Variations, Geometry and Physics in Honour of Demeter Krupka's Sixty-Fifth Birthday

VARIATIONS , G EOMETRY AND P HYSICS
No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or
by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no
expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of information
contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in
rendering legal, medical or any other professional services.
VARIATIONS , G EOMETRY AND P HYSICS
IN HONOUR OF D EMETER K RUPKA S SIXTY- FIFTH BIRTHDAY
O LGA K RUPKOV A
AND
DAVID S AUNDERS
Nova Science Publishers, Inc.

New York

c 2009 by Nova Science Publishers, Inc.
All rights reserved. No part of this book may be reproduced, stored in a retrieval system or
transmitted in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical
photocopying, recording or otherwise without the written permission of the Publisher.
For permission to use material from this book please contact us:
Telephone 631-231-7269; Fax 631-231-8175
Web Site: http://www.novapublishers.com
NOTICE TO THE READER
The Publisher has taken reasonable care in the preparation of this book, but makes no expressed or
implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of
information contained in this book. The Publisher shall not be liable for any special, consequential,
or exemplary damages resulting, in whole or in part, from the readers use of, or reliance upon, this
material.
Independent verification should be sought for any data, advice or recommendations contained in
this book. In addition, no responsibility is assumed by the publisher for any injury and/or damage
to persons or property arising from any methods, products, instructions, ideas or otherwise
contained in this publication.
This publication is designed to provide accurate and authoritative information with regard to the
subject matter cover herein. It is sold with the clear understanding that the Publisher is not engaged
in rendering legal or any other professional services. If legal, medical or any other expert assistance
is required, the services of a competent person should be sought. FROM A DECLARATION OF
PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE AMERICAN BAR
ASSOCIATION AND A COMMITTEE OF PUBLISHERS.
Library of Congress Cataloging-in-Publication Data

Variations, geometry & physics / Olga Krupkova and David Saunders (editor).
p. cm.
ISBN 978-1-61324-186-8 (eBook)
1. Calculus of variations. 2. Global analysis (Mathematics) 3. Mathematical physics. I. Krupkov,
Olga, 1960- II. Saunders, D. J., 1964- III. Title: Variations, geometry, and physics.
QA315.V38 2008
510dc22
2008024454
Published by Nova Science Publishers, Inc. New York

Contents
Preface vii
PART I VARIATIONAL P RINCIPLES ON J ET B UNDLES 1
Chapter 1 Lepage Forms in Variational Theories: From Lepages Idea

to the Variational Sequence 3
Jana Musilova and Michal Lenc
Chapter 2 Lepage Forms in the Calculus of Variations 27

Olga Krupkova
Chapter 3 On a Generalization of the Poincare-Cartan Form

in Higher-Order Field Theory 57
D. R. Grigore
Chapter 4 Krupkas Fundamental Lepage Equivalent and

the Excess Function of Wilkins 77
D. J. Saunders
Chapter 5 Lepage Congruences in Discrete Mechanics 85

Antonio Fernandez and Pedro L. Garca
Chapter 6 Finite Order Variational Sequences: a Short Review 99

Raffaele Vitolo
Chapter 7 Concatenating Variational Principles and

the Kinetic Stress-Energy-Momentum Tensor 117
M. C. Lopez, M. J. Gotay and J. E. Marsden
Chapter 8 A Geometric Hamilton-Jacobi Theory for Classical

Field Theories 129
M. de Leon, J. C. Marrero and D. M. de Diego
vi Contents
PART II N ATURAL B UNDLES AND D IFFERENTIAL I NVARIANTS 141
Chapter 9 Natural Lagrangian Structures 143

Josef Janyska
Chapter 10 Connections on Higher Order Frame Bundles

and Their Gauge Analogies 167
Ivan Kolar
Chapter 11 Natural Lifts in Riemannian Geometry 189

Oldrich Kowalski and Masami Sekizawa
Chapter 12 Invariant Variational Problems and Invariant

Flows via Moving Frames 209
Peter J. Olver
Chapter 13 Differential Invariants of the Motion Group Actions 237

Boris Kruglikov and Valentin Lychagin
PART III D IFFERENTIAL E QUATIONS AND G EOMETRICAL

S TRUCTURES 253
Chapter 14 Remarks on the History of the Notion

of Lie Differentiation 255
Andrzej Trautman
Chapter 15 Second-Order Differential Equation Fields with Symmetry 261

M. Crampin and T. Mestdag
Chapter 16 Dimensional Reduction of Curvature-Dependent

Central Potentials on Spaces of Constant Curvature 277
J. F. Carinena, M. F. Ranada and M. Santander
Chapter 17 Direct Geometrical Method in Finsler Geometry 293

L. Tamassy
Chapter 18 Linear Connections Along the Tangent Bundle Projection 315

W. Sarlet
Chapter 19 On the Inverse Problem for Autoparallels 341

G. E. Prince
PART IV A PPENDIX 353
Demeter Krupka: List of Publications 355
Index 363
Preface
This book is a commemorative volume to celebrate the

sixty-fifth birthday of Demeter Krupka, an internation-
ally recognized researcher in several fields of mathemat-
ics and mathematical physics.
Demeter was born on 22 January 1942 in Levoca,
Slovakia. He studied first at Masaryk University, Brno
and in 1976 he received a Ph.D. degree from Charles
University, Prague; during his studies he also spent a
year at Warsaw University under the supervision of Pro-
fessor Andrzej Trautman. In 1981 he was awarded the
degree of Doctor of Sciences in geometry and topology
by the Czechoslovak Academy of Sciences, Prague.
Demeter has held posts at Masaryk University and at the Silesian University in Opava,
and is currently Professor in the Department of Algebra and Geometry at Palacky Univer-
sity, Olomouc. He also holds an honorary position as an IAS Distinguished Fellow at the
Institute for Advanced Study, La Trobe University, Melbourne. His main research areas are
global variational analysis, differential geometry, tensor algebra and mathematical physics.
Among his recognized achievments we should mention the development of a global theory
of variational functionals in fibred spaces in the early 1970s, and its higher-order generali-
sations in the early 1980s, based on his concept of a Lepage form and the EulerLagrange
form. This approach provides a completely intrinsic setting for EulerLagrange theory
and the invariance of variational functionals, including Noether theorem and its generalisa-
tions. It also facilitates the study of regular variational functionals, casting new light on the
concept of regularity in the calculus of variations, Hamilton theory and HamiltonJacobi
theory. He also was among the first to find necessary and sufficient conditions for a system
of higher-order PDEs in covariant form to be variational, and he solved the local problem
of the structure of higher-order null Lagrangians.
In 1990 Demeter published his key paper on variational sequences, a finite order alter-
native to Vinogradovs C -spectral sequence and the variational bicomplex introduced by,
for example, Dedecker and Tulczyjew. Another important impact of his work concerns
the geometric foundations of theory of differential invariants developed in the 1970s and
1980s, based on the concept of a higher order differential group, and the creation of a rigor-
ous setting for natural variational principles. While solving technical problems appearing in
the higher order variational calculus, he also contributed to linear algebra by studying and
viii Preface
solving the trace decomposition problem in general tensor spaces over the real numbers.
During his career, Demeter has written numerous articles for publication (a list is given
in the Appendix to this volume) and taught a wide variety of topics in mathematics and
mathematical physics; other highlights include being a founding editor of two major jour-
nals (Journal of Geometry and Physics and Differential Geometry and its Applications) and
holding the post of Rector at the Silesian University in Opava. Most recently he has been
joint editor of a major reference work (Handbook of Global Analysis).
To celebrate Demeters sixty-fifth birthday, a Colloquium was held in Olomouc on 25th
and 26th August, 2007 with twenty-one invited speakers, all with established reputations
in their field. Many of the speakers have prepared versions of their talks for the present
volume, and we have also been able to include some articles from authors who were not
able to attend. We have divided the articles into three groups, covering three major areas
of Demeters research interests: Variational principles on jet bundles, Natural bundles and
differential invariants and Differential equations and geometric structures. We are grateful
to all the authors who have provided these articles, and we are sure that they will be of great
interest to the community.
Olga Krupkova, David Saunders (Editors)

Olomouc, January 2008
Part I
VARIATIONAL P RINCIPLES ON J ET
B UNDLES
1
In: Variations, Geometry and Physics ISBN 978-1-60456-920-9
Editors: O. Krupkova and D. Saunders, pp. 3-26
c 2009 Nova Science Publishers, Inc.
Chapter 1
L EPAGE F ORMS IN VARIATIONAL T HEORIES :

F ROM L EPAGE S I DEA TO THE VARIATIONAL
S EQUENCE
Jana Musilova and Michal Lenc
Institute of Theoretical Physics and Astrophysics, Faculty of Science,
Masaryk University, Kotlarska 2, 611 37 Brno, Czech Republic
Abstract
In the presented paper the development of the concept of Lepage forms is followed
from the initial idea given by Lepage in 1930s to its important role in the contempo-
rary geometrical analysis and variational theories in physics. The attention is paid
especially to the role of the concept of Lepage forms for the main formula of calculus
of variations (called the first variational formula) as well as to their meaning as geo-
metrical objects resulting by a natural way from the theory of variational sequences
on finite order prolongations of fibered manifolds. As a possible application Lepage
forms are studied in the context of our results concerning the representation of the
variational sequence by differential forms on the one hand, and in concrete physical
examples on the other. For the illustration we show that the concept of Lepage forms
can be effectively used in such physical situations in variational field theories, as are
e.g. variational problems in the classical string and brane theories.
1. Introduction
The contemporary general concept of Lepage forms is closely related to fundamentals of
local and global aspects of the modern calculus of variations. It represents differential
forms of a special type defined on finite order jet prolongations of fibered manifolds. These
manifolds are chosen as appropriate underlying geometrical structures for the majority of

This paper is dedicated to 65th birthday of our colleague and friend professor Demeter Krupka, whose
contribution to the development of the concept of Lepage forms as the key geometrical objects of modern
calculus of variations and its application in physical theories is undoubtedly important.

E-mail address: janam@physics.muni.cz

E-mail address: lenc@physics.muni.cz
4 Jana Musilova and Michal Lenc
physical theories. Theories developed on fibered manifolds with one-dimensional bases

are called mechanics, while for n-dimensional bases of underlying fibered manifolds
we speak about field theories. There are specific geometrical (i.e. coordinate free) ob-
jects defined on a fibered manifold and its jet prolongations, closely connected with the
fibered structure of the manifold: projectable and vertical vector fields, horizontal and con-
tact differential forms. These objects play the key role in formulation of the fundamental
variational problem which lies in searching such sections of the underlying fibered mani-
fold, e.g. trajectories of studied physical systems, representing critical (stationary) points
of a functional appropriately chosen with respect to a given theory. Such a functional is,
for a theory formulated on a fibered manifold (Y, , X) with the n-dimensional base X,
the (m + n)-dimensional total space Y , and the projection , defined by the integral of a
horizontal n-form = L 0 , called r-th order Lagrangian, where L is a function on the
r-th prolongation (J r Y, r , X) of the fibered manifold and 0 is the volume element on the
base, in local coordinates 0 = dx1 . . . dxn . The general questions most intensively
studied in last decades are especially:
the variational equivalence problem leading to the concept of Lepage equivalents of

a given Lagrangian, i.e. differential forms giving the same variational integral as the
corresponding Lagrangian, and in general to the concept of Lepage n-forms,
the structure of the Euler-Lagrange mapping assigning to a Lagrangian its Euler-
Lagrange form representing the equations of motion; the study of this structure en-
ables us to find trivial Lagrangians, leading to identically zero Euler-Lagrange forms,
the structure of Helmholtz-Sonin mapping; this enables us to solve the inverse prob-
lem of calculus of variations, i.e. stating whether given equations of motion arise
from a variational problem, and leads to the concept of Lepage (n + 1)-forms and
their equivalence with respect to given equations of motion,
the theory of the finite-order variational sequences and the problem of their rep-
resentation by differential forms, both inspired by the above exposed equivalence
problems, namely by the close relation between the Euler-Lagrange mapping and the
exterior derivative operator,
a cohomological structure of the variational sequence in which both Euler-Lagrange
and Helmholtz-Sonin mappings are included, as an effective tool for describing glo-
bal aspects the theory,
an idea of the quite general definition of Lepage forms arising naturally from the
structure of the variational sequence itself.
Both primary ideas leading to the concept of Lepage forms (the importance of the exte-
rior derivative operator and the idea of equivalence of forms in variational problems) were
formulated by Lepage in [35] and [36], and presented by Elie Cartan and Th. de Donder,
respectively. They were later developed by Paul Dedecker in [6]. In Sec. 2 we present the
brief review of considerations concerning these historical ideas and their consequences.
The concept of a Lepage equivalent of a Lagrangian = L 0 as a form leading to the
same variational problem, strictly based on the geometrical structure of fibered manifolds
Lepage Forms in Variational Theories 5
and their jet prolongations, was introduced by Krupka (see e.g. his key paper [16]), and then
developed up to a coherent theory. His studies of the Euler-Lagrange mapping in 1970s
and 1980s inspired the later considerations concerning the variational sequence. (See e.g.
in [17], [18], [19], [20], [21], [22], [24], a complete review with contemporary results see
in [26].) The concept of a Lepage (n + 1)-form and Lepage class of a dynamical (n + 1)-
form E = E 0 , important from the point of view of equations of motion E J s
(variational or non-variational) for trajectories : X Y , U X, = IdU of a
dynamical system, was introduced by Krupkova for mechanics, i.e. n = 1 (see [30], [31],
[32]). It was well elaborated to a coherent theory by the same author (see the key book
[33]). The concept was later extended to non-holonomic constrained mechanical systems
in [34]. Main ideas and results obtained in studies of Lepage forms made by Krupka and
Krupkas coworkers are summarized in Sec. 3.
The above mentioned special type of a Lepage form, Lepage equivalent of a Lagrangian,
is important namely for so called first variation formula. This formula represents an ap-
propriate decomposition of the basic object of the calculus of variations the variational
functional (the integral first variation formula), or the decomposition of the integrand it-
self with respect to the fibered structure of the underlying manifold (the infinitesimal first
variation formula). Moreover, the first variational formula gives information concerning
some properties of symmetry transformations of variational problems. Thus, the applica-
tion of Lepage equivalents of Lagrangians for formulation of variational physical theories
is straightforward. In Sec. 4 we give some examples of such an application in modern
variational field theories in physics string and brane theories. We present there the Lep-
age equivalents in three physical situations: a relativistic particle (first order mechanics),
the classical variational functional for a boson string (first order theory, see e.g. [5]), the
Polyakov functional for D-dimensional space and its special case for Minkowski space
(first order field theory). With help of results following from the first variational formula
we use these Lepage equivalents for obtaining quantities important for conservation laws.
The quite general definition of Lepage forms arises by a natural way from the structure
of the variational sequence on finite order jet prolongations of fibered manifolds. Especially,
it is closely related to the problem of representation of the variational sequence by forms.
The idea of the finite order variational sequence was exposed by Krupka in 1989 (see [23]),
who studied the structure of morphisms of the sequence important for calculus of variations,
as well as the cohomology aspects important for global questions. Recall that the infinite-
order counterpart of Krupkas theory is the theory of variational bicomplex, formulated in
1970s and 1980s by Anderson and Duchamp in [3], Dedecker and Tulczyjew in [7], [43],
Takens in [42], and Vinogradov and his school in [44], [45], [46], [4], [11].
Both theories, the variational sequence as well as variational bicomplex, are based on
the analysis of the exterior derivative operator d. (Recall that the r-th order variational
sequence is the quotient of the well-known De Rham sequence by its exact subsequence
of forms of certain kind of contactness or variational irrelevance, the quotient being ex-
act as well.) However, there are some differences resulting from the mentioned different
choice of basic structures as domains of definition, finite order jet prolongations J r Y of
a fibered manifold Y and infinite jets J Y . The main difference lies in the fact that the
finite order theory keeps the order of all sequence operation fixed (contrary to the case of
bicomplex, where the order can be increased arbitrarily). This gives a deeper understand-
ing of the role of the order, of the analytic structure of fundamental variational concepts.
Moreover, it is more appropriate for application in physical situations, always described
as finite order problems. The instructive comparison of both theories is presented in [25]
and [48]. The finite order variational sequence was intensively studied from 1990s to the
present by Krupka and coauthors or coworkers, and by other authors, mainly Grigore [9],
[10] and Vitolo [47], [48], [49], [50]. (The practically complete bibliography related to both
the variational bicomplex and mainly to the variational sequence is presented in [50].)
A specific problem connected with the finite order variational sequence is its represen-
tation by differential forms. The idea lies in the requirement to represent every class [] of
differential forms in the r-th order variational sequence by an appropriately chosen form
with specific properties: R-linearity, coordinate invariance, exactness of the sequence of
representatives, projection property of the operator assigning to a class its representative.
The representation problem was also studied namely by Krupkas school and important
partial results were obtained (except for cited Krupkas works see also e.g. [37], [12], [13],
[14], [28], [38], [39], [29], [41].) A complete solution of the representation problem in
r-th order field theory was inspired by works of Anderson [1], [2]. The complete solu-
tion of the representation problem for finite order variational sequence in field theories is
presented in [15]. It was inspired by the Euler representation operator introduced by Ander-
son for the variational bicomplex: the Andersons concept of this operator was adapted to
the finite order case and the required properties of the obtained representation were proved
by taking into account specific features of the finite-order problem. It appeared that the
general definition of Lepage forms arises immediately from the structure of the variational
sequence representation. Note that the general concept of Lepage forms was introduced by
an alternative way in [29] for the case of mechanics.
Some aspects of the representation of the variational sequence in the context with the
general concept of Lepage forms are discussed in Sec. 5.
2. Lepages Idea a Brief Historical Review

The famous mathematician Theophile Lepage gained the degree of Doctor of Science de
lUniversite de Liege in 1924. He was probably the student of Elie Cartan. He had 11 doc-
toral students and 168 descendants in the period 19371973. He was the dean of the Faculte
des Sciences de lUniversite Libre de Bruxelles in 19531955. His bibliography amounts
to 19 scientific papers published in 19291955, mostly devoted to geometry and calculus
of variations. As a certain curiosity we can mention that Theophile Lepage had introduced
a symplectic analog of Hodge theory before the Hodge theory itself was presented.
His paper [36] notifying the concept of future Lepage forms was published in 1936.
However, in 1933 some Lepages ideas was published in Comptes Rendus des Seances de
l Academie des Sciences, being presented on the seance 18-th December 1933 by Elie
Cartan (see [35]). In the presentation called Sur certaines formes differentielles exterieures
et la variation des integrales doubles (corresponds to the field theory on a fibered manifold
with the two-dimensional base) a predecessor of a contact 1-form is introduced and some
factorization of 2-forms with respect to contact forms is made. Let us now present some
results from the Lepages key paper [36] important for the concept of Lepage forms and
discuss them from the point of the contemporary theory. In [36] a double variational integral
defined by the formula

z1 zn z1 zn
Z Z
I(zi ) = f x, y; z1 , . . . , zn ; ,..., ; ,..., dx dy (2.1)
x x y y
is studied. There x and y are independent variables, zi (x, y) are declared as unknown
functions. It is evident that this corresponds to a first order variational problem on a fibered
manifold (Y, , X), where dim X = 2, dim Y = n + 2, dim J 1 Y = 2 + n + 2n = 3n + 2.
A 2-form is defined (in the Lepages terminology called the symbolic quadratic form)
= f (x, y; z1 , . . . , zn ; p1 , . . . , pn ; q1 , . . . , qn ) dx dy, (2.2)
where function f is differentiable. (The exterior product is not explicitly written but it
is evidently taken into account.) This form corresponds to a Lagrangian defined on J 1 Y .
Thus, the integral (2.1) is a variational functional of the type
Z
J 1 , : (x, y) (z1 (x, y), . . . , zn (x, y))

and our aim is to find its stationary paths . The forms
i = dzi pi dx qi dy, 1 i n, (2.3)
are introduced as Pfaffian forms, being the predecessor of todays contact forms on J 1 Y .
Moreover, congruences (i.e. equivalence classes) are introduced as follows: Denote
i = Xi dx + Yi dy + Aij j , Aij + Aji = 0, 1 i, j n,
Xi , Yi , Aij being differentiable functions of (x, y, zj , pj , qj ). The relation
= f dx dy + i i , (mod 1 , . . . , n ) (2.4)
defines the equivalence relation of forms giving the same variational integral (2.1). Then
there is requirement to calculate all forms defining congruences
d 0 (mod 1 , . . . , n ). (2.5)
The result is as follows:

f f

= f dx dy + dx dy i + Aij i j . (2.6)
qi pi
It can be easily verified that from the point of view of the contemporary theory the form
given by first two summands of (2.6), i.e.
f f

= f dx dy + dx dy i ,
qi pi
is exactly the principal component of the Lepage equivalent of the Lagrangian (compare
with the relation (3.5) for n = 2, s = 1), while the form Aij i 2 is 2-contact in the
present terminology. (Note that the aim of the paper is to study so called geodesic vector
fields relative to the form . These fields given by functions
v : (x, y, z1 , . . . , zn ) (pi (x, y, z1 , . . . , zn ), qi (x, y, z1 , . . . , zn ))
are defined by the condition d[] = 0, where [] = v and [i ] = i v.)

Finally let us mention an important contribution of Paul Dedecker to the concept of
Lepage forms. It is presented in the paper [6]. The notation used therein is close to that
used at present. The variational integral in mechanics is introduced by
Z
I1 = L(t, q i , q i ) dt, 1 i n, (2.7)
c
where c is an arc in the space W with local coordinates (t, q i , q i ) satisfying the equations
i = 0. In fact, this means that c can be interpreted as a first prolongation J 1 of a section
of a fibered manifold with one-dimensional base. The so called Pfaffian form is defined
by
L
= L dt + i i , i = dq i q i dt. (2.8)
q
We can see that from the point of view of the todays theory the form (2.8) is exactly the
Lepage equivalent of the first order Lagrangian in mechanics.
The semibasic forms are then introduced as
= L dt mod i
and it is stated and proved that the the relation (2.8) gives the unique semibasic form such
that
d 0 mod i .
The procedure is then generalized for the field theory and a version of the first variational
formula is derived.
3. Lepage Forms on Fibered Manifolds

The concept of a Lepage form and the Lepage equivalent of a Lagrangian on fibered man-
ifolds and their jet prolongations was introduced by Krupka in one of his basic works [16]
in the connection of properties of the variational integral as well as the so called first vari-
ational formula. This formula is closely related to the fibered structure of the underlying
manifold. The definition and properties of both a Lepage form and a Lepage equivalent of a
Lagrangian were then studied by Krupka in [22]. In this section we follow the development
and improvement of the concept of Lepage forms culminating in Krupkas latest work on
this topic [26]. First let us briefly expose basic geometrical structures for the variational
integral, variational principle and the first variation formula. We use the standard defini-
tions and standard notations given e.g. in [26]. Moreover, we often present definitions of
geometrical objects by their chart expressions, which is more appropriate for their practical
use.
3.1. Lagrange Structures, Variational Functionals

Let (Y, , X), in shortened notation Y or , be a fibered manifold with the n-dimensional
base X, (m + n)-dimensional total space Y , m-dimensional fibers 1 ({x}), x X and
the projection : Y X (surjective submersion). Note that for n = 1 or n > 1 this
structure represents the geometrical background for mechanics or field theories, respec-
tively. Denote by (J r Y, r , X), r 0, the r-th jet prolongation of (Y, , X), where we
put J 0 Y = Y , 0 = . Canonical projections are denoted by s,r : J s Y J r Y for
s > r 0. Let (V, ) be a fibered chart on Y , V being an open subset of Y , = (xi , y ),
1 i n, 1 m. The pair (U, ), U = (V ), = (xi ), is the associated chart on
X. A smooth mapping : U Y such that = Id U is called a section of defined
on U . Denote by U a set of all such sections. The pair (Vr , r ) where Vr = r1 (V ),
k y
r = (xi , y , yi1 , . . . , yi1 ...ir ), yi1 ...ik = ,
xi1 . . . xik
1 k r, 1 i1 , . . . , ik n, is the associated fibered chart on (J r Y, r , X).

A vector field field on Y is called -projectable, if there exists a vector field 0 on X
such that T = 0 . It is called -vertical, if 0 = 0. A -projectable vector field is
expressed as

= i (xj ) i + (xj , y ) .
x y
For -vertical vector field we have i = 0, 1 i n. Definitions of s -projectable or
s,r -projectable vector fields, and definitions of s -vertical or s,r -vertical vector fields on
J s Y , s > r 0, are quite analogous. A vector field on J s Y is of the form
s

= i
X
i
+
+ i1 ...ik ,
x y k=1
yi1 ...ik
where the components of are functions on J s Y , in general. Let be a -projectable

vector field in Y . Then its s-jet prolongation is the vector field on J s Y given by the chart
expression
s

J s = i (xj ) j
X
i
+ (x , y )
+ i1 ...ik (xj , y , . . . , yj1 ...jk ) ,
x y k=1
yi1 ...ik
di1 ...ik1 d j
i1 ...ik = yi1 ...ik1 j ,
dxik dxik
where
s
df f X X f
dj f = = + yi1 ...ik j
dxj xj k=0 1i1 ...ik n
yi1 ...ik
is the total derivative of a function f on J s Y with respect to xj .

A differential form on J s Y is called r -horizontal if i = 0 for every s -vertical
vector field . A form on J s Y is called s,r -horizontal if i = 0 for every s,r -vertical
vector field . A form on J s Y is called contact if J s = 0 for every section of .

Differential forms
= dy yi dxi , i1 = dyi1 yi1 j dxj ,

(3.1)
i1 ...is1 = dyi1 ...is1 yi1 ...is1 j dxj , dyi1 ...is ,
where 1 m, 1 i1 , . . . , is n, form the basis of 1-forms called the basis adapted

to the contact structure. Note that these forms are contact with the exception of dyi1 ...is .

A form on J s Y is called k-contact if every summand of the chart expression of s+1,s
contains the exterior product of exactly k 1-contact factors of the type (3.1).
Let be a differential q-form on J s Y . Then there exists a unique decomposition
q

X
s+1,s = h + p = pq , p0 = h, (3.2)
k=0
where h is the horizontal form called the horizontal or 0-contact component of and
p is the contact form called the contact component of . The contact component p is
uniquely decomposed into its k-contact components, 1 k q.
Let (V, ) be a fibered chart on Y . We denote by s0 V the ring of functions on Vs ,
and by sq V a s0 V -module of q-forms on Vs . Note that the sets sq,X V of s -horizontal
q-forms and sq,Y V of s,0 -horizontal q-forms on Vs are submodules of sq V .
Now, let us introduce the concept of variational functional. Let W Y be an open set.
1
A horizontal n-form defined on Wr = r,0 (W ) is called r-th order Lagrangian. It holds
= L(xi , y , yi1 , . . . , yi1 ...ir ) 0 , 0 = dx1 . . . dxn .
The pair (, ) is so called Lagrange structure. Let U = (W ). Let U be a compact

n-dimensional submanifold of X with the boundary . Denote by ,W a set of all
sections U such that supp = and () W . The mapping
Z
() : ,W () = J r (3.3)

is called the variational functional or variational integral over . In physics it is often

called the action function. For a -projectable vector field on Y the expression
Z
(J r ) () = J r J r (3.4)

represents the variation of the variational functional, called its variational derivative or the
first variation induces by the vector field . A section ,W is called the extremal of
the variational functional (3.3) if it is its stationary point, i.e. the variational derivative of
vanishes for . The aim of the calculus of variations is to find equations for extremals
(in physics equations of motion of physical systems particles and fields) and in general to
study properties of variational functionals.
3.2. Lepage Equivalents of Lagrangians

The definition of the Lepage equivalent of a Lagrangian as one of concepts of coherent the-
ory on fibered manifolds was introduced by Krupka in [16]. It was based on primary ideas
of Lepage and Dedecker and on some generalizations of Sniatycki [40]. In the above cited
Krupkas paper horizontal and contact differential forms on jet prolongations of fibered
manifolds are introduced practically in the todays form, being called horizontal and pseu-
dovertical forms. Especially, for a differential p-form on J r Y a form h() on J r+1 Y is
defined (the notation of the time inclusive) by the relation
hh()(jxr+1 ), 1 p i
= h(jxr ), Tx j r T r+1 1 Tx j r T r+1 p i,
for r-jets of all sections of at x, and every collection of vector arguments 1 , . . . , p

on J r+1 Y . By this relation the form h() is defined. A form with the property that the
above value is zero whenever one of vector arguments is vertical, i.e. T r+1 i = 0, is
called horizontal. Inspired by the relation j r+1 p() = 0, where p() = r+1,r
h()
r r
a pseudovertical p-form on J Y , is defined by the property j = 0 for any section
of . Then, the properties of the mapping h : h() for n-forms are studied. The
definition of the horizontalization mapping is extended for (n + 1)-form by the relation

h(i() r+1,r ) = i()
for every r+1 -vertical vector field on J r+1 Y , proving the existence and uniqueness of
. The symbol i() denotes the contraction of the form by the vector field . Finally, a
mapping h : h() = is defined.
A Lepage n-form is then defined as follows: Let be an n-form on J r Y . is called
Lepagian form, if the (n+1)-form h(d) is horizontal with respect to r+1,0 . The properties
of Lepagian forms (in todays terminology Lepage forms) are studied and explicit formula
for h(d) is derived for the second order r = 2 and for a n-form on J 1 Y . The relation of
mappings h and h to Lepage congruencies [36] is mentioned and the connections of Lepage
forms with equations of extremals of Lagrangians is studied, giving rise to Euler-Lagrange
mapping by the formula
nX (J 1 Y ) E() 1Y (J 2 Y ), h(d) = E(h()) ,
where = F 0 is the volume element on X and F > 0 is a function. Note that here
E 1Y (J 2 Y ) is a uniquely defined pseudovertical 1-form. The Lepage equivalent of the
Lagrangian is in fact introduced by an example, but it is not explicitly defined includ-
ing the terminology. (Note that qX (J s Y ) or 1Y (J s Y ) in the previous definition of the
mapping E() means sq,X V or s1,Y V in the todays notation.)
Let us now present the contemporary definition of Lepage equivalents of Lagrangians,
as presented in [26]. The concept of Lepage forms is introduced by the following lemma.
Lemma. Let W Y be an open set, and let sn W . The following conditions are
equivalent:
(1) The (n + 1)-form p1 d is s+1,0 horizontal.
1
(2) For every s,0 -vertical vector field on W s = s,0 (W ) it holds h i d = 0,

(3) The form s+1,s has the chart expression
s

X
s+1,s = f0 0 + fi,j1 ...jk j1 ...jk i + , i = i 0 ,
xi
k=0
f0 j ,j ...j
dp fp,j1 ...jk fk 1 k1 = 0, sym(j1 . . . jk ), 1 k s,
yj1 ...jk
f0
fjs+1 ,j1 ...js = 0, sym(j1 . . . js+1 ).
yj1 ...js+1
Any form satisfying the Lemma is called the Lepage form. The structure of Lepage
forms is given by the following explicit formulas.
Theorem. Let W Y be an open set. A form sn W is a Lepage form if and only

if for every fibered chart (V, ), = (xi , y ), on Y such that V W , s+1,s has an
expression

s+1,s = + d + , where
s sk
!
X X
f0
= f0 0 + (1) dp1 . . . dp j1 ...jk i ,
k=0 =0
yj1 ...jk p1 ...p i
where f0 is a function, defined in coordinates by h = f0 0 , is a contact (n 1)-form,

and is a form of contactness 2.
The Lepage equivalent of a Lagrangian = L 0 is defined as such a Lepage form for

which h = up to a possible projection. This leads to the conclusion that the function f0
in the preceding relation is equal to L, i.e.
s sk
!
X X
L
= L 0 + (1) dp1 . . . dp j1 ...jk i . (3.5)
k=0 =0
yj1 ...jk p1 ...p i
For mechanics (n=1) and of order r we have

r1
!
Xdk k L
= L d + (1) k . (3.6)
k=0
d k qk+1
The Euler-Lagrange mapping assigning to every Lagrangian its Euler-Lagrange form

defined by the standard way as a dynamical form
r
!

X
L
E = E (L) 0 = (1) dj1 . . . dj 0
=0
yj1 ...j
is closely related to the concept of Lepage equivalents. For every Lepage equivalent of a
Lagrangian it holds p1 d = E , and, moreover, the following theorem can be proved:
Theorem. Let rn,X W be a Lagrangian and let sn W be its Lepage equivalent.

