each other in space. The resulting time-space population 𝑠𝛼−1
reveals fractal characteristics and patterns resembling those L {𝐸𝛼 (±𝑎𝑡𝛼 )} = , (3)
𝑠𝛼 ∓ 𝑎
of percolation [25, 26], in the inanimate world, or of lichens,
when thinking in living organisms [27, 28]. The possible we verify the generalization from the exponential up to the
examples correspond only to possible interpretations of the Mittag-Leffler function, that is, from integer up to fractional
abstract algorithm implemented in the paper. Percolation powers of 𝑠.
is the phenomenon involved in the movement and filtering These results mean that standard methods in mod-
of fluids through porous materials. Nevertheless, in the last elling and control, such as transfer functions and frequency
years percolation brought a new light into many topics response, can be directly applied as long as we allow the
such as material science, epidemiology, or geology. On the substitution of integer orders by their fractional counterparts.
other hand, lichens are organisms consisting of two partners,
namely, a fungus and a green alga growing in a symbiotic 2.2. Genetic Algorithms. GAs are a computer method to
relationship. The body of a lichen consists of fungal filaments find approximate solutions in optimization problems. GAs
surrounding the cells of the algae. The basis of the symbiosis are implemented such that a population of 𝑁 possible
in lichens is that the fungus provides the algal protection and solutions evolves with successive iterations towards better
gains nutrients in return. Therefore, such examples are merely approximations. In the GA formulation it is necessary to
possible interpretations of the simulation results, but, in fact, define the genetic representation of the problem and the
an abstract formulation is the basis of the proposed study that fitness function that measures how successfully a given
primarily intends to model the GA evolution with FC tools. individual approximates the solution. In the GA execution
Abstract and Applied Analysis 3

the population is initialized randomly and after it is improved


applying iteratively the operations of mutation, crossover, and
selection that mimic Darwin’s theory. During the evolution
a given part of the population is selected to breed the new
generation. Solutions are selected by means of the fitness
function. Therefore, those individuals that have the best
fitness values are preferred. The GA execution is ended when 𝑖, 𝑗 Neighbour 𝒜
some predefined condition is obeyed, such as when the
maximum number of generations 𝑡max is reached or when
a satisfactory fitness value is obtained. The technique of
“elitism” is often adopted that allows the better individuals to
carry over, unaltered, to the next generation.
The pseudo-code of a GA is as follow:
(1) generate randomly the initial population of individu- Figure 1: Hexagonal tessellation of the plane and definition of
neighbour set A of cell (𝑖, 𝑗).
als (solutions);
(2) evaluate the fitness function for each individual in the
population; (or identical) types of objects. Therefore, the logical operation
(3) repeat: (𝑖, 𝑗) ⊥ A yields a value between 0 and 6, with each term “1”
(or “0”) being the result of verifying if cell (𝑖, 𝑗) is occupied
(a) select the individuals with best fitness value for with “object type 1” (or “object type 2”) and a given cell in A is
reproducing; occupied with “object type 2” (or “object type 1”). For “empty”
(b) treat the population by means of the crossover cells, or for cells near the boundary of the tessellation space,
and mutation operators and produce offspring; no logical action is performed.
(c) evaluate the fitness value of each individual in In the numerical experiments neighbours with more
the offspring; cells were tested, namely, with a second and a third ring
having 12 and 18 hexagons and having different weights.
(d) replace the worst ranked part of previous pop-
Nevertheless, the results were qualitatively of the same type,
ulation by the best individuals of the produced
and, therefore, these experiments are not reported in this
offspring;
paper. Furthermore, for testing the surrounding cells different
(e) until termination. fitness functions were also evaluated. Again, while leading to
different plots, the results were not significantly distinct and
3. Numerical Experiments are not described in the sequel.
We must note that fitness (4) is straightforward to calcu-
In this section we describe the experiments with the GA late, leading to a fast computational implementation, while
and we analyse the results in the perspective of fractional accessing as “good” a genetic species that includes a high
dynamics. number of variations between objects in neighbour cells. This
abstract notion of discontinuity can be interpreted according
3.1. Genetic Algorithm Using Hexagonal Tessellation. We to the type of application. As mentioned in Introduction
consider a two-dimensional space subdivided in discrete we can interpret for percolation as some type of interface
cells and where three types of situation may occur. The between two distinct materials or for lichens as the interface
cells consists of a tessellation using regular hexagons as between two symbiotic species.
represented in Figure 1, where (𝑖, 𝑗) denotes the cell indexing For the crossover operation a simple one-point scheme
and the six gray cells define the set A of neighbours. is considered. First, the indices 𝑖 and 𝑗, for the one-point
Each cell has three possible situations, namely, “empty,” crossover, are randomly generated. Second, the individual 1
“occupied with object type 1,” and “occupied with object type of the offspring is generated by selecting the upper left and
2.” lower right corners of parent 1 and the upper right and lower
In the GA these objects interact by means of a fitness left corners of parent 2. The individual 2 of the offspring is
function 𝐽 defined as obtained using the complementary selection of the corners
in parents 1 and 2.
𝑖max 𝑗max
𝐽 = ∑ ∑ {(𝑖, 𝑗) ⊥ A} , (4)
𝑖=1 𝑗=1 3.2. Numerical Experiments and Fractional Dynamics. In
this subsection experiments with GA populations of 𝑁 =
where 𝑖max and 𝑗max denote the maximum values for indices {100, 200, 500, 1000} individuals and a two-dimensional
𝑖 and 𝑗, respectively. The notation (𝑖, 𝑗) ⊥ A describes the space are developed such that 𝑖max = 30 and 𝑗max = 60.
logical operation of comparing the object present in cell The GA terminates for 𝑡max = 103 generations. In the plots
(𝑖, 𝑗) with the set of six neighbours A and incrementing the are adopted the colours white, red, and green for the cases of
value of 𝐽 by “1” (or “0”) each time the cells have different cells “empty,” “occupied with object type 1,” and “occupied with
4 Abstract and Applied Analysis

0.7

0.6

0.5

0.4

𝑑(𝑡)
0.3

Figure 2: GA result for 𝑁 = 100, 𝑃0 = {0.2, 0.4, 0.4}. 0.2

0.1

0
0 200 400 600 800 1000
𝑡

Figure 4: Plot of 𝑑(𝑡) for 𝑁 = 100, 𝑃0 = {0.33, 0.33, 0.33}.

Re
0 10 20 30 40
0

Figure 3: GA result for 𝑁 = 100, 𝑃0 = {0.8, 0.2, 0.2}.

−5
object type 2,” respectively. Moreover, different initializations
of the GA population, such that the three types of cells have
distinct probabilities, are also tested. Let us represent the −10
Im

initialization probabilities 𝑃0 = {𝑤ℎ𝑖𝑡𝑒, 𝑟𝑒𝑑, 𝑔𝑟𝑒𝑒𝑛}. In the


sequel the cases 𝑃0 = {0.2, 0.4, 0.4}, 𝑃0 = {0.33, 0.33, 0.33},
𝑃0 = {0.5, 0.25, 0.25}, and 𝑃0 = {0.8, 0.1, 0.1} are considered,
−15
corresponding to environmental conditions varying from
“fertile” up to “arid.” For the GA parameters the one-
point, crossover with tournament selection is adopted, 100%
crossover rate and elitism, a mutation probability of 0.05. −20
Figures 2 and 3 show the plots resulting for 𝑁 =
Experimental
100 with 𝑃0 = {0.2, 0.4, 0.4} and 𝑃0 = {0.8, 0.2, 0.2}, Approximation
respectively. We observe clearly the fractal structure in space
and the interlacing between the two distinct types of objects. Figure 5: Polar diagram of 𝑑(𝑡) and its approximation (7) for 𝑁 =
Moreover, the effect of the initial conditions is also clear, that 100, 𝑃0 = {0.33, 0.33, 0.33} and 0.001 ≤ 𝜔 ≤ 0.03.
is, the outcome of having rich or poor populating conditions
of the tessellated plane.
The analysis of the GA dynamics [34, 35] in space time where the notation (𝑖, 𝑗)𝑡 ¬(𝑖, 𝑗)𝑡−1 describes the logical opera-
requires the definition and clarification of several concepts. tion of comparing the objects present in cell (𝑖, 𝑗) at iterations
In this line of thought, the term “space” represents the plane 𝑡 and 𝑡−1 and incrementing the value of 𝑑 by “1” (or “0”) if the
where the objects are laid. Since we consider an hexagonal cells have different (or identical) types of objects. Therefore,
tessellation, the space points are represented by cells (𝑖, 𝑗), 𝑑 yields the value 0 (𝑖max × 𝑗max ) for two completely identical
𝑖 = 1, . . . , 𝑖max , and 𝑗 = 1, . . . , 𝑗max . The term “time” (distinct) consecutive best individuals.
denotes the iteration time consisting of the successive GA Figure 4 depicts 𝑑(𝑡) for 𝑁 = 100, 𝑃0 = {0.33, 0.33, 0.33}.
generations 𝑡 = 1, . . . , 𝑡max . For the characterization of We observe a vanishing transient with severe discontinuities.
the GA population the individual with best fitness value The vanishing value means the convergence towards the final
𝐽 is considered, since usually in GA applications only the value while the discontinuities reveal that often successive
best solution is selected. Furthermore, for describing the generations change only slightingly or even do not evolve
dynamics, 𝑑, an index measuring the “distance” between two at all. These transients vary with the initial conditions,
consecutive best individuals is considered, defined as that is, with the probabilities 𝑃0 and the GA parameters,
namely, the population size 𝑁. Therefore, in the sequel several
𝑖max 𝑗max combinations of values of 𝑃0 and 𝑁 are tested.
𝑑 (𝑡) = ∑ ∑ {(𝑖, 𝑗)𝑡 ¬(𝑖, 𝑗)𝑡−1 } , (5) It was decided to identify a parametric model in the
𝑖=1 𝑗=1 Fourier domain since usually it leads to a simple and robust
Abstract and Applied Analysis 5

90
0.012
80
70 0.01
60 0.008
𝐾 50 𝑝 0.006
40
30 0.004
20
0.002
10
0 0
0.1 0.1

0.8 0.8
𝑁
𝑁 (× 0.6
(× 0.6 10
10 00
00 )
) 0.4 𝑃0 0.4 𝑃0

10.2 1 0.2

Figure 6: Variation of the transfer function parameter 𝐾 versus 𝑁 Figure 7: Variation of the transfer function parameter 𝑝 versus 𝑁
and 𝑃0 . and 𝑃0 .

procedure. Therefore, the Fourier transform of each time


response, 𝐷(𝚤𝜔) = F[𝑑(𝑡)], was determined. For the
identification several transfer functions were tested trying to
establish a compromise between accuracy and complexity,
1.25
while trying to preserve the same type of expression for all
cases under study. The final choice fell on a function 𝐺(𝑠) with 1.2
3 parameters given by the expression
1.15
𝐾 −𝑠𝜏 𝛼
𝐺 (𝑠) = 𝛼𝑒 , (6)
1 + (𝑠/𝑝) 1.1

1.05
where 𝐾 denotes the gain, 𝑝 represents a pole of fractional
order 𝛼, and 𝜏 stands for a time delay. 1
For example, Figure 5 depicts the polar diagram of the 0.1
experimental result and approximation (6), that is, Re =
R{𝐷(𝚤𝜔)} versus Im = I{𝐷(𝚤𝜔)}, for 𝑁 = 100 and 𝑃0 = 0.8
{0.33, 0.33, 0.33}. Several experiments demonstrated that the 𝑁
(× 0.6
low frequency content of 𝐷(𝚤𝜔) is invariant with different 10
00
GA seeds, in opposition with the high frequency behaviour ) 0.4 𝑃0
that reflects the stochastic nature of the algorithm and reveals
noisy characteristics in the Fourier domain. Therefore, in the 10.2
sequel a bandwidth limitation is considered such that 𝜔 ≤ Figure 8: Variation of the transfer function parameter 𝛼 versus 𝑁
0.03. and 𝑃0 .
Given the large number of combinations of values for
𝑁 and 𝑃0 , the identification was performed automatically by
means of a second GA having a population of 500 individuals Figures 6, 7, 8, and 9 show the variation of the transfer
and terminating after calculating 500 iterations. Several tests function parameters {𝐾, 𝑝, 𝛼, 𝜏} versus 𝑁 and 𝑃0 .
demonstrated that good identification results were produced We observe that the parameters of the transfer function
by the fitness function: (6) have the following behaviour:

𝜔=0.03 (i) 𝐾 grows with 𝑁 and has a maximum for 𝑃0 = {0.5,


∑ {R [𝐷 (𝚤𝜔)] −R [𝐺 (𝚤𝜔)]}2 +{I [𝐷 (𝚤𝜔)] −I [𝐺 (𝚤𝜔)]} . 0.25, 0.25};
𝜔=0.001
(7) (ii) 𝑝 decreases with 𝑁 and is independent of 𝑃0 ;
6 Abstract and Applied Analysis

between distinct objects in the population. This scheme has


an abstract nature but can be interpreted as representing
a simplified version of some kind of interaction between
distinct materials or, even, as symbiotic relation between
0.5 two different species. During the experiments several condi-
tions for the GA evolution were tested, namely, with initial
0.4 populations having different number of objects and distinct
0.3
percentages of each type. For describing the space-time GA
dynamics a simple measure was defined. The index, inspired
𝜏
0.2 in the notion of distance, captures the differences between
two consecutive best individuals. It is verified, in all cases, that
0.1 the proposed description leads to simple models capable of
0 being traduced by analytical expressions of fractional order.
0.1

References
0.8
[1] S. G. Samko, A. A. Kilbas, and O. I. Marichev, Fractional Inte-
𝑁
(× 0.6 grals and Derivatives, Gordon and Breach Science, Amsterdam,
10
00 The Netherlands, 1993.
) 0.4 𝑃0
[2] K. S. Miller and B. Ross, An Introduction to the Fractional
1 0.2 Calculus and Fractional Differential Equations, John Wiley &
Sons, New York, NY, USA, 1993.
Figure 9: Variation of the transfer function parameter 𝜏 versus 𝑁 [3] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory and Ap-
and 𝑃0 . plications of Fractional Differential Equations, vol. 204 of North-
Holland Mathematics Studies, Elsevier Science B.V., Amsterdam,
The Netherlands, 2006.
(iii) 𝛼 has a fractional value and seems to be almost [4] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional
independent of 𝑁 and grows slightly with 𝑃0 ; Calculus: Models and Numerical Methods, vol. 3 of Series on
(iv) 𝜏 has not a clear relationship with 𝑁 or 𝑃0 . It has a Complexity, Nonlinearity and Chaos, World Scientific, Boston,
Mass, USA, 2012.
small value and its average is 𝜏av = 0.3.
[5] A. B. Malinowska and D. F. M. Torres, Introduction to the Frac-
We verify that we can model the GA dynamical behaviour tional Calculus of Variations, Imperial College Press, London,
in terms of a simple fractional order model. In fact, stochastic UK, 2012.
results of the GA seem to be of minor influence in the [6] O. Heaviside, “On operators in physical mathematics,” Proceed-
proposed modelling scheme, not only due to the com- ings of the Royal Society, vol. 52, pp. 504–529, 1893.
prehensive variation of the parameters, but also because [7] R. L. Magin, Fractional Calculus in Bioengineering, Begell
several numerical experiments with distinct seeds lead to House, Redding, Calif, USA, 2006.
similar results. Nevertheless, we have a phenomenal mod- [8] C. M. Ionescu and R. De Keyser, “Relations between fractional-
elling perspective based on the analytical approximation and order model parameters and lung pathology in chronic obstruc-
supported by the outcome results. Therefore, a new challenge tive pulmonary disease,” IEEE Transactions on Biomedical
Engineering, vol. 56, no. 4, pp. 978–987, 2009.
is the inverse problem. By other words, the problem of
defining the fractional order model remains open and, as [9] C. Ionescu and J. T. MacHado, “Mechanical properties and
impedance model for the branching network of the sapping
a consequence, designing the GA rules that produce such
system in the leaf of Hydrangea Macrophylla,” Nonlinear
dynamical behaviour. In the scope of this problematic it can Dynamics, vol. 60, no. 1-2, pp. 207–216, 2010.
not be forgotten that the GA optimizes a symbiotic behaviour
[10] F. Mainardi, Fractional Calculus and Waves in Linear Viscoelas-
using a regular hexagonal space tessellation. Therefore, at a ticity: An Introduction to Mathematical Models, Imperial College
higher level several problems remain open such as the design Press, London, UK, 2010.
of other fitness functions, the effect of other tessellation [11] J. T. Machado, V. Kiryakova, and F. Mainardi, “Recent history of
methods, or the phenomena produced by a larger number of fractional calculus,” Communications in Nonlinear Science and
object types in the GA population. Numerical Simulation, vol. 16, no. 3, pp. 1140–1153, 2011.
[12] J. H. Holland, Adaptation in Natural and Artificial Systems, Uni-
4. Conclusions versity of Michigan Press, Ann Arbor, Mich, USA, 1975.
[13] D. E. Goldberg, Genetic Algorithms in Search Optimization, and
This paper presented a GA dynamical evolution and its Machine Learning, Addison-Wesley, Reading, Mass, USA, 1989.
description by means of a fractional model. The GA adopts [14] J. R. Koza, Genetic Programming: On the Programming of Com-
a tessellation of the space using regular hexagons in order to puters by Means of Natural Selection, MIT Press, Cambridge,
provide an efficient scheme to handle the neighbour cells in Mass, USA, 1992.
the numerical discretization. Furthermore, the GA includes [15] Z. Michalewicz, Genetic Algorithms + Data Structures = Evolu-
a fitness function that evaluates positively the interface tion Programs, Springer, Berlin, Germany, 1996.
Abstract and Applied Analysis 7

[16] T. Bäck, U. Hammel, and H.-P. Schwefel, “Evolutionary compu-


tation: comments on the history and current state,” IEEE
Transactions on Evolutionary Computation, vol. 1, no. 1, pp. 3–17,
1997.
[17] B. Grünbaum and G. C. Shephard, Tilings and Patterns, W. H.
Freeman and Company, New York, NY, USA, 1987.
[18] D. Chavey, “Tilings by regular polygons. II. A catalog of tilings,”
Computers & Mathematics with Applications, vol. 17, no. 1–3, pp.
147–165, 1989.
[19] T. C. Hales, “The honeycomb conjecture,” Discrete & Computa-
tional Geometry, vol. 25, no. 1, pp. 1–22, 2001.
[20] I. Stojmenovic, “Honeycomb networks: topological properties
and communication algorithms,” IEEE Transactions on Parallel
and Distributed Systems, vol. 8, no. 10, pp. 1036–1042, 1997.
[21] C. Bays, “A note on the game of life in hexagonal and pentagonal
tessellations,” Complex Systems, vol. 15, no. 3, pp. 245–252, 2005.
[22] D. R. Hofstadter, Gödel, Escher, Bach: An Eternal Golden Braid,
Basic Books, New York, NY, USA, 1979.
[23] J. Ma, A. Michaelides, and D. Alfè, “Binding of hydrogen on
benzene, coronene, and graphene from quantum Monte Carlo
calculations,” Journal of Chemical Physics, vol. 134, no. 13, Article
ID 134701, 2011.
[24] J. Kappraff, “Growth in plants: a study in number,” Forma, vol.
19, no. 4, pp. 335–354, 2004.
[25] H. Kesten, “What is...percolation?” Notices of the American
Mathematical Society, vol. 53, no. 5, pp. 572–573, 2006.
[26] G. R. Grimmett and l. Manolescu, “Inhomogeneous bond per-
colation on square, triangular, and hexagonal lattices,” submit-
ted to Annals of Probability, http://arxiv.org/abs/1105.5535.
[27] R. W. Sumner, Pattern formation in lichen [M.S. thesis], Massa-
chusetts Institute of Technology, Cambridge, Mass, USA, 2001.
[28] B. Desbenoit, E. Galin, and S. Akkouche, “Simulating and mod-
eling lichen growth,” LIRIS Research Report 2672, LIRIS FRE,
Lyon, France, 2004.
[29] I. Podlubny, “Fractional-order systems and 𝑃𝐼𝜆 𝐷𝜇 -controllers,”
IEEE Transactions on Automatic Control, vol. 44, no. 1, pp. 208–
214, 1999.
[30] K. B. Oldham and J. Spanier, The Fractional Calculus: Theory
and Applications of Differentiation and Integration to Arbitrary
Order, Academic Press, New York, NY, USA, 1974.
[31] V. S. Kiryakova, Generalized Fractional Calculus and Applica-
tions, vol. 301 of Pitman Research Notes in Mathematics Series,
Longman Scientific & Technical, New York, NY, USA, 1994.
[32] I. Podlubny, Fractional Differential Equations: An Introduction
to Fractional Derivatives, Fractional Differential Equations, to
Methods of Their Solution and Some of Their Applications, vol.
198 of Mathematics in Science and Engineering, Academic Press,
San Diego, Calif, USA, 1999.
[33] H. J. Haubold, A. M. Mathai, and R. K. Saxena, “Mittag-
Leffler functions and their applications,” Journal of Applied
Mathematics, vol. 2011, Article ID 298628, 51 pages, 2011.
[34] E. J. S. Pires, J. A. T. Machado, and P. B. Oliveira, “Fractional
order dynamics in a GA planner,” Signal Processing, vol. 83, no.
11, pp. 2377–2386, 2003.
[35] E. J. S. Pires, J. A. T. Machado, and P. B. Oliveira, “Dynamical
modelling of a genetic algorithm,” Signal Processing, vol. 86, no.
10, pp. 2760–2770, 2006.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 505903, 29 pages
http://dx.doi.org/10.1155/2013/505903

Research Article
An Operational Matrix Based on Legendre Polynomials for
Solving Fuzzy Fractional-Order Differential Equations

Ali Ahmadian,1,2 Mohamed Suleiman,1 and Soheil Salahshour3


1
Institute for Mathematical Research, Universiti Putra Malaysia, 43400 Serdang, Selangor, Malaysia
2
Department of Mathematics, Science Faculty, Universiti Putra Malaysia, 43400 Serdang, Selangor, Malaysia
3
Young Researchers and Elite Club, Islamic Azad University, Mobarakeh Branch, P.O. Box 9189945113, Mobarakeh, Iran

Correspondence should be addressed to Ali Ahmadian; ahmadian.hosseini@gmail.com

Received 20 April 2013; Accepted 16 May 2013

Academic Editor: Dumitru Baleanu

Copyright © 2013 Ali Ahmadian et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper deals with the numerical solutions of fuzzy fractional differential equations under Caputo-type fuzzy fractional
derivatives of order 𝛼 ∈ (0, 1). We derived the shifted Legendre operational matrix (LOM) of fuzzy fractional derivatives for the
numerical solutions of fuzzy fractional differential equations (FFDEs). Our main purpose is to generalize the Legendre operational
matrix to the fuzzy fractional calculus. The main characteristic behind this approach is that it reduces such problems to the
degree of solving a system of algebraic equations which greatly simplifies the problem. Several illustrative examples are included to
demonstrate the validity and applicability of the presented technique.

1. Introduction On the other hand, the modeling of natural phenomena


is stated using mathematical tools (mathematical arithmetic,
The subject of fractional calculus has gained considerable mathematical logics, etc.). However, obtaining a determinis-
popularity and importance during the past three decades. tic model of such problems is not easy, even does not occur
Some of the most recent works on this topic, that is, the and always has some errors and vagueness. So, investigating
theory of derivatives and integrals of fractional (noninteger) a popular way to interpret such vagueness is important.
order, are such as the book of Podlubny [1], Baleanu et Since 1965 with Zadeh’s well-known paper on introducing
al. [2], Diethelm [3], Baleanu et al. [4], and Sabatier et al. fuzzy sets, applications of fuzzy concept to the structure
[5]. Only in the last few years, the various applications of of any modeling has appeared more and more, instead of
fractional calculus have been extended in the area of physics deterministic case. So the topic of fuzzy differential equations
and engineering such as the modeling of nonlinear oscillation (FDEs) of integer order has been rapidly growing in recent
of earthquake [6], the fluid-dynamic models [7], continuum years [14–20]. Additionally, the application of various tech-
and statistical mechanics [8], and solid mechanics [9]. In niques has been expanded by means of the interpolations and
a notably enormous number of recent works, one can find polynomials for approximating the fuzzy solutions of fuzzy
the robustness upon the usefulness of fractional calculus integral equations vastly, like Bernstein polynomials [21, 22],
to derive particular solutions of different kinds of classical Lagrange interpolation [23, 24], Chebyshev interpolation
differential equations like Bessel differential equation of [25], Legendre wavelets [26], and Galerkin-type technique
general order [10, 11]. Also, the most significant advantage [27].
of applying FDEs is their nonlocal property, which interprets Recently, Agarwal et al. [28] proposed the concept
that the next state of a system relies not only pon its current of solutions for the fractional differential equations with
phase but also pon all of its past records of phases [12]. uncertainty. They have considered the Riemann-Liouville’s
For example, with the fractional differentiability, the fluid differentiability with a fuzzy initial condition to solve FFDEs.
dynamic traffic model can get rid of the shortage arising from In [29, 30], the authors considered the generalization of H-
the hypothesis of continuum traffic flow [12, 13]. differentiability for the fractional case. Discovering a suitable
2 Abstract and Applied Analysis

approximate or exact solution for FFDEs is a significant task generalized differentiability. Also, we introduce a suitable way
which has been aroused simultaneously with the emerging to estimate the nonlinear fuzzy fractional initial problems on
of FFDES, except for a few number of these equations, the interval [0, 1], by spectral shifted Legendre collocation
and we have hardship in finding their analytical solutions. method based on Legendre operational matrix, to find the
Consequently, there have been limited efforts to develop new approximate fuzzy solution. Finally, the accuracy of the pro-
methods for gaining approximate solutions which reasonably posed algorithms is demonstrated by several test problems.
estimate the exact solutions. Salahshour et al. [31] considered We note that the two shifted Legendre and shifted Jacobi
fuzzy laplace transforms for solving FFDEs under Riemann- operational matrices have been introduced by Saadatmandi
Liouville H-differentiability. Also Mazandarani and Kamyad and Dehghan [36] and Doha et al. [39], respectively, in the
[32] generalized the fractional Euler method for solving crisp concept. We, therefore, motivated our interest in the
FFDEs under Caputo-type derivative. shifted Legendre operational matrix in the fuzzy settings.
From another point of view, several methods have been This paper is organized as follows: In Section 2, we begin
exploited to solve fractional differential equations, and frac- by introducing some necessary definitions and mathematical
tional partial differential equations, fractional integrodiffer- preliminaries of the fuzzy calculus and fractional calculus.
ential equations such as Adomian’s decomposition method Some basic concepts, properties and theorems of fuzzy
[7], He’s variational iteration method [33], homotopy per- fractional calculus are presented in Section 3. Section 4 is
turbation method [34], and spectral methods [35, 36]. In devoted to the fuzzy Legendre functions and their properties.
this way, orthogonal functions have received considerable The shifted Legendre operational matrix of fuzzy fractional
attention in dealing with the various kinds of fractional derivative for solving fuzzy fractional differential equation is
differential equations. The main characteristic behind the obtained in Section 5. Section 6 illustrates the effectiveness
approach using this technique is that it reduces these prob- of the proposed method through solving several examples
lems to those of solving a system of algebraic equations thus which some of them are modelled based on the real phenom-
greatly simplifying the problem. Saadatmandi and Dehghan ena. Finally, a conclusion is given in the last section.
[36] presented the shifted Legendre operational matrix for
fractional derivatives and applied it with tau method for the 2. Preliminaries
numerical solution of fractional differential equations subject
to initial conditions. Also in [37–39], the authors derived new We give some definitions and introduce the necessary nota-
formulas using shifted Chebyshev polynomials and shifted tion which will be used throughout the paper, see, for exam-
Jacobi polynomials of any degree, respectively and applied ple, [40, 41]. Also for some definitions related to generalized
them together with tau and collocation spectral methods for fuzzy difference, one can find more in [42, 43].
solving multiterm linear and nonlinear fractional differential We denote the set of all real numbers by R. A fuzzy
equations. number is a mapping 𝑢̃ : R → [0, 1] with the following
The essential target of this paper is to recommend a properties:
suitable way to approximate FFDEs using a shifted Leg-
endre tau approach. This strategy demands a formula for (a) 𝑢̃ is upper semicontinuous,
fuzzy fractional-order Caputo derivatives of shifted Legendre (b) 𝑢̃ is fuzzy convex, that is, 𝑢̃(𝜆𝑥 + (1 − 𝜆)𝑦 ≥
polynomials of any degree which is provided and applied min{̃ 𝑢(𝑥), 𝑢̃(𝑦)} for all 𝑥, 𝑦 ∈ R, 𝜆 ∈ [0, 1],
together with the tau method for solving FFDEs with initial
conditions. Up till now, and to the best of our knowledge, (c) 𝑢̃ is normal, that is, ∃𝑥0 ∈ R for which 𝑢̃(𝑥0 ) = 1,
few methods corresponding to those mentioned previously (d) supp 𝑢̃ = {𝑥 ∈ R | 𝑢̃(𝑥) > 0} is the support of the 𝑢,
have been devoted to solve FFDEs and are traceless in and its closure cl (supp 𝑢̃) is compact.
the literature for FFDEs under Caputo differentiability. This
partially motivates our interest in the operational matrix of Let E be the set of all fuzzy number on R. The 𝛼-level set
fuzzy fractional derivative of shifted Legendre polynomials. of a fuzzy number 𝑢̃ ∈ E, 0 ≤ 𝑟 ≤ 1, denoted by [̃ 𝑢]𝑟 , is
Also another motivation is based on the reality that only a defined as
few terms of expansion of the shifted Legendre function is {𝑥 ∈ R | 𝑢̃ (𝑥) ≥ 𝑟} if 0 < 𝑟 ≤ 1
needed to reach to a high accuracy, therefore, it does not 𝑢]𝑟 = {
[̃ (1)
need to implement the method frequently for finding the cl (supp 𝑢̃) if 𝑟 = 0.
approximate results in each particular point.
It is clear that the 𝑟-level set of a fuzzy number is a closed
For finding the fuzzy solution, the shifted Legendre
operational matrix is generalized for the fuzzy fractional and bounded interval [̃ 𝑢(𝑟), 𝑢̃(𝑟)], where 𝑢̃(𝑟) denotes the
derivative (0 < 𝛼 < 1) which is based on the Legendre tau left-hand endpoint of [̃ 𝑢]𝑟 and 𝑢̃(𝑟) denotes the right-hand
method for solving numerically FFDEs with the fuzzy initial endpoint of [̃𝑢]𝑟 . Since each 𝑦 ∈ R can be regarded as a fuzzy
conditions. It is worthy to note here that the method based number 𝑦̃ defined by
on using the operational matrix of the Legendre orthogonal
function for solving FFDEs is computer oriented. 1 if 𝑡 = 𝑦,
𝑦̃ (𝑡) = { (2)
The aim of this paper is to introduce the shifted Legendre 0 if 𝑡 ≠𝑦,
operational matrix of fuzzy fractional derivative which is
based on Legendre tau method for solving FFDEs under R can be embedded in E.
Abstract and Applied Analysis 3

The addition and scalar multiplication of fuzzy number Definition 5 (see [47]). Let 𝑥, 𝑦 ∈ E. If there exists 𝑧 ∈ E such
in E are defined as follows: that 𝑥 = 𝑦 ⊕ 𝑧, and then 𝑧 is called the H-difference of 𝑥 and
𝑦, and it is denoted by 𝑥 ⊖ 𝑦.
𝑢 + ̃V, 𝑢̃ + ̃V) ,
(1) 𝑢̃ ⊕ ̃V = (̃
In this paper, the sign “⊖” always stands for H-difference
{(𝜆̃ 𝑢 (𝑟) , 𝜆̃
𝑢 (𝑟)) 𝜆 ≥ 0, (3)
and note that 𝑥 ⊕ 𝑦 ≠𝑥 + (−𝑦). Also throughout the paper
(2) (𝜆 ⊙ 𝑢̃) = {
(𝜆̃𝑢 (𝑟) , 𝜆̃
𝑢 (𝑟)) 𝜆 < 0. is assumed that the Hukuhara difference and Hukuhara
{ generalized differentiability existed.
The metric structure is given by the Hausdorff distance
𝐷 : E × E → R+ ⋃ 0, Definition 6 (see [42]). The generalized difference (g-
difference for short) of two fuzzy numbers 𝑢, V ∈ E is given
󵄨 󵄨 󵄨 󵄨
𝑢, ̃V) = sup max {󵄨󵄨󵄨𝑢̃ (𝑟) − ̃V (𝑟)󵄨󵄨󵄨 , 󵄨󵄨󵄨󵄨𝑢̃ (𝑟) − ̃V (𝑟)󵄨󵄨󵄨󵄨} . (4)
𝐷 (̃ by its level sets as
𝑟∈[0,1]

It is easy to see that 𝐷 is a metric in E and has the following [𝑢 ⊖ g V]𝛼 = cl ⋃ ([𝑢]𝛽 ⊖gH [V]𝛽 ) ∀𝛼 ∈ [0, 1] , (8)
properties 𝛽≥𝛼

(i) 𝐷(̃
𝑢⊕𝑤 ̃, ̃V ⊕ 𝑤
̃) = 𝐷(̃𝑢, ̃V), for all 𝑢̃, ̃V, 𝑤
̃ ∈ E,
where the gH-difference ⊖gH is with interval operands [𝑢]𝛽
(ii) 𝐷(𝑘 ⊙ 𝑢̃, 𝑘 ⊙ ̃V) = |𝑘|𝐷(̃
𝑢, ̃V), for all 𝑘 ∈ R, 𝑢̃, ̃V ∈ E, and [V]𝛽 .
𝑢 ⊕ ̃V, 𝑤
(iii) 𝐷(̃ ̃ ⊕ 𝑒̃) ≤ 𝐷(̃ ̃) + 𝐷(̃V, 𝑒̃), for all 𝑢̃, ̃V, 𝑤
𝑢, 𝑤 ̃ ∈ E,
(iv) 𝐷(̃ ̃ ̃ ̃
𝑢 + ̃V, 0) ≤ 𝐷(𝑢, 0) + 𝐷(V, 0), for all 𝑢, V ∈ E, Proposition 7. The g-difference in Definition 6 is given by the
expression
(v) (E, 𝐷) is a complete metric space.

