EE550: Project Progress Report 1

QM Molecular System Modelling and Associated Control Objectives

Hamood ur Rehman Khan ID#: 200503210

Electrical Engineering Department, KFUPM Submitted 6 October, 2013 (Term 131)

Introduction
This project is exclusively concerned with system forms derived from (diatomic) molecular models governed by Schrodingers wave equation in one dimension.
In[4]:=

I h D@@x, tD, tD h yH0,1L H x, tL

HH0 + UL @x, tD

Out[4]=

HH0 + U L yH x, tL

This is essentially a PDE. For a detailed description/definition of various objects in this equations, please see the attached handwritten notes/explaination. For our purporses, it suffices to mention that y(x, t) is the state (not necessarily the state-variable of choice) of the physcial system to be controlled. y(x, t) corresponds to the wave function of the diatomic molecule as given by its quantum mechanical (QM) model. x is the spatial location of the moleclue, t is time. H0 is the nominal Hamiltonian operator derived from the postulates of classical QM theory. H0 contains the Laplacian Operator D = x x HL, plus another term known as the Morse Potential. U is the laser control input which operats on the state, y(x, t), of the system as well. The nonlinearity in the system arises because of the way in which the operator U acts on y(x, t) (in the absence of laser control, the system is essentially linear).

A. Controllability: The Quest for an Explicit (Constructable) Control Imput U

In contrast to the classical linear control systems issue of deciding whether it is controlable or not (I would guess that for linear systems all definitions of controllability would be equivalent to some sense) the objective here is to come up with an explicit control function U (living in some appropriate function space [e.g., Lp ]) such that the wave packet, y(x, t) can be steered from a given initial state at t = 0, ` y0 (X ) (which has a Gaussian shape with respect to X, a dimensionaless quantity) to another target ` state y1 (X ) (which is also Gaussian) in a finite time T. Note that the QM state of the system, y(x, T), which is a solution to the wave equation might not match the desired Gaussian target wave packet, so ` have to settle for a control input that minimizes some meaure between y1 (X ) and y(x, T). This is a problem in non-linear optimal control. There are various strategies to solve this problem for the system at hand. A thorough description and computation of these optimal solution will be covered in the next progress report. For now we restrict ourselves to approximating the system through what is known as

[EE550] First Progress Report.nb

the bilinear state space model and make pertinent observations regarding their linearizability (in some sense).

B. The Bilinear State Space Model (and its simulation in Mathematica)

Through various considerations (i.e., Galerikin approximation to the QM state function y(x, t)), it is shown in the main reference [1], that the diatomic molecule can be accurately QM modelled by what is known as a homegeneous bilinear system given as:
In[1]:=

x == A x + B x u x Ax+Bux

Out[1]=

Where all matrices and variables have the usual meaning as is customary in linear state space models. u is the (scalar) control input and x is the state variable vector. How x is related to the the QM state of the molecule y(x, t) requires a more elaborate and involved exposition. We will leave that for future communications. For now, we concentrate on how to simulate the above system for special forms for matrices A and B, and an optimal control input obtained in [1]. In [1] the distributed parameter system given by the wave equation is converted into a lumped parameter system in which the state vector is finite dimensional. For simulation purposes we consider 2dimensional state space model in which the state variables evolve in some subset of C2 (which, according to [1] is the 3-dimensional sphere in C4 [the dimensions need checking]). The dimension of the state space depends on the the application at hand and is a matter of accurate yet efficient system modelling of QM systems.
In[207]:=

H A is diagonal matrix over C, compoesed of the Eigenvalues of the Hamilitonian of the system, Each eigenvalues corresponds to energy state of the molecule, measured either through spectrosopic measurements or numerically through the molecules potential function L A = 88I, 0<, 80, .2 I<<

Out[207]=

0 0 0. + 0.2

In[208]:=

H The matrix B is I times the "interaction matrix of the Hamiltonian" and basically related to the potential function component. For the sake of simplicity we assume a constant flat potential, Hthough this is not accurate for the diamotic molecule which follows the space dependent Morse Potential, we will extend this to the general case L B = I 883, 2<, 81, 5<< 3 2 5