Then the section ,W is the extremal of the corresponding variational functional
(3.3) if and only if for every -vertical vector field defined on W such that the support
supp ( ) it holds
J s iJ s d = 0,
or equivalently, the Euler-Lagrange form of vanishes along J 2 .
Another important result lies in the first variational formula resulting directly from
properties of the Lepage equivalent of the Lagrangian , the expression for the vari-
ational derivative (3.4) of the variational functional and the formula for Lie derivative
= i d + di . Let be a section of , a -projectable vector field on Y . Then it
holds
J r J r = J s iJ s d + dJ s iJ s , (3.7)
and using the Stokes theorem
Z Z Z
J r J r = J s iJ s d + J s iJ s , (3.8)

The relations (3.7) and (3.8) give the infinitesimal first variational formula and the integral
first variational formula, respectively. There are very important consequences resulting
from the first variational formula for applications in variational physical theories.
-projectable vector field on V is called the generator of invariance transformations
of Lagrangian rn,X V , if
J r = 0. (3.9)
For such a vector field the left hand side of the integral first variational formula (3.8) van-
ishes. Let be an extremal of the functional , (V ), i.e. J s iJ s d = 0. Then
the first variational formula gives
Z
J s iJ s = 0. (3.10)

On the other hand, the integral (3.10) represents the flow of the quantity corresponding to
the integrand through the boundary of . It holds
J s iJ s = J s+1 s+1,s

iJ s = h iJ s .
So, we can interpret the expression
= h iJ s J s (3.11)
as a quantity obeying a conservation law along the extremal (so called elementary flow).
At the end of this section let us note that a concept of Lepage forms introduced primarily
for n-forms was later extended by Krupkova (see e.g. [33]) for (n + 1)-forms. However, in
this paper we dont discuss the properties of Lepage (n + 1)-forms separately. On the other
hand in Sec. 5 we mention the quite general definition of a Lepage form as resulting from
the structure of the variational sequence.
4. Application Lepage Equivalents of Lagrangians for String

Theories
Relativistic particles and strings are physical systems typically described by variational
theories. So, their equations of motion are derived from conditions for stationary sections of
a variational functional. These two situations are appropriate for application of the concept
of Lepage equivalents of Lagrangians, representing the case of first order mechanics and
first order field theory, respectively. There are of course physical arguments for the choice
of variational integral (for strings see [5]). Apart from these arguments, the obtained results
enables us to formulate these variational functionals as variational integrals for a certain
Lagrange structure of an appropriately chosen fibered manifold. In this section we give
corresponding Lagrange structures (, ) and calculate Lepage equivalents . Moreover,
using the infinitesimal first variational formula we can study invariance transformations of
Lagrangians.
Let the underlying fibered manifold be again (Y, , X). We denote fibered charts on
Y , associated charts on X and associated fibered chart on J 1 Y by the standard way (V, ),
(U, ), (V1 , 1 ). We will use the standard relativistic notation.
Recall that considering the infinitesimal first variational formula in the form (3.7) we
can see that quantities obeying conservation laws along the prolongations of extremals of
the Lagrangian are (n 1)-forms given by (3.11), where is a invariance transformation
of . From the physical point of view it is suitable to take into account the full Lep-
age equivalent, i.e. including a trivial Lagrangian of the same order as the Lagrangian
given by the physical situation itself. The last assumption is induced by the standard form
of boundary conditions of the variational problem. As a good example we can mention the
Lagrangian L = (1/2)mx2 +U (x) of a non-relativistic particle with one degree of freedom
(the motion along x-axis). For non-constant potential energy U (x) only the time transla-
tion is the invariance transformation of the Lagrangian. However, adding an appropriately
chosen trivial Lagrangian we can show that for linear potential energy U (x) = Kx + C (a
homogeneous field) the translation along x is invariance transformation as well.
4.1. Relativistic Particle

In the case of relativistic mechanics the base X of the underlying manifold is one-dimen-
sional (n=1), the unique coordinate being a non-physical parameter. Every fiber is a
time-space with metric g = g dx dx . Thus the fiber dimension is m = 4. Thus, we
have dim Y = 5. So on (Y, , X) we have coordinates
= ( ), = (, x ), 1 = (, x , x ), 0 3.
The Lagrange structure (, ) with classical Lagrangian is
q
1 = L1 d, L1 = mc g x x .
The Lagrange structure appropriate for both the zero and nonzero mass is given by the
Lagrangian !
1 g x x 2 2
2 = L2 d, L2 = + ( )m c ,
2 ( )
where ( ) is a function of the parameter. Both Lagrangians are reparametrizable. The

unique Lepage equivalent of this Lagrangian given by the general expression (3.6) is of the
form
L L L

= L d + = L x d + dx .
x x x
By the simple calculation we finally obtain
g x
1 = mc q dx , (4.1)
g x x
!
1 g x x g x
2 = ( )m2 c2 d dx . (4.2)
2 ( ) ( )
Now, let us study conservation laws resulting from invariance transformations of full
Lagrangians ( + 0 ) for both cases, 1 and 2 , 0 being a trivial Lagrangian. For our
situation it holds r = s = 1, and thus
d d
J 1 = + + , = x .
x x d d
Every trivial Lagrangian of the first order has the form h df for a function f = f (, x ).
Using Lepage equivalents (4.1) and (4.2) we obtain for full Lagrangians the resulting
general expressions for quantities given by (3.11)

f mc g x
f
1 = q , (4.3)

g x x x
and ! !
1 g x x f g x f
2 = ( )m2 c2 + . (4.4)
2 ( ) ( ) x
4.2. Classical Bosonic String

Strings can be studied within the field theory. The base of the underlying manifold is in case
of strings two dimensional, i.e. n = 2. The base coordinates denoted by 0 and 1 have the
meaning of the time evolution parameter and the position on the string, respectively. Every
fiber is again time-space, i.e. m = 4. Metric is denoted again by g = g dx dx . For
such a structure Lagrangian has a form
= L( i , x , xi ) d 0 d 1 .
The principal component of Lepage equivalent is, following (3.5)
L
= L d 0 d 1 + i (d 0 d 1 ),
xi i
L L L L

= L x x d 0 d 1 + dx d 1 dx d 0 . (4.5)
x0 0 x1 1 x0 x1
For an example we consider a bosonic string. Let us denote

h = hij d i d j = g xi xj d i d j .
The Lagrangian of a string is defined as

= L d 0 d 1 = T det h d 0 d 1 ,
q
L = T (g x0 x1 )2 (g x0 x0 )(g x1 x1 )
q
= T (g g g g ) x0 x0 x1 x1 . (4.6)
Note that in the original derivation of the variational functional of the string the Nambu-
Goto idea formulated e.g. in [8] is based on the variational integral (action)
Z
S = T d,

where d is understood as the elementary surface area. This means that dS is, correctly
speaking, the (unique) volume element of the relativistic time-space. Thus, in this approach
h can be understood as the induced metric for a mapping
: ( 0 , 1 ) (x0 ( 0 , 1 ), x1 ( 0 , 1 ), x2 ( 0 , 1 ), x3 ( 0 , 1 )),
x x i
h = g = (g ) d d j .
i j
However, in the approach based on Lagrange structures formalism on fibered manifolds
and their jet prolongations, the object h is defined on first jet prolongation of (Y, , X).
Using (4.5) for Lagrangian (4.6), we obtain after some simple calculations the principal
component of Lepage equivalent

= T det h d 0 d 1
T
+ (g g g g ) dx (x0 x0 x1 d 0 x1 x1 x0 d 1 ). (4.7)
det h
Now, let us study Lepage equivalents of trivial Lagrangians for the considered physical
situation. Taking into account that every trivial Lagrangian of first order has the form h d
(see [27]), where is a (n 1)-form on J r1 Y , we can obtain the corresponding principal
component of Lepage equivalents. In coordinates

0 = h d = di j + xj di X d i d j , (4.8)
where
= i d i + X dx .
Thus, we have 0 = L0 0 , where
1 0 1 X

L0 = 0
1
+ x0
x 1
0 X X

x1 + (x x x0 x1 ) . (4.9)
x 0 x 1 0
The corresponding Lepage equivalent has the form

0 = 0 + + d,
where is a 2-contact 2-form and is a contact 1-form. For conservation laws only the
principal component 0 is important, because h d has the character of a special type of
trivial Lagrangian and J s = 0.
1 0 X

0 = + (x x x1 x0 ) d 0 d 1
0 1 x 0 1
1 X X X

+ + x1 dx d 1
x 1 x x
0 X X X

+ + x0 dx d 0 . (4.10)
x 0 x x
Let

= 0
0
+ 1 1 +
x
be an invariance transformation of the Lagrangian ( + 0 ), where 0 is a trivial La-
grangian. We obtain the corresponding elementary flow
= + 0 = h iJ 1 + h iJ 1 0 , (4.11)
where

T
= T det h 1 + (g g g g )
det h
h io
x0 x0 x1 x0 x0 x0 x1 0 x1 x1 x0 1 d 0

T
+ T det h 0 + (g g g g )
det h
h i
x1 x1 x0 x1 x0 x0 x1 0 x1 x1 x0 1 d 1 (4.12)
and

0 = A00 0 + A01 1 + B0 d 0 + A10 0 + A11 1 + B1 d 1 , (4.13)
where
X 0 X X

A00 = x0 + x0 x0 ,
0 x x x
0 1 X 1

A01 = + x0 ,
1 0 1 x
0 X X X

B0 = + x0 ,
x 0 x x
1 0 X 0

A10 = 0
1
+ x1 ,
0 x
X 1 X X

A11 = x1 + x1 x1 ,
1 x x x
1 X X X

B1 =
1
+ x1
.
x x x
The invariance transformations of the Lagrangian ( + 0 ) given by (4.6) and (4.9) can
be obtained from (3.9) in principle. Conservation laws can be then studied using (4.11).
4.3. Polyakov Action for D-Dimensional Space

The Polyakov action is a variational integral arising from generalization of string Lagran-
gian to a D-dimensional space by the following way:
= L( i , x , xi ) d 0 d 1 , 0 i, j 1, 0 D 1,
T p
L= det f f ij g xi xj . (4.14)
2
Here f is the metric on the base, without a relation to g. Using again relation (4.5) we
can calculate the principal component of Lepage equivalent of the Lagrangian (4.14) in the
following form.
T p
= det f f ij g xi xj d 0 d 1
2p
+ T det f g xi dx (f 1i d 0 f 0i d 1 ). (4.15)
Note that the quantities for conservation laws can be calculated for the Polyakov action
by the quite analogous procedure as for a bosonic string.
5. Lepage Forms in Variational Sequences

In previous sections a concept of a Lepage equivalent of a Lagrangian was discussed and its
usefulness for application to concrete situations occurring in variational theories in physics
was demonstrated. In this section we present the generalized concept of a Lepage form
in the context of the variational sequence representation (for the complete solution of the
representation problem in field theories see [14] and [15], for mechanics see [29]). We
follow the general ideas and structures employed by Krupka in [23]. The main results are
formulated in the language of the theory of sheaves. (We use sheaves of germs of C -
differential q-forms.)
5.1. Variational Sequence

Let rq , q 0, be the direct image of the sheaf of smooth q-forms over J r Y by the jet
projection r,0 (functions are labeled as 0-forms). Denote by
(
ker p0 for 1 q n,
rq,c = (5.1)
ker pqn for n + 1 q dim J r Y,
where p0 and pqn are morphisms of sheaves induced by mappings p0 and pqn , assigning
to a form its horizontal and (q n)-contact component, respectively. The dimension of
r n+r
J Y is N = m n + n. We further denote
rq = rq,c + d rq1,c , (5.2)

d rq1,c is the image sheaf of rq1,c by d. Let us consider the sequence
{0} r1 rn rn+1 rn+2 rP {0}, (5.3)
with P = m n+r1

n + 2n 1 being the maximal nontrivial degree of forms of the type
(5.2). The arrows (except the first one) are given by exterior derivatives d. It can proved
that the sequence (5.2) is an exact subsequence of the de Rham sequence of forms
{0} r1 rn rn+1 rn+2 rN {0}.
The resulting quotient sequence is called the variational sequence of order r. The situation
is shown in the following figure. The variational sequence is, of course, also exact. We
denote the quotient mappings as follows
Eqr : rq /rq [] Eqr ([]) = [d ] rq+1 /rq+1 . (5.4)
The mappings Enr and En+1

r correspond to the Euler-Lagrange mapping and Helmholtz-
Sonin mapping of calculus of variations, respectively.
The global properties of the variational sequence are described using the abstract de
Rham theorem and can be found in [23]:
(1) Each sheaf rq is soft.
(2) The variational sequence (concisely {0} RY V) is an acyclic resolution of the

constant sheaf RY over Y .
(3) For every q 0 it holds H q ((RY , V)) = H q (Y, R), where
(Y, V) : {0} (Y, RY ) (Y, r0 ) (Y, rN ) {0}
is the cochain complex of global sections and H q ((RY , V)) denotes its q-th coho-
mology group.
5.2. Representation of the Variational Sequence

In this subsection we present the representation of the r-th order variational sequence by
forms. All details and proofs of theorems can be found in [15]. By the representation we
mean a mapping assigning to every class [] of the variational sequence of the r-th order
an appropriately chosen representative with specific properties mentioned in Sec. 1: R-
linearity, coordinate invariance, exactness of there sequence of representatives, projection
property of the operator assigning to a class its representative. Such a mapping is given
by Euler operator. Its construction was inspired by the idea of Anderson in [1], [2] for
the variational bicomplex. For finite order variational sequences the construction uses the
concept of the Lie derivative of differential forms with respect of vector fields along maps,
and is based on the generalized integration by parts procedure, both introduced in [15]. Let
us give here main results only. Integration by parts is described by the following lemma.
(Note that J = (j1 . . . j ), 0 r, is a multiindex and |J| = is its length.)
{0} {0} {0} {0}
d1 d2 dq1 dq dP 1 dP
{0} r1 r2 rq rP {0}
d0 d1 d2 dq1 dq dP 1 dP dP +1 dN 1 dN
{0} R r0 r1 r2 rq rP rP +1 rN {0}
EP +1 EN 1 EN
E0 EP
r1 /r1 r2 /r2 r /rq r /rP

E1 E2 Eq1 q Eq EP 1 P
{0} {0} {0} {0}

Lemma. Let (V, ), = (xi , y ) be a fibered chart on Y and rn+k V a form. Let
pk be expressed as
r
X
pk = J J .
|J|=0
Then there exists the decomposition
pk = I() + pk d pk R() (5.5)
where
r

(1)|J| dJ J .
X
I() = = (5.6)
|J|=0
and R() is a local k-contact (n + k 1)-form.

Moreover, we can write the operator I in the form
r
!
|J|
rn+k W 1
2r+1
X
I: I() = k (1) dJ y pk n+k W,
|J|=0
yJ
where we used the obvious identity

r
!
X
pk = 1
J y pk
k
|J|=0
yJ
We can see that I is clearly R-linear. (Note that the operator y stands for the contraction,
i.e. y = i means the contraction of a form by the vector field .)
It can be proved that the decomposition 5.5 is valid globally (the proof see in [15]) and
following the terminology in [2] we call the operator I interior Euler operator.
The following Theorem summarizes all properties of the interior Euler operator impor-
tant for the representation of the variational sequence.
Theorem. Let W Y be an open set and let rn+k W , 1 k N n, be a form.
Then
(1) (2r+1,r ) I() 2r+1

n+k W, (5.7)
(2) I(pk d pk R()) = 0, (5.8)
2
(3) I () = (4r+3,2r+1 ) I(), (5.9)
(4) ker I = rn+k W. (5.10)
Now let us define a family of mappings Rq :
Rq : rq W/rq W [] Rq ([]) sq W,
as follows
p0 for 0 q n, s = r + 1

Rq ([]) = I() for n + 1 q P, s = 2r + 1 (5.11)

for P + 1 q N, s = r
Evidently, in all three cases the mapping Rq assigns to every class [] of q-forms on J r Y
a correctly defined representative Rq ([]). In this sense, every class [] is represented by a
form Rq ([]).
Consider a sequence
{0} R0 (r0 W ) R1 (r1 W/r1 W ) RP (rP W/rP W )

RP +1 (rP +1 W ) RN (rN W ), (5.12)
where arrows denote the mappings
E q : Rq (rq W/rq W ) Rq+1 (rq+1 W/rq+1 W ) (5.13)
induced by the commutativity of diagrams
Eqr
rq W/rq W rq+1 W/rq+1 W (5.14)
Rq Rq+1
Rq (rq W/rq W ) Rq+1 (rq+1 W/rq+1 W )

Eq
i.e. E q Rq ([]) = Rq+1 Eqr ([]) = Rq+1 ([d ]).

The sequence (5.12) is called the representation sequence of the r-th order variational
sequence. The representation sequence is exact, as it is proved in [15]. There are practical
applications od the representation of the variational sequence. Especially, the representa-
tives Rq ([]) of classes of q-forms [] for q = n 1, n, n + 1, n + 2 are the well-known
forms on calculus of variations, i.e. r-th order currents, r-th order Lagrangians, r-th order
dynamical forms and r-th order Helmholtz-Sonin forms. So, the exactness of the represen-
tation sequence (5.12) enables us e.g. to solve the problem of trivial Lagrangians and the
inverse problem of calculus of variations.
5.3. Generalized Lepage Forms

One may generalize the concept of Lepage forms to all (n + k)-forms, k 0. Such a
generalization is directly induced by the structure of the representation of the variational
sequence. An (n + k)-form on J r Y is called Lepage if
pk+1 d = I(d ), i.e. pk+1 d = I(pk+1 d ). (5.15)
The equivalent definition resulting from the decomposition (5.5) can be given in the form
pk+1 d pk+1 R(d ) = pk+1 d R(pk+1 d ) = 0.
A Lepage form is also called the Lepage equivalent of the class [].
The definition of Lepage forms as well as the general definition of Lepage equivalents
was introduced also in [29] for mechanics. There the properties of Lepage forms for me-
chanics are studied in details and useful examples of chart expressions are presented. Note
that in the terminology of [29] the Lepage equivalent of the class [] is called the Lepage
equivalent of the source form. This is a form [] such that pk = I(), i.e. the
representative Rn+k [] (see (5.11)) in our notation.
6. Conclusion
The aim of the paper was to show the brief history of the concept of Lepage forms from
the initiating Lepages idea up to the most general definition resulting by the quite natural
way from geometrical structures built on underlying fibered manifolds and their jet prolon-
gations. The development of this concept supports the unquestionable meaning of Lepage
forms for the calculus of variations. Moreover, numerous examples following from varia-
tional physical theories, as e.g. classical Newtonian and relativistic mechanics, theory of
fields, string theories, show the usefulness of Lepage forms, especially Lepage equivalents
of Lagrangians, for many physical applications, as e.g. studies of conservation laws, etc.
It is evident that the relation (5.15) represents the most general possibility to define the
concept of a Lepage form. It ought to be mentioned that this definition inspired primarily by
Lepage is undoubtedly based on Krupkas complex theoretical considerations on Lagrange
structures and Lepage equivalents of Lagrangians and on his idea and studies of the finite-
order variational sequence.
Acknowledgements
The research is supported by the grant 201/06/0922 of the Czech Grant Agency and by
the grant MSM 0021622409 of the Ministry of Education, Youth and Sports of the Czech
Republic.
References
[1] I. M. Anderson, The Variational Bicomplex (book preprint, technical report of the
Utah State University, 1989; www.math.usu.edu/-fg mp).
[2] I. M. Anderson, Introduction to the variational bicomplex, Contemporory Math. 132
(1992) 5173.
[3] I. M. Anderson and T. Duchamp, On the existence of global variational principles,
Amer. J. Math. 102 (1980) 781867.
[4] A. V. Bocharov, V. N. Chetverikov, S. V. Duzhin, N. G. Khorkova, I. S. Krasilschik, A.
V. Samokhin, Yu. N. Thorkov, A. M. Verbovetsky and A. M. Vinogradov, Symmetries
and Conservation Laws for Differential Equations of Mathematical Physics ((I. S.
Krasilschik and A. M. Vinogradov, Eds.) Amer. Math. Soc., 1999).
[5] L. Brink, P. di Vecchia and P. Howe, A locally supersymmetric and reparametrization
invariant action for the spinning string, Phys. Lett. 65B(5) (1976) 471474.
[6] P. Dedecker, A property of differential forms in calculus of variations, Pac. J. Math. 7
(1957) 15451549.
[7] P. Dedecker and W. M. Tulczyjew, Spectral sequences and the inverse problem of
calculus of variations, In: Differential Geometric Methods in Mathematical Physics
(Proc. Internat. Coll., Aix-en-Provence, France, 1979, Lecture Notes in Math. 836,
Springer, Berlin, 1980) 498503.
[8] T. Goto, Prog. Theor. Phys. 46 (1971) 1560; Derived also in unpublished lectures of
Y. Nambu presented in Copenhagen, 1970.
[9] D. R. Grigore, The variational sequence on finite order bundle extensions and the
Lagrange formalism, Diff. Geom. Appl. 10 (1999) 4377.
[10] D. R. Grigore, Variationally trivial Lagrangians and locally variational equations of

arbitrary order, Diff. Geom. Appl. 10 (1999) 79105.
[11] I. S. Krasilschik, V. V. Lychagin and A. M. Vinogradov, Geometry of Jet Spaces and

Nonlinear Partial Differential Equations (Advanced Studies in Contemporary Mathe-
matics 1, Gordon & Breach, New York, 1986).
[12] M. Krbek, J. Musilova and J. Kasparova, The variational sequence: Local and global
properties, In: Proceedings of the Seminar on Differential Geometry (Math. Publica-
tions 2, Silesian University, Opava, 2000) 1538.
[13] M. Krbek, J. Musilova and J. Kasparova, The representation of the variational se-
quence in field theories, In: Steps in Differential Geometry (Proc. Colloq. Differential
Geometry, Debrecen, Hungary, 2000, (L. Kozma, P. T. Nagy and L. Tamassy, Eds.)
Debrecen, 2001) 147160.
[14] M. Krbek and J. Musilova, Representation of the variational sequence, Rep. Math.
Phys. 51 (2003) 251258.
[15] M. Krbek and J. Musilova, Representation of the Variational Sequence by Differential

Forms, Acta Appl. Math. 88 (2005) 177199.
[16] D. Krupka, Some geometric aspects of variational problems in fibred manifolds, Folia
Fac. Sci. Nat. Univ. Purk. Brunensis XIV (10) (1973) pp. 65.
[17] D. Krupka, On the structure of Euler-Lagrange mapping, Arch. Math. (Brno) 10

(1974) 353358.
[18] D. Krupka, A map associated to the Lepagean forms of the calculus of variations in
fibered manifolds, Czechoslovak Math. J. 27 (1977) 114118.
[19] D. Krupka, Natural Lagrangian Structures, In: Differential geometry (Semester in

Differential Geometry, Banach Center, Warsaw, 1979, Banach Center Publications
12, 1984) 185210.
[20] D. Krupka, On the local structure of Euler-Lagrange mapping of the calculus of vari-
ations, In: Differential Geometry and its Applications (Proc. Conf., Nove Mesto na
Morave, Czechoslovakia, 1980, (O. Kowalski, Ed.) Charles University, Praha, 1981)
181188.
[21] D. Krupka, Regular lagrangians and lepagean forms, In: Differential Geometry and
its Applications (Proc. Conf., Brno, Czechoslovakia, 1986, (D. Krupka and A. Svec,
Eds.) D. Reidel, Dordrecht, 1986) 85101.
[22] D. Krupka, Geometry of Lagrangean structures 2, 3, In: 14th Winter School on Ab-
stract Analysis (Proc. Conf., Srn, Czechoslovakia, 1986, Suppl. Rend. del Circ. Mat.
di Palermo 14, 1987) 178224.
[23] D. Krupka, Variational sequences on finite order jet spaces, In: Differential Geometry
and its Applications (Proc. Conf., Brno, Czechoslovakia, 1989, (J. Janyska and D.
Krupka, Eds.) World Scientific, Singapore, 1990) 236254.
[24] D. Krupka, The Geometry of Lagrange Structures, Preprint Series in Global Analysis
GA 7 (1997), Mathematical Institute, Silesian University, Opava 1987, pp. 82.
[25] D. Krupka, Variational sequences and bicomplexes, In: Differential Geometry and its
Applications (Proc. Conf., Brno, Czech Republic, 1998, (O. Kowalski, D. Krupka and
J. Slovak, Eds.) Masaryk University, Brno, 1998) 525532.
[26] D. Krupka, Global variational theory in fibred spaces, In: Handbook of Global Anal-
ysis ((D. Krupka and D. Saunders, Eds.) Elsevier, Amsterdam, 2008) 773836.
[27] D. Krupka and J. Musilova, Trivial Lagrangians in Field Theory, Diff. Geom. and
Appl. 9 (1998) 393505.
[28] D. Krupka and J. Musilova, Recent results in variational sequence theory, In: Steps
in Differential Geometry (Proc. Colloq. Differential Geometry, Debrecen, Hungary,
2000, (L. Kozma, P. T. Nagy and L. Tamassy, Eds.) Debrecen, 2001) 161186.
[29] D. Krupka and J. Sedenkova, Variational sequences and Lepage forms, In: Differential
Geometry and its Applications (Proc. Conf., Prague, Czech Republic, 2004, (J. Bures,
O. Kowalski, D. Krupka and J. Slovak, Eds.) Charles University, Prague, 2005) 617
627.
[30] O. Krupkova, Lepagean 2-forms in higher order Hamiltonian mechanics. I. Regularity,
Arch. Math. (Brno) 22 (1986) 97120.
[31] O. Krupkova, Lepagean 2-forms in higher order Hamiltonian mechanics. II. Inverse
problem, Arch. Math. (Brno) 23 (1987) 155170.
[32] O. Krupkova, Variational analysis on fibered manifolds over one-dimensional bases,
PhD Thesis, Silesin University, Opava and Charles University, Prague, 1992, pp. 67.
[33] O. Krupkova, The Geometry of Ordinary Variational Equations (Lecture Notes in
Mathematics, 1678, Springer, Berlin, 1997).
[34] O. Krupkova, Mechanical systems with non-holonomic constraints, J. Math. Phys. 38
(1997) 50985126.
[35] Th. H. J. Lepage, Calcul des Variations. Sur certaines formes differentielles
exterieures et la variation des integrales doubles, Comptes rendus des seances de
lAcademie des sciences, Tome cent-quatre-vingt-dix-septieme (1933) 17181719.
[36] Th. H. J. Lepage, Sur les champs geodesiques du Calcul des Variations I, II, Bull.
Acad. Roy. Belg., Cl. Sci. 22 (1936) 716729, 10361046.
[37] J. Musilova, Variational sequence in higher order mechanics, In: Differential Geom-
etry and its Applications (Proc. Conf., Brno, Czech Republic, 1995, (J. Janyska, I.
Kolar and J. Slovak, Eds.) Masaryk Unversity, Brno, 1996) 611624.
[38] J. Musilova and M. Krbek, A note to the representation of the variational sequence in
mechanics, In: Differential Geometry and its Applications (Proc. Conf., Brno, Czech
Republic, 1998, (I. Kolar, D. Krupka, O. Kowalski and J. Slovak, Eds.) Brno, 1999)
511623.
[39] J. Sedenkova, On the invariant variational sequences in mechanics, In: Rend. Cont.
Mat. Palermo (Proc. 22-nd Winter School Geometry nad Physics, Srn, Czech Repub-
lic, 2002) 185190.
[40] J. Sniatycki, On the geometric structure of classical field theory in Lagrangian formu-
lation, Proc. Camb. Phil. Soc. 68 (1970) 475484.
[41] J. Stefanek, A representation of the variational sequence in higher order mechanics,

In: Differential Geometry and its Applications (Proc. Conf., Brno, Czech Republic,
1995, (J. Janyska, I. Kolar and J. Slovak, Eds.) Masaryk University, Brno, 1996) 469
478.
[42] F. Takens, A global version of the inverse problem of calculus of variations, J. Diff.
Geom. 14 (1979) 543562.
[43] W. M. Tulczyjew, The Euler-Lagrange resolution, In: Differential Geometric Meth-

ods in Mathematical Physics (Proc. Internat. Coll., Aix-en-Provence, France, 1979,
Lecture Notes in Math. 836, Springer, Berlin, 1980) 2248.
[44] A. M. Vinogradov, On the algebro-geometric foundations of Lagrangian field theory,

Soviet. Math. Dokl. 18 (1977) 12001204.
[45] A. M. Vinogradov, A spectral sequence associated with a non-linear differential equa-

tion, and algebro-geometric foundations of Lagrangian field theory with constraints,
Soviet. Math. Dokl. 18 (1978) 144148.
[46] A. M. Vinogradov, The C-spectral sequence, Lagrangian formalism and conservation

laws I and II, J. Math. Anal. Appl. 100 (1) (1984) 1129.
[47] R. Vitolo, Finite order Lagrangian bicomplexes, Math. Proc. Cambridge Philos. Soc.
125 (1999) 321333.
[48] R. Vitolo, On different geometric formulations of Lagrangian formalism, Diff. Geom.

Appl. 10 (3) (1999) 225255.
[49] R. Vitolo, Finite order formulation of Vinogradovs C-spectral sequence, Acta Appl.
Math. 70 (1-2) (2002) 133154.
[50] R. Vitolo, Variational sequences, In: Handbook of Global Analysis ((D. Krupka and
D. Saunders, Eds.) Elsevier, Amsterdam, 2008) 11151163.
Chapter 2
L EPAGE F ORMS IN THE C ALCULUS

OF VARIATIONS
Olga Krupkova
Department of Algebra and Geometry, Faculty of Science,
Palacky University, Tomkova 40, 779 00 Olomouc, Czech Republic
and
Department of Mathematics, La Trobe University,
Bundoora, Victoria 3086, Australia
Abstract
Lepage forms represent a fundamental concept in the global calculus variations.
Inspired by the work of Lepage, they were introduced by Demeter Krupka in 1973 in
his seminal paper Some Geometric Aspects of Variational Problems in Fibered Man-
ifolds, published in the journal of Brno University Folia Fac. Sci. Nat. Univ. Purk.
Brunensis (see also arXiv:math-ph/0110005 for an electronic transcription). In this
paper we recall Lepage forms, their recent generalisations, the role they play in the
current global variational analysis, and their applications in the geometric theory of
differential equations, the theory of variational sequences, and higher-order mechan-
ics and field theory.
2000 Mathematics Subject Classification. 58-02, 70-02, 58E15, 70H50.

Key words and phrases. Lagrangian, Lepage form, the first variation formula, Euler
Lagrange form, Noether theorem, conservation law, Hamilton equations, regular La-
grangian, the variational sequence, null-Lagrangian, the inverse problem of the calculus
of variations, Helmholtz form.
1. Introduction
This paper is a survey of the theory of Lepage forms, its current development and numerous
applications in the calculus of variations on fibred manifolds.

Paper dedicated to Demeter Krupka on the occasion of his 65th birthday

E-mail address: krupkova@inf.upol.cz
28 Olga Krupkova
The concept of Lepage form (more precisely, Lepage n-form where n is the dimension
of the base manifold) was introduced in the early 1970s by D. Krupka, inspired by a clas-
sical paper of Th. Lepage [61]. D. Krupka was among the first who realised that fibred
and prolongation structures introduced by Ehresmann [14] represent an appropriate back-
ground for the investigation of first and higher order variational functionals on manifolds
(cf. [7, 8, 17, 22, 36, 77]). In his seminal paper [36] published 35 years ago he introduced
the fundamental concept of the Lepage equivalent of a Lagrangian, and since that time he
has systematically developed techniques and concepts of a global higher order variational
theory on fibred spaces where Lepage forms and variational sequences play a central role.
I am grateful to Demeter that from the beginning of my university education I had the pos-
sibility of joining his seminar, to learn his theory of Lepage forms in jet bundles, and to
participate in the fascinating process of discovering geometric structures in the calculus of
variations.
Figure 1. Demeter Krupka in the early 1980s lecturing on Lepage forms.
This paper, reviewing the current status of the theory of Lepage forms, starts by briefly
recalling the calculus of vector fields and horizontal and contact forms on jet bundles over
fibred manifolds [36, 37, 41], including the fundamental Krupkas decomposition formula
of a differential form into contact components, concepts and techniques that are essential
for understanding intrinsic constructions and properties of geometric objects appearing in
fibred variational analysis. Based on papers [36, 37, 38, 39, 41, 42, 43, 45, 49], the next
section is devoted to Lepage n-forms and their role in the theory of global variational func-
tionals. Lepage n-forms are fundamental for obtaining the intrinsic first variation formula,
that gives various geometric formulations of the EulerLagrange equations on one hand,
and, combined with the concepts of invariant variational functionals, of Noether theorems
on the other hand. Moreover, it is demonstrated that this setting provides a new look at
Hamiltonian theory and regularity of variational problems, that has not yet been completely
explored.
The reader should be aware that the family of Lepage equivalents of a Lagrangian con-
tains the well-known Cartan form [5, 78], and the PoincareCartan form [11, 17, 22, 36]
and its higher order generalisations [11, 15, 18, 31, 33, 41], that are very popular in vari-
Lepage Forms in the Calculus of Variations 29
ous geometric formulations of the calculus of variations (see e.g. [22, 12, 16, 18, 33, 34],
or the quite recent multisymplectic formalism [13, 21, 24, 62] and others). It should be
stressed, however, that Cartan-like forms, though often preferred, are not the only pos-
sibility; in some situations they are too restrictive (exterior differential systems for varia-
tional equations, Hamilton theory, regularity), or even apparently unavailable (higher order
field theories, homogeneous Lagrangians, non-fibred structures), so that some other Lepage
equivalents of a Lagrangian need to be considered (we refer to [4, 39, 68] for constructions
and [8, 36, 41, 54, 58, 59, 65, 71] for discussions and applications).
The last section of the present paper is concerned with generalisations of Lepage forms.
Roughly speaking, a motivation for introducing Lepage forms of higher degrees is to trans-
fer variational operators to the exterior derivative [8, 40]; as a first step, to introduce Lep-
age equivalents of EulerLagrange forms [50] in such a way that the following diagram
should be commutative:
EulerLagrange mapping
Lagrangian EulerLagrange form

y y
d
Lepage n-form Lepage (n + 1)-form
The idea comes from the solution of the inverse problem of the calculus of variations and of
the problem of the structure of null Lagrangians: with help of Lepage n-forms these prob-
lems are transferred to the application of the Poincare Lemma [40, 41]. Lepage equivalents
of EulerLagrange forms introduced in [50] and further studied in [25, 28, 29, 53, 54], play
a fundamental role in investigating variational equations by means of exterior differential
systems methods, enabling the study of the geometry of these equations and structure of
their solutions, symmetries and conservation laws, and exact methods of integration (see
e.g. [25, 51, 52, 55, 56, 57]). In particular, a new approach to Hamiltonian theory and
regular variational problems is achieved, associating Hamilton equations directly to the
EulerLagrange form (not to individual Lagrangians) [50, 51, 53, 54]. Finally, it is shown
how the concept of Lepage form can be generalised to arbitrary k-forms (k n), and how
it fits in with Krupkas variational sequence [35, 44, 48, 57].
2. Horizontal and Contact Forms on Fibred Manifolds

Throughout the paper, all manifolds and mappings are smooth and summation over repeated
indices is understood. We refer to [66] for the theory of jet bundles and to [36, 41] or [45, 51]
for details on the calculus of vector fields and differential forms on fibred manifolds. We
denote by : Y X a fibred manifold, dim X = n, dim Y = m + n, r : J r Y X the
r-jet prolongation of , (V, ), = (xi , y ), 1 i n, 1 m, a fibred chart on Y ,
and (Vr , r ), r = (xi , y , yj1 , . . . , yj1 ...jr ) the associated chart on J r Y . If dim X = 1,
we write (t, q ), and (t, q , q1 , . . . qr ), respectively, (t, q , q , q ) if r = 2. Next we put
0 = dx1 dx2 ... dxn , i1 ...ik = i/xik i1 ...ik1 , 1 k n, (2.1)
for a local volume on X and its contractions, and

di = i
+ yi + yji + + yj1 ...jr i (2.2)
x y yj yj1 ...jr
30 Olga Krupkova
if dim X > 1, respectively,
d
= + q1 + q2 + + qr+1

(2.3)
dt t q q1 qr
if dim X = 1, the operators of total derivative.

A mapping : U Y defined on an open subset U X is called a section of the
fibred manifold , if = idU . A section of r is called holonomic if = J r , i.e.,
the r-jet prolongation of a section of .
A vector field on J r Y , r 0, is called r -projectable if there exists a vector field
0 on X such that T r . = 0 , and r -vertical if projects onto a zero vector field
on X, i.e., T r . = 0. Quite similarly one can define a r,s -projectable or a r,s -vertical
vector field on J r Y , where r > s. Local flows of projectable vector fields transfer sections
into sections; consequently, -projectable vector fields on Y can be naturally prolonged to
vector fields on J r Y as follows. Given a -projectable vector field on Y with projection
0 , denote by {u } and {0u } the corresponding local one-parameter groups. For every u,
the mapping u is an isomorphism of the fibred manifold , i.e. u = 0u . Then
for every section , the composition = u 1 0u is a section of and we can define
the r-jet prolongation of u by
J r u (Jxr ) = Jr0u (x) (u 1

0u ). (2.4)
Then J r is a local flow corresponding to a vector field on J r , denoted by J r and called

the r-jet-prolongation of . The vector field J r is both r -projectable and r,s -projectable
for 0 s < r, and its r -projection, (resp. r,s -projection) is 0 (resp. , resp. J s ,
1 s r 1).
A q-form on J r Y is called contact if J r = 0 for every section of . The ideal
of contact forms on J r Y is generated by local contact 1-forms
j1 ...jk = dyj1 ...jk yj1 ...jk l dxl , 0 k r 1, (2.5)
and 2-forms dj1 ...jr1 . If dim X = 1 we write
k = dqk qk+1

dt, 0 k r 1. (2.6)
It is important to notice that contact 1-forms (2.5) give rise to a local basis of 1-forms on
J r Y adapted to the contact structure, namely, (dxi , , . . . , j1 ...jr1 , dyj1 ...jr ). In coordi-
nate expressions and computations in fibred coordinates this basis is much more convenient
than the canonical basis (dxi , dy , . . . , dyj1 ...jr ).
A q-form on J r Y is called r -horizontal (or 0-contact) if the contraction i vanishes
for every r -vertical vector field on J r Y . Similarly, is called r,s -horizontal if it vanishes
whenever at least one of its arguments is a r,s -vertical vector field. A contact q-form on
J r Y is called i-contact, 1 i q, if for every r -vertical vector field on J r Y , i is
(i 1)-contact.
By Krupkas theorem, the module of r+1,r -horizontal q-forms on J r+1 Y is the direct
sum of the submodules of i-contact forms, 0 i q, meaning that every r+1,r -horizontal
q-form on J r+1 Y has a unique decomposition into a sum of i-contact forms, 0 i q.
If we denote by h and pi , 1 i q, projectors onto the submodules of the horizontal and

i-contact forms, where 1 i q, we obtain the above theorem in the following form:
Every q-form on J r Y admits a unique decomposition

r+1,r = h + p1 + + pq (2.7)
into a sum of i-contact forms, 0 i q, on J r+1 Y . The operators h and pi are called the
horizontalisation and i-contactisation, respectively. We also say that h is the horizontal
part of and pi is the i-contact part of .
Note that every q-form where q > n is contact; it is called strongly contact if pqn =
0.
3. Lepage Forms and the First Variation

3.1. Variations
Let us start with Lagrangians and variations, following [36]. Consider a fibred manifold
: Y X, where X is an orientable manifold of dimension n. A Lagrangian of order r
for is defined to be a horizontal n-form on J r Y . In a fibred chart (V, ), = (xi , y ),
on Y ,
= L 0 ,
1
where L is a real function on Vr = r,0 (V ), i.e., L = L(xi , y , yj1 , . . . , yj1 j2 ...jr ).
Denote by a piece of X, i.e. a compact n-dimensional submanifold of X with bound-
ary, and () the set of smooth sections of over . The real valued function
Z
() J r R (3.1)

is called the variational function or action function of the Lagrangian over . Since
J r = 0 for any contact n-form on J r Y , the action function remains the same if one
considers the form + instead of . In other words, for any n form such that h = ,
Z Z
J r = J r . (3.2)

Let : U Y be a section defined on an open set U X. Let be a -projectable

vector field defined in a neighborhood of (U ), and 0 the projection of . If t is the local
1-parameter group of and 0t is the projection of t (i.e., 0t is the local 1-parameter
1
group of 0 ), then t = t 0t is a 1-parameter family of sections of , depending
smoothly on the parameter t. The family {t } is called the variation or deformation of the
section induced by the vector field . Often the vector field itself is called a variation
of . Now, taking the action function (3.1) for a fixed section (), and considering
a variation of , we get the following real-valued function, defined in a neighborhood I of
the point 0 in R: Z
It J r t R.
0t ()
32 Olga Krupkova
Differentiating this function at the point t = 0 we arrive at

nd Z o Z
J r t = J r LJ r . (3.3)
dt 0t () 0
The real number (3.3) is called variation of the action function at the point , induced by
the vector field . The function
Z
() J r LJ r R (3.4)

is then called the first variation of the action function (3.1) by the vector field .
Let us recall the key idea that lead Krupka to the discovery of Lepage forms: The prob-
lem is to find an intrinsic form of the first variation formula, i.e. a decomposition of the
integral in (3.4) to a sum of a term characterising extremals (EulerLagrange term) and
a boundary term (connected with conservation laws). To this end, one would like (and
should) utilize the well-known Cartan formula for the decomposition of the Lie deriva-
tive. Unfortunately, the result cannot be obtained by a direct application of this formula
to LJ r , since iJ r d depends not only upon the variation itself, but also upon pro-
longations (derivations) of . However, taking into account property (3.2) of the action
function, one can study the following question: Given a Lagrangian , is it possible to add
to a contact form in such a way that after the decomposition of LJ r ( + ) in (3.4) the
first term would depend upon the variation vector field only? Then, for such forms, Car-
tans formula would provide the desired intrinsic first variation formula, and, consequently,
a geometric form of the EulerLagrange equations and of conservation laws.
3.2. Lepage n-Forms and the First Variation Formula

Recall that n = dim X, and let s 0. Formalization of the above idea leads to the
following definition [36]:
Definition 3.1. A n-form on J s Y is called a Lepage n-form of order s if for every s+1,0 -
vertical vector field on J s Y
hi d = 0. (3.5)
Note that if is a Lepage n-form on J s Y then h is a Lagrangian on J s+1 Y . We write
h = = L0 . (3.6)
From the definition we obtain three Krupka theorems on the structure of Lepage
n-forms [36, 41]:
Theorem 3.2. Let be an n-form on J s Y . The following conditions are equivalent:
(1) is a Lepage form of order s.
(2) The (n + 1)-form p1 d is s+1,0 -horizontal.
(3) d satisfies

s+1,s d = E + F, (3.7)
where E is 1-contact and s+1,0 -horizontal, and F is at least 2-contact.