Definition 1. The property (iv) in the properties of the above [𝑢 ⊖ g V]𝛼 = [inf min {𝑢𝛽 − V𝛽 , 𝑢𝛽 − V𝛽 } ,
metric space suggests the definition of a function ‖ ⋅ ‖E : 𝑅 → 𝛽≥𝛼
̃ for all 𝑢 ∈ E that has the properties
E that ‖𝑢‖E := 𝐷(𝑢, 0), (9)
of usual norms. In [44], the properties of this function are
sup max {𝑢𝛽 − V𝛽 , 𝑢𝛽 − V𝛽 }] .
presented as follows: 𝛽≥𝛼
(i) ‖𝑢‖ ≥ 0, for all 𝑢 ∈ E and ‖𝑢‖ = 0 if and only if
𝑢 = 0̃, Proof. See [42].
(ii) ‖𝜆⋅𝑢‖ = |𝜆|⋅‖𝑢‖ and ‖𝑢+V‖ ≥ ‖𝑢‖+‖V‖, for all 𝑢, V ∈
E, for all 𝜆 ∈ R. The next proposition gives simplified notation for 𝑢 ⊖ g V
(iii) |‖𝑢‖ − ‖V‖| ≤ 𝐷(𝑢, V) and 𝐷(𝑢, V) ≤ ‖𝑢‖ + and V ⊖ g 𝑢.
‖V‖ for all 𝑢, V ∈ E.
Proposition 8. For any two fuzzy numbers 𝑢, V ∈ E the two g-
Definition 2 (see [45]). Let 𝑓 and 𝑔 be the two fuzzy- difference 𝑢 ⊖ g V and V ⊖ g 𝑢 exist and, for any 𝛼 ∈ [0, 1], one
number-valued functions on the interval [𝑎, 𝑏], that is, 𝑓, 𝑔 : 𝑢 ⊖ g V = −(V ⊖ g 𝑢) with
[𝑎, 𝑏] → E. The uniform distance between fuzzy-number-
valued functions is defined by [𝑢 ⊖ g V]𝛼 = [𝑑𝛼 , 𝑑𝛼 ] , [𝑢 ⊖ g V]𝛼 = [−𝑑𝛼 , −𝑑𝛼 ] , (10)

𝐷 (𝑓, 𝑔) := sup 𝐷 (𝑓 (𝑥) , 𝑔 (𝑥)) . (5)
𝑥∈[𝑎,𝑏] where
Remark 3 (see [46]). Let 𝑓 : [𝑎, 𝑏] → E be fuzzy continuous.
Then from property (iv) of Hausdorff distance, we can define 𝑑𝛼 = inf (𝐷𝛼 ) , 𝑑𝛼 = sup (𝐷𝛼 ) , (11)
󵄨 󵄨 󵄨󵄨 𝑟 󵄨󵄨
𝐷 (𝑓 (𝑥) , 0̃) = sup max {󵄨󵄨󵄨󵄨𝑓𝑟 (𝑥)󵄨󵄨󵄨󵄨 , 󵄨󵄨󵄨𝑓 (𝑥)󵄨󵄨󵄨} , ∀𝑥 ∈ [𝑎, 𝑏] . and the sets 𝐷𝛼 are
𝑟∈[0,1] 󵄨 󵄨
(6) 𝐷𝛼 = {𝑢𝛽 − V𝛽 | 𝛽 ≥ 𝛼} ∪ {𝑢𝛽 − V𝛽 | 𝛽 ≥ 𝛼} . (12)
Definition 4 (see [42]). Let 𝐾𝑐𝑛
be the space of nonempty
𝑛
compact and convex sets of R . The generalized Hukuhara Proof. See [42].
difference of two sets 𝐴, 𝐵 ∈ 𝐾𝑐𝑛 (gH-difference for short) is
defined as follows:
The following proposition prove that the g-difference is
(a) 𝐴 = 𝐵 + 𝐶 or well-defined.
𝐴 ⊖gH 𝐵 = 𝐶 ⇐⇒ { (7)
(b) 𝐵 = 𝐴 + (−1) 𝐶.
Proposition 9 (see [14]). For any fuzzy numbers 𝑢, V ∈ E the
In case (a) of the above equation, the gH-difference is g-difference 𝑢 ⊖ g V exists and it is a fuzzy number.
coincident with the H-difference. Thus the gH-difference is
a generalization of the H-difference. Proof. See [42].
4 Abstract and Applied Analysis

The following property holds for g-derivative. Theorem 13 (see [17]). Let 𝑓 : (𝑎, 𝑏) → E be a function and
denote [𝐹(𝑡)]𝑟 = [𝑓𝑟 (𝑡), 𝑔𝑟 (𝑡)], for each 𝑟 ∈ [0, 1]. Then
Proposition 10. Let 𝑢, V ∈ E be two fuzzy numbers, and then
(1) if 𝑓 is (1)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are
(i) 𝑢 ⊖ g V = 𝑢 ⊖gH V whenever the expressions on the right differentiable functions and
exist, in particular, 𝑢 ⊖ g 𝑢 = 0,
𝑟
(ii) (𝑢 + V)⊖ g V = 𝑢, [𝐹󸀠 (𝑡)] = [𝑓𝑟󸀠 (𝑡) , 𝑔𝑟󸀠 (𝑡)] , (17)
(iii) 0 ⊖ g (𝑢 ⊖ g V) = V ⊖ g 𝑢,
(iv) 𝑢 ⊖ g V = V ⊖ g 𝑢 = 𝑤 if and only if 𝑤 = −𝑤. (2) if 𝑓 is (2)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are
Furthermore, 𝑤 = 0 if and only if 𝑢 = V. differentiable functions and
𝑟
In this paper, we consider the following definition which [𝐹󸀠 (𝑡)] = [𝑔𝑟󸀠 (𝑡) , 𝑓𝑟󸀠 (𝑡)] . (18)
was introduced by Bede and Gal in [14].
Definition 14 (see [42]). Let 𝑓 : (𝑎, 𝑏) → E and 𝑥0 ∈
Definition 11 (see [14]). Let 𝑓 : (𝑎, 𝑏) → E and 𝑥0 ∈ (𝑎, 𝑏). (𝑎, 𝑏). We say that 𝑓 is g-differentiable at 𝑥0 , if there exists
One says that 𝑓 is strongly generalized differentiable at 𝑥0 , if an element 𝑓󸀠 (𝑥0 ) ∈ E such that
there exists an element 𝑓󸀠 (𝑥) ∈ E, such that
𝑓 (𝑥 + ℎ) ⊖ g 𝑓 (𝑥)
(i) for all ℎ > 0 sufficiently small, ∃𝑓(𝑥0 + ℎ) ⊖ 𝑓󸀠 (𝑥0 ) = lim . (19)
𝑓(𝑥0 ), ∃𝑓(𝑥0 )⊖𝑓(𝑥0 −ℎ), and the limits (in the metric ℎ→0 ℎ
𝐷) Next we review one of the main results from Bede [15] for
𝑓 (𝑥0 + ℎ) ⊖ 𝑓 (𝑥0 ) fuzzy initial value problem (FIVP) under (1)-differentiability
lim which Nieto et al. [48] generalized this results for FIVP under
ℎ → 0+ ℎ
(13) (2)-differentiability (let ‖⋅‖ denote the usual Euclidean norm).
𝑓 (𝑥0 ) ⊖ 𝑓 (𝑥0 − ℎ)
= lim+ = 𝑓󸀠 (𝑥0 ) , Theorem 15 (see [15], characterization theorem). Let one
ℎ→0 ℎ
consider the fuzzy initial value problem
(ii) for all ℎ > 0 sufficiently small, ∃𝑓(𝑥0 ) ⊖ 𝑓(𝑥0 +
ℎ), ∃𝑓(𝑥0 − ℎ) ⊖ 𝑓(𝑥0 ), and the limits (in the metric 𝑦󸀠 = 𝑓 (𝑥, 𝑦 (𝑥)) ,
𝐷) (20)
𝑦 (𝑡0 ) = 𝑦0 ,
𝑓 (𝑥0 ) ⊖ 𝑓 (𝑥0 + ℎ)
lim+ where 𝑓 : [𝑥0 , 𝑥0 + 𝑎] × E → E is such that
ℎ→0 −ℎ
(14) 𝑟
𝑓 (𝑥0 − ℎ) ⊖ 𝑓 (𝑥0 ) (i) [𝑓(𝑥, 𝑦)]𝑟 = [𝑓𝑟 (𝑥, 𝑦, 𝑦), 𝑓 (𝑥, 𝑦, 𝑦)],
= lim+ = 𝑓󸀠 (𝑥0 ) ,
ℎ→0 −ℎ 𝑟
(ii) 𝑓𝑟 and 𝑓 are equicontinuous (i.e., for any 𝜖 > 0 there
(iii) for all ℎ > 0 sufficiently small, ∃𝑓(𝑥0 + ℎ) ⊖ is a 𝛿 > 0 such that |𝑓𝑟 (𝑥, 𝑦, 𝑧) − 𝑓𝑟 (𝑥1 , 𝑦1 , 𝑧1 )| < 𝜖
𝑟 𝑟
𝑓(𝑥0 ), ∃𝑓(𝑥0 −ℎ)⊖𝑓(𝑥0 ), and the limits (in the metric and |𝑓 (𝑥, 𝑦, 𝑧) − 𝑓 (𝑥1 , 𝑦1 , 𝑧1 )| < 𝜖 for all 𝑟 ∈ [0, 1],
𝐷) whenever (𝑥, 𝑦, 𝑧), (𝑥1 , 𝑦1 , 𝑧1 ) ∈ [𝑥0 , 𝑥0 + 𝑎] × R2 and
𝑓 (𝑥0 + ℎ) ⊖ 𝑓 (𝑥0 ) ‖(𝑥, 𝑦, 𝑧) − (𝑥1 , 𝑦1 , 𝑧1 )‖ < 𝛿 and uniformly bounded on
lim any bounded set,
ℎ → 0+ ℎ
(15) (iii) there exists an 𝐿 > 0 such that |𝑓𝑟 (𝑥2 , 𝑦2 , 𝑧2 ) −
𝑓 (𝑥0 − ℎ) ⊖ 𝑓 (𝑥0 )
= lim+ = 𝑓󸀠 (𝑥0 ) , 𝑓𝑟 (𝑥1 , 𝑦1 , 𝑧1 )| ≤ 𝐿 max{|𝑦2 − 𝑦1 |, |𝑧2 − 𝑧1 |} for all 𝑟 ∈
ℎ→0 −ℎ 𝑟 𝑟
[0, 1], |𝑓 (𝑥2 , 𝑦2 , 𝑧2 ) − 𝑓 (𝑥1 , 𝑦1 , 𝑧1 )| ≤ 𝐿 max{|𝑦2 −
(iv) for all ℎ > 0 sufficiently small, ∃𝑓(𝑥0 ) ⊖ 𝑓(𝑥0 + 𝑦1 |, |𝑧2 − 𝑧1 |} for all 𝑟 ∈ [0, 1].
ℎ), ∃𝑓(𝑥0 ) ⊖ 𝑓(𝑥0 − ℎ), and the limits (in the metric
𝐷) Then the FIVP (20) and system of ODEs
󸀠
𝑓 (𝑥0 ) ⊖ 𝑓 (𝑥0 + ℎ) (𝑦𝑟 (𝑥)) = 𝑓𝑟 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) ,
lim+
ℎ→0 −ℎ
(16) 󸀠 𝑟
(𝑦𝑟 (𝑥)) = 𝑓 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) ,
𝑓 (𝑥0 ) ⊖ 𝑓 (𝑥0 − ℎ)
= lim+ = 𝑓󸀠 (𝑥0 ) . (21)
ℎ→0 ℎ
𝑦𝑟 (𝑥0 ) = (𝑦0𝑟 ) ,
Remark 12. Throughout this paper, we say that 𝑓 is (1)-
differentiable on (𝑎, 𝑏), if 𝑓 is differentiable in the sense (i) 𝑦 (𝑥0 ) = (𝑦0𝑟 ) ,
of Definition 11 and also 𝑓 is (2)-differentiable on (𝑎, 𝑏), if 𝑓
is differentiable in the sense (ii) of Definition 11. are equivalent.
Abstract and Applied Analysis 5

Corollary 16 (see [48]). If we consider FIVP (20) under (2)- The matrix form of the above equations is
differentiability then the FIVP (20) and the following system of
ODEs are equivalent: 𝐴𝑋 = 𝑌, (28)

󸀠 𝑟 where the coefficient matrix 𝐴 = (𝑎𝑖𝑗 ), 1 ≤ 𝑖, 𝑗 ≤ 𝑛 is a crisp


(𝑦𝑟 (𝑥)) = 𝑓 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) , 𝑛 × 𝑛 matrix and 𝑦𝑖 ∈ E, 1 ≤ 𝑖 ≤ 𝑛. This system is called a
fuzzy linear system (FLS).
󸀠
(𝑦𝑟 (𝑡)) = 𝑓𝑟 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) ,
(22) Definition 20 (see [50]). A fuzzy number vector (𝑥1 , 𝑥2 ,
𝑟
𝑦 (𝑥0 ) = (𝑦0𝑟 ) , . . . , 𝑥𝑛 )𝑡 given by 𝑥𝑖 = (𝑥𝑖 (𝑟), 𝑥𝑖 (𝑟)), 1 ≤ 𝑖 ≤ 𝑛, 0 ≤ 𝑟 ≤ 1
is called a solution of the fuzzy linear system (27) if
𝑦 (𝑥0 ) = (𝑦0𝑟 ) . 𝑛 𝑛
∑ 𝑎𝑖𝑗 𝑥𝑗 = ∑ 𝑎𝑖𝑗 𝑥𝑗 = 𝑦𝑖 ,
Theorem 17 (see [49]). Let 𝑓(𝑥) be a fuzzy-valued function 𝑗=1 𝑗=1

on [𝑎, ∞) and it is represented by (𝑓(𝑥; 𝑟), 𝑓(𝑥; 𝑟)). For any (29)
𝑛 𝑛
fixed 𝑟 ∈ [0, 1], assume that (𝑓(𝑥; 𝑟) and 𝑓(𝑥; 𝑟)) are Riemann- ∑𝑎𝑖𝑗 𝑥𝑗 = ∑𝑎𝑖𝑗 𝑥𝑗 = 𝑦𝑖 .
integrable on [𝑎, 𝑏] for every 𝑏 ≥ 𝑎, and assume that there are 𝑗=1 𝑗=1
two positive 𝑀(𝑟) and 𝑀(𝑟) such that:
If for a particular 𝑘, 𝑎𝑘𝑗 > 0, 1 ≤ 𝑗 ≤ 𝑛, we simply get
𝑏 𝑏
󵄨 󵄨 󵄨 󵄨 𝑛 𝑛
∫ 󵄨󵄨󵄨󵄨𝑓 (𝑥; 𝑟)󵄨󵄨󵄨󵄨 𝑑𝑥 ≤ 𝑀 (𝑟) , ∫ 󵄨󵄨󵄨󵄨𝑓 (𝑥; 𝑟)󵄨󵄨󵄨󵄨 𝑑𝑥 ≤ 𝑀 (𝑟) ∑ 𝑎𝑘𝑗 𝑥𝑗 = 𝑦𝑘 , ∑𝑎𝑘𝑗 𝑥𝑗 = 𝑦𝑘 .
𝑎 𝑎 (30)
𝑗=1 𝑗=1
for every 𝑏 ≥ 𝑎.
To solve fuzzy linear systems, one can refer to [51–53].

Now we define some notations which are used for the


(23)
fuzzy fractional calculus throughout the paper.
Then 𝑓(𝑥) is improper fuzzy Riemann-integrable on [𝑎, ∞) (i) 𝐿E𝑝 (𝑎, 𝑏), 1 ≤ 𝑝 < ∞ is the set of all fuzzy-valued
and the improper fuzzy Riemann-integral is a fuzzy number. measurable functions 𝑓 on [𝑎, 𝑏] where ||𝑓||𝑝 =
Further more, we have 1 1/𝑝
(∫0 (𝑑(𝑓(𝑡), 0))𝑝 𝑑𝑡) .
∞ ∞ ∞
∫ 𝑓 (𝑥) 𝑑𝑥 = [∫ 𝑓 (𝑥; 𝑟) 𝑑𝑥, ∫ 𝑓 (𝑥; 𝑟) 𝑑𝑥] . (24) E
(ii) 𝐶 [𝑎, 𝑏] is a space of fuzzy-valued functions which are
𝑎 𝑎 𝑎
continuous on [𝑎, 𝑏].
Definition 18 (see [45]). 𝑓(𝑥) : [𝑎, 𝑏] → E. We say that 𝑓 (iii) 𝐴𝐶E [𝑎, 𝑏] denotes the set of all fuzzy-valued functions
Fuzzy-Riemann integrable to 𝐼 ∈ E, if for any 𝜖 > 0, there which are absolutely continuous.
exists 𝛿 > 0 such that for any division 𝑃 = {[𝑢, V]; 𝜉} of [𝑎, 𝑏]
with the norms Δ(𝑃) < 𝛿, we have One can easily find this definition in the crisp sense in
[1, 54].

𝐷 (∑ (𝑢 − V) ⊙ 𝑓 (𝜁) , 𝐼) < 𝜖, (25) Definition 21 (see [54]). The Riemann-Liouville fractional
𝑃 integral operator of order 𝛼, 𝑛 − 1 < 𝛼 ≤ 𝑛, of a function
𝑓 ∈ 𝐶[𝑎, 𝑏] is defined as
where ∑∗ means addition with respect to ⊕ in E,
𝑥 𝑓 (𝑡)
𝛼 1
𝑏 (𝐼𝑎+ 𝑓) (𝑥) = ∫ 𝑑𝑡 𝑥 > 𝑎,
Γ (𝛼) 𝑎 (𝑥 − 𝑡)1−𝛼
𝐼 := (𝐹𝑅) ∫ 𝑓 (𝑥) 𝑑𝑥. (26)
(31)
𝑎 𝑏
𝛼 1 𝑓 (𝑡)
(𝐼𝑏− 𝑓) (𝑥) = ∫ 𝑑𝑡 𝑥 < 𝑏.
We also call an 𝑓 as above, (𝐹𝑅)-integrable. Γ (𝛼) 𝑥 (𝑡 − 𝑥)1−𝛼
Definition 19 (see [50]). Consider the 𝑛 × 𝑛 linear system of Properties of the operator 𝐼𝛼 can be found in [1, 54, 55].
equations we refer to only the following
For 𝑓 ∈ 𝐶[𝑎, 𝑏], 𝛼, 𝛽 ≥ 0 and 𝛾 > −1,
𝑎11 𝑥1 + 𝑎12 𝑥2 + ⋅ ⋅ ⋅ + 𝑎1𝑛 𝑥𝑛 = 𝑦1 ,
(1) 𝐼𝛼 𝐼𝛽 𝑓 (𝑥) = 𝐼𝛼+𝛽 𝑓 (𝑥) ,
𝑎21 𝑥1 + 𝑎22 𝑥2 + ⋅ ⋅ ⋅ + 𝑎2𝑛 𝑥𝑛 = 𝑦2 ,
(27) (2) 𝐼𝛼 𝐼𝛽 𝑓 (𝑥) = 𝐼𝛽 𝐼𝛼 𝑓 (𝑥) ,
.. (32)
.
Γ (𝛾 + 1) 𝛼+𝛾
(3) 𝐼𝛼 𝑥𝛾 = 𝑥 .
𝑎𝑛1 𝑥1 + 𝑎𝑛2 𝑥2 + ⋅ ⋅ ⋅ + 𝑎𝑛𝑛 𝑥𝑛 = 𝑦𝑛 . Γ (𝛼 + 𝛾 + 1)
6 Abstract and Applied Analysis

Definition 22 (see [54]). The Riemann-Liouville fractional Lemma 26. Let 0 < 𝛼 < 1 and 𝑓 ∈ 𝐶[𝑎, 𝑏] ∩ 𝐿 𝑝 [𝑎, 𝑏]. Then
derivatives of order 0 < 𝛼 < 1 of a function 𝑓 ∈ 𝐶[𝑎, 𝑏] the Caputo fractional derivatives are bounded for any 𝑥 ∈ [𝑎, 𝑏]
are expressed by and (1 ≤ 𝑝 < ∞) as
1 𝑑 𝑥 𝑓 (𝑡) 𝑑𝑡 󵄩󵄩󵄩𝑓󸀠 (𝑥)󵄩󵄩󵄩
𝛼
(𝐷𝑎+ 𝑓) (𝑥) = ∫ (𝑥 > 𝑎) , 󵄨󵄨 𝑐 𝛼 󵄨󵄨 󵄩󵄩 󵄩󵄩𝑝
Γ (1 − 𝛼) 𝑑𝑥 𝑎 (𝑥 − 𝑡)𝛼 󵄨󵄨( 𝐷𝑎+ 𝑓) (𝑥)󵄨󵄨 ≤ (𝑥 − 𝑎)1−𝛼 = 𝑀𝑎𝛼 ,
|Γ (1 − 𝛼)| [1 − 𝛼]
𝑥𝑓
(𝑡) 𝑑𝑡 󵄩󵄩 󸀠 󵄩󵄩 (40)
𝛼 1 𝑑 1
(𝐷𝑏− 𝑓)(𝑥) = − ∫ (𝑥 < 𝑏) . 󵄩󵄩𝑓 (𝑥)󵄩󵄩
Γ (1 − 𝛼) 𝑑𝑥 Γ (1 − 𝛼) 𝑎 (𝑡 − 𝑥)𝛼 󵄨󵄨 𝑐 𝛼 󵄨󵄨 󵄩 󵄩𝑝 1−𝛼
󵄨󵄨( 𝐷𝑏+ 𝑓) (𝑥)󵄨󵄨 ≤ |Γ (1 − 𝛼)| [1 − 𝛼] (𝑏 − 𝑥) = 𝑀𝑏𝛼 .
(33)

Definition 23 (see [55]). The fractional Caputo derivatives Proof. See [54, 57].
𝑐 𝛼
𝐷𝑎+ 𝑓(𝑥) and 𝑐 𝐷𝑏−
𝛼
𝑓(𝑥) on [𝑎, 𝑏] for 0 < 𝛼 < 1 are defined
via the above Riemann-Liouville fractional derivatives by 3. Fuzzy Caputo Fractional Derivatives
(𝑐 𝐷𝑎+
𝛼 𝛼
𝑓) (𝑥) = (𝐷𝑎+ [𝑓 (𝑡) − 𝑓 (𝑎)]) (𝑥) , In this section, some definitions and theorems related to the
𝛼
(34) fuzzy Caputo fractional derivatives are presented which are
( 𝑐 𝐷𝑏− 𝑓) (𝑥) = (𝐷𝑏−
𝛼
[𝑓 (𝑡) − 𝑓 (𝑏)]) (𝑥) , an extension of the fractional derivative in the crisp sense. The
generalized differentiability should be considered to expand
which can be simplified as the concept of Caputo fractional derivatives for the fuzzy
𝑓 (𝑎) space. For more details, see [14, 30].
𝛼
( 𝑐 𝐷𝑎+ 𝑓) (𝑥) = (𝐷𝑎+
𝛼
𝑓) (𝑥) − (𝑥 − 𝑎)−𝛼 ,
Γ (1 − 𝑎) Definition 27. Let 𝑓 : 𝐿E ∩ 𝐶E be a fuzzy set-value function
(35) 𝑥
𝛼 𝑓 (𝑏) and Φ(𝑥) = (1/Γ(1 − 𝛼)) ∫𝑎 (𝑓(𝑡)𝑑𝑡/(𝑥 − 𝑡)𝛼 ), and then 𝑓 is
( 𝑐 𝐷𝑏− 𝑓) (𝑥) = (𝐷𝑏−
𝛼
𝑓) (𝑥) − (𝑏 − 𝑥)−𝛼 . said to be g-Caputo fuzzy fractional differentiable at 𝑥, when
Γ (1 − 𝑎)
Also, the fractional Caputo derivative can be defined in a 𝛼
Φ (𝑥 + ℎ) ⊖ g Φ (𝑥)
( g 𝐷𝑎+ 𝑓) (𝑥) = lim , (41)
sense of integral form described in Definition 24. ℎ→0 ℎ
where
Definition 24 (see [56]). The Caputo definition of the
󸀠
fractional-order derivative is defined as 𝑥 𝑓 (𝑡; 𝑟) 𝑑𝑡
g 𝛼 1
( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) =[ ∫ ],
𝑐 𝛼 1 𝑥 𝑓𝑛 (𝑡) Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼
𝐷 𝑓 (𝑥) = ∫ 𝑑𝑡, (42)
Γ (𝑛 − 𝛼) 0 (𝑥 − 𝑡)𝛼+1−𝑛 (36) 󸀠
𝑥 𝑓
g 𝛼 1 (𝑡; 𝑟) 𝑑𝑡
𝑛 − 1 < 𝛼 ≤ 𝑛, 𝑛 ∈ N, ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) =[ ∫ ].
Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼
where 𝛼 > 0 is the order of the derivative and 𝑛 is the smallest
Remark 28. A fuzzy-valued function 𝑓 is 𝐶[1 − 𝛼]-
integer greater than 𝛼. For the Caputo derivative, we have
differentiable, if it is differentiable as in Definition 27, Case
𝑐
𝐷𝛼 𝐶 = 0, (𝐶 is a constant) , (37) (i), and it is 𝐶[2 − 𝛼]-differentiable, if it is differentiable as in
Definition 27, case (ii).
{ 0, for 𝛽 ∈ N0 , 𝛽 < ⌈𝛼⌉ ,
{
{
{ Theorem 29. Let 0 < 𝛼 < 1 and 𝑓 ∈ 𝐴𝐶E [𝑎, 𝑏], then Caputo
{
{ fuzzy fractional derivative exists almost everywhere on (𝑎, 𝑏)
𝑐 𝛼 𝛽
𝐷 𝑥 = { Γ (𝛽 + 1) 𝛽−𝛼 and for all 0 ≤ 𝑟 ≤ 1 we have
{
{ 𝑥 , for 𝛽 ∈ N0 , 𝛽 ≥ ⌈𝛼⌉
{
{ Γ (𝛽 + 1 − 𝛼)
{
{ or 𝛽 ∉ N, 𝛽 > ⌊𝛼⌋ . (𝐶𝐷𝑎𝛼+ 𝑓) (𝑥; 𝑟)
(38) 󸀠 󸀠
𝑥 𝑓 (𝑡) 𝑥 𝑓
1 1 (𝑡)
The ceiling function ⌈𝛼⌉ is used to denote the smallest integer =[ ∫ 𝛼 𝑑𝑡, ∫ 𝑑𝑡]
Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡) Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼
greater than or equal to 𝛼, and the floor function ⌊𝛼⌋ to denote
the largest integer less than or equal to 𝛼. Also N = {1, 2, . . .} (43)
and N0 = {0, 1, 2, . . .}.
when 𝑓 is (1)-differentiable, and
Definition 25 (see [1]). Similar to the differential equation of
integer order, the Caputo’s fractional differentiation is a linear (𝐶𝐷𝑎𝛼+ 𝑓) (𝑥; 𝑟)
operation, that is, 󸀠 󸀠
𝑥 𝑓 𝑥 𝑓 (𝑡)
1 (𝑡) 1
𝑐 𝛼 𝑐 𝛼
𝐷 (𝜆𝑓 (𝑥) + 𝜇𝑔 (𝑥)) = 𝜆 𝐷 𝑓 (𝑥) + 𝜇 𝐷 𝑔 (𝑥) , 𝑐 𝛼 =[ ∫ 𝑑𝑡, ∫ 𝑑𝑡]
(39) Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼 Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼
where 𝜆 and 𝜇 are constants. (44)
Abstract and Applied Analysis 7

when 𝑓 is (2)-differentiable, in which (𝐼𝑎𝛼+ 𝑓)(𝑥) = Proof. See [32].


𝑥
(1/Γ(𝛼)) ∫𝑎 (𝑓(𝑡)𝑑𝑡/(𝑥 − 𝑡)1−𝛼 ) for 𝑥 > 𝑎.
Now, characterization theorem (Theorem 15), which was
Proof. It is straightforward by applying Definitions 20 and 24. introduced by Bede in [15] and established by Pederson and
Sambandham in [59] for hybrid fuzzy differential equations,
is extended for fuzzy Caputo-type fractional differential
Theorem 30. Let one assume that 𝑓 ∈ 𝐶E [𝑎, 𝑏], and then one
equations. To this end, we first consider the FFDEs under
has the following:
Caputo’s H-differentiability for 0 < 𝛼 < 1 as follows:
(𝐼𝑎𝛼+ 𝐶𝐷𝑎𝛼+ 𝑓) (𝑥) = 𝑓 (𝑥) ⊖ 𝑓 (𝑎) , 0 < 𝛼 < 1, (45)
(𝑐 𝐷𝑎𝛼+ 𝑦) (𝑥) = 𝑓 (𝑥, 𝑦 (𝑥)) , 𝑦 (𝑎) ∈ E. (50)
𝐶
when 𝑓 is [1 − 𝛼]-differentiable and
Theorem 33 (characterization theorem). Let one consider
(𝐼𝑎𝛼+ 𝐶𝐷𝑎𝛼+ 𝑓) (𝑥) = −𝑓 (𝑎) ⊖ (−𝑓 (𝑥)) , 0 < 𝛼 < 1, (46) the fuzzy fractional differential equation under Caputo’s H-
differentiability (50) where 𝑓 : [𝑥0 , 𝑥0 + 𝑎] × E → E and
𝐶 such that:
when 𝑓 is [2 − 𝛼]-differentiable.
𝑟
Proof. See [30]. (i) [𝑓(𝑥, 𝑦)]𝑟 = [𝑓𝑟 (𝑥, 𝑦, 𝑦), 𝑓 (𝑥, 𝑦, 𝑦)],
𝑟
Lemma 31. Let 0 < 𝛼 < 1 and 𝑓 ∈ 𝐴𝐶E [𝑎, 𝑏], then the (ii) 𝑓𝑟 and 𝑓 are equicontinuous (i.e., for any 𝜖 > 0 there
fuzzy Caputo derivative can be expressed by means of the fuzzy is a 𝛿 > 0 such that |𝑓𝑟 (𝑥, 𝑦, 𝑧) − 𝑓𝑟 (𝑥1 , 𝑦1 , 𝑧1 )| < 𝜖
Riemann-Liouville integral as follows: 𝑟 𝑟
and |𝑓 (𝑥, 𝑦, 𝑧) − 𝑓 (𝑥1 , 𝑦1 , 𝑧1 )| < 𝜖 for all 𝑟 ∈ [0, 1],
(𝐶𝐷𝑎𝛼+ 𝑓) (𝑥; 𝑟) whenever (𝑥, 𝑦, 𝑧), (𝑥1 , 𝑦1 , 𝑧1 ) ∈ [𝑥0 , 𝑥0 + 𝑎] × R2 and
‖(𝑥, 𝑦, 𝑧) − (𝑥1 , 𝑦1 , 𝑧1 )‖ < 𝛿 and uniformly bounded on
= (𝐼𝑎1−𝛼
+ 𝐷𝑓) (𝑥; 𝑟) (47) any bounded set,
(iii) there exists an 𝐿 > 0 such that |𝑓𝑟 (𝑥2 , 𝑦2 , 𝑧2 ) −
= [(𝐼𝑎1−𝛼 1−𝛼
+ 𝐷𝑓) (𝑥; 𝑟) , (𝐼𝑎+ 𝐷𝑓) (𝑥; 𝑟)] ,
𝑓𝑟 (𝑥1 , 𝑦1 , 𝑧1 )| ≤ 𝐿 max{|𝑦2 − 𝑦1 |, |𝑧2 − 𝑧1 |} for all 𝑟 ∈
𝑟 𝑟
when 𝑓 is (1)-differentiable, and [0, 1], |𝑓 (𝑥2 , 𝑦2 , 𝑧2 ) − 𝑓 (𝑥1 , 𝑦1 , 𝑧1 )| ≤ 𝐿 max{|𝑦2 −
𝑦1 |, |𝑧2 − 𝑧1 |} for all 𝑟 ∈ [0, 1].
(𝐶𝐷𝑎𝛼+ 𝑓) (𝑥; 𝑟)
Then, (50) and the following system of FDEs are equivalent
= (𝐼𝑎1−𝛼
+ 𝐷𝑓) (𝑥; 𝑟) (48) when 𝑦(𝑥) is 𝐶[1 − 𝛼]-differentiable

= [(𝐼𝑎1−𝛼 1−𝛼
+ 𝐷𝑓) (𝑥; 𝑟) , (𝐼𝑎+ 𝐷𝑓) (𝑥; 𝑟)] , (𝑐 𝐷𝑥𝛼+ 𝑦) (𝑥; 𝑟) = 𝑓𝑟 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) ,
0

when 𝑓 is (2)-differentiable. 𝑟
(𝑐 𝐷𝑥𝛼+ 𝑦) (𝑥; 𝑟) = 𝑓 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) ,
0

Now we consider the generalization of Taylor’s formula (51)


for the fuzzy Caputo fractional derivative which was intro- 𝑦𝑟 (𝑥0 ) = (𝑐 𝑦0𝑟 ) ,
duced in [32, 45]. It should be mentioned that this theorem
is the extension of Taylor’s formula for the Caputo fractional 𝑦 (𝑥0 ) = (𝑐 𝑦0𝑟 ) ,
derivative in the crisp context [58].
also (50) and the following system of FDEs is equivalent when
̃
Theorem 32. Let 𝑓(𝑥) ∈ 𝐴𝐶E (0, 𝑏] and suppose that 𝑦(𝑥) is 𝐶[2 − 𝛼]-differentiable
𝑐 𝑘𝛼 ̃
𝐷 𝑓(𝑥) ∈ 𝐶E (0, 𝑏] for 𝑘 = 0, 1, . . . , 𝑛 + 1 where 0 < 𝛼 < 1,
𝑟
0 ≤ 𝑥0 ≤ 𝑥 and 𝑥 ∈ (0, 𝑏]. Then we have (𝑐 𝐷𝑥𝛼+ 𝑦) (𝑥; 𝑟) = 𝑓 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) ,
0
𝑟 𝑟
[𝑓̃ (𝑥)] = [𝑓𝑟 (𝑥) , 𝑓 (𝑥)] , (𝑐 𝐷𝑥𝛼+ 𝑦) (𝑥; 𝑟) = 𝑓𝑟 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) ,
0

𝑛 𝑖𝛼 𝑐 (𝑛+1)𝛼 𝛼 (52)
𝑟 𝑥 𝑐 𝑖𝛼 𝑟 + 𝐷 𝑓 (𝑥0 ) (𝑛+1)𝛼 𝑦𝑟 (𝑥0 ) = (𝑐 𝑦0𝑟 ) ,
𝑓 (𝑥) = ∑ 𝐷 𝑓 (0 ) + 𝑥 ,
𝑖=0 Γ (𝑖𝛼 + 1) Γ (𝑛𝛼 + 𝛼 + 1)
𝑐 (𝑛+1)𝛼 𝛼
𝑦 (𝑥0 ) = (𝑐 𝑦0𝑟 ) .
𝑛 𝑖𝛼
𝑟 𝑥 𝑐 𝑖𝛼 𝑟 + 𝐷 𝑓 (𝑥0 ) (𝑛+1)𝛼
𝑓 (𝑥) = ∑ 𝐷 𝑓 (0 ) + 𝑥 ,
𝑖=0 Γ (𝑖𝛼 + 1) Γ (𝑛𝛼 + 𝛼 + 1) Proof. In the papers [15, 59], the authors proved for fuzzy
(49) ordinary differential equations and hybrid fuzzy differential
equations. The result for FFDEs is obtained analogously by
𝑟 𝑟
where 𝑐 𝐷𝛼 𝑓𝑟 (0)=𝑐 𝐷𝛼 𝑓𝑟 (𝑥)| 𝑥=0 , 𝑐 𝐷𝛼 𝑓 (0)=𝑐 𝐷𝛼 𝑓 (𝑥)| 𝑥=0 . using Theorem 2 in [15] and Theorem 3.1 in [59].
8 Abstract and Applied Analysis

4. Proposed Method for Solving FFDEs 1


0.8
Saadatmandi and Dehghan [36] introduced the shifted
Legendre operational matrix for derivative with fractional 0.6
order using a spectral method which has been followed by 0.4
Doha et al. [37–39]. They presented the shifted Chebyshev 0.2
polynomials and Jacobi polynomials for solving fractional

L m (x)
differential equations by tau method. In this section, we 0
try to approximate fuzzy solution using shifted Legendre −0.2
polynomials under H-differentiability as follows. −0.4
−0.6
4.1. Properties of Shifted Legendre Polynomials. The Legendre
polynomials, denoted by 𝑃𝑛 (𝑧), are orthogonal with Legendre −0.8
weight function: 𝑤(𝑧) = 1 over [−1, 1], namely [60], −1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1 x
2
∫ 𝑃𝑛 (𝑧) 𝑃𝑚 (𝑧) 𝑑𝑧 = 𝛿 , (53)
−1 2𝑛 + 1 𝑛𝑚 m=3 m=6
m=4 m=7
where 𝛿𝑛𝑚 is the Kronecker function and can be specified with m=5 m=8
the help of following recurrence formula:
Figure 1: The shifted Legendre functions for different 𝑚.
𝑃0 (𝑧) = 1, 𝑃1 (𝑧) = 𝑧,
2𝑖 + 1 𝑖 (54)
𝑃𝑖+1 (𝑧) = 𝑧𝑃𝑖 (𝑧) − 𝑃 (𝑧) , 𝑖 = 1, 2, . . . . A function 𝑢(𝑥) of independent variable defined for 0 ≤ 𝑥 ≤ 1
𝑖+1 𝑖 + 1 𝑖−1 may be expanded in terms of shifted Legendre polynomials as
In order to use these polynomials on the interval [0, 1], 𝑚
we define the so-called shifted Legendre polynomials by 𝑢 (𝑥) = ∑𝑏𝑖 𝐿 𝑖 (𝑥) = Λ𝑇 Φ (𝑥) , (60)
introducing the change of variable 𝑧 = 2𝑥 − 1. Let the shifted 𝑖=0
Legendre polynomials 𝑃𝑛 (2𝑥 − 1) be denoted by 𝐿 𝑛 (𝑥). The where the shifted Legendre coefficients matrix 𝐵 and the
shifted Legendre polynomials are orthogonal with respect to shifted Legendre vector Φ(𝑥) are given by
the weight function 𝑤𝑠 (𝑥) = 1 in the interval (0, 1) with the
following orthogonality property: Λ𝑇 = [𝜆 1 , 𝜆 2 , . . . , 𝜆 𝑚 ] ,
1
(61)
1 Φ (𝑥) = [𝐿 0 (𝑥) , 𝐿 1 (𝑥) , . . . , 𝐿 𝑚 (𝑥)] .
𝑇
∫ 𝐿 𝑛 (𝑥) 𝐿 𝑚 (𝑥) 𝑑𝑥 = 𝛿 . (55)
0 2𝑛 + 1 𝑛𝑚
Also, the derivative of Φ(𝑥) can be expressed by
The shifted Legendre polynomials are generated from the
following three-term recurrence relation: 𝑑Φ (𝑥)
= 𝐷Φ (𝑥) , (62)
𝑑𝑥
(2𝑖 + 1) (2𝑥 − 1) 𝑖
𝐿 𝑖+1 (𝑡) = 𝐿 𝑖 (𝑥) − 𝐿 (𝑥) , where 𝐷 is the (𝑚+1)×(𝑚+1) operational matrix of derivative
𝑖+1 𝑖 + 1 𝑖−1
given by
𝑖 = 1, 2, . . . (56)
{ 2 (2𝑗 + 1) , for 𝑗 = 𝑖 − 𝑘,
𝐿 0 (𝑥) = 1, 𝐿 1 (𝑥) = 2𝑥 − 1. {
{
{ 𝑘 = 1, 3, . . . , 𝑚, if 𝑚 odd,
𝐷 = (𝑑𝑖𝑗 ) = { {
{
{ 𝑘 = 1, 3, . . . , 𝑚 − 1, if 𝑚 even,
The analytic form of the shifted Legendre polynomial 𝐿 𝑛 (𝑥) {
of degree 𝑛 is given by {0, otherwise.
(63)
𝑛 𝑛
(𝑛 + 𝑖)! 𝑥𝑖
𝐿 𝑛 (𝑥) = ∑(−1)𝑛+𝑖 = ∑ 𝑒 𝑥𝑖 , (57) The graph of some shifted Legendre polynomials (for 3 ≤ 𝑚 ≤
(𝑛 − 𝑖)! (𝑖!)2 𝑖=0 𝑖,𝑛
𝑖=0 8) shown in Figure 1 to depict their behaviors.
Now, we use the shifted Legender functions due to
in which
approximate a fuzzy function.
(𝑛 + 𝑖)!
𝑒𝑖,𝑛 = (−1)𝑛+𝑖 , (58)
(𝑛 − 𝑖)!(𝑖!)2 4.2. The Approximation of Fuzzy Function. In this section,
we propose a shifted Legendre approximation for the fuzzy-
where valued functions. To this end, we use Legendre’s nodes and
fuzzy shifted Legendre polynomials to calculate the fuzzy best
𝐿 𝑛 (0) = (−1)𝑛 , 𝐿 𝑛 (1) = 1. (59) approximation. For more details, see [61–67].
Abstract and Applied Analysis 9

Definition 34. For 𝑦 ∈ 𝐿E𝑝 (0, 1) ∩ 𝐶E (0, 1) and Legendre and also assume that 𝑦(𝑥) is 𝑐 [1 − 𝛼]-differentiable. Therefore
polynomial 𝐿 𝑛 (𝑥) a real valued function over (0, 1), the fuzzy using Theorem 32, we have
function is approximated by 󵄨󵄨 𝑚 𝑖𝛼 󵄨󵄨
󵄨󵄨 𝑟
󵄨󵄨𝑦 (𝑥) − ∑ 𝑥 𝑖𝛼 𝑟 + 󵄨󵄨󵄨 𝑥(𝑚+1)𝛼
󵄨󵄨 𝐷 𝑦 (0 ) 󵄨
󵄨󵄨 ≤ 𝑀𝛼 ,
𝑖=0 Γ (𝑖𝛼 + 1) Γ (𝑚𝛼 + 𝛼 + 1)
∞∗ 󵄨󵄨 󵄨󵄨
𝑦 (𝑥) = ∑ 𝑐𝑗 ⊙ 𝐿 𝑗 (𝑥) , 𝑥 ∈ (0, 1) , (64) 󵄨󵄨 𝑚 𝑖𝛼 󵄨󵄨
𝑗=0 󵄨󵄨 𝑟
󵄨󵄨𝑦 (𝑥) − ∑ 𝑥 𝑖𝛼 𝑟 + 󵄨󵄨󵄨 𝑥(𝑚+1)𝛼
󵄨󵄨 𝐷 𝑦 (0 ) 󵄨
󵄨󵄨 ≤ 𝑀𝛼 ,
where the fuzzy coefficients 𝑐𝑗 are obtained by 󵄨󵄨 𝑖=0 Γ (𝑖𝛼 + 1) 󵄨󵄨 Γ (𝑚𝛼 + 𝛼 + 1)
(71)
1
𝑐𝑗 = (2𝑗 + 1) ⊙ ∫ 𝑦 (𝑥) ⊙ 𝐿 𝑗 (𝑥) 𝑑𝑥, (65) in which |𝐷(𝑚+1)𝛼 𝑦𝑟 (𝑥0 )| < 𝑀𝛼 and |𝐷(𝑚+1)𝛼 𝑦𝑟 (𝑥0 )| < 𝑀𝛼 .
0
Proof. From Theorem 32, we have
in which 𝐿 𝑗 (𝑥) is the same as in (57), and ∑∗ means addition
𝑐 (𝑚+1)𝛼
with respect to ⊕ in E. 𝑚
𝑥𝑖𝛼 𝑐 𝑖𝛼 𝑟
𝐷 𝑦𝑟 (𝑥0 ) (𝑚+1)𝛼
𝑦𝑟 (𝑥) = ∑ +
𝐷 𝑦 (0 ) + 𝑥 ,
Remark 35. In actuality, only the first (𝑚 + 1)-terms shifted 𝑖=0 Γ (𝑖𝛼 + 1) Γ (𝑚𝛼 + 𝛼 + 1)
Legendre polynomials are considered. So we have 𝑐 (𝑚+1)𝛼
𝑚
𝑥𝑖𝛼 𝑐 𝑖𝛼 𝑟 𝐷 𝑦𝑟 (𝑥0 ) (𝑚+1)𝛼
𝑦𝑟 (𝑥) = ∑ +
𝐷 𝑦 (0 ) + 𝑥 ,
𝑚∗
𝑇 𝑖=0 Γ (𝑖𝛼 + 1) Γ (𝑚𝛼 + 𝛼 + 1)
𝑦 (𝑥) ≃ 𝑦̃𝑚 (𝑥) = ∑ 𝑐𝑗 ⊙ 𝐿 𝑗 (𝑥) = 𝐶𝑚 ⊙ Φ𝑚 (𝑥) , (66) (72)
𝑗=0
and the following relation can be obtained:
hence 󵄨󵄨 𝑚 𝑖𝛼 󵄨󵄨
󵄨󵄨 𝑟 + 󵄨󵄨󵄨
󵄨󵄨𝑦 (𝑥) − ∑ 𝑥 𝑖𝛼 𝑟
𝐷 𝑦 (0 )󵄨󵄨
𝑚∗ 󵄨󵄨 󵄨󵄨
𝑦𝑟 (𝑥) ≃ 𝑦̃𝑚
𝑟
(𝑥) = ∑ 𝑐𝑗𝑟 ⊙ 𝐿 𝑗 (𝑥) ,
󵄨󵄨 𝑖=0 Γ (𝑖𝛼 + 1) 󵄨
(67)
𝑗=0 󵄨󵄨𝑐 (𝑚+1)𝛼 𝑟 󵄨󵄨
󵄨󵄨 𝐷 𝑦 (𝑥0 )󵄨󵄨󵄨 (𝑚+1)𝛼 𝑥(𝑚+1)𝛼
𝑇
≤󵄨 𝑥 ≤ 𝑀𝑟𝛼 ,
that the fuzzy shifted Legendre coefficient vector 𝐶𝑚+1 and Γ (𝑚𝛼 + 𝛼 + 1) Γ (𝑚𝛼 + 𝛼 + 1)
shifted Legendre function vector Φ𝑚+1 (𝑥) are defined as 󵄨󵄨󵄨 𝑚 󵄨󵄨󵄨
󵄨󵄨 𝑟 𝑥𝑖𝛼 󵄨
󵄨󵄨𝑦 (𝑥) − ∑ 𝐷𝑖𝛼 𝑦𝑟 (0+ )󵄨󵄨󵄨
𝑇
𝐶𝑚 = [𝑐0 , 𝑐2 , . . . , 𝑐𝑚 ] , (68) 󵄨󵄨 Γ (𝑖𝛼 + 1) 󵄨󵄨
󵄨 𝑖=0 󵄨
󵄨󵄨𝑐 (𝑚+1)𝛼 𝑟 󵄨󵄨
Φ𝑚 (𝑥) = [𝐿 0 (𝑥) , 𝐿 1 (𝑥) , . . . , 𝐿 𝑚 (𝑥)] . (69) 󵄨󵄨 𝐷 𝑦 (𝑥0 )󵄨󵄨󵄨 (𝑚+1)𝛼 𝑥(𝑚+1)𝛼
≤󵄨
𝑟
𝑥 ≤ 𝑀𝛼 ,
Γ (𝑚𝛼 + 𝛼 + 1) Γ (𝑚𝛼 + 𝛼 + 1)
Definition 36 (see [68]). A fuzzy-valued polynomial 𝑝̃∗ ∈ (73)
̃ is the best approximation to fuzzy function 𝑓 on 𝜒 =
∏ 𝑁 𝑟
{𝑥0 , 𝑥1 , 𝑥2 , . . . , 𝑥𝑁}, if that 𝑀𝑟𝛼 = |𝑐 𝐷(𝑚+1)𝛼 𝑦𝑟 (𝑥0 )| and 𝑀𝛼 = |𝑐 𝐷(𝑚+1)𝛼 𝑦𝑟 (𝑥0 )|.