Out[208]=

[EE550] First Progress Report.nb

In[211]:=

H what remains is to define the time dependent input function u@tD and the initial conditions, which we take to be 881<,82<< L u@t_ D := .3; H this is a piecewise constant inputL H in the paper @1D there are mor complicated piecewise constant input designs L H The state vecotr consistis of two time functions x1@tD and x2@tD L H Here is an attempt to compute the state trajectory from initial rest to time 1 unt L x@t_ D := Transpose@88x1@tD, x2@tD<<D; H the state vector is a column vector L H the state space model as described in Mathematica L statespace = D@x@tD, tD A.x@tD + B.x@tD u@tD

Out[211]=

x1 HtL x2 HtL

x1HtL + 0.3 H3 x1HtL + 2 x2HtLL H0. + 0.2 L x2HtL + 0.3 H x1HtL + 5 x2HtLL

In[212]:=

LHS = Flatten@statespace@@1DDD H Left hand side L 8x1 HtL, x2 HtL< RHS = Flatten@statespace@@2DDD H Right hand side L 80.3 H3 x1HtL + 2 x2HtLL + x1HtL, H0. + 0.2 L x2HtL + 0.3 H x1HtL + 5 x2HtLL< equations = Thread@LHS 8x1 HtL RHSD H system of ODE to be solve L H0. + 0.2 L x2HtL + 0.3 H x1HtL + 5 x2HtLL<

Out[212]=

In[213]:=

Out[213]=

In[214]:=

Out[214]=

0.3 H3 x1HtL + 2 x2HtLL + x1HtL, x2 HtL

All that remains is to solve this system of equation, in this case, becuase of the simplicity of the system an exact solution might be obtainable, lets try this.
In[215]:=

solution = DSolve@equations Join 8x1@0D 99x1 I8t< I-0.991202 - 2.19759 10-16 M IH0. + 1. L 2.7755610 x2 I8t<
-17

1, x2@0D

2<, 8x1, x2<, tD

Out[215]=

sinH1.36411 tL + H1. + 0. L 2.7755610

-16
-17

-17

cosH1.36411 tL -

I2.00888 - 2.23678 10 H1. + 0. L

2.7755610
-17

M cosH2.23589 tL - I2.23678 10-16 + 2.00888 M sinH2.23589 tLMM,

t

I0.885292 + 1.2784 10-16 M IH0. + 1. L 2.7755610

t

sinH1.36411 tL +

cosH1.36411 tL + I1.25914 - 3.26229 10-16 M cosH2.23589 tL +

I3.26229 10-16 + 1.25914 M sinH2.23589 tLMM==

Complicated, but exact nonetheless. Lets obtain some plots.

[EE550] First Progress Report.nb

In[216]:=

x1func@t_D := Evaluate@x1@tD . solution@@1, 1DDD; x2func@t_D := Evaluate@x2@tD . solution@@1, 2DDD; x1re = Plot@Re@x1func@tDD, 8t, 0, 1<, PlotStyle Thick, Frame True, Axes False, PlotLabel "Real part of x1", FrameLabel 8"t", "x1HtL"<, GridLines AutomaticD; x1im = Plot@Im@x1func@tDD, 8t, 0, 1<, PlotStyle Thick, Frame True, Axes False, PlotLabel "Imaginary part of x1", FrameLabel 8"t", "x1HtL"<, GridLines AutomaticD; x2re = Plot@Re@x2func@tDD, 8t, 0, 1<, PlotStyle Thick, Frame True, Axes False, PlotLabel "Real part of x2", FrameLabel 8"t", "x1HtL"<, GridLines AutomaticD; x2im = Plot@Im@x2func@tDD, 8t, 0, 1<, PlotStyle Thick, Frame True, Axes False, PlotLabel "Imaginary part of x2", FrameLabel 8"t", "x1HtL"<, GridLines AutomaticD;

In[218]:=

Finally:
In[226]:=

GraphicsGrid@88x1re, x1im<, 8x2re, x2im<<, Spacings > 8Scaled@0D, Scaled@0D<D

Real part of x1
1.0 0.5 x1HtL x1HtL 0.0 -0.5 -1.0 0.0 0.2 0.4 t 0.6 0.8 1.0 1.2 1.0 0.8 0.6 0.4 0.2 0.0 0.0

Imaginary part of x1

0.2

0.4 t

0.6

0.8

1.0

Out[226]=

Real part of x2
2.0 1.5 x1HtL x1HtL 1.0 0.5 0.0 -0.5 0.0 0.2 0.4 t 0.6 0.8 1.0 1.5 1.0 0.5 0.0 0.0

Imaginary part of x2

0.2

0.4 t

0.6

0.8

1.0

What we have is the state space solution to the nonlinear bilinear system used to model a 2 level QM model of diatomic molecule.