(4) In every fibred chart, s+1,s is expressed as follows:
s

X
s+1,s = L 0 + fi, j1 j2 ...jk j1 j2 ...jk i + , (3.8)
k=0
where is an arbitrary at least 2-contact n-form, and
L
f(js+1 , j1 j2 ...js ) = ,
yj1 j2 ...js+1
(3.9)
(j , j1 j2 ...jk1 ) L
f k = di fi, j1 j2 ...jk , 1 k s,
yj1 j2 ...jk
where (. . . ) means symmetrisation in the indicated indices, i.e. (i, j1 j2 . . . jp ) =

sym(i, j1 j2 . . . jp ).
Theorem 3.3. A form on J s Y is Lepage n-form if and only if in every fibred chart

s+1,s = + d + , (3.10)
where
s sk
X X L
= L 0 + (1)l dp1 dp2 . . . dpl j1 j2 ...jk i , (3.11)
k=0 l=0
yj1 j2 ...jk p1 p2 ...pl i
is a contact (n 1)-form, and is an at least 2-contact form.
Note that

s+1,s = + d + = + p1 d + , (3.12)
and that in this decomposition, the forms + p1 d and are global (being the at most

1-contact and the at least 2-contact part of s+1,s , respectively). We stress that, on the
contrary, the decomposition + p1 d need not be invariant under the change of fibred
coordinates, meaning that in general is not a global differential form.
Theorem 3.4. If is a Lepage n-form then the 1-contact part p1 d of d depends only on
h = = L0 and reads
L s+1
X L
p1 d = E = (1)l+1 dp1 dp2 . . . dpl 0 . (3.13)
y l=1
yp1 p2 ...pl
Definition 3.5 ([36]). Given a Lagrangian , we say that a form is a Lepage equivalent
of if is a Lepage n-form and h = . The (n + 1)-form E = p1 d is then called
the EulerLagrange form of , and its fibred-chart components are called EulerLagrange
expressions.
Theorem 3.6 (Krupka [36, 41], Marvan [63]). Every Lagrangian has a Lepage equivalent.
Taking into account the above theorems we obtain:

34 Olga Krupkova
Corollary 3.7.
(1) Every Lagrangian of order r has a Lepage equivalent of order 2r 1, that is
generally nonunique. Every Lepage equivalent of takes the form as described by theorem
3.3 or 3.2 (4).
(2) Every Lagrangian has a Lepage equivalent that is at most 2-contact. If is of
order r that Lepage equivalent is of order 2r 1, is generally nonunique, and reads
= + p1 d, (3.14)
where is given by formula (3.11) and is a contact (n 1)-form. In particular, ev-
ery Lagrangian of order r has (global) 2-contact Lepage equivalents that are 2r1,r1 -
horizontal; they take the form (3.14) where is an arbitrary 1-contact 2r1,r1 -horizontal
(n 1)-form.
(3) To every Lagrangian there exists a uniquely determined EulerLagrange form E .
If is defined on J r Y then E is generally on J 2r Y and is given by formula
L r
X L
E = (1)l+1 dp1 dp2 . . . dpl 0 = p1 d, (3.15)
y l=1
yp1 p2 ...pl
where is any Lepage equivalent of .

It should be stressed that the splitting of in (3.14) into the term that is determined
by the Lagrangian, and an auxiliary term p1 d generally is not invariant under changes of
fibred coordinates. This means that the expression (3.11) may have only a local meaning,
not providing a differential form on J 2r1 Y . If happens to be a global differential form,
we speak about the (higher-order) PoincareCartan form associated to the Lagrangian ,
or, about the PoincareCartan equivalent of .
Equipped with Lepage n-forms, we are able to write down the intrinsic first variation
formula.
Theorem 3.8 (Krupka [36, 41]). Let be a Lagrangian of order r, its Lepage equivalent,
a -projectable vector field on Y . Then
(i) the Lie derivative LJ r is expressed by the formula
LJ r = hiJ 2r1 d + hdiJ 2r1 , (3.16)
(ii) for any section of ,

J r LJ r = J 2r1 iJ 2r1 d + dJ 2r1 iJ 2r1 , (3.17)
(iii) for any piece of X with boundary , and any section of such that dom ,
Z Z Z
r 2r1
J LJ r = J iJ 2r1 d + J 2r1 iJ 2r1 , (3.18)

and, in all the above formulas, the first term on the right-hand side depends on the vector
field only (not on its prolongations).
Formula (3.16) or (3.17) is called the infinitesimal first variation formula, (3.18) is the
integral first variation formula.
3.3. Examples of Lepage Equivalents of Lagrangians

We have seen that to a Lagrangian one has in general many Lepage equivalents. However, a
few particular cases are known, when either the Lepage equivalent is unique, or, the family
of Lepage equivalents contains some interesting distinguished representatives.
When discussing special cases, first of all, one has to mention classical and higher order
mechanics. In the case dim X = 1 we denote local fibred coordinates on Y by (t, q ), and
the associated higher order coordinates by (t, q , q1 , . . . , qr ), or (t, q , q , q , . . . ). The
corresponding local contact 1-forms then read j = dqj qj+1 dt, 0 j r. Taking into
account that now is a 1-form, i.e., at most 1-contact, and is a contact function (i.e. = 0),
we immediately obtain:
Theorem 3.9. Let dim X = 1. To every Lagrangian on J r Y there exists a unique Lepage
equivalent. In fibred coordinates where = Ldt, it reads
Ld L d2 L r1 d
r1 L
= L dt + + + (1)
q1 dt q2 dt2 q3 dtr1 qr
L (3.19)
d L L
+ +
r2 + r1 .
qr1 dt qr qr
The above form is called (higher order) Cartan form. For a first order Lagrangian
the formula is reduced to the well-known one, introduced to the calculus of variations and
classical mechanics by Whittaker [78] and Cartan [5]:
L L L
= Ldt + = L q dt + dq . (3.20)
q q q
Another important special case is dim X = n > 1 and r = 1, the so-called first order
field theory. If is a first order Lagangian then theorem 3.3 gives all first-order Lepage
equivalents of as follows:
= + p1 d +
L
ik

ij ijk (3.21)
= L0 + d k i (g + dk g ) j i + ,
g
yi
where is an arbitrary at least 2-contact n-form, and is an arbitrary 1-contact (n 1)-

form, = 12 gij ij + 16 gijk d ijk , where we assume the components be functions
skewsymmetric in the upper indices. Obviously the part of , uniquely determined by the
Lagrangian, i.e.
L
= L0 + i , (3.22)
yi
is invariant with respect to changes of fibred coordinates; it is called the PoincareCartan
form. We can see that the form is uniquely characterised and intrinsically defined as the
Lepage equivalent of that is 1,0 -horizontal and at most 1-contact.
It is known that behind the PoincareCartan form, the family (3.21) contains the fol-
lowing two global Lepage forms, completely determined by the Lagrangian:
36 Olga Krupkova
Caratheodory form
1 L 1
L n
C = L dx1 +
L dxn
+
Ln1 y1 1
ynn
(3.23)
L 1 L L 1
= L 0 + j + 2 j 1 j 2 + ,
yj 2L yj11 yj22
this form is invariant under changes of any (not only fibred) coordinates (cf. [65]), and
Krupka form [39] (later also found by Betounes [4])
n
1 kL
K
X
= L 0 + 1 k j1 jk . (3.24)

k=1
(k!)2 yj1 yjkk
1
This form has an important property dK

= 0 if and only if the Euler-Lagrange form of
vanishes identically.
As pointed out in [36], an interesting situation arises when dim X = n > 1 and r =
2 (second order Lagrangians in field theory). First of all, in this case is no longer
intrinsically characterised similarly as for r = 1, i.e. as the Lepage equivalent of that is
3,1 -horizontal and at most 1-contact: indeed, such form is no longer unique and is given
by formula
L L L
= L 0 + dk + gik i + gij j i , (3.25)
yi
yik
yij
where gjk are arbitrary functions skewsymmetric in the upper indices. However, one can
check by a direct calculation that the part of determined by the Lagrangian, i.e.
L L L
= L 0 + dk i + j i , (3.26)
yi
yik yij
is invariant under fibred coordinate transformations [41]. This means that for second order
Lagrangians we still have a well-defined PoincareCartan form. Also the question on the
existence of a higher order Caratheodory form has an affirmative answer; the form is defined
on J 3 Y by the following formula [68]:
1 L
L 1 L 1
C = L dx1 + + d j1
1 j 1 . . .
Ln1
y1j11y11y1j 1
L (3.27)

n L n L n
L dx + d j n n + n j n .
ynn ynjn ynjn
Quite a different story concerns a possible Krupka form K

for second order Lagrangians:
yet, the question of whether such a (global) form may be found is still open, however,
some arguments coming from the geometry of homogeneous Lagrangian systems support
the suggestion that it does not exist [68].
For dim X = n > 1 and r > 2 (higher order field theory) the situation is even more
complicated. Surprisingly, in the class we have no global differential form determined
by the Lagrangian, meaning that a higher-order PoincareCartan form does not exist;
so we have no distinguished Lepage equivalent of of Cartan type. The non-existence

of a global higher-order PoincareCartan form initiated studies of geometric constructions
providing a globalization of the expression (see e.g. ([6, 15, 18, 23, 31, 33]). From the
point of view of the general theory of Lepage forms the problem of globalization of
is very simple: a desired form always exists and is obtained by adding to an appropriate
term p1 d where is 1-contact and horizontal with respect to the projection 2r1,r1 (cf.
corollary 3.7).
3.4. Extremals and EulerLagrange Equations

As a first application of Lepage forms one obtains different global expressions for the Euler
Lagrange equations [36, 42].
A section () is called a stable point of the action function of with respect to
a variation if Z
J r LJ r = 0. (3.28)

is said to be an extremal of on if (3.28) holds for all variations of such that

supp( ) ; here supp( ) denotes the support of the vector field along the
section , defined as the closure of the set {x dom |( )(x) 6= 0}. Finally, is
called an extremal of the Lagrangian if it is an extremal of on every in the domain of
definition of .
Necessary and sufficient conditions for a section be an extremal of a Lagrangian follow
from the first variation formula:
Theorem 3.10. Let be a Lagrangian on J r Y , a Lepage equivalent of . Let be a

section of . The following conditions are equivalent:
(1) is an extremal of .
(2) For every vector field on J 2r1 Y
J 2r1 i d = 0 . (3.29)
(3) For every -projectable vector field on Y
J 2r1 iJ 2r1 d = 0 . (3.30)
(4) For every -vertical vector field on Y
J 2r1 iJ 2r1 d = 0 . (3.31)
(5) J 2r1 is an integral section of the exterior differential system generated by the sys-
tem of n-forms
i d, where runs over all vertical vector fields on J 2r1 Y . (3.32)

38 Olga Krupkova
(6) The Euler-Lagrange form E vanishes along J 2r , i.e.,
E J 2r = 0 . (3.33)
(7) In every fibred chart, satisfies the following system of differential equations
r
!
X
k L
(1) dj1 djk J 2r = 0, 1 m. (3.34)
k=0
yj1 jk
Any of the equivalent conditions (2)(7) above may be called the Euler-Lagrange equa-
tions of the Lagrangian . Note that (4) means that although in place of variations one can
take arbitrary vector fields, extremals are determined only by the vertical part of variations
defined on Y (they do not depend upon components at /yj , /yjk , . . . ). Thus among
equations (2)(4), (4) represents the most simple and most frequently used form of intrinsic
EulerLagrange equations.
We stress that for dim X = 1 (3.32) is a system of 1-forms, annihilating a distribution
on J 2r1 Y , called the EulerLagrange distribution [46].
Notice that for any Lepage equivalent of , d = d + F , where F is at least
2-contact. Hence i F is contact and vanishes along J 2r1 . This means that for the Euler
Lagrange equations (3.31) the at most 2-contact part of a Lepage equivalent is essential, so
that we can simply consider them in the form
J 2r1 iJ 2r1 d = 0, . (3.35)
Finally, recall that Lagrangians 1 , 2 are said to be equivalent if E1 = E2 (possibly

up to a projection). A Lagrangian is called null if E = 0. Obviously, 1 and 2 are
equivalent if and only if 1 2 is null.
3.5. Hamilton Equations, Regular Lagrangians

We have seen that EulerLagrange equations of a Lagrangian have a geometric mean-
ing as equations for holonomic integral sections of the exterior differential system (3.32),
where is a Lepage equivalent of . However, one can study all integral section of the cor-
responding EDS, and explore that for different Lepage equivalents of different exterior
differential systems are obtained.
Let be a Lagrangian on J r Y , its Lepage equivalent. We call the EDS defined by
(3.32) Hamiltonian system of . Equations for integral sections of a Hamiltonian system,
i.e.,
i d = 0, 2r1 -vertical , (3.36)
are called Hamilton equations, the integral sections are then called Hamilton extremals
of [42]. Obviously, the set of prolongations of extremals of is a subset of the set of
Hamilton extremals. Properties of this subset, however, depend both upon properties of the
Lagrangian and the choice of its Lepage equivalent : due to Dedecker [8] and Krupka
and his collaborators, this understanding is a key to the concept of a regular variational
problem [49, 53, 54, 58, 59, 70, 71].
In higher-order mechanics we associate to a unique Lepage equivalent, , and hence

a unique Hamiltonian system, generated by 1-forms i d , where runs over vertical vec-
tor fields. A well-known result says that if a Lagrangian on J r Y satisfies the regularity
condition
2L
det 6= 0, (3.37)
qr qr
then every Hamilton extremal of is the (2r 1)-th prolongation of an extremal of ,
i.e. the sets of extremals and Hamilton extremals of are in 1-1 correspondence; moreover,
the mapping (t, q , . . . , q2r1
, p , . . . pr1 ), where pi are components
) (t, q , . . . , qr1

of at dqi , 0 i r 1, is a local diffeomorphism (called Legendre transformation)
[9, 46].
In the case dim X = n > 1, the Hamiltonian theory is richer, since the Lepage equiv-
alent of is no longer unique. Going back to De Donder and Goldschmidt and Stern-
berg [11, 22], the most commonly considered Hamilton equations are those based on the
PoincareCartan form (cf. also Hamilton equations in the multisymplectic formalism
[13, 21, 24, 62]); we call these Hamilton equations De DonderHamilton equations. The
result now is completely analogous to that of mechanics: if a Lagrangian on J 1 Y satisfies
the regularity condition
2L
det 6= 0, (3.38)
yi yj
then every Hamilton extremal of is the prolongation of an extremal of , meaning that

De DonderHamilton equations and EulerLagrange equations are equivalent. Writing
L L
= L yi 0 + dy i = H0 + pi dy i , (3.39)
yi yi
we get the Hamiltonian and momenta of . The above regularity condition then guarantees
that (xi , y , yj ) (xi , y , pj ) is a local coordinate transformation on J 1 Y (Legendre
transformation). In Legendre coordinates, De DonderHamilton equations read
y H pj H
j
= j, j
= . (3.40)
x p x y
A higher-order version of this result was first considered by De Donder [11]. However,
the generalisation is not so straightforward, since may be not globally well-defined, and
if globalised, is non-unique. Saving the property of being determined completely by the
Lagrangian, one has to resign on global Hamilton equations. Given a Lagrangian of order
r, and its local PoincareCartan equivalent (3.11), Hamilton equations now read
i d = 0 for every 2r1 -vertical vector field on J 2r1 Y (3.41)
and are defined on the domain W of the coordinates (xi , y , yj1 , yj1 j2 , . . . , yj1 j2 ...jr ). Put
= H0 + pi dy i + pj1 i dyj1 i + + pj1 ...jr1 i dyj1 ...jr1 i , (3.42)

40 Olga Krupkova
where
rk1
L
pj1 ...jk i =
X
(1)l dp1 dp2 . . . dpl , 0 k r 1,
l=0
yj1 ...jk p1 ...pl i
r
(3.43)
X
H = L + pj1 ...jk yj1 ...jk ,
k=1
and denote by [q1 . . . qs ] the number of all different sequences arising by permuting the
sequence q1 , . . . , qs . As proved by Shadwick [69], if the rank of all the matrices
!
1 2L
(3.44)
[j1 . . . j2rs (pr+1 . . . ps ] [p1 . . . pr )] yj1 ...j2rs (pr+1 ...ps yp1 ...pr )

is maximal, where r s 2r 1, the , j1 j2rs label columns, , p1 ps

label rows, and the brackets ( ) denote symmetrization in the indicated indices, then every
Hamilton extremal of passing in W is of the form 2r1,r = J r where is an
extremal of . Moreover, the system of functions
xi , y , yj1 , . . . , yj1 ...jr1 , pj1 ...jr , . . . , pj1 , j1 jr (3.45)
forms a part of a coordinate system on W (called Legendre coordinates). In Legendre

coordinates De DonderHamilton equations read
yj1 ...jk H pj1 ...jk l H

i
= j1 ...jk i , l
= , (3.46)
x x yj1 ...jk

sym{j1 ...jk i} p
where 0 k r 1, and in the second set of equations, summation over l takes place. We
note that Shadwicks regularity condition above can be put into a geometric form and can
be expressed equivalently by means of certain bilinear foms or by a linear mapping [20, 34].
Considering in place of a global form = +p1 d we get to non-unique global
De DonderHamilton equations. There arises the question of whether Shadwicks regular-
ity condition still can guarantee an analogous nice correspondence between extremals and
Hamilton extremals, According to Krupka [43] and Gotay [24] the answer is affirmative: If
Shadwicks regularity condition is satisfied then every solution of De DonderHamilton
equations of is of the form 2r1,r = J r where is an extremal of . However,
note that De DonderHamilton equations of in Legendre coordinates may have a more
complicated form compared to (3.46). For other interesting aspects of the theory see e.g.
[1, 10, 16, 24, 34, 42, 43, 67].
It is important to notice that for higher-order regular Lagrangians De DonderHamilton
equations are no longer equivalent with the EulerLagrange equations: the subset of Hamil-
ton extremals that is in bijective correspondence with extremals consists of sections that are
holonomic up to the order r (note r = order of the Lagrangian).
Within De DonderHamilton theory the concept of regularity can be reconsidered to
give regularity conditions for higher order Lagrangians different from Shadwicks regularity
condition. The idea proposed in our joint paper [49] was that the true order of the De
DonderHamilton equations in essential. Namely, for some Lagrangians of order r 2,
their PoincareCartan form is 2r1,s -projectable, where s < 2r 1; in this case, it is

apparently inappropriate to apply the standard procedure leading to considering Hamilton
equations of order 2r 1. The problem should instead be studied as a problem of order s.
In the above mentioned paper this idea was applied to the class of second order Lagrangians
with 3,1 -projectable . This concerns the following second order Lagrangians affine in
the second derivatives: = L0 where
L = L0 (xi , y , yj ) + hpq i
(x , y ) ypq . (3.47)
In [49] we called a Lagrangian regular if all its Hamilton extremals were holonomic, and
we proved the following theorem:
Theorem 3.11 (Krupka and Stepankova). Let be a Lagrangian of the form (3.47). If the
condition !
2 L0 hik
hki

det 6= 0 (3.48)
yi yk y y
is satisfied then is regular, its EulerLagrange and De DonderHamilton equations are
equivalent, and the mapping
L0 hjk
hjk
hkj

(xi , y , yj ) (xi , y , pj ), pj = + y (3.49)
yj xk y y k
is a local coordinate transformation on J 1 Y (Legendre transformation).
It should be stressed that for second order Lagrangians (3.47) the momenta (3.49) and
Hamiltonian
L0 hjk
H = L0 + yj yj yk , (3.50)
yj y
are first order functions. De DonderHamilton equations expressed in Legendre coordinates
(3.49) then take the usual form (3.40). The above results directly apply to the Einstein
Hilbert Lagrangian (scalar curvature) of the General Relativity Theory (see [49] for explicit
computations). Thus, within this setting, gravity naturally appears as a first order regular
theory (without constraints). Later the above ideas were applied to study also some other
kinds of higher order Lagrangians with projectable PoincareCartan forms by Garcia and
Munoz [19, 20].
We have seen that regularity is a property of a Lepage form, rather than of a Lagrangian
itself, and it carries a geometric content that all Hamilton extremal are holonomic (for first
order Hamiltonian systems), or holonomic up to a proper order (for higher order systems).
However, in field theory we have for a given Lagrangian a family of Hamiltonian systems,
defined by different Lepage equivalents of . In this way we expect to obtain regularity
conditions depending on and some auxiliary functions coming from parts of that are
not uniquely determined by the Lagrangian, and we may consider the Hamiltonian theory
in a completely new setting: instead of asking whether a Lagrangian is regular (that usually
means regularity of its De DonderHamilton system) we may ask a question if the family
of Hamiltonian systems associated with a Lagrangian contains a Hamiltonian system that
is regular. This possibility was first observed by Dedecker [8], and recently studied for
42 Olga Krupkova
Lepage equivalents of first and higher order Lagrangians in [53, 54, 58, 59, 70, 71]. In
[59] a procedure of regularisation of conventionally singular Lagrangians was applied to
important physical field Lagrangians (Dirac field and electromagetic field Lagrangian), and
corrected Hamiltonians and momenta were found, providing Hamilton equations without
constraints.
3.6. Symmetries and Conservation Laws, Noether Theorem

Lepage forms play an important role also in the theory of invariant variational functionals
developed by Krupka in the early 1970s [36, 38].
Let be a Lagrangian on J r Y , E its Euler-Lagrange form. An isomorphism of
the fibred manifold : Y X is called an invariance transformation of , respectively,
generalised invariance transformation of if
J r = 0, respectively, J 2r E = 0. (3.51)
If is a -projectable vector field on Y and {u } is a local one-parameter group of such

that u is an invariance transformation, respectively, generalised invariance transformation,
for every u, we get the following infinitesimal version of the above invariance conditions:
LJ r = 0, respectively, LJ 2r E = 0. (3.52)
The above conditions are called the Noether equation and the NoetherBessel-Hagen equa-
tion, respectively.
Noether and Noether-Bessel-Hagen equations can be used to find the group of invari-
ance or generalised invariance transformations of a given Lagrangian, or conversely, to find
all Lagrangians (EulerLagrange forms) invariant with respect to a given group of transfor-
mations of Y ; of course, solving the NoetherBessel-Hagen equation in this case we obtain
invariant differential equations that need not come from a Lagrangian: variationality is an
additional property to be satisfied.
We have an important theorem due to Krupka [38]:
Theorem 3.12. Let be a Lagrangian on J r Y and E its EulerLagrange form. Given an

isomorphism of the fibred manifold : Y X, respectively, a -projectable vector field
on Y , it follows that
J 2r E = EJ r , LJ 2r E = ELJ r . (3.53)
Corollary 3.13. Every invariance transformation of a Lagrangian is its generalised in-

variance transformation.
Corollary 3.14. If LJ 2r E = 0 then LJ r is a null Lagrangian.
As proved in [26, 47] (see also the next section), the above condition means that around
every point in J r1 Y there is an (n 1)-form such that LJ r = hd.
Equipped with the concept of invariant Lagrangian, we obtain a classical result of
Emmy Noether [64] and its generalisation, that in the setting of Lepage forms appear as
an easy consequence of the first variation formula (Krupka [38]):
Theorem 3.15 (Noether Theorem). Let be a Lagrangian on J r Y , its Lepage equivalent.

If a -projectable vector field on Y generates invariance transformations of , and if is
an extremal then
d(J 2r1 iJ 2r1 ) = 0. (3.54)
Theorem 3.16 (Generalised Noether Theorem). Let be a Lagrangian on J r Y , its Lep-

age equivalent. If a -projectable vector field on Y generates generalised invariance
transformations of , and if is an extremal then
d(J 2r1 (iJ 2r1 )) = 0, (3.55)
where is a (local) (n 1)-form such that LJ r = hd.
Equation (3.54), respectively (3.55), is called a conservation law, and the (n 1)-
form iJ 2r1 respectively, iJ 2r1 , that is closed along prolongations of extremals,
is called a conserved current. Using the Poincare lemma we get around every point x
X an (n 2)-form for which d = f i i is a conserved current on X, so that the
corresponding conservation law d(f i i ) = 0 takes a divergence form div f = 0. In
mechanics (dim X = 1) the situation is simpler: a conserved current is a function, F ,
and a conservation law reads F J 2r1 = const. Therefore, F is called constant of the
motion.
Since in field theory the Lepage equivalent of is not unique, there arises a question
on how a conserved current depends upon a choice of a Lepage equivalent of . Using
formulas (3.10) and (3.14), i.e. = + where is an at least 2-contact form, and taking
into account that contraction of gives a contact form that vanishes along prolongations
of sections of , we can see that the conservation law corresponding to depends merely
upon the at most 2-contact part of ; hence, for any Lepage equivalent of an invariant
Lagrangian , the conserved current is the (n 1)-form iJ 2r1 . Of course, in higher
order field theory we must take into account the non-uniqueness of and the non-existence
of a global PoincareCartan form , discussed in the previous sections.
4. Lepage Forms and Differential Equations

4.1. The Inverse Problem of the Calculus of Variations
By a dynamical form of order s we understand a 1-contact and s,0 -horizontal (n + 1)-
form E on J s Y [51]. This means that in every fibred chart (V, ), = (xi , y ) on Y ,
1
E = E 0 , where the components E are functions on Vs = s,0 (V ). A section of
satisfying
E J s = 0 (4.1)
is called a path of E. Equation (4.1) in fibred coordinates takes the form E J s = 0,
1 m, that is a system of m partial, respectively ordinary, differential equations of
order s if dim X > 1, respectively if dim X = 1.
Let E be a dynamical form on J s Y . We have the following important definitions [40]:
E is called locally variational if to every point in J s Y there exists a neighbourhood U , an
integer r s, and a Lagrangian defined on s,r (U ) such that E|U = E . E is called
44 Olga Krupkova
globally variational if is defined on J r Y , i.e., if E is locally variational and has a global

Lagrangian. Paths of a locally variational form E are called extremals of E.
Note that extremals of a globally variational form coincide with extremals of any of
its global Lagrangians. On the other hand, if E is locally but not globally variational,
the family of extremals of E contains, but is not equal to, the family of extremals of any
individual Lagrangian of E.
Remark 4.1. We have seen that a dynamical form on J s Y determines a system of m sth
order differential equations (4.1). Conversely, given a system of m sth-order differential
equations E J s = 0, 1 m, for graphs of mappings from Rn to Rm (n 1) we
can represent it by a dynamical form E defined on an open subset of J s (Rn Rm ) setting
E = E dt if dim X = 1, respectively, E = E 0 if dim X > 1. Such equations
are called variational if the associated dynamical form E is locally variational.
There is a close connection between locally variational forms and closed forms:
Theorem 4.2 (Krupka [40, 41]). Let E be a dynamical form on J s Y . E is locally varia-
tional if and only if to every point x J s Y there exists a neighbourhood W and an at least
2-contact form FW on W such that the form W = E + FW is closed.
Since dW = 0, we have by the Poincare lemma to every point z W a neighborhood

U and a form on U such that d = U . However, the one-contact part p1 d of d equals
the dynamical form E (restricted to U ), i.e. p1 d is horizontal with respect to the projection
onto Y . This means that is a Lepage n-form. Now, = h is a Lagrangian such that
E = E|U . In this way we can obtain a local Lagrangian for E simply by means of the
Poincare lemma. Moreover, the geometric structure of J s Y enables one to introduce a
modified homotopy operator, A, adapted to the contact structure [41] with the following
properties: (1) Ad + dA = d, (2) if is k-contact then A is (k 1)-contact. Then,
however,
= hA = Ap1 = AE, (4.2)
in fibered coordinates
Z 1
L = q E (t, uq , uq1 , . . . , uqs )du (4.3)
0
if dim X = 1, and
Z 1

L=y E (xi , uy , uyj , . . . , uyj1 ...js )du (4.4)
0
if dim X > 1, respectively. The formula for L was obtained in [74] and [75], and is called
TontiVainberg Lagrangian.
Necessary and sufficient conditions for a dynamical form be locally variational were
first studied by Helmholtz for the case of second order ordinary differential equations [30].
A generalisation to ordinary differential equations of an arbitrary order is due to Vander-

bauwhede [76], and to higher-order partial differential equations to Anderson and Duchamp
[3], and Krupka [40, 41]. The celebrated conditions read as follows:
Theorem 4.3. A dynamical form E on J s Y is locally variational if and only if its compo-
nents in every fibred chart satisfy the following conditions:
s
!
E l E
X k dkl E
dim X = 1 : (1) (1)k = 0, (4.5)
ql ql k=l+1
l dtkl qk
dim X = n :
s
!
E l E X k E (4.6)
(1) (1)k djl+1 djl+2 . . . djk = 0,
yj1 j2 ...jl yj1 j2 ...jl k=l+1 l yj1 j2 ...jk
for all , , and 0 l s. To a locally variational form a local Lagrangian is given by
formula (4.3), respectively, (4.4).
4.2. Lepage Equivalents of Locally Variational Forms

Now we are able to extend the concept of Lepage n-form to (n + 1)-forms as proposed in
[50, 53].
Theorem 4.2 shows that for a dynamical form be locally variational it is essential that it
can be completed (at least locally) to a closed form. This property was a motivation for the
following definitions:
Definition 4.4. Let s 0. A closed (n + 1)-form on J s Y is called a Lepage (n + 1)-
form if p1 is a dynamical form. A Lepage (n + 1)-form is called Lepage equivalent of
a dynamical form E if p1 = E.
With the help of the Poincare lemma, and taking account of the definition of a Lepage
n-form, we immediately obtain relations between Lepage (n + 1)-forms and n-forms:
Theorem 4.5.
(1) If is a Lepage (n + 1)-form then locally = d where is a Lepage n-form.
(2) If is a Lepage (n + 1)-form then the dynamical form E = p1 is locally varia-
tional.
(3) Every Lepage equivalent of a locally variational form E locally equals to d
where is a Lepage equivalent of a Lagrangian for E. Conversely, if is a Lepage
equivalent of a Lagrangian then d is a Lepage equivalent of the EulerLagrange form
E .
Note that we can say that Lepage (n + 1)-forms are closed counterparts of variational
equations.
With help of Lepage (n + 1)-forms we realise that the difference between local and
global variationality is the same as the difference between local and global exactness of
(n+1)-forms. In the following theorem we denote by H n+1 (Y ) the (n+1)-st cohomology
group of Y . We also note that due to the structure of the fibres of the bundle J s Y Y , the
cohomology groups of J s Y coincide with the corresponding cohomology groups of Y .
46 Olga Krupkova
Theorem 4.6.
(1) A locally variational form E is globally variational if and only if it has an exact
Lepage equivalent.
(2) If H n+1 (Y ) = {0} then every locally variational form on J s Y is globally varia-
tional.
The question of the existence of Lepage equivalents of locally variational forms for
general n and s is still open. However, it is known that if n 1 and the order of E is 2,
a Lepage equivalent, if exists, is nonunique [53, 54], and splits into the sum
= E + d, (4.7)
where E is determined uniquely by the components E of E.

On the other hand, for n = 1, s arbitrary, and n > 1, s = 1 every locally variational
form has a unique minimal-order Lepage equivalent [50] (cf. also [32]), and [29]: If n = 1,
s 1,
s1
= E dt +
X
jk
F j k , (4.8)
j,k=0
where
sjk1 !
jk 1 X
j+l j+l dl E
F = (1) , (4.9)
2 l=0
l dtl qj+k+l+1
whenever 0 j + k s 1, and F jk = 0 otherwise. Note that the Lepage equivalent of a
dynamical form of order s is projectable onto J s1 Y . If n > 1, s = 1, the unique minimal

order Lepage equivalent of E is defined on Y and reads
n
1 k E
= E 0 +
X
1 k j1 jk . (4.10)
k=1
k!(k + 1)! yj11 yjkk
We refer the reader to the article [27] by Grigore (in this volume), where the structure
of closed (n + 1)-forms that are counterparts of variational equations (on higher order
Grassmann bundles) is clarified.
4.3. Lepage (n + 1)-Forms in the Geometry of Variational Equations

Lepage (n + 1)-forms are useful not only in the study of the inverse variational problem.
They play an even more important role for investigations of the structure and symmetries
of variational equations, and are essential for the formulation of exact integration methods.
In this section we just mention very briefly a few important applications in the geometric
theory of differential equations. However, for more details we refer to [29, 51, 52, 53, 54,
55, 56] and references therein.
Let E be a locally variational form on J s Y (s 1). Given a Lepage equivalent of E
on J r Y , we define a Hamiltonian system of to be an exterior differential system on J r Y ,
locally generated by n-forms i , where runs over all r -vertical vector fields on J r Y .
Due to the nonuniqueness of we usually have many Hamiltonian systems associated

with a given locally variational form (EulerLagrange equations), however, all have the
same holonomic integral sections. Moreover, for any Lepage equivalent of E, holo-
nomic integral sections of the Hamiltonian system of coincide with (prolongations of)
extremals of E. Therefore equations for holonomic integral sections of a Hamiltonian sys-
tem, i.e.,
J r i = 0, r -vertical vector fields , (4.11)
are EulerLagrange equations, and equations for (all) its integral sections, i.e.,
i = 0, r -vertical vector fields , (4.12)
are called Hamilton equations. We stress that every Hamiltonian system of a locally varia-
tional form provides a complete family of extremals (global solutions of the EulerLagrange
equations) and a complete family of Hamilton extremals (global solutions of Hamilton equa-
tions (even if the locally variational form is not globally variational).
Now one can study EulerLagrange and Hamilton equations and their solutions from a
geometric point of view by investigating properties of the corresponding exterior differential
systems. Let us mention at least regularity (more generally classification of equations with
respect to geometric properties of their solutions), and symmetries and conservations laws,
including corresponding integration methods.
Note that the above geometric representation by means of exterior differential systems
demonstrates a deep difference between ordinary and partial differential equations:
For n = 1 the Lepage equivalent of E is unique, hence to ordinary variational equations
we have unique Hamilton equations, determined by E. This means that, in particular, we
get a Hamiltonian and momenta determined by the EulerLagrange expressions (not by an
individual Lagrangian). Moreover, the Hamiltonian EDS is locally generated by 1-forms,
this means that Hamilton extremals are integral sections of a distribution, called the Euler
Lagrange distribution, uniquely determined by the EulerLagrange form E, and defined
on J s1 Y , where s is the order of E. We obtain a natural concept of a regular variational
system: a locally variational form E is called regular if the rank of its EulerLagrange
distribution = 1. Hence, we get the regularity condition expressed by means of the Euler
Lagrange expressions.
For n > 1 Hamilton extremals are no longer described by a distribution, moreover, we
have a family of Hamilton equations associated with an EulerLagrange form. Hamilton
equations (Hamiltonians, momenta) are (similarly as in the case of dim X = 1) associated
directly to the EulerLagrange form, not to individual Lagrangians. They do, however,
depend upon a choice of a Lepage equivalent of E. In this case, regularity is understood
as a property of a Hamiltonian EDS, i.e., again it does not depend upon a choice of an
individual Lagrangian for E, but depends however upon a choice of for E. Thus, in field
theory, we can look for a regularisation of given variational equations: the aim is to find a
Hamiltonian system for E that is regular.
The EDS description also helps us look for symmetries and conservation laws for Euler
Lagrange and Hamilton equations. In particular, we can easily study symmetries of Lepage
n and (n + 1)-forms (i.e., vector fields along which the Lie derivative of , or vanishes).
48 Olga Krupkova
Moreover, every symmetry of a Lepage n-form (resp. (n + 1)-form ) generates a con-

served current (i.e. a closed n-form, belonging to the Hamiltonian EDS of d (resp. )).
For n = 1 this gives functions constant along extremals of E.
5. Lepage Forms in the Variational Sequence

Within the theory of varational sequences developed by Krupka in [44], the concept of
Lepage equivalent can be generalised to arbitrary k-forms, k n.
Denote by rq the sheaf q-forms on J r Y , r0,c = {0}, and rq,c the sheaf of contact
q-forms if q n, respectively strongly contact q-forms if q > n, on J r Y . Set
rq = rq,c + drq1,c
where drq1,c is the image sheaf of rq1,c by the exterior derivative d. There arises an
exact sequence of soft sheaves 0 r1 r2 r3 , where the morphisms are
the exterior derivative, called contact sequence. It is a subsequence of De Rham sequence
0 R r0 r1 r2 r3 . The quotient sequence
0 R r0 r1 /r1 r2 /r2 r3 /r3
which is also exact, is called the r-th order variational sequence on . It is important to
stress that elements of the quotient sheaf rq /rq are not forms, but classes of (local) q-
forms of order r. We denote by []rv an element of rq /rq , that is the (variational) class of
rq . The quotient mappings are denoted by
Eqr : rq /rq rq+1 /rq+1 .
As proved by Krupka in [44], the variational sequence is an acyclic resolution of the con-
stant sheaf R over Y . Due to the Abstract De Rham theorem, the cohomology groups of
the cochain complex of global sections of the variational sequence are identified with the
De Rham cohomology groups H q Y of Y .
Quotient sheaves rq /rq are determined up to natural isomorphisms of Abelian groups.
In this way classes in rq /rq admit representations by differential forms. Source forms [73]
for the quotient sheaf rq /rq arise by applying to q-forms the so-called interior Euler
Lagrange operator I [2, 35, 48]. Source forms for q = n and q = n + 1 are called
Lagrangians and dynamical forms, respectively. In the source forms representation, the
quotient mapping Enr : rn /rn rn+1 /rn+1 coincides with the Euler-Lagrange mapping
E : E . The next mapping is called Helmholtz mapping: if E is a dynamical form
representing a class in rn+1 /rn+1 then the image of E is an (n + 2)-form HE , called
the Helmholtz form of E [44] (components of the Helmhotz form are the left-hand-sides of
variationality conditions (4.6)).
The variational sequence helps us to learn the structure of the EulerLagrange mapping
and of null Lagrangians as follows: Condition HE = 0 for elements of the quotient sheaves
r ([]r ) = 0, and by exactness of the variational sequence means that there exists
reads En+1 v
a class []rv rn /rn such that []rv = [d]rv . The source form = h for []rv is then a
local Lagrangian for E. If H n+1 Y = {0}, may be chosen globally defined on J r Y , so
we get a global Lagrangian for E. Similarly, if E = 0, i.e., Enr ([]rv ) = 0, there exists
a class []rv rn1 /rn1 such that []rv = [d]rv , for source forms, = h = hd. If
H n (Y ) = {0} then every null Lagrangian is globally null (there is a defined on J r1 Y ).
The motivation for the generalisation of the concept of Lepage form is to obtain a
representation of classes in the variational sequence such that the sequence morphisms
would be the exterior derivatives [48, 57]: A q-form , q n, is called a Lepage form
if pqn+1 d = Id (i.e. is a source form). If is a source q-form, we say that is a
Lepage equivalent of if is a Lepage q-form and pqn = . Note that for q = n,
and q = n + 1 and locally variational, we get definitions introduced in the previous sec-
tions. Summarising, we can express the source form representation and the Lepage form
representation of the variational sequence as follows:
En1 E En+1 En+2

rn /rn n rn+1 /rn+1 rn+2 /rn+2

Iy Iy Iy
E H
sn 2s
n+1 4s
n+2

Lepy Lepy Lepy

d d d d
kn kn+1 kn+2
Generalisations of Lepage forms are studied also within the theory of non-holonomic
systems [60, 72].
Acknowledgments
Research supported by grants GACR 201/06/0922 of the Czech Science Foundation, and
MSM 6198959214 of the Czech Ministry of Education, Youth and Sports. The author also
highly appreciates support and hospitality of the Mathematics Department and the Institute
for Advanced Study at La Trobe University, Melbourne, Australia.
References
[1] V. Aldaya and J. de Azcarraga, Higher order Hamiltonian formalism in field theory,
J. Phys. A: Math. Gen. 13 (1982) 25452551.
[2] I. Anderson, The Variational Bicomplex (Utah State University, Technical Report,
1989).
[3] I. Anderson and T. Duchamp, On the existence of global variational principles, Am.
J. Math. 102 (1980) 781867.
[4] D. E. Betounes, Extension of the classical Cartan form, Phys. Rev. D 29 (1984) 599
606.
[5] E. Cartan, Lecons sur les invariants integraux (Hermann, Paris, 1922).
50 Olga Krupkova
[6] M. Crampin and D. J. Saunders, The Hilbert-Caratheodory and Poincare-Cartan

forms for higher-order multiple-integral variational problems, Houston J. Math. 30
(2004) 657689.
[7] P. Dedecker, Calcul des Variations et Topologie Algebrique, Thesis, Univ. de Liege,
Faculte des Sciences, 1957.
[8] P. Dedecker, On the generalization of symplectic geometry to multiple integrals in

the calculus of variations, In: Lecture Notes in Math. (570, Springer, Berlin, 1977)
395456.
[9] P. Dedecker, Le theoreme de Helmholtz-Cartan pour une integrale simple dordre

superieur, C. R. Acad. Sci. Paris, Ser. A 288 (1979) 827830.
[10] P. Dedecker, Sur le formalisme de HamiltonJacobiE. Cartan pour une integrale

multiple dordre superieur, C. R. Acad. Sci. Paris, Ser. I 299 (1984) 363366.
[11] Th. De Donder, Theorie Invariantive du Calcul des Variations (GauthierVillars,

Paris, 1930).
[12] M. de Leon and P. R. Rodrigues, Generalized Classical Mechanics and Field Theory
(North-Holland, Amsterdam, 1985).
[13] A. Echeverria-Enriquez, M. C. Munoz-Lecanda and N. Roman-Roy, Geometry of

multisymplectic Hamiltonian first-order field theories, J. Math. Phys. 41 (2000)
74027444.
[14] Ch. Ehresmann, Les prolongements dune space fibre differentiable, C. R. Acad. Sci.
Paris 240 (1955) 17551757.
[15] M. Ferraris, Fibered connections and global PoincareCartan forms in higher order
calculus of variations, In: Geometrical Methods in Physics (Proc. Conf. on Diff.
Geom. and Appl. Vol. 2, Nove Mesto na Morave, Sept. 1983, (D. Krupka, Ed.) J. E.
Purkyne Univ. Brno, Czechoslovakia, 1984) 6191.
[16] M. Ferraris and M. Francaviglia, On the global structure of the Lagrangian and
Hamiltonian formalisms in higher order calculus of variations, In: Geometry and
Physics (Proc. Int. Meeting, Florence, Italy 1982, (M. Modugno, Ed.) Pitagora,
Bologna, 1983) 4370.
[17] P. L. Garcia, The Poincare-Cartan Invariant in the Calculus of Variations, Symp.