max 𝐷 (𝑝̃∗ (𝑥𝑖 ) , 𝑓𝑖 ) = min { max 𝐷 (𝑝̃ (𝑥𝑖 ) , 𝑓𝑖 )} , Remark 39. If we consider Lemma 38 and define 𝑀𝛼 =
𝑟
𝑖=0,1,2,...,𝑁 ̃ ∏
𝑝∈ ̃
𝑁
𝑖=0,1,2,...,𝑁 max{𝑀𝑟𝛼 , 𝑀𝛼 }, then (71) can be stated in the following form,
(70) regarding to Section 2:
𝑚
̃ is the set of all fuzzy valued polynomials. 𝑥𝑖𝛼
in which ∏ 𝑁 𝐷 (𝑦 (𝑥) , ∑ 𝐷𝑖𝛼 𝑦 (0+ ))
𝑖=0 Γ (𝑖𝛼 + 1)
The problem is referred to as the best shifted Legendre 󵄨󵄨 󵄨󵄨
𝑚
approximation, as we use Legendre’s nodes. 󵄨󵄨 𝑥𝑖𝛼 󵄨󵄨
= sup max {󵄨󵄨󵄨𝑦𝑟 (𝑥) − ∑ 𝐷𝑖𝛼 𝑦𝑟 (0+ )󵄨󵄨󵄨 ,
󵄨
󵄨󵄨 𝑖=0 Γ (𝑖𝛼 + 1)
󵄨󵄨
Theorem 37. The best approximation of a fuzzy function
𝑟∈[0,1] 󵄨
󵄨󵄨 𝑚 𝑖𝛼 󵄨󵄨
based on the Legendre nodes exists and is unique. 󵄨󵄨 𝑟 + 󵄨󵄨󵄨
󵄨󵄨𝑦 (𝑥) − ∑ 𝑥 𝐷 𝑖𝛼 𝑟
𝑦 (0 ) 󵄨󵄨}
󵄨󵄨 󵄨󵄨
Proof. The proof is an instantaneous outcome of Theo- 󵄨󵄨 𝑖=0 Γ (𝑖𝛼 + 1) 󵄨
rem 4.2.1 in [68].
𝑥(𝑛+1)𝛼 𝑟 𝑥(𝑛+1)𝛼
≤ sup max {𝑀𝑟𝛼 , 𝑀𝛼 }
Now, we want to show that the fuzzy approximation 𝑟∈[0,1] Γ (𝑛𝛼 + 𝛼 + 1) Γ (𝑛𝛼 + 𝛼 + 1)
converges of Legendre functions to function 𝑦(𝑥).
𝑥(𝑛+1)𝛼
E ≤ 𝑀𝛼 .
Lemma 38. Suppose that 𝑦(𝑥) ∈ 𝐴𝐶 (0, 1] ∩ and 𝐿E𝑝 (0, 1] Γ (𝑛𝛼 + 𝛼 + 1)
𝑐 𝛼 E (74)
𝐷 𝑦(𝑥) ∈ 𝐶 (0, 1], 0 < 𝛼 < 1, 0 ≤ 𝑥0 ≤ 𝑥, and 𝑥 ∈ (0, 1],
10 Abstract and Applied Analysis

Theorem 40. Let 𝑐 𝐷𝛼 𝑦(𝑥) ∈ 𝐶E (0, 1] ∩ 𝐿E𝑝 (0, 1], and 0 < From Remark 39 and Lemma 38, we have
𝛼 < 1. Also consider a sequence of finite dimensional fuzzy
󵄩󵄩 𝑟 󵄩 𝑀𝛼 1
space 𝑋E ⊂ 𝑋, 𝑚 ≥ 1, in which 𝑋E have dimension 𝑑𝑚+1 . 󵄩󵄩𝑦 (𝑥) − 𝑦̃𝑟 (𝑥)󵄩󵄩󵄩 ≤
Additionally, 𝑋E have a basis {𝐿 𝑖 (𝑥)}𝑚 󵄩󵄩 𝑚 󵄩󵄩 Γ (𝑚𝛼 + 𝛼 + 1) √ (2 (𝑚 + 1) 𝛼 + 1) ,
𝑖=0 . If one assumes that
𝑦̃𝑚 (𝑥) = 𝐶𝑇 Φ is the best fuzzy approximation for fuzzy
function 𝑦(𝑥) from {𝐿 𝑖 (𝑥)}𝑚 󵄩󵄩 𝑟 󵄩 𝑀𝛼 1
𝑖=0 , then the error estimation is as 󵄩󵄩𝑦 (𝑥) − 𝑦̃𝑟 (𝑥)󵄩󵄩󵄩 ≤
follows: 󵄩󵄩 𝑚 󵄩󵄩 Γ (𝑚𝛼 + 𝛼 + 1) √ (2 (𝑚 + 1) 𝛼 + 1) ,
(79)
lim 𝐷 (𝑦 (𝑥) , 𝑦̃𝑚 (𝑥)) = 0. (75)
𝑚→∞
and if 𝑚 → ∞, we get ‖𝑦𝑟 (𝑥) − 𝑦̃ 𝑟 (𝑥)‖ → 0, ‖𝑦𝑟 (𝑥) −
𝑚
𝑟
Proof. Let 𝑓 : 𝑅 → E be a fuzzy valued function such 𝑦̃𝑚 (𝑥)‖ → 0. Therefore, from Remark 39 and the definition
𝑟
that [𝑓(𝑥)]𝑟 = [𝑓𝑟 (𝑥), 𝑓 (𝑥)]. Also consider the fuzzy of Hausdorff distance in Section 2, it can be implied that
Taylor’s formula in Theorem 32, 𝑓𝑟 (𝑥) = ∑𝑚 𝑖𝛼
𝑖=0 (𝑥 /Γ(𝑖𝛼 + lim 𝐷 (𝑦 (𝑥) , 𝑦̃𝑚 (𝑥)) = 0,
𝑟 𝑚→∞ (80)
𝑖𝛼 𝑖𝛼 𝑟
1)) 𝑐 𝐷 𝑓𝑟 (0+ ) and 𝑓 (𝑥) = ∑𝑚 𝑖𝛼 𝑐 +
𝑖=0 (𝑥 /Γ(𝑖𝛼 + 1)) 𝐷 𝑓 (0 ).
From Lemma 38 we have which completes the proof.

󵄨󵄨 𝑟 󵄨 𝑥(𝑚+1)𝛼 Remark 41. The same result can be obtained for 𝑦(𝑥) under
󵄨󵄨𝑦 (𝑥) − 𝑓𝑟 (𝑥)󵄨󵄨󵄨 ≤ 𝑀𝑟𝛼 , 𝑐
[2 − 𝛼]-differentiability.
󵄨 󵄨 Γ (𝑚𝛼 + 𝛼 + 1)
(76)
󵄨󵄨 𝑟 󵄨
󵄨󵄨𝑦 (𝑥) − 𝑓𝑟 (𝑥)󵄨󵄨󵄨 ≤ 𝑀𝑟 𝑥(𝑚+1)𝛼 4.3. Operational Matrix of Caputo Fractional Derivative
󵄨󵄨 󵄨󵄨 𝛼 .
Γ (𝑚𝛼 + 𝛼 + 1)
Lemma 42. The fuzzy Caputo fractional derivative of order
𝑇 0 < 𝛼 < 1 over the shifted Legendre functions can be gained
From the assumption, 𝐶 Φ is the best fuzzy approximation
in the form of
to 𝑦 from {𝐿 𝑖 (𝑥)}𝑚
𝑖=0 , and 𝑓 ∈ 𝑋E , 𝑥 ∈ (0, 1]. So one has
𝑖
𝑐 Γ (𝑘 + 1) 𝑘−𝛼
󵄨󵄨 𝑟 󵄨 𝑥(𝑚+1)𝛼 𝐷𝛼 𝐿 𝑖 (𝑥) = ∑ 𝑒𝑘,𝑖
󸀠
𝑥 ,
󵄨󵄨𝑦 (𝑥) − 𝑦̃𝑟 (𝑥)󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨󵄨𝑦𝑟 (𝑥) − 𝑓𝑟 (𝑥)󵄨󵄨󵄨󵄨 ≤ 𝑀𝑟
(81)
󵄨󵄨 󵄨󵄨 󵄨 , Γ (𝑘 − 𝛼 + 1)
𝑚 󵄨 𝛼
Γ (𝑚𝛼 + 𝛼 + 1) 𝑘=0

󸀠
where 𝑒𝑘,𝑖 = 0 when 𝑖 < ⌈𝛼⌉ and for 𝑖 ≥ ⌈𝛼⌉, and so one has
󵄨󵄨 𝑟 󵄨 󵄨 󵄨
󵄨󵄨𝑦 (𝑥) − 𝑦̃𝑟 (𝑥)󵄨󵄨󵄨 ≤ 󵄨󵄨󵄨𝑦𝑟 (𝑥) − 𝑓𝑟 (𝑥)󵄨󵄨󵄨 ≤ 𝑀𝑟 𝑥(𝑚+1)𝛼 󸀠
󵄨󵄨 𝑚 󵄨󵄨 󵄨󵄨 󵄨󵄨 𝛼 , 𝑒𝑘,𝑖 = 𝑒𝑘,𝑖 .
Γ (𝑚𝛼 + 𝛼 + 1)
(77) Proof. Employing the analytic form of shifted Legendre
polynomials explained in Section 4.1 and Definition 25, we
and thus, taking into account Theorem 1 in [69] and above have:
relations, we have
𝑖
󵄩󵄩 𝑟 󵄩2 𝐷𝛼 𝐿 𝑖 (𝑥) = ∑ 𝑒𝑘,𝑖 𝐷𝛼 ⊙ 𝑥𝑘−𝛼 . (82)
󵄩󵄩𝑦 (𝑥) − 𝑦̃𝑟 (𝑥)󵄩󵄩󵄩
󵄩󵄩 𝑚 󵄩󵄩 𝑘=0

󵄩 󵄩2 Now, by exploiting Definition 24, the lemma can be proved.


≤ 󵄩󵄩󵄩󵄩𝑦𝑟 (𝑥) − 𝑓𝑟 (𝑥)󵄩󵄩󵄩󵄩

𝑀𝑟,2 1
Lemma 43. Let 0 < 𝛼 < 1, and the integral of the product
≤ 𝛼
2
∫ 𝑥2(𝑚+1)𝛼 𝑑𝑥
Γ(𝑚𝛼 + 𝛼 + 1) 0 of the fuzzy Caputo fractional derivative with order 𝛼 over the
shifted legendre functions can be obtained by
𝑀𝑟,2
𝛼
≤ , 𝑗 󸀠
Γ(𝑚𝛼 + 𝛼 + 1)2 (2 (𝑚 + 1) 𝛼 + 1)2 1 𝑖 𝑒𝑘,𝑖 𝑒𝑙,𝑗
Γ (𝑘 + 1)
∫ 𝐷𝛼 𝐿 𝑖 (𝑥) 𝐿 𝑗 (𝑥) 𝑑𝑥 = ∑ ∑ .
󵄩󵄩 𝑟 󵄩2 (78) 0 (𝑘 + 𝑙 + 1) − 𝛼 Γ (𝑘 − 𝛼 + 1)
󵄩󵄩𝑦 (𝑥) − 𝑦̃𝑟 (𝑥)󵄩󵄩󵄩 𝑘=0 𝑙=0
󵄩󵄩 𝑚 󵄩󵄩 (83)
󵄩󵄩 𝑟 󵄩󵄩2 Proof. Using Lemma 42 and the analytic form of shifted Leg-
≤ 󵄩󵄩󵄩𝑦𝑟 (𝑥) − 𝑓 (𝑥)󵄩󵄩󵄩
󵄩 󵄩 endre polynomials explained in Section 4.1, we can acquire:
𝑟,2
𝑀𝛼 1 1
≤ ∫ 𝑥2(𝑚+1)𝛼 𝑑𝑥 ∫ 𝐷𝛼 𝐿 𝑖 (𝑥) 𝐿 𝑗 (𝑥) 𝑑𝑥
Γ(𝑚𝛼 + 𝛼 + 1)2 0 0

𝑟,2 𝑗 󸀠
(84)
𝑀𝛼
𝑖 𝑒𝑘,𝑖 𝑒𝑙,𝑗 Γ (𝑘 + 1) 1
(𝑘+𝑙+1)−𝛼−1
≤ . = ∑∑ ∫ 𝑥 𝑑𝑥.
Γ(𝑚𝛼 + 𝛼 + 1)2 (2 (𝑚 + 1) 𝛼 + 1)2 𝑘=0 𝑙=0
Γ (𝑘 − 𝛼 + 1) 0
Abstract and Applied Analysis 11

The operational matrix of different orthogonal functions where ∼ Λ is an (𝑚 + 1)-square product operational matrix
for solving differential equations was introduced in the crisp for the fuzzy vector Λ ̃ = [Λ, ̃ Using the above equation
̃ Λ].
concept [36, 37, 39]. Here, the Legendre operational matrix and by the orthogonal property equation (56), the elements
(LOM) in [36] is applied to the FFDEs using Caputo-type { ∼ Λ𝑖𝑗 }𝑚
𝑖,𝑗=0
can be computed from
derivative.
The Caputo fractional derivatives operator of order 0 <
𝑚 ∗
𝛼 < 1 of the vector Φ defined in (69) can be stated by ∼
Λ 𝑖𝑗 = (2𝑗 + 1) ∑ 𝜆 𝑘 ⊙ 𝑔𝑖𝑗𝑘 , (91)
𝐷𝛼 Φ (𝑥) ≃ 𝐷(𝛼) Φ (𝑥) , (85) 𝑘=0

where 𝐷(𝛼) is the (𝑚 + 1)-square operational matrix of where ∑∗ denotes the fuzzy summation and ⊙ indicates fuzzy
fractional Caputo’s derivative of Legendre functions. Regard- multiplication. Moreover, 𝑔𝑖𝑗𝑘 are obtained by
(𝛼)
ing the following theorem, the LOM elements 𝐷𝑖,𝑗 are
determined under Caputo fractional derivative. This theorem 1

is generalizing the operational matrix of derivatives of shifted 𝑔𝑖𝑗𝑘 = ∫ 𝐿 𝑖 (𝑥) 𝐿 𝑗 (𝑥) 𝐿 𝑘 (𝑥) 𝑑𝑥, (92)
0
Legendre given in Section 4.1 to the fractional calculus.
(𝛼)
that in the simpler form, it is given by
Theorem 44. Let Φ be Legendre functions vector. 𝐷𝑖,𝑗 is
the (𝑚 + 1)-square operational matrix of fractional Caputo’s 𝑑𝑗−𝑙 𝑑𝑙 𝑑𝑖−𝑙
(𝛼)
derivative of order 0 < 𝛼 < 1. Then the elements of 𝐷𝑖,𝑗 are {
{ , 𝑘 = 𝑖 + 𝑗 − 2𝑙;
{
{ (2𝑖 + 2𝑗 − 2𝑙 + 1) 𝑑𝑖+𝑗−𝑙
achieved as {
{
𝑔𝑖𝑗𝑘 ={ 𝑙 = 0, 1, . . . , 𝑗, (93)
𝑖 {
{
{
(𝛼)
𝐷𝑖,𝑗 = ∑ 𝜃𝑖,𝑗,𝑘 , (86) {0,
{ 𝑘 ≠𝑖 + 𝑗 − 2𝑙;
𝑘=⌈𝛼⌉ { 𝑙 = 0, 1, . . . , 𝑗,

in which 𝜃𝑖,𝑗,𝑘 are acquired by


in which 𝑑𝑙 = (2𝑙)!/2𝑙 (𝑙!)2 .
𝜃𝑖,𝑗,𝑘
Remark 46. As a matter of fact, from (93), it is obvious that
= (2𝑗 + 1) the elements of the product of shifted Legendre operational
matrix are independent from the fuzzy vector.
𝑗
(−1)𝑖+𝑗+𝑘+𝑙 (𝑖 + 𝑘)! (𝑙 + 𝑗)!
×∑ .
𝑙=0 (𝑖 − 𝑘)!𝑘!Γ (𝑘 − 𝛼 + 1) (𝑗 − 𝑙)!(𝑙!)2 (𝑘 + 𝑙 − 𝛼 + 1) 5. Application of the LOM for Solving FFDEs
(87)
In this section, Legendre operational matrix of fractional
(𝛼)
Consider that in 𝐷 , the first ⌈𝛼⌉ rows, are all zero. Caputo’s derivative is considered to present its significance for
solving fuzzy fractional differential equation which is alike to
Proof. Employing the relation (85) and the orthogonal prop- the concept of the Caputo-type derivative in crisp case [36].
erties of shifted Legendre functions (56), we have The existence and uniqueness of the solutions under fuzzy
Caputo’s type derivative to this problem are discussed in [30].
𝐷(𝛼) = ⟨𝐷𝛼 Φ (𝑥) , Φ(𝑥)𝑇 ⟩ 𝐸−1 , (88)

in which ⟨𝐷𝛼 Φ(𝑥), Φ(𝑥)𝑇 ⟩ and 𝐸−1 are (𝑚 + 1)-square 5.1. General Linear FFDEs. Let us consider the general linear
matrixes defined as fuzzy fractional differential equation
1
⟨𝐷𝛼 Φ (𝑥) , Φ(𝑥)𝑇 ⟩ = {∫ 𝐷𝛼 𝐿 𝑖 (𝑥)𝐿 𝑗 (𝑥)𝑑𝑥} , (𝑐 𝐷0𝛼+ 𝑦) (𝑥) = 𝑎1 ⊙ 𝑦 (𝑥) ⊕ 𝑎2 ⊙ 𝑓 (𝑥) , 0 < 𝛼 ≤ 1,
0 𝑖,𝑗=0 (89) (94)
𝑦 (0) = 𝑦0 ∈ E,
𝐸−1 = diag {(2𝑖 + 1)}𝑚−1
𝑖=0 .
in which 𝑎𝑗 for 𝑗 = 1, 2 are fuzzy constant coefficients, 𝑦(𝑥) :
Hence, applying Lemma 43 and inserting the above matrixes
in the product 𝐷(𝛼) , the theorem be proved. 𝐿E ∩ 𝐶E is a continuous fuzzy-valued function, and 𝑐 𝐷0𝛼+
represents the fuzzy Caputo fractional derivative of order 𝛼.
Remark 45. If 𝛼 = 𝑛 ∈ N, then Theorem 44 gives the
mentioned result as in Section 4.1. Remark 47. In this section, considering more simplicity, it is
The following property of the product of two shifted assumed that for the fuzzy set-valued function 𝑦(𝑥) : 𝐼 ⊂
Legendre polynomials fuzzy vectors will also be applied R → 𝐿E ∩ 𝐶E , we present the functions 𝑦𝑟 , 𝑦𝑟 : 𝑅 → 𝐼 ⊂
which is the extension of the crisp case introduced in [36], 𝑅, 𝑟 ∈ [0, 1] by 𝑦−𝑟 (𝑥) = (𝑦(𝑥))𝑟− , 𝑦+𝑟 (𝑥) = (𝑦(𝑥))𝑟+ , for all 𝑥 ∈
𝐼, for all 𝑟 ∈ [0, 1], which are in the same previous manner,
̃ ≃ ∼ ΛΦ,
ΦΦ𝑇Λ (90) called the left and right 𝑟-cut functions of 𝑦(𝑥).
12 Abstract and Applied Analysis

Let 𝑟 ∈ [0, 1] and 0 < 𝛼 < 1. Then from (94), we have that Also, using relations (85) and (100), we obtain
𝑟 𝑟 𝑟
[𝑎1 ⊙ 𝑦 (𝑥)] = [(𝑎1 ⊙ 𝑦 (𝑥))− , (𝑎1 ⊙ 𝑦 (𝑥))+ ] 𝑐
𝐷𝛼 𝑦 (𝑥) ≃ 𝐶𝑇 ⊙ 𝐷𝛼 Φ𝑚 (𝑥) ≃ 𝐶𝑇 ⊙ 𝐷(𝛼) Φ𝑚 (𝑥) . (103)
𝑟 𝑟
= [(𝑎1 )− 𝑦(𝑥)𝑟− , (𝑎1 )+ 𝑦(𝑥)𝑟+ ] Substituting (100)–(103) in problem (94), the coefficients
(95) 𝑚
{𝑐𝑗(𝑟) } are specified by imposing the equation to be almost
= 𝑎1(𝑟) 𝑦(𝑥)(𝑟) , 𝑗=0
fuzzy exact in the Legendre operational matrix form. Now, we
𝑟 𝑟
[𝑎1 𝑦(𝑥)]± = (𝑎1 )± 𝑦(𝑥)𝑟± . establish fuzzy residual for the approximation of (94), when
[𝑦(𝑥)](𝑟) ≈ [𝑦̃𝑚 (𝑥)](𝑟) .
Therefore,
𝛼 𝑟 𝑟
( 𝑐 𝐷0+ 𝑦)± (𝑥) = (𝑎1 )± 𝑦(𝑥)𝑟± + (𝑎2 )± 𝑓(𝑥)𝑟± ,
𝑟
(96) Theorem 48. Let 𝑦𝑟 ∈ 𝐶E [0, 1] and 0 < 𝛼 ≤ 1, and then

(𝑟) (𝑟)
and thus [(𝑐 𝐷𝛼 𝑦̃𝑚 )(𝑥)] = [𝑅𝑚 (𝑥) ⊕ 𝑎2 ⊙ 𝑓̃𝑚 (𝑥) ⊕ 𝑎1 ⊙ 𝑦̃𝑚 (𝑥)] .
𝛼 𝑟 𝛼 𝑟
[( 𝑐 𝐷0+ 𝑦)− (𝑥) , ( 𝑐 𝐷0+ 𝑦)+ (𝑥)] (104)
(𝑟)
𝑟 𝑟 𝑟
= [(𝑎1 )− 𝑦(𝑥)𝑟− + (𝑎2 )− 𝑓(𝑥)𝑟− , (𝑎1 )+ 𝑦(𝑥)𝑟+ + (𝑎2 )+ 𝑓(𝑥)𝑟+ ] .
𝑟 Proof. Let 𝑟 ∈ [0, 1]. We have [(𝑐 𝐷0𝛼+ 𝑦̃𝑚 )(𝑥)] =
(97) [(𝑐 𝐷0𝛼+ 𝑦̃𝑚 )𝑟− (𝑥), (𝑐 𝐷0𝛼+ 𝑦̃𝑚 )𝑟+ (𝑥)], [𝑦̃𝑚 (𝑥)](𝑟) = (𝑟)
[𝑦𝑚− (𝑥),
(𝑟)
𝑦𝑚+ (𝑥)] and regarding to what was conversed previously,
Hence, we obtain
𝑟 𝑟
(𝑥) − (𝑎1 )± 𝑦̃𝑚 (𝑥)𝑟± − (𝑎2 )± 𝑓̃𝑚 (𝑥)𝑟± ,
𝛼 (𝑟)
𝛼 (𝑟) ( 𝑐 𝐷 𝑦̃𝑚± (𝑟)
) (𝑥) = 𝑅𝑚±
[( 𝑐 𝐷0+ 𝑦) (𝑥)]
(105)
𝛼 𝑟 𝛼 𝑟
= [( 𝑐 𝐷0+ 𝑦)− (𝑥) , ( 𝑐 𝐷0+ 𝑦)+ (𝑥)] so
𝑟 𝑟 𝑟 𝑟
= [(𝑎1 )− 𝑦(𝑥)𝑟− +(𝑎2 )− 𝑓(𝑥)𝑟− , (𝑎1 )+ 𝑦(𝑥)𝑟+ +(𝑎2 )+ 𝑓(𝑥)𝑟+ ] 𝛼
[( 𝑐 𝐷 𝑦̃𝑚 ) (𝑥)]
(𝑟)

(𝑟) (𝑟)
= [𝑎1 ⊙ 𝑦(𝑥)] + [𝑎2 ⊙ 𝑓(𝑥)] , 𝑟 ∈ [0, 1] , 𝛼
= [( 𝑐 𝐷 𝑦̃𝑚−
(𝑟)
) (𝑥) , ( 𝑐 𝐷 𝑦̃𝑚+
(𝑟) 𝛼
) (𝑥)]
(𝑟)
= [𝑎1 ⊙ 𝑦(𝑥) ⊕ 𝑎2 ⊙ 𝑓(𝑥)] , 𝑟 ∈ [0, 1] . (𝑟)
= [𝑅𝑚− (𝑟)
(𝑥) , 𝑅𝑚+ (𝑥)]
(98)
𝑟 𝑟
− [(𝑎1 )− 𝑦̃𝑚 (𝑥)𝑟− , (𝑎1 )+ 𝑦̃𝑚 (𝑥)𝑟+ ]
We can rewrite (94) in the operator form
𝑟 𝑟
(𝑎1(𝑟) 𝐼⊖ g 𝑐 𝐷𝛼 ) 𝑦(𝑟) = 𝑎2(𝑟) 𝑓(𝑟) , (99) − [(𝑎2 )− 𝑓̃𝑚 (𝑥)𝑟− , (𝑎2 )+ 𝑓̃𝑚 (𝑥)𝑟+ ]
(𝑟) (𝑟) (𝑟)
in which the fuzzy operator 𝑐 𝐷𝛼 = 𝐼1−𝛼 𝐷 is supposed to be = [𝑅𝑚 (𝑥)] ⊕ [(𝑎1 ) ⊙ 𝑦̃𝑚 (𝑥)] ⊕ [(𝑎2 ) ⊙ 𝑓̃𝑚 (𝑥)] .
compact on a fuzzy Banach space 𝑋E to 𝑋E [45, 70] and ⊖ g (106)
is the notation for g-difference.
For solving fuzzy fractional differential equation (94), Using g-difference, we have
(𝑟)
we attempt to find a fuzzy function 𝑦𝑚 ∈ 𝑋E ; there-
𝑐 𝛼 (𝑟) (𝑟)
fore, let ( 𝐷0+ 𝑦)(𝑥), 𝑦(𝑥) and 𝑓(𝑥) be approximated using [𝑅𝑚 (𝑥)] = [(𝑐 𝐷𝛼 𝑦̃𝑚 )(𝑥)]
Definition 34 as: (107)
(𝑟) (𝑟)
𝑚∗ ⊖ g [(𝑎1 ) ⊙ 𝑦̃𝑚 (𝑥)] ⊖ g [(𝑎2 ) ⊙ 𝑓̃𝑚 (𝑥)] ,
𝑇
𝑦 (𝑥) ≃ 𝑦̃𝑚 (𝑥) = ∑ 𝑐𝑗 ⊙ 𝐿 𝑗 (𝑥) = 𝐶𝑚 ⊙ Φ𝑚 , (100)
𝑗=0 or in the sense of fuzzy operator,
that (𝑟)
[𝑅𝑚 ] = (𝑎1(𝑟) 𝐼 ⊖ g 𝑐 𝐷𝛼 ) 𝑦(𝑟) ⊖ g 𝑎2(𝑟) 𝑓(𝑟) . (108)
𝑚∗
(𝑟)
[𝑦̃𝑚 (𝑥)] =∑ 𝑐𝑗(𝑟) ⊙ [𝐿 𝑗 (𝑥)] 𝑥 ∈ 𝐼 ⊂ 𝑅,
𝑗=0
(101)
𝑚∗ It is anticipated that the deriving fuzzy function [𝑦̃𝑚 (𝑥)](𝑟)
𝑓 (𝑥) ≃ 𝑓̃𝑚 (𝑥) = ∑ 𝑓𝑗 ⊙ 𝐿 𝑗 (𝑥) = 𝐹𝑚𝑇 ⊙ Φ𝑚 , will be a suitable approximation of the exact solution
(𝑟)
𝑗=0 ̃
[𝑦(𝑥)] . To this end, let 𝑋E = 𝐿2E ([0, 1]), and let ⟨⋅, ⋅⟩E
indicate the fuzzy inner product for 𝑋E . It is demanded that
where 𝐹𝑚+1 = [𝑓0 , 𝑓1 , . . . , 𝑓𝑚 ]𝑇 is obtained as (𝑟)
𝑅𝑚 satisfy
1
𝑓𝑗 = (2𝑗 + 1) ⊙ ∫ 𝑓 (𝑥) ⊙ 𝐿 𝑗 (𝑥) 𝑑𝑥. (102) (𝑟)
⟨𝑅𝑚 ̃
, 𝐿 𝑖 ⟩𝐸 = 0, 𝑖 = 0, 1, . . . , 𝑚 − 1, 𝑟 ∈ [0, 1] , (109)
0
Abstract and Applied Analysis 13

1
in which ⟨𝑅𝑚 (𝑟)
, 𝐿 𝑖 ⟩𝐸 = [(𝐹𝑅) ∫0 𝑅𝑚 (𝑥) ⊙ 𝐿 𝑖 (𝑥)𝑑𝑥](𝑟) . The left of this concept in the crisp context (see more in [69, 74–76]).
side is the shifted legendre coefficients associated with 𝐿 𝑗 . If Initially, we state the following lemma which can offer an
{𝐿 𝑖 }𝑚
𝑖=0 are the main members of shifted Legendre family Φ =
upper bound for approximating the error of Caputo fractional
{𝐿 𝑖 }∞ derivative. So we define the error vector 𝐸𝛼 as
𝑖=0 which is complete in 𝑋E , then (109) needs the main
(𝑟)
terms to be zero in the Fourier extension of 𝑅𝑚 with respect 𝑇
to Φ which is called tau method in the crisp context and it 𝐸𝛼 = 𝐷𝛼 Φ − 𝐷(𝛼) Φ = [𝐸0,𝛼 , 𝐸1,𝛼 , . . . , 𝐸𝑚,𝛼 ] , (114)
is in a similar manner with the meaning of fuzzy (for more
details, see [71–73]). where
To discover 𝑦̃𝑛(𝑟) , implementing (109) to (108), regarding 𝑚
the (100)–(103), we generate 𝑚 fuzzy linear equations as 𝐸𝑖,𝛼 = 𝐷𝛼 𝐿 𝑖 (𝑥) − ∑𝐷𝑖𝑗(𝛼) 𝐿 𝑗 (𝑥) , 𝑖 = 0, 1, . . . , 𝑚. (115)
𝑗=0
(𝑟)
⟨[𝑅𝑚 (𝑥)] , 𝐿 𝑖 (𝑥)⟩
Now, we can propose the next lemma by using Theorem 1
𝑐 𝛼 (𝑟) (𝑟) in [69] to depict the error bound of Caputo fractional
= ⟨[( 𝐷 𝑦̃𝑚 )(𝑥)] ⊖ g [(𝑎1 ) ⊙ 𝑦̃𝑚 (𝑥)] (110) derivative operator for the shifted Legendre polynomials.
(𝑟)
⊖ g [(𝑎2 ) ⊙ 𝑓̃𝑚 (𝑥)] , 𝐿 𝑖 (𝑥)⟩ = 0,
̃ Lemma 49. Let the error function of Caputo fractional deriva-
tive operator for Legendre polynomials 𝐸𝑖,𝛼 : [𝑥0 , 1] → R
for 𝑖 = 0, 1, . . . , 𝑚 − 1. Equation (110) generate (𝑚) set of fuzzy be 𝑚 + 1 times continuously differentiable for 0 < 𝑥0 ≤ 𝑥,
linear equations. These fuzzy linear equations can be acquired 𝑥 ∈ (0, 1]. Also 𝐸𝑖,𝛼 ∈ 𝐶𝑚+1 [𝑥0 , 1] and 𝛼 < 𝑚 + 1 and then the
as the following system of equations: error bound is presented as follows:

𝑚−1∗ 󵄩󵄩 󵄩󵄩 |Γ (𝑖 + 1)| 𝑀𝛼 1
∑ 𝑐𝑗(𝑟) ⊙ {⟨𝐷(𝛼) 𝐿 𝑗 , 𝐿 𝑖 ⟩ − ⟨𝑎1(𝑟) 𝐿 𝑗 , 𝐿 𝑖 ⟩} 󵄩󵄩𝐸𝑖,𝛼 󵄩󵄩 ≤ √ 𝑥−𝛼 .
|Γ (1 − 𝛼)| Γ (𝑚𝛼 + 𝛼 + 1) (2 (𝑚 + 1) 𝛼 + 1) 0
𝑗=0
(111) (116)
𝑚−1∗
= ∑ 𝑓𝑗(𝑟) ⊙ ⟨𝑎2(𝑟) 𝐿 𝑗 , 𝐿 𝑖 ⟩ , 𝑖 = 0, 1, . . . , 𝑚 − 1 Proof. Firstly, we acquire a bound for 𝑐
𝐷𝛼 𝑥𝑖 , regarding
𝑗=0 Definition 24, as
or
𝑐 Γ (𝑖 + 1) 𝑖−𝛼
𝐷𝛼 𝑥 𝑖 = 𝑥
𝑚−1∗ 1 Γ (𝑖 + 1 − 𝛼)
∑ 𝑐𝑗(𝑟) ⊙ {(𝐹𝑅) ∫ 𝐷(𝛼) 𝐿 𝑗 (𝑥) 𝐿 𝑖 (𝑥) 𝑑𝑡 (117)
𝑗=0 0 |Γ (𝑚 + 2)| −𝛼
≤ 𝑥 , 𝑖 = 0, 1, . . . , 𝑚.
|Γ (1 − 𝛼)| 0
1
− (𝐹𝑅) ∫ 𝑎1(𝑟) 𝐿 𝑗 (𝑥) 𝐿 𝑖 (𝑥) 𝑑𝑡} (112)
0 Then using Lemma 1 in [69], we have:
𝑚−1∗ 1 𝑖
= ∑ 𝑓𝑗(𝑟) ⊙ (𝐹𝑅) ∫ 𝑎2(𝑟) 𝐿 𝑗 (𝑥) 𝐿 𝑖 (𝑥) 𝑑𝑡. |Γ (𝑖 + 1)| −𝛼 𝑖
𝐷𝛼 𝐿 𝑖 (𝑥) = ∑ 𝑒𝑘,𝑖 𝐷𝛼 𝑥𝑘 ≤ 𝑥 ∑𝑒
𝑗=0 0
𝑘=0
|Γ (1 − 𝛼)| 0 𝑘=0 𝑘,𝑖
(118)
Afterwards, substitution of (100) in the initial condition of |Γ (𝑖 + 1)| −𝛼 |Γ (𝑖 + 1)| −𝛼
= 𝑥 𝐿 (1) = 𝑥 ,
(94) yields |Γ (1 − 𝛼)| 0 𝑖 |Γ (1 − 𝛼)| 0
𝑚∗
in which
𝑦 (0) = ∑ 𝑐𝑗(𝑟) ⊙ 𝐿 𝑗 (0) = 𝑦0 , (113)
𝑗=0 (𝑘 + 𝑖)!
𝑒𝑘,𝑖 = (−1)𝑘+𝑖 . (119)
that this equation be coupled with the previous fuzzy linear (𝑖 − 𝑘)!(𝑘!)2
equations and constructed (𝑚 + 1) fuzzy linear equations.
Now, utilizing Theorem 1 in [69], the lemma can be proved.
Clearly, after solving this fuzzy system, the coefficients {𝑐𝑗 }𝑚
𝑗=0
will be gained.
Therefore, the maximum norm of error vector 𝐸𝛼 is
5.1.1. Error Analysis. The aim of this section is to acquire the attained as
error bound for the Legendre approximation using shifted
Legendre polynomials. We consider the best shifted Legendre |Γ (𝑚 + 1)| 𝑀𝛼 1
󵄩󵄩 󵄩󵄩
approximation of a smooth fuzzy function under Caputo 󵄩󵄩𝐸𝛼 󵄩󵄩∞ ≤ √ 𝑥−𝛼 .
derivative for 0 < 𝛼 ≤ 1 to reach the result. It should be |Γ (1 − 𝛼)| Γ (𝑚𝛼 + 𝛼 + 1) (2 (𝑚 + 1) 𝛼 + 1) 0
considered that the results of this section are the extension (120)
14 Abstract and Applied Analysis