C. Linearization of Bilinear Systems

Linearization of highly nonlinear systems like the one describe in subsection D is by no means an obvious task. Local linearization based on Taylor series expansions and Jacobian computation usually lead to unsatisfactory solutions. Global strategies for linearizing nonlinear systems usually require tools from manifold theory, differential topological considerations and concepts from differential geometry etc. Generally speaking nonlinear systems resist a purely (linear- or otherwise) algebraic attack with regards to linearization. Fortunately for bilinear systems, the concepts of global and local linearization coincide, as first investigated in [2] which also gives sufficient conditions for to obtain globally linearized models. [2] also shows

[EE550] First Progress Report.nb

that for homogeneous bilinear systems, an alebgraic solution is possible using Lie algebra and the properties of Lie bracket. I will not attempt to carry out that anlaysis because of the limited time resources. My main objective in writing this report was to present a unified and coherent task/issue list to be tackled by this project. A full global linearization anlaysis will be carried out in a time frame of 2-3 weeks from now. For the same reason a simulation of the linearized system is omitted from this project report.

D. Stability Analysis
There is siginificant literature on how to effectively stabilize bilinear systems including feedback strategies. I have been able to find the most relevant references, however a full stability analysis of bilinear systems is not within the (defined) scope of this proejct. It is my intention to carry out stability anlaysis of QM system if time permites. Most of the questions regarding stability of bilinear systems have been resolved to the best of my knowledge. It is a matter of time and effort to carry out stability analysis in the context of a bilinear state space model for QM moldecular model.

E. Functions/Quantities of Interest
This section relates to the design of the optimal control input U. This will become more clear as material is added to this report to generate the second progress report. For the time being, note that the Morse Potential is used to model the potential function for the net actractive force field between the atoms bonded in a diatomic molecule.

[EE550] First Progress Report.nb

In[37]:=

D0 = 4; H arbitrary constant, physical signifcance will be made clear later on L = .2; H another arbitrary constant L x0 = 0; H origin of space, where the molecule nominal remains L H Morse Potential Function for diatomic used in constructed the nominal Hamiltonian operator H0HL = HMorseP@xD + LHL L MorseP@x_ D := D0 H1 Exp@ Hx x0LDL ^ 2; H Lets see how the morse potential function looks like L Plot@MorseP@xD, 8x, 2, 2<, PlotStyle Thick, PlotLabel "The Morse fucntion for =.2", Frame True, Axes False, FrameLabel 8"position x", "Morse Potential"< D
The Morse fucntion for g=.2

0.8

Morse Potential

0.6

Out[39]=

0.4

0.2

0.0 -2 -1 0 position x 1 2

The potential function seems to be asymmetric around x0 = 0. Lets try to find the minimum of MoresP[x]
In[33]:=

Clear@, D0, x0D; H first derivative of MorseP L D@MorseP@xD, xD

Out[34]=

2 D0

Hxx0L

I1

Hxx0L

Setting the above derivative to zero:

In[35]:=

Solve@D@MorseP@xD, xD

0, xD x0 + 2 C@1D , C@1D IntegersF>>

Out[35]=

::x ConditionalExpressionB

Setting the constant of integration to 0.

[EE550] First Progress Report.nb

In[36]:=

% . C@1D 0 88x x0<<

Out[36]=

We see that the Morse Potential has a minimum at x0 regardless of g, D0.

Rerferences Cited in this Report (Not the Complete Project)

[1] Dahleh, M.; Peirce, A.; Rabitz, H.A.; Ramakrishna, V., Control of molecular motion, Proceedings of the IEEE, vol.84, no.1, pp.7, 15, Jan 1996. [2] S. elikovsk, On the global linearization of bilinear systems, Systems & Control Letters, Volume 15, Issue 5, December 1990, Pages 433-439