Math. XIV (1974) 219246.
[18] P. L. Garcia and J. Munoz, On the geometrical structure of higher order variational
calculus, In: Modern Developments in Analytical Mechanics I: Geometrical Dynam-
ics (Proc. IUTAM-ISIMM Symposium, Torino, Italy 1982, (S. Benenti, M. Fran-
caviglia and A. Lichnerowicz, Eds.) Accad. delle Scienze di Torino, Torino, 1983)
127147.
[19] P. L. Garcia and J. Munoz Masque, Le probleme de la regularite dans le calcul des
variations du second ordre, C. R. Acad. Sci. Paris, Ser. I 301 (1985) 639642.
[20] P. L. Garcia Perez and J. Munoz Masque, Higher order regular variational
problems, In: Proc. Internat. Colloquium Geometrie Symplectique et Physique
Mathematique (Aix-en-Provence 1990, (P. Donato, C. Duval, J. Elhadad and G.
M. Tuynman, Eds.) Prog. Math. 99, Birkhauser, Boston, 1991) 136159.
[21] G. Giachetta, L. Mangiarotti and G. Sardanashvily, New Lagrangian and Hamilto-

nian Methods in Field Theory (World Scientific, Singapore, 1997).
[22] H. Goldschmidt and S. Sternberg, The Hamilton-Cartan formalism in the calculus of

variations, Ann. Inst. Fourier 23 (1973) 203267.
[23] M. J. Gotay, An exterior differential systems approach to the Cartan form, In: Proc.
Internat. Colloquium Geometrie Symplectique et Physique Mathematique (Aix-
en-Provence 1990, (P. Donato, C. Duval, J. Elhadad and G. M. Tuynman, Eds.) Prog.
Math. 99, Birkhauser, Boston, 1991) 160188.
[24] M. J. Gotay, A multisymplectic framework for classical field theory and the calculus
of variations, I. Covariant Hamiltonian formalism, In: Mechanics, Analysis and Ge-
ometry: 200 Years After Lagrange ((M. Francaviglia and D. D. Holm, Eds.) North
Holland, Amsterdam, 1990) 203235.
[25] D. R. Grigore, Higher-order Lagrangian theories and Noetherian symmetries, Roma-

nian Journ. Phys. 39 (1995) 11-35; arXiv:hep-th/9305009.
[26] D. R. Grigore, Variationally trivial Lagrangians and locally variational differential

equations of arbitrary order, Diff. Geom. Appl. 10 (1999) 79105.
[27] D. R. Grigore, On a Generalization of the Poincare-Cartan Form in Higher-Order

Field Theory, In: Variations, Geometry and Physics ((O. Krupkova and D. Saunders,
Eds.) Nova Science Publishers, 2008) 5776.
[28] D. R. Grigore and O. T. Popp, On the LagrangeSouriau form in classical field the-
ory, Math. Bohemica 123 (1998) 7386.
[29] A. Hakova and O. Krupkova, Variational first-order partial differential equations, J.

Differential Equations 191 (2003) 6789.
[30] H. Helmholtz, Ueber die physikalische Bedeutung des Prinzips der kleinsten
Wirkung, J. fur die reine u. angewandte Math. 100 (1887) 137166.
[31] M. Horak and I. Kolar, On the higher order PoincareCartan forms, Czechoslovak
Math. J. 33 (108) (1983) 467475.
[32] L. Klapka, EulerLagrange expressions and closed twoforms in higher order me-
chanics, In: Geometrical Methods in Physics (Proc. Conf. on Diff. Geom. and Appl.
Vol. 2, Nove Mesto na Morave, Sept. 1983, (D. Krupka, Ed.) J. E. Purkyne Univ.
Brno, Czechoslovakia, 1984) 149153.
52 Olga Krupkova
[33] I. Kolar, Some geometric aspects of the higher order variational calculus, In: Geo-
metrical Methods in Physics (Proc. Conf. on Diff. Geom. and Appl. Vol. 2, Nove
Mesto na Morave, Sept. 1983, (D. Krupka, Ed.) J. E. Purkyne University, Brno,
Czechoslovakia, 1984) 155166.
[34] I. Kolar, A geometric version of the higher order Hamilton formalism in fibered
manifolds, J. Geom. Phys. 1 (1984) 127137.
[35] M. Krbek and J. Musilova, Representation of the variational sequence, Rep. Math.
Phys. 51 (2003) 251258.
[36] D. Krupka, Some geometric aspects of variational problems in fibered manifolds,

Folia Fac. Sci. Nat. Univ. Purk. Brunensis, Physica 14, Brno, Czechoslovakia (1973)
pp. 65; arXiv:math-ph/0110005.
[37] D. Krupka, A geometric theory of ordinary first order variational problems in fibered
manifolds. I. Critical sections, J. Math. Anal. Appl. 49 (1975) 180206.
manifolds. II. Invariance, J. Math. Anal. Appl. 49 (1975) 469476.
[39] D. Krupka, A map associated to the Lepagean forms of the calculus of variations in
fibered manifolds, Czechoslovak Math. J. 27 (1977) 114118.
[40] D. Krupka, On the local structure of the Euler-Lagrange mapping of the calculus of
variations, In: Proc. Conf. on Diff. Geom. Appl. (Nove Mesto na Morave, Czechoslo-
vakia, 1980, Charles Univ. Prague, 1982) 181188; arXiv:math-ph/0203034.
[41] D. Krupka, Lepagean forms in higher order variational theory, In: Modern Devel-
opments in Analytical Mechanics I: Geometrical Dynamics (Proc. IUTAM-ISIMM
Symposium, Torino, Italy 1982, (S. Benenti, M. Francaviglia and A. Lichnerowicz,
Eds.) Accad. Sci. Torino, Torino, 1983) 197238.
[42] D. Krupka, On the higher order Hamilton theory in fibered spaces, In: Geometrical
Methods in Physics (Proc. Conf. Diff. Geom. Appl., Nove Mesto na Morave, 1983,
(D. Krupka, Ed.) J.E. Purkyne University, Brno, Czechoslovakia, 1984) 167183.
[43] D. Krupka, Regular Lagrangians and Lepagean forms, In: Differential Geometry and
Its Applications (Proc. Conf., Brno, Czechoslovakia, 1986, (D. Krupka and A. Svec,
Eds.) D. Reidel, Dordrecht, 1986) 111148.
[44] D. Krupka, Variational sequences on finite order jet spaces, In: Differential Geometry
and Its Applications (Proc. Conf., Brno, Czechoslovakia, 1989, (J. Janyska and D.
Krupka, Eds.) World Scientific, Singapore, 1990) 236254.
[45] D. Krupka, Global variational theory in fibred spaces, In: Handbook of Global Anal-
ysis (Elsevier, 2008) 755839.
[46] D. Krupka and J. Musilova, Hamilton extremals in higher order mechanics, Arch.
Math. (Brno) 20 (1984) 2130.
[47] D. Krupka and J. Musilova, Trivial Lagrangians in field theory, Diff. Geom. Appl. 9
(1998) 293305.
[48] D. Krupka and J. Sedenkova, Variational sequences and Lepagean forms, In: Dif-
ferential Geometry and its Applications (Proc. Conf., Prague, 2004, (J. Bures, O.
Kowalski, D. Krupka and J. Slovak, Eds.) Charles Univ., Prague, Czech Republic,
2005) 605615.
[49] D. Krupka and O. Stepankova, On the Hamilton form in second order calculus of
variations, In: Geometry and Physics (Proc. Internat. Meeting, Florence, 1982, (M.
Modugno, Ed.) Pitagora, Bologna, 1983) 85102.
[50] O. Krupkova, Lepagean 2-forms in higher order Hamiltonian mechanics, I. Regular-
ity, Arch. Math. (Brno) 22 (1986) 97120.
Mathematics 1678, Springer, Berlin, 1997).
[52] O. Krupkova, Differential systems in higher-order mechanics, In: Proceedings of
the Seminar on Differential Geometry ((D. Krupka, Ed.) Mathematical Publications,
Vol. 2, Silesian University in Opava, Opava, 2000) 87130.
[53] O. Krupkova, Hamiltonian field theory revisited: A geometric approach to regularity,
In: Steps in Differential Geometry (Proc. Colloq. Diff. Geom., Debrecen, July 2000,
(L. Kozma, P. T. Nagy and L. Tamassy, Eds.) University of Debrecen, Debrecen,
2001) 187207.
[54] O. Krupkova, Hamiltonian field theory, J. Geom. Phys. 43 (2002) 93132.
[55] O. Krupkova, Variational equations on manifolds, In: Theory and Applications of
Differential Equations (Nova Science Publ., pp. 88, in press).
[56] O. Krupkova and G. E. Prince, Second Order Ordinary Differential Equations in Jet
Bundles and the Inverse Problem of the Calculus of Variations, In: Handbook of
Global Analysis (Elsevier, 2008) 841908.
[57] O. Krupkova and G. E. Prince, Lepage forms, closed two-forms and second-order
ordinary differential equations; paper in honour of N.I. Lobachevskii, Russian Math-
ematics (Iz VUZ), 51 (12) (2007) N.I. Lobachevskii Anniversary Volume, 116; DOI
10.3103/S1066369X07120018.
[58] O. Krupkova and D. Smetanova, On regularization of variational problems in first-
order field theory, In: Proc. of the 20th Winter School Geometry and Physics (Srn,
2000, Rend. Circ. Mat. Palermo (2) Suppl. No. 66, 2001) 133140.
[59] O. Krupkova and D. Smetanova, Legendre transformation for regularizable La-
grangians in field theory, Letters in Math. Phys. 58 (2001) 189204.
[60] O. Krupkova, J. Volna and P. Volny, Constrained Lepage forms In: Proc. 10th Int.
Conf. on Diff. Geom. and Appl. (Olomouc 2007, (O. Kowalski, D. Krupka, O. Krup-
kova and J. Slovak, Eds.) World Scientific, Singapore, 2008) 627633.
54 Olga Krupkova
[61] Th. Lepage, Sur les champs geodesiques du Calcul des Variations, Bull. Acad. Roy.
Belg., Cl. des Sciences 22 (1936) 716729.
[62] J. E. Marsden, S. Pekarsky, S. Shkoller and M. West, Variational methods, multisym-

plectic geometry and continuum mechanics, J. Geom. Phys. 38 (2001) 253284.
[63] M. Marvan, On global Lepagean equivalents, In: Geometrical Methods in Physics

(Proc. Conf. on Diff. Geom. and Appl. Vol. 2, Nove Mesto na Morave, Czechoslo-
vakia, 1983, (D. Krupka, Ed.) J. E. Purkyne University, Brno, 1984) 185190.
[64] E. Noether, Invariante Variationsprobleme, Nachr. kgl. Ges. Wiss. Gottingen, Math.
Phys. Kl. (1918) 235257.
[65] P. J. Olver, Equivalence and the Cartan form, Acta Appl. Math. 31 (1993) 99136.
[66] D. J. Saunders, The Geometry of Jet Bundles (London Math. Soc. Lecture Notes
Series 142, Cambridge Univ. Press, Cambridge, 1989).
[67] D. J. Saunders, The regularity of variational problems, Contemporary Math. 132

(1992) 573593.
[68] D. J. Saunders, The Cartan form, 20 years on, In: Proc. 10th Int. Conf. on Diff.
Geom. and Appl. (Olomouc 2007, (O. Kowalski, D. Krupka, O. Krupkova and J.
Slovak, Eds.) World Scientific, Singapore, 2008) 527537.
[69] W. F. Shadwick, The Hamiltonian formulation of regular r-th order Lagrangian field
theories, Letters in Math. Phys. 6 (1982) 409416.
[70] D. Smetanova, On Hamilton p2 -equations in second-order field theory, In: Steps

in Differential Geometry (Proc. of Colloquium on Differential Geometry, Debrecen
2000, University of Debrecen, Debrecen 2001) 329341.
[71] D. Smetanova, On regularization of second order Lagrangians, In: Global Anal-

ysis and Applied Mathematics (Proc. International Workshop on Global Analysis,
Ankara, 2004, American Institute of Physics Proc. 729, 2004) 289296.
[72] M. Swaczyna, On the nonholonomic variational principle, In: Global Analysis and
Applied Mathematics (International Workshop on Global Analysis, Ankara, Turkey,
15-17 April, 2004, AIP Conference Proceedings, Melville, New York 2004) 297
306.
[73] F. Takens, A global version of the inverse problem of the calculus of variations, J.
Diff. Geom. 14 (1979) 543562.
[74] E. Tonti, Variational formulation of nonlinear differential equations I, II, Bull. Acad.
Roy. Belg. Cl. Sci. 55 (1969) 137165, 262278.
[75] M. M. Vainberg, Variational methods in the theory of nonlinear operators (GITL,

Moscow, 1959, in Russian).
[76] A. L. Vanderbauwhede, Potential operators and variational principles, Hadronic J. 2

(1979) 620641.
[77] A. M. Vinogradov, A spectral sequence associated with a non-linear differential

equation, and algebro-geometric foundations of Lagrangian field theory with con-
straints, Soviet Math. Dokl. 19 (1978) 144148.
[78] E. T. Whittaker, A Treatise on Analytical Dynamics of Particles and Rigid Bodies

(The University Press, Cambridge, 1917).
Chapter 3
O N A G ENERALIZATION
OF THE P OINCAR E -C ARTAN F ORM
IN H IGHER -O RDER F IELD T HEORY
D.R. Grigore
Dept. of Theor. Phys., National Inst. Phys.
Nucl. Engeneering Horia Hulubei,
Bucharest-Magurele, P. O. Box MG 6, Romania
Abstract
We present here a possible generalization of the Poincare-Cartan form in classical
field theory to the most general case: arbitrary dimension, arbitrary order of the theory
and the absence of a fiber bundle structure. We use for the kinematical description
of the system the (r, n)-Grassmann manifold associated to a given manifold X, i.e.
the manifold of r-contact elements of n-dimensional submanifolds of X. The idea is
to define globally a n + 1 form on this Grassmann manifold, more precisely its class
with respect to a certain subspace and to write it locally as the exterior derivative of
a n form which is a kind of Poincare-Cartan form in the higher-order and non-fibred
situation.
2000 Mathematics Subject Classification. 53A55, 77S25, 58A20.

Key words and phrases. Grassmann bundles, Lagrangian Formalism.
1. Introduction
It is widely accepted that the variational principles should be given in a coordinate indepen-
dent formulation. This idea was first realized for a dynamical system with a finite number
of degrees of freedom (i.e. particle mechanics), using a differential 1-form instead of the
Lagrangian, by Poincare and Cartan [43], [8]. There are a number of generalizations of this
idea for classical field theory [27], [5], [6], [44], [45], [24], [11], [10], [12], [46], [42]. A re-
lated concept is that of Lepage equivalent of a Lagrangian form (see for instance [28], [29],

E-mail address: grigore@theory.nipne.ro, grigore@ifin.nipne.ro
58 D.R. Grigore
[32]). All these generalizations use as geometric framework for classical field theory the jet
bundle formalism (more explicitly the space-time and field variables are local coordinates
on a fiber bundle X over a space-time manifold M ) and the derivative of the fields, up to
order s, are variables in the s-th order jet bundle extension J s X of X.
It was later [26], [47] suggested that it is more convenient to work with the exterior
differential of the above Poincare-Cartan form. For the case of finite number of degrees
of freedom this 2-form is in general presymplectic and was used by Souriau and others
[47], [23] to obtain the phase space as a symplectic manifold in a deductive way. The idea
is to consider that the fundamental mathematical object for a Lagrangian system must be
this 2-form and not the Lagrangian function or the Poincare-Cartan 1-form. This point of
view leads to the main features of the Lagrangian and the Hamiltonian formalism and to a
natural definition of the Noetherian symmetries. For higher-order mechanics this approach
has been developed in [33], [34], [35] (see also [38]).
One can generalize this Lagrange-Souriau form without using the fibration hypothesis
mentioned above in two particular but important cases: for classical field theory of first
order [13] and for systems with a finite number of degrees of freedom and of arbitrary order
[14]. Moreover, this Lagrange-Souriau form can be locally written as the exterior differ-
ential of a Poincare-Cartan form related to some chosen local chart. This Poincare-Cartan
form is the same as that given by Krupka [27], Betounes [5]-[6] and Rund [44]. The La-
grangian is locally determined up to a variationally trivial Lagrangian, i.e. a Lagrangian
giving trivial Euler-Lagrange equations. As a consequence, one can define in a geometri-
cally nice way the Noetherian symmetries using the Lagrange-Souriau form.
In this paper we give a generalization of the Lagrange-Souriau and of the Poincare-
Cartan forms in the most general case used in higher order field theory. We consider an
arbitrary manifold X without a fiber bundle structure over some space-time manifold so
instead of the s-th order jet bundle extension one must use the s-th order Grassmann bundle
Pns X associated to X which was recently considered in the literature [20]. In the next
Section we will summarize the main features of this construction. Next, in Section 3, we
will be able to define globally a n + 1 differential form, but one will be able to see that,
in general, one cannot determine this form uniquely. Fortunately one can consider the
equivalence class of this form to a certain globally defined subspace of differential forms.
This equivalence class is the physical object we are looking for. It is interesting to note
that this subspace of differential forms is in fact identically zero exactly in the two particular
cases mentioned above (s = 2, n arbitrary and n = 1, s arbitrary). Some combinatorial
tricks introduced in [16] must be used to simplify the analysis of some tensorial identities.
In Section 4 we locally exhibit the n + 1 differential form as the exterior derivative of
a locally defined Poincare-Cartan n-form and in this way the (local) Lagrangian function
appears also. Some functions called hyper-Jacobians [7], [41] emerge naturally in this
context and can be used to provide the most general expression for a variationally trivial
Lagrangian of arbitrary order already obtained in [17] by a different method. Finally, in
Section 5 we present two particular but very important cases, namely n = 1 s arbitrary and
s = 2 n arbitrary. Some ideas related to the ones from this paper also appear in [31]. In the
fibered situation, similar (n + 1)-forms were studied in [39], [40] and [22].
We will skip some proofs which are similar to the proofs from [16]-[19] and can be
found in the web version of this paper [15].
On a Generalization of the Poincare-Cartan Form 59
2. Grassmann Manifolds
2.1. The Basic Constructions of the Grassmann Manifolds
In this Section we present the basic construction of Grassmannian manifolds following [20]
and [18]. We will skip all the proofs. We consider N , n 1 and r 0 integers such that
n N , and let X be a smooth manifold of dimension N which is the mathematical model
for the kinematical degrees of freedom of a certain classical field theory.
Let U Rn be a neighborhood of the point 0 Rn , x X and let (0,x) be the set
of smooth maps : U X such that (0) = x. On (0,x) one has the the equivalence
relationship iff there exists a chart (V, ) = (xA ), A = 1, . . . , N on X
such that the functions , : Rn RN have the same partial derivatives up to
order r in the point 0. The equivalence class of will be denoted by j0r and it is called a
r
(r, n)-velocity. The set of (r, n)-velocities at x is denoted by T(0,x) (Rn , Y ) (0,x) / .
We denote
Tnr X =
[
r
T(0,x) (Rn , X),
xX
and define surjective mappings nr,s : Tnr X Tns X, where 0 < s r, by nr,s (j0r ) = j0s
and nr,0 : Tnr X X, where 1 r, by nr,0 (j0r ) = (0).
If (V, ), = (xA ), is a chart on X we define the couple (Vnr , nr ) where Vnr =
(n )1 (V ), nr = (xA , xA
r,0 A
j , , xj1 ,j2 ,...,jr ) 1 j1 j2 jr n, and

k
xA r
j1 ,...,jk (j0 ) j1 xA
, 0 k r. (2.1)

t . . . tjk
0
The expressions xA r
j1 ,jk (j0 ) are defined for all indices j1 , . . . , jr in the set {1, . . . , n}
but because of the symmetry property
xA r A r
jP (1) ,...,jP (k) (j0 ) = xj1 ,...,jk (j0 ) (k = 2, ..., n) (2.2)
for all permutations P Pk of the numbers 1, . . . , k we consider only the independent

components given by the restrictions 1 j1 j2 jr n. This allows one to use
multi-index notations i.e. nr = (xA
J ), |J| = 0, ..., r where by definition xA A
x . The
same comment is true for the partial derivatives xA .
j1 ,...,jk
The couple (Vnr , nr ) is a chart on Tnr X called the associated chart of the chart (V, )
and the system of charts give a smooth structure on this set; moreover Tnr X is a fiber bundle
over X with the canonical projection nr,0 . The set Tnr Y endowed with the smooth structure
defined by the associated charts defined above is called the manifold of (r, n)-velocities
over X.
In the chart (Vnr , nr ) one introduces the following differential operators:
r1 ! . . . rn !
jA1 ,...,jk , j1 , . . . , jk {1, . . . , n} (2.3)
k! xA
j1 ,...,jk
where rk is the number of times the index k shows up in the sequence j1 , . . . jk .

60 D.R. Grigore
The combinatorial factors are such that the following relation is true:
BS+ i1 ik
(
A j1 ,...,jk j1 . . . jk if k = l
iA1 ,...,ik xB
j1 ,...,jl = (2.4)
0 if k 6= l.
Here we use the notations from [14], namely Sj1 ,...,jk are the symmetrization (for the
sign +) and respectively the antisymmetrization (for the sign ) projector operators defined
by
1 X
Sj1 ,...,jk fj1 ,...,jk (P )fjP (1) ,...,jP (k) (2.5)
k! P P
k
where the sum runs over the permutation group Pk of the numbers 1, . . . , k and
+ (P ) 1, (P ) (1)|P | , P Pk ;
here |P | is the signature of the permutation P .

In this way one takes care of overcounting the indices. More precisely, for any smooth
function on V r , the following formula is true:
r
(jA1 ,...,jk f )dxA
X X
df = j1 ,...,jk = (IA f )dxA
I (2.6)
k=0 |I|r
where we have also used the convenient multi-index notation.

The formal derivatives are:
r1
j1 ,...,jk
Dir
X
xA
X
i,j1 ,...,jk A = xA J
iJ A . (2.7)
k=0 |J|r1
The last expression uses the multi-index notation; if I and J are two such multi-indices
we mean by IJ the juxtaposition of the two sets I, J.
When no danger of confusion exists we simplify the notation putting simply Di = Dir .
The formal derivatives give a conveniently expression for the change of charts on the
velocity manifold induced by a change of charts on X.
By definition the differential group of order r is the set
Lrn {j0r J0,0

r
(Rn , Rn )| Dif f (Rn )} (2.8)
i.e. the group of invertible r-jets with source and target at 0 Rn . The group multiplication
in Lrn is defined by the jet composition Lrn Lrn (j0r , j0r ) 7 j0r ( ) Lrn .
The canonical (global) coordinates on Lrn are defined by

k i
aij1 ,...,jk (j0r ) = j1 , j1 j2 jk , k = 0, ..., r (2.9)

t . . . tjk 0
where i are the components of a representative of j0r .

We denote
a (aij , aij1 ,j2 , . . . , aij1 ,...,jk ) = (aiJ )1|J|r
and notice that one has

det(aij ) 6= 0. (2.10)
The group Lrn is a Lie group. The manifolds of (r, n)-velocities Tnr Y admits a (natural)
smooth right action of the differential group Lrn , defined by the jet composition
(x a)A A r
I xI (j0 ( )) (2.11)
where the connection between xA I and is given by (2.1) and the connection between aI
i
and is given by (2.9).

The group Lrn has a natural smooth left action on the set of smooth real functions defined
on Tnr X , namely for any such function f we have:
(a f )(x) f (x a). (2.12)
We say that a (r, n)-velocity j0r Tnr X is regular, if (or any other representative) is
an immersion. We have the central result:
Theorem 2.1. The set Pnr X ImmTnr X/Lrn has a unique differential manifold struc-
ture such that the canonical projection rn is a submersion. The group action defines on
ImmTnr X the structure of a right principal Lrn -bundle.
A point of Pnr X containing a regular (r, n)-velocity j0r is called an (r, n)-contact
element, or an r-contact element of an n-dimensional submanifold of X, and is denoted by
[j0r ]. As in the case of r-jets, the point 0 Rn (resp. (0) X) is called the source (resp.
the target) of [j0r ]. The manifold Pnr is called the (r, n)-Grassmannian bundle, or simply
a higher order Grassmannian bundle over X.
Besides the quotient projection rn : ImmTnr X Pnr we have for every 1 s r, the
canonical projection of Pnr X onto Pns X defined by r,s r s
n ([j0 ]) = [j0 ] and the canonical
r r r
projection of Pn X onto X defined by n ([j0 ]) = (0).
On Pnr X there are total differential operators; as expected, in the chart rn (W I,r ) they
have the expression:
r1 ! . . . rn !
j1 ,...jk (2.13)
k! yj1 ,...,jk
We note for further use the following formula:
i1 ,...,ik yj1 ,...,jk = Sj+1 ,...,jk Qij11 . . . Qijkk Q , k = 1, ..., r; (2.14)
here we have defined:

Q y yi ( xi ) (2.15)
and Q is the inverse of the matrix:
Pji dj xi , Pji Qjl = li . (2.16)
Next we define the total derivative operators on the Grassmann manifold:

r1
X

j1 ,...,jk =
X
J
di i
+ yi,j1 ,...,jk i
+ yiJ . (2.17)
x k=0
x |J|r1
62 D.R. Grigore
We note that:
(rn ) (zij Dj ) = di . (2.18)
In particular, we have for any smooth function f on rn (W r ) the following formula:
Di (f rn ) = xji (dj f ) rn . (2.19)
The formula for the chart change on Pnr X. can be written with this operators: let us
consider two overlapping charts: (rn (V r ), (xi , yI )) and respectively (rn (Vr ), (xi , yI ));
then we have on the overlap:

yiI = Qji dj yI . |I| r 1 (2.20)
We also note that:

Pij dj = di . (2.21)
2.2. Contact Forms on Grassmann Manifolds

By a contact form on Pnr X we mean any form rq (P X) verifying
[j r ] = 0 (2.22)
for any immersion : Rn X. We denote by rq(c) (P X) the set of contact forms of

degree q n. Here [j r ] : Rn Pnr is given by: [j r ] (t) [jtr ] . We mention some of
properties verified by these forms.
If one considers only the contact forms on an open set rn (V r ) Pnr X then we em-
phasize this by writing rq(c) (V ). The ideal of all contact forms is denoted by C(r ). By
elementary computations one finds out that, as in the case of a fiber bundle, for any chart
(V, ) on X, every element of the set r1(c) (V ) is a linear combination of the following
expressions:
j1 ,...,jk dyj1 ,...,jk yi,j

1 ,...,jk
dxi , k = 0, ..., r 1 (2.23)
or, in multi-index notations:
J dyJ yiJ

dxi , |J| r 1. (2.24)
We have the formula
dJ = Ji

dxi , |J| r 2. (2.25)
Any form rq (P X), q = 2, ..., n is contact iff it is generated by J , |J| r1

and dI , |I| = r 1. In the end we present the transformation formula relevant for
change of charts.
Proposition 2.2. Let (V, ) and (V , ) two overlapping charts on X and let (W r , r ),
r = (xi , yI , xiI ) and (W r , r ), r = (xi , yI , xiI ) the corresponding charts on Tnr X.
Then the following formula is true on rn (W r W r ) Pnr X:
|I|
I =
X
(J yI )J QI, , 1 |I| r 1. (2.26)
|J|=1
where we have defined:
QI, yI yjI

( xj ), 0 |I| r 1 (2.27)
and
= Q (2.28)
where Q is given by the formula (2.15).
As a consequence we have:
Corollary 2.3. If for a q-form has the expression

X X X
=
p+s=k |J1 |,...,|Jp |r1 |I1 |=...=|Is |=r1

J11 Jpp dI11 dIss J11,...,Jp ,I1 ,...,Is
,...,p ,1 ,...,s , k q (2.29)
is valid in one chart, then it is valid in any other chart.
This corollary allows us to define for any q = 1, dots, dim(J r Y ) = m n+r

n a contact
form with order of contactness k to be any rq such that it has in one chart (thereafter
in any other chart) the expression above. We denote these forms by rq,k .
3. A Lagrange-Souriau Form on a Grassmann Manifold

3.1. Some Invariant Conditions
As in the preceding Section we consider a differential manifold X and the associated (s, n)-
Grassmann manifold Pns X. We start we the following general result:
Proposition 3.1. Let sq (X) a q-differential form on Pns X verifying:
i = 0 (3.1)
for any s,s1

n -vertical vector field on Pns X (i.e. (s,s1
n ) = 0). Then this form has the
local expression:
q
TI11,...,
,...,Ik
1 Ikk dxik+1 dxiq
X X
= k ,ik+1 ,...,iq I1
(3.2)
k=0 |I1 |,...,|Ik |s1
where TI11,...,,...,Ik
k ,ik+1 ,...,iq
are smooth functions depending on the variables
i
(x , y , yj , . . . , yj1 ,...,js ) and are antisymmetric in the couples (Ip , p ) p = 1, . . . k
and in the indices ik+1 , . . . , iq .
We denote the space of these forms by sq, (X). By elementary computations

from (2.26) and (2.28) we can obtain the transformation formulas for the coefficients
TI11,...,
,...,Ik
k ,ik+1 ,...,iq
in the overlap of two charts. As a consequence of these transformation
64 D.R. Grigore
formulas we can prove that some constraints on the coefficients TI11,...,

,...,Ik
k ,ik+1 ,...,iq
of the
form are in fact globally defined. These relations are:
TI11,...,
,...,Ik
k ,ik+1 ,...,iq
= 0, |I1 | + + |Ik | t t N, (3.3)
TI11,i2 ,...,iq = 0, I1 6= . (3.4)

Ti1 ,...,iq = 0. (3.5)
and the tracelessness condition
TlI11,...,
,...,Ik
k ,l,ik+2 ,...,iq
=0 k = 1, . . . , q. (3.6)
Some notations will be usefull. We denote the subset of the forms verifying (3.3) by
s,t
q, (X). If the condition (3.4) is fulfilled we say that the form verifies the Lepage
condition. We denote the subset of the forms verifying the conditions (3.3) and (3.4) by
s,t,Lep
q, (X) and
s,t s s,t
q,,k (X) q,k (X) q, (X);
these are contact forms with the order of contactness equal to k and the definition is globally
true. We call the subset of the forms verifying the conditions (3.3), (3.4) (i.e Lepage) and
(3.6) (i.e. tracelessness) by s,t,Lep
q,,tr (X). We finally stress again that all the spaces of the
type ...
... are globally defined.
3.2. The Definition of the Lagrange-Souriau Form

We need to consider one more condition on the Lepage forms, namely closeness. First we
have:
Proposition 3.2. Let s,s1 s,s1 -projectable, i.e. there
q, (X) be closed. Then is n
exists a q-form 0 s1
q (X) such that
= (s,s1
n ) 0 . (3.7)
Moreover, the form 0 is closed.
Proof. We exhibit the dependence of the form on the highest-order derivatives; according
to the preceding corollary these derivatives can appear in two places: in the coefficients
T,...,
1 ,...,k ,ik+1 ,...,iq
, k = 0, . . . , q and in the contact forms I , |I| = s 1. It is not very
hard to write now as follows:
q
T,...,
X
= 1 ,...,k ,ik+1 ,...,iq
1 k dxik+1 dxiq +
k=0
q
kTI11,...,
,,...,
1 2 k dxik+1 dxiq +
X X
k ,ik+1 ,...,iq I1
k=1 |I1 |=s1
where the form is s,s1

n -projectable.
The expression above is, as said before, at most linear in the highest-order derivatives.
One computes explicitly the coefficient of this derivatives and finds out that they are zero.
This proves the first assertion. The closeness of the form 0 follows from the surjectivity
of the map s,s1
n .
s,s1,Lep
It is clear that there are strong conditions on any closed form q,,tr (X). We
will give a structure theorem for such a form in the case q = n + 1 which is relevant for
physical applications. First we note that in this case we have

T,i1 ,...,in
= T i1 ,...,in (3.8)
for some smooth functions T on Pns X. Here i1 ,...,in is the completely antisymmetric ten-
sor. Then the general structure formula is:
Theorem 3.3. Let s,s,Lep
n+1,,tr (X) be closed. Then admits the following decomposi-
tion:
= T0 + dT1 (3.9)
where:
- T1 s,s2 s,s1
n,,2 (X) and dT1 n+1,,2 (X).
- T0 s,s1,1
n+1, (X) has the local structure given by the formula (3.2) with the tensors
Tk given by formulas of the type
Tk = Pk T , k = 2, . . . , n + 1 (3.10)
with Pk some linear differential operators which can be recursively determined.
The proof relies heavily on induction on k and uses some creation and annihilation
operators introduced in [16] and [17]. As a corollary of this formula we can make now the
connection with the Lagrangian formalism. Namely, we have:
Corollary 3.4. The expressions T defined according to (3.8) verify the generalized Hel-
mholtz equations.
Proof. We write explicitly the closeness condition and select only those equations contain-

ing the expressions T,i1 ,...,in
. As a result one obtains the following set of equations
!
|I| |J| |J| + |I|
I 1 T2
X
= (1) (1) dJ IJ2 T1 (3.11)
|J|s|I|
|J|
for |I| = s, |I| = 1, . . . , s1 and |I| = 0 respectively. But (3.11) are exactly the Helmholtz
equations (see [1], [3], [4], [9], [30]).
From the theorem above we also obtain:

Proposition 3.5. The decomposition (3.9) proved in the preceding theorem determines in
an unique way the form T0 .
Proof. It is sufficient to prove that
= T0 + dT1 = 0 = T0 = 0.
Indeed, from = 0 we have in particular T,i1 ,...,in = 0 and it follows from the formula
(3.10) that we have T0 = 0.
66 D.R. Grigore
As a corollary, let us denote by [] the equivalence class of the form s,s,Lep

n+1,,tr (X)
s,s2 s,s1
modulo dn,,2 (X) n+1,,2 (X). Then we have
[] = 0 T = 0 (3.12)
which says that the class of is uniquely determined by the so-called Euler-Lagrange
components of : T , = 1, . . . , m N n.
We call the globally defined class [] of a certain form s,s,Lep
n+1,,tr (X) a Lagrange-
Souriau class. We note in closing this Section that there are two particular but important
cases when the class of the form is formed only from the form ; obviously this happens
when
ds,s2 s,s1
n,,2 (X) n+1,,2 (X) = 0. (3.13)
One can see that if s = 2, and n arbitrary, or n = 1, and s arbitrary, the equality above
becomes an identity. So in this cases one can speak of a globally defined Lagrange-Souriau
form as in [13], [21] and [14] respectively.
4. The Associated Poincare-Cartan n-Form

4.1. The General Construction
Because the Euler-Lagrange expressions T are at most linear in the higher-order derivatives
h i
yi1 ,...,is it is to be expected that they follow from a Lagrangian of minimal order r s+1
2 .
We prove this fact in this Section. The key observation is
Proposition 4.1. Let s,s

q+1, (X) be closed. Then one can write it locally in the form
= d (4.1)
where sn (X) is a s,s1

n -projectable form and has the coordinates expression:
q
LI11,...,I 1 k
X X
ik+1
= ,...,k ,ik+1 ,...,iq I1 Ik dx
k
dxiq (4.2)
k=0 |I1 |,...,|Ik |r1
where LI11,...,I
,...,k ,ik+1 ,...,iq ,
k
|I1 |, . . . , |Ik | r 1 are smooth functions depending on the
i
variables (x , y , yj , . . . , yj1 ,...,js ) and verify the (anti)symmetry properties in the couples
(Ip , p ) p = 1, . . . k and in the indices ik+1 , . . . , iq .
Proof. Let us use the proposition 3.2 and write s,s

q+1, (X) in the form (3.7). It is easy
to see that 0 has the following generic form:
q+1
AI11,...,I 1 k
X X
ik+1
0 = ,...,k ,ik+1 ,...,iq+1 dyI1 dyIk dx
k
dxiq+1
k=0 |I1 |++|Ik |s1
where AI11,...,I
,...,k ,ik+1 ,...,iq+1 ,
k
|I1 |, . . . , |Ik | s 1 are smooth functions depending on the
i
variables (x , y , yj , . . . , yj1 ,...,js ) and verify the (anti)symmetry properties in the couples
(Ip , p ) p = 1, . . . k and in the indices ik+1 , . . . , iq . In particular the differentials

dyI , |I| r can appear at most once in every term of the preceding sum because if
there would exists a term with at least two such differentials we would get a contradiction
according to the obvious inequality 2r s. So, one can write
X
0 = + dyI I (4.3)
|I|r
where and I are (q + 1) (resp. q) forms which do not contain the differentials dyI ,
|I| r.
But the form 0 is closed (see proposition 3.2) so one can write it, locally, as follows:
0 = d0 (4.4)
with 0 having a structure similar to (4.3):
X
0 = + dyI I ; (4.5)
|I|r
here and I are q (resp. (q 1)) forms which do not contain the differentials dyI , |I|
r. If we substitute (4.3) and (4.5) into (4.4) we easily obtain the consistency condition:
J I = I J , |I|, |J| r.
Applying the usual Poincare lemma one gets from here that I , |I| r have the
following expression
I = I , |I| r
where is a (q 1)-form which does not contain the differentials dyI , |I| r. Now one
substitute this into the expression (4.5) above and gets:
X
0 = + d dyI I dxi .
|I|r
xi
It follows that one can take in (4.4)

X
0 = dyI I dxi
|I|r
xi
without affecting it. But it is clear that this form has the structure
q
BI11,...,I 1 k
X X
ik+1
0 = ,...,k ,ik+1 ,...,iq dyI1 dyIk dx
k
dxiq .
k=0 |I1 |++|Ik |r1
If we define (s,s1
n ) 0 then we have the equality from the statement.
If q = n then we have similarly to (3.8):

Li1 ,...,in = i1 ,...,in L (4.6)
where L is a smooth real function called the (local) Lagrangian.
Now we have a result similar to theorem 2 from [32], ch. 3.2, but as we can see the
proof is much simpler and do not make use of the Young diagrams technique.
68 D.R. Grigore
Theorem 4.2. Let s,s n+1, (X) be closed and verifying the Lepage condition (3.4).
Suppose that we have written it as in proposition 4.1. Then the following formula is true:
= 0 + d + (4.7)
where
0 n!Ldx1 dxn + n (1)|J| dJ I11Ji1 Li1 ,...,in I11 dxi2 dxin ;

X
|J|r1|I1 |
(4.8)
also s,r2
n1,,1 and s,s1
n,,2 .
Proof. We start from the formula (4.2) obtained before and notice that the term correspond-
ing to k = 1 can be written as follows:
LI11 ,i2 ,...,in I11 dxi2 dxin

(1)|J| dJ I11Ji1 Li1 ,...,in I11 dxi2 dxin + d
X
=n
|J|r1|I1 |
where
I11 ,i2 ,...,in1 I11 dxi2 dxin1 .
X

|I1 |r2
is a (n 1)-form with the order of contactness equal to 1. Now we define the form to be
the sum of the terms corresponding to the contributions k 2 in the expression (4.2); this
gives us a n-form with the order of contactness equal to 2. The formula from the statement
follows.
We prove now that the Euler-Lagrange expressions T are following from a Lagrangian
of order r.
Proposition 4.3. In the conditions from above the following result is true:
(1)|J| dJ J L.
X
T = (4.9)
|J|r
Proof. We have by direct computation

n
1X
T,i1 ,...,in
= Li1 ,...,in + (1)p dip L,i ,...,i ,...,i (4.10)
n p=1 1 p n
and
L1 ,i2 ,...,in = n (1)|J| dJ Ji1 1 Li1 ,...,in + ()1 ,i2 ,...,in .
X
(4.11)
|J|r1
If we substitute the second relation into the first one, we obtain the formula from the
statement.
Let us comment this result. First we can say that because the expressions T have
the usual Euler-Lagrange expression, they verify the generalized Helmholtz equations, so
we have an alternative proof of the corollary 3.4. Next, we notice that in fact we have a
sharper result, namely the expressions T follow from a Lagrangian of order r which is
the minimal possible order. Indeed, if T would follow from a Lagrangian of order strictly
smaller than r, the the Euler-Lagrange equations would have the order strictly smaller than
s which would contradict the basic stating point of our analysis. So, we can say that we
have obtained above a form of the conjecture regarding the reduction to the minimal order
in the higher-order Lagrangian formalism.
We close this Subsection with the following result.
Proposition 4.4. In the conditions of the proposition 4.1, let us suppose that q = n and
moreover that the tensors LI11,...,I
,...,k ,ik+1 ,...,in ,
k
|I1 | = = |Ik | = r 1, k = 1, . . . , n
are traceless. The we have the following formula:
!
n k (r 1)!
LI11,...,Ik
= r i1 I1 ikkIk Li1 ,...,in ,
,...,k ,ik+1 ,...,in
k (k + r 1)! 1
|I1 | = = |Ik | = r 1, k = 0, . . . , n (4.12)
The proof goes by induction and is based on the condition (3.6). We do not give the
details but we only mention that the preceding formula appears also in [31].
4.2. Hyper-Jacobians and Variationally Trivial Lagrangians

By definition, the hyper-Jacobians of order s are the following expressions:
k
,..., ,ik+1 ,...,in
i1 ,...,in yIllil ,
Y
JI11,...,Ik k |I1 | = = |Ik | = s 1, k = 0, . . . , n.
l=1
(4.13)
Then we have a result of combinatorial nature.
Proposition 4.5. In the conditions of proposition 4.1 we have for k 2 and |I1 | + +
|Ik | r the following formulas:
for q n:
! q !
1 n X l
LI11,...,Ik
,...,k ,ik+1 ,...,iq = (1) k(q+1)
i1 ,...,in
n! q k l=k
k
,..., ,jl+1 ,...,jq ,i1 ,...,ik ,iq+1 ,...,in
LI11,...,I
X
k+1 l
,...,l ,jl+1 ,...,jq JIk+1 ,...,Il
l
; (4.14)
|Ik+1 |==|Il |=r1
for q n:
! q !
1 n X l
LI11,...,Ik
,...,k ,ik+1 ,...,iq = (1) k(n+1)
iqn+1 ,...,iq
n! q k l=k
k
,..., ,jl+1 ,...,jq ,iqn+1 ,...,ik
LI11,...,I
X
k+1 l
,...,l ,jl+1 ,...,jq JIk+1 ,...,Il
l
(4.15)
|Ik+1 |==|Il |=r1
70 D.R. Grigore
,..., ,j ,...,j
where JIk+1 k+1
,...,Il
l l+1 q
are the hyper-Jacobians of order r and LI11,...,I l
,...,l ,jl+1 ,...,jq ,
|I1 |, . . . , |Ik | r 1 are tensors depending on the variables (x , y , . . . , yj1 ,...,jr1 )
1
and with the symmetry properties in the couples (Ip , p ) p = 1, . . . k and in the indices
ik+1 , . . . , iq . If s = 2r 1 then the formulas above are valid for k = 1 also.
Based on the preceding two results we can give a new proof of an important result from
[17] namely that a(local) Lagrangian L of order r which is variationally trivial (i.e. the
associated Euler-Lagrange expressions are identically zero) must be a linear combination
of hyper-Jacobians of order r (however the coefficients are not completely arbitrary).
5. Two Particular Cases

In this Section we present two particular case which do to have physical relevance. We will
obtain, essentially, known results but we think that it is profitable to see how they follow as
particularizations of the main framework developed in this paper. There will also be some
refinements of these old results.
5.1. The Case s = 2 and n Arbitrary

If we particularize in this case the main theorem of Section 3, we get:
Theorem 5.1. Let X be a differential manifold of dimension N > n and let Pn2 X be
the second order Grassmann manifold associated to it. Let 2,2 n+1 (X) be closed and
verifying the Lepage condition (3.4) and the tracelessness condition (3.6). Then has the
following local expression:
n
Ti00 ,...,k ,ik+1 ,...,in i00 1 k dxik+1 dxin
X
=
k=1
n
X
+ T0 ,...,k ,ik+1 ,...,in 0 k dxik+1 dxin , (5.1)
k=1
where:
the coefficients Ti00 ,...,k ,ik+1 ,...,in and T0 ,...,k ,ik+1 ,...,in are smooth functions of the
variables (xi , y , yj , yjl
);
they are antisymmetric in the indices 1 , . . . , k (resp. in 0 , . . . , k ) and in the

indices ik+1 , . . . , in ;
the following traceless condition is valid:
Tj0 ,...,k ,j,ik+2 ,...,in = 0. (5.2)
The form is globally defined by these conditions.
The proof follows directly from Theorem 3.3 if we take note that the condition (3.13) is
true in this particular case. Let us remark that the previous tracelessness condition is weaker
than the global condition K = 0 introduced in [21], [13].
One can express all the coefficients of the form in terms of the expressions T ; as a
consequence we have = 0 T = 0. This is a stronger form of the relation (3.12).
Finally we have the analogue of Theorem 4.1:
Theorem 5.2. In the conditions of the preceding theorem one can write locally as follows:
= d (5.3)
where
n
!
X 1 n i1
= 1 ikk Li1 ,...,in 1 k dxik+1 dxin . (5.4)
k=0
k! k
Here we have
Li1 ,...,in = i1 ,...,in L (5.5)
where L is a smooth local function depending on the variables (xi , y , yj ).
It is instructive to prove all these results directly in this particular case. We remark in
the end that there are no restriction on the local Lagrangian L other that the independence
on the second order derivatives.
5.2. The Case n = 1 and s Arbitrary

This case was studied in [25], [33], [35] and [14] with minor differences. As before we
have:
Let X be a differential manifold of dimension N = n+1 and let P1s X be the associated
Grassmann manifold of order s. We will denote the local coordinates on it as follows:

x1 7 t and y1,
. . . , 1 7 qk i.e. we have the coordinates (t, q0 , q1 , . . . , qs ); it is natural to
| {z }
ktimes
put also: 1, . . . , 1 7 k .

| {z }
ktimes
Theorem 5.3. In the conditions described above, let s,s

2 (X) closed and verifying the
tracelessness and the Lepage conditions. Then one can write it locally as follows:
= T dt +
X
ij
T i j (5.6)
i+js1
with T , ij smooth functions depending on the variables (t, q , q , . . . , q ) verifying:

T 0 1 k
ij ji
T = T . (5.7)
In this case the form is globally defined.
We define now the total derivative operator by:

s1

dst
X

dt + qj+1 (5.8)
t j=0 qj
and assume that

ij
T = 0, i + j s. (5.9)
72 D.R. Grigore
The closeness conditions for this local description is:

T s1,0
+ 2T = 0, (5.10)
qs
ij
T
= 0, i, j = 0, . . . , s 1. (5.11)
qs
1 T j,0 j1,0
+ dt T + T = 0, j = 1, . . . , s 1, (5.12)
2 qj
ij i1,j i,j1
dt T + T + T = 0, i, j = 1, . . . , s 1, (5.13)
ij jk
T ki
T
T
+ + = 0, i, j, k = 0, . . . , s 1, (5.14)
qk qi qj
1 T T

00
+ dt T = 0. (5.15)
2 q0 q0
One can relax the condition (5.9). Indeed one can accept that the form is given by
the expression (5.6) with the summations restricted only to i, j s 1. In that case one
obtains from the closeness condition, beside the relations (5.10) - (5.15) above, also:
s1,j
T = 0, j, . . . , s 1. (5.16)
Now one uses (5.12) to prove by induction that we have in fact (5.9) [35].
Let us mention two more facts. First, we have from in our particular case directly from
(5.10) - (5.13) + (5.15):
s1ij
ij 1 X j+k T
T = (1)k+j (dt )k (5.17)
2 k=0
k qi+j+k+1
so in particular we have = 0 T = 0.
From this expression one can obtain the Helmholtz equations as in Corollary 3.4. They
are:
s
!
T X
k k T
= (1) (dt )kj , j = 0, . . . , s. (5.18)
qj k=j
j qk
We also note that the expressions (5.17) from above verify identically the system (5.10)
- (5.15). Indeed, only the equation (5.14) should be investigated because the others are used
completely in the induction process to obtain (5.17). But it is not very hard to prove that
(5.17) verify identically (5.14) [33], [35].
Finally we give the analogue of Theorem 4.1 in this case:
Theorem 5.4. In the conditions of the theorem above one can write locally in the form
= d (5.19)
where
r1
X
= Lj j . (5.20)
j=0
Here r = [(s + 1)/2] as before,
r1j
L
Lj
X
(1)i (dt )i , j = 0, . . . , r 1 (5.21)
i=0
qi+j+1
and L is a smooth function depending only on the variables: (t, q , qj , . . . , qr ) which

remains arbitrary for s = 2r and is constrained to be at most linear in qr for s = 2r 1.
The proof is elementary. We provide finally the expressions of the coefficients of the
form in terms of L:
r
X L
T = (1)j (dt )j (5.22)
j=0
qj
(i.e. the usual Euler-Lagrange expressions) and
ij Lj Li
T = , i, j = 0, . . . , s 1. (5.23)
qi qi
6. Conclusions
We first mention that one can use the formalism developed in this paper to analyse higher
order Lagrangian systems with Noetherian symmetries, as in [13], [14], [35], [36],[37] and
[38]. Indeed, if is a diffeomorphisms of the manifold X then one can see that its lift J s
to Pns X leaves invariant the subspace of forms appearing in the left hand side of (3.13).
This means that we can define a Noetherian symmetry as a map such that J s leaves the
Lagrange-Souriau class invariant. It is to be expected that the computations will be much
more difficult than in the two particular cases from the last Section.
Next, we mention that it is not clear if in the general case studied here, the only restric-
tion on the Euler-Lagrange expressions are given by the generalized Helmholtz equations,
but it is reasonable to conjecture that this is true.
Last, we remark that the formalism above could be generalized, in principle, to the case
when the Euler-Lagrange expressions are not restricted by the condition of linearity in the
highest order derivatives, trying for instance to relax the condition (3.3) i.e to factorize to
a smaller subspace.
References
[1] H. F. Ahner and A. E. Moose, Covariant Inverse Problem of the Calculus of Variation,
Journ. Math. Phys. 18 (1977) 13671373.
[2] I. M. Anderson, The Variational Bicomplex (Utah State Univ. preprint, 1989, Aca-
demic Press, Boston, to appear).
[3] I. M. Anderson and T. Duchamp, On the Existence of Global Variational Principles,

American Journ. Math. 102 (1980) 781868.
74 D.R. Grigore
[4] R. W. Atherton and G. M. Homsy, On the Existence and Formulation of Variational

Principles for Nonlinear Differential Equations, Studies in Appl. Math. LIV (1975)
3160.
[5] D. E. Betounes, Extensions of the Classical Cartan Form, Phys. Rev. D 29 (1984)
599606.
[6] D. E. Betounes, Differential Geometric Aspects of the Cartan Form: Symmetry The-
ory, J. Math. Phys. 28 (1987) 23472353.
[7] J. M. Ball, J. C. Currie and P. J. Olver, Null Lagrangians, Weak Continuity, and Vari-
ational Problems of Arbitrary Order, Journ. Functional Anal. 41 (1981) 135174.
[8] E. Cartan, Lecons sur les Invariants Integraux (Hermann, 1922).
[9] A. Galindo and L. Martnez Alonso, Kernels and Ranges in the Variational Formalism,
Lett. Math. Phys. 2 (1978) 385390.
[10] P. L. Garcia, The Poincare-Cartan Invariant in the Calculus of Variations, Symp. Math.
XIV (1974) 219246.
[11] H. Goldschmit and S. Sternberg, The Hamilton-Cartan Formalism in the Calculus of

Variations, Ann. Inst. Fourier 23 (1973) 203267.
[12] M. J. Gotay, A Multisymplectic Framework for Classical Field Theory and the Cal-
culus of Variations. I. Covariant Hamiltonian Formalism, In: Mechanics, Analysis
and Geometry: 200 Years after Lagrange ((M. Francaviglia and D. D. Holms, Eds.)
North-Holland, Amsterdam, 1990) 203235.
[13] D. R. Grigore, A Generalized Lagrangian Formalism in Particle Mechanics and Clas-

sical Field Theory, Fortschr. der Phys. 41 (1993) 569617.
[14] D. R. Grigore, Higher-Order Lagrangian Theories and Noetherian Symmetries, Ro-

manian Journ. Phys. 39 (1994) 1135.
[15] D. R. Grigore, On a Generalisation of the Poincare-Cartan Form to Classical Field

Theory; arXiv:math.DG/9801073.
[16] D. R. Grigore, The Variational Sequence on Finite Jet Bundle Extensions and the
Lagrangian Formalism, Diff. Geom. Appl. 10 (1999) 4377; arXiv:dg-ga/9702016.
[17] D. R. Grigore, Variationally Trivial Lagrangians and Locally Variational Differential

Equations of Arbitrary Order, Diff. Geom. Appl. 10 (1999) 79105.
[18] D. R. Grigore, Higher-Order Lagrangian Formalism on Grassmann Manifolds;

arXiv:dg-ga/9709005.
[19] D. R. Grigore, Lagrangian Formalism on Grassmann Manifolds, In: Handbook of

Global Analysis ((D. Krupka and D. Saunders, Eds.) Elsevier, 2008) 327373.
[20] D. R. Grigore and D. Krupka, Invariants of Velocities and Higher Order Grassmann
Bundles, Journ. Geom. Phys. 24 (1997) 244266; arXiv:dg-ga/9708013.
[21] D. R. Grigore and O. T. Popp, On the Lagrange-Souriau Form in Classical Field The-
ory, Mathematica Bohemica 123 (1998) 7386.
[22] A. Hakova and O. Krupkova, Variational first-order partial differential equations,

Journ. Differential Equations 191 (2003) 6789.
[23] P. Horvathy, Variational Formalism for Spinning Particles, Journ. Math. Phys. 20
(1979) 4952.
[24] I. Kijowski, A Finite-Dimensional Canonical Formalism in Classical Field Theory,

Comm. Math. Phys. 30 (1973) 99128.
[25] L. Klapka, Euler-Lagrange Expressions and Closed Two-Forms in Higher Order Me-
chanics, In: Geometrical Methods in Physics (Conf. on Differential Geometry and
Applications, Czechoslovakia, 1983, Univ. Brno, (D. Krupka, Ed.)).
[26] J. Klein, Espaces Variationels et Mecanique, Ann. Inst. Fourier 12 (1962) 1124.
[27] D. Krupka, A Map Associated to the Lepagean Forms of the Calculus of Variations in
Fibered Manifolds, Czech. Math. Journ. 27 (1977) 114118.
[28] D. Krupka, Lepagean Forms in Higher Order Variational Theory, In: Proceedings
of the IUTAM-ISIMM Symposium on Modern Developments in Analytical Mechanics
(Turin, 1982, Atti della Academia delle Scienze di Torino, Suppl. al Vol. 117, 1983)
198238.
[29] D. Krupka, Geometry of Lagrangian Structures, In: Proceedings of the 14th Winter
School on Abstract Analysis (Srn, 1986, Suppl. ai Rendiconti del Circolo Matematico
di Palermo, Serie II, no.14, 1987) 187224.
[30] D. Krupka, Variational Sequence on Finite Order Jet Spaces, In: Proceedings of the
Conference Differential Geometry and its Applications (August, 1989, World Scien-
tific, Singapore, 1990) 236254.
[31] D. Krupka, Topics in the Calculus of Variation: Finite Order Variational Sequences,
In: Differential Geometry and its Applications (Proceedings Conf. Opava, 1992, Open
Univ. Press) 437495.
[32] D. Krupka, The Geometry of Lagrange Structures, Preprint Series in Global Analysis,
GA 7/97, Dept. of Math., Opava Univ., Czech Rep.
[33] O. Krupkova, Lepagean 2-forms in higher order Hamiltonian mechanics, I. Regularity,

Arch. Math. (Brno) 22 (1986) 97120.
[34] O. Krupkova, Lepagean 2-forms in Higher Order Hamiltonian Mechanics. II Inverse

Problem, Arch. Math. (Brno) 23 (1987) 155170.
76 D.R. Grigore
[35] O. Krupkova, Variational Analysis on Fibered Manifolds over One-Dimensional

Bases, Ph. D. Thesis, Opava Univ., 1992.
[36] O. Krupkova, Liouville and Jacobi theorems for vector distributions, In: Differential
Geometry and Its Applications (Proc. Conf., Opava, August 1992, (O. Kowalski and
D. Krupka, Eds.) Mathematical Publications 1, Silesian University, Opava, Czechoslo-
vakia, 1993) 7588.
[37] O. Krupkova, Symmetries and first integrals of time-dependent higher-order con-

strained systems, Journ. Geom. Phys. 18 (1996) 3858.

Mathematics 1678, Springer, Berlin, 1997).
[39] O. Krupkova, Hamiltonian field theory revisited: A geometric approach to regularity,

In: Steps in Differential Geometry (Proc. Colloq. Diff. Geom., Debrecen, July 2000,
(L. Kozma, P. T. Nagy and L. Tamassy, Eds.) Debrecen University, Debrecen, 2001)
187207.
[40] O. Krupkova, Hamiltonian field theory, Journ. Geom. Phys. 43 (2002) 93132.
[41] P. J. Olver, Hyperjacobians, Determinant Ideals and the Weak Solutions to Variational
Problems, Proc. Roy. Soc. Edinburgh 95A (1983) 317340.
[42] P. J. Olver, Applications of Lie Groups to Differential Equations (Springer, 1986).
[43] H. Poincare, Lecons sur les Methodes Nouvelles de la Mecanique Celeste (Gauthier-
Villars, Paris, 1892).
[44] H. Rund, A Cartan Form for the Field Theory of Charatheodory in the Calculus of
Variations of Multiple Integrals, Lect. Notes in Pure and Appl. Math. 100 (1985) 455
469.
[45] H. Rund, Integral Formulae Associated with the Euler-Lagrange Operator of Multiple
Integral Problems in the Calculus of Variation, quationes Math. 11 (1974) 212229.
[46] D. J. Saunders, An Alternative Approach to the Cartan Form in the Lagrangian Field
Theories, J. Phys. A 20 (1987) 339349.
[47] J. M. Souriau, Structure des Systemes Dynamique (Dunod, Paris, 1970).

Chapter 4
K RUPKA S F UNDAMENTAL L EPAGE E QUIVALENT

AND THE E XCESS F UNCTION OF W ILKINS
D.J. Saunders
Palacky University, Olomouc, Czech Republic
Abstract
We recall the features of Lepage equivalents of first-order Lagrangians on jet
spaces of fibred manifolds, and the corresponding structures associated with homo-
geneous Lagrangians. We demonstrate the correspondence between Krupkas funda-
mental Lepage equivalent and a variant of the Weierstrass excess function introduced
by J. E. Wilkins.
1. Introduction
I first met Demeter Krupka in 1989 at the DGA conference; although we have not published
any joint work, there are many areas of our subject where we take similar approaches, and
over the years we have held many useful discussions. In this short note (which I state at
the outset contains no original material) I should like to make some remarks about one of
Demeters constructions which has fascinated me for several years: this is the Fundamental
Lepage equivalent of a Lagrangian [6]. This object also appears in a paper by David Be-
tounes [1], published a few years later but as a result of independent work. It may be
described as follows.
Given a Lagrangian m-form
= L ( = dx1 dxm )
in n dependent variables y and their derivatives yj , consider the m-form
min{m,n}
X 1 rL
1 r i1 ir
(r!)2 yj11 yjrr
r=0

E-mail address: david@symplectic.demon.co.uk
78 D.J. Saunders
where s = dy s yks dxk are contact forms, and i1 ir = /xir i1 ir1 is defined
recursively. This m-form has the property that it is closed precisely when the original
Lagrangian m-form is null: that is, when the Euler-Lagrange equations of the Lagrangian
vanish identically.
In this note, I want to relate the circle of ideas surrounding this object which arises,
of course, in the exterior differential forms approach to the calculus of variations to
those of an apparently different approach, associated with the Weierstrass excess function.
2. Single-Integral Problems
Consider the variational problem with fixed endpoints given in classical notation by
Z b
L(x, y , yx )dx = 0 .
a
If we modify the integrand L by adding to it a total derivative the problem retains the same
extremals: Z b
d
L(x, y , yx ) + f (x, y ) dx = 0 ,
a dx
because Z b
d
f (x, y ) dx = [f (x, y )]ba

a dx
is independent of the path along which the integral is taken.
We may rewrite this problem in modern notation, using the language of fibred manifolds
and differential forms. Take a fibred manifold : E R with coordinate x on R, and
fibred coordinates (x, y ) (1 n) on E. Given a Lagrangian 1-form = L dx on the
first-order jet manifold J 1 , the problem is to find local sections such that
Z b
(j 1 )
a
takes extreme values. For any f : E R, adding the horizontal differential dh f =

(df /dx)dx does not affect the extremals, because
Z b Z b
1
(j ) dh f = d( f ) = [ f ]ba
a a
is independent of the section as the endpoints are fixed.

Instead of adding a horizontal differential dh f to the Lagrangian, we could add a contact
form, in other words some linear combination of the forms written locally as dy yx dx
where yx are the jet coordinates. Once again, we would have
(j 1 ) (L dx + g (dy yx dx)) = (j 1 ) (L dx)
because (j 1 ) (dy yx dx) = 0. Choosing in particular the coefficient functions g to be

L
g =
yx
Krupkas Fundamental Lepage Equivalent and the Excess Function of Wilkins 79
gives the Cartan form

L
= L dx + (dy yx dx) .
yx
The Cartan form is a Lepage form [7]: that is,
iZ d is a contact form
whenever the vector field Z on J 1 is vertical over E. Furthermore, if is an extremal

section for then
(j 1 ) iX 1 d = 0
for the prolongation X 1 of any variation field X on E; and then
(j 1 ) iY d = 0
for any vector field Y on J 1 by the Lepage property and the fixed endpoints. So j 1 is an
extremal section for .
A straightforward calculation in jet coordinates shows that the Euler-Lagrange form,
d :
which is a 2-form on J 2 , is the 1-contact part of 2,1

L d L
d = (dy yx dx) dx + . . . .
y dx yx
Thus it is obvious that, when is closed, is automatically null. In this present situation,
the converse also holds: we can see this because a null Lagrangian is necessarily of the form
= L dx ehere
df f f
L= = + yx ,
dx x y
and then
f f f
= + yx dx + (dy yx dx) = df .
x y y
3. Multiple-Integral Problems
The theory for multiple-integral problems, unlike that for single-integral problems, is by no
means as straightforward. Two quite distinct approaches are associated with De Donder [5]
and Weyl [9], and with Caratheodory [2].
We start with the De Donder-Weyl approach. Take a fibred manifold : E M where
x (1 i m) are coordinates on M and (xi , y ) (1 n) are fibred coordinates on
i
E; and suppose given a Lagrangian m-form = L on J 1 , where = dx1 . . . dxm

is a fixed volume form on the m-dimensional base manifold M (and the same symbol is
used for its pull-back to J 1 ). The variational problem is to find local sections such that
Z
(j 1 )
C
80 D.J. Saunders
takes extreme values, where C M is a simply-connected compact m-dimensional sub-

manifold with boundary and is given on C. If = i i is a horizontal (m 1)-form
on E, adding the horizontal differential dh = (di /dxi ) does not affect the extremals:
Z Z Z
1
(j ) dh = d( ) =
C C C
is independent of the section as the boundary C is fixed. Alternatively, instead of adding

a horizontal differential dh , we could add a 1-contact form. Put
b = L + L i ,

yi
where = dy yi dxi are local contact forms; then
(j 1 )
b = (j 1 ) (L)
because (j 1 ) = 0. The m-form b is often called the Cartan form, by analogy with
the single-integral case, and it is a Lepage equivalent of ; but, unlike the single-integral
case, there are other possible Lepage equivalents.
The Euler-Lagrange form is the 1-contact part of d b (it is easy to see that this will be
the case for any Lepage equivalent, not merely the one we have described):

L d L
d =
b + . . . .
y dxi yi
b is closed then clearly is null. And for a partial converse, if a null Lagrangian is
If
given by = dh then b is closed. But a complete converse is not available: if i are
functions on E then
i
d
= det = dh 1 dh m
dxj
is also a null Lagrangian, and now

b is not closed.
A somewhat different approach was taken by Caratheodory . Again we consider local
sections such that Z
(j 1 )
C
takes extreme values; but now, instead of adding a trace, we add a determinant. For any m
functions i on E, adding
det di /dxj = dh 1 dh m

does not affect the extremals:

Z Z
1 i j
1 dh 2 dh m

(j ) det d /dx =
C C
is independent of the section as the boundary is fixed. Alternatively, instead of adding a

i i
horizontal form det d /dx , we could add a contact form in such a way that the result
was a decomposable m-form. Put
m
1 ^ L

e = i
L dx + ;
Lm1 yi
i=1
then
(j 1 )
e = (j 1 ) (L)
because (j 1 ) = 0. The m-form e is another Lepage equivalent of , and is called the

Caratheodory form. As before, the Euler-Lagrange form is the 1-contact part of d e , and
if
e is closed then is null. Furthermore, if a null Lagrangian is given by the determinant
= det di /dxj then

e is closed.
Now, however, we have the problem that e is not linear, so that taking the sum of two
determinants we obtain
i i
d1 d2
= det j
+ det
dx dxj
as another null Lagrangian, where e is not closed.

The fundamental Lepage equivalent is designed to avoid these problems; in order to do
this, it is necessary to use higher derivatives of the Lagrangian function. Put
min{m,n}
X 1 rL
= 1 r i1 ir ;
(r!)2 yj11 yjrr
r=0
then, once again, is a Lepage equivalent of . If = det di /dxj then it may be

shown [4] that = e . But now is linear, and d = 0 for any null lagrangian .
4. Homogeneity
We turn now to a different, but related, type of variational problem. A parametric varia-
tional problem is one where the submanifolds we consider are given parametrically, and are
not the images of sections. The basic example of such a problem arises in Finsler geometry,
where we consider 1-dimensional parametric problems:
Z b
(j 1 ) L dt = 0
a
where E is a manifold with local coordinates ua (1 a 1 + n), : [a, b] E is a

curve, and the Lagrangian L is a function. In order for the extremals to be independent
of parametrization (but nevertheless to have a particular orientation), it is necessary and
sufficient for L to be positively homogeneous of degree 1:
L
ua = L.
ua
82 D.J. Saunders
A homogeneous Lagrangian L gives rise to its Hilbert form

L a
L = du ;
ua
if is an extremal curve for L then j 1 is an extremal curve for L . Any Lagrangian
involving x explicitly (on a jet bundle) gives rise, in a standard way, to a homogeneous
Lagrangian, and the Hilbert form of this homogeneous Lagrangian then projects to the
Cartan form of the original Lagrangian.
Parametric m-dimensional variational problems, on a manifold E with coordinates ua
(1 a m + n) and its manifold of regular m-velocities FE with coordinates (ua , uai )
(1 i m) are considered in the same way:
Z
(j 1 ) L dm t = 0
C
where L now satisfies the homogeneity condition

L
uaj = ji L .
uai
There are now m Hilbert 1-forms
i L a
L = du ,
uai
and any Lagrangian involving xi explicitly (on a jet bundle) gives rise to a homogeneous
Lagrangian.
The Cartan form from the De Donder-Weyl theory makes no sense in this context. But
we can construct
m
eL = 1
^
i
L
Lm1
i=1
and this projects to the Caratheodory form of the original Lagrangian [3]. If we define the
tensors

S i = dua a
ui
then we can construct the Hilbert forms by
i
L = S i dL .
We can also construct the fundamental form

1 1 2
L = S dS d S m dL
m!
and this projects to the fundamental Lepage equivalent of the original Lagrangian [4]. The
coordinate formula for L is
1 mL
L = dua1 duamm . (1)
m! ua11 uamm 1
5. The Weierstrass Excess Function

We return to inhomogeneous variational problems, and consider the classical question of
whether an extremal is a genuine local minimum of the action functional; a useful reference
for this is [8]. For questions like these, we need to distinguish between weak and strong
local minima, with the distinction arising from the topology on the space of sections: if
we regard sections as close if their values are close then we obtain a strong local mini-
mum, whereas if we also require their derivatives to be close then we obtain a weak local
minimum.
In order to consider this question, one approach for a single-integral problem is to define
the Weierstrass excess function: in classical notation, this is given by
L
E(x, y , yx , z ) = L(x, y , z ) L(x, y , yx ) (z yx ) ,
yx
or in modern notation, with 1,0 ((x)) = (x), we would write
E(jx1 , (x)) = L((x)) L(jx1 ) dF L((x) jx1 ) .
The Weierstrass necessary condition is then that, if is a strong minimum of the variational
problem, then
E(jx1 , (x)) 0
for any section of J 1 R projecting to .
For multiple-integral problems the question is, as we might expect, more complicated.
The De Donder-Weyl theory suggests that we should use
b i , y , yi , zi ) = L(xi , y , zi ) L(xi , y , yi ) L (zi yi )

E(x
yi
as an excess function, whereas the Caratheodory theory suggests

i i 1 i L
E(x , y , yi , zi ) = L(x , y , zi ) p1 det j L + (zj yj )
e
L yi
instead. It turns out, however, that neither Eb 0 nor E

e 0 is, in general, a necessary
condition for a strong minimum.
Similar difficulties arise for homogeneous problems. For these problems the De
Donder-Weyl theory is not appropriate. The excess function for the Caratheodory theory
becomes
a a a a a 1 L a
E(u , ui , vi ) = L(u , vi ) p1 det
e v
L uai j
in the homogeneous case, but again E
e 0 is not a necessary condition for a strong mini-
mum. Consider, for example,
L = u11 u22 u12 u21 + u31 u42 u32 u41 :
this is a null Lagrangian, but E

e 6= 0.
84 D.J. Saunders
It is, however, possible to define an excess function for homogeneous problems which
does not have this problem; a suitable definition was proposed in a paper by J. E. Wilkins
in 1944 [10]. This paper defines an excess function E by the formula
1 j1 jm mL
E(ua , uai , via ) = L(ua , via ) v 1 vjmm , (2)
(m!)2 i1 im ui11 uimm j1
and demonstrates that E = 0 whenever the Lagrangian is null. The structure of this
formula is interesting, as it involves higher derivatives of the Lagrangian. Indeed, by com-
paring equations (1) and (2) it may be seen that the derivatives of the Lagrangian, and their
coefficients, are combined in exactly the same arrangement here as in the fundamental form
L of the homogeneous Lagrangian, a form directly related to Krupkas fundamental Lep-
age equivalent. It seems, therefore, that this type of structure, discovered in its two different
manifestations in 1944 and 1977, is of some importance in studying the properties of null
Lagrangians.
Acknowledgements
The author expresses his acknowledgments to the Czech Science Foundation (grant
no. 201/06/0922 for Global Analysis and its Applications).
References
[1] D. E. Betounes, Extensions of the classical Cartan form, Phys. Rev. D 29 (1984), 599
606.
[2] C. Caratheodory, Uber die Variationsrechnung bei mehrfachen Integralen, Acta
Szeged. Sect. Scient. Mathem. 4 (1929), 193216.
[3] M. Crampin and D. J. Saunders, The Hilbert-Caratheodory form for parametric multi-
ple integral problems in the calculus of variations, Acta Appl. Math. 76 (2003), 3755.
[4] M. Crampin and D. J. Saunders, On null Lagrangians, Diff. Geom. Appl. 22 (2005),
131146.
[5] Th. De Donder, Theorie invariantive du calcul des variations (nouvelle edit.: Paris,
Gauthier-Villars, 1935).
[6] D. Krupka, A map associated to the Lepagean forms in the calculus of variations,
Czech Math. J. 27 (1977), 114118.
[7] Th.-H.-J. Lepage, Sur les champs geodesiques du calcul des variations, Bull. Acad.
Roy. Belg. Cl. Sci. V Ser 22 (1936), 716729, 10361046.
[8] H. Rund, The Hamilton-Jacobi theory in the calculus of variations (London: Van
Nostrand, 1966).
[9] H. Weyl, Geodesic fields in the calculus of variations for multiple integrals, Ann. Math.
(2nd Ser.) 36 (1935), 607629.
[10] J. E. Wilkins, Multiple integral problems in parametric form in the calculus of varia-
tions, Ann. Math. (2nd Ser.) 45 (1944), 312334.
Chapter 5
L EPAGE C ONGRUENCES
IN D ISCRETE M ECHANICS
Antonio Fernandez1 and Pedro L. Garca2

1
Dpto. de Matematica Aplicada, Universidad de Salamanca.
2
Dpto. de Matematicas, Universidad de Salamanca
Abstract
We introduce the concept of contact 1-form in Discrete Mechanics. In terms of this
concept, we express the Poincare-Cartan form of a discrete Lagrangian by two formu-
las that generalizes to the Discrete Mechanics the classical Lepage congruences. The
requirement for these congruences to have similar properties to those in the continuous
case leads to a special class of mechanical systems, which interest is illustrated with
some examples.
1. Introduction
One of the most beautiful geometrical doctrines of the last century is, without any doubt,
that known in the literature from the beginning of the 70s as the Hamilton-Cartan formal-
ism of the Variational Calculus. This doctrine starts in the 30s with the works of De Don-
der [4] and Weyl [18] and its generalization by Lepage [15, 16] some years after, its modern
formulation in terms of jet fiber bundles of bundled manifolds was stablished 30 years after.
Two key facts of this formulation were in its starts the identification of the Poincare-Cartan
form with the boundary term of the formula of variation of the integral of the action for
first order variational problems and the characterization of this form from a natural glob-
alization of the classical Lepage congruences (three basic references from this first stage
are: Goldschmidt and Sternberg [10], Garca [7] and Krupka [12, 13]. The generaliza-
tion of this setup to higher order problems was more problematic (Garca and Munoz [8],
Ferraris and Francaviglia [6], Krupka and Stepankova [14], Horak and Kolar [11] and ref-
erences therein), and moreover, the recent treatment of the constrained problems and its

Dedicated to Demeter Krupka

E-mail address: anton@usal.es

E-mail address: pgarcia@usal.es
86 Antonio Fernandez and Pedro L. Garca
relation with the Lagrangian reduction topic (Fernandez, Garca and Rodrigo [5], Bibbona,
Fatibene and Francaviglia [1], Garca and Rodrigo [9], etc.).
In its simplest case Analytical Mechanics the setup is well known:
Given a fibration p : Q = Q R R (Q : configuration manifold of a mechanical
system and R the time line), the natural space were the Lagrangian density Ldt lives is
the affine bundle : J 1 (Q) Q of the 1-jets of the local sections of p which geometry
is driven by a 1-form with values on the vector bundle V (Q) (V (Q) : bundle of the
p-vertical vector fields of Q) defined by the rule:
jt1 s (D) = (dvert.
t s)( D) = D st (p D) D Tjt1 s (J 1 (Q)).
In the standard set of local coordinates (t, q i , q i ) of J 1 (Q):

X
dq i q i dt i .