Now we extend these results for the fuzzy case. Hence, the approximate and exact solutions of the fuzzy fractional
error of fuzzy Caputo fractional derivative is defined as differential equations, respectively. Then we rewrite (94) as
𝛼 (𝑟) (𝑟)
𝐷∗ (𝐷𝛼 Φ (𝑥) , 𝐷(𝛼) Φ (𝑥)) ( 𝑐 𝐷 𝑦𝑚±
(𝑟)
) (𝑥) − (𝑎1 )± 𝑦𝑚 (𝑥)(𝑟) (𝑟) (𝑟)
± = 𝐻𝑚± (𝑥) + (𝑎2 )± 𝑓𝑚 (𝑥)± ,
(124)
= sup 𝐷 (𝐷𝛼 Φ (𝑥) , 𝐷(𝛼) Φ (𝑥)) (𝑟)
𝑥∈[0,1] that 𝐻𝑚± (𝑥) is the fuzzy perturbation function that depends
󵄨 󵄨 only on 𝑦𝑚 (𝑥)(𝑟) ± . By subtracting (94) from above equation
= sup sup max {󵄨󵄨󵄨󵄨𝐷𝛼 Φ(𝑥)𝑟− (𝑥) − 𝐷(𝛼) Φ(𝑥)𝑟− (𝑥)󵄨󵄨󵄨󵄨 , and using Lemmas 38 and 50, Theorem 40, and Definition 2,
𝑥∈[0,1] 𝑟∈[0,1]
one can obtain
󵄨󵄨 𝛼 󵄨 󵄩󵄩 (𝑟) 󵄩󵄩
󵄨󵄨𝐷 Φ(𝑥)𝑟+ (𝑥) − 𝐷(𝛼) Φ(𝑥)𝑟+ (𝑥)󵄨󵄨󵄨}
󵄨 󵄨 󵄩󵄩󵄩𝐻𝑚± (𝑥)󵄩󵄩󵄩E
󵄩󵄩 𝛼 󵄩 󵄩 󵄩
= sup max {󵄩󵄩󵄩𝐷 Φ(𝑥)𝑟− (𝑥) − 𝐷(𝛼) Φ(𝑥)𝑟− (𝑥)󵄩󵄩󵄩󵄩∞ , ≤ sup max {󵄩󵄩󵄩𝐷𝛼 Φ(𝑥)𝑟± − 𝐷𝛼 Φ(𝑥)𝑟± 󵄩󵄩󵄩∞ }
𝑟∈[0,1] 𝑟∈[0,1] (125)
󵄩󵄩 𝛼 󵄩 󵄩 󵄩
󵄩󵄩𝐷 Φ(𝑥)𝑟+ (𝑥) − 𝐷(𝛼) Φ(𝑥)𝑟+ (𝑥)󵄩󵄩󵄩 } + (𝑎1 )± sup max {󵄩󵄩󵄩󵄩𝑦±𝑟 (𝑥) − 𝑦𝑚+1,±
(𝑟) 𝑟
(𝑥)󵄩󵄩󵄩󵄩∞ } .
󵄩 󵄩∞
(121) 𝑟∈[0,1]

If we substitute the error bound from Lemmas 38 and 50 and


regarding Lemmas 38 and 49 and Theorem 40, and we have: the proof of Theorem 40, then we have
󵄩 󵄩 󵄩󵄩󵄩𝐻(𝑟) (𝑥)󵄩󵄩󵄩
sup max {󵄩󵄩󵄩󵄩𝐷𝛼 Φ(𝑥)𝑟− − 𝐷(𝛼) Φ(𝑥)𝑟− 󵄩󵄩󵄩󵄩∞ , 󵄩󵄩 𝑚± 󵄩󵄩𝐸
𝑟∈[0,1]
(𝑟)
󵄩󵄩 𝛼 󵄩 |Γ (𝑚 + 1)| 𝑀𝛼,± 1
󵄩󵄩𝐷 Φ(𝑥)𝑟+ (𝑥) − 𝐷(𝛼) Φ(𝑥)𝑟+ (𝑥)󵄩󵄩󵄩 } ≤ √ 𝑥−𝛼
󵄩 󵄩∞ |Γ (1 − 𝛼)| Γ (𝑚𝛼 + 𝛼 + 1) (2 (𝑚 + 1) 𝛼 + 1) 0
𝑟
Lemma 49 |Γ (𝑚 + 1)| 𝑀𝛼,− (𝑟)
≤ sup max { (𝑟) 𝑀𝛼,± 1
𝑟∈[0,1] |Γ (1 − 𝛼)| Γ (𝑚𝛼 + 𝛼 + 1) + (𝑎1 )± √ ,
Γ (𝑚𝛼 + 𝛼 + 1) (2 (𝑚 + 1) 𝛼 + 1)
1 (126)
×√ 𝑥−𝛼 ,
(2 (𝑚 + 1) 𝛼 + 1) 0 or
󵄩󵄩 (𝑟) 󵄩󵄩
𝑟 󵄩󵄩𝐻𝑚± (𝑥)󵄩󵄩 ≤ 𝐶 (1 + 𝑥0𝛼 ) , (127)
|Γ (𝑚 + 1)| 𝑀𝛼,+ 󵄩 󵄩E
|Γ (1 − 𝛼)| Γ (𝑚𝛼 + 𝛼 + 1) at which 𝐶 = (|Γ(𝑚 + 1)|/|Γ(1 − 𝛼)|)(𝑀𝛼(𝑟) /Γ(𝑚𝛼 + 𝛼 +
1))√1/(2(𝑚 + 1)𝛼 + 1), for 0 < 𝑥0 ≤ 𝑥, 𝑥 ∈ (0, 1]. Therefore,
1
×√ 𝑥−𝛼 } , 𝐻𝑚(𝑟) (𝑥) is bounded.
(2 (𝑚 + 1) 𝛼 + 1) 0
(122) 5.2. Nonlinear FFDEs. Consider the nonlinear FFDE
where 0 < 𝑥0 ≤ 𝑥, 𝑥 ∈ (0, 1]. (𝑐 𝐷0𝛼+ 𝑦) (𝑥) = 𝐹 (𝑥, 𝑦 (𝑥) , 𝑦 (𝑥)) , 0 < 𝛼 < 1,
(128)
Lemma 50. Consider the error function of the fuzzy Caputo 𝑦 (0) = [𝑦 , 𝑦0 ] ,
fractional derivative operator for shifted Legendre polynomials 0
which is continuously differentiable and 0 < 𝛼 < 1. Then the where 𝐹 can be nonlinear in general, 𝑦(𝑥) : 𝐿E ∩ 𝐶E is a
error bound is given as follows: continuous fuzzy-valued function, and 𝑐 𝐷0𝛼+ indicates the
fuzzy Caputo fractional derivative of order 𝛼.
𝐷∗ (𝐷𝛼 Φ (𝑥) , 𝐷(𝛼) Φ (𝑥)) In order to use LOM for this problem, we first approxi-
mate 𝑦(𝑥) and (𝑐 𝐷0𝛼+ 𝑦)(𝑥) as (100) and (103), respectively. By
|Γ (𝑚 + 1)| 𝑀𝛼 1 replacing these equations in (128), we have
≤ √ 𝑥−𝛼 ,
|Γ (1 − 𝛼)| Γ (𝑚𝛼 + 𝛼 + 1) (2 (𝑚 + 1) 𝛼 + 1) 0 𝑇
(123) 𝐶𝑇 ⊙ 𝐷(𝛼) Φ (𝑥) ≃ 𝐹 (𝑥, 𝐶𝑇 Φ (𝑥) , 𝐶 Φ (𝑥)) . (129)

that 𝑀𝛼 = max{𝑀𝛼,− , 𝑀𝛼,+ }. We intend to find the fuzzy coefficients 𝐶𝑚 . Also by substitut-
ing (100) in the initial condition of nonlinear FFDE (128), we
Proof. It is straightforward from the definition of (𝐸𝛼 ), have
Definition 1, and Remark 3. 𝑦 (0) = 𝐶𝑇 Φ (0) = 𝑦 ,
0
(130)
For an error assessment of the approximation solution 𝑇
of (94), we assume that 𝑦𝑚 (𝑥) and 𝑦(𝑥) reveal the fuzzy 𝑦 (0) = 𝐶 Φ (0) = 𝑦0 .
Abstract and Applied Analysis 15

To find the approximate fuzzy solution 𝑦̃𝑚 (𝑥), we first Remark 53. If 𝛼 = 1, then the fuzzy fractional differential
collocate (129) at (𝑚) points. For appropriate collocation equations (132) are converted to fuzzy integer-order differen-
points we use the first (𝑚) shifted Legendre polynomials tial equations. So the exact solution of problem under (1)-
roots. These equations together with (130) generate (𝑚 + differentiability using Theorem 15 is as follows:
1) nonlinear fuzzy equations which can be solved using
Newton’s iterative method presented in [77]. 𝑌 (𝑥; 𝑟) = (0.5 + 0.5𝑟) 𝑒𝑥 ,
(135)
Remark 51. The solvability of system (129) and (130) is a 𝑌 (𝑥; 𝑟) = (1.5 − 0.5𝑟) 𝑒𝑥 , 0 < 𝑟 ≤ 1.
complicated problem and we cannot prove the existence and
uniqueness of such a fuzzy solution. But in our accomplish- By utilizing the technique explained in Section 5, (131) with
ment, we have solved this system, regarding the method in 𝜆 = 1 becomes:
[77], using MATLAB functions. In the assumed example, 𝑚
these functions have prospered to gain an accurate fuzzy ∑ 𝑐𝑗𝑟 [𝑑𝑖,𝑗
(𝛼)
− 𝐼] 𝐿 𝑗 (𝑥; 𝑟) = 0 𝑖 = 0, . . . , 𝑚,
approximate solution of the system, even starting with a zero 𝑗=0
initial guess. (136)
𝑚
∑ 𝑐𝑟 𝑗 [𝑑𝑖,𝑗
(𝛼)
− 𝐼] 𝐿 𝑗 (𝑥; 𝑟) = 0 𝑖 = 0, . . . , 𝑚,
6. Numerical Examples 𝑗=0

In this section, to demonstrate the effectiveness of the or we can rewrite it in the matrix form as
proposed method in the present paper, two different test
𝑇
examples are carried out. Also, the obtained numerical 𝐶𝑚,− [𝐷(𝛼) − 𝐼] Φ (𝑥) = 0,
solutions be compared with exact solutions. (137)
𝑇
𝐶𝑚,+ [𝐷(𝛼) − 𝐼] Φ (𝑥) = 0,
Example 52. Let us consider the following FFDE:
𝛼 in which (𝑐𝑗𝑟 ) = [𝑐𝑗,− 𝑟 𝑟
, 𝑐𝑗,+ ] and for 𝑗 = 0, 1, . . . , 𝑚. As it
( 𝑐 𝐷0+ 𝑦) (𝑥) = 𝜆 ⊙ 𝑦 (𝑥) , 0 < 𝛼 ≤ 1,
was described in Section 5, we produce 𝑚 fuzzy algebraic
(131)
equations multiplied above fuzzy residual system by 𝐿 𝑖 (𝑥) for
𝑦 (0; 𝑟) = [𝑦 (𝑟) , 𝑦0 (𝑟)] , 0 < 𝑟 ≤ 1,
0 𝑖 = 0, 1, . . . , 𝑚−1 using orthogonal property, and so we obtain
in which 𝑦(𝑥) : 𝐿E ∩ 𝐶E is a continuous fuzzy-valued function 𝑇
𝐶𝑚,− [𝐷(𝛼) − 𝐼] = 0,
and 𝑐 𝐷0𝛼+ denotes the fuzzy Caputo fractional derivative of (138)
order 𝛼. We solve this example according to two following 𝑇
𝐶𝑚,+ [𝐷(𝛼) − 𝐼] = 0.
cases for 𝜆 = 1, −1.

Case 1. Suppose that 𝜆 = 1, and then using 𝑐 [1 − 𝛼]- Now, using the initial condition (131), we have
differentiability and Theorem 33, we have the following: 𝑇
𝑦 (0; 𝑟) ≃ 𝐶𝑚,− Φ𝑚 = (0.5 + 0.5𝑟) ,
𝛼
( 𝑐 𝐷0+ 𝑦) (𝑥; 𝑟) = 𝑦 (𝑥) , 0 < 𝛼 ≤ 1, 𝑇
(139)
𝑦 (0; 𝑟) ≃ 𝐶𝑚,+ Φ𝑚 = (1.5 − 0.5𝑟) .
𝑦 (0; 𝑟) = 𝑦0 (𝑟) , 0 < 𝑟 ≤ 1,
(132) The above three equations generate a set of 𝑚 + 1 fuzzy linear
𝛼
( 𝑐 𝐷0+ 𝑦) (𝑥; 𝑟) = 𝑦 (𝑥) , 0 < 𝛼 ≤ 1, algebraic equations. As a result, for 𝛼 = 0.85 and 𝑚 = 3, one
can gain
𝑦 (0; 𝑟) = 𝑦0 (𝑟) , 0 < 𝑟 ≤ 1,
0 0 0 0
̃ 𝑟) = [0.5 + 0.5𝑟, 1.5 − 0.5𝑟].
in which 𝑦(0; [ 1.8639 0.3901 −0.1755 0.1095 ]
𝐷 (0.85)
=[
[−0.3901 4.5267 0.8696 −0.4078]
] (140)

[ 1.6885 −0.4794 6.7831 1.3797 ]


The exact solution of FFDE is as follows:
𝑌 (𝑥; 𝑟) = (0.5 + 0.5𝑟) 𝐸𝛼,1 [𝑥𝛼 ] , 0 < 𝛼 ≤ 1,
(133) the unknown coefficients 𝑐𝑗 can be gained by substituting
𝛼
𝑌 (𝑥; 𝑟) = (1.5 − 0.5𝑟) 𝐸𝛼,1 [𝑥 ] , 0 < 𝑟 ≤ 1, matrix 𝐷(0.85) in the previous mentioned systems.

where 𝐸𝛼,𝛼 is the classical Mittag-Leffler function Case 2. Suppose that 𝜆 = −1, then using 𝑐 [2 − 𝛼]-
differentiability and Theorem 33, the FDEs are obtained as

𝑧𝑘 same as (132). Also the exact solution is given by
𝐸𝛼,𝛼 (𝑧) = ∑ . (134)
𝑘=0
Γ (𝛼 (𝑘 + 1))
𝑌 (𝑥; 𝑟) = (0.5 + 0.5𝑟) 𝐸𝛼,1 [−𝑥𝛼 ] , 0 < 𝛼 ≤ 1,
Note that for 𝛼 = 1 this representation is still valid since (141)
Γ(1) = 1 and 𝐸1,1 (𝑧) = 𝑒𝑧 . 𝑌 (𝑥; 𝑟) = (1.5 − 0.5𝑟) 𝐸𝛼,1 [−𝑥𝛼 ] , 0 < 𝑟 ≤ 1.
16 Abstract and Applied Analysis

10−1 10−2

10−2

Absolute error
Absolute error

10−3 10−3

10−4

10−5 10−4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cut r-cut
m=2 m=5
m=4 m=9
m=8 m = 11

Figure 2: The absolute error for different 𝑚 of Example 52, Case 1. Figure 4: The Absolute Error for different 𝑚 of Example 52, Case 2.
𝛼 = 0.85. 𝛼 = 0.75.

differential equation. So the exact solution of problem under


5
(2)-differentiability using Corollary 16 is as follows:
Fuzzy approximate solution (y(x))

4.5
4 𝑌 (𝑥; 𝑟) = (0.5 + 0.5𝑟) 𝑒−𝑥 ,
3.5 (143)
3 𝑌 (𝑥; 𝑟) = (1.5 − 0.5𝑟) 𝑒−𝑥 , 0 < 𝑟 ≤ 1.
2.5
2 The approximate solution, exact solution and absolute errors
are depicted for Case 1 of Example 52 in Table 1 for 𝑥 = 1 with
1.5
𝛼 = .85. It can be seen that a few terms of the shifted Legendre
1
functions are required to achieve a suitable approximation
0.5 which demonstrate the applicability of the proposed method
0 0.2 0
0.2 0.4
0.4 0.6 0.8 0.8 0.6 for this problem. Also the absolute errors for 𝑚 = 2, 4, 8
1 1 r-cut
x with 𝛼 = 0.85 are plotted in Figure 2 which show the
decreasing of the error with the increasing of the number of
Figure 3: The Fuzzy approximate solution of Example 52, Case 1.
Legendre functions. The fuzzy approximate solution is shown
𝛼 = 0.85, 𝑚 = 4.
in Figure 3 for 𝛼 = 0.85, 𝑚 = 4.

̃ 𝑟) of
Remark 55. Figure 2 depicts the absolute errors for 𝑦(𝑥;
By applying the technique described in Sections 4 and 5, ̃ 𝑟),
Example 52, Case 1. In the same way, if we consider 𝑦(𝑥;
namely 𝑚 = 2, we may write the approximate solution for then analogously to the demonstration of Figure 2, we can
Cases 1 and 2 in the forms obtain the absolute errors.

In the Table 2, the fuzzy approximate solution for Case 2


𝑦̃ (𝑥) = 𝑐0 𝐿 0 (𝑥) + 𝑐1 𝐿 1 (𝑥) + 𝑐2 𝐿 2 (𝑥) , of Example 52 is considered. The fractional Caputo derivative
2
(142) is 𝛼 = 0.75 and the number of Legendre functions are 𝑚 =
𝑦̃2 (𝑥) = 𝑐0 𝐿 0 (𝑥) + 𝑐1 𝐿 1 (𝑥) + 𝑐2 𝐿 2 (𝑥) . 9. The result are computed for 𝑥 = 1 with different 𝑟-cut.
Again we can see that Table 2 demonstrates the validity of the
method for this kind of problems. Furthermore, the absolute
So regarding the (100)–(110) and (142), the unknown param- errors of Example 52, Case 2, are plotted in Figure 4 with
eters, 𝑐𝑗 , 𝑐𝑗 , 𝑗 = 0, 1, 2, are achieved for both cases. 𝑚 = 5, 9, 11 and the fuzzy approximate solution is shown in
Figure 5 for 𝛼 = 0.75, 𝑚 = 9.

Remark 54. If 𝛼 = 1, then the fuzzy fractional differential Remark 56. Figure 4 shows the absolute errors for 𝑦(𝑥; 𝑟) of
equation (131) with 𝜆 = −1 is converted to fuzzy integer-order Example 52, Case 2. In the same way, if we consider 𝑦(𝑥; 𝑟),
Abstract and Applied Analysis 17

Table 1: The result of the proposed method for Case 1 of Example 52 with 𝛼 = 0.85 and 𝑚 = 4.

𝑟 𝑦4𝑟 𝑌(𝑟) Error 𝑦4𝑟 𝑌(𝑟) Error


0 1.5625 1.5627 2.455𝑒 − 4 4.6875 4.6882 7.364𝑒 − 4
0.1 1.7188 1.7190 2.270𝑒 − 4 4.5313 4.5320 7.119𝑒 − 4
0.2 1.8750 1.8753 2.946𝑒 − 4 4.3750 4.3757 6.873𝑒 − 4
0.3 2.0313 2.0316 3.191𝑒 − 4 4.2188 4.2194 6.628𝑒 − 4
0.4 2.1875 2.1878 3.437𝑒 − 4 4.0625 4.0631 6.382𝑒 − 4
0.5 2.3438 2.3441 3.682𝑒 − 4 3.9063 3.9069 6.137𝑒 − 4
0.6 2.5000 2.5004 3.928𝑒 − 4 3.7500 3.7506 5.891𝑒 − 4
0.7 2.6563 2.6567 4.173𝑒 − 4 3.5938 3.5943 5.646𝑒 − 4
0.8 2.8125 2.8129 4.419𝑒 − 4 3.4375 3.4380 5.400𝑒 − 4
0.9 2.9688 2.9692 4.664𝑒 − 4 3.2813 3.2818 5.155𝑒 − 4
1 3.1250 3.1255 4.909𝑒 − 4 3.1250 3.1255 4.909𝑒 − 4

Table 2: The result of the proposed method for Case 2 of Example 52 with 𝛼 = 0.75 and 𝑚 = 9.

𝑟 𝑦5𝑟 𝑌(𝑟) Error 𝑦5𝑟 𝑌(𝑟) Error


0 0.1962 0.1966 3.149𝑒 − 4 0.5887 0.5897 9.446𝑒 − 4
0.1 0.2159 0.2162 3.464𝑒 − 4 0.5691 0.5700 9.131𝑒 − 4
0.2 0.2355 0.2359 3.778𝑒 − 4 0.5495 0.5504 8.816𝑒 − 4
0.3 0.2551 0.2555 4.093𝑒 − 4 0.5298 0.5307 8.501𝑒 − 4
0.4 0.2747 0.2752 4.408𝑒 − 4 0.5102 0.5110 8.186𝑒 − 4
0.5 0.2944 0.2948 4.723𝑒 − 4 0.4906 0.4914 7.872𝑒 − 4
0.6 0.3140 0.3145 5.038𝑒 − 4 0.4710 0.4717 7.557𝑒 − 4
0.7 0.3336 0.3341 5.353𝑒 − 4 0.4514 0.4521 7.242𝑒 − 4
0.8 0.3532 0.3538 5.668𝑒 − 4 0.4317 0.4324 6.927𝑒 − 4
0.9 0.3729 0.3735 5.982𝑒 − 4 0.4121 0.4128 6.612𝑒 − 4
1 0.3925 0.3931 6.297𝑒 − 4 0.3925 0.3931 6.297𝑒 − 4

Here, suppose that 𝜆 = −1, then using 𝑐 [2 − 𝛼]-


1.5
Fuzzy approximate solution (y(x))

differentiability and Theorem 33 we have the following:


𝛼
( 𝑐 𝐷0+ 𝑦) (𝑥; 𝑟) = −𝑦 (𝑥) + 𝑥 + 1, 0 < 𝛼 ≤ 1,
1
𝑦 (0; 𝑟) = 𝑦 (𝑟) , 0 < 𝑟 ≤ 1,
0
(145)
𝛼
0.5 ( 𝑐 𝐷0+ 𝑦) (𝑥; 𝑟) = −𝑦 (𝑥) + 𝑥 + 1, 0 < 𝛼 ≤ 1,

𝑦 (0; 𝑟) = 𝑦0 (𝑟) , 0 < 𝑟 ≤ 1,


0
0 ̃ 𝑟) = [0.5 + 0.5𝑟, 1.5 − 0.5𝑟]. Also, regarding (145),
where 𝑦(0;
0.5 1 the exact solution of FFDE (144) be determined as
x 0.6 0.8
0.4
1 0 0.2
r-cut 𝑌 (𝑥; 𝑟) = (0.5 + 0.5𝑟) 𝐸𝛼,1 [𝑥𝛼 ]
𝑥
Figure 5: The fuzzy approximate solution of Example 52, Case 2.
+ ∫ (𝑥 − 𝑡)𝛼−1 𝐸𝛼,𝛼 [−(𝑥 − 𝑡)𝛼 ] (𝑥 + 1) 𝑑𝑡,
𝛼 = 0.75, 𝑚 = 9. 0

0 < 𝛼 ≤ 1,
then analogously to the demonstration of Figure 4, we can
obtain the absolute errors. 𝑌 (𝑥; 𝑟) = (1.5 − 0.5𝑟) 𝐸𝛼,1 [𝑥𝛼 ]
𝑥
Example 57. Let us consider the following FFDE: + ∫ (𝑥 − 𝑡)𝛼−1 𝐸𝛼,𝛼 [−(𝑥 − 𝑡)𝛼 ] (𝑥 + 1) 𝑑𝑡,
0
(𝑐 𝐷0𝛼+ 𝑦) (𝑥) = 𝜆 ⊙ 𝑦 (𝑥) ⊕ (𝑥 + 1) , 0 < 𝛼 ≤ 1,
(144) 0 < 𝑟 ≤ 1.
𝑦 (0; 𝑟) = [𝑦̃ (𝑟) , 𝑦0 (𝑟)] 0 < 𝑟 ≤ 1. (146)
0
18 Abstract and Applied Analysis

Table 3: The result of the proposed method for Example 57 with 𝛼 = 0.95 and 𝑚 = 6.

𝑟 𝑦6𝑟 𝑌(𝑟) Error 𝑦6𝑟 𝑌(𝑟) Error


0 1.191161 1.191169 8.916561𝑒 − 6 1.562753 1.562743 9.986616𝑒 − 6
0.1 1.209740 1.209748 7.971402𝑒 − 6 1.544173 1.544164 9.041457𝑒 − 6
0.2 1.228320 1.228327 7.026243𝑒 − 6 1.525594 1.525586 8.096298𝑒 − 6
0.3 1.246899 1.246905 6.081084𝑒 − 6 1.507014 1.507007 7.151139𝑒 − 6
0.4 1.265479 1.265484 5.135925𝑒 − 6 1.488435 1.488428 6.205980𝑒 − 6
0.5 1.284059 1.284063 4.190766𝑒 − 6 1.469855 1.469850 5.260821𝑒 − 6
0.6 1.302638 1.302642 3.245608𝑒 − 6 1.451275 1.451271 4.315663𝑒 − 6
0.7 1.321218 1.321220 2.300449𝑒 − 6 1.432696 1.432692 3.370504𝑒 − 6
0.8 1.339798 1.339799 1.355290𝑒 − 6 1.414116 1.414114 2.425345𝑒 − 6
0.9 1.358377 1.358378 4.101313𝑒 − 7 1.395536 1.395535 1.480186𝑒 − 6
1 1.376957 1.376956 5.350274𝑒 − 7 1.376957 1.376956 5.350274𝑒 − 7

Applying the shifted Legendre method with LOM technique Finally, the corresponding fuzzy approximate solution from
explained in Sections 4 and 5, we have (151) can be acquired.
For case 𝛼 = 0.95 and 𝑚 = 6, 𝑥 = 1 with 𝑟 = 0.1, we
𝑚 𝑚
obtain the fuzzy approximate solution in a series expansion
∑ 𝑐𝑗𝑟 [𝑑𝑖,𝑗
(𝛼)
+ 𝐼] 𝐿 𝑗 (𝑥; 𝑟) = ∑ 𝑓𝑗𝑟 𝐿 𝑗 (𝑥; 𝑟) 𝑖 = 0, . . . , 𝑚,
as
𝑗=0 𝑗=0

𝑚 𝑚 𝑦̃ (𝑥, 0.1) = 0.5500 + 0.5533𝑥 − 0.1915𝑥2 + 1.1965𝑥3


𝑟 (𝛼) 𝑟
∑𝑐 𝑗 [𝑑𝑖,𝑗 + 𝐼] 𝐿 𝑗 (𝑥; 𝑟) = ∑ 𝑓 𝑗 𝐿 𝑗 (𝑥; 𝑟) 𝑖 = 0, . . . , 𝑚,
𝑗=0 𝑗=0 − 1.9554𝑥4 + 1.5046𝑥5 − 0.4478𝑥6 ,
(147) (152)
𝑦̃ (𝑥, 0.1) = 1.4500 − 0.5496𝑥 + 1.3466𝑥2 − 1.9084𝑥3
or it can be written in the matrix form as
+ 2.4940𝑥4 − 1.8195𝑥5 − 1.8195𝑥6 .
𝑇 (𝛼) 𝑇
𝐶𝑚,− [𝐷 + 𝐼] Φ (𝑥) = 𝐹𝑚,− Φ (𝑥) ,
(148) Remark 58. If 𝛼 = 1, then the fuzzy fractional differential
𝑇
𝐶𝑚,+ [𝐷(𝛼) + 𝐼] Φ (𝑥) = 𝐹𝑚,+
𝑇
Φ (𝑥) , equation (144) with 𝜆 = −1 is converted to fuzzy integer-
order differential equation. Therefore, the exact solution of
in which (𝑐𝑗𝑟 ) = [𝑐𝑗,− 𝑟 𝑟
, 𝑐𝑗,+ ] and (𝑓𝑗𝑟 ) = [𝑓𝑗,−
𝑟 𝑟
, 𝑓𝑗,+ ] for 𝑗 = problem under (2)-differentiability using Corollary 16 is as
0, 1, . . . , 𝑚. As was illustrated in Section 5, we make 𝑚 fuzzy follows:
algebraic equations by the result of the inner product of fuzzy
𝑌 (𝑥; 𝑟) = 𝑥 + (0.5 + 0.5𝑟) 𝑒−𝑥 ,
residual with 𝐿 𝑖 (𝑥), 𝑖 = 0, 1, . . . , 𝑚 − 1, so we gain
(153)
𝑇 𝑌 (𝑥; 𝑟) = 𝑥 + (1.5 − 0.5𝑟) 𝑒−𝑥 .
𝐶𝑚,− [𝐷(𝛼) + 𝐼] = 𝐹𝑚,−
𝑇
,
(149) In order to evaluate the advantages and the accuracy of using
𝑇
𝐶𝑚,+ [𝐷(𝛼) + 𝐼] = 𝐹𝑚,+
𝑇
. the presented method for the fuzzy fractional differential
equations, Example 57 is considered. The results are illus-
Also for the initial condition (144), we have trated in Table 3 are the approximate solutions are compared
𝑇 with exact solutions and the absolute errors are derived for
𝑦 (0; 𝑟) ≃ 𝐶𝑚,− Φ𝑚 = (0.5 + 0.5𝑟) , 𝛼 = 0.95 with 𝑚 = 6 at 𝑥 = 1. It is obvious that the fuzzy
(150) approximate solutions are in high agreement with the fuzzy
𝑇
𝑦 (0; 𝑟) ≃ 𝐶𝑚,+ Φ𝑚 = (1.5 − 0.5𝑟) . exact solutions. Moreover, different numbers of Legendre
polynomials are applied to obtain the absolute errors for this
The above equations produce a set of 𝑚 + 1 fuzzy linear problem which can be seen in Figure 6. This graph shows
algebraic equations. As a result, namely, with 𝑚 = 2, the that the method has a good convergence rate. Also, the fuzzy
approximate fuzzy solution should be written in the form approximate solution is shown for 𝛼 = 0.95 in Figure 7. The
2
absolute error of different order of fractional differentiability
𝑦̃ (𝑥) = ∑ 𝑐𝑗 𝐿 𝑗 (𝑥) = [𝑐0 𝑐1 𝑐2 ] Φ (𝑥) ,
𝑇 can be considered for 𝑥 = 1 in Figure 8 and the approximate
𝑗=0
solutions are shown in Figure 9 that shows that this approach
(151) can apply to solve the problem effectively with different fuzzy
2 fractional order of derivatives with suitable errors. It can bee
𝑇
𝑦̃ (𝑥) = ∑ 𝑐𝑗 𝐿 𝑗 (𝑥) = [𝑐0 𝑐1 𝑐2 ] Φ (𝑥) . seen that the 𝛼 approaches an integer order, and the error has
𝑗=0 a tendency to decrease, as expected.
Abstract and Applied Analysis 19

10−5 10−3

10−4
−6
10
Absolute error

Absolute error
10−5

10−7
10−6

10−8 10−7
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cut r-cut
m=6 𝛼 = 0.75
m=9 𝛼 = 0.85
m = 12 𝛼 = 0.95

Figure 6: The absolute error for different 𝑚 of Example 57, 𝛼 = 0.95. Figure 8: The absolute error for different 𝛼 of Example 57, 𝑚 = 8.

1
1.8 0.9
Fuzzy approximate solution (y(x))

1.6 0.8
1.4 0.7
1.2 0.6
r-cut

1 0.5

0.8 0.4
0.3
0.6
0.2
0.4
0 0.1
0.5
x 0
1 0.8 1 1.15 1.2 1.25 1.3 1.35 1.4 1.45 1.5 1.55 1.6 1.65
0 0.2 0.4 0.6
r-cut y(1; r)

Figure 7: The fuzzy approximate solution of Example 57, 𝛼 = Figure 9: The approximate solution for 𝛼 = 0.75(∗), 0.85(∘),
0.95, 𝑚 = 6. and 0.95(+) of Example 57, 𝑚 = 8.

in which 𝑦(𝑥) : 𝐿E [0, 1] ∩ 𝐶E [0, 1] is a continuous fuzzy-


̃ 𝑟)
Remark 59. Figure 6 illustrates the absolute errors for 𝑦(𝑥; valued function and 𝑐 𝐷0𝛼+ denotes the fuzzy Caputo frac-
̃ 𝑟),
of Example 57. In the same way, if we consider for 𝑦(𝑥; tional derivative of order 𝛼.
then analogously to the demonstration of Figure 6, we can
obtain the absolute errors. Now, using [1 − 𝛼]-differentiability and Theorem 33, we
have the following:

2𝑥2−𝛼 𝑥1−𝛼
Example 60. For our third example, consider the inhomoge- (𝑐 𝐷0+
𝛼
𝑦) (𝑥; 𝑟) + 𝑦 (𝑥; 𝑟) = − + 𝑥2 − 𝑥,
neous linear equation in [78] with fuzzy initial values, and so Γ (3 − 𝛼) Γ (2 − 𝛼)
we have:
𝑦 (0; 𝑟) = −1 + 𝑟, 0 < 𝛼 ≤ 1, 0 ≤ 𝑥 ≤ 1,

𝑐 𝛼 2𝑥2−𝛼 𝑥1−𝛼 𝛼 2𝑥2−𝛼 𝑥1−𝛼


𝐷0+ 𝑦 (𝑥) + 𝑦 (𝑥) = − + 𝑥2 − 𝑥, ( 𝑐 𝐷0+ 𝑦) (𝑥; 𝑟) + 𝑦 (𝑥; 𝑟) = − + 𝑥2 − 𝑥,
Γ (3 − 𝛼) Γ (2 − 𝛼) Γ (3 − 𝛼) Γ (2 − 𝛼)
𝑦 (0; 𝑟) = [−1 + 𝑟, 1 − 𝑟] , 0 < 𝛼 ≤ 1, 0 ≤ 𝑥 ≤ 1, 𝑦 (0; 𝑟) = 1 − 𝑟, 0 < 𝛼 ≤ 1, 0 ≤ 𝑥 ≤ 1.
(154) (155)
20 Abstract and Applied Analysis

Solving (155) leads to determining the solution of FFDE (154) 100


as follows:
𝑥
𝑦 (𝑥; 𝑟) = (−1 + 𝑟) 𝐸𝛼,1 [−𝑥𝛼 ] + ∫ (𝑥 − 𝑡)𝛼−1 𝐸𝛼,𝛼 [−(𝑥 − 𝑡)𝛼 ]
0 10−5

2𝑥2−𝛼 𝑥1−𝛼

Absolute error
×( − + 𝑥2 − 𝑥) 𝑑𝑡,
Γ (3 − 𝛼) Γ (2 − 𝛼) 10−10

0 ≤ 𝑟 ≤ 1,
𝑥 10−15
𝑦 (𝑥; 𝑟) = (1 − 𝑟) 𝐸𝛼,1 [−𝑥𝛼 ] + ∫ (𝑥 − 𝑡)𝛼−1 𝐸𝛼,𝛼 [−(𝑥 − 𝑡)𝛼 ]
0
2−𝛼 1−𝛼
2𝑥 𝑥 10−20
×( − + 𝑥2 − 𝑥) 𝑑𝑡, 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Γ (3 − 𝛼) Γ (2 − 𝛼)
r-cut
0 ≤ 𝑟 ≤ 1.
𝛼 = 0.65 𝛼 = 0.85
(156) 𝛼 = 0.75 𝛼 = 0.95
By exploiting the method proposed in Section 5, the Figure 10: The absolute error for different 𝛼 of Example 60, 𝑚 = 5.
equations are acquired by
𝑚 𝑚
∑ 𝑐𝑗𝑟 [𝑑𝑖,𝑗
(𝛼)
+ 𝐼] 𝐿 𝑗 (𝑥; 𝑟) = ∑ 𝑓𝑗𝑟 𝐿 𝑗 (𝑥; 𝑟) 𝑖 = 0, . . . , 𝑚,
𝑗=0 𝑗=0 The approximate and exact solution of FFDEs (154) is
illustrated in Table 4 at 𝑥 = 1 with the absolute errors. In
𝑚 𝑚
𝑟 this table, three shifted Legendre polynomials are considered
∑𝑐𝑟 𝑗 [𝑑𝑖,𝑗
(𝛼)
+ 𝐼] 𝐿 𝑗 (𝑥; 𝑟) = ∑ 𝑓 𝑗 𝐿 𝑗 (𝑥; 𝑟) 𝑖 = 0, . . . , 𝑚, to derive the approximate solution. It is clear that with the
𝑗=0 𝑗=0
lower number of shifted Legendre functions, using proposed
(157) method, one can reach a resealable approximate solution.
or we can rewrite it in the matrix form as Also Figure 10 depicts the absolute error with 𝑚 = 5 and
𝑇 various values of 𝛼. This figure exhibits the applicability
𝐶𝑚,− [𝐷(𝛼) + 𝐼] Φ (𝑥) = 𝐹𝑚,−
𝑇
Φ (𝑥) ,
and validity of the proposed technique for this example.
(158) Furthermore, the fuzzy approximate solution for 𝛼 = 0.75
𝑇
𝐶𝑚,+ [𝐷(𝛼) + 𝐼] Φ (𝑥) = 𝐹𝑚,+
𝑇
Φ (𝑥) , with 𝑚 = 5 is shown in Figure 11 at 𝑥 = 1. Additionally,
in which (𝑐𝑗𝑟 ) = [𝑐𝑗,− 𝑟 𝑟
, 𝑐𝑗,+ ] and (𝑓𝑗𝑟 ) = [𝑓𝑗,−
𝑟 𝑟
, 𝑓𝑗,+ ] for 𝑗 = in Figure 12, the fuzzy approximate solution is compared
0, 1, . . . , 𝑚. As it was described in Section 5, we create 𝑚 fuzzy with different values of 𝛼. It can be seen that as 𝛼 tends to
algebraic equations multiplied in above system by 𝐿 𝑖 (𝑥) for 1, the solution of the fuzzy fractional differential equations
𝑖 = 0, 1, . . . , 𝑚−1 and implemented in the inner product using tends to that of the fuzzy integer-order differential equations.
orthogonal property, and so we gain: Finally, Figure 13 displays the absolute error for 𝛼 = 0.75 with
different values of 𝑚 which is obvious that with increasing the
𝑇
𝐶𝑚,− [𝐷(𝛼) + 𝐼] = 𝐹𝑚,−
𝑇
, number of shifted Legendre functions, the absolute error of
(159) the problem has been decreased gently.
𝑇
𝐶𝑚,+ [𝐷(𝛼) + 𝐼] = 𝐹𝑚,+
𝑇
.
Remark 61. Figures 10 and 13 illustrate the absolute errors for
Also for the initial condition (154), we have ̃ 𝑟) of Example 60. In the same way, if we consider 𝑦(𝑥;
𝑦(𝑥; ̃ 𝑟),
𝑇
𝑦 (0; 𝑟) ≃ 𝐶𝑚,− Φ𝑚 = −1 + 𝑟, then analogously to the demonstration of Figures 10 and 13,
(160) we can obtain the absolute errors.
𝑇
𝑦 (0; 𝑟) ≃ 𝐶𝑚,+ Φ𝑚 = 1 − 𝑟.
Now, we consider two applicable examples which are new
Equations (159) and (160) produce a set of 𝑚 + 1 fuzzy under uncertainty represented by fuzzy-valued functions.
linear algebraic equations. As a result, for 𝛼 = 0.75𝑟 = Firstly, a brief history is given in the deterministic case of
0.1 and 𝑚 = 3 the unknown coefficients 𝑐𝑗 can be Example 62. For more details, see [1, 79].
𝑇
achieved as 𝐶3,− = [−0.6914, 0.3147, 0.1144, 0.0083] and There are two kinds of electrical circuits which are related
𝑇
𝐶3,+ = [0.3725, −0.1418, 0.2954, −0.0903]. Also with these to the fractional calculus. Circuits of the first types are
assumptions, one has supposed to consist of capacitors and resistors, which are
described by conventional (integer-order) models. A circuit
0 0 0 0 expressing fractional order behavior is called fractance. Cir-
[ 1.7652 0.5884 −0.2263 0.1305 ]
𝐷 (0.75)
=[ ]
[−0.5884 3.6662 1.2114 −0.4979] . (161)
cuits of the second type may consist of resistors, capacitors,
and fractances. The term fractance was suggested initially
[ 1.5389 −0.6230 5.1081 1.8265 ] by Mehaute and Crepy [80]. Now, in order to scrutinize the
Abstract and Applied Analysis 21

Table 4: The result of the proposed method for Example 60 with 𝛼 = 0.95 and 𝑚 = 3.