=
q
i
This differential form, known in the literature as the contact 1-form of the 1-jet bundle,
allows toR obtain an intrinsic expression of the differential s L of the integral of the action
L(s) = j 1 s Ldt on a section s in the following terms:
Z Z
E(s) (D1 ) dt + d (j 1 s) iD1 ,

s L = LD1 Ldt =
j1s j1s
where j 1 s and D1 are the 1-jet extensions of the section s and a vector field D on Q respec-
tively, E : s E(s) s (V E) is a second order differential operator (Euler-Lagrange
operator), is a 1-form on J 1 (Q) (Poincare-Cartan form) and is the duality pairing.
Even more, it is possible to characterize the Poincare-Cartan form by the conditions:
= p + Ldt, ,
d = E
where p and E are, respectively, a function and a 1-form on J 1 (Q) with values on V (Q)
and where the products and are taken with respect to the duality pairing.
These are in the case of the mechanics the so-called Lepage congruences, from where
it is possible to recover the variation of the integral of the action by restricting to j 1 s the
Lie derivative with respect to D1 of this congruences bearing in mind that (j 1 s) = 0 and
y (j 1 s) L1D = 0.
Established the problem in these terms, in the present work we are going to start the
study of this question in Discrete Mechanics, which we believe that it has not been treated
up to now, perhaps because for the non existence of a clear concept of tangency to a discrete
curve in this field.
After a brief outline of the basic principles of the Discrete Mechanics on section 2.
(see [17, 2, 3]), in section 3. we introduce a general notion of contact 1-form characterizing
intrinsically one of them in terms of which the usual discretization of the derivative as an
incremental quotient in two nearby instants is codified (Theorem 3). From this concept,
the Poincare-Cartan 1-form of a discrete Lagrangian is characterized by a set of conditions
that extend to the discrete case the classical Lepage congruences (Theorem 4). The require-
ment for this congruences to have similar properties to those in the continuous case leads to
a special class of mechanical systems, which interest is illustrated in section 5. with some
examples.
Lepage Congruences in Discrete Mechanics 87
2. Discrete Mechanics
Let Q = R Q be the configuration space.
Definition 1. A discrete section S d is a collection of points of Q
sd = {(t0 , q0 ), (t1 , q1 ), . . . , (tN , qN )}
or, equivalently, a point of QN +1 .

Definition 2. A discrete action is a differentiable application Sd : QN +1 R
From now onwards, we are only going to consider the discrete actions of the form
N
X 1
d
S (t0 , q0 , t1 , q1 , . . . , tN , qN ) = Lk (tk , qk , tk+1 , qk+1 )(tk+1 tk ), (1)
k=0
where Lk : Q Q R is a differentiable function that will be called local discrete La-

grangian.
Definition 3. A critical discrete section of the discrete action Sd is a discrete section sd
such that
dsd Sd (0, D1 , . . . , DN 1 , 0) = 0, Dk T(tk ,qk ) Q k = 1, . . . , N 1.
If we calculate dSd we get:

1
NX
d
dS =d Lk (tk , qki , tk+1 , qk+1
i
)(tk+1 tk ) =
k=0

L0 X L0
= (t1 t0 ) L0 dt0 + i
(t1 t0 )dq0i
t0 q0
i
N 1
X Lk1 Lk (2)
+ (tk tk1 ) + Lk1 + (tk+1 tk ) Lk dtk
tk tk
k=1
X Lk1 !
Lk
+ i
(tk tk1 ) + i
(tk+1 tk ) dqki
i
qk qk

LN 1 LN 1 i
+ (tN tN 1 ) + LN 1 )dtN + i
(tN tN 1 dqN
tN qN
and, hence, the following theorem holds
Theorem 1. A discrete section sd = (t0 , q0 , t1 , q1 , . . . , tN , qN ) is critical of the discrete
action (1) if and only if satisfies the discrete Euler-Lagrange equations
Lk1 Lk

0= (tk tk1 ) + (tk+1 tk ), i = 1, . . . , n
qki qki

k = 1, . . . , N 1.
Lk1 Lk
0= (tk tk1 ) + Lk1 + (tk+1 tk ) Lk
tk tk
If the Euler-Lagrange equations allow us to define (tk+1 , qk+1 ) as functions of

(tk1 , qk1 ) and (tk , qk ), we can define a set of maps
k : Q Q Q Q
((tk1 , qk1 ), (tk , qk )) 7 ((tk , qk ), (tk+1 , qk+1 ))
that is called the discrete flux of the critical section sd .

In terms of the discrete flux, the Euler-Lagrange equations are

Lk Lk1
k (tk+1 t k ) = (tk tk1 ),
qki qki
(3)
Lk Lk1
k (tk+1 tk ) + Lk = (tk tk1 ) + Lk1
tk tk
and, hence, a discrete flux k defines a critical section sd if and only if equations (3) hold.
On the other hand, if we look at the first and last terms of the differential of the ac-
tion (2), we can define a pair of 1-forms:
X Lk
Lk
= i
(tk+1 t k )dq i
k + (tk+1 t k ) + Lk dtk ,
qk tk
i
X Lk
+ i Lk
= i
(tk+1 tk )dqk+1 + (tk+1 tk ) + Lk dtk+1
qk+1 tk+1
i
and, then, a 1-form ( , + ) on Q Q that will be called Poincare-Cartan form.

The two 1-forms and + are related by the formula
+ = d Lk (tk , qk , tk+1 , qk+1 )(tk+1 tk )

(4)
and, if is the discrete flux of a critical section, by (3), we have that and + in two con-
secutive stages are k related, that is k = + . Finally, combining this two equations
we have
Theorem 2 (Discrete Cartan Equation). A discrete flux k defines a critical section sd if

and only if
k + + = d k Lk (tk+1 tk ) .

(5)
Finally, observe that if we differentiate the discrete Cartan Equation (5) we get the
simplecticity of the discrete flux k
Corollary 1. If k is the discrete flux of a critical section sd , then
k d+ = d+ .
As we have done in the continuous mechanics, we have obtained the Poincare-Cartan

form from the discrete variational principle. The next two sections of the paper will be
devoted to obtain (for a particular class of Lagrangians) a characterization of the Poincare-
Cartan form on a stage Q Q from the local discrete Lagrangian only.
3. Discrete Contact 1-Forms

In order to introduce the notion of contact 1-form in the discrete realm, we need two defi-
nitions in the direct product of a manifold.
Definition 4. Let M be a differentiable manifold. The conjugation of M M is the

diffeomorphism of M M that interchanges the factors:
: M M M M
(x1 , x2 ) 7 (x2 , x1 ).
The conjugation diffeomorphism acts in a natural way on the vector fields of M M ,

and thus defining a 1-1 tensor on M M , T , which local expression is the following:
If (x1 , . . . , xn ) is a coordinate system in M , let (x11 , . . . , xn1 , x12 , . . . , xn2 ) be the induced
coordinate system in M M . As (x11 , . . . , xn1 , x12 , . . . , xn2 ) = (x12 , . . . , xn2 , x11 , . . . , xn1 ),
it is (xi2 ) = xi1 and (xi1 ) = xi2 , and, hence,

T = , T = .
xi1 xi2 xi2 xi1
Definition 5. A contact 1-form on Q Q is a 1-1 tensor on Q Q, ( , + ) verifying
1 ( , + ) is a projector of Q Q R R.
2 T + =
Local expression: Let (q 1 , . . . , q n ) be a set of local coordinates on Q, and let

(t0 , q01 , . . . , q0n , t1 , q11 , . . . , q1n ) be the induced coordinates on Q Q. Then, the general
expression of and + is:
X
= (dq0i + f0i dt0 ) ,
i
q0i
X
+ = (dq1i + f1i dt1 )
i
q1i
and the condition T + = implies that f1i = f0i .

Hence, the local expression of the contact 1-form is
X
= (dq0i ui dt0 ) ,
i
q0i
(6)
+
X
= (dq1i i
u dt1 ) i .
i
q1
Definition 6. The contact distribution D will be the kernel of the contact 1-form ( , + )
D = {(D0 , D1 ) : ( (D0 ), + (D1 )) = (0, 0)}.

From the local expresion of the contact 1-form (6), it follows that the contact distribution
is locally spanned by the vector fields D and D+ given by
X
D = + ui i ,
t0 q0
i
X
D+ = + ui i .
t1 q1
i
Theorem 3. There exists a unique contact 1-form ( , + ) which contact distribution D is

integrable and its first integrals are -invariant.
Proof. From the local expression of the local vector fields D and D+ , it follows that the
integrability condition is
X
+ j + j
[D , D ] = D (u ) j D (u ) j = 0
j q1 q0
and, hence,
D (uj ) = D+ (uj ) = 0, j = 1, . . . , n
ui
and the functions are first integrals of the distribution. From this set of first integrals we
can construct a set of n additional first integrals
v0j = q0j uj t0 , j = 1, . . . , n.
Now, by imposing the condition of -invariance of the first integrals, we have that, being
v0j = q1j uj t1
it should be v0j = v0j , that is
q1j q0j
q0j uj t0 = q1j uj t1 , uj = , j = 1, . . . n.
t1 t0
From now onwards we are going to consider only this canonically determined contact
1-form.
4. Discrete Lepage Congruences

Let L : Q Q R a discrete Lagrangian, and let ( , + ) be its Poincare-Cartan form
as defined in Section 2..
In these conditions, we have:
Theorem 4. If L is a first integral of the contact distribution D on Q Q, its Poincare-
Cartan form ( , + ) is univocally determined by the following conditions:
= p + Ldt0 ; + = p+ + + Ldt1 ,
(7)
+ = d L(t1 t0 ) ,

where p and p+ are functions on Q Q with values on T Q0 and T Q1 respectively.

And, conversely, if the Poincare-Cartan form of a discrete Lagrangian L satisfies the
conditions (7), then L is a first integral of the contact distribution D
Proof. Let (q 1 , . . . , q n ) be local coordinates on Q, and (t0 , q01 , . . . , q0n , t1 , q11 , . . . , q1n ) the
induced coordinates on Q Q. In these coordinates, the local expression of the Poincare-
Cartan form is
X L
+ i L
= (t1 t0 )dq1 + (t1 t0 ) + L dt1 =
i
q1i t1
X L
(t1 t0 ) dq1i ui dt1 + Ldt1

= i
i
q1
X
L i L
+ (t1 t0 ) i
u +
q 1 t1
i
X
+ + L i L
=p + Ldt1 + (t1 t0 ) u + ,
q1i t1
i
where
X L
p+ = (t t0 )dq1i
i 1
i
q1
and the local expression of is

X
L i L
= p + Ldt0 + (t1 t0 ) u + ,
i
q0i t0
where
X L
p = (t t0 )dq0i .
i 1
i
q0
From these two local expressions, it follows that the discrete Lepages congruences (7) hold
if and only if the Lagrangian function L is a first integral of D.
On the other hand, if a 1-form ( , + ) satisfies the Lepages congruences (7) we have
that X X
p = pi dq i
0 , p +
= p+ i
i dq1
i i
with

p
i =
, p+
i = +
q0i q1i
and, hence,

L
p+
i = +
= d L(t1 t0 ) = i (t1 t0 ),
q1i q1i i
q1 q
1
L
p +
+

i = = d L(t1 t0 ) = i (t1 t0 )
q0i q0i q0i q1
and, then, it is
= , + = + .
Remark 1. Despite its apparent restriction, there is a rather natural way of obtaining local
discrete Lagrangians that fullfil the hypothesis of Theorem (4):
Starting from a continuous Lagrangian L(t, q, q), we can construct a family of local
discrete Lagrangians, simply by evaluating the Lagrangian in an intermediate point of the
segment joining the points (tk , qk ) and (tk+1 , qk+1 )
i
qk+1 qki

Ld (tk , qki , tk+1 , qk+1
i
, ) i i
= L (1 )tk + tk+1 , (1 )qk + qk+1 , .
tk+1 tk
Then, the condition of being Ld a first integral of the contact distribution can be ex-
pressed as: !
L X i L
D Ld =(1 ) + uk i = 0,
t q
i
!
L X i L
D+ Ld = + uk i = 0.
t q
i
On the other hand, as
dLd L X L
= (tk+1 tk ) + (q i qki )
d t q i k+1
i
it follows that Ld is a first integral of the contact distribution if and only if dLd /d = 0,
that is, if and only if Ld is independent of the parameter .
5. Examples
In this section we are going to deal with discrete local Lagrangians obtained from continu-
ous Lagrangians of the form
n
1X i 2
L(ui , v i ) = (u ) (v 1 , . . . , v n ),
2
i=1
where ui = q i and v i = q i tq i . The Euler-Lagrange equations of the variational problem

defined by this Lagrangians are:

d L L
i
i = 0, (8)
dt q q
where
L L
i
= ui + i t, i
= i
q v q v
and, hence, equations (8) are:
dui

d
+2 i + = 0.
dt v dt v i
dv i i
But, given that dt = t du
dt we can express the Euler-Lagrange equations in terms of
i
v only:
1 dv i dv i

d d 2
0= +2 i + = t (9)
t dt v dt v i dt dt v i

that is, the variables v i are related by the equations v i = t2 v i + . Let us assume that this
equations allow us to determine v as functions of t. Given that v i = q tq, we have that
i
vi q i tq i q
Z i
v (t)
2
= 2
= , q(t) = t dt.
t t t t2
By Theorem (4), the Poincare-Cartan form ( , + ) in each stage (t0 , q0 , t1 , q1 ), is

given by the canonical momenta
L X L uj L v j

L L
+
pi = i (t1 t0 ) = (t1 t0 ) j i
+ j i = i
t0 i ,
q1 u q1 v q1 u v
j
L X L uj

L v j

L L

pi = i (t1 t0 ) = (t1 t0 ) j i
+ j i = i
t1 i
q1 u q0 v q0 u v
j
in our particular case,

p+ i
i = u + t0,
v i (10)

p i
i = u + t1 .
v i
On the other hand, as we have seen, the Cartan equation (5) is equivalent to
k = +
and, hence, by (7) to
k p +
i,k =pi,k ,
X X
(11)
k p i
i,k uk Lk = p+ i
i,k1 uk1 Lk1 .
i i
Example 1 (The P free particle). This mechanical system is given by the Lagrangian func-
tion L(ui ) = 21 i (ui )2 ; and its Euler-Lagrange equations are
ui = 0 ui (t) = i t + i , i = 1, . . . , n. (12)
On the other hand, the local discrete Lagrangians defined by this Lagrangian function
are
i
qk+1 qki
1X i 2
Lk (uik ) = (uk ) , uik =
2 tk+1 tk
i
and, hence, the momenta (10) are

p+ i
i,k = uk = pi,k
and the discrete Cartan equations (11) are

i
qk+1 qki i
q i qk1
k uik = uik1 = k
tk+1 tk tk tk1
that is, the discrete Cartan equations do not impose conditions on all the discrete variables;
in particular, every time discretization t0 , . . . , tN is compatible with them. If we choose
a regular discretization, whith tk+1 tk = tNNt0 = h, we have the following discrete
equations:
i
qk+1 2qki + qk1 i
=0
h
that is the standard central-point discretization (multiplied by h) of the Euler-Lagrange
equations (12).
Example 2. Let us now add a linear potential to the free particle Lagrangian:
1X i 2 X 1
L(ui , v i ) = (u ) i v i = u ut A v t ,
2 2
i i
where we have wrote u = (u1 , . . . , un ), v = (v 1 , . . . , v n ), A = (a1 , . . . , an ).

The Euler-Lagrange equations are, then
du
+ 2A = 0 q = 2A.
dt
The solutions of the Euler-Lagrange equations can be obtained by the general procedure
previously introduced:
Z i
i 2 i v (t) i
v (t) = t i + i , q (t) = t dt = t ti + i = i t2 i t + i .
t2 t
The local discrete Lagrangians defined by this Lagrangian function can be written in
vectorial form as
1
Lk (uik , vki ) = uk utk A vkt
2
and, then, the canonical momenta are:
p+
k = uk + Atk , p
k = uk + Atk+1 .
The first equation of (11) is, then
k (uk + Atk+1 ) = uk1 + Atk1 .
Let us now drop the +

k in order to simplify the explanation; and, then, we have that
uk = uk1 A(hk + hk1 ), hk = tk+1 tk , hk1 = tk tk1 . (13)
The second equation is

1 1
p + t t t t t t
k Lk = pk1 Lk1 2 uk uk +A(uk tk+1 +qk+1 ) = 2 uk1 uk1 +A(uk1 tk1 +qk1 )
and, given that vk = qk tk uk = qk+1 tk+1 uk+1 , this equation is
1 1
uk utk + A utk hk = uk1 utk1 A utk1 hk1 .
2 2
If we substitute uk using (13), we get, after a straightforward computation, that
1
A At (h2k h2k1 ) = 0
2
that is, the time discretization should be uniform:
tN t0
hk = hk1 = h =
N
and, returning to the first equation, we have that
qk+1 qk qk qk1 qk+1 2qk + qk1
= 2Ah = 2Ah
h h h
that is, once again, the central-point discretization of the Euler Lagrange equations.
Example 3. Finally, we are going to consider a quadratic potential:

1X i 2 1X i 1 1
L(ui , v i ) = (u ) v ij v j = u ut v A v t ,
2 2 2 2
i i,j
where A is a symmetric matrix.

Once again, the solutions of the Euler-Lagrange equations can be easily obtained:
v(t) =t2 v(t) A + v(t) = (I t2 A)1 ,

(I t2 A)1
Z
q(t) = t dt,
t2
where = (1 , . . . , n ) is a vector of constants and we have assumed that the matrix

I t2 A is invertible.
The local discrete Lagrangians defined by this Lagrangian function can be written in
vectorial form as
1 1
Lk (uk , vk ) = uk utk v A vkt
2 2
and, then, the canonical momenta are:
p+
k = uk + tk vk A, p
k = uk + tk+1 vk A.
Using the same conventions that in the previous example, the first equation of (11) is,
then
uk + tk+1 vk A = uk1 + tk1 vk1 A
and, hence
uk = uk1 Bk1 Bk1 + (tk1 tk+1 )qk A Bk1 , (14)
where we have denoted by Bk = I tk+1 tk A. From this expression, it is easy to obtain

that
vk = vk1 Bk1 Bk1 . (15)
The second equation is
1 1
p + t t t t t
k Lk = pk1 Lk1 2 uk uk +vk A(uk tk+1 +vk ) = 2 uk1 uk1 +vk1 Aqk1
and, by substituting uk and vk using (14) and (15) it is possible to obtain tk+1 in terms
of (tk1 , qk1 , tk , qk ); however, in this case the time distribution it is not the regular one
tk = t0 + kh in general, as it is easy to see in the simplest case of being A a diagonal
matrix.
Acknowledgements
This work has been partially supported by the Spanish Ministerio de Ciencia y Tecnologa,
project number MTM2004-01683.
References
[1] E. Bibbona, L. Fatibene and M. Francaviglia, Gauge-natural parameterized variational
problems, vakonomic field theories and relativistic hydrodynamics of a charged fluid,
Int. J. Geom. Methods Mod. Phys. 3 (8) (2006) 15731608.
[2] J.-B. Chen, H.-Y. Guo and K. Wu, Total variation and variational symplectic-energy-
momentum integrators, (2001); arXiv:hep-th/0109178.
[3] J.-B. Chen, H.-Y. Guo and K. Wu, Discrete total variation calculus and Lees discrete
mechanics, Appl. Math. Comput. 177 (1) (2006) 226234.
[4] Th. De Donder, Theorie invariantive du calcul des variations (Nuov. ed. Gauthier-
Villars, Paris, 1935).
[5] A. Fernandez, P. L. Garca and C. Rodrigo, Lagrangian reduction and constrained vari-
ational calculus, In: Proceedings of the IX Fall Workshop on Geometry and Physics
(Vilanova i la Geltru, 2000, Publ. R. Soc. Mat. Esp., vol. 3, R. Soc. Mat. Esp., Madrid,
2001) 5364.
[6] M. Ferraris and M. Francaviglia, On the global structure of Lagrangian and Hamilto-
nian formalisms in higher order calculus of variations, In: Proceedings of the Interna-
tional Meeting on Geometry and Physics (Florence, 1982, Bologna, Pitagora, 1983)
4370.
[7] P. L. Garca, The Poincare-Cartan invariant in the calculus of variations, In: Symposia
Mathematica (Vol. XIV, Convegno di Geometria Simplettica e Fisica Matematica, IN-
DAM, Rome, 1973, Academic Press, London, 1974) 219246.
[8] P. L. Garca and J. Munoz, On the geometrical structure of higher order variational
calculus, In: Proceedings of the IUTAM-ISIMM Symposium on Modern Developments
in Analytical Mechanics (Vol. I, Torino, 1982, vol. 117, 1983) 127147.
[9] P. L. Garca and C. Rodrigo, The momentum map in vakonomic mechanics, In: Pro-
ceedings of the XII Fall Workshop on Geometry and Physics (Publ. R. Soc. Mat. Esp.,
vol. 7, R. Soc. Mat. Esp., Madrid, 2004) 111123.
[10] H. Goldschmidt and S. Sternberg, The Hamilton-Cartan formalism in the calculus of

variations, Ann. Inst. Fourier (Grenoble) 23 (1) (1973) 203267.
[11] M. Horak and I. Kolar, On the higher order Poincare-Cartan forms, Czechoslovak
Math. J. 33 (108) (3) (1983) 467475.
manifolds. I. Critical sections, J. Math. Anal. Appl. 49 (1975) 180206.
manifolds. II. Invariance, J. Math. Anal. Appl. 49 (1975) 469476.
[14] D. Krupka and O. Stepankova, On the Hamilton form in second order calculus of
variations, In: Proceedings of the International Meeting on Geometry and Physics
(Florence, 1982, Bologna, Pitagora, 1983) 85101.
[15] Th. Lepage, Sur les champs geodesiques des integrales multiples, Acad. Roy. Bel-
gique. Bull. Cl. Sci. (5) 27 (1941) 2746.
[16] Th. Lepage, Champs stationnaires, champs geodesiques et formes integrables. II,
Acad. Roy. Belgique. Bull. Cl. Sci. (5) 28 (1942) 247265.
[17] J. E. Marsden and M. West, Discrete mechanics and variational integrators, Acta Nu-
mer. 10 (2001) 357514.
[18] H. Weyl, Geodesic fields in the calculus of variations for multiple integrals, Ann. of
Math. (2) 36 (3) (1935) 607629.
Chapter 6
F INITE O RDER VARIATIONAL S EQUENCES :

A S HORT R EVIEW
Raffaele Vitolo
Department of Mathematics E. De Giorgi,
University of Lecce, Via per Arnesano, 73100 Lecce, Italy
Abstract
Variational sequences are complexes of modules or sheaf sequences in which one

of the maps is the EulerLagrange operator, i.e., the differential operator taking a La-
grangian into its EulerLagrange form. In this review paper we discuss variational
sequences on finite order jets, with special emphasis on Krupkas approach. We also
discuss recent results on this topic as well as possible research directions.
2000 Mathematics Subject Classification. Primary 58J10, secondary 58A12, 58A20.

Key words and phrases. Jet spaces, variational sequence, variational bicomplex.
Introduction
In the Seventies, during a process of geometrization of the calculus of variations, it was re-
alized that operations like passing from a Lagrangian to its EulerLagrange form were part
of a complex, namely, the variational sequence. Foundational contributions to variational
sequences are in the papers [3, 7, 12, 39, 40, 41, 42, 43, 44, 45].
Among the problems which were solved by the variational sequence was the so-called
global inverse problem of the calculus of variations: given a set of EulerLagrange equa-
tions, the vanishing of Helmholtz conditions is a necessary and sufficient condition for
the existence of a local Lagrangian for the given equations; does there exist a global La-
grangian? It was proved that the answer is in the cohomology of the variational sequence.
More precisely, the cohomological obstruction for always having a global Lagrangian is the
n + 1-st de Rham cohomology of the space of independent and dependent variables.

E-mail address: raffaele.vitolo@unile.it
100 Raffaele Vitolo
The geometric framework for variational sequences is that of jet spaces. Infinite order
jet spaces were used as a rule, with the exception of [3]. There are some technical reasons
for that choice: the first and most important is that on infinite order jet spaces the contact
distribution is integrable and admits an intrinsic direct summand. This fact leads to much
simpler computations.
On the other hand, using infinite order jets one simply drops any information on the
order of the objects involved in the computations. In this sense, the use of finite order
jets can lead to finer results. A first approach in this sense was in [3]. In that paper the
finite order variational sequence was truncated after the space of EulerLagrange forms.
Moreover, in order to obtain the solution of the global inverse problem the authors resorted
to infinite order jets. Another approach was through C-spectral sequences in [8, 9]. But it
used one conjecture about the structure of contact forms (see Theorem 1.3).
In [23] Krupka proved the above conjecture and was able to give the first formulation
of the (long) variational sequence on finite order jets. The formulation was different from
both the so-called variational bicomplex [2, 37] and the C-spectral sequence [7, 44]. The
idea is rather simple: consider the de Rham complex on jets of order r. Then a subsequence
of forms which yield trivial contribution to action-like functionals is defined. The quotient
of the former sequence with the latter one yields the finite order variational sequence.
In this paper, after a preliminary section on jet spaces and contact forms, we describe
Krupkas finite order variational sequence. In the final section we discuss the state of the
research on this topic.
1. Jet Spaces
Manifolds and maps between manifolds are C . All morphisms of fibred manifolds (and
hence bundles) will be morphisms over the identity of the base manifold, unless otherwise
specified. In particular, when speaking of forms we will always mean C differential
forms.
We recall some basic facts on jet spaces. Our framework is a fibred manifold
: Y X,
with dim X = n, dim Y = n + m, and n, m 1. We have the vector subbundle

def
VY = ker T of T Y , which is made by vectors which are tangent to the fibres of Y .
For 1 r, we are concerned with the r-th jet space J r ; we also set J 0 Y . For
0 s < r we recall the natural fibrings
r,s : J r J s , r : J r X,
and the affine bundle r,r1 : J r J r1 associated with the vector bundle r T X
J r1 V Y J r1 .
Charts on Y adapted to the fibring are denoted by (xi , y ). Latin indices i, j, . . . run
from 1 to n and label base coordinates, Greek indices , , . . . run from 1 to m and label
fibre coordinates, unless otherwise specified. We denote by (/xi , /y ) and (dxi , dy ),
respectively, the local bases of vector fields and 1-forms on Y induced by an adapted chart.
Finite Order Variational Sequences: A Short Review 101
We denoteP(symmetrized) multi-indices by capital letters: I = (i1 , . . . , in ) Nn . We

def def
also set |I| = k ik and I! = 1 ! n !. The sum of a multiindex with a Latin index I + i
will denote the sum of I and the multiindex (0, . . . , i, 0, . . . , 0), where 1 is at the i-th entry.
The charts induced on J r are denoted by (xi , yI ), where 0 |I| r and y0 = def
y .
The local vector fields and forms of J r induced by the fibre coordinates are denoted by
(/yI ) and (dyI ), 0 |I| r, 1 i m, respectively.
An r-th order (ordinary or partial) differential equation is, by definition, a submanifold
S J r .
We denote by jr s :X J r the jet prolongation of a section s : X Y and by
J f : J r J r the jet prolongation of a fibred morphism f : Y Y over a diffeo-
r
morphism f: X X. Any vector field : Y T Y which projects onto a vector field

: X T X can be prolonged to a vector field r : J r T J r by prolonging its flow;
its coordinate expression is well-known (see, e.g., [5, 37]).
The fundamental geometric structure on jets is the contact distribution (or Cartan dis-
tribution) C r T J r . It is the distribution on J r generated by all vectors which are
tangent to the image jr s(X) J r of a prolonged section jr s. It is locally generated by
the vector fields

Di = + yI+i , , (1)
x i yI yJ
with 0 |I| r 1, |J| = r. It is easy to show that this distribution is not involutive and
does not admit any natural direct summand that complement it to T J r . While the contact
distribution has an essential importance in the symmetry analysis of PDE [5], in this context
the dual concept of contact differential forms plays a central role.
Let us denote by Fr the sheaf of smooth functions on J r . We denote by kr the sheaf
of k-forms on J r . We denote by r the sheaf of forms of any degree on J r .
Definition 1.1. We say that a form kr is a contact k-form if
(jr s) = 0
for all sections s of . We denote by C 1 kr the sheaf of contact k-forms on J r . We denote

by C 1 r the sheaf of contact forms of any degree on J r .
Note that if k > n then every form is contact, i.e., C 1 kr = kr .
It is obvious from the commutation of d and pull-back that dC 1 kr C 1 k+1 r . More-
over, it is obvious that C r is a sheaf of ideals (with respect to the exterior product) in r .
1
Unfortunately, C 1 r does not coincide with the ideal generated by 1-forms which annihilate
the contact distribution (for this would contradict the non-integrability). More precisely, the
following lemma can be easily proved (see, e.g., [23]).
Lemma 1.2. The sheaf C 1 1r is locally generated (on Fr ) by the 1-forms
I = dyI yI+i

dxi ,
def
0 |I| r 1.
The above differential forms generate an ideal of r . However, such an ideal is not
differential, hence it does not coincide with C 1 r . To realize it, the following formula can
be easily proved
dI = I+i
dxi , (2)
102 Raffaele Vitolo
from which it follows that, when |I| = r 1, then dI , which is a contact 2-form, cannot
be expressed through the 1-forms of lemma 1.2 because I+i contains derivatives of order
r + 1.
The following theorem is an important achievement by Krupka. It has been first con-
jectured in [9] (C 1 -hypothesis), then proved in [23, 24].
Theorem 1.3. Let k 2. The sheaf C 1 kr is locally generated (on Fr ) by the forms
I , dJ , 0 |I| r 1, |J| = r 1.
We can consider forms which are generated by p-th exterior powers of contact forms.
More precisely, we have the following definition.
Definition 1.4. Let p 1. We say that a form kr is a p-contact k-form if it is generated
by p-th exterior powers of contact forms.
We denote by C p kr the sheaf of p-contact k-forms on J r .
We denote by C p r the sheaf of p-contact forms of any degree on J r .
Finally, we set C 0 r =
def
r .
In other words, C p r is the p-th power of the ideal C 1 r in r . Of course, a 1-contact
form is just a contact form. We have the obvious inclusion
C p+1 r C p r .
It follows that C p+1 r is a sheaf of ideals of C p r , hence of r . Moreover, dC p+1 r

C p+1 r .
Now, we would like to introduce a tool to extract from a form kr the non-trivial
part (to the purposes of calculus of variations). In other words, we would like to introduce
a map whose kernel is precisely the set of contact forms. Such forms yield no contribution
to action-like functionals (see Remark 1.10). First of all, we observe that eq. (2) and Theo-
rem 1.3 suggest that such a map can be constructed if we allow it to increase the jet order by
1. More precisely, it can be easily proved that the contact 1-forms I , with 0 |I| r 1
generate a natural subbundle Cr J r J r1 T J r1 T J r [46]. We have the
following lemma (see [32, 37]).
Lemma 1.5. We have the splitting

r+1 r r+1
J T J = J T X Cr+1 , (3)
Jr X J r+1
with projections
Dr+1 : J r+1 T X T J r , r+1 : J r+1 T J r V J r ,

X Jr
and coordinate expression

Dr+1 = dxi Di = dxi + y
I+i ,
xi yI

r+1 = I = (dyI yI+i

dxi ) .
yI yI
Note that the above construction makes sense through the natural inclusions V J r
T J r and J r+1 X T X J r+1 J r T J r , the latter being provided by T r .
From elementary multilinear algebra it turns out that we have the splitting
M
r+1 k r r+1 q
J J r T J = J T X p Cr+1 .
X J r+1
p+q=k
Now, we observe that a form kr fulfills

r+1,r () : J r+1 k T J r k T J r+1 ,
where the inclusion is realized through the map T r+1,r . Hence, r+1,r
() can be split
into k + 1 factors which, respectively, have 0 contact factors, 1 contact factor, . . . , k contact
factors. More precisely, let us denote by Hrq the set of q-forms of the type
: J r q T X.
We have the following proposition (for a proof, see [23, 46, 48]).
Proposition 1.6. We have the natural decomposition
C p pr+1 Hr+1
q
M

r+1,r (kr ) ,
p+q=k
with splitting projections prp,q : kr C p pr+1 Hr+1

q
defined by

p,q p+q p q
pr () = iDr+1 ir+1 r+1,r ,
q
where iDr+1 , ir+1 stand for contractions followed by a wedge product.
Note that the above maps prp,q are not surjective. See [46] for more details.
Definition 1.7. We say the horizontalization to be the map
hp,q : C p p+q
r C p pr+1 Hr+1
q
, 7 prp,q ().
We denote by
p,q
def p,q
r = h (C p p+q
r ) (4)
0,q
the image of the horizontalization; we say an element r to be a horizontal form.
Probably the first occurrence of horizontalization is in [22]. Of course, horizontalization
is just the projection on forms which have no contact factors. Note that, if q > n, then
horizontalization is the zero map. In coordinates, if 0 < q n, then
= I11I 1 h
...h ih+1 iq dyI1 dyIh dx
h ih+1
dxiq
and the coordinate expression of the horizontalization is
h0,q () = yI11+i1 yIhh+ih I11I i1 iq

...h ih+1 iq dx dx ,
h
(5)
104 Raffaele Vitolo
where 0 h q. The coordinate expressions of hp,q can be obtained in a similar way

(see [3, 23, 24, 46]).
Note that if n > 1 then the above form is not the most general polynomial in (r + 1)-st
derivatives, even if q = 1. For q > 1 the skew-symmetrization in the indexes i1 ,. . . , ih
yields a peculiar structure in the polynomial, in which the sums of all terms of the same
degree are said to be hyperjacobians. Finally, we observe that if n = 1 then the horizontal-
ization is surjective on the space of forms with affine coefficients with respect to r + 1-st
derivatives [25].
The technical importance of horizontalization is in the next two results.
Lemma 1.8. Let qr , with 0 q n, and s : X Y be a section. Then
(jr s) () = (jr+1 s) (h0,q ())
Proposition 1.9. Let p 0. The kernel of hp,q coincides with p + 1-contact q-forms, i.e.,
C p+1 q = ker hp,q .
For a proof of both results, see, for example, [48].
The above decomposition also affects the exterior differential. Namely, the pull-back of
the differential can be split in two operators, one of which raises the contact degree by one,
and the other raises the horizontal degree by one. More precisely, in view of proposition 1.6
and following [37], we introduce the maps

iH : kr kr+1 , iH = iDr+1 r+1,r , (6a)

iV : kr kr+1 , iV = ir+1 r+1,r . (6b)
The maps iH and iV are two derivations along r+1,r of degree 0. Together with the exterior
differential d they yield two derivations along r+1,r of degree 1, the horizontal and vertical
differential
def
dH = iH d d iH : kr kr+1 ,
def
dV = iV d d iV : kr kr+1 ,
It can be proved (see [37]) that dH and dV fulfill the properties
d2H = d2V = 0, dH dV + dV dH = 0, (7a)
dH + dV = (rr+1 ) d, (7b)

(jr+1 s) dV = 0, d (jr s)
= (jr+1 s) dH . (7c)
The action of dH and dV on functions f : J r Y R and oneforms on J r Y uniquely
characterizes dH and dV . We have the coordinate expressions

i f f
dH f = Di f dx = + yI+i dxi ,

(8a)
xi yI
dH dxi = 0, dH dyI = dyI+i

dxi , dH I = I+i

dxi , (8b)
f
dV f = I , (8c)
yI
dV dxi = 0 , dV dyI = dyI+i

dxi , dV I = 0. (8d)
We note that dH dyI = dH I .
Remark 1.10. A form nr defines an action functional

Z
A(s, U ) = (jr s) ,
def
(9)
U
where U X is any oriented n-dimensional submanifold of X with regular boundary.