𝑟 𝑦3𝑟 𝑌(𝑟) Error 𝑦3𝑟 𝑌(𝑟) Error


0 −0.371581 −0.371573 7.857218𝑒 − 5 0.371581 0.371573 7.857218𝑒 − 5
0.1 −0.334423 −0.334416 7.071496𝑒 − 5 0.334423 0.334416 7.071496𝑒 − 5
0.2 −0.297265 −0.297258 6.285775𝑒 − 5 0.297265 0.297258 6.285775𝑒 − 5
0.3 −0.260107 −0.260101 5.500053𝑒 − 5 0.260107 0.260101 5.500053𝑒 − 5
0.4 −0.222948 −0.222944 4.714331𝑒 − 5 0.222948 0.222944 4.714331𝑒 − 5
0.5 −0.185790 −0.185786 3.928609𝑒 − 5 0.185790 0.185786 3.928609𝑒 − 5
0.6 −0.148632 −0.148629 3.142887𝑒 − 5 0.148632 0.148629 3.142887𝑒 − 5
0.7 −0.111474 −0.111472 2.357165𝑒 − 5 0.111474 0.111472 2.357165𝑒 − 5
0.8 −0.074316 −0.074314 1.571443𝑒 − 6 0.074316 0.074314 1.571443𝑒 − 6
0.9 −0.037158 −0.037157 7.857218𝑒 − 6 0.037158 0.037157 7.857218𝑒 − 6
1 0.000000 0.000000 9.520160𝑒 − 14 0.000000 0.000000 9.520160𝑒 − 14

1
Fuzzy approximate solution (y(x))

1
0.9
0.5 0.8
0.7
0
0.6
r-cut

0.5
−0.5
0.4
−1 0.3
1
0.2
0.5
x 0.1
0.8 1
0 0.4 0.6
0 0.2 0
r-cut −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4
Figure 11: The fuzzy approximate solution of Example 60, 𝛼 = y(1; r)
0.75, 𝑚 = 5. Figure 12: The fuzzy approximate solution for different 𝛼:(1: star,
0.95: square, 0.85: ∘, 0.75: ×, 0.65: pentagram) of Example 60, 𝑚 = 5.

mentioned problem in a real case, we use the fuzzy initial


value 𝑦(0; 𝑟) and the concept of Caputo’s H-differentiability
for fractional derivative of 𝑦(𝑥). So we have the following 10−2
fuzzy fractional oscillation differential equation.
10−4
Example 62. Consider an electrical circuit (LR circuit) with
10−6
an AC source. The current equation of this circuit can be
written as follows: 10−8
Absolute error

10−10
𝑐 𝛼 𝑅
( 𝐷0+ 𝑦) (𝑥) = − 𝑦 (𝑥) + V (𝑥) , 0 ≤ 𝑥 ≤ 1,
𝐿 (162) 10−12

𝑦 (0; 𝑟) = [0.96 + 0.04𝑟, 1.01 + 0.01𝑟] , 10−14

10−16
in which 𝑅 is the circuit resistance, and 𝐿 is a coefficient,
corresponding to the solenoid and 0 ≤ 𝑟 ≤ 1. Suppose that 10−18
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
V(𝑥) = sin(𝑥), 𝑅 = 1 ohm and 𝐿 = 1H, so (162) can be
r-cut
rewritten as
m=5 m = 14
𝑐 m = 10 m=3
( 𝐷0𝛼+ 𝑦) (𝑥) = −𝑦 (𝑥) + sin (𝑥) , 0 ≤ 𝑥 ≤ 1,
(163) Figure 13: The absolute error for different 𝑚 of Example 60, 𝛼 =
𝑦 (0; 𝑟) = [0.96 + 0.04𝑟, 1.01 + 0.01𝑟] . 0.75.
22 Abstract and Applied Analysis

Table 5: The result of the proposed method for Example 62 with 𝛼 = 0.85 and 𝑚 = 8.

𝑟 𝑦8𝑟 𝑌(𝑟) Error 𝑦8𝑟 𝑌(𝑟) Error


0 0.7208 0.7206 1.4374𝑒 − 4 0.7398 0.7397 1.5112𝑒 − 4
0.1 0.7223 0.7221 1.4433𝑒 − 4 0.7394 0.7393 1.5098𝑒 − 4
0.2 0.7238 0.7237 1.4492𝑒 − 4 0.7391 0.7389 1.5083𝑒 − 4
0.3 0.7253 0.7252 1.4551𝑒 − 4 0.7387 0.7385 1.5068𝑒 − 4
0.4 0.7269 0.7267 1.4610𝑒 − 4 0.7383 0.7381 1.5053𝑒 − 4
0.5 0.7284 0.7282 1.4669𝑒 − 4 0.7379 0.7378 1.5039𝑒 − 4
0.6 0.7299 0.7298 1.4729𝑒 − 4 0.7375 0.7374 1.5024𝑒 − 4
0.7 0.7314 0.7313 1.4788𝑒 − 4 0.7372 0.7370 1.5009𝑒 − 4
0.8 0.7330 0.7328 1.4847𝑒 − 4 0.7368 0.7366 1.4994𝑒 − 4
0.9 0.7345 0.7343 1.4906𝑒 − 4 0.7364 0.7362 1.4980𝑒 − 4
1 0.7360 0.7359 1.4965𝑒 − 4 0.7360 0.7359 1.4965𝑒 − 4

Now, using [2 − 𝛼]-differentiability and Theorem 33, we conditions of (164), we can obtain the unknown fuzzy
have the following: coefficients {𝑐𝑗 }3𝑗=0 as

𝛼 𝑐0 = 0.7222, 𝑐1 = −0.0497,
( 𝑐 𝐷0+ 𝑦) (𝑥; 𝑟) + 𝑦 (𝑥; 𝑟) = sin (𝑥) ,
𝑐2 = 0.1272, 𝑐3 = −0.0650,
𝑦 (0; 𝑟) = (0.96 + 0.04𝑟) , 0 < 𝛼 ≤ 1, 0 ≤ 𝑥 ≤ 1, (168)
(164) 𝑐0 = 0.7488, 𝑐1 = −0.0611,
𝑐 𝛼
( 𝐷0+ 𝑦) (𝑥; 𝑟) + 𝑦 (𝑥; 𝑟) = sin (𝑥) ,
𝑐2 = 0.1317, 𝑐3 = −0.0674,
𝑦 (0; 𝑟) = (1.01 + 0.01𝑟) , 0 < 𝛼 ≤ 1, 0 ≤ 𝑥 ≤ 1.
and finally, the fuzzy approximate solution is given by
The exact solution of (163) under (2)-differentiability for
integer order is given by 𝑦 (𝑥, 0.1) = 0.9640 − 1.6420𝑥 + 2.7118𝑥2 − 1.2990𝑥3 ,
4

1 𝑦4 (𝑥, 0.1) = 1.0090 − 1.7215𝑥 + 2.8128𝑥2 − 1.3483𝑥3 .


𝑌1 (𝑡; 𝑟) = (sin (𝑡) − cos (𝑡)) + 𝑒−𝑡 (1.46 + 0.04r) , (169)
2
(165)
1 In Table 5, the numerical results of (162) are depicted at
𝑌1 (𝑡; 𝑟) = (sin (𝑡) − cos (𝑡)) + 𝑒−𝑡 (1.51 − 0.01r) .
2 𝑥 = 1 for 𝛼 = 0.85 and 𝑚 = 8. The absolute error
confirms that the proposed method approximates the fuzzy
By applying the technique described in Section 5, we solution with a suitable accuracy. Figure 14 shows the absolute
approximate fuzzy solution as error of the proposed method with a different value of 𝛼
with 𝑚 = 10 which is clear as 𝛼 tends to 1, and the error
𝑐 is decreasing dramatically, also with the increasing of the
𝐷𝛼 𝑦 (𝑥) ≃ 𝐶𝑇 ⊙ 𝐷(𝛼) Φ𝑚 (𝑥) ,
number of Legendre functions, the absolute error is dereasing
𝑇
𝑦 (𝑥) ≃ 𝑦̃𝑚 (𝑥) = 𝐶𝑚 ⊙ Φ𝑚 , (166) that is, shown in Figure 16. Finally the fuzzy approximate
solution of (162) is demonstrated in Figure 15 in the interval
sin (𝑥) ≃ 𝑓̃𝑚 (𝑥) = 𝐹𝑚𝑇 ⊙ Φ𝑚 , 0 ≤ 𝑥 ≤ 1.
Now we consider the second application of FFDES in
the real world. But before we consider the problem of fuzzy
where vector 𝐹𝑚𝑇 is obtained as (102). With 𝑚 = 3, 𝛼 = 0.75, model, We glimpse the vision of a non-fuzzy case.
and 𝑟 = 0.1, we have A pharmacodynamic model is usually separated in two
parts: a link model and a transduction model. The link model
0 0 0 0 depicts the distribution of drug from an observed compart-
[ 1.7652 0.5884 −0.2263 0.1305 ]
𝐷(0.75) =[ ]
[−0.5884 3.6662 1.2114 −0.4979] ,
ment (e.g., plasma) into a biophase (the effect compartment).
(167) Drug in the biophase induces a pharmacodynamic response
[ 1.5389 −0.6230 5.1081 1.8265 ] (the effect), which is represented by a transduction model. For
more details, see [81, 82].
𝐹4𝑇 = [0.4597 0.4279 −0.0392 −0.0072] , The model describing the transfer of drug to the effect
compartment, from the mechanistic interpretation, it can be
and subsequently, we generate 𝑚 + 1 fuzzy linear equations considered as the pharmacokinetics of the drug. However,
using (110) and also from substituting (100) in the initial practically, and because drug condensation in the effect
Abstract and Applied Analysis 23

10−2 compartment does not appears, it is not often clear if the link
model is actually linked with a site of action, or if it exhibits
some pharmacodynamic related delay, or a combination of
10−3 the two. It is begun by representing drug concentration in
the effect compartment by the (Caputo) fractional differential
Absolute error

equation as follows:
10−4
(𝛼 𝐷0+
𝛼
𝐶𝑒) (𝑡) = 𝑘eo (−𝐶𝑒 (𝑡) + 𝑓 (𝑡)) , (170)

in which 𝑓(𝑡) describes the drug input into the central


10−5 compartment and 𝑘eo is the elimination rate constant from
the effect compartment. Also the action of drug is described
by a nonlinear memory-less function of 𝐶𝑒.
10−6 Now, we consider the action of drug delivery by a fuzzy
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 function under the fuzzy Caputo’s H-differentiability which
r-cut is modelled by fuzzy fractional differential equation.
𝛼 = 0.75 𝛼 = 0.85
𝛼 = 0.65 𝛼 = 0.95 Example 63. Consider the following fuzzy pharmacody-
namic model
Figure 14: The absolute error for different 𝛼 of Example 62, 𝑚 = 10.
(𝛼 𝐷0+
𝛼
𝐶𝑒) (𝑡) = 𝑘eo (−𝐶𝑒 (𝑡) + 𝑓 (𝑡)) , 0 < 𝛼 < 1, 0 ≤ 𝑡 ≤ 1,
𝐶𝑒 (0; 𝑟) = (0.5 + 0.5𝑟, 1.5 − 0.5𝑟) ,
Fuzzy approximate solution (y(x))

(171)
1.1
1 in which 𝑓(𝑡) = (Dose 𝑘𝑎 /𝑉(𝑘el −𝑘𝑎 ))(𝑒−𝑘el 𝑡 −𝑒−𝑘𝑎 𝑡 ) represents
0.9 drug concentration in the central compartment following, for
0.8 example, an oral administration of a dose, in this example
0.7 dose = 1 and parameters values are 𝑉 = 1, 𝑘𝑎 = 5.0, 𝑘el =
0.5, and 𝑘eo = 1.
1
0.8 The exact solution of (171) with 𝛼 = 1 under generalized
0.6 1
differentiability is given by
0.4 0.8
x 0.6
0.2 0.4 20 𝑡/2 5 −4𝑡
0 0
0.2 r-cut 𝑌 (𝑡; 𝑟) = 𝑒−𝑡 (− 𝑒 − 𝑒 + 3 + 0.5𝑟) ,
9 18
(172)
Figure 15: The fuzzy approximate solution of Example 62, 𝛼 = 20 5
0.75, 𝑚 = 10. 𝑌 (𝑡; 𝑟) = 𝑒 (− 𝑒𝑡/2 − 𝑒−4𝑡 + 4 − 0.5𝑟) .
−𝑡
9 18
We apply the method presented in Section 5 and solve this
10−2
problem for 𝑚 = 3, 𝛼 = 0.95, and 𝑟 = 0.2. Hence, we have

0 0 0 0
[ 1.9566 0.1432 −0.0743 0.0501 ]
𝐷 (0.95)
=[ ]
[−0.1432 5.4849 1.3447 −0.1823] ,
10−3
Absolute error

(173)
[ 1.8823 −0.2015 8.8258 0.5752 ]
𝐹4𝑇 = [−0.6537 −0.1886 0.2729 −0.1322] ,
10−4
and thereafter the fuzzy unknown coefficients for the fuzzy
approximation of (171) can be acquired easily by replacing the
above results in (110) and solving this algebraic fuzzy linear
equations system. So we have
10−5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
𝑐0 = 0.1744, 𝑐1 = −0.4368,
r-cut
m=3 m=9 𝑐2 = 0.0573, 𝑐3 = 0.0185,
m=6 m = 12 (174)
𝑐0 = 0.6417, 𝑐1 = −0.6635,
Figure 16: The absolute error for different 𝑚 of Example 62, 𝛼 =
0.85. 𝑐2 = 0.1043, 𝑐3 = 0.0095,
24 Abstract and Applied Analysis

Table 6: The result of the proposed method for Example 63 with 𝛼 = 0.95 and 𝑚 = 12.
𝑟
𝑟 𝑦1 2 𝑌(𝑟) Error 𝑦1 2𝑟 𝑌(𝑟) Error
0 −0.243121 −0.243123 1.232190𝑒 − 6 0.128454 0.128450 3.891957𝑒 − 6
0.1 −0.224543 −0.224544 1.365176𝑒 − 6 0.109875 0.109871 3.758969𝑒 − 6
0.2 −0.205964 −0.205965 1.498166𝑒 − 6 0.091296 0.091293 3.625981𝑒 − 6
0.3 −0.187385 −0.187387 1.631153𝑒 − 6 0.072717 0.072714 3.492989𝑒 − 6
0.4 −0.168806 −0.168808 1.764141𝑒 − 6 0.054139 0.054135 3.360001𝑒 − 6
0.5 −0.150227 −0.150229 1.897131𝑒 − 6 0.035560 0.035557 3.227013𝑒 − 6
0.6 −0.131648 −0.131651 2.030121𝑒 − 6 0.016981 0.016978 3.094025𝑒 − 6
0.7 −0.113070 −0.113072 2.163107𝑒 − 6 −0.001597 −0.001600 2.961033𝑒 − 6
0.8 −0.094491 −0.094493 2.296095𝑒 − 6 −0.020176 −0.020178 2.828050𝑒 − 6
0.9 −0.075912 −0.075914 2.429084𝑒 − 6 −0.038754 −0.038757 2.695061𝑒 − 6
1 −0.057333 −0.057336 2.562075𝑒 − 6 −0.057333 −0.057336 2.562075𝑒 − 6

and ultimately the fuzzy approximate solution is given by 10−2

𝑦 (𝑥, 0.1) = 0.6 − 0.9392𝑥 − 2.2482𝑥2 + 0.3823𝑥3 , 10−3


4
(175)
Absolute error

𝑦4 (𝑥, 0.1) = 1.4 − 1.8391𝑥 + 0.3419𝑥2 + 0.1891𝑥3 .


10−4

In Table 6, the approximate fuzzy solutions (171) are com-


pared with the exact solution at 𝑥 = 1 for 𝛼 = 0.95 10−5
and 𝑚 = 12. The absolute error shows that the proposed
method approximates the fuzzy solution with a high accu-
racy. Moreover, Figure 17 describes the absolute error of the 10−6
proposed method with a different value of 𝛼 with 𝑚 = 8. It 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
is obvious that the method reaches a good approximation as r-cut
𝛼 approaches 1; also again with the increasing of the number 𝛼 = 0.65 𝛼 = 0.85
of Legendre functions, the absolute error is decreasing that 𝛼 = 0.75 𝛼 = 0.95
is, shown in Figure 19. The fuzzy approximate solution of the
Figure 17: The absolute error for different 𝛼 of Example 63, 𝑚 = 8.
problem is displayed in Figure 18 in the interval 0 ≤ 𝑥 ≤ 1.

Example 64. Consider the following initial value problem of


𝑚 𝑚 2
nonlinear FFDE:
∑ 𝑐𝑗𝑟 𝑑𝑖,𝑗
(𝛼)
𝐿 𝑗 (𝑥; 𝑟) − 3 (3 − 𝑟) ∑ (𝑐𝑗𝑟 𝐿 𝑗 (𝑥; 𝑟)) = 0
𝑗=0 𝑗=0

𝛼
( 𝑐 𝐷0+ 𝑦) (𝑥) = 3𝐴 ⊙ 𝑦2 , 0 < 𝛼 < 1, 0 ≤ 𝑥 ≤ 1, 𝑖 = 0, . . . , 𝑚,
(176) (177)
𝑚 𝑚 2
𝑦 (0; 𝑟) = [0.5√𝑟, 0.2√1 − 𝑟 + 0.5] , ∑ 𝑐𝑗𝑟 𝑑𝑖,𝑗
(𝛼)
𝐿𝑗 (𝑥; 𝑟) − 3 (1 + 𝑟) ∑ (𝑐𝑗𝑟 𝐿 𝑗 (𝑥; 𝑟)) = 0
𝑗=0 𝑗=0

𝑖 = 0, . . . , 𝑚,
in which 𝑦(𝑥) : 𝐿E [0, 1] ∩ 𝐶E [0, 1] is a continuous fuzzy-
valued function and 𝑐 𝐷0𝛼+ denotes the fuzzy Caputo frac- or we can rewrite it in the matrix form as
tional derivative of order 𝛼. Also 𝐴 = [1+𝑟, 3−𝑟] is a constant
fuzzy number. 𝑇 2
𝐶𝑚,− 𝐷(𝛼) Φ (𝑥) − 3 (3 − 𝑟) [𝐶𝑚,−
𝑇
Φ (𝑥)] = 0,
For approximating the fuzzy solution of (162), using the (178)
𝑇 2
method described in Sections 4 and 5, if we approximate the 𝐶𝑚,+ 𝐷(𝛼) Φ (𝑥) − 3 (1 + 𝑟) [𝐶𝑚,+
𝑇
Φ (𝑥)] = 0,
solution by 𝑚 shifted Legendre functions, then it needs to
consider the first 𝑚 roots of the shifted Legendre polynomial in which (𝑐𝑗𝑟 ) = [𝑐𝑗,−
𝑟 𝑟
, 𝑐𝑗,+ ] and for 𝑗 = 0, 1, . . . , 𝑚. As it
𝐿 𝑚+1 (𝑥). Initially using (162), we have was described in Section 5, we produce 𝑚 fuzzy algebraic
Abstract and Applied Analysis 25

1
Fuzzy approximate solution (y(x))

1.5 0.9

1 0.8
0.7
0.5
0.6
0

r-cut
0.5
−0.5 0.4
1
0.8 0.3
0.6
0.2
0.4
x 1
0.2 0.8 0.1
0.6
0.4
0 0.2
0 r-cut 0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Figure 18: The fuzzy approximate solution of Example 63, 𝛼 = y(1; r)
0.95, 𝑚 = 8.
m=3
m=5
m=7
10−3
Figure 20: The approximate solution for different 𝑚 of Example 64,
𝛼 = 0.75.
10−4
Absolute error

Approximate solution y(x; r))

10−5 1
0.8
0.6
10−6
0.4
0.2
10−7 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1
1 0.8
r-cut 0.8 0.6
0.6 0.4
0.4 t
m=3 m=9 x 0.2
0 0
0.2 r-cu
m=6 m = 12
Figure 21: The fuzzy approximate solution of Example 64, 𝛼 =
Figure 19: The absolute error for different 𝑚 of Example 63, 𝛼 = 0.75, 𝑚 = 3.
0.95.

which for 𝛼 = 0.95 and 𝑟 = 0.1 are


equations by collecting (162) in the first 𝑚 root of 𝐿 𝑚+1 (𝑥). 𝑐0 = 0.2593, 𝑐1 = 0.1106,
Also using the initial condition (162), we have
𝑐2 = −0.0032, 𝑐3 = 0.0054,
𝑇
(181)
𝑦 (0; 𝑟) ≃ 𝐶𝑚,− Φ𝑚 (0) = (0.5√𝑟) , 𝑐0 = 0.1171, 𝑐1 = −0.0203,
(179)
𝑇
𝑦 (0; 𝑟) ≃ 𝐶𝑚,+ Φ𝑚 (0) = (0.2√1 − 𝑟 + 0.5) . 𝑐2 = 0.0045, 𝑐3 = 0.0155.

Finally, using the above results and (100), the approximate


These roots can be put in the (178) and deriving three solution is computed in each particular point easily.
nonlinear fuzzy equation, then these equations be coupled The approximate solution for Example 64 with different
with (179). Finally with solving the fuzzy nonlinear equations fractional Caputo order is derived in Table 7. The solution
system, the unknown coefficients 𝑐̃𝑗 are obtained. has been estimated using 3 shifted Legendre functions. The
Now, as an explanation, assume that 𝑚 = 3, then the first results in this table are at 𝑥 = 1. It is clear that the method
three roots of 𝐿 4 (𝑥) are as follows: is applicable and valid for different fuzzy fractional order 𝛼.
Furthermore, by changing the number of shifted Legendre
functions, the approximate solution for 𝛼 = 0.75 at 𝑥 =
𝑥0 = 0.0694, 𝑥1 = 0.9306, 𝑥2 = 0.3300, (180) 1 is revealed in Figure 20 which shows that with only a
26 Abstract and Applied Analysis

Table 7: The approximate solution of the proposed method for Example 64 at 𝑥 = 1 with 𝑚 = 3.
𝛼=0.5 𝑟 𝛼=0.75 𝑟 𝛼=0.85 𝑟 𝛼=0.5 𝑟 𝛼=0.75 𝑟 𝛼=0.85 𝑟
𝑟 𝑦̃3 𝑦̃3 𝑦̃3 𝑦̃3 𝑦̃3 𝑦̃3
0 0.1402 0.1019 0.1040 0.8224 0.8856 0.9381
0.1 0.1434 0.1053 0.1080 0.3061 0.3250 0.3456
0.2 0.1468 0.1091 0.1123 0.2769 0.2917 0.3111
0.3 0.1503 0.1131 0.1170 0.2539 0.2646 0.2826
0.4 0.1539 0.1175 0.1222 0.2356 0.2418 0.2586
0.5 0.1578 0.1224 0.1278 0.2209 0.2223 0.2380
0.6 0.1618 0.1278 0.1340 0.2087 0.2056 0.2202
0.7 0.1662 0.1328 0.1409 0.1986 0.1912 0.2046
0.8 0.1706 0.1388 0.1485 0.1908 0.1785 0.1909
0.9 0.1752 0.1455 0.1571 0.1838 0.1674 0.1788
1 0.1778 0.1576 0.1680 0.1778 0.1576 0.1680

few number of shifted Legendre polynomials, desired results For future research, we will consider this method for
are available. Ultimately, the fuzzy approximate solution is solving FFDEs with order 1 < 𝛼 < 2. Also we will apply
described in Figure 21 with 𝛼 = 0.75 and 𝑚 = 3. it under Riemann-Liouville’s H-differentiability. Apart from
this, the other orthogonal functions like Jacobi polynomials
Remark 65. It is consequential to note that, for a crisp differ- will be extended for solving FFDEs.
ential equation of integer or fractional order, the problem

(𝑐 𝐷0𝛼+ 𝑦) (𝑥) + 𝐹 (𝑥, 𝑦 (𝑥)) = 0, 𝑥 ∈ [0, 𝑋] , (182) Acknowledgments


is the same as The authors would like to thank Professor Dumitru Baleanu,
the Leader Guest Editor, for his consideration about this
(𝑐 𝐷0𝛼+ 𝑦) (𝑥) = −𝐹 (𝑥, 𝑦 (𝑥)) , 𝑥 ∈ [0, 𝑋] , (183) paper and the anonymous reviewers for their helpful sugges-
tions and comments.
in which is 𝐹 is nonlinear operator, and in the case of space
E, both problems are not equivalent [83].
References
7. Conclusion [1] I. Podlubny, Fractional Differential Equations, vol. 198 of Math-
ematics in Science and Engineering, Academic Press, San Diego,
In this paper, a cluster of orthogonal functions named Calif, USA, 1999.
shifted Legendre function is used to solve fuzzy fractional [2] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional
differential equations under Caputo type. The benefit of the Calculus. Models and Numerical Methods, vol. 3 of Series on
shifted Legendre operational matrices method over other Complexity, Nonlinearity and Chaos, World Scientific Publish-
existing orthogonal polynomials is its simplicity of execution ing, Hackensack, NJ, USA, 2012.
besides some other advantages. A general formulation of the [3] K. Diethelm, The Analysis of Fractional Differential Equations.
proposed method is provided in details. Also a complete error An Application-Oriented Exposition Using Differential Operators
analysis is considered in Section 5. In this paper, we tried to of Caputo Type, vol. 2004 of Lecture Notes in Mathematics,
answer some questions: Springer, Berlin, Germany, 2010.
[4] D. Baleanu, Z. B. Güvenc, and J. A. Tenreiro Machado, New
(i) how to derive a shifted Legendre operational matrice Trends in Nanotechnology and Fractional Calculus Applications,
method over the fuzzy fractional integration and Springer, New York, NY, USA, 2010.
differentiation [5] J. Sabatier, O. P. Agrawal, and J. A. Tenreiro Machado, Advances
(ii) how to solve the fuzzy fractional order differential in Fractional Calculus: Theoretical Developments and Appli-
equations via the shifted Legendre operational matri- cations in Physics and Engineering, Springer, Dordrecht, The
ces of the fuzzy fractional derivative. Netherlands, 2007.
[6] J. He, “Nonlinear oscillation with fractional derivative and
Moreover, it should be mentioned that another advantage its applications,” in Proceedings of International Conference on
of this technique is that it decreases these problems to the Vibrating Engineering, pp. 288–291, Dalian, China, 1998.
degree of solving a system of fuzzy algebraic equations thus [7] S. Momani and N. Shawagfeh, “Decomposition method for
extremely making the problems easier. Several examples was solving fractional Riccati differential equations,” Applied Math-
carried out to depict the effectiveness and the absence of ematics and Computation, vol. 182, no. 2, pp. 1083–1092, 2006.
complexity of the proposed method. The achieved solutions [8] F. Mainardi, “Fractional calculus: some basic problems in
have a satisfactory results obtained with only a small number continuum and statistical mechanics,” in Fractals and Fractional
of Legendre polynomials. Calculus in Continuum Mechanics (Udine, 1996), A. Carpinteri
Abstract and Applied Analysis 27

and F. Mainardi, Eds., vol. 378 of CISM Courses and Lectures, [25] M. Barkhordari Ahmadi and M. Khezerloo, “Fuzzy bivariate
pp. 291–348, Springer, Vienna, Austria, 1997. Chebyshev method for solving fuzzy Volterra-Fredholm inte-
[9] Y. A. Rossikhin and M. V. Shitikova, “Applications of fractional gral equations,” International Journal of Industrial Mathematics,
calculus to dynamic problems of linear and nonlinear heredi- vol. 3, no. 2, pp. 67–77, 2011.
tary mechanics of solids,” Applied Mechanics Reviews, vol. 50, [26] H. S. Goghary and M. S. Goghary, “Two computational meth-
no. 1, pp. 15–67, 1997. ods for solving linear Fredholm fuzzy integral equation of the
[10] S.-D. Lin, W.-C. Ling, K. Nishimoto, and H. M. Srivastava, second kind,” Applied Mathematics and Computation, vol. 182,
“A simple fractional-calculus approach to the solutions of the no. 1, pp. 791–796, 2006.
Bessel differential equation of general order and some of its [27] T. Lotfi and K. Mahdiani, “Fuzzy Galerkin method for solving
applications,” Computers & Mathematics with Applications, vol. Fredholm integral equations with error analysis,” International
49, no. 9-10, pp. 1487–1498, 2005. Journal of Industrial Mathematics, vol. 3, no. 4, pp. 237–249, 2011.
[11] P.-Y. Wang, S.-D. Lin, and H. M. Srivastava, “Remarks on [28] R. P. Agarwal, V. Lakshmikantham, and J. J. Nieto, “On the
a simple fractional-calculus approach to the solutions of the concept of solution for fractional differential equations with
Bessel differential equation of general order and some of its uncertainty,” Nonlinear Analysis. Theory, Methods & Applica-
applications,” Computers & Mathematics with Applications, vol. tions, vol. 72, no. 6, pp. 2859–2862, 2010.
51, no. 1, pp. 105–114, 2006. [29] T. Allahviranloo, S. Salahshour, and S. Abbasbandy, “Explicit
solutions of fractional differential equations with uncertainty,”
[12] Z. Odibat and S. Momani, “The variational iteration method:
Soft Computing, vol. 16, no. 2, pp. 297–302, 2012.
an efficient scheme for handling fractional partial differential
equations in fluid mechanics,” Computers & Mathematics with [30] S. Salahshour, T. Allahviranloo, S. Abbasbandy, and D. Baleanu,
Applications, vol. 58, no. 11-12, pp. 2199–2208, 2009. “Existence and uniqueness results for fractional differential
equations with uncertainty,” Advances in Difference Equations,
[13] J. He, D. Huang, and Z. He, “Study on sedimentary facies vol. 2012, article 112, 2012.
classified by main amplitude parameters of seismic wave and
[31] S. Salahshour, T. Allahviranloo, and S. Abbasbandy, “Solving
their variation characteristics,” Kuangwu Yanshi, vol. 19, no. 4,
fuzzy fractional differential equations by fuzzy Laplace trans-
pp. 86–89, 1999.
forms,” Communications in Nonlinear Science and Numerical
[14] B. Bede and S. G. Gal, “Generalizations of the differentiability Simulation, vol. 17, no. 3, pp. 1372–1381, 2012.
of fuzzy-number-valued functions with applications to fuzzy [32] M. Mazandarani and A. V. Kamyad, “Modified fractional Euler
differential equations,” Fuzzy Sets and Systems, vol. 151, no. 3, method for solving fuzzy fractional initial value problem,” Com-
pp. 581–599, 2005. munications in Nonlinear Science and Numerical Simulation, vol.
[15] B. Bede, “Note on “Numerical solution of fuzzy differential 18, no. 1, pp. 12–21, 2013.
equations by predictor-corrector method”,” Information Sci- [33] S. Momani and M. Aslam Noor, “Numerical methods for
ences. An International Journal, vol. 178, no. 7, pp. 1917–1922, fourth-order fractional integro-differential equations,” Applied
2008. Mathematics and Computation, vol. 182, no. 1, pp. 754–760,
[16] J. J. Buckley and T. Feuring, “Fuzzy differential equations,” Fuzzy 2006.
Sets and Systems, vol. 110, no. 1, pp. 43–54, 2000. [34] S. Momani and Z. Odibat, “Homotopy perturbation method
[17] Y. Chalco-Cano and H. Román-Flores, “On new solutions of for nonlinear partial differential equations of fractional order,”
fuzzy differential equations,” Chaos, Solitons and Fractals, vol. Physics Letters A, vol. 365, no. 5-6, pp. 345–350, 2007.
38, no. 1, pp. 112–119, 2008. [35] E. A. Rawashdeh, “Numerical solution of fractional integro-
[18] O. Kaleva, “Fuzzy differential equations,” Fuzzy Sets and Sys- differential equations by collocation method,” Applied Mathe-
tems, vol. 24, no. 3, pp. 301–317, 1987. matics and Computation, vol. 176, no. 1, pp. 1–6, 2006.
[19] H. Kim and R. Sakthivel, “Numerical solution of hybrid [36] A. Saadatmandi and M. Dehghan, “A new operational matrix
fuzzy differential equations using improved predictor-corrector for solving fractional-order differential equations,” Computers
method,” Communications in Nonlinear Science and Numerical & Mathematics with Applications, vol. 59, no. 3, pp. 1326–1336,
Simulation, vol. 17, no. 10, pp. 3788–3794, 2012. 2010.
[37] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “Efficient
[20] S. Salahshour and T. Allahviranloo, “Applications of fuzzy
Chebyshev spectral methods for solving multi-term fractional
Laplace transforms,” Soft Computing Journal, pp. 1–14, 2012.
orders differential equations,” Applied Mathematical Modelling,
[21] R. Ezzati and S. Ziari, “Numerical solution and error estimation vol. 35, no. 12, pp. 5662–5672, 2011.
of fuzzy Fredholm integral equation using fuzzy Bernstein [38] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A Chebyshev
polynomials,” Australian Journal of Basic and Applied Sciences, spectral method based on operational matrix for initial and
vol. 5, no. 9, pp. 2072–2082, 2011. boundary value problems of fractional order,” Computers &
[22] M. M. M. Otadi, “Numerical solution of fuzzy integral equations Mathematics with Applications, vol. 62, no. 5, pp. 2364–2373,
using Bernstein polynomials,” Australian Journal of Basic and 2011.
Applied Sciences, vol. 5, no. 7, pp. 724–728, 2011. [39] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A new
[23] M. A. F. Araghi and N. Parandin, “Numerical solution of fuzzy Jacobi operational matrix: an application for solving fractional
Fredholm integral equations by the Lagrange interpolation differential equations,” Applied Mathematical Modelling, vol. 36,
based on the extension principle,” Soft Computing, vol. 15, no. no. 10, pp. 4931–4943, 2012.
12, pp. 2449–2456, 2011. [40] R. Goetschel Jr. and W. Voxman, “Elementary fuzzy calculus,”
[24] N. Parandin and M. A. F. Araghi, “The approximate solution Fuzzy Sets and Systems, vol. 18, no. 1, pp. 31–43, 1986.
of linear fuzzy fredholm integral equations of the second kind [41] H.-J. Zimmermann, Fuzzy Set Theory—and Its Applications,
by using iterative interpolation,” World Academy of Science, Kluwer Academic Publishers, Dordrecht, The Netherlands,
Engineering and Technology, vol. 37, pp. 1036–1042, 2009. 1991.
28 Abstract and Applied Analysis

[42] B. Bede and L. Stefanini, “Generalized differentiability of fuzzy- [62] S. Abbasbandy and M. Amirfakhrian, “A new approach to
valued functions,” Fuzzy Sets and Systems. universal approximation of fuzzy functions on a discrete set
[43] L. Stefanini, “A generalization of Hukuhara difference and of points,” Applied Mathematical Modelling, vol. 30, no. 12, pp.
division for interval and fuzzy arithmetic,” Fuzzy Sets and 1525–1534, 2006.
Systems, vol. 161, no. 11, pp. 1564–1584, 2010. [63] S. Abbasbandy and M. Amirfakhrian, “Numerical approx-
[44] B. Bede and S. G. Gal, “Quadrature rules for integrals of fuzzy- imation of fuzzy functions by fuzzy polynomials,” Applied
number-valued functions,” Fuzzy Sets and Systems, vol. 145, no. Mathematics and Computation, vol. 174, no. 2, pp. 1001–1006,
3, pp. 359–380, 2004. 2006.
[45] G. A. Anastassiou, Fuzzy Mathematics: Approximation Theory, [64] S. Abbasbandy and B. Asady, “The nearest trapezoidal fuzzy
vol. 251 of Studies in Fuzziness and Soft Computing, Springer, number to a fuzzy quantity,” Applied Mathematics and Compu-
Berlin, Germany, 2010. tation, vol. 156, no. 2, pp. 381–386, 2004.
[46] G. A. Anastassiou and S. G. Gal, “On a fuzzy trigonometric [65] O. Kaleva, “Interpolation of fuzzy data,” Fuzzy Sets and Systems,
approximation theorem of Weierstrass-type,” Journal of Fuzzy vol. 61, no. 1, pp. 63–70, 1994.
Mathematics, vol. 9, no. 3, pp. 701–708, 2001. [66] R. Lowen, “A fuzzy Lagrange interpolation theorem,” Fuzzy Sets
[47] O. Kaleva, “A note on fuzzy differential equations,” Nonlinear and Systems, vol. 34, no. 1, pp. 33–38, 1990.
Analysis. Theory, Methods & Applications, vol. 64, no. 5, pp. 895– [67] C.-T. Yeh, “Trapezoidal and triangular approximations preserv-
900, 2006. ing the expected interval,” Fuzzy Sets and Systems, vol. 159, no.
[48] J. J. Nieto, A. Khastan, and K. Ivaz, “Numerical solution of 11, pp. 1345–1353, 2008.
fuzzy differential equations under generalized differentiability,” [68] S. Abbasbandy and M. Amirfakhrian, “Best Approximation of
Nonlinear Analysis. Hybrid Systems, vol. 3, no. 4, pp. 700–707, Fuzzy Functions,” Journal of Nonlinear Studies, vol. 14, pp. 88–
2009. 103, 2007.
[49] H.-C. Wu, “The improper fuzzy Riemann integral and its [69] S. Kazem, S. Abbasbandy, and Sunil Kumar, “Fractional-order
numerical integration,” Information Sciences, vol. 111, no. 1–4, pp. Legendre functions for solving fractional order differential
109–137, 1998. equations,” Applied Mathematical Modelling, vol. 37, no. 7, pp.
[50] B. Asady, S. Abbasbandy, and M. Alavi, “Fuzzy general linear 5498–5510, 2013.
systems,” Applied Mathematics and Computation, vol. 169, no. 1, [70] G. A. Anastassiou, “Fuzzy approximation by fuzzy convolution
pp. 34–40, 2005. type operators,” Computers & Mathematics with Applications,
[51] T. Allahviranloo and M. Afshar Kermani, “Solution of a fuzzy vol. 48, no. 9, pp. 1369–1386, 2004.
system of linear equation,” Applied Mathematics and Computa- [71] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang,
tion, vol. 175, no. 1, pp. 519–531, 2006. Spectral Methods in Fluid Dynamics, Springer Series in Com-
[52] M. Dehghan and B. Hashemi, “Iterative solution of fuzzy linear putational Physics, Springer, New York, NY, USA, 1988.
systems,” Applied Mathematics and Computation, vol. 175, no. 1, [72] J. H. Freilich and E. L. Ortiz, “Numerical solution of systems of
pp. 645–674, 2006. ordinary differential equations with the Tau method: an error
[53] M. Friedman, M. Ming, and A. Kandel, “Fuzzy linear systems,” analysis,” Mathematics of Computation, vol. 39, no. 160, pp. 467–
Fuzzy Sets and Systems, vol. 96, no. 2, pp. 201–209, 1998. 479, 1982.
[54] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory [73] E. L. Ortiz, “The tau method,” SIAM Journal on Numerical
and Applications of Fractional Differential Equations, vol. 204 Analysis, vol. 6, no. 3, pp. 480–492, 1969.
of North-Holland Mathematics Studies, Elsevier Science B.V., [74] S. Kazem, “An integral operational matrix based on Jacobi
Amsterdam, The Netherlands, 2006. polynomials for solving fractional-order differential equations,”
[55] K. S. Miller and B. Ross, An Introduction to the Fractional Calcu- Applied Mathematical Modelling, vol. 37, no. 3, pp. 1126–1136,
lus and Fractional Differential Equations, A Wiley-Interscience 2013.
Publication, John Wiley & Sons, New York, NY, USA, 1993. [75] J. C. Mason and D. C. Handscomb, Chebyshev Polynomials,
[56] K. Diethelm, N. J. Ford, A. D. Freed, and Yu. Luchko, “Algo- Chapman & Hall/CRC, Boca Raton, Fla, USA, 2003.
rithms for the fractional calculus: a selection of numerical [76] S. Nemati, P. M. Lima, and Y. Ordokhani, “Numerical solution
methods,” Computer Methods in Applied Mechanics and Engi- of a class of two-dimensional nonlinear Volterra integral equa-
neering, vol. 194, no. 6–8, pp. 743–773, 2005. tions using Legendre polynomials,” Journal of Computational
[57] S. G. Samko, A. A. Kilbas, and O. I. Marichev, Fractional and Applied Mathematics, vol. 242, pp. 53–69, 2013.
Integrals and Derivatives. Theory and Applications, Gordon and [77] S. Abbasbandy and R. Ezzati, “Newton’s method for solving a
Breach Science Publishers, Yverdon, Switzerland, 1993. system of fuzzy nonlinear equations,” Applied Mathematics and
[58] Z. M. Odibat and N. T. Shawagfeh, “Generalized Taylor’s Computation, vol. 175, no. 2, pp. 1189–1199, 2006.
formula,” Journal of Applied Mathematics and Computing, vol. [78] Z. Odibat and S. Momani, “An algorithm for the numerical
186, no. 1, pp. 286–293, 2007. solution of differential equations of fractional order,” Journal of
[59] S. Pederson and M. Sambandham, “Numerical solution of Applied Mathematics & Informatics, vol. 26, pp. 15–27, 2008.
hybrid fuzzy differential equation IVPs by a characterization [79] S. Das, Functional Fractional Calculus for System Identification
theorem,” Information Sciences, vol. 179, no. 3, pp. 319–328, 2009. and Controls, Springer, Berlin, Germany, 2008.
[60] J. Shen and T. Tang, High Order Numerical Methods and [80] A. Le Mehaute and G. Crepy, “Introduction to transfer and
Algorithms, Chinese Science Press, Beijing, China, 2005. motion in fractal media: the geometry of kinetics,” Solid State
[61] S. Abbasbandy and M. Amirfakhrian, “The nearest approx- Ionics, vol. 9-10, no. 1, pp. 17–30, 1983.
imation of a fuzzy quantity in parametric form,” Applied [81] C. Csajka and D. Verotta, “Pharmacokinetic-pharmacodynamic
Mathematics and Computation, vol. 172, no. 1, pp. 624–632, modelling: history and perspectives,” Journal of Pharmacokinet-
2006. ics and Pharmacodynamics, vol. 33, no. 3, pp. 227–279, 2006.
Abstract and Applied Analysis 29

[82] D. Verotta, “Fractional dynamics pharmacokinetics-pharma-


codynamic models,” Journal of Pharmacokinetics and Pharma-
codynamics, vol. 37, no. 3, pp. 257–276, 2010.
[83] J. J. Nieto, R. Rodrı́guez-López, and D. Franco, “Linear first-
order fuzzy differential equations,” International Journal of
Uncertainty, Fuzziness and Knowledge-Based Systems, vol. 14, no.
6, pp. 687–709, 2006.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 546502, 10 pages
http://dx.doi.org/10.1155/2013/546502

Research Article
Two Efficient Generalized Laguerre Spectral Algorithms for
Fractional Initial Value Problems

D. Baleanu,1,2,3 A. H. Bhrawy,4,5 and T. M. Taha5


1
Department of Mathematics and Computer Sciences, Faculty of Arts and Sciences, Cankaya University,
Eskisehir Yolu 29.Km, 06810 Ankara, Turkey
2
Department of Chemical and Materials Engineering, Faculty of Engineering, King Abdulaziz University, Jeddah, Saudi Arabia
3
Institute of Space Sciences, Magurele-Bucharest, Romania
4
Department of Mathematics, Faculty of Science, King Abdulaziz University, Jeddah, Saudi Arabia
5
Department of Mathematics, Faculty of Science, Beni-Suef University, Beni Suef, Egypt

Correspondence should be addressed to A. H. Bhrawy; alibhrawy@yahoo.co.uk

Received 15 April 2013; Accepted 21 May 2013

Academic Editor: Soheil Salahshour

Copyright © 2013 D. Baleanu et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We present a direct solution technique for approximating linear multiterm fractional differential equations (FDEs) on semi-
infinite interval, using generalized Laguerre polynomials. We derive the operational matrix of Caputo fractional derivative of
the generalized Laguerre polynomials which is applied together with generalized Laguerre tau approximation for implementing a
spectral solution of linear multiterm FDEs on semi-infinite interval subject to initial conditions. The generalized Laguerre pseudo-
spectral approximation based on the generalized Laguerre operational matrix is investigated to reduce the nonlinear multiterm
FDEs and its initial conditions to nonlinear algebraic system, thus greatly simplifying the problem. Through several numerical
examples, we confirm the accuracy and performance of the proposed spectral algorithms. Indeed, the methods yield accurate
results, and the exact solutions are achieved for some tested problems.