This is slightly more general than the usual notion, where a horizontal form of the type
: J r n T X is used (see, e.g., [37]). It follows that contact forms yield no contribu-
tion to action-like functionals. The definition (9) is a first motivation for the computations
of the above section.
2. Finite Order Variational Sequence

The first statement of a partial version of finite order variational sequence was in [3]. This
finite order variational sequence stopped with a trivial projection to 0 just after the space
of finite order source forms (see below). The local exactness of this sequence was proved,
together with an original solution of the global inverse problem (despite the fact that in
order to do that the authors used infinite order jets). For more detailed comments about that
variational sequence see remark 2.8.
The first formulation of a (long) variational sequence on finite order jet spaces is due to
Krupka [23] (see [25] for the case n = 1). Below we will describe the main points of the
approach of [23], and compare it with other approaches.
In [23] a natural exact subsequence of the de Rham sequence on J r is defined. This
subsequence is made by contact forms and their differentials. Then we define the rth
order variational sequence to be the quotient of the de Rham sequence on J r by means of
the above exact subsequence. Local and global results about the variational sequence are
proved using the fact that the above subsequence is globally exact and using the abstract de
Rham theorem.
Let us consider the sheaf of 1-contact forms C 1 r , and denote by (dC p kr )the sheaf
generated by the presheaf dC p kr . We set
qr =
def 1 q
C r + (dC 1 q1
r ) 0 q n,
(10)
p+n
r
def p p+n p p+n1
= C r + (dC r ) 1 p dim J r .
We observe that dC 1 rq1 C 1 qr , so that the second summand of the above first equation
yields no contribution to C 1 qr . The sheaves p+n
r become trivial when p + n > P , where
the value of P is computed in [23] using Theorem 1.3. Moreover, we have the following
property (proved in [23]).
Lemma 2.1. Let 0 k dim J r . Then the sheaves kr are soft sheaves.
We have the following natural soft subsequence of the de Rham sequence on J r
d d ... d d
0 1r 2r Pr 0 (11)
Definition 2.2. The sheaf sequence (11) is said to be the contact sequence.
106 Raffaele Vitolo
Theorem 2.3. The contact sequence is an exact soft resolution of C 1 1r , hence the coho-
mology of the associated cochain complex of sections on any open subset of J r vanishes.
The above theorem is proved in [23] by first proving the local exactness of the contact
sequence and then using standard results from sheaf theory (for which an adequate source
is [50]).
Standard arguments of homological algebra prove that the diagram in Figure 1 (p.106)
is commutative, and its rows and columns are exact.
0 0 0
d d ... d d
0 1r 2r Pr 0
d d d ... d d d
0 R 0r 1r 2r Pr Pr +1 0
EI
E1 E2 EP 1
1r /1r 2r /2r ... Pr /Pr
0 0 0
Figure 1. The r-th order variational bicomplex.
Definition 2.4. The diagram in Figure 1 is said to be the r-th order variational bicom-
plex associated with the fibred manifold : Y X. We say the bottom row of the
above diagram to be the r-th order variational sequence associated with the fibred man-
ifold : Y X.
Due to theorem 2.3 the finite order variational sequence is an exact sheaf sequence (this
means that the sequence is locally exact, [50]). Hence both the de Rham sequence and the
variational sequence are acyclic resolutions of the constant sheaf R (acyclic means that the
sequences are locally exact with the exception of the first sheaf R). Next corollary follows
by the abstract de Rham theorem.
Corollary 2.5. The cohomology of the variational sequence is naturally isomorphic to the
de Rham cohomology of J r .
The above finite order diagram yields a variational sequence which can be proved to
be equal to the finite order variational sequence obtained from a finite order analogue of
the C-spectral sequence [49]. Moreover, as one could expect, for 0 s < r pull-back via
r,s yields a natural inclusion of the s-th order variational bicomplex into the r-th order
variational bicomplex. More precisely, we have the following lemma (see [23]).
Lemma 2.6. Let 0 s < r. Then we have the injective sheaf morphism

rs : ks /ks kr /kr , [] 7 [r,s ].
Hence, there is an inclusion of the sth order variational bicomplex into the rth order
variational bicomplex. The inclusion commutes with the operators of the variational bi-
complexes of orders s and r.
Having already dealt with local and global properties of the r-th order variational se-
quence, we are left with the problem of representing the quotient sheaves. This problem
has been independently solved by many authors in the infinite order case. We recognize
two different approaches to the problem: with differential forms (see for example [41, 42])
and with differential operators [43, 44]. The restriction to finite order jets of the former ap-
proach has been developed in [46] for p = 1, p = 2, and in [20, 21] for all p. See [49] for a
finite order differential operator approach. We will describe the differential forms approach.
First of all, it is obvious that, for 0 q n, horizontalization provides such a repre-
sentation (see [23, 46]).
Proposition 2.7. Let 0 q n. Then we have the isomorphism

0,q
Hq : qr /qr r , [] 7 h0,q ().
The quotient differential Eq reads through the above isomorphism as
Hq+1 (Eq ([])) = Hq+1 ([d]) = h0,q+1 (d) = dH h0,q ().
The last equality of the above equation is the least obvious, and was first proved in [3].

The proof depends on the fact that Di yI+j
= yI+j+i , and that the indexes i, j are skew-
symmetrized in the coefficients of dH h () (see the coordinate expression of h0,q ).
0,q
Remark 2.8. In [3] the finite order variational sequence is developed starting from the idea
of finding a subsequence of forms whose order do not change under dH . The authors prove
that the above property characterizes the forms which are in the image of h0,q (see also [2]).
Conversely, in [23] the idea is to start with forms on finite order jets, but the result is the
same up to the degree q = n.
When the degree of forms is greater than n we are able to provide isomorphisms of the
quotient sheaves with other quotient sheaves made with proper subsheaves. This helps both
to the purpose of representing quotient sheaves and to the purpose of comparing the current
approach with others, as we will see.
Proposition 2.9. Let p 1. The horizontalization hp,n induces the natural sheaf isomor-
phism
p,n
Hp+n : p+n
r /p+n
r r /hp,n ((dC p p+n1
r )), [] 7 [hp,n ()].
The quotient differential Ep+n reads through the above isomorphism as
Hp+1+n (Ep+n ([])) = Hp+1+n ([d]) = [hp+1,n (d)].

108 Raffaele Vitolo
For a proof, see [46, 48].

Following [2, 41, 42], let us introduce the map
1
Ip : C p pr Hrn C p p2r H2r
n
, Ip () = (1)|I| DI i/uI (12)
p
where DI stands for the iterated Lie derivative (LD1 )i1 (LDn )in . We say the map Ip to
be the interior Euler operator. It can be proved [2, 20, 42] that the following properties of
Ip hold
- Ip is a natural map, i.e., LX 2r (Ip ()) = Ip (LX r ()), hence Ip is a global map;
- if C p pr Hrn then there exists a unique form C p p2r H2r

n , which is of the
p n1
type = dH with C p 2r1 H2r1 , such that
= I() + . (13)
Remark 2.10. The above form is not uniquely defined, in general. For p = 1, if the order
of is 1 it is easily proved that is uniquely defined; if the order of is 2 then there exists
a unique fulfilling a certain intrinsic property; if the order is 3 it is proved in [16, 17] that
no natural of the above type exists. However, suitable linear connections on M and on
the fibres of : E M can be used to determine a unique . See [1, 2] for the case of
p > 1.
It follows from the above theorem that if C p p2r1 H2r1
n1
then Ip (dH ) = 0, so
2
that Ip = Ip .
Theorem 2.11. We have the isomorphism
p+n
r /p+n
r Vrp , [] 7 Ip (Hp+n ([])),
where Vrp C p p2r+1 H2r+1

n is a suitable subspace (see [46] for a characterization for
p = 1, p = 2).
For a proof, see [46] (p = 1, p = 2) and [20, 21] for any p. The above theorem also
mean that, despite the fact that the denominator in proposition 2.9 is made by forms which
are locally total divergences, only global divergences really matter. We say the elements of
V p to be the p-th degree variational forms; for p = 1 they are also known as source forms.
The map Ip+1 allows us to represent the differentials Ep+n through forms:
Ip+1 (Hp+1+n (Ep+n ([]))) = Ip+1 (Hp+1+n ([d])). (14)
From the coordinate expression of Ip it follows that En is just the EulerLagrange op-
def
erator and E1+n is just the Helmholtz operator. In fact, let = dx1 dxn . Then, if
0,n
r , then = h0,n () = L, where L is a function with polynomial structure in
r + 1-st order derivatives as in (5). Now we can use (14) on , but if is not known the
computational problem of finding it can be technically difficult in principle. On the other
hand, we can use the commutativity of the inclusion of Lemma 2.6 with the operators Ep+n
and consider nr+1 . Then h0,n () = and En () is the standard EulerLagrange
operator on the r + 1-st order Lagrangian . A similar reasoning proves that E1+n coincide
with the Helmholtz operator.
A different, computational approach to the problem of the representation of quotients is
presented in [13, 14].
A further approach to the problem of representation appeared in [30] for the case n = 1.
Here the concept of Lepagean equivalent is introduced in full generality (older version
of this concept can be found e.g., in [22], with references to older foundational works).
Namely, let p+nr . Then a Lepage equivalent of [] p+nr /p+n
r is a differential
p+n
form r such that
hp,n () = hp,n (), hp+1,n (d) = Ip+1 (hp+1,n (d)).
The most important example of a Lepagean equivalent is the PoincareCartan form of a

Lagrangian (see, e.g., [22]).
3. Some Related Problems

In this section we will briefly describe what are the most recent results which involve finite
order variational sequences.
3.1. Variationally Trivial Lagrangians

A variationally trivial Lagrangian is an element [] nr /nr such that En ([]) = 0.
If [] is a variationally trivial Lagrangian, then by the local exactness of the variational
sequence we have h0,n () = dH (h0,n1 ()) with [] n1 r /n1
r a local form. A
n1 n1
global horizontal n 1-form [] r /r such that [] = dH [] exists if and only
if [] induces the zero cohomology class in the variational sequence. A refinement of this
result is the following theorem.
Theorem 3.1. Let : J r n T X induce a variationally trivial Lagrangian []. Then,

locally, = dH , where = h0,n1 () and n1
r1 .
In other words, = h0,n (d), hence is the representative of a class En1 ([]) =
[d] n1 r1
r1 /n1 . This means that depends on r-th order derivatives through hyper-
jacobians. This result has been proved in [3], [4] (here the proof is for the special case
when the Lagrangian does not depend on (xi )), [14, 29] (here the proof uses the finite order
variational sequence). See also [27] for another approach to the problem. Of course, the
result is sharp: the order cannot be further lowered.
3.2. Locally Variational Source Forms

A locally variational source form is an element [] n+1
r /n+1
r such that E1+n ([]) = 0.
If [] is a locally variational source form, then by the local exactness of the variational
sequence [] is the EulerLagrange expression of a local Lagrangian, i.e., [] = En ([])
with [] nr /nr . A global Lagrangian [] nr /nr such that [] = En ([]) exists if
and only if [] = 0 H n+1 (Y ).
110 Raffaele Vitolo
The previous result is sharp with respect to the order [23, 46]. However, it can be very
difficult to check that a source form is in the space n+1 r /n+1
r . A result proved in [2] is
(r)
helpful in this sense. Let y denote all derivative coordinates of order r on a jet space. Let
f C (J 2r ), and suppose that f (xi , y (0) , . . . , y (r) , ty (r+1) , t2 y (r+2) , . . . , tr y (2r) ) is a
polynomial of degree less than or equal to r in y (s) , with r + 1 s 2r. Then f is said to
be a weighted polynomial of degree r in the derivative coordinates of order r + 1 s 2r.
Theorem 3.2. Let [] be a locally variational source form, with : J 2r C0 n T X.

Suppose that the coefficients of are weighted polynomials of degree less than or equal to
r. Then = E(), where : J r n T X.
Again, the result is sharp with respect to the order of the jet space where the Lagrangian
is defined. The above theorem is complemented in [2] by a rather complex algorithm for
building the lowest order Lagrangian. This algorithm is an improvement of the well-known
Volterra Lagrangian Z 1
L= y (xi , tyI )dt
0
for a locally variational source form . In fact, the above Lagrangian is defined on the same
jet space as . The finite order variational sequence yields another method for computing
lower order Lagrangians, provided we know that = [] n+1 r /n+1
r . Namely, we
n+2
apply the contact homotopy operator to the closed form d r , finding n+1 r
such that d = d. Using the (standard) homotopy operator we find nr such that
def 0,n
d = , and = h () is the required Lagrangian. Of course, the most difficult point
is to invert the representation of quotients in the variational sequence, i.e., to find a least
order such that = [].
The above theorem does not exhaust the finite order inverse problem. A locally varia-
tional source form on J 2r seems to have a definite form of the coefficients with respect
to its derivatives of order s, with r + 1 s 2r. It is an open problem to determine such
a structure, e.g. prove that such forms always lie in n+1
s /n+1
s for a minimal value of s; a
least order Lagrangian would follow from the local exactness of the variational sequence.
Finally, we recall that recently some geometric results on variational first-order partial
differential equations have been obtained in [15]. Such equations arise in multisymplectic
field theories.
3.3. Contact Elements

Let Y be an n + m-dimensional manifold, and x Y . We say that two n-dimensional
submanifolds L1 , L2 such that x L1 L2 are r-equivalent if they have a contact of order
r at x. It is possible to choose a chart of Y at x of the form (xi , y ), 1 i n, 1 i m,
where both L1 and L2 can be expressed as graphs y = f1 (xi ), y = f2 (xi ). Then the
contact condition is the equality of the derivatives of the above functions f1 , f2 at x up to
the order r. This is an equivalence relation whose quotient set is J r (Y, n), the manifold
of r-th order n-dimensional contact elements. This construction was first formalized in
[10], and is also known as r-th order jet space of n-dimensional submanifolds of Y [45] or
extended jet bundle [34]). If Y is endowed with a fibring , then J r is the open and dense
subspace of J r (Y, n) which is made by submanifolds which are transverse to the fibring at
a point (which, of course, can be locally identified with the images of sections, hence with
local sections themselves).
Of course, manifolds of contact elements have a contact distribution, hence a variational
sequence can be formulated through the C-spectral sequence [7, 43, 44]. Manifolds of
contact elements can also be seen as jets of parametrizations of submanifolds (i.e., jets of
local n-dimensional immersions) up to the action of the reparametrization group [18]. In
this setting another approach to the variational sequence is [38]. In [33] the finite-order
C-spectral sequence on the manifold of contact elements is computed. Research based on
Krupkas approach on a variational sequence on finite order contact elements is in progress
[31].
Another interesting research topic is the development of finite order variational struc-
tures on differential equations, i.e. submanifolds of jet spaces. This would possibly lead to
a classification of their conservation laws of a certain order [35].
3.4. Variational Sequence and Symmetries

The Lie derivative of variational forms is interesting for the determination of symmetries
of Lagrangians and source forms. However, the result of a Lie derivative with respect to a
prolonged vector field is a form which, in general, contains dH -exact terms. For this reason
it is natural to use a new operator, the variational Lie derivative, which is defined up to
dH -exact terms. Such a formula first appeared in [45] (infinitesimal Stokesformula) in
the infinite order formalism. The finite order case has been dealt with in [11, 28]. See also
[6] for symmetries of source forms which are locally but not globally variational. This topic
has clear connections with Noethers theorem, for which we invite the reader to consult the
above literature.
3.5. Further Topics

We already mentioned that other approaches to variational sequences exist in literature,
mostly on infinite order jets.
It can be proved that there exists an infinite order analogue of Krupkas r-th order
variational bicomplex [47]. This is defined in view of Lemma 2.6 via a direct limit of the
injective family of r-th order variational bicomplexes. Nonetheless the direct limit infinite
order bicomplex will be a bicomplex of presheaves, because gluing forms defined on jets of
increasing order provides forms which are only locally of finite order.
The C-spectral sequence on jets of fibrings yields an infinite order variational sequence
[7, 43, 44]. See [26, 48] for a comparison with Krupkas approach and [49] for some finite
order C-spectral sequence computations.
In [36] the relationship between a part of the finite order variational sequence and the
Spencer sequence are stressed. This relationship was already explored in [43, 44] in the
case of infinite order jet spaces.
Acknowledgements
It is a pleasure for me to acknowledge Professor Demeter Krupka. His hospitality and his
inspiring seminars during my Ph.D. studies have been an invaluable contribution to my
112 Raffaele Vitolo
mathematical education.
This research has been supported by PRIN 2005/2007 Simmetrie e supersimmetrie
classiche e quantistiche, by the section GNSAGA of the Istituto Nazionale di Alta Matem-
atica http://www.altamatematica.it, and by the Dipartimento di Matematica E.
De Giorgi of the University of Salento.
References
[1] R. Alonso Blanco, On the Green-Vinogradov formula, Acta Appl. Math. 72 (1-2)
(2002) 1932.
[2] I. M. Anderson, The Variational Bicomplex (book preprint, technical report of the
Utah State University, 1989); available at http://www.math.usu.edu/fg mp/).
[3] I. M. Anderson and T. Duchamp, On the existence of global variational principles,

American J. of Math. 102 (1980) 781868; http://www.jstor.org/.
[4] J. M. Ball, J. C. Currie and P. J. Olver, Null Lagrangians, weak continuity and varia-
tional problems of arbitrary order, J. Funct. Anal. 41 (1981) 135174.
[5] A. V. Bocharov, V. N. Chetverikov, S. V. Duzhin, N. G. Khor kova, I. S. Kra-

sil shchik, A. V. Samokhin, Yu. N. Torkhov, A. M. Verbovetsky and A. M. Vino-
gradov, Symmetries and Conservation Laws for Differential Equations of Mathemat-
ical Physics ((I. S. Krasil shchik and A. M. Vinogradov, Eds.) Amer. Math. Soc.,
1999).
[6] J. Brajerck and D. Krupka, Variational principles for locally variational forms, J.
Math. Phys. 46 (2005) 052903.
[7] P. Dedecker, On applications of homological algebra to calculus of variations and

mathematical physics, In: Proceedings of the IV international colloquium on differ-
ential geometry (Santiago de Compostela, Universidad de Santiago de Compostela,
1978, Cursos y Congresos de la Universidad de Santiago de Compostela 15) 285
294.
[8] S. V. Duzhin, C-spectral sequence on the manifold J 1 M , Uspekhi Math. Nauk 38

(1983) 165166 (in Russian); English translation: Russian Math. Surveys 38 (1)
(1983) 179181.
[9] S. V. Duzhin, C-spectral sequence on manifolds of jets of finite order, preprint

VINITI UDK 514.763.8 (1982) (in Russian).
[10] C. Ehresmann, Les prolongements dune variete differentiable. IV. Elements de con-
tact et elements denveloppe, C. R. Acad. Sci. Paris 234 (1952) 10281030.
[11] M. Francaviglia, M. Palese and R. Vitolo, Symmetries in finite order variational se-
quences, Czech. Math. J. 52 (127) (2002), 197213; http://poincare.unile.it/vitolo.
[12] I. M. Gel fand and L. A. Dikii, Asymptotic behaviour of the resolvent of Sturm
Liouville equations and the algebra of the KortexegDe Vries equations, Uspekhi
Mat. Nauk 30 (5) (1975) 67100 (in Russian); Russian Math. Surveys 30 (5) (1975)
77113 (English translation).
[13] D. R. Grigore, The Variational Sequence on Finite Jet Bundle Extensions and the
Lagrangian Formalism, Diff. Geom. Appl. 10 (1999) 4377.
[14] D. R. Grigore, Variationally trivial Lagrangians and locally variational differential

equations of arbitrary order, Diff. Geom. Appl. 10 (1999) 79105.
[15] A. Hakova and O. Krupkova, Variational first-order partial differential equations, J.

Diff. Equat. 191 (2003) 6789.
[16] I. Kolar, A geometrical version of the higher order Hamilton formalism in fibred
manifolds, J. Geom. Phys. 1 (2) (1984) 127137.
[17] I. Kolar, Natural operators related with the variational calculus, In: Proc. Conf. Diff.
Geom. Appl. 1992 (Silesian Univ. Opava, 1993) 461472.
[18] I. Kolar, P. Michor and J. Slovak, Natural Operations in Differential Geometry

(Springer-Verlag, 1993); http://www.emis.de/monographs/index.html.
[19] I. S. Krasil shchik and A. M. Verbovetsky, Homological methods in equations of

mathematical physics, Open Education and Sciences, Opava (Czech Rep.) 1998;
arXiv:math.DG/9808130.
[20] M. Krbek and J. Musilova, Representation of the variational sequence by differential

forms, Rep. Math. Phys. 51 (2/3) (2003) 251258.
[21] M. Krbek and J. Musilova, Representation of the variational sequence by differential

forms, Acta Appl. Math. 88 (2) (2005) 177199.
[22] D. Krupka, Some geometric aspects of the calculus of variations in fibred mani-
folds, Folia Fac. Sci. Nat. UJEP Brunensis, Brno University, 14 (1973); arXiv:math-
ph/0110005.
[23] D. Krupka, Variational sequences on finite order jet spaces, In: Proc. Conf. Diff.
Geom. Appl. 1989 (World Scientific, New York, 1990) 236254.
[24] D. Krupka, The contact ideal, Diff. Geom. Appl. 5 (1995) 257276.
[25] D. Krupka, Variational sequences in mechanics, Calc. Var. 5 (1997) 557583.
[26] D. Krupka, Variational sequences and variational bicomplexes, In: Differential Ge-
ometry and Applications (Proc. Conf., (I. Kolar, O. Kowalski, D. Krupka and J.
Slovak, Eds.) Brno, August 1998, Masaryk Univ., Brno, Czech Republic, 1999) 525
531; http://www.emis.de/proceedings.
[27] D. Krupka, The total divergence equation, Lobachevskii Journal of Mathematics 23

(2006) 7193.
114 Raffaele Vitolo
[28] D. Krupka, O. Krupkova, G. Prince and W. Sarlet, Contact symmetries and varia-
tional sequences, In: Differential Geometry and Its Applications (Proc. Conf., (J.
Bures, O. Kowalski, D. Krupka and J. Slovak, Eds.) Prague, August 2004, Charles
University in Prague, Czech Republic, 2005) 605615.
[29] D. Krupka and J. Musilova, Trivial Lagrangians in Field Theory, Diff. Geom. and
Appl. 9 (1998) 293505.
[30] D. Krupka and J. Sedenkova, Variational sequences and Lepage forms, In: Differen-
tial Geometry and Its Applications (Proc. Conf., (J. Bures, O. Kowalski, D. Krupka
and J. Slovak, Eds.) Prague, August 2004, Charles University in Prague, Czech Re-
public, 2005) 617627.
[31] D. Krupka and Z. Urban, Differential invariants and higher order Grassmann bundles,
In: Differential Geometry and its Applications (Proc. 10th Int. Conf. on Diff. Geom.
and Appl., Olomouc 2007, World Scientific, Singapore, 2008) 463473.
[32] L. Mangiarotti and M. Modugno, Fibered Spaces, Jet Spaces and Connections for
Field Theories, In: Proc. of the Int. Meeting on Geometry and Physics (Pitagora
Editrice, Bologna, 1983) 135165.
[33] G. Manno and R. Vitolo, Variational sequences on finite order jets of subman-
ifolds, In: Proc. of the VIII Conf. on Diff. Geom. and Appl. (Opava 2001,
Czech Republic, http://www.emis.de/proceedings); see also the longer preprint
[34] P. J. Olver, Applications of Lie Groups to Differential Equations (GTM 107, 2nd
edition, Springer 1993).
[35] P.J. Olver, private communication (2007).
[36] J. F. Pommaret, Spencer sequence and variational sequence, Acta Appl. Math. 41
(1995) 285296.
[37] D. J. Saunders, The Geometry of Jet Bundles (Cambridge Univ. Press, 1989).
[38] D. J. Saunders, Homogeneous variational complexes and bicomplexes;
[39] F. Takens, A global version of the inverse problem of the calculus of variations, J.
Diff. Geom. 14 (1979) 543562.
[40] W. M. Tulczyjew, Sur la differentielle de Lagrange, C. R. Acad. Sc. Paris, serie A
280 (1975) 12951298.
[41] W. M. Tulczyjew, The Lagrange Complex, Bull. Soc. Math. France 105 (1977) 419
431.
[42] W. M. Tulczyjew, The EulerLagrange Resolution, In: Internat. Coll. on Diff. Geom.
Methods in Math. Phys. (AixenProvence, 1979; Lecture Notes in Mathematics, n.
836, SpringerVerlag, Berlin, 1980) 2248.
[43] A. M. Vinogradov, On the algebrogeometric foundations of Lagrangian field theory,

Soviet Math. Dokl. 18 (1977) 12001204; http://diffiety.ac.ru/.
[44] A. M. Vinogradov, A spectral sequence associated with a nonlinear differential

equation, and algebrogeometric foundations of Lagrangian field theory with con-
straints, Soviet Math. Dokl. 19 (1978) 144148; http://diffiety.ac.ru/.
[45] A. M. Vinogradov, The CSpectral Sequence, Lagrangian Formalism and Conserva-

tion Laws I and II, J. Math. Anal. Appl. 100 (1) (1984); http://diffiety.ac.ru/.
[46] R. Vitolo, Finite order variational bicomplexes, Math. Proc. of the Camb. Phil. Soc.
125 (1998) 321333; http://poincare.unile.it/vitolo; a more detailed version is avail-
able at arXiv:math-ph/0001009.
[47] R. Vitolo, A new infinite order formulation of variational sequences, Arch. Math. Un.
Brunensis 4 (34) (1998) 483504; http://poincare.unile.it/vitolo.
[48] R. Vitolo, On different geometric formulations of Lagrangian formalism,

Diff. Geom. and its Appl. 10 (3) (1999) 225255; updated version at
http://poincare.unile.it/vitolo.
[49] R. Vitolo, Finite order formulation of Vinogradov Cspectral sequence, Acta Appl.
Math. 70 (1-2) (2002) 133154; updated version at http://poincare.unile.it/vitolo.
[50] R. O. Wells, Differential Analysis on Complex Manifolds (GTM, n. 65, Springer

Verlag, Berlin, 1980).
Chapter 7
C ONCATENATING VARIATIONAL P RINCIPLES

AND THE K INETIC S TRESS -E NERGY-M OMENTUM
T ENSOR
Marco Castrillon Lopez1, Mark J. Gotay2 and Jerrold E. Marsden3
1
Departamento de Geometra y Topologa, Facultad de Ciencias Matematicas,
Universidad Complutense de Madrid, 28040 Madrid, Spain,
2
Department of Mathematics, University of Hawaii,
Honolulu, Hawaii 96822, USA,
3
Control and Dynamical Systems 107-81, California Institute of Technology,
Pasadena, California 91125, USA
Abstract
We show how to concatenate variational principles over different bases into one
over a single base, thereby providing a unified Lagrangian treatment of interacting
systems. As an example we study a KleinGordon field interacting with a mesically
charged particle. We employ our method to give a novel group-theoretic derivation of
the kinetic stress-energy-momentum tensor density corresponding to the particle.
1. Introduction and Setup

Let us recall the geometric setting of a classical variational principle [3]: We are given a
fibration Y X, with dim X = n + 1, and we wish to extremize an action of the form
Z
S() = L(j 1 )
X
where : X Y is a section and L : J 1 Y n+1 X is a specified Lagrangian density.1

E-mail address: mcastri@mat.ucm.es

E-mail address: gotay@math.hawaii.edu

E-mail address: marsden@cds.caltech.edu
1
For simplicity we consider only first order theories. We also ignore technical issues and proceed formally.
118 Marko Castrillon Lopez, Mark J. Gotay and Jerrold E. Marsden
One commonly encounters several (say K) such variational principles simultaneously,

for instance when one studies the Newtonian dynamics of a swarm of charged particles
(in a background electromagnetic field), or the interaction between Dirac and YangMills
fields. In the cases cited, the relevant fibrations have the form Yi X for the ith variational
principle; the key point being that each fibration has the same base X. To combine these
variational principles into a single principle is a straightforward matter: one builds the fiber
product Y1 X X YK X, and then on the first jet of this bundle one takes as the
Lagrangian density an expression of the form L1 + + LK + Lint for some interaction
terms Lint .
It is less clear how to deal with variational principles with disparate bases, that is, fibra-
tions Yi Xi in which the Xi are all different. A simple example is a nucleon moving in a
dynamic KleinGordon field. (Here the configuration bundle for the nucleon is X R R,
where X is 4-dimensional spacetime and R is the material world line of the nucleon. The
fibration for the KleinGordon field is R X X, sections of which are scalar fields on
spacetime.) Even if the bases are identical, it may be desirable to distinguish them. This
is the case, for instance, in relativistic multiparticle systems (cf. [1]), when one wants to
parametrize each particles trajectory by its own proper time, as opposed to a single uni-
versal time.
In this context of disparate bases, one standard way to proceed is as follows. Construct
an action functional using sections i : Xi Yi for the ith bundle by setting
K Z
X Z
1
S(i , . . . , K ) = Li (j i ) + Lint (j 1 1 , . . . , j 1 K ). (1.1)
i=1 Xi X1 XK
Then varying these fields i , one obtains the EulerLagrange equations for the problem.
See equation (2.1) for a specific example.
However, while producing the EulerLagrange equations, this approach has the unsat-
isfactory feature of not yielding a field theory in the usual sense, in which the fields are
sections of a single bundle and which has a well-defined Lagrangian density. This or some
other formalism is needed if one wishes to tap into the machinery of multisymplectic ge-
ometry, multimomentum maps, stress-energy-momentum (SEM) tensors, and constraint
theory, etc.
To concatenate variational principles with disparate bases in such a way as to recap-
ture a genuine field theory, we proceed as follows. To begin, construct the product bun-
dle Y1 YK X1 XK , which we denote Y X for short. In agree-
ment with experience we restrict attention to product sections of this bundle of the form
= (1 , . . . , K ), where each i is a section of Yi Xi . With = (1 , . . . , K ) such a
section,
j 1 (x) = j 1 1 (x1 ), . . . , j 1 K (xK )

where x = (x1 , . . . , xK ). Denote by J1 Y the subbundle of J 1 Y consisting of all such jets;

equivalently, J1 Y = J 1 Y1 J 1 YK .
Given Lagrangian densities Li on the jet bundles J 1 Yi , it is simple enough to lift them
to maps, still denoted by Li , on the concatenated jet bundle J1 Y by composing with pro-
jections:
j 1 (x) 7 Li j 1 i (xi ) .

Concatenating Variational Principles 119
But how do we concatenate these Li into a single Lagrangian density? Even ignoring
interaction terms, we cannot just add the Li as they take values in different spaces, viz.
ni +1 Xi and so need not be forms of equal rank. The trick is to suspend the Li : J1 Y
ni +1 Xi to maps J1 Y N +K X, where N = n1 + +nK , by inserting suitable tensor
densities in the Li to even out their ranks in the target.
First, we pull Li back via the projection X Xi to an (ni + 1)-form on X. Sec-
ond, for each i choose scalar densities Di of weight 1 on X1 Xci XK . Now in
n +1
Li = Li d i xi the coefficient Li transforms as a scalar density of weight 1 on Xi , so the
coefficient in
Li := Li d ni +1xi Di d n1 +1x1 d\
ni +1x d nK +1x = L D d N +Kx
i K i i
will also transform as a scalar density of weight 1 on X under the subgroup
Diff(X1 ) Diff(XK ) Diff(X)
(which is sufficient for our purposes). The densities Di are to be chosen by hand, depending
on the precise structure of the system; see the examples in 2. and 3.. Thus modified, we
may assemble L1 + + LK into a map
L : J1 Y N +K X.
Interaction terms, which are typically defined over several of the bases Xi (again, see the
following examples) are treated similarly. Finally, it is straightforward to deal with com-
posite situations in which some of the bases are identical and others are not.
Ultimately, the specific choice of the Di will not matter as long as
Z Z
N +K
Li Di d x= Li d ni +1x
X Xi
for each i, that is, the concatenated action reduces to the original action. Specifically, this
means that Z
L(j 1 ) = S()
X
where the right hand side is given by (1.1). In particular, the EulerLagrange equations
remain unaltered when the Lagrangian L is used in place of the action functional (1.1).
Once we have a total Lagrangian density in hand (albeit possibly a distributional one),
we may proceed in the usual fashion. Thus we may compute the equations of motion
and various geometric objects, such as SEM tensors. To extract physical information from
these objects, however, it will normally be necessary to project them from X to some Xi
or products thereof; this projection is accomplished by integration over the remaining Xj .
Rather than continuing to try to describe the procedure in generality, it is more instruc-
tive to illustrate it via a simple example. (It really is easier done than said!)
In 2. we apply this method to a system consisting of a KleinGordon field interacting
with a mesically charged particle. (Think of a pion field interacting with a nucleon.) Beyond
illustrating concatenation, this example has interesting features which are worth elucidating.
In particular, we study the SEM tensor density of this system. Its computation, following
[5], is interesting in that it naturally produces the Minkowski, or kinetic, SEM tensor for
a moving particle as a matter of course. To our knowledge, this SEM tensor has never
been derived via a Lagrangian from first principles; it has always been inserted into the
formalism in an ad hoc manner. An important point therefore is that our method is not
merely a tidy means of packaging variational principles; it is capable of providing, in an
entirely straightforward fashion, quantities which otherwise cannot be obtained except in
makeshift ways.
Finally in 3. we briefly indicate some other contexts in which our results should be
useful.
2. Motion of a Mesically Charged Particle in a KleinGordon

Field
Let X be an oriented spacetime with metric G. We consider a real KleinGordon field
: X R of mass M interacting with a particle of mass m and mesic charge . The
particles trajectory in spacetime (or placement field) is z : R X. The base for
the system is thus X R, the second factor being thought of as a time axis,2 and the
configuration bundle Y is then
(R X) (X R) X R
with coordinates (, X a ) on the fiber and (x , ) on the base. We set z a = X a z.