1. Introduction Jacobi spectral approximation, in which they generalized the


Chebyshev spectral methods [6] and quadrature Legendre
Fractional calculus has been used to develop accurate models tau method [9]; moreover, other very important cases can
of many phenomena of science, engineering, economics, and be obtained for that approach. Maleki et al. [18] proposed
applied mathematics. These models are found to be best an efficient and accurate spectral collocation method based
described by FDEs [1–4]. on shifted Legendre-Gauss quadrature nodes for solving
One of the best methods, in terms of the accuracy,
fractional boundary value problems in finite interval. The
for investigating the numerical solution of various kinds of
authors of [19] used the spline functions methods for tackling
differential equations is spectral method (see, for instance,
[5–8]). Because all types of spectral methods are global and the linear and nonlinear FDEs. Recently, Bhrawy et al. [20]
numerical computational methods, they are very convenient investigated the fractional integrals of modified generalized
for approximating linear and nonlinear FDEs [6, 7, 9]. We Laguerre operational matrix to implement a numerical solu-
refer also to recent numerical and analytical methods for tion of the integrated form of the linear FDEs on semi-
solving FEEs [10–16]. infinite interval. Furthermore, Yuzbasi [21] proposed a new
In the last few years, theory and numerical solution of collocation method based on Bessel functions to introduce an
FDEs by using spectral methods have received an increasing approximate solution of a class of FDEs. We refer also to the
attention. In this direction, Doha et al. [17] proposed an recent papers [22–26] where operational matrices of several
effective way to approximate solutions of linear and nonlinear orthogonal polynomials are developed for solving linear and
multiterm FDEs with constant and variable coefficients using nonlinear ODEs and FDEs.
2 Abstract and Applied Analysis

In this paper, the Caputo fractional derivative of gen- where ⌈]⌉ and ⌊]⌋ are the ceiling and floor functions,
eralized Laguerre operational matrix (GLOM) is stated respectively, while 𝑁 = {1, 2, . . .} and 𝑁0 = {0, 1, 2, . . .}.
and proved. The main aim of this paper is to extend the The Caputo’s derivative operator is a linear operation:
application of generalized Laguerre spectral tau method
based on GLOM to develop a direct solution technique 𝐷] (𝜆𝑓 (𝑥) + 𝜇𝑔 (𝑥)) = 𝜆𝐷] 𝑓 (𝑥) + 𝜇𝐷] 𝑔 (𝑥) , (5)
for the numerical solution of linear multi-term FDEs on
where 𝜆 and 𝜇 are constants.
a semi-infinite interval. Moreover, we develop the gener-
We recall below some relevant properties of the general-
alized Laguerre pseudo-spectral approximation based on
ized Laguerre polynomials (Szego [27] and Funaro [28]). let
the GLOM for reducing the nonlinear multi-term FDEs
subject to nonhomogeneous initial conditions to a system Λ = (0, ∞) and let 𝑤(𝛼) (𝑥) = 𝑥𝛼 𝑒−𝑥 be a weight function on
of nonlinear algebraic equations. Finally, the accuracy of the Λ in the usual sense. Define
proposed algorithms is demonstrated by test problems. The 𝐿2𝑤(𝛼) (Λ)
numerical results are given to show that the proposed spectral (6)
algorithms based on generalized Laguerre operational matrix = {V | V is measurable on Λ and ‖V‖𝑤(𝛼) < ∞} ,
of Caputo fractional derivatives are very effective for linear
and nonlinear FDEs. with the inner product and norm
The outline of the paper is as follows. In Section 2, we
present some preliminaries. Section is devoted to drive the (𝑢, V)𝑤(𝛼) = ∫ 𝑢 (𝑥) V (𝑥) 𝑤(𝛼) (𝑥) 𝑑𝑥,
Λ
GLOM of Caputo fractional derivative. In Section 4, we (7)
extend the generalized Laguerre spectral tau and pseudo- ‖V‖𝑤(𝛼) = (V, V)1/2
𝑤(𝛼)
.
spectral approximations based on the GLOM of fractional
derivative for solving multiorder linear and nonlinear FDEs. For 𝛼 > −1, the generalized Laguerre polynomials are given
Some numerical experiments are presented in Section 5. by
Finally, we conclude the paper with some remarks.
1 −𝛼 𝑥 𝑖 𝑖+𝛼 −𝑥
𝐿(𝛼)
𝑖 (𝑥) = 𝑥 𝑒 𝜕𝑥 (𝑥 𝑒 ) , 𝑖 = 1, 2, . . . . (8)
𝑖!
2. Some Basic Preliminaries
According to [29] for 𝛼 > −1, we get
The two most commonly used definitions are the Riemann-
Liouville operator and the Caputo operator. We give some 𝜕𝑥 𝐿(𝛼) (𝛼+1)
𝑖 (𝑥) = −𝐿 𝑖−1 (𝑥) ,
definitions and properties of fractional derivatives and gen-
eralized Laguerre polynomials. 𝐿(𝛼)
𝑖+1 (𝑥)

Definition 1. The fractional integral operator of Riemann- 1


Liouville sense is defined as = [(2𝑖 + 𝛼 + 1 − 𝑥) 𝐿(𝛼) (𝛼)
𝑖 (𝑥) − (𝑖 + 𝛼) 𝐿 𝑖−1 (𝑥)] ,
𝑥 𝑖+1
1
𝐽] 𝑓 (𝑥) = ∫ (𝑥 − 𝑡)]−1 𝑓 (𝑡) 𝑑𝑡, ] > 0, 𝑥 > 0,
Γ (]) 0 (1) 𝑖 = 1, 2, . . . ,
0 (9)
𝐽 𝑓 (𝑥) = 𝑓 (𝑥) .
Definition 2. The Caputo fractional derivatives is given by where 𝐿(𝛼) (𝛼)
0 (𝑥) = 1 and 𝐿 1 (𝑥) = 1 + 𝛼 − 𝑥.
]
𝐷 𝑓 (𝑥) The generalized Laguerre polynomials are the 𝐿2𝑤(𝛼) (Λ)-
orthogonal system;
= 𝐽𝑚−] 𝐷𝑚 𝑓 (𝑥) ∞
𝑥 𝑚 (2) ∫ 𝐿(𝛼) (𝛼)
𝑗 (𝑥) 𝐿 𝑘 (𝑥) 𝑤
(𝛼)
(𝑥) 𝑑𝑥 = ℎ𝑘 𝛿𝑗𝑘 , (10)
1 𝑑
= ∫ (𝑥 − 𝑡)𝑚−]−1 𝑚 𝑓 (𝑡) 𝑑𝑡, 0
Γ (𝑚 − ]) 0 𝑑𝑡 where ℎ𝑘 = Γ(𝑘 + 𝛼 + 1)/𝑘!.
𝑚 − 1 < ] < 𝑚, 𝑥 > 0, The generalized Laguerre polynomials on Λ are obtained
𝑚
from
where 𝐷 is 𝑚th order differential operator.
𝑖
Γ (𝑖 + 𝛼 + 1)
The Caputo fractional derivative operator satisfies 𝐿(𝛼)
𝑖 (𝑥) = ∑ (−1)
𝑘
𝑥𝑘 , 𝑖 = 0, 1, . . . .
𝑘=0
Γ (𝑘 + 𝛼 + 1) (𝑖 − 𝑘)!𝑘!
𝐷] 𝐶 = 0, (𝐶 is a constant) , (3) (11)

𝐷] 𝑥 𝛽 The special value


𝑖−𝑞
{ 0, for 𝛽 ∈ 𝑁0 , 𝛽 < ⌈]⌉ , (𝑖 − 𝑗 − 1)!
{ (4) 𝐷𝑞 𝐿(𝛼) 𝑞
𝑖 (0) = (−1) ∑ 𝐿(𝛼)
𝑗 (0) , 𝑖 ⩾ 𝑞,
= { Γ (𝛽 + 1) 𝛽−] for 𝛽 ∈ 𝑁0 , 𝛽 ≥ ⌈]⌉ or 𝑗=0 (𝑞 − 1)! (𝑖 − 𝑗 − 𝑞)!
{ 𝑥 ,
{ Γ (𝛽 + 1 − ]) 𝛽 ∉ 𝑁, 𝛽 > ⌊]⌋ , (12)
Abstract and Applied Analysis 3

where 𝐿(𝛼)
𝑗 (0) = Γ(𝑗 + 𝛼 + 1)/Γ(𝛼 + 1)𝑗!, will be of important Then the derivative of the vector 𝜙(𝑥) can be expressed by
use later, for treating the initial conditions of the given FDEs.
Let 𝑢(𝑥) ∈ 𝐿2𝑤(𝛼) (Λ), then 𝑢(𝑥) may be expressed in terms 𝑑𝜙 (𝑥)
= D(1) 𝜙 (𝑥) , (21)
of generalized Laguerre polynomials as 𝑑𝑥

𝑢 (𝑥) = ∑𝑎𝑗 𝐿(𝛼)
𝑗 (𝑥) , (13) where D(1) is the (𝑁 + 1) × (𝑁 + 1) operational matrix of the
𝑗=0 derivative given by
1 ∞
𝑎𝑗 = ∫ 𝑢 (𝑥) 𝐿(𝛼)
𝑗 (𝑥) 𝑤
(𝛼)
(𝑥) 𝑑𝑥, 𝑗 = 0, 1, 2, . . . . 0 0 0 0
0 0 ⋅⋅⋅ 0
ℎ𝑘 0 1 0 0 0
0 0 ⋅⋅⋅ 0
(14)
(1 1 0 0
0 0 ⋅⋅⋅ 0)
( 0)
In particular applications, the generalized Laguerre polyno- D(1) = − (1 1 1 0
0 0 ⋅⋅⋅ ). (22)
(1 1 1 1
0 0 ⋅⋅⋅ 0)
mials up to degree 𝑁 + 1 are considered. Then we have
.. .. .. ..
. .. ..
𝑁 . . ⋅ ⋅ ⋅ ..
. . . .
𝑢𝑁 (𝑥) = ∑𝑎𝑗 𝐿(𝛼)
𝑗 (𝑥) . (15) (1 1 1 1 1 ⋅ ⋅ ⋅ 1 0)
𝑗=0
By using (21), it is clear that
We will present the Laguerre-Gauss quadrature. Let
{𝑥𝑗(𝛼) , 𝜛𝑗(𝛼) } be the set of generalized Laguerre-Gauss quadra- 𝑑𝑛 𝜙 (𝑥) 𝑛
ture nodes and weights: 𝑛
= (D(1) ) 𝜙 (𝑥) , (23)
𝑑𝑥
𝑁
∫ 𝜙 (𝑥) 𝑤(𝛼) (𝑥) 𝑑𝑥 = ∑𝜙 (𝑥𝑗(𝛼) ) 𝜛𝑗(𝛼) . (16) where 𝑛 ∈ 𝑁 and the superscript in D(1) denotes matrix
Λ 𝑗=0 powers. Thus
For the generalized Laguerre-Gauss quadrature, {𝑥𝑗(𝛼) } are the 𝑛
D(𝑛) = (D(1) ) , 𝑛 = 1, 2, . . . . (24)
zeros of 𝐿(𝛼)
𝑖+1 (𝑥), and

Γ (𝑖 + 𝛼 + 1) Lemma 3. Let 𝐿(𝛼)


𝜛𝑗(𝛼) = − 𝑖 (𝑥) be a generalized Laguerre polynomial
(𝑖 + 1)!𝐿(𝛼)
𝑖 (𝑥𝑗(𝛼) ) 𝜕𝑥 𝐿(𝛼)
𝑖+1 (𝑥𝑗(𝛼) ) Then
(17)
Γ (𝑖 + 𝛼 + 1) 𝑥𝑗(𝛼) 𝐷] 𝐿(𝛼)
𝑖 (𝑥) = 0, 𝑖 = 0, 1, . . . , ⌈]⌉ − 1, ] > 0. (25)
= , 0 ≤ 𝑗 ≤ 𝑖.
(𝛼) 2
(𝑖 + 𝛼 + 1) (𝑖 + 1)![𝐿(𝛼)
𝑖 (𝑥𝑗 )] In the following theorem we prove the operational matrix
of Caputo fractional derivative for the generalized Laguerre
3. GLOM of Fractional Derivatives vector (20).
Let 𝑢(𝑥) ∈ 𝐿2𝑤(𝛼) (Λ), and then 𝑢(𝑥) may be expanded using Theorem 4. Suppose ] > 0, the fractional derivative of order
generalized Laguerre polynomials as ] of 𝜙(𝑥) is given by

𝑢 (𝑥) = ∑𝑎𝑗 𝐿(𝛼)
𝑗 (𝑥) , 𝐷] 𝜙 (𝑥) ≃ D(]) 𝜙 (𝑥) , (26)
𝑗=0

1 ∞ where D(]) is the (𝑁+1)×(𝑁+1) operational matrix of Caputo


𝑎𝑗 = ∫ 𝑢 (𝑥) 𝐿(𝛼)
𝑗 (𝑥) 𝑤
(𝛼)
(𝑥) 𝑑𝑥, 𝑗 = 0, 1, 2, . . . . fractional derivative and is given by
ℎ𝑘 0
(18)
D(])
In particular applications, the generalized Laguerre poly-
nomials up to degree 𝑁 + 1 are considered. Then we have 0 0 0 ⋅⋅⋅ 0
.. .. .. ..
𝑁 . . . ⋅⋅⋅ .
𝑢𝑁 (𝑥) = ∑ 𝑎𝑗 𝐿(𝛼) 𝑇
𝑗 (𝑥) = 𝐶 𝜙 (𝑥) , (19) ( 0 0 0 ⋅⋅⋅ 0 )
𝑗=0 (𝑆] (⌈]⌉ , 0) 𝑆] (⌈]⌉ , 1) 𝑆] (⌈]⌉ , 2) ⋅⋅⋅ 𝑆] (⌈]⌉ , 𝑁))
( )
=( .. .. .. .. ),
where the vector 𝐶 and vector 𝜙(𝑥) are given by ( . . . ⋅⋅⋅ . )
( )
𝑆] (𝑖, 0) 𝑆] (𝑖, 1) 𝑆] (𝑖, 2) ⋅⋅⋅ 𝑆] (𝑖, 𝑁)
𝐶𝑇 = [𝑐0 , 𝑐1 , . . . , 𝑐𝑁] , .. .. .. ..
. . . ⋅⋅⋅ .
𝑇
(20) ( 𝑆] (𝑁, 0) 𝑆] (𝑁, 1) 𝑆] (𝑁, 2) ⋅⋅⋅ 𝑆] (𝑁, 𝑁) )
𝜙 (𝑥) = [𝐿(𝛼)
0 (𝑥) , 𝐿(𝛼)
1 (𝑥) , . . . , 𝐿(𝛼)
𝑁 (𝑥)] . (27)
4 Abstract and Applied Analysis

where Also according to Lemma 3, we can write


𝑆] (𝑖, 𝑗) 𝐷] 𝐿(𝛼)
𝑖 (𝑥) ≃ [0, 0, 0, . . . , 0] 𝜙 (𝑥) , 𝑖 = 0, 1, . . . , ⌈]⌉ − 1.
𝑖 𝑗 (35)
= ∑ ∑ ((−1)𝑘+ℓ 𝑗!Γ (𝑖 + 𝛼 + 1)
𝑘=⌈]⌉ℓ=0
A combination of (34) and (35) leads to the desired result.
(28)
× Γ (𝑘 − ] + 𝛼 + ℓ + 1) )
Remark 5. In the case of ] = 𝑛 ∈ 𝑁, Theorem 4 gives the
× ((𝑖 − 𝑘)! (𝑗 − ℓ)!ℓ!Γ (𝑘 − ] + 1) same result as (23).
−1
× Γ (𝑘 + 𝛼 + 1) Γ (𝛼 + ℓ + 1) ) . 4. Applications of GLOM for Multiterm FDEs
Proof. Applying (4) to (11) gives In this section, we are interested in using GLOM in combi-
nation with two types of spectral methods for solving linear
𝐷] 𝐿(𝛼)
𝑖 (𝑥)
and nonlinear FDEs.
𝑖
Γ (𝑖 + 𝛼 + 1)
= ∑ (−1)𝑘 𝐷] 𝑥 𝑘 4.1. Linear Multiorder FDEs. Here, we propose a direct
𝑘=0
(𝑖 − 𝑘)!𝑘!Γ (𝑘 + 𝛼 + 1)
solution technique to approximate linear multi-term FDEs
(29)
𝑖 with constant coefficients using the generalized Laguerre tau
Γ (𝑖 + 𝛼 + 1)
= ∑ (−1)𝑘 𝑥𝑘−] , method in combination with GLOM.
𝑘=⌈]⌉
(𝑖 − 𝑘)!Γ (𝑘 − ] + 1) Γ (𝑘 + 𝛼 + 1) Consider the linear multi-order FDE
𝑘
𝑖 = ⌈]⌉ , . . . , 𝑁.
𝐷] 𝑢 (𝑥) = ∑𝛾𝑗 𝐷𝛽𝑗 𝑢 (𝑥) + 𝛾𝑘+1 𝑢 (𝑥) + 𝑔 (𝑥) , in Λ, (36)
𝑘−] 𝑗=1
Now, 𝑥 can be approximated by 𝑁 + 1 terms of the
generalized Laguerre polynomials to get with initial conditions
𝑁
𝑢(𝑖) (0) = 𝑑𝑖 , 𝑖 = 0, . . . , 𝑚 − 1, (37)
𝑥𝑘−] = ∑ 𝑏𝑘𝑗 𝐿(𝛼)
𝑗 (𝑥) , (30)
𝑗=0 where 𝛾𝑗 (𝑗 = 1, . . . , 𝑘 + 1) are constants and 𝑚 − 1 <
where 𝑏𝑗 is directly obtained from (18), and ] ≤ 𝑚, 0 < 𝛽1 < 𝛽2 < ⋅ ⋅ ⋅ < 𝛽𝑘 < ]. Also 𝐷] 𝑢(𝑥) ≡
𝑢(]) (𝑥) is ] order of Caputo fractional derivative for 𝑢(𝑥),
𝑗 𝑑𝑖 (𝑖 = 0, . . . , 𝑚 − 1) are the initial values, and 𝑔(𝑥) is a source
𝑗!Γ (𝑘 − ] + 𝛼 + ℓ + 1)
𝑏𝑗 = ∑ (−1)ℓ . (31) function.
ℓ=0 (𝑗 − ℓ)! (ℓ)!Γ (ℓ + 𝛼 + 1)
To solve the fractional FDE (36)-(37), we approximate
Employing (29)–(31) yields 𝑢(𝑥) and 𝑔(𝑥) by generalized Laguerre polynomials as
𝑁 𝑁
𝐷] 𝐿(𝛼) (𝛼)
𝑖 (𝑥) = ∑ 𝑆] (𝑖, 𝑗) 𝐿 𝑗 (𝑥) , 𝑖 = ⌈]⌉ , . . . , 𝑁, (32) 𝑢 (𝑥) ≃ ∑𝑐𝑖 𝐿(𝛼) 𝑇
𝑖 (𝑥) = 𝐶 𝜙 (𝑥) , (38)
𝑗=0 𝑖=0

where 𝑁
𝑔 (𝑥) ≃ ∑𝑔𝑖 𝐿(𝛼) 𝑇
𝑖 (𝑥) = 𝐺 𝜙 (𝑥) , (39)
𝑖 𝑗
𝑘+ℓ 𝑖=0
𝑆] (𝑖, 𝑗) = ∑ ∑ ((−1) 𝑗!Γ (𝑖 + 𝛼 + 1)
𝑘=⌈]⌉ℓ=0
where vector 𝐺 = [𝑔0 , . . . , 𝑔𝑁]𝑇 is known.
×Γ (𝑘 − ] + 𝛼 + ℓ + 1) ) By using Theorem 4 (relation equations (26), and (38)) we
have
(33)
((𝑖 − 𝑘)! (𝑗 − ℓ)!ℓ! 𝐷] 𝑢 (𝑥) ≃ 𝐶𝑇 𝐷] 𝜙 (𝑥) = 𝐶𝑇 𝐷(]) 𝜙 (𝑥) ,
× Γ (𝑘 − ] + 1) Γ (𝑘 + 𝛼 + 1)
𝐷𝛽𝑗 𝑢 (𝑥) ≃ 𝐶𝑇 𝐷𝛽𝑗 𝜙 (𝑥) = 𝐶𝑇 𝐷(𝛽𝑗 ) 𝜙 (𝑥) , 𝑗 = 1, . . . , 𝑘.
−1 (40)
× Γ (𝛼 + ℓ + 1) ) .
Accordingly, (32) can be written in a vector form as follows: Making use of (38)–(40), the residual 𝑅𝑁(𝑥) for (36) can be
given from
𝐷] 𝐿(𝛼)
𝑖 (𝑥) 𝑘

≃ [𝑆] (𝑖, 0) , 𝑆] (𝑖, 1) , 𝑆] (𝑖, 2) , . . . , 𝑆] (𝑖, 𝑁)] 𝜙 (𝑥) , (34) 𝑅𝑁 (𝑥) = (𝐶𝑇 𝐷(]) − 𝐶𝑇 ∑𝛾𝑗 D(𝛽𝑗 ) − 𝛾𝑘+1 𝐶𝑇 − 𝐺𝑇 ) 𝜙 (𝑥) .
𝑗=1
𝑖 = ⌈]⌉ , . . . , 𝑁. (41)
Abstract and Applied Analysis 5

The use of generalized Laguerre tau approximation gen- 5. Numerical Results


erates (𝑁 − 𝑚 + 1) system of linear equations
This section considers several numerical examples to demon-
⟨𝑅𝑁 (𝑥) , 𝐿(𝛼)
𝑗 (𝑥)⟩
strate the accuracy and applicability of the proposed spectral
algorithms based on operational matrix of fractional deriva-
∞ tives of generalized Laguerre polynomials. A comparison
= ∫ 𝑤 (𝑥) 𝑅𝑁 (𝑥) 𝐿(𝛼)
𝑗 (𝑥) 𝑑𝑥 = 0,
(42)
of the results obtained by adopting different choices of the
0
generalized Laguerre parameter 𝛼 reveals that the present
𝑗 = 0, 1, . . . , 𝑁 − 𝑚. algorithms are very convenient for all choices of 𝛼 and
produces accurate solutions to multi-term FDEs on semi-
Substituting (23) and (38) into (37) generates 𝑚 set of linear infinite interval.
equations
Example 1. Consider the linear FDE
𝑢(𝑖) (0) = 𝐶𝑇 D(𝑖) 𝜙 (0) = 𝑑𝑖 , 𝑖 = 0, 1, . . . , 𝑚 − 1. (43)
𝐷2 𝑢 (𝑥) + 𝐷3/2 𝑢 (𝑥) + 𝑢 (𝑥) = 1 + 𝑥,
The combination of (42) and (43) reduces the solution of (48)
(36)-(37) to a linear system of algebraic equations, which can 𝑢 (0) = 1, 𝑢󸀠 (0) = 1, 𝑥 ∈ Λ,
be solved for unknown coefficients of the vector 𝐶 by any
direct solver technique to find the spectral solution 𝑢(𝑥). where 𝑢(𝑥) = 1 + 𝑥 is the exact solution.

4.2. Nonlinear Multiorder FDEs. In this section, we present If we apply the operational matrix formulation, the
the generalized Laguerre pseudo-spectral approximation in generalized Laguerre spectral tau method with 𝑁 = 2, we
combination with GLOM of fractional derivative to find the get
approximate solution 𝑢𝑁(𝑥).
Let us consider the nonlinear multi-term FDE 𝑢 (𝑥) = 𝑐0 𝐿(𝛼) (𝛼) (𝛼) 𝑇
0 (𝑥) + 𝑐1 𝐿 1 (𝑥) + 𝑐2 𝐿 2 (𝑥) = 𝐶 𝜙 (𝑥) .
(49)

𝐷] 𝑢 (𝑥) = 𝐹 (𝑥, 𝑢 (𝑥) , 𝐷𝛽1 𝑢 (𝑥) , . . . , 𝐷𝛽𝑘 𝑢 (𝑥)) , in Λ, Here, we have


(44) 0 0 0
D(1) = − (1 0 0) ,
subject to the nonhomogeneous initial conditions (37), where
1 1 0
𝐹 can be nonlinear in general. In [30], the authors studied the
existence of solutions of a class of nonlinear FDEs. 0 0 0
Now, we will implement the generalized Laguerre oper- D(2) = (0 0 0) ,
ational matrix for treating this nonlinear problem. To do 1 0 0
this, firstly, we approximate 𝑢(𝑥), 𝐷] 𝑢(𝑥), and 𝐷𝛽𝑗 𝑢(𝑥), for (50)
𝑗 = 1, . . . , 𝑘 by using (38), (40), respectively, and then the 0 0 0
operational matrices formulation of (44) can be expressed as D(3/2) =( 0 0 0 ),
𝑆3/2 (2, 0) 𝑆3/2 (2, 1) 𝑆3/2 (2, 2)
𝐶𝑇 D(]) 𝜙 (𝑥)
(45) 𝑔0
≃ 𝐹 (𝑥, 𝐶𝑇 𝜙 (𝑥) , 𝐶𝑇 D(𝛽1 ) 𝜙 (𝑥) , . . . , 𝐶𝑇 D(𝛽𝑘 ) 𝜙 (𝑥)) . 𝐺 = (𝑔1 ) ,
𝑔2
Also, making use of (38) and (23) in (44) yields
where 𝑔𝑗 and 𝑆] (𝑖, 𝑗) are defined in (18) and (27).
(𝑖) 𝑇 (𝑖) Applying variational formulation of the tau method of
𝑢 (0) = 𝐶 D 𝜙 (0) = 𝑑𝑖 , 𝑖 = 0, 1, . . . , 𝑚 − 1. (46)
(48) yields
Collocating the operational matrix equation (45) at (𝑁−𝑚+1)
nodes of the generalized Laguerre-Gauss quadrature on Λ, 𝑐0 + 𝑐2 + 𝑆3/2 (2, 0) = 𝑔0 . (51)

𝐶𝑇 D(]) 𝜙 (𝑥𝑗(𝛼) ) The treatment of initial conditions using (43) gives

(𝛼 + 1) (𝛼 + 2)
≃ 𝐹 (𝑥, 𝐶𝑇 𝜙 (𝑥𝑗(𝛼) ) , 𝐶𝑇 D(𝛽1 ) 𝜙 (𝑥𝑗(𝛼) ) , (47) 𝑐0 + (𝛼 + 1) 𝑐1 + 𝑐2 = 1,
2 (52)
𝑇 (𝛽𝑘 )
...,𝐶 D 𝜙 (𝑥𝑗(𝛼) )) . −𝑐1 − (𝛼 + 2) 𝑐2 − 1 = 0.

Combining (𝑁 − 𝑚 + 1) algebraic equations (47) with 𝑚 Solving the resulted system of algebraic equations (51)-
initial conditions (46) generates a system of (𝑁 + 1) non- (52) provides the unknown coefficients in terms of the
linear algebraic equations. This system may be evaluated by parameter 𝛼:
implementing Newton’s iterative method to find the spectral
solution 𝑢(𝑥). 𝑐0 = 𝛼 + 2, 𝑐1 = −1, 𝑐2 = 0. (53)
6 Abstract and Applied Analysis

Table 1: 𝑐0 , 𝑐1 , and 𝑐2 for different values of 𝛼 for Example 1. Table 2: 𝑐0 , 𝑐1 , and 𝑐2 for different values of 𝛼 for Example 2.

𝛼 𝑐0 𝑐1 𝑐2 𝛼 𝑐0 𝑐1 𝑐2
−0.5 1.5 −1 0 −0.5 0.75 −3 2
0 2 −1 0 0 2 −4 2
0.5 2.5 −1 0 0.5 3.75 −5 2
1 3 −1 0 1 6 −6 2
2 4 −1 0 2 12 −8 2
3 5 −1 0 3 20 −10 2

Accordingly, the approximate solution can be written as


Now, making use of (43) yields
𝐿(𝛼)
0 (𝑥)

𝐿(𝛼) (𝛼 + 1) (𝛼 + 2)
1 (𝑥) 𝑐0 + (𝛼 + 1) 𝑐1 + 𝑐2 = 0.
𝑢 (𝑥) = (𝑐0 , 𝑐1 , 𝑐2 ) ( ) = 1 + 𝑥, (54) 2 (59)
𝐿(𝛼)
2 (𝑥) −𝑐1 − (𝛼 + 2) 𝑐2 = 0.

which is the exact solution. Finally by solving (58)-(59), then, we get


Table 1 lists the values of 𝑐0 , 𝑐1 , and 𝑐2 with different
choices of (𝛼). Indeed, we can achieve the exact solution
of this problem with all choices of the generalized Laguerre 𝑐0 = 𝛼2 + 3𝛼 + 2, 𝑐1 = −2𝛼 − 4, 𝑐2 = 2. (60)
parameters 𝛼.

Example 2. Consider linear FED Thus we can write

𝐷2 𝑢 (𝑥) + 𝐷1/2 𝑢 (𝑥) + 𝑢 (𝑥) 𝐿(𝛼)


0 (𝑥)

2.6666666667 1.5 𝐿(𝛼)


1 (𝑥)
= 𝑥2 + 2 + 𝑥 , (55) 𝑢 (𝑥) = (𝑐0 , 𝑐1 , 𝑐2 ) ( ) = 𝑥2 , (61)
Γ (0.5)
𝐿(𝛼)
2 (𝑥)
𝑢 (0) = 0, 𝑢󸀠 (0) = 0, 𝑥 ∈ Λ.

The analytical solution is 𝑢(𝑥) = 𝑥2 .


which is the exact solution.
The use of technique described in Section 4.1 with 𝑁 = 2 In Table 2, we exhibit the values of 𝑐0 , 𝑐1 and 𝑐2 with
enables one to approximate the solution as different choices of (𝛼).

𝑢 (𝑥) = 𝑐0 𝐿(𝛼) (𝛼) (𝛼) 𝑇 Example 3. Consider linear initial value problem of fractional
0 (𝑥) + 𝑐1 𝐿 0 (𝑥) + 𝑐2 𝐿 2 (𝑥) = 𝐶 𝜙 (𝑥) .
(56)
order (see [31])
Here, we have

0 0 0 𝐷2 𝑢 (𝑥) − 2𝐷𝑢 (𝑥) + 𝐷1/2 𝑢 (𝑥) + 𝑢 (𝑥)


D(2) = (0 0 0) , 16 2.5
1 0 0 = 𝑥3 − 6𝑥2 + 6𝑥 + 𝑥 , (62)
5√𝜋
0 0 0
𝑢 (0) = 0, 𝑢󸀠 (0) = 0, 𝑥 ∈ Λ,
(1/2) 𝑆 (1, 0) 𝑆1/2 (1, 1) 𝑆1/2 (1, 2)
D = ( 1/2 ), (57)
𝑆1/2 (2, 0) 𝑆1/2 (2, 0) 𝑆1/2 (2, 0) whose exact solution is given by 𝑢(𝑥) = 𝑥3 .

𝑔0 If we apply the operational matrix formulation, the


𝐺 = (𝑔1 ) . generalized Laguerre spectral tau method with 𝑁 = 2, we
𝑔2 get

Therefore using (42), we obtain 3


𝑢 (𝑥) = ∑𝑐𝑖 𝐿(𝛼) 𝑇
𝑖 (𝑥) = 𝐶 𝜙 (𝑥) . (63)
𝑐0 + 𝑆1/2 (1, 0) 𝑐1 + [𝑆1/2 (2, 0) + 1] 𝑐2 = 6. (58) 𝑖=0
Abstract and Applied Analysis 7

Also, Table 3: 𝑐0 , 𝑐1 , 𝑐2 , and 𝑐3 for different values of 𝛼 for Example 3.

0 0 0 0 0 0 0 0 𝛼 𝑐0 𝑐1 𝑐2 𝑐3
(2) 0 0 0 0 (1) 1 0 0 0 −0.5 15/8 −45/4 15 −6
D =( ), D = −( ),
1 0 0 0 1 1 0 0 0 6 −18 18 −6
2 1 0 0 1 1 1 0 0.5 105/8 −105/4 21 −6
0 0 0 0 1 24 −36 24 −6
2 60 −60 30 −6
𝑆1/2 (1, 0) 𝑆1/2 (1, 1) 𝑆1/2 (1, 2) 𝑆1/2 (1, 3)
3 120 −90 36 −6
(1/2) ( )
D =( ,
𝑆1/2 (2, 0) 𝑆1/2 (2, 1) 𝑆1/2 (2, 2) 𝑆1/2 (2, 3))

𝑆1/2 (3, 0) 𝑆1/2 (3, 1) 𝑆1/2 (3, 2) 𝑆1/2 (3, 3) Example 4. Consider the following FDE:
( )
𝐷5/2 𝑢 (𝑥) + 𝐷2 𝑢 (𝑥) − 2𝐷1/2 𝑢 (𝑥) + 4𝑢 (𝑥) = 𝑔 (𝑥) ,
𝑔0 (68)
𝑔1 𝑢 (0) = 0, 𝑢󸀠 (0) = 0, 𝑥 ∈ Λ,
𝐺 = ( ).
𝑔2
𝑔3 where
(64)
2Γ (10) 17/2 Γ (10) 13/2
Therefore using (42), we obtain 𝑔 (𝑥) = 4𝑥9 + 72𝑥7 − 𝑥 + 𝑥 (69)
Γ (19/2) Γ (15/2)
𝑐0 + [2𝑐1 + 𝑆1/2 (1, 0)] 𝑐1
and the exact solution is 𝑢(𝑥) = 𝑥9 .
+ [3 + 𝑆1/2 (2, 0)] 𝑐2
Now, if we use the spectral tau approximation based on
+ [4 + 𝑆1/2 (3, 0)] 𝑐3 − 𝑔0 = 0, with 𝑁 = 9 and 𝑥 ∈ Λ, then we obtain

[1 + 𝑆1/2 (1, 1)] 𝑐1 9


𝑢 (𝑥) = ∑𝑐𝑖 𝐿(𝛼) 9
𝑖 (𝑥) = 𝑥 , (70)
+ [2 + 𝑆1/2 (2, 1)] 𝑐2 𝑖=0

+ [3 + 𝑆1/2 (3, 1)] 𝑐3 − 𝑔1 = 0, which is the exact solution.

(𝛼 + 1) (𝛼 + 2) Example 5. Consider the nonhomogeneous fractional initial


𝐶𝑇 𝜙 (0) = 𝑐0 + (𝛼 + 1) 𝑐1 + 𝑐2 value problem
2
(𝛼 + 1) (𝛼 + 2) (𝛼 + 3) 𝐷2 𝑢 (𝑥) + 𝐷] 𝑢 (𝑥) + 𝑢 (𝑥) = 𝑔 (𝑥) ,
+ 𝑐3 = 0,
6 (71)
(𝛼 + 3) (𝛼 + 2) 𝑢 (0) = 1, 𝑢󸀠 (0) = 0, 𝑥 ∈ (0, 10) ,
𝐶𝑇 D(1) 𝜙 (0) = −𝑐1 − (𝛼 + 2) 𝑐2 − 𝑐3 = 0.
2 where
(65)
Accordingly, we get 𝑔 (𝑥)
1 1
𝑐0 = 𝛼3 + 6𝛼 + 11𝛼 + 6,
3 3
= ]6 (𝑒((1/] )𝑥) + sin ( 3
𝑥)) + 𝑒((1/] )𝑥) − sin ( 3 𝑥)
] ]
𝑐1 = −3𝛼2 − 15𝛼 − 18, (66) 1 𝑥 3 1
+ ∫ (𝑥 − 𝑡)−]−1 (𝑒((1/] )𝑥) − sin ( 3 𝑥)) 𝑑𝑡
𝑐2 = 6𝛼 + 18, 𝑐3 = −6. Γ (−]) 0 ]
(72)
Thus we can write
3

𝐿(𝛼) and the exact solution is given by 𝑢(𝑥) = 𝑒((1/] )𝑥) −


0 (𝑥)
sin((1/]3 )𝑥).
𝐿(𝛼) (𝑥)
( 1 )
𝑢 (𝑥) = (𝑐0 , 𝑐1 , 𝑐2 , 𝑐3 ) ( ) 3
(𝐿(𝛼) (𝑥)) = 𝑥 , (67) Table 4 introduces the maximum absolute errors, using
2 the tau method based on GLOM of fractional derivative, at
] = 1.5 and with different choices of the parameters 𝛼 and
𝐿(𝛼)
3 (𝑥) 𝑁. The curves of exact solutions and approximate solutions
( ) obtained by the proposed method for 𝛼 = 1, 𝑁 = 10, and ] =
which is the exact solution. The 4 unknown coefficients with 1.5, 1.7, 1.9 are shown in Figures 1 and 2. From these figures,
various choices of 𝛼 are listed in Table 3. the exact and approximate solutions are completely conciet.
8 Abstract and Applied Analysis

Table 4: Maximum absolute error for various choices of 𝛼 and 𝑁


40000
for Example 5.

Exact and approximate solutions


𝑁 ] 𝛼 GLOM 𝛼 GLOM 𝛼 GLOM 30000
5 1.5 0 1.69 ⋅ 10−2 1 4.41 ⋅ 10−2 2 1.07 ⋅ 10−1
10 1.08 ⋅ 10−3 4.17 ⋅ 10−4 6.52 ⋅ 10−4
20000
15 4.99 ⋅ 10−6 1.46 ⋅ 10−5 1.33 ⋅ 10−5
20 3.57 ⋅ 10−8 1.22 ⋅ 10−7 2.41 ⋅ 10−7
10000
25 3.50 ⋅ 10−10 1.30 ⋅ 10−9 4.07 ⋅ 10−9
30 4.03 ⋅ 10−12 1.55 ⋅ 10−11 3.47 ⋅ 10−11
35 5.10 ⋅ 10−14 2.08 ⋅ 10−13 6.50 ⋅ 10−13 0
40 6.87 ⋅ 10−16 2.64 ⋅ 10−15 5.29 ⋅ 10−15 0 5 10 15 20 25 30 35
x
uN (x, 1.5) u(x, 1.7)
7 u(x, 1.5) uN (x, 1.9)
Exact and approximate solutions

uN (x, 1.7) u(x, 1.9)


6
Figure 2: Comparing the exact solution and the approximate
5 solutions at 𝑁 = 10, 𝛼 = 1, and ] = 1.5, 1.7, 1.9.
4
×10−5
3
8
2
The absolute error

1 6
0 1 2 3 4 5 6 7
x
4
uN (x, 1.5) u(x, 1.7)
u(x, 1.5) uN (x, 1.9)
uN (x, 1.7) u(x, 1.9) 2

Figure 1: Comparing the exact solution and the approximate


0
solutions at 𝑁 = 10, 𝛼 = 1, and ] = 1.5, 1.7, 1.9.
0 5 10 15 20 25 30
x

Figure 3: The absolute error for 𝛾 = 1/100, 𝛼 = 1 at 𝑁 = 30.


Example 6. Let us consider the nonlinear fractional initial
value problem
Example 7. We consider the nonlinear fractional initial value
𝐷2 𝑢 (𝑥) + 𝐷1/2 𝑢 (𝑥) + 𝑢2 (𝑥) = 𝑔 (𝑥) , problem
(73)
𝑢 (0) = 1, 𝑢󸀠 (0) = 0, 𝑥 ∈ (0, 20) , 𝐷𝜉 𝑢 (𝑥) + 𝐷𝜂 𝑢 (𝑥) 𝐷𝜃 𝑢 (𝑥) + 𝑢2 (𝑥) = 𝑔 (𝑥) ,

where 𝑢 (0) = 0, 𝑢󸀠 (0) = 0, 𝑢󸀠󸀠 (0) = 0, 𝑥 ∈ (0, 1) ,

𝜁 ∈ (2, 3) , 𝜂 ∈ (1, 2) , 𝜃 ∈ (0, 1) ,


𝑔 (𝑥) = cos2 (𝛾𝑥) − 𝛾2 cos (𝛾𝑥) (75)
1 𝑥 (74)
+ ∫ (𝑥 − 𝑡)−]−1 cos (𝛾𝑡) 𝑑𝑡 where
Γ (−1/2) 0
6𝑥3−𝜉 36𝑥6−𝜂−𝜃
𝑔 (𝑥) = 𝑥6 + + (76)
and the exact solution is given by 𝑢(𝑥) = cos(𝛾𝑥). Γ (4 − 𝜉) Γ (4 − 𝜂) Γ (4 − 𝜃)

The solution of this problem is obtained by applying and the exact solution is 𝑢(𝑥) = 𝑥3 .
the generalized Laguerre-Gauss collocation method based on
generalized Laguerre operational matrix. The absolute error The solution of this problem is obtained by applying
between the exact and the approximate solution obtained by the generalized Laguerre-Gauss collocation method based on
the proposed method 𝛾 = 0.01, 𝛼 = 1 and 𝑁 = 30 is given in generalized Laguerre operational matrix for 𝜉 = 2.001, 𝜂 =
Figure 3. 1.001, and 𝜃 = 0.001. The exact solution and approximate
Abstract and Applied Analysis 9

30 [6] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A Chebyshev


spectral method based on operational matrix for initial and
25 boundary value problems of fractional order,” Computers &
Mathematics with Applications, vol. 62, no. 5, pp. 2364–2373,
20 2011.
[7] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A new
uN (x)

15
Jacobi operational matrix: an application for solving fractional
10
differential equations,” Applied Mathematical Modelling, vol. 36,
no. 10, pp. 4931–4943, 2012.
5 [8] D. Funaro, “Estimates of Laguerre spectral projectors in Sobolev
spaces,” in Orthogonal Polynomials and Their Applications, C.
0 Brezinski, L. Gori, and A. Ronveaux, Eds., vol. 9, pp. 263–266,
0 0.5 1 1.5 2 2.5 3 Scientific Publishing, Singapore, 1991.
x [9] A. H. Bhrawy, A. S. Alofi, and S. S. Ezz-Eldien, “A quadrature
tau method for fractional differential equations with variable
Exact solution coefficients,” Applied Mathematics Letters, vol. 24, no. 12, pp.
uN (x) at N = 5
2146–2152, 2011.
uN (x) at N = 10
[10] J. Song, F. Yin, X. Cao, and F. Lu, “Fractional variational
Figure 4: Comparing the exact solution and the approximate iteration method versus Adomian’s decomposition method in
solutions at 𝑁 = 5, 10 and 𝛼 = 1. some fractional partial differential equations,” Journal of Applied
Mathematics, vol. 2013, Article ID 392567, 10 pages, 2013.
[11] H. Jafari, H. Tajadodi, and D. Baleanu, “A modified variational
iteration method for solving fractional Riccati differential equa-
solutions obtained by the proposed method for 𝛼 = 1 and tion by Adomian polynomials,” Fractional Calculus and Applied
two choices of 𝑁 are shown in Figure 4. Analysis, vol. 16, no. 1, pp. 109–122, 2013.
[12] A. A. Hemeda, “New iterative method: an application for
solving fractional physical differential equations,” Abstract and
6. Conclusion Applied Analysis, vol. 2013, Article ID 617010, 9 pages, 2013.
In this paper, the generalized Laguerre operational matrix of [13] G.-C. Wu and D. Baleanu, “Variational iteration method for
Caputo fractional derivative was derived. Furthermore, we fractional calculus: a universal approach by Laplace transform,”
have implemented the generalized Laguerre tau approxima- Advances in Difference Equations, vol. 2013, article 18, 2013.
tion in combination with the GLOM with the generalized [14] C. Yang and J. Hou, “An approximate solution of nonlinear
Laguerre family to solve the linear FDEs. In addition, com- fractional differential equation by Laplace transform and ado-
mian polynomials,” Journal of Information and Computational
bining the pseudo-spectral approximation and the GLOM
Science, vol. 10, no. 1, pp. 213–222, 2013.
of fractional derivative was applied to develop an accurate
[15] M. Gulsua, Y. Ozturka, and A. Anapalia, “Numerical approach
approximate solution of nonlinear FDEs. The generalized for solving fractional Fredholm integro-differential equation,”
Laguerre-Gauss quadrature points were used as a collocation International Journal of Computer Mathematics, 2013.
nodes. This method reduces the nonlinear FDEs to a system [16] H. S. Nik, S. Effati, S. S. Motsa, and S. Shateyi, “A new piecewise-
of algebraic equation in the expansion coefficients which spectral homotopy analysis method for solving chaotic systems
can be solved by any standard technique. The numerical of initial value problems,” Mathematical Problems in Engineer-
results demonstrate that the proposed spectral algorithms are ing, vol. 2013, Article ID 583193, 13 pages, 2013.
accurate and efficient. [17] E. H. Doha, A. H. Bhrawy, D. Baleanu, and S. S. Ezz-Eldien, “On
shifted Jacobi spectral approximations for solving fractional
differential equations,” Applied Mathematics and Computation,
References vol. 219, no. 15, pp. 8042–8056, 2013.
[1] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional [18] M. Maleki, I. Hashim, M. Tavassoli Kajani, and S. Abbas-
Calculus: Models and Numerical Methods, vol. 3 of Series on bandy, “An adaptive pseudospectral method for fractional order
Complexity, Nonlinearity and Chaos, World Scientific Publish- boundary value problems,” Abstract and Applied Analysis, vol.
ing, Hackensack, NJ, USA, 2012. 2012, Article ID 381708, 19 pages, 2012.
[2] K. S. Miller and B. Ross, An Introduction to the Fractional [19] A. Al-Rabtah, S. Momani, and M. A. Ramadan, “Solving linear
Calculus and Fractional Differential Equations, John Wiley & and nonlinear fractional differential equations using spline
Sons, New York, NY, USA, 1993. functions,” Abstract and Applied Analysis, vol. 2012, Article ID
426514, 9 pages, 2012.
[3] M. D. Ortigueira, “Introduction to fractional linear systems.
[20] A. H. Bhrawy, M. M. Alghamdi, and T. M. Taha, “A new
Part 1: continuous-time case,” IEE Proceedings: Vision, Image
modified generalized Laguerre operational matrix of fractional
and Signal Processing, vol. 147, no. 1, pp. 62–70, 2000.
integration for solving fractional differential equations on the
[4] S. Das, Functional Fractional Calculus for System Identification half line,” Advances in Difference Equations, vol. 2012, article 179,
and Controls, Springer, New York, NY, USA, 2008. 2012.
[5] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, [21] S. Yuzbasi, “Numerical solution of the Bagley-Torvik equation
Spectral Methods in Fluid Dynamics, Springer, New York, NY, by the Bessel collocation method,” Mathematical Methods in the
USA, 1988. Applied Sciences, vol. 36, no. 3, pp. 300–312, 2013.
10 Abstract and Applied Analysis

[22] M. Shaban, S. Kazem, and J. A. Rad, “A modification of the


homotopy analysis method based on Chebyshev operational
matrices,” Mathematical and Computer Modelling, vol. 57, pp.
1227–1239, 2013.
[23] K. Parand and S. A. Kaviani, “Application of the exact opera-
tional matrices based on the Bernstein polynomials,” Journal of
Mathematics and Computer Science, vol. 6, pp. 36–59, 2013.
[24] E. Tohidi, F. Soleymani, and A. Kilicman, “Robustness of
operational matrices of differentiation for solving state-space
analysis and optimal control problems,” Abstract and Applied
Analysis, vol. 2013, Article ID 535979, 9 pages, 2013.
[25] A. H. Bhrawy, D. Baleanu, L. M. Assas, and J. A. Tenreiro
Machado, “On a generalized laguerre operational matrix of
fractional integration,” Mathematical Problems in Engineering,
vol. 2013, Article ID 569286, 7 pages, 2013.
[26] M. El-Kady and A. El-Sayed, “Fractional differentiation matri-
ces for solving fractional orders differential equations,” Interna-
tional Journal of Pure and Applied Mathematics, vol. 84, no. 2,
pp. 1–13, 2013.
[27] G. Szego, Orthogonal Polynomials, American Mathematical
Society, Providence, RI, USA , 4th edition, 1975.
[28] D. Funaro, Polynomial Approximation of Differential Equations,
vol. 8, Springer, Berlin, Germany, 1992.
[29] D. K. Dimitrov, F. Marcellán, and F. R. Rafaeli, “Monotonicity
of zeros of Laguerre-Sobolev-type orthogonal polynomials,”
Journal of Mathematical Analysis and Applications, vol. 368, no.
1, pp. 80–89, 2010.
[30] D. Baleanu, H. Mohammadi, and S. Rezapour, “Positive solu-
tions of an initial value problem for nonlinear fractional
differential equations,” Abstract and Applied Analysis, vol. 2012,
Article ID 837437, 7 pages, 2012.
[31] E. A. Rawashdeh, “Numerical solution of fractional integro-
differential equations by collocation method,” Applied Mathe-
matics and Computation, vol. 176, no. 1, pp. 1–6, 2006.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 351057, 5 pages
http://dx.doi.org/10.1155/2013/351057

Research Article
Local Fractional Series Expansion Method for Solving
Wave and Diffusion Equations on Cantor Sets

Ai-Min Yang,1,2 Xiao-Jun Yang,3 and Zheng-Biao Li4


1
College of Mechanical Engineering, Yanshan University, Qinhuangdao 066004, China
2
College of Science, Hebei United University, Tangshan 063009, China
3
Department of Mathematics and Mechanics, China University of Mining and Technology, Xuzhou Campus, Xuzhou,
Jiangsu 221008, China
4
College of Mathematics and Information Science, Qujing Normal University, Qujing, Yunnan 655011, China

Correspondence should be addressed to Ai-Min Yang; aimin heut@163.com

Received 6 May 2013; Accepted 22 May 2013

Academic Editor: Dumitru Baleanu

Copyright © 2013 Ai-Min Yang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We proposed a local fractional series expansion method to solve the wave and diffusion equations on Cantor sets. Some examples
are given to illustrate the efficiency and accuracy of the proposed method to obtain analytical solutions to differential equations
within the local fractional derivatives.

1. Introduction different approaches in definition of the local fractional


derivative exist, among them the local fractional derivative
Fractional calculus theory [1–3] has been applied to a wide of Kolwankar et al. [32–38], the fractal derivative of Chen et
class of complex problems encompassing physics, biology, al. [39, 40], the fractal derivative of Parvate et al. [41, 42], the
mechanics, and interdisciplinary areas [4–9]. Various meth- modified Riemann-Liouville of Jumarie [43, 44], and versions
ods, for example, the Adomian decomposition method [10], described in [45–52].
the Rach-Adomian-Meyers modified decomposition method In order to deal with local fractional ordinary and partial
[11], the variational iteration method [12, 13], the homotopy differential equations, there are some developed technologies,
perturbation method [13, 14], the fractal Laplace and Fourier for example, the local fractional variational iteration method
transforms [15], the homotopy analysis method [16], the heat- [45, 46], the local fractional Fourier series method [47,
balance integral method [17–19], the fractional variational 48], the Cantor-type cylindrical-coordinate method [49],
iteration method [20–22], the fractional subequation method the Yang-Fourier transform [50, 51], and the Yang-Laplace
[23, 24], and the generalized Exp-function method [25], have transform [52].
been utilized to solve fractional differential equations [3, 15]. The local fractional derivative is defined as follows [26–
The characteristics of fractal materials have local and 31, 45–52]:
fractal behaviors well described by nondifferential functions.
However, the classic fractional calculus is not valid for 𝑑𝛼 𝑓 (𝑥) 󵄨󵄨󵄨󵄨 Δ𝛼 (𝑓 (𝑥) − 𝑓 (𝑥0 ))
𝑓(𝛼) (𝑥0 ) = 󵄨
󵄨 = lim 𝛼 , (1)
differential equation on Cantor sets due to its no-local nature. 𝑑𝑥𝛼 󵄨󵄨 𝑥=𝑥 0
𝑥 → 𝑥0 (𝑥 − 𝑥0 )
In contrast, the local fractional calculus is one of the best
candidates for dealing with such problems [26–44]. The local
fractional calculus theory has played crucial applications in where Δ𝛼 (𝑓(𝑥) − 𝑓(𝑥0 )) ≅ Γ(1 + 𝛼)Δ(𝑓(𝑥) − 𝑓(𝑥0 )), and 𝑓(𝑥)
several fields, such as theoretical physics, transport problems is satisfied with the condition [26, 47]
in fractal media described by nondifferential functions. There
󵄨󵄨 󵄨 𝛼󵄨 󵄨𝛼
are some versions of the local fractional calculus where 󵄨󵄨𝑓 (𝑥) − 𝑓 (𝑥0 )󵄨󵄨󵄨 ≤ 𝜏 󵄨󵄨󵄨𝑥 − 𝑥0 󵄨󵄨󵄨 (2)
2 Abstract and Applied Analysis

so that [26–31] Hence, from (10) we can obtain a recursion; namely,


󵄨󵄨 󵄨 𝑋𝑖+𝑛 (𝑥) = (𝐿 𝛼 𝑋𝑖 ) (𝑥) ,
󵄨󵄨𝑓 (𝑥) − 𝑓 (𝑥0 )󵄨󵄨󵄨 < 𝜀
𝛼 (11)
(3)
with 𝑛 = 1; we arrive at the following relation:
with 𝑈 : |𝑥 − 𝑥0 | < 𝛿, for 𝜀, 𝛿 > 0 and 𝜀, 𝛿 ∈ 𝑅.
The main idea of this paper is to present the local 𝑋𝑖+1 (𝑥) = (𝐿 𝛼 𝑋𝑖 ) (𝑥) , (12)
fractional series expansion method for effective solutions with 𝑛 = 2; we may rewrite (11) as
of wave and diffusion equations on Cantor sets involving
local fractional derivatives. The paper has been organized 𝑋𝑖+2 (𝑥) = (𝐿 𝛼 𝑋𝑖 ) (𝑥) . (13)
as follows. Section 2 gives a local fractional series expansion
By the recursion formulas, we can obtain the solution of (4)
method. Some illustrative examples are shown in Section 3.
as
The conclusions are presented in Section 4.

𝑡𝑖𝛼
𝑢 (𝑥, 𝑡) = ∑ 𝑋𝑖 (𝑥) . (14)
2. Analysis of the Method 𝑖=0 Γ (1 + 𝑖𝛼)

Let us consider the local fractional differential equation The convergent condition is
𝑡𝑖𝛼
𝑢𝑡𝑛𝛼 = 𝐿 𝛼 𝑢, (4) lim [ 𝑋 (𝑥)] = 0. (15)
𝑛→∞ Γ (1 + 𝑖𝛼) 𝑖
where 𝐿 is a linear local operator with respect to 𝑥, 𝑛 ∈ {1, 2}. This approach is termed the local fractional series expansion
In accordance with the results in [28, 47], there are method (LFSEM)
multiterm separated functions of independent variables 𝑡 and
𝑥, namely,
3. Applications to Wave and Diffusion
∞ Equations on Cantor Sets
𝑢 (𝑥, 𝑡) = ∑𝑇𝑖 (𝑡) 𝑋𝑖 (𝑥) , (5)
𝑖=0 In this section, four examples for wave and diffusion equa-
tions on Cantor sets will demonstrate the efficiency of
where 𝑇𝑖 (𝑡) and 𝑋𝑖 (𝑥) are local fractional continuous func- LFSEM.
tions.
Moreover, there is a nondifferential series term Example 1. Let us consider the diffusion equation on Cantor
𝑖𝛼 set
𝑡
𝑇𝑖 (𝑡) = 𝑝𝑖 , (6)
𝑢𝑡𝛼 (𝑥, 𝑡) − 𝑢𝑥2𝛼 (𝑥, 𝑡) = 0, 0<𝛼≤1 (16)
Γ (1 + 𝑖𝛼)
with the initial condition
where 𝑝𝑖 is a coefficient.
In view of (6), we may present the solution in the form 𝑥𝛼
𝑢 (𝑥, 0) = . (17)
Γ (1 + 𝛼)
∞ 𝑖𝛼
𝑡
𝑢 (𝑥, 𝑡) = ∑𝑝𝑖 𝑋 (𝑥) . (7) Following (12), we have recursive formula
𝑖=0 Γ (1 + 𝑖𝛼) 𝑖
𝜕2𝛼 𝑋𝑖 (𝑥)
𝑋𝑖+1 (x) = ,
Then, following (7), we have 𝜕𝑥2𝛼
(18)

𝑡 𝑖𝛼 𝑥𝛼
𝑢 (𝑥, 𝑡) = ∑ 𝑋 (𝑥) . 𝑋0 (𝑥) = .
Γ (1 + 𝑖𝛼) 𝑖
(8) Γ (1 + 𝛼)
𝑖=0
Hence, we get
Hence,
𝑥𝛼
∞ ∞
𝑋0 (𝑥) = ,
1 1 Γ (1 + 𝛼)
𝑢𝑡𝑛𝛼 = ∑ 𝑡𝑖𝛼 𝑋𝑖+1 (𝑥) = ∑ 𝑡𝑖𝛼 𝑋𝑖+𝑛 (𝑥) ,
𝑖=0 Γ (1 + 𝑖𝛼) 𝑖=0 Γ (1 + 𝑖𝛼) 𝑋1 (𝑥) = 0,
(19)
∞ ∞
𝑡𝑖𝛼 𝑡𝑖𝛼 𝑋2 (𝑥) = 0,
𝐿 𝛼 𝑢 = 𝐿 𝛼 [∑ 𝑋𝑖 (𝑥)] = ∑ (𝐿 𝛼 𝑋𝑖 ) (𝑥) .
𝑖=0 Γ (1 + 𝑖𝛼) 𝑖=0 Γ (1 + 𝑖𝛼) ..
(9) .

In view of (9), we have and so on.


Therefore, through (19) we get the solution
∞ ∞
1 𝑡𝑖𝛼 𝑥𝛼
∑ 𝑡𝑖𝛼 𝑋𝑖+𝑛 (𝑥) = ∑ (𝐿 𝛼 𝑋𝑖 ) (𝑥) . (10) 𝑢 (𝑥, 𝑡) = . (20)
𝑖=0 Γ (1 + 𝑖𝛼) 𝑖=0 Γ (1 + 𝑖𝛼) Γ (1 + 𝛼)
Abstract and Applied Analysis 3

Example 2. Let us consider the diffusion equation on Cantor Hence, using the relations (29), the recursive calculations
set yield
𝑥2𝛼
𝑢𝑡𝛼 (𝑥, 𝑡) − ⋅ 𝑢2𝛼 (𝑥, 𝑡) = 0, 0<𝛼≤1 (21) 𝑥𝛼
Γ (1 + 2𝛼) 𝑥 𝑋0 (𝑥) = ,
Γ (1 + 𝛼) (29)
with the initial condition
𝑋1 (𝑥) = 1,
𝑥2𝛼
𝑢 (𝑥, 0) = . (22) 𝑋2 (𝑥) = 0,
Γ (1 + 2𝛼)
𝑋3 (𝑥) = 0,
Following (12), we get
(30)
𝑋4 (𝑥) = 0,
2𝛼 2𝛼
𝑥 𝜕 𝑋i (𝑥)
𝑋𝑖+1 (𝑥) = , ..
Γ (1 + 2𝛼) 𝜕𝑥2𝛼 .
(23)
𝑥2𝛼
𝑋0 (𝑥) = . and so on.
Γ (1 + 2𝛼)
Finally, we obtain
By using the recursive formula (23), we get consequently
𝑥𝛼 𝑡2𝛼
𝑥2𝛼 𝑢 (𝑥, 𝑡) = + . (31)
𝑋0 (𝑥) = , Γ (1 + 𝛼) Γ (1 + 2𝛼)
Γ (1 + 2𝛼)
𝑥2𝛼 Example 4. Let us consider the wave equation on Cantor sets
𝑋1 (𝑥) = , [26, 30]
Γ (1 + 2𝛼)
(24)
2𝛼
𝑥 𝑢𝑡2𝛼 (𝑥, 𝑡) − cu2𝛼
𝑋2 (𝑥) = , 𝑥 (𝑥, 𝑡) = 0, 0 < 𝛼 ≤ 1, (32)
Γ (1 + 2𝛼)
.. where c is a constant.
. The initial condition is

As a direct result of these recursive calculations, we arrive at 𝑢 (𝑥, 0) = 𝐸𝛼 (𝑥𝛼 ) . (33)


2𝛼 ∞ 𝑖𝛼 2𝛼
𝑥 𝑡 𝑥
𝑢 (𝑥, 𝑡) = ∑ = 𝐸 (𝑡𝛼 ) .
Γ (1 + 2𝛼) 𝑖=0 Γ (1 + 𝑖𝛼) Γ (1 + 2𝛼) 𝛼 By using (14) we have
(25)
𝜕2𝛼 𝑋𝑖 (𝑥)
𝑋𝑖+2 (𝑥) = 𝑐 ,
Example 3. Let us consider the following wave equation on 𝜕𝑥2𝛼
Cantor sets:
𝑋0 (𝑥) = 𝑢 (𝑥, 0) = 𝐸𝛼 (𝑥𝛼 ) ,
2𝛼
𝑥 (34)
𝑢𝑡2𝛼 (𝑥, 𝑡) − ⋅ 𝑢2𝛼 (𝑥, 𝑡) = 0, 0<𝛼≤1 (26) 𝜕2𝛼 𝑋𝑖 (𝑥)
Γ (1 + 2𝛼) 𝑥 𝑋𝑖+2 (𝑥) = 𝑐 ,
𝜕𝑥2𝛼
with the initial condition
𝑋0 (𝑥) = 𝑢𝑥(𝛼) (𝑥, 0) = 𝐸𝛼 (𝑥𝛼 ) .
𝑥𝛼
𝑢 (𝑥, 0) = . (27)
Γ (1 + 𝛼) Then, through the iterative relations (35), we have
In view of (14), we obtain
𝑋0 (𝑥) = 𝐸𝛼 (𝑥𝛼 ) ,
2𝛼 2𝛼
𝑥 𝜕 𝑋𝑖 (𝑥) (35)
𝑋i+2 (𝑥) = , 𝑋1 (𝑥) = 𝐸𝛼 (𝑥𝛼 ) ,
Γ (1 + 2𝛼) 𝜕𝑥2𝛼
𝑥𝛼 𝑋2 (𝑥) = 𝑐𝐸𝛼 (𝑥𝛼 ) ,
𝑋0 (𝑥) = 𝑢 (𝑥, 0) = ,
Γ (1 + 𝛼) (28) 𝑋3 (𝑥) = 𝑐𝐸𝛼 (𝑥𝛼 ) ,
𝑥2𝛼 𝜕2𝛼 𝑋𝑖 (𝑥) (36)
𝑋𝑖+2 (𝑥) = , 𝑋4 (𝑥) = 𝑐2 𝐸𝛼 (𝑥𝛼 ) ,
Γ (1 + 2𝛼) 𝜕x2𝛼
..
𝑋1 (𝑥) = 𝑢𝑥(𝛼) (𝑥, 0) = 1. .
4 Abstract and Applied Analysis

Therefore, we obtain [7] B. J. West, M. Bologna, and P. Grigolini, Physics of Fractal


Operators, Springer, New York, NY, USA, 2003.

𝑡2𝑖𝛼
𝑢 (𝑥, 𝑡) = 𝐸𝛼 (𝑥𝛼 ) ∑𝑐𝑖 [8] V. E. Tarasov, Fractional Dynamics: Applications of Fractional
𝑖=0 Γ (1 + 2𝑖𝛼) Calculus to Dynamics of Particles, Fields and Media, Springer,
Berlin, Germany, 2011.

𝑡(2𝑖+1)𝛼 (37) [9] J. A. Tenreiro Machado, A. C. J. Luo, and D. Baleanu, Nonlinear
+ 𝐸𝛼 (𝑥𝛼 ) ∑𝑐𝑖 Dynamics of Complex Systems: Applications in Physical, Biologi-
𝑖=0 Γ (1 + (2𝑖 + 1) 𝛼)
cal and Financial Systems, Springer, New York, NY, USA, 2011.
= 𝐸𝛼 (𝑥 ) [cosh𝛼 (𝑐𝑡𝛼 ) + sin𝛼 (𝑐𝑡𝛼 )] ,
𝛼 [10] J. S. Duan, T. Chaolu, R. Rach, and L. Lu, “The Adomian decom-
position method with convergence acceleration techniques
where for nonlinear fractional differential equations,” Computers &
Mathematics With Applications, 2013.

𝑡2𝑖𝛼 [11] J. S. Duan, T. Chaolu, and R. Rach, “Solutions of the initial value
cosh𝛼 (𝑡𝛼 ) = ∑ , problem for nonlinear fractional ordinary differential equa-
𝑖=0 Γ (1 + 2𝑖𝛼)
(38) tions by the Rach-Adomian-Meyers modified decomposition
∞ method,” Applied Mathematics and Computation, vol. 218, no.
𝛼 𝑡(2𝑖+1)𝛼
sinh𝛼 (𝑡 ) = ∑ . 17, pp. 8370–8392, 2012.
𝑖=0 Γ (1 + (2𝑖 + 1) 𝛼) [12] S. Das, “Analytical solution of a fractional diffusion equation by
variational iteration method,” Computers & Mathematics with
For more details concerning (38), we refer to [26–28]. Applications, vol. 57, no. 3, pp. 483–487, 2009.
[13] S. Momani and Z. Odibat, “Comparison between the homotopy
4. Conclusions perturbation method and the variational iteration method for
linear fractional partial differential equations,” Computers &
In this work, the local fractional series expansion method Mathematics with Applications, vol. 54, no. 7-8, pp. 910–919,
is demonstrated as an effective method for solutions of a 2007.
wide class of problems. Analytical solutions of the wave and [14] S. Momani and Z. Odibat, “Homotopy perturbation method
diffusion equations on Cantor sets involving local fractional for nonlinear partial differential equations of fractional order,”
derivatives are successfully developed by recurrence relations Physics Letters A, vol. 365, no. 5-6, pp. 345–350, 2007.
resulting in convergent series solutions. In this context, the [15] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional
suggested method is a potential tool for development of Calculus Models and Numerical Methods, Series on Complexity,
approximate solutions of local fractional differential equa- Nonlinearity and Chaos, World Scientific, Boston, Mass, USA,
tions with fractal initial value conditions, which, of course, 2012.
draws new problems beyond the scope of the present work. [16] H. Jafari and S. Seifi, “Homotopy analysis method for solving
linear and nonlinear fractional diffusion-wave equation,” Com-
munications in Nonlinear Science and Numerical Simulation, vol.
Acknowledgments 14, no. 5, pp. 2006–2012, 2009.
The first author was supported by the National Scientific [17] J. Hristov, “Heat-balance integral to fractional (half-time) heat
diffusion sub-model,” Thermal Science, vol. 14, no. 2, pp. 291–316,
and Technological Support Projects (no. 2012BAE09B00),
2010.
the National Natural Science Foundation of China (no.
11126213 and no. 61170317), and the National Natural Science [18] J. Hristov, “Integral-balance solution to the stokes’ first problem
of a viscoelastic generalized second grade fluid,” Thermal
Foundation of Hebei Province (no. E2013209123). The third
Science, vol. 16, no. 2, pp. 395–410, 2012.
author is supported in part by NSF11061028 of China and
Yunnan Province NSF Grant no. 2011FB090. [19] J. Hristov, “Transient flow of a generalized second grade fluid
due to a constant surface shear stress: an approximate integral-
balance solution,” International Review of Chemical Engineering,
References vol. 3, no. 6, pp. 802–809, 2011.
[20] G. C. Wu and E. W. M. Lee, “Fractional variational iteration
[1] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory method and its application,” Physics Letters A, vol. 374, no. 25,
and Applications of Fractional Differential Equations, Elsevier, pp. 2506–2509, 2010.
Amsterdam, The Netherlands, 2006.
[21] Y. Khan, N. Faraz, A. Yildirim, and Q. Wu, “Fractional
[2] F. Mainardi, Fractional Calculus and Waves in Linear Viscoelas- variational iteration method for fractional initial-boundary
ticity, Imperial College Press, London, UK, 2010. value problems arising in the application of nonlinear science,”
[3] I. Podlubny, Fractional Differential Equations, Academic Press, Computers & Mathematics with Applications, vol. 62, no. 5, pp.
San Diego, Calif, USA, 1999. 2273–2278, 2011.
[4] R. L. Magin, Fractional Calculus in Bioengineering,, Begerll [22] G. C. Wu, “A fractional variational iteration method for solving
House, West Redding, Conn, USA, 2006. fractional nonlinear differential equations,” Computers & Math-
[5] J. Klafter, S. C. Lim, and R. Metzler, Fractional Dynamics in ematics with Applications, vol. 61, no. 8, pp. 2186–2190, 2011.
Physics: Recent Advances, World Scientific, Singapore, 2011. [23] S. Zhang and H.-Q. Zhang, “Fractional sub-equation method
[6] G. M. Zaslavsky, Hamiltonian Chaos and Fractional Dynamics, and its applications to nonlinear fractional PDEs,” Physics
Oxford University Press, Oxford, UK, 2008. Letters A, vol. 375, no. 7, pp. 1069–1073, 2011.
Abstract and Applied Analysis 5

[24] H. Jafari, H. Tajadodi, N. Kadkhoda, and D. Baleanu, “Frac- information of non-random functions,” Chaos, Solitons and
tional subequation method for Cahn-Hilliard and Klein- Fractals, vol. 40, no. 3, pp. 1428–1448, 2009.
Gordon equations,” Abstract and Applied Analysis, vol. 2013, [44] G. Jumarie, “Laplace’s transform of fractional order via the
Article ID 587179, 5 pages, 2013. Mittag-Leffler function and modified Riemann-Liouville deriv-
[25] S. Zhang, Q. A. Zong, D. Liu, and Q. Gao, “A generalized exp- ative,” Applied Mathematics Letters, vol. 22, no. 11, pp. 1659–1664,
function method for fractional Riccati differential equations,” 2009.
Communications in Fractional Calculus, vol. 1, pp. 48–52, 2010. [45] X. J. Yang and D. Baleanu, “Fractal heat conduction problem
[26] X. J. Yang, Advanced Local Fractional Calculus and Its Applica- solved by local fractional variation iteration method,” Thermal
tions, World Science, New York, NY, USA, 2012. Science, vol. 17, no. 2, pp. 625–628, 2013.
[27] X. J. Yang, “Local fractional integral transforms,” Progress in [46] W. H. Su, D. Baleanu, X.-J. Yang, and H. Jafari, “Damped wave
Nonlinear Science, vol. 4, pp. 1–225, 2011. equation and dissipative wave equation in fractal strings within
[28] X. J. Yang, Local Fractional Functional Analysis and Its Applica- the local fractional variational iteration method,” Fixed Point
tions, Asian Academic, Hong Kong, 2011. Theory and Applications, vol. 2013, no. 1, pp. 89–102, 2013.
[29] W. P. Zhong, X. J. Yang, and F. Gao, “A Cauchy problem for [47] M. S. Hu, R. P. Agarwal, and X.-J. Yang, “Local fractional Fourier
some local fractional abstract differential equation with fractal series with application to wave equation in fractal vibrating
conditions,” Journal of Applied Functional Analysis, vol. 8, no. 1, string,” Abstract and Applied Analysis, vol. 2012, Article ID
pp. 92–99, 2013. 567401, 15 pages, 2012.
[30] W. H. Su, X. J. Yang, H. Jafari, and D. Baleanu, “Fractional [48] G. A. Anastassiou and O. Duman, Advances in Applied Math-
complex transform method for wave equations on Cantor ematics and Approximation Theory, Springer, New York, NY,
sets within local fractional differential operator,” Advances in USA, 2013.
Difference Equations, vol. 2013, no. 1, pp. 97–107, 2013. [49] X. J. Yang, H. M. Srivastava, J. H. He, and D. Baleanu, “Cantor-
[31] M. S. Hu, D. Baleanu, and X. J. Yang, “One-phase problems type cylindrical-coordinate method for differential equations
for discontinuous heat transfer in fractal media,” Mathematical with local fractional derivatives,” Physics Letters A, vol. 377, no.
Problems in Engineering, vol. 2013, Article ID 358473, 3 pages, 28–30, pp. 1696–1700, 2013.
2013. [50] X. Yang, M. Liao, and J. Chen, “A novel approach to pro-
[32] K. M. Kolwankar and A. D. Gangal, “Local fractional Fokker- cessing fractal signals using the Yang-Fourier transforms,” in
Planck equation,” Physical Review Letters, vol. 80, no. 2, pp. 214– Proceedings of the International Workshop on Information and
217, 1998. Electronics Engineering (IWIEE ’12), vol. 29, pp. 2950–2954,
March 2012.
[33] A. Carpinteri and A. Sapora, “Diffusion problems in fractal
media defined on Cantor sets,” ZAMM Journal of Applied [51] W. Zhong, F. Gao, and X. Shen, “Applications of Yang-Fourier
Mathematics and Mechanics, vol. 90, no. 3, pp. 203–210, 2010. transform to local fractional equations with local fractional
derivative and local fractional integral,” Advanced Materials
[34] A. Carpinteri and P. Cornetti, “A fractional calculus approach to Research, vol. 461, pp. 306–310, 2012.
the description of stress and strain localization in fractal media,”
Chaos, Solitons and Fractals, vol. 13, no. 1, pp. 85–94, 2002. [52] W. P. Zhong and F. Gao, “Application of the Yang-Laplace trans-
forms to solution to nonlinear fractional wave equation with
[35] A. Carpinteri, B. Chiaia, and P. Cornetti, “The elastic problem fractional derivative,” in Proceedings of the 3rd International
for fractal media: basic theory and finite element formulation,” Conference on Computer Technology and Development, pp. 209–
Computers and Structures, vol. 82, no. 6, pp. 499–508, 2004. 213, ASME, 2011.
[36] A. Babakhani and V. Daftardar-Gejji, “On calculus of local
fractional derivatives,” Journal of Mathematical Analysis and
Applications, vol. 270, no. 1, pp. 66–79, 2002.
[37] F. Ben Adda and J. Cresson, “About non-differentiable func-
tions,” Journal of Mathematical Analysis and Applications, vol.
263, no. 2, pp. 721–737, 2001.
[38] Y. Chen, Y. Yan, and K. Zhang, “On the local fractional deriv-
ative,” Journal of Mathematical Analysis and Applications, vol.
362, no. 1, pp. 17–33, 2010.
[39] W. Chen, “Time-space fabric underlying anomalous diffusion,”
Chaos, Solitons and Fractals, vol. 28, no. 4, pp. 923–925, 2006.
[40] W. Chen, H. Sun, X. Zhang, and D. Korošak, “Anomalous diffu-
sion modeling by fractal and fractional derivatives,” Computers
& Mathematics with Applications, vol. 59, no. 5, pp. 1754–1758,
2010.
[41] A. K. Golmankhaneh, V. Fazlollahi, and D. Baleanu, “Newto-
nian mechanics on fractals subset of real-line,” Romania Reports
in Physics, vol. 65, pp. 84–93, 2013.
[42] A. Parvate and A. D. Gangal, “Fractal differential equations and
fractal-time dynamical systems,” Pramana, vol. 64, no. 3, pp.
389–409, 2005.
[43] G. Jumarie, “Probability calculus of fractional order and frac-
tional Taylor’s series application to Fokker-Planck equation and
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 160681, 8 pages
http://dx.doi.org/10.1155/2013/160681

Research Article
Nonlinear Fractional Jaulent-Miodek and Whitham-Broer-Kaup
Equations within Sumudu Transform

Abdon Atangana1 and Dumitru Baleanu2,3,4


1
Institute for Groundwater Studies, Faculty of Natural and Agricultural Sciences, University of the Free State,
Bloemfontein 9300, South Africa
2
Department of Mathematics and Computer Sciences, Faculty of Art and Sciences, Cankaya University, Balgat, 06530 Ankara, Turkey
3
Department of Chemical and Materials Engineering, Faculty of Engineering, King Abdulaziz University, P.O. Box 80204,
Jeddah 21589, Saudi Arabia
4
Institute of Space Sciences, P.O. Box MG-23, R 76900, Magurele-Bucharest, Romania

Correspondence should be addressed to Abdon Atangana; abdonatangana@yahoo.fr

Received 15 April 2013; Accepted 28 April 2013

Academic Editor: Soheil Salahshour

Copyright © 2013 A. Atangana and D. Baleanu. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

We solve the system of nonlinear fractional Jaulent-Miodek and Whitham-Broer-Kaup equations via the Sumudu transform
homotopy method (STHPM). The method is easy to apply, accurate, and reliable.

1. Introduction The nonlinear FWBK equation which will be considered


in this paper has the following form:
Nonlinear partial differential equations arise in various areas 𝜂
of physics, mathematics, and engineering [1–4]. We notice 𝜕𝑡 𝑢 + 𝑢𝑢𝑥 + 𝑢𝑥 + 𝛽𝑢𝑥𝑥 = 0, 0 < 𝜂, 𝜇 ≤ 1,
that in fluid dynamics, the nonlinear evolution equations 𝜇
show up in the context of shallow water waves. Some 𝜕𝑡 V + (𝑢V)𝑥 + 𝛼𝑢𝑥𝑥𝑥 − 𝛽V𝑥𝑥 = 0, (𝑥, 𝑡) ∈ [𝑎, 𝑏] × [0, 𝑇] ,
of the commonly studied equations are the Korteweg-de (1)
Vries (KdV) equation, modified KdV equation, Boussinesq
equation [5], Green-Naghdi equation, Gardeners equation, and the nonlinear FJM equation is
and Whitham-Broer-Kaup and Jaulent-Miodek (JM) equa- 3 9 3
tions. Analytical solutions of these equations are usually 𝜕𝑡𝛼 𝑢 + 𝑢𝑥𝑥𝑥 + VV𝑥𝑥𝑥 + V𝑥 V𝑥𝑥 − 6𝑢𝑢𝑥 + 6𝑢VV𝑥 − 𝑢𝑥 V𝑥2
2 2 2
not available. Since only limited classes of equations are
solved by analytical means, numerical solution of these = 0,
nonlinear partial differential equations is of practical impor-
𝜇 15 2
tance. Therefore, finding new methods and techniques to 𝜕𝑡 V + V𝑥𝑥𝑥 − 6𝑢𝑥 V𝑥 − V V = 0, (𝑥, 𝑡) ∈ [𝑎, 𝑏] × [0, 𝑇] .
deal with these type of equations is still an open prob- 2 𝑥
lem in this area. The purpose of this paper is to find an (2)
approximated solution for the system of fractional Jaulent- The system of (1) and (2) is subjected to the following initial
Miodek and Whitham-Broer-Kaup equations (FWBK) via conditions:
the Sumudu transform method. The fractional systems of par-
tial differential equations under investigation here are given 𝑢 (𝑥, 0) = 𝑓 (𝑥) ,
below. (3)
V (𝑥, 0) = 𝑔 (𝑥) .
2 Abstract and Applied Analysis

FWBK equation (1) describes the dispersive long wave in Definition 5. The Jumarie Fractional order derivative is given
shallow water, where 𝑢(𝑥, 𝑡) is the field of horizontal velocity, as follows [24]:
V(𝑥, 𝑡) is the height which deviates from the equilibrium
position of liquid, and 𝛼 and 𝛽 are constants that represent 1 𝑑𝑛
𝐷𝑥𝛼 (𝑓 (𝑥)) =
different powers. If 𝛼 = 0 and 𝛽 = 1, (1) reduces to Γ (𝑛 − 𝛼) 𝑑𝑥𝑛
the classical long-wave equations which describe the shallow 𝑥
water wave with diffusion [6]. If 𝛼 = 1 and 𝛽 = 0, (8)
× ∫ (𝑥 − 𝑡)𝑛−𝛼−1 {𝑓 (𝑡) − 𝑓 (0)} 𝑑𝑡,
(1) becomes the modified Boussinesq equations [7, 8]. FJM 0
equation (2) appears in several areas of science such as 𝑛 − 1 ≤ 𝛼 ≤ 𝑛.
condense matter physics [9], fluid mechanics [10], plasma
physics [11], and optics [12] and associates with energy-
Lemma 6. If 𝑚 − 1 < 𝛼 ≤ 𝑚, 𝑚 ∈ N and 𝑓 ∈ 𝐶𝜇𝑚 , 𝜇 ≥ −1,
dependent Schrödinger potential [13, 14].
then
The paper is organized as follows. In Section 2, we intro-
duce briefly some of the basic tools of fractional order and 𝐷𝛼 𝐽𝛼 𝑓 (𝑥) = 𝑓 (𝑥) ,
of the Sumudu transform method. We show the numerical
𝑚−1
results in Section 4. The conclusions can be seen in Section 5. 𝑥𝑘 (9)
𝐽𝛼 𝐷0𝛼 𝑓 (𝑥) = 𝑓 (𝑥) − ∑ 𝑓(𝑘) (0+ ) , 𝑥 > 0.
𝑘=0
𝑘!
2. Basic Tools
Definition 7 (partial derivatives of fractional order [15, 16, 19]).
2.1. Properties and Definitions Assume now that 𝑓(x) is a function of 𝑛 variables 𝑥𝑖 , 𝑖 =
Definition 1 (see [15–24]). A real function 𝑓(𝑥), 𝑥 > 0, is said 1, . . . , 𝑛 also of class 𝐶 on 𝐷 ∈ R𝑛 . As an extension of
to be in the space 𝐶𝜇 , 𝜇 ∈ R if there exists a real number Definition 3, we define partial derivative of order 𝛼 for 𝑓 with
respect to 𝑥𝑖 the function
𝑝 > 𝜇, such that 𝑓(𝑥) = 𝑥𝑝 ℎ(𝑥), where ℎ(𝑥) ∈ 𝐶[0, ∞), and
it is said to be in space 𝐶𝜇𝑚 if 𝑓(𝑚) ∈ 𝐶𝜇 , 𝑚 ∈ N. 1 𝑥𝑖 󵄨󵄨
𝑚−𝛼−1 𝑚 󵄨
𝑎𝜕𝛼x 𝑓 = ∫ (𝑥𝑖 − 𝑡) 𝜕𝑥𝑖 𝑓 (𝑥𝑗 )󵄨󵄨󵄨 𝑑𝑡, (10)
Γ (𝑚 − 𝛼) 𝑎 󵄨󵄨𝑥𝑗=𝑡
Definition 2 (see [15–24]). The Riemann-Liouville fractional
integral operator of order 𝛼 ≥ 0, of a function𝑓 ∈ 𝐶𝜇 , 𝜇 ≥ −1,
where 𝜕𝑥𝑚𝑖 is the usual partial derivative of integer order 𝑚.
is defined as
𝑥
1
𝐽𝛼 𝑓 (𝑥) = ∫ (𝑥 − 𝑡)𝛼−1 𝑓 (𝑡) 𝑑𝑡, 𝛼 > 0, 𝑥 > 0, 3. Background of Sumudu Transform
Γ (𝛼) 0 (4)
Definition 8 (see [25]). The Sumudu transform of a function
𝐽0 𝑓 (𝑥) = 𝑓 (𝑥) . 𝑓(𝑡), defined for all real numbers 𝑡 ≥ 0, is the function 𝐹𝑠 (𝑢),
Properties of the operator can be found in [15–23]; we defined by
mention only the following. ∞
For 𝑓 ∈ 𝐶𝜇 , 𝜇 ≥ −1, 𝛼, 𝛽 ≥ 0 and 𝛾 > −1 1 𝑡
𝑆 (𝑓 (𝑡)) = 𝐹𝑠 (𝑢) = ∫ exp [− ] 𝑓 (𝑡) 𝑑𝑡. (11)
0 𝑢 𝑢
𝐽𝛼 𝐽𝛽 𝑓 (𝑥) = 𝐽𝛼+𝛽 𝑓 (𝑥) , 𝐽𝛼 𝐽𝛽 𝑓 (𝑥) = 𝐽𝛽 𝐽𝛼 𝑓 (𝑥) ,
Theorem 9 (see [26]). Let 𝐺(𝑢) be the Sumudu transform of
𝛼 𝛾 Γ (𝛾 + 1) 𝛼+𝛾 (5) 𝑓(𝑡) such that
𝐽 𝑥 = 𝑥 .
Γ (𝛼 + 𝛾 + 1) (i) (𝐺(1/𝑠)/𝑠) is a meromorphic function, with singulari-
Definition 3. The Caputo fractional order derivative is given ties having Re[𝑠] ≤ 𝛾;
as follows [15–18]: (ii) there exist a circular region Γ with radius 𝑅 and positive
𝐶 𝛼 constants 𝑀 and 𝐾 with |𝐺(1/𝑠)/𝑠| < 𝑀𝑅−𝐾 , then the
0 𝐷𝑥 (𝑓 (𝑥))
function 𝑓(𝑡) is given by
1 𝑥 𝑑𝑛 𝑓 (𝑡)
= ∫ (𝑥 − 𝑡)𝑛−𝛼−1 𝑑𝑡, 𝑛 − 1 ≤ 𝛼 ≤ 𝑛. 1 𝛾+𝑖∞ 1 𝑑𝑠
Γ (𝑛 − 𝛼) 0 𝑑𝑡𝑛 𝑆−1 (𝐺 (𝑠)) = ∫ exp [𝑠𝑡] 𝐺 ( )
(6) 2𝜋𝑖 𝛾−𝑖∞ 𝑠 𝑠
(12)
Definition 4. The Riemann-Liouville fractional order deriva- 𝐺 (1/𝑠)
= ∑ residual [exp [𝑠𝑡] ].
tive is given as follows [16–24]: 𝑠
𝐷𝑥𝛼 (𝑓 (𝑥))
For the proof see [26].
𝑛 𝑥
1 𝑑
= ∫ (𝑥 − 𝑡)𝑛−𝛼−1 𝑓 (𝑡) 𝑑𝑡, 𝑛 − 1 ≤ 𝛼 ≤ 𝑛.
Γ (𝑛 − 𝛼) 𝑑𝑥𝑛 0 3.1. Basics of the Sumudu Transform Homotopy Perturbation
(7) Method. We illustrate the basic idea of this method [27–32]
Abstract and Applied Analysis 3

by considering a general fractional nonlinear nonhomoge- which is the coupling of the Sumudu transform and the
neous partial differential equation with the initial condition HPM using He’s polynomials. Comparing the coefficients of
of the following form: like powers of 𝑝, the following approximations are obtained
[29, 30]:
𝐷𝑡𝛼 𝑈 (𝑥, 𝑡) = 𝐿 (𝑈 (𝑥, 𝑡)) + 𝑁 (𝑈 (𝑥, 𝑡)) + 𝑓 (𝑥, 𝑡) , 𝛼 > 0,
(13)
𝑝0 :𝑈0 (𝑥, 𝑡) = 𝐺 (𝑥, 𝑡) ,
subject to the initial condition
𝑝1 :𝑈1 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑈0 (𝑥, 𝑡)) + 𝐻0 (𝑈)]] ,
𝐷0𝑘 𝑈 (𝑥, 0) = 𝑔𝑘 , (𝑘 = 0, . . . , 𝑛 − 1) ,
(14)
𝐷0𝑛 𝑈 (𝑥, 0) = 0, 𝑛 = [𝛼] , 𝑝2 :𝑈2 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑈1 (𝑥, 𝑡)) + 𝐻1 (𝑈)]] , (22)

where 𝐷𝑡𝛼 denotes without loss of generality the Caputo 𝑝3 :𝑈3 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑈2 (𝑥, 𝑡)) + 𝐻2 (𝑈)]] ,
fraction derivative operator, 𝑓 is a known function, 𝑁 is
the general nonlinear fractional differential operator, and 𝐿 𝑝𝑛 :𝑈𝑛 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑈𝑛−1 (𝑥, 𝑡)) + 𝐻𝑛−1 (𝑈)]] .
represents a linear fractional differential operator.
Applying the Sumudu transform on both sides of (10), we Finally, we approximate the analytical solution 𝑈(𝑥, 𝑡) by
obtain truncated series:
𝑆 [𝐷𝑡𝛼 𝑈 (𝑥, 𝑡]) = 𝑆 [𝐿 (𝑈 (𝑥, 𝑡))]
𝑁
(15)
+ 𝑆 [𝑁 (𝑈 (𝑥, 𝑡))] + 𝑆 [𝑓 (𝑥, 𝑡)] . 𝑈 (𝑥, 𝑡) = lim ∑ 𝑈𝑛 (𝑥, 𝑡) . (23)
𝑁→∞
𝑛=0
Using the property of the Sumudu transform, we have
𝑆 [𝑈 (𝑥, 𝑡)] = 𝑢𝛼 𝑆 [𝐿 (𝑈 (𝑥, 𝑡))] + 𝑢𝛼 𝑆 [𝑁 (𝑈 (𝑥, 𝑡))] The above series solutions generally converge very rapidly
(16) [29, 30].
+ 𝑢𝛼 𝑆 [𝑓 (𝑥, 𝑡)] + 𝑔 (𝑥, 𝑡) .
Now applying the Sumudu inverse on both sides of (12) we 4. Applications
obtain
In this section, we apply this method for solving the system
𝑈 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑈 (𝑥, 𝑡))] + 𝑢𝛼 𝑆 [𝑁 (𝑈 (𝑥, 𝑡))]] of the fractional differential equation. We will start with (1).
(17)
+ 𝐺 (𝑥, 𝑡) ,
4.1. Approximate Solution of (1). Following carefully the
where 𝐺(𝑥, 𝑡) represents the term arising from the known
steps involved in the STHPM, after comparing the terms of
function 𝑓(𝑥, 𝑡) and the initial conditions.
the same power of 𝑝 and choosing the appropriate initials
Now we apply the following HPM:
conditions, we arrive at the following series solutions:

𝑈 (𝑥, 𝑡) = ∑ 𝑝𝑛 𝑈𝑛 (𝑥, 𝑡) . (18) 𝑐1
𝑛=0 𝑢0 (𝑥, 𝑡) = 𝐺 (𝑥, 𝑡) = − + 2𝑐1 √−𝛼 − 𝛽2 sech (𝑐1 𝑥) ,
𝑐2
The nonlinear term can be decomposed to
∞ V0 (𝑥, 𝑡) = 𝐺1 (𝑥, 𝑡)
𝑁𝑈 (𝑥, 𝑡) = ∑ 𝑝𝑛 H𝑛 (𝑈) , (19) 2
𝑛=0 = − 𝑐12 (𝛼 + 𝛽2 ) + 2𝑐12 (𝛼 + 𝛽2 ) sech(𝑐1 𝑥)
using the He’s polynomial H𝑛 (𝑈) given as
+ 2𝑐12 𝛽√−𝛼 − 𝛽2 sech (𝑐1 𝑥) tanh (𝑐1 𝑥) ,
𝑛 ∞
1 𝜕 [
H𝑛 (𝑈0 , . . . , 𝑈𝑛 ) = 𝑁 ( ∑𝑝𝑗 𝑈𝑗 (𝑥, 𝑡))] , 𝑢1 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑢0 (𝑥, 𝑡)) + 𝐻0 (𝑢)]]
𝑛! 𝜕𝑝𝑛 𝑗=0 (20)
[ ]
3
𝑛 = 0, 1, 2 . . . . 𝑐12 𝑡𝜂 sech(𝑐1 𝑥)
=
𝑐2 Γ (𝜂 + 1)
Substituting (15) and (16) gives

× (𝑐1 𝑐2 𝛽√−𝛼 − 𝛽2 cos (2𝑐1 𝑥)
∑ 𝑝𝑛 𝑈𝑛 (𝑥, 𝑡)
𝑛=0


+ 4𝑐1 𝑐2 (𝛼 + 𝛽2 ) sinh (𝑐1 𝑥) + √−𝛼 − 𝛽2
= 𝐺 (𝑥, 𝑡) + 𝑝 [𝑆−1 [𝑢𝛼 𝑆 [𝐿 ( ∑ 𝑝𝑛 𝑈𝑛 (𝑥, 𝑡))] (21)
𝑛=0 × ( − 3𝑐1 𝑐2 𝛽

+ 𝑢𝛼 𝑆 [𝑁 ( ∑ 𝑝𝑛 𝑈𝑛 (𝑥, 𝑡))]]] , + (𝑐1 − 𝑐2 ) sinh (2𝑐1 𝑥) )) ,
𝑛=0
4 Abstract and Applied Analysis

V1 (𝑥, 𝑡) 𝑢(𝑥, 20)


𝑥
−150 −100 −50 50 100 150
= 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (V0 (𝑥, 𝑡)) + 𝐻0 (V)]]
0.99
1
=
𝑐2 Γ (1 + 𝜇) 0.98

× (2𝑐12 𝑡𝜇 ( − 2𝑐1 sech (𝑥) (𝛼 + 𝛽2 ) 0.97


4
− 2𝑐2 𝛽 (𝛼 + 𝛽2 ) sech (𝑐1 𝑥)
0.96

1 5
− √−𝛼 − 𝛽2 sech (𝑐1 𝑥) 0.95
2

× (𝛽 − 28𝑐1 𝑐2 𝛽2 + 𝛽 (1 + 18𝑐1 𝑐2 𝛽) 0.94

Figure 1: Approximate solution for FWBK equation.


× cosh (2𝑐1 𝑥) − 5𝑐2 sinh (2𝑐1 𝑥) )
2 2
+ 2𝑐2 𝛽 (𝛼 + 𝛽2 ) sech (𝑐1 𝑥) tanh (𝑐1 𝑥)
+ 22𝜂+1 𝑐1 𝑐2 √−𝛼 − 𝛽2 cosh (4𝑐1 𝑥)
+ √−𝛼 − 𝛽2 sech (𝑐1 𝑥)
− 4𝜂 𝑐22 √−𝛼 − 𝛽2 cosh (4𝑐1 𝑥)
× (4𝑐1 𝑐2 sech (𝑥) (𝛼 + 𝛽2 )
2
+ tanh (𝑐1 𝑥)
2 − 4𝜂 𝑐22 𝑐1 𝛽2 √−𝛼 − 𝛽2 cosh (4𝑐1 𝑥)

× (𝛽 + 𝑐2 tanh (𝑐1 𝑥) + 11 × 41+𝜂 𝑐2 𝑐21 𝛽2 √−𝛼 − 𝛽2 sinh (2𝑐1 𝑥)

× (−1 + 𝑐1 𝛽2 tanh (𝑐1 𝑥)))))) , − 11 × 41+𝜂 𝑐1 𝑐22 𝛽2 √−𝛼 − 𝛽2 sinh (2𝑐1 𝑥)


𝑢2 (𝑥, 𝑡)
+ 23+2𝜂 𝑐12 𝑐22 𝛽 (𝛼 + 𝛽2 )
−1 𝛼
= 𝑆 [𝑢 𝑆 [𝐿 (𝑢1 (𝑥, 𝑡)) + 𝐻1 (V)]]
× (37 sinh (𝑐1 𝑥) − 3 sinh (3𝑐1 𝑥))

1 − 4𝜂+1 𝑐2 𝑐21 𝛽√−𝛼 − 𝛽2 sinh (4𝑐1 𝑥)


=
𝑐22 Γ (1 + 2𝜂)
5 +4𝜂+1 𝑐1 𝑐22 𝛽√−𝛼 − 𝛽2 sinh (4𝑐1 𝑥)) .
× (4−1−𝜂 𝑐13 𝑡2𝜂 sech(𝑐1 𝑥)
(24)
× (−5 × 4𝜂 √−𝛼 − 𝛽2
And so on in the same manner one can obtain the rest of the
components. However, here, few terms were computed and
× (2𝑐1 𝑐2 − 𝑐22 + 𝑐12 (16𝛼 + 39𝛽2 )) ) the asymptotic solution is given by

+ 3 × 42+𝜂 𝑐1 (𝑐1 − 𝑐2 ) 𝑐2 (𝛼 + 𝛽2 ) cosh (𝑐1 𝑥) 𝑢 (𝑥, 𝑡) = 𝑢0 (𝑥, 𝑡) + 𝑢1 (𝑥, 𝑡) + 𝑢2 (𝑥, 𝑡) + 𝑢3 (𝑥, 𝑡) + ⋅ ⋅ ⋅ ,

+ 41+𝜂 √−𝛼 − 𝛽2 ( − 2𝑐1 𝑐2 + 𝑐22 + 𝑐12 V (𝑥, 𝑡) = V0 (𝑥, 𝑡) + V1 (𝑥, 𝑡) + V2 (𝑥, 𝑡) + V3 (𝑥, 𝑡) + ⋅ ⋅ ⋅ .
(25)
× (1 + 𝑐22 (12𝛼 + 31𝛽2 )))
Figures 1, 2, 3, and 4 show the graphical representation of the
× cosh (2𝑐1 𝑥) + 42+𝜂 𝑐1 𝑐2 (−𝑐1 + 𝑐2 ) (𝛼 + 𝛽2 )
approximated solution of the system of nonlinear fractional
Whitham-Broer-Kaup equation for 𝜂 = 0.9, 𝜇 = 0.98, 𝑐1 =
× cosh (3𝑐1 𝑥) − 4𝜂 𝑐12 √−𝛼 − 𝛽2 cosh (4𝑐1 𝑥) 𝑐2 = 0.1, and 𝛽 = 𝛼 = 0.1.
Abstract and Applied Analysis 5

4.2. Approximate Solution of (2). For (2), in the view of


the Sumudu transform method, by choosing the appropriate
initials conditions we are at the following series solutions:

𝑐2 𝑐𝑥 2
𝑢0 (𝑥, 𝑡) = (1 − sech ( ) ) ,
8 2
1
𝑐𝑥 2
0.99 2200 V0 (𝑥, 𝑡) = 𝑐 sech ( ),
𝑢(𝑥, 𝑡)

2
150
0.98
𝑐5 𝑡𝜂 sech(𝑐𝑥/2)5
𝑢1 (𝑥, 𝑡) = −
100 128Γ (𝜂 + 1)
−100 𝑡
𝑐𝑥 3𝑐𝑥
0 50 × (192 cosh [ ] − 32 cosh [ ]
𝑥 2 2
100 𝑐𝑥 3𝑐𝑥
0 +3𝑐 (3 sinh [ ] + sinh [ ]))
2 2
Figure 2: Approximate solution of FWBK equation.
𝑐𝑥
× tanh [ ],
2

𝑐4 𝑡𝜇 sech(𝑐𝑥/2)3 tanh [𝑐𝑥/2]


V1 (𝑥, 𝑡) = −
(𝑥, 20) 16Γ (𝜇 + 1)

𝑐𝑥
𝑥 × (71 − cosh [𝑐𝑥] + 6𝑐 tanh [ ]) ,
−150 −100 −50 50 100 150 2

−0.0005 𝑢2 (𝑥, 𝑡)

4−10−𝜂 𝑐5 𝑡𝜂 (𝑐𝑥/2)15
−0.001
=(
Γ (1 + 𝜇) Γ (1 + 𝜂) Γ (0.5 + 𝜂) Γ (1 + 𝜇 + 𝜂)

𝑐𝑥 4
−0.0015 × (−32𝑐3 √𝜋𝑡𝜂 𝜇 cosh ( ) Γ (𝜇)
2

× Γ (1 + 𝜂 + 𝜇) Γ (1 + 2𝜂 + 𝜇)
Figure 3: Approximate solution of FWBK equation.
𝑐𝑥
× (221184 − 20532𝑐2 ) cosh ( )
2
3𝑐𝑥
+ 6 (−11008 + 4813𝑐2 ) cosh ( )
2
5𝑐𝑥 5𝑐𝑥
− 69120 cosh ( ) − 8622𝑐2 cosh ( )
2 2
7𝑐𝑥 7𝑐𝑥
0.002 + 10368 cosh ( ) + 267𝑐2 cosh ( )
2 2
0.001 2
200
(𝑥, 𝑡)

9𝑐𝑥 9𝑐𝑥
0 150 − 128 cosh ( ) + 9𝑐2 cosh ( )
−0.001 2 2
100 𝑐𝑥 𝑐𝑥
𝑡
−100 + 61032𝑐 sinh ( ) − 2772𝑐3 sinh ( )
2 2
0 50
𝑥 3𝑐𝑥 3𝑐𝑥
100 + 29040𝑐 sinh ( ) + 828𝑐3 sinh ( )
2 2
5𝑐𝑥 5𝑐𝑥
Figure 4: Approximate solution of FWBK equation. − 27312𝑐 sinh ( ) + 108𝑐3 sinh ( )
2 2
6 Abstract and Applied Analysis

7𝑐𝑥 − 52680𝑐 sinh (𝑐𝑥) − 391458𝑐3 sinh (𝑐𝑥)


+ 4596𝑐 sinh ( )
2
− 240𝑐 sinh (2𝑐𝑥) + 196824𝑐3 sinh (2𝑐𝑥)
3 7𝑐𝑥 9𝑐𝑥
−36𝑐 sinh ( ) − 84𝑐 sinh ( ))
2 2 + 17580𝑐 sinh (3𝑐𝑥) − 24207𝑐3 sinh (3𝑐𝑥)
+ 3 × 4𝜂 Γ (0.5 + 𝜂)
+ 120𝑐 sinh (4𝑐𝑥) − 156𝑐3 sinh (4𝑐𝑥)
9𝑐𝑥 9
× (65536𝜇 cosh ( ) Γ (𝜇) Γ (1 + 𝜂 + 𝜇) Γ
2 − 12𝑐 sinh (5𝑐𝑥) + 3𝑐3 sinh (4𝑐𝑥) )) ,

𝑐𝑥 2
× (1 + 2𝜂 + 𝜇) sinh ( ) V2 (𝑥, 𝑡)
2
𝑐𝑥 13
× (−2𝑐 + sinh (𝑐𝑥)) = (2−17−2𝜂 𝑐4 𝑡𝜇 sech ( )
2
+ 1024𝑐3 𝑡𝜂 𝜂Γ (𝜂) Γ (1 + 𝜇) Γ 2
× (Γ(1 + 𝜇) Γ (1 + 𝜂) Γ (0.5 + 𝜇) Γ
𝑐𝑥 2 −1
× (1 + 2𝜂 + 𝜇) sinh ( ) × (1 + 𝜇 + 𝜂) Γ (1 + 3𝜇))
2
𝑐𝑥 3𝑐𝑥 𝑐𝑥 4
× (−15745 cosh ( ) + 12951 cosh ( ) × (3 × 44+𝜇 𝑐4 𝑡𝜂 cosh ( )
2 2 2

3𝑐𝑥 7𝑐𝑥 2
− 1175 cosh ( ) + cosh ( ) × Γ(1 + 𝜇) Γ (1 + 𝜂) Γ (0.5 + 𝜇)
2 2
𝑐𝑥
𝑐𝑥 × Γ (1 + 3𝜇) sinh ( )
− 6240𝑐 sinh ( ) 2
2
𝑐𝑥 3𝑐𝑥
3𝑐𝑥 × (896 cosh ( ) − 608 cosh ( )
+ 1728𝑐 sinh ( ) 2 2
2
5𝑐𝑥 𝑐𝑥
5𝑐𝑥 + 32 cosh ( ) + 78𝑐 sinh ( )
−96𝑐 sinh ( )) 2 2
2
3𝑐𝑥 5𝑐𝑥
2 𝑐𝑥 +3𝑐 sinh ( ) − 5𝑐 sinh ( ))
6 𝜂+𝜇
+ 2𝑐 𝑡 Γ(1 + 𝜂 + 𝜇) sinh ( ) 2 2
2
+ Γ (𝜂 + 1) Γ (1 + 𝜂 + 𝜇)
× (−235648 − 1154128𝑐2 + 15804𝑐4
× (15 × 4𝜇 Γ (0.5 + 𝜇)
2 4
− 16 (5584 − 7358𝑐 + 1125𝑐 )
𝑐𝑥 9
× (− 65536𝜇 cosh ( ) Γ (𝜇)
× cosh (𝑐𝑥) 2

+ 16 (15904 − 60016𝑐2 + 99𝑐4 ) 3𝑐𝑥


× Γ (1 + 3𝜇) sinh ( ))
2
× cosh (2𝑐𝑥)
+ 2𝑐7 𝑡2𝜇 Γ (1 + 2𝜇)
+ 89216 cosh (3𝑐𝑥) − 296896𝑐2
𝑐𝑥 3𝑐𝑥
× (994 cosh ( ) − 435 cosh ( )
× cosh (3𝑐𝑥) 2 2
5𝑐𝑥 𝑐𝑥
+ 720𝑐4 cosh (3𝑐𝑥) − 18816 cosh (4𝑐𝑥) + cosh ( ) + 204𝑐 sinh ( )
2 2
+ 14672𝑐2 cosh (4𝑐𝑥) − 108𝑐4 3𝑐𝑥 3𝑐𝑥 2
− 36𝑐 sinh ( ) −5𝑐 sinh ( )) )
× cosh (4𝑐𝑥) 2 2
𝑐𝑥 4
+128 cosh (5𝑐𝑥) − 32𝑐2 cosh (5𝑐𝑥)) + 16𝑐3 √𝜋𝑡𝜇 𝜇 cosh ( )
2
Abstract and Applied Analysis 7

0.0012
0.1 0.0011

𝑢(𝑥, 𝑡)
20
0.075 20 0.00088
(𝑥, 𝑡)

0.05 0.00066 15
0.0004
04
0.025 15 10 𝑡
0 −100
10 𝑡 0 5
−100
𝑥
5 100
0 0
𝑥
100
0 Figure 6: Approximate solution of FJM equation.
Figure 5: Approximate solution of FJM equation.

The STHPM is chosen to solve this nonlinear problem


because of the following advantages that the method has
× Γ (𝜇) Γ (1 + 3𝜇) over the existing methods. This method does not require
the linearization or assumptions of weak nonlinearity. The
× (−79965 − 22032𝑐2 solutions are not generated in the form of general solution
as in the Adomian decomposition method (ADM) [33, 34].
+ 8 (−2633 + 3240𝑐2 ) cosh (𝑐𝑥) No correction functional or Lagrange multiplier is required in
the case of the variational iteration method [35, 36]. It is more
+ (54940 − 3888𝑐2 ) cosh (2𝑐𝑥) realistic compared to the method of simplifying the physical
problems. If the exact solution of the partial differential
− 3960 cosh (3𝑐𝑥) + cosh (4𝑐𝑥) equation exists, the approximated solution via the method
+ 130152𝑐 sinh (𝑐𝑥) − 39504𝑐 sinh (2𝑐𝑥) converges to the exact solution. STHPM provides us with a
convenient way to control the convergence of approximation
series without adapting ℎ, as in the case of [37] which is a
− 216𝑐 sinh (3𝑐𝑥) ) )) . fundamental qualitative difference in the analysis between
STHPM and other methods. And also there is nothing like
(26)
solving a partial differential equation after comparing the
And so on in the same manner one can obtain the rest of the terms of same power of 𝑝 like in the case of homotopy
components. However, here, few terms were computed, and perturbation method (HPM) [38].
the asymptotic solution of the nonlinear fractional Jaulent-
Miodek is given by References
𝑢 (𝑥, 𝑡) = 𝑢0 (𝑥, 𝑡) + 𝑢1 (𝑥, 𝑡) + 𝑢2 (𝑥, 𝑡) + 𝑢3 (𝑥, 𝑡) + ⋅ ⋅ ⋅ , [1] F. Mainardi, “Fractional calculus: some basic problems in
continuum and statistical mechanics,” in Fractals and Fractional
V (𝑥, 𝑡) = V0 (𝑥, 𝑡) + V1 (𝑥, 𝑡) + V2 (𝑥, 𝑡) + V3 (𝑥, 𝑡) + ⋅ ⋅ ⋅ . Calculus in Continuum Mechanics, vol. 378 of CISM Courses and
(27) Lectures, pp. 291–348, Springer, Vienna, Austria, 1997.
[2] A. Atangana and E. Alabaraoye, “Solving system of fractional
Figures 5 and 6 show the graphical representation of the partial differential equations arisen in the model of HIV
approximated solution of the system of nonlinear fractional infection of CD4+ cells and attractor one-dimensional Keller-
Jaulent-Miodek equation for 𝜂 = 0.98, 𝜇 = 0.48, and 𝑐 = 0.1. Segel equation,” Advances in Difference Equations, vol. 2013,
article 94, 2013.
Figures 5 and 6 show the approximate solution of the main
problem. [3] A. Atangana and A. Kılıçman, “Analytical solutions of the space-
time fractional derivative of advection dispersion equation,”
Mathematical Problems in Engineering, vol. 2013, Article ID
5. Conclusion 853127, 9 pages, 2013.
[4] A. Atangana and J. F. Botha, “Analytical solution of the ground
We derived approximated solutions of nonlinear fractional water flow equation obtained via homotopy decomposition
Jaulent-Miodek and Whitham-Broer-Kaup equations using method,” Journal of Earth Science & Climatic Change, vol. 3, p.
the relatively new analytical technique the STHPM. We 115, 2012.
presented the brief history and some properties of fractional [5] N. I. Mahmudov, “Approximate controllability of fractional
derivative concept. It is demonstrated that STHPM is a sobolev-type evolution equations in banach spaces,” Abstract
powerful and efficient tool for the system of FPDEs. In and Applied Analysis, vol. 2013, Article ID 502839, 9 pages, 2013.
addition, the calculations involved in STHPM are very simple [6] A. Mohebbi and Z. Asgari, “Efficient numerical algorithms
and straightforward. for the solution of “good” Boussinesq equation in water wave
8 Abstract and Applied Analysis

propagation,” Computer Physics Communications, vol. 182, no. [25] G. K. Watugala, “Sumudu transform: a new integral trans-
12, pp. 2464–2470, 2011. form to solve differential equations and control engineering
[7] F. Xie, Z. Yan, and H. Zhang, “Explicit and exact traveling wave problems,” International Journal of Mathematical Education in
solutions of Whitham-Broer-Kaup shallow water equations,” Science and Technology, vol. 24, no. 1, pp. 35–43, 1993.
Physics Letters A, vol. 285, no. 1-2, pp. 76–80, 2001. [26] S. Weerakoon, “Application of Sumudu transform to partial
[8] B. A. Kupershmidt, “Mathematics of dispersive water waves,” differential equations,” International Journal of Mathematical
Communications in Mathematical Physics, vol. 99, no. 1, pp. 51– Education in Science and Technology, vol. 25, no. 2, pp. 277–283,
73, 1985. 1994.
[9] W.-X. Ma, C.-X. Li, and J. He, “A second Wronskian formulation [27] J. Singh, D. Kumar, and A. Kılıçman, “Homotopy perturbation
of the Boussinesq equation,” Nonlinear Analysis. Theory, Meth- method for fractional gas dynamics equation using Sumudu
ods & Applications, vol. 70, no. 12, pp. 4245–4258, 2009. transform,” Abstract and Applied Analysis, vol. 2013, Article ID
934060, 8 pages, 2013.
[10] T. Hong, Y. Z. Wang, and Y. S. Huo, “Bogoliubov quasiparticles
carried by dark solitonic excitations in non-uniform Bose- [28] H. Eltayeb and A. Kılıçman, “A note on the Sumudu transforms
Einstein condensates,” Chinese Physics Letters, vol. 15, pp. 550– and differential equations,” Applied Mathematical Sciences, vol.
552, 1998. 4, no. 21-24, pp. 1089–1098, 2010.
[11] J.-f. Zhang, “Multiple soliton solutions of the dispersive long- [29] A. Kılıçman, H. Eltayeb, and P. R. Agarwal, “On Sumudu
wave equations,” Chinese Physics Letters, vol. 16, no. 1, pp. 4–5, transform and system of differ ential equations,” Abstract and
1999. Applied Analysis, vol. 2010, Article ID 598702, 11 pages, 2010.
[12] G. C. Das, J. Sarma, and C. Uberoi, “Explosion of a soliton in a [30] V. G. Gupta and B. Sharma, “Application of Sumudu transform
multicomponent plasma,” Physics of Plasmas, vol. 4, no. 6, pp. in reaction-diffusion systems and nonlinear waves,” Applied
2095–2100, 1997. Mathematical Sciences, vol. 4, no. 9–12, pp. 435–446, 2010.
[31] H. Eltayeb and A. Kılıçman, “A note on the Sumudu transforms
[13] S. Y. Lou, “A direct perturbation method: nonlinear Schrodinger
and differential equations,” Applied Mathematical Sciences, vol.
equation with loss,” Chinese Physics Letters, vol. 16, pp. 659–661,
4, no. 21–24, pp. 1089–1098, 2010.
1999.
[32] F. B. M. Belgacem, A. A. Karaballi, and S. L. Kalla, “Analyt-
[14] A. Atangana and A. H. Cloot, “Stability and convergence of the
ical investigations of the Sumudu transform and applications
space fractional variable-order Schrödinger equation,” Advan-
to integral production equations,” Mathematical Problems in
ces in Difference Equations, vol. 2013, article 80, 2013.
Engineering, no. 3-4, pp. 103–118, 2003.
[15] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional
[33] A. Abdon, “New cass of boundary value problems,” Informa-
Calculus Models and Numerical Methods, vol. 3 of Series on
tions Sciences Letters, vol. 1, no. 2, pp. 67–76, 2012.
Complexity, Nonlinearity and Chaos, World Scientific, Hacken-
sack, NJ, USA, 2012. [34] G.-C. Wu, “Adomian decomposition method for non-smooth
initial value problems,” Mathematical and Computer Modelling,
[16] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory and
vol. 54, no. 9-10, pp. 2104–2108, 2011.
Applications of Fractional Differential Equations, vol. 204 of
North-Holland Mathematics Studies, Elsevier Science B. V., [35] G. C. Wu and D. Baleanu, “Variational iteration method for
Amsterdam, The Netherlands, 2006. the Burgers’ flow with fractional derivatives-New Lagrange
multipliers,” Applied Mathematical Modelling, vol. 5, pp. 1012–
[17] K. B. Oldham and J. Spanier, The Fractional Calculus, Mathe-
1018, 2012.
matics in Science and Engineering, vol. 111, Academic Press, New
York, NY, USA, 1974. [36] G.-C. Wu and D. Baleanu, “Variational iteration method for
fractional calculus—a universal approach by Laplace trans-
[18] B. Ross, Ed., Fractional Calculus and Its Applications, vol. 457 of form,” Advances in Difference Equations, vol. 2013, article 18,
Lecture Notes in Mathematics, Springer, Berlin, Germany, 1975. 2013.
[19] Y. Luchko and R. Gorenflo, “The initial value problem for some [37] H. Xu, S.-J. Liao, and X.-C. You, “Analysis of nonlinear frac-
fractional differential equations with the Caputo derivative,” tional partial differential equations with the homotopy analysis
Preprint Series A08-98, Fachbreich Mathematik and Infor- method,” Communications in Nonlinear Science and Numerical
matik, Freic Universitat Berlin, 1998. Simulation, vol. 14, no. 4, pp. 1152–1156, 2009.
[20] A. Atangana and A. Secer, “A note on fractional order deriva- [38] J.-H. He, “Homotopy perturbation method with an auxiliary
tives and table of fractional derivatives of some special func- term,” Abstract and Applied Analysis, vol. 2012, Article ID
tions,” Abstract and Applied Analysis, vol. 2013, Article ID 857612, 7 pages, 2012.
279681, 8 pages, 2013.
[21] I. Podlubny, Fractional Differential Equations, vol. 198 of Math-
ematics in Science and Engineering, Academic Press, San Diego,
Calif, USA, 1999.
[22] S. G. Samko, A. A. Kilbas, and O. I. Marichev, Fractional
Integrals and Derivatives: Theory and Applications, Gordon and
Breach Science, Yverdon, Switzerland, 1993.
[23] K. S. Miller and B. Ross, An Introduction to the Fractional Calcu-
lus and Fractional Differential Equations, A Wiley-Interscience,
John Wiley & Sons, New York, NY, USA, 1993.
[24] G. Jumarie, “On the representation of fractional Brownian
motion as an integral with respect to (d𝑡)𝑎 ,” Applied Mathemat-
ics Letters, vol. 18, no. 7, pp. 739–748, 2005.

You might also like