Our presentation is based upon the excellent exposition in Chapter 8 of [1], to which
we refer the reader for further information. The action (1.1) for the system in this case is
usually written
1
Z p
S(, z) = G (x), (x), (x) M 2 (x)2 G(x) d 4x
X 2
Z Z
4 4
(x)kz()k (x z()) d x d mkz()k d, (2.1)
XR R
p
where the dot denotes differentiation with respect to and kzk = Gab z a z b . Observe
that the bases for the free KleinGordon term and the free particle term are different, and
that the interaction term in the middle lives on the product of these.
Before proceeding, there are two technical issues that need to be resolved, stemming
from the presence of the two factors of X in the configuration bundle. First, note that
in the leading term of S, G is regarded as living on the X in the base, while in the last
term it evidently resides on the X in the fiber. It is necessary to know precisely where G
lives, as this has an effect on the subsequent analysis: if on the base, then G is treated as a
field, while if on the fiber it is simply thought of as a geometric object. We reconcile these
two interpretations by taking G to be anchored to the base, and then pulling it back to the
fiber by means of the following construction.3,4 Introduce yet another factor of X in the
2
Not necessarily proper time.
3
This is a variant of the Kuchar method of parametrizing a classical field theory; see [6] and [2] for details.
4
At the end of this section we will briefly examine what happens if instead we anchor G to the fiber.
fiber along with diffeomorphisms : X X, viewed as sections of X X X, with

corresponding configuration and multivelocity variables a = X a and a = (X a
)/x , respectively. (We can, and do, regard the two copies of X in the fiber as identical.)
We use these auxiliary nondynamic fields, the diffeomorphism fields (or diffields for short,
to (i) identify the copies of X in the fiber with that in the base, and (ii) endow the new copy
of X in the fiber with the metric g = G with components
gab = G a b ,
where ( a ) = ( a )1 . All this is summarized in the following figure.
The general set up for the introduction of diffeomorphism fields.
Second, the delta function 4 (x z()) must be modified, as it compares elements

x in the base with elements z() in the fiber. As just indicated we can use the diffields
fields to remedy this problem as well: we need only write 4 x 1 (z()) instead. It is
sometimes convenient to replace
4 x 1 (z()) = 4 ((x) z())(det )

(2.2)
using the properties of delta functions (cf. the Appendix), where is the Jacobian of
. From this we see that 4 x 1 (z()) (i) is a scalar density on X (again, see the
Appendix), and (ii) depends upon the spacetime derivatives of , even though this is not
obvious at first glance. The reason we do not insist on a fixed identification of the base X
with the fiber X, and instead allow a variable identification by means of the diffields, is to
allow some gauge freedom in the fields; see also footnote 8 below.
Remark. Analogous fields , called covariance fields,, are introduced in [6] and [2], but
there they have a different purpose, namely, to make a field theory on a given background
generally covariant and in doing so, they are introduced as dynamic fields.
In addition to the diffields , we introduce a (positive-definite) metric K on R as a

nondynamic field. We suppose that K is chosen so that R has metric volume 1.
With these fixes we may R now concatenate the three action terms over the composite
4
base X R as S(, z) = L d x d, with the Lagrangian
1
L(x , , , , , z b , z b ; a , a , G , K) = G , , M 2 2 G K
2
4

m + kzk ( z)(det ). (2.3)
Notice that the interaction term itself needs no essential modification, as the corresponding
term in (2.1) is already an integral over X R,but informs our choice of scalar density
in the free particle term, viz. 4 x 1 (z()) , when we suspend the latter to X R.
We have also written this delta function in the form (2.2) to make it clear that L is defined
pointwise.

Remark. The choice of R scalar density D = K in the KleinGordon term is hardly unique;
all we require is that R D d = 1. For instance, we could instead take () for D with no
essential difference.
As evident from (2.3), the modified configuration bundle is taken to be
Y = Y X X 2 X Lor(X) R Riem(R),
where we abbreviate the bundle X X X by X 2 , Lor(X) is the bundle whose sections
are Lorentz metrics on X and, similarly, Riem(R) is the bundle whose sections are Rie-
mannian metrics on R. However, in our approach and z are variational, while , G and
K are nondynamic fields. As per the above, we now regard
q
kzk = G a b z a z b .
Remark. Occasionally, as in [9], one encounters what one might call noncovariant concate-
nations. In the current example, this amounts to writing the terms in the action as integrals
over X alone and is effectively accomplished by imposing the coordinate condition x0 = .
As this procedure is not covariant, it can lead to problems [10].
We compute the EulerLagrange equations. Varying with respect to and employing
(2.2), we obtain
M 2 (x) G(x) K() kz()k 4 x 1 (z())
p p
p
G , G (x) K() = 0.
Integrating with respect to , using the fact that volK (R) = 1, and rearranging, this reduces
to the KleinGordon equation
+ M 2 = (2.4)
where denotes the G-covariant derivative and
Z
12
kz()k 4 x 1 (z()) d

(x) = (G)
R
is the source density.

Similarly, varying with respect to z and employing (2.2) yield
h 4 1
i
m + (x) kz()k x (z())
z a
gab (z())z b ()

1

+ m + (x) x (z()) = 0.
kz()k
Carrying out the differentiation and then integrating over X, some manipulations give
" #
d g (z()) z b ()
ab
m + 1 (z())

d kz()k
= a , 1 (z()) kz()k

gbc,a (z())z b ()z c ()

1
+ m + (z()) . (2.5)
2kz()k
To give insight into these equations, note that in the special case when (X, G) is Minkowski
spacetime, = IdX , and is taken to be proper time along the particles world line, these
equations simplify in a global Lorentz frame to
d h i
m + (z()) za () = ,a (z()).
d
This is the mesic analogue of the Lorentz force law in electrodynamics.
Neither K, the G , nor the a have field equations, since they are not variational. Thus
one is free to assign them whatever values one wishes in (2.4) and (2.5). Often, however,
one has specific values of G and in mind, e.g., the given spacetime metric for G and IdX
for .
Turning now to the SEM tensor, let Diffc (X) Diffc (R) (that is, the group of diffeo-
morphisms that are the identity outside a compact set) act on the modified configuration
bundle Y according to

( f ) (x, , , z; , G, K) = (x), f (), , z; , G, f K .
(We assume that all diffeomorphisms are positively oriented.) The Lagrangian density L =
L d 4x d is then visibly equivariant with respect to the induced action on J1 Y , that is,5
L ( f ) j 1 (, z; , G, K = ( f ) L j 1 (, z; , G, K) .

We may thus use equation (3.12) in [5] to compute the 5-dimensional SEM tensor den-
sity
T T 4

T =
T 4 T 44
5
Even though the pointwise action of Diff(X) on the fiber of thediffeomorphism bundle X X X
is taken to be trivial, its action on sections thereof is not: = 1 . Thus the identification of the factor
of X in the base with that in the fiber can fluctuate, which is one of the reasons we allow to be variable in the
first place.
of the interacting system (where x4 = ).6 Integrating over and raising an index, we
project out the spacetime SEM tensor density:
Z

T = T d = t + (m + ) , (2.6)
R
where
1
t = (2G G G G ), , + G M 2 2 G
2
is the canonical SEM tensor density of the (free) KleinGordon field and
z a ()z b () 4
Z

x 1 (z()) d

(x) = a b
R kz()k
is the Minkowski tensor density. (m is then the kinetic SEM tensor density). As well, we
compute T 4 = 0 = T 4 . Finally, we find that when integrated over X, T 4 4 is effectively
the KleinGordon action:

Z
4 1 2 2 4
T 4= G , , M G d x K.
2 X
Remark. The kinetic SEM tensor density is a familiar object in microscopic continuum
mechanics, cf. Chapter 8 of [1] and 33 of [9]. Minkowski [11] originally introduced it
in flat-spacetime electrodynamics in order to recover the continuity equation T , = 0 in
view of the fact that TEM , 6= 0 when currents are present. In the continuum limit of a
noninteracting clutch of particles, goes over to the SEM tensor density for a perfect
fluid as in 9.1-2 of [1]. It is interesting that in this limit, the infinite time integrals in the
kinetic SEM tensor density disappear and one is left with a local tensor density.
To our knowledge, ours is the first genuine derivation of the Minkowski tensor density
from first principles in a variational context, once again illustrating the power of multi-
symplectic geometry in classical field theory and in particular, the usefulness of having a
concatenated theory for which one can make use of concepts such as the SEM tensor.
As we have defined it, the Minkowski tensor density depends upon the diffields as well
as the particle placement field. However, note that when = IdX , reduces to the more
familiar expression
z a ()z b () 4
Z
ab
(x) = (xz()) d.
R kz()k
Remark. Suppose we focus solely on the particle dynamics so that the (original) config-
uration bundle is X R R. The corresponding Lagrangian density mkz()kd is
Diffc (R)-covariant, and so we may compute the corresponding SEM scalar density as in
Example a, Interlude II of [4]. We obtain T = E, the energy of the particle, which
vanishes as the Lagrangian is time reparametrization-invariant. (This is reflected by the
6
Using the product metric G K on X R, one could also compute T via the Hilbert formula (4.2) in
[5]. See also [10].
vanishing of T 4 4 in the 5-dimensional context when there is no KleinGordon field.) Thus

only when the spacetime X is part of the base of the variational principle do we encounter
the kinetic SEM tensor density; it does not appear in standard particle dynamics per se.
To reiterate, even in the absence of other fields, our technique yields yet another (5-
dimensional!) treatment of the relativistic free particle that has the advantage of automati-
cally incorporating the Minkowski tensor.
Remark. Note that the term in (2.6) arises from the interaction of with the mesi-
cally charged particle. This term has no analogue in the electrodynamics of particles; there
we get simply
T = TEM + m .

Charged strings behave similarly, as we show in 3A (cf. equation (3.1)). This can be
traced to the fact that the electromagnetic field is a covector, while the KleinGordon field
is a scalar.
The SEM tensor density T is symmetric. It is also divergence-free, as can be seen
from general principles (cf. Proposition 5 in [5]). One may verify this directly, via a long
calculation.
We end with a discussion of an alternate treatment of this system.
Suppose we consider the physical metric as a geometric object g on the fiber as
opposed to a field on spacetime. Then we would define G = g with components
G = a b gab . Proceeding as in the above, the Lagrangian density would be
L(x , , , , , z b , z b ; a , a , K)
1 ab
= a b g , , M 2 2 g (det ) K
2
m + kzk 4 ( z)(det )

p
where G = g (det ) and kzk = gab (z)z a z b .
Computing the SEM tensor density in this formulation, we obtain T 0 and the
other components as before. That the spacetime components vanish is actually a conse-
quence of the generalized Hilbert formula (3.13) in [5], since the nondynamic fields and
K do not transform under Diff(X). (In the original formulation, the nondynamic metric
G on X does transform under the spacetime diffeomorphism group with the result that
(2.6) is nonzero.) The difference between this SEM tensor density and the previous one
stems from: (i) the spacetime metric no longer being regarded as a field, so that it cannot
contribute to the energy, momentum, and stress content of the system, and (ii) the subtly
different manners in which the diffields appear in the two formulations.
That one can encounter several SEM tensor densities for the same system may seem
surprising, but is unavoidable and can also be regarded as different packaging of the same
information. What the SEM tensor density turns out to be depends upon what the fields are,
whether they are dynamic, and precisely how they appear in the Lagrangian. And even the
size of the SEM tensor density depends upon how the system is formulated! For instance,
for something as simple as a relativistic free particle, we can have a 1 1 SEM tensor
density (which vanishes identically)as noted in a previous remark, or a 5 5 SEM tensor
density (which does or doesnt, depending on where the spacetime metric is anchored). And
in the latter case, the 5 5 object reduces to the 4 4 Minkowski tensor density! Thus how
the system is formulated plays a substantial role insofar as how various quantities, and in
particular the SEM tensor density, are to be understood.
3. Further Examples and Outlook
To conclude we briefly mention some other systems for which our techniques should prove
helpful. We begin by upping the dimension of the matter from 1 to 2, that is, we replace the
particle by a string. For variety, we also replace the mesic interaction by an electromagnetic
one.
Charged Strings. We closely follow the exposition in 2. Let (X, G) and (W =

R B, H = HR HB ) be 4- and 2-dimensional Lorentzian spacetimes, respectively.
worldsheet in X, this being a map z : W X. We use coordinates
We consider a string
x , A = (, ) as coordinates on X W . The configuration bundle Y is correspond-
ingly
1 X W (X W ) X Lor(X) W Lor(W ) X W.
Assume that the string carries a charge density : B R and interacts with a dynamic
electromagnetic field described by a potential 1-form A on X. We also take the metric H
on W to be dynamic; thus we adopt the Polyakov approach as in [7]. The action for this
system is
1
Z p
S(A, z, H) = F (x)F (x) G(x) d 4x
4
ZX
z p
+ A (x)() (, ) 4 (x z(, )) HB () d 4x d d
XW
T
Z p
H AB () G (z())z (),A z (),B H() d 2
2 W
where T is the tension.

As in the case of the meson, we see that z takes values in X, which is the base for the
electromagnetic field. So we need to introduce diffields as before. As well, we take the
spacetime metric to reside on the factor of X in the base. Finally, let K be a nondynamic
Riemannian metric on W with total volume 1. The modified configuration bundle for the
concatenated variational principle is then
Y X X 2 W Riem(W ) X W
and the Lagrangian reads
L(x , , , A , A, , z a , z a ,A , HAB ; a , a , G , KAB )

1
= G G F F G K
4
z a 4 p
+ A a z)(det ) HB

T AB
H G a b z a ,A z b ,B 4 ( z)(det ) H.
2
Now build the product metric G K on X W . Using the Hilbert formula and then
integrating as before, we compute the 6-dimensional SEM tensor density as follows: the
spacetime components are
T = TEM + T

(3.1)
where

1
T
EM

= G F F + G F F G
4
is the free electromagnetic SEM tensor density, and

Z

= a b H AB z a ,A z b ,B 4 x 1 (z()) H d 2
W
is the analogue of the Minkowski tensor density for strings. The extra T A and T A
components are zero and, after integrating over X, the T AB subblock reduces to

Z
AB 1
T = G G F F G d x K AB K.
4
4 X
Continua. Another intriguing example that we intend to pursue in future works is a

charged elastic body, fluid, or plasma, in which one concatenates a continuum with electro-
magnetism on a given background metric spacetime. Such theories will likely have signifi-
cant differences with the particle and string examples presented above. Amongst these dif-
ferences, we expect that, unlike mesically or electrically charged particles, continua should
have well-defined initial value problems (see also the discussion of this point in [1]). Evi-
dence for this can be found in works such as [8] and [12].
One other interesting aspect of a charged elastic body is the following. If B is the
body manifold, then its motion in spacetime is determined by a map z : R B X.
The main difference from our previous two examples is that rather than the delta functions
4 1
((x) z()) we must now use characteristic functions (z(R B)) . We expect
that examples such as this will be key players in the future development of the point of view
given in this paper.
Appendix
Let M be a manifold with coordinates x = (x1 , . . . , xm ). Here we prove that the delta
function m (x x0 ) transforms as a scalar density of weight 1.
Let : M M be a diffeomorphism and f C (X). On the one hand,

Z
(f )(x0 ) = (f )(x) m (x x0 ) d m x.
X
On the other hand, by the change of variables theorem with y = (x),
Z
f ((x0 )) = f (y) m (y (x0 )) d m y
X
Z
= f ((x)) m ((x) (x0 )) |J(x)| d m x.
X
where J is the Jacobian determinant of . Since f is arbitrary the desired result follows
upon comparing these two formul.
References
[1] J. L. Anderson, Principles of Relativity Physics (Academic Press, New York, 1967).
[2] M. Castrillon Lopez, M. J. Gotay and J. E. Marsden, Parametrization and stress-
energy-momentum tensors in metric theories, J. Phys. A:Math. Theor. (2008); to ap-
pear.
[3] M. J. Gotay, J. A. Isenberg, J. E. Marsden and R. Montgomery, Momentum maps and
classical fields, I: Covariant field theory, (1998); arXiv:physics/9801019.
[4] M. J. Gotay, J. A. Isenberg and J. E. Marsden, Momentum maps and classical fields,
II: Canonical analysis of field theories, (2004); arXiv:math-ph/0411032.
[5] M. J. Gotay and J. E. Marsden, Stress-energy-momentum tensors and the Belinfante
Rosenfeld formula, Contemp. Math. 132 (1992) 367391.
[6] M. J. Gotay and J. E. Marsden, Parametrization theory, (2008); in preparation.
[7] M. B. Green, J. H. Schwarz and E. Witten, Superstring Theory, Volume I: Introduction
(Cambridge Univ. Press, Cambridge, 1987).
[8] M. Kunzinger, G. Rein, R. Steinbauer and G. Teschl, On classical solutions of the rela-
tivistic VlasovKleinGordon system, Electronic J. Differential Equations (1) (2005)
pp. 17.
[9] L. D. Landau and E. M. Lifshitz, The Classical Theory of Fields (Fourth revised En-
glish ed., Permagon Press, Aberdeen, 1979).
[10] M. Leclerc, Canonical and gravitational stress-energy tensors, (2006); arXiv:gr-
qc/0510044.
[11] H. Minkowski, Die Grundgleichungen fur die elektromagnetischen Vorgange in be-
wegten Korpern, Nach. Ges. Wiss. Gottingen (1908) 53111.
[12] G. Rein, Generic global solutions of the relativistic VlasovMaxwell system of plasma
physics, Comm. Math. Phys. 135 (1990) 4178.
Chapter 8
A G EOMETRIC H AMILTON -JACOBI T HEORY

FOR C LASSICAL F IELD T HEORIES
Manuel de Leon1, Juan Carlos Marrero2 and David Martn de Diego3

1
Instituto de Ciencias Matematicas, CSIC-UAM-UC3M-UCM,
Serrano 123, 28006 Madrid, Spain
2
Departamento de Matematica Fundamental, Universidad de La Laguna,
La Laguna, Canary Islands, Spain,
3
Instituto de Ciencias Matematicas, CSIC-UAM-UC3M-UCM,
Serrano 123, 28006 Madrid, Spain
Abstract
In this paper we extend the geometric formalism of the Hamilton-Jacobi theory

for hamiltonian mechanics to the case of classical field theories in the framework of
multisymplectic geometry and Ehresmann connections.
2000 Mathematics Subject Classification. 70S05, 49L99.
Key words and phrases. Multisymplectic field theory, Hamilton-Jacobi equations.
1. Introduction
The standard formulation of the Hamilton-Jacobi problem is to find a function S(t, q A )
(called the principal function) such that

S A S
+ H q , A = 0. (1.1)
t q

To Prof. Demeter Krupka in his 65th birthday

E-mail address: mdeleon@imaff.cfmac.csic.es

E-mail address: jcmarrer@ull.es

E-mail address: d.martin@imaff.cfmac.csic.es
130 M. de Leon, J.C. Marrero and D. Martn de Diego
If we put S(t, q A ) = W (q A ) tE, where E is a constant, then W satisfies

A W
H q , A = E; (1.2)
q
W is called the characteristic function.

Equations (1.1) and (1.2) are indistinctly referred as the Hamilton-Jacobi equation.
There are some recent attempts to extend this theory for classical field theories in the
framework of the so-called multisymplectic formalism [15, 16]. For a classical field theory
the hamiltonian is a function H = H(x , y i , pi ), where (x ) are coordinates in the space-
time, (y i ) represent the field coordinates, and (pi ) are the conjugate momenta.
In this context, the Hamilton-Jacobi equation is [17]
S

i S
+ H x ,y , =0 (1.3)
x y i
where S = S (x , y j ).
In this paper we introduce a geometric version for the Hamilton-Jacobi theory based in
two facts: (1) the recent geometric description for Hamiltonian mechanics developed in [6]
(see [8] for the case of nonholonomic mechanics); (2) the multisymplectic formalism for
classical field theories [3, 4, 5, 7] in terms of Ehresmann connections [9, 10, 11, 12].
We shall also adopt the convention that a repeated index implies summation over the
range of the index.
2. A Geometric Hamilton-Jacobi Theory for Hamiltonian

Mechanics
First of all, we give a geometric version of the standard Hamilton-Jacobi theory which will
be useful in the sequel.
Let Q be the configuration manifold, and T Q its cotangent bundle equipped with the
canonical symplectic form
Q = dq A dpA
where (q A ) are coordinates in Q and (q A , pA ) are the induced ones in T Q.
Let H : T Q R a hamiltonian function and XH the corresponding hamiltonian
vector field:
iXH Q = dH
The integral curves of XH , (q A (t), pA (t)), satisfy the Hamilton equations:
dq A H dpA H
= , = A.
dt pA dt q
Theorem 2.1 (Hamilton-Jacobi Theorem). Let be a closed 1-form on Q (that is, d = 0
and, locally = dW ). Then, the following conditions are equivalent:
Hamilton-Jacobi Theory for Classical Field Theories 131
(i) If : I Q satisfies the equation
dq A H
=
dt pA
then is a solution of the Hamilton equations;
(ii) d(H ) = 0.
To go further in this analysis, define a vector field on Q:

XH = T Q XH
as we can see in the following diagram:
XH
T Q T (T Q)
Q T Q

XH
Q TQ
Notice that the following conditions are equivalent:
(i) If : I Q satisfies the equation
dq A H
=
dt pA
then is a solution of the Hamilton equations;

, then is an integral curve of X ;
(i) If : I Q is an integral curve of XH H
are -related, i.e.

(i) XH and XH

T (XH ) = XH
so that the above theorem can be stated as follows:
Theorem 2.2 (Hamilton-Jacobi Theorem). Let be a closed 1-form on Q. Then, the fol-
lowing conditions are equivalent:
and X are -related;

(i) XH H
(ii) d(H ) = 0.
3. The Multisymplectic Formalism

3.1. Multisymplectic Bundles
The configuration manifold in Mechanics is substituted by a fibred manifold
: E M
such that
(i) dim M = n, dim E = n + m
(ii) M is endowed with a volume form .
We can choose fibred coordinates (x , y i ) such that
= dx1 dxn .
We will use the following useful notations:
dn x = dx1 dxn ,
dn1 x = i dn x .
x
Denote by V = ker T the vertical bundle of , that is, their elements are the tangent
vectors to E which are -vertical.
Denote by
: n E E
the vector bundle of n-forms on E.
The total space n E is equipped with a canonical n-form :
()(X1 , . . . , Xn ) = (e)(T (X1 ), . . . , T (Xn )),
where X1 , . . . , Xn T (n E) and is an n-form at e E.

The (n + 1)-form
= d ,
is called the canonical multisymplectic form on n E.
Denote by nr E the bundle of r-semibasic n-forms on E, say
nr E = { n E | iv1 vr = 0, whenever v1 , . . . , vr are -vertical}.
Since nr E is a submanifold of n E it is equipped with a multisymplectic form r , which

is just the restriction of .
Two bundles of semibasic forms play an special role: n1 E and n2 E. The elements of
these spaces have the following local expressions:
n1 E : p0 dn x,
n2 E : p0 dn x + pi dy i dn1 x
which permits to introduce local coordinates (x , y i , p0 ) and (x , y i , p0 , pi ) in n1 E and

n2 E, respectively.
Since n1 E is a vector subbundle of n2 E over E, we can obtain the quotient vector
space denoted by J 1 which completes the following exact sequence of vector bundles:
0 n1 E n2 E J 1 0 .
We denote by 1,0 : J 1 E and 1 : J 1 M the induced fibrations.
3.2. Ehresmann Connections in the Fibration 1 : J 1 M

A connection (in the sense of Ehresmann) in 1 is a horizontal subbundle H which is
complementary to V 1 ; namely,
T (J 1 ) = H V 1
where V 1 = ker T 1 is the vertical bundle of 1 . Thus, we have:
(i) there exists a (unique) horizontal lift of every tangent vector to M ;

(ii) in fibred coordinates (x , y i , pi ) on J 1 , then

V 1 = span , , H = span {H } ,
y i pi

where H is the horizontal lift of x .
(iii) there is a horizontal projector h : T J 1 H.
3.3. Hamiltonian Sections

Consider a hamiltonian section
h : J 1 n2 E
of the canonical projection : n2 E J 1 which in local coordinates read as
h(x , y i , pi ) = (x , y i , H(x, y, p), pi ) .
Denote by h = h 2 , where 2 is the multisymplectic form on n2 E.

The field equations can be written as follows:
ih h = (n 1) h , (3.1)
where h denotes the horizontal projection of an Ehresmann connection in the fibred mani-
fold 1 : J 1 M .
The local expressions of 2 and h are:
2 = d(p0 dn x + pi dy i dn1 x ),
h = d(H dn x + pi dy i dn1 x ) .
3.4. The Field Equations

Next, we go back to the Equation (3.1).
The horizontal subspaces are locally spanned by the local vector fields

H = h = + i i + ( )j ,
x x y pj
where i and ( )j are the Christoffel components of the connection.

Assume that is an integral section of h; this means that : M J 1 is a local
section of the canonical projection 1 : J 1 M such that T (x)(Tx M ) = H (x) , for
all x M .
If (x ) = (x , i (x), i (x)) then the above conditions becomes
i H i H

= ,
= i
x pi x y
which are the Hamilton equations.
4. The Hamilton-Jacobi Theory

Let be a 2-semibasic n-form on E; in local coordinates we have
= 0 (x, y) dn x + i (x, y) dy i dn1 x .
Alternatively, we can see it as a section : E n2 E, and then we have
(x , y i ) = (x , y i , 0 (x, y), i (x, y)) .
A direct computation shows that

0 i

d = dy i dn x
y i x

i
+ dy j dy i dn1 x .
y j
Therefore, d = 0 if and only if
0 i
= , (4.1)
y i x
i j
= . (4.2)
y j y i
Using and h we construct an induced connection in the fibred manifold : E M

by defining its horizontal projector as follows:
he : Te E Te E,
he (X) = T 1,0 h()(e) (X)
where (X) T()(e) (J 1 ) is an arbitrary tangent vector which projects onto X.

From the above definition we immediately proves that
(i) h is a well-defined connection in the fibration : E M .
(ii) The corresponding horizontal subspaces are locally spanned by

H = h = + i (( )(x, y)) i .
x x y
The following theorem is the main result of this paper.

Theorem 4.1. Assume that is a closed 2-semibasic form on E and that h is a flat con-
nection on : E M . Then the following conditions are equivalent:
(i) If is an integral section of h then is a solution of the Hamilton equations.
(ii) The n-form h is closed.
Before to begin with the proof, let us consider some preliminary results.
We have
(h )(x , y i ) = (x , y i , H(x , y i , i (x, y)), i (x, y)) ,
that is
h = H(x , y i , i (x, y)) dn x + i dy i dn1 x .
Notice that h is again a 2-semibasic n-form on E.

A direct computation shows that
!
H H j i i
d(h ) = + + dy i dn x + dy j dy i dn1 x .
y i pj y i x y j
Therefore, we have the following result.

Lemma 4.2. Assume d = 0; then
d(h ) = 0
if and only if
H H j i
+ + =0.
y i pj y i x
Proof of the Theorem.

(i) (ii)
It should be remarked the meaning of (i).
Assume that
(x ) = (x , i (x))
is an integral section of h; then

i H
= .
x pi
(i) states that in the above conditions,
( )(x ) = (x , i (x), j = j ((x)))
is a solution of the Hamilton equations, that is,

i i i j H
= + = i.
x x y j x y
Assume (i). Then

H H j i
+ +
y i pj y i x
H H i i
= + + , (since d = 0)
y i pj y j x
H j i i
= + + , (since the first Hamilton equation)
y i x y j x
=0 (since (i))
which implies (ii) by Lemma 4.2.

(ii) (i)
Assume that d(h ) = 0.
Since h is a flat connection, we may consider an integral section of h. Suppose that
(x ) = (x , i (x)).
Then, we have that

i H
= .
x pi
Thus,
j j j i
= + ,
x x y i x
j i i
= + , (since d = 0)
x y j x
j i H
= + , (since the first Hamilton equation)
x y j pi
H
= j , (since (ii)). 2
y
Assume that = dS, where S is a 1-semibasic (n 1)-form, say
S = S dn1 x
Therefore, we have
S S
0 = , i =
x y i
and the Hamilton-Jacobi equation has the form

S i S
+ H(x , y , ) =0.
y i x y i
The above equations mean that
S

i S
+ H x , y , = f (x )
x y i
so that if we put H = H f we deduce the standard form of the Hamilton-Jacobi equation

(since H and H give the same Hamilton equations):
S

i S
+ H x , y , =0.
x y i
An alternative geometric approach of the Hamilton-Jacobi theory for Classical Field

Theories in a multisymplectic setting was discussed in [15, 16].
5. Time-dependent Mechanics
A hamiltonian time-dependent mechanical system corresponds to a classical field theory
when the base is M = R.
We have the following identification 12 E = T E and we have local coordinates
(t, y i , p0 , pi ) and (t, y i , pi ) on T E and J 1 , respectively. The hamiltonian section is
given by
h(t, y i , pi ) = (t, y i , H(t, y, p), pi ) ,
and therefore we obtain
h = dH dt dpi dy i .
If we denote by = dt the different pull-backs of dt to the fibred manifolds over M ,

we have the following result.
The pair (h , dt) is a cosymplectic structure on E, that is, h and dt are closed forms
and dt nh = dt h h is a volume form, where dimE = 2n + 1. The Reeb
vector field Rh of the structure (h , dt) satisfies
iRh h = 0 , iRh dt = 1.
The integral curves of Rh are just the solutions of the Hamilton equations for H.
The relation with the multisymplectic approach is the following:
h = Rh dt ,
or, equivalently,

h = Rh .
t
A closed 1-form on E is locally represented by
= 0 dt + i dy i .
Using we obtain a vector field on E:
(Rh ) = T 1,0 Rh
such that the induced connection is
h = (Rh ) dt
Therefore, we have the following result.
Theorem 5.1. The following conditions are equivalent:
(i) (Rh ) and Rh are ( )-related.
(ii) The 1-form h is closed.
Remark 5.2. An equivalent result to Theorem 5.1 was proved in [14] (see Corollary 5 in
[14]).
Now, if
S S
= dS = dt + i dy i ,
t y
then we obtain the Hamilton-Jacobi equation

S i S
+ H t, y , =0.
y i t y i
Acknowledgement
This work has been partially supported by MEC (Spain) Grants MTM 2006-03322, MTM
2007-62478, project Ingenio Mathematica (i-MATH) No. CSD 2006-00032 (Consolider-
Ingenio 2010) and S-0505/ESP/0158 of the CAM.
References
[1] R. Abraham and J. E. Marsden, Foundations of Mechanics (2nd edition, Benjamin-
Cumming, Reading, 1978).
[2] V. I. Arnold, Mathematical Methods of Classical Mechanics (Graduate Texts in Math-

ematics 60, Springer-Verlag, Berlin, 1978).
[3] E. Binz, J. Sniatycki and H. Fischer, Geometry of Classical Fields (North-Holland

Mathematics Studies, 154, North-Holland Publishing Co., Amsterdam, 1988).
[4] F. Cantrijn, A. Ibort and M. de Leon, On the geometry of multisymplectic manifolds,

J. Austral. Math. Soc. (Series A) 66 (1999) 303330.
[5] F. Cantrijn, A. Ibort and M. de Leon, Hamiltonian structures on multisymplectic man-

ifolds, Rend. Sem. Mat. Univ. Pol. Torino 54 (3) (1996) 225236.
[6] J. F. Carinena, X. Gracia, G. Marmo, E. Martnez, M. Munoz-Lecanda and N. Roman-

Roy, Geometric Hamilton-Jacobi theory, Int. J. Geom. Meth. Mod. Phys. 3 (7) (2006)
14171458.
[7] M. J. Gotay, J. Isenberg, J. E. Marsden and R. Montgomery, Momen-

tum Maps and Classical Relativistic Fields. Part I: Covariant Field Theory;
arXiv:physics/9801019v2.
[8] M. de Leon, D. Iglesias-Ponte and D. Martn de Diego, Hamilton-Jacobi Theory for

Nonholonomic Mechanical Systems, J. Phys. A: Math. Theor. 41 (2008) 015205, pp.
14.
[9] M. de Leon, J. C. Marrero and D. Marn, A geometrical approach to Classical Field

Theories: a constraint algorithm for singular theories, In: New Developments in Dif-
ferential Geometry (Debrecen, 1994, Mat. Appl. 350, Kluwer, Dordrecht, 1996) 291
312.
[10] M. de Leon, J. C. Marrero and D. Marn, Ehresmann connections in Classical Field

Theories, In: Differential Geometry and its Applications (Granada, 1994, Anales de
Fsica, Monografas 2, 1995) 7389.
[11] M. de Leon, D. Martn de Diego and A. Santamara-Merino, Tulczyjew s triples and

lagrangian submanifolds in classical field theory, In: Applied Differential Geometry
and Mechanics ((W. Sarlet and F. Cantrijn, Eds.) Universiteit Gent, 2003) 2147.
[12] M. de Leon, D. Martn de Diego and A. Santamara-Merino, Symmetries in Field

Theory, Int. J. Geom. Meth. Mod. Phys. 1 (5) (2004) 651710.
[13] M. de Leon, P. R. Rodrigues, Generalized Classical Mechanics and Field Theory

(North-Holland, Amsterdam, 1985).
[14] J. C. Marrero and D. Sosa, The Hamilton-Jacobi equation on Lie Affgebroids, Int. J.
Geom. Meth. Mod. Phys. 3 (3) (2006) 605622.
[15] C. Paufler and H. Romer, De Donder-Weyl equations and multisymplectic geometry,

In: XXXIII Symposium on Mathematical Physics (Torun, 2001, Rep. Math. Phys. 49,
2002) 325334.
[16] C. Paufler and H. Romer, Geometry of Hamiltonean n-vector fields in multisymplectic

field theory, J. Geom. Phys. 44 (2002) 5269.
[17] H. Rund, The Hamilton-Jacobi theory in the Calculus of Variations (Robert E. Krieger
Publ. Co., Nuntington, N.Y. 1973).
[18] D. J. Saunders, The Geometry of Jet Bundles (London Mathematical Society Lecture
Notes Ser. 142, Cambridge Univ. Press, Cambridge, 1989).
Part II
N ATURAL B UNDLES AND

D IFFERENTIAL I NVARIANTS
141
Chapter 9
N ATURAL L AGRANGIAN S TRUCTURES

Josef Janyska
Department of Mathematics and Statistics
Masaryk University, Janackovo nam 2a, 602 00 Brno
Czech Republic
Abstract
We use the theory of natural and gauge-natural bundles and natural differential
operators to give a general description of invariant and gauge invariant Lagrangian
structures on natural and gauge natural bundles.
2000 Mathematics Subject Classification. 58A32, 53C05, 53C80, 70G45, 70G50, 70S05,
70S15
Key words and phrases. Natural bundle, gauge-natural bundle, natural differential opera-
tor, invariant Lagrangian, infinitesimal symmetr
1. Introduction
Let : Y M be a fibered manifold and r : J r Y M its r-jet prolongation, Saunders
[43]. An r-order Lagrangian is a fibered morphism, over M ,
L : J r Y m T M . (1.1)
A vector field on Y projectable on a vector field on M is said to be an infinitesimal

symmetry of L if the Lie derivative of L with respect to the r-jet lift J r of vanishes,
Krupka and Trautman [35], i.e.
LJ r L(y r ) = 0 , (1.2)
where y r J r Y . But from the general theory of Lie derivatives
LJ r L(y r ) := L(J r ,m T ) L(y r ) = (T L J r )(y r ) m T (L(y r )) ,

E-mail address: janyska@math.muni.cz
144 Josef Janyska
where m T , m = dim M , is the vector field on m T M obtained by the flow prolonga-

tion of , see Subsection 2.13..
So the condition for infinitesimal symmetries LJ r L = 0 reads as
T L(J r )(y r ) = m T (L(y r )) . (1.3)
Usually, in physical theories, Y is a bundle (tensor bundle, bundle of connections, gauge

bundle, ...) with a geometrical structure given by an action of a Lie group. Then, if is
a vector field preserving the given structure, a Lagrangian satisfying (1.3) is said to be
invariant with respect to the given structure.
There are two main geometrical structures of Y :

1. Y is a natural bundle (in the sense of Nijenhuis [40]) with a geometrical structure
given by the differential group Grm = invJ0r (Rm , Rm )0 . Examples of such bundles are
tensor bundles (r = 1) and the bundle of classical (linear) connections (r = 2). Invariant
Lagrangians on tensor bundles were studied by many authors, see for instance Krupka [26,
27, 28, 30, 31] and Novotny [41].
2. Y is a gauge-natural bundle (in the sense of Eck [6]) with a geometrical structure
induced by a gauge group G. In this case a special attention is devoted to invariant La-
grangians on the bundle of principal connections, see for instance Betounes [1], Eck [6],
Horndeski [10], Utiyama [51] and others.
In this paper we will study natural (invariant and gauge invariant) Lagrangian struc-
tures on natural and gauge-natural bundles. To describe invariant Lagrangians on natural
or gauge-natural bundles we can use general properties of natural differential operators on
natural bundles or gauge-natural bundles.
The paper is organized as follows. In Section 2. we recall basic properties of natu-
ral bundles and natural differential operators on natural bundles. In Section 3. we study
invariant (natural) Lagrangians on natural bundles, especially on the natural bundle of clas-
sical connections. In Section 4. we recall basic properties of gauge-natural bundles and
natural differential operators on gauge-natural bundles and finally, in Section 5., we study
gauge invariant (natural) Lagrangians on gauge-natural bundles, especially on the gauge-
natural bundle of general linear and principal connections by using higher order versions of
Utiyamas reduction method, Janyska [15, 17, 18].
In what follows we will use the following notations. M is the category of all smooth
manifolds and smooth mappings, Mm is the category of all m-dimensional smooth man-
ifolds and local diffeomorphisms, F M m is the category of all fibred manifolds with m-
dimensional bases and smooth fibred morphisms covering local diffeomorphisms of bases,
V B m (A B m ) is the category of all vector (affine) bundles with m-dimensional bases and
smooth linear (affine) fibred morphisms covering local diffeomorphisms of bases and, fi-
nally, PB m (G) is the category of all principal G-bundles with m-dimensional bases and
smooth principal fibred morphisms covering local diffeomorphisms of bases.
In what follows all manifolds and maps are supposed to be smooth.
Natural Lagrangian Structures 145
2. Natural Bundles and Operators

We recall here definitions and basic properties of the theory of natural bundles and natural
differential operations, for details see [9, 11, 19, 24, 33, 40, 48, 53]. As examples we
mention functors and operators which will be used later.
Natural differential operators on natural bundles are closely related with the term ge-
ometric invariant (or concomitant) which has been used in differential geometry since the
end of the 19th century. In the 1930s Schouten and his collaborators, [45], used the notion
of geometric object and invariant operations with geometric objects. A modern functorial
approach to the theory of geometrical objects and invariant operations with geometric ob-
jects was introduced by Nijenhuis [39] in the 1950s. Starting from the famous paper by
Nijenhuis [40] geometrical objects and invariant operations with geometrical objects have
been very intensively studied by using the concepts of natural bundles and natural differ-
ential operators. Nijenhuis defined natural bundles as lifting functors on the category Mm .
Lifting functors are supposed to satisfy three conditions: prolongation, localization and
regularity (continuity).
Kolar [22] generalized lift functors to the category M of all differentiable manifolds
and their smooth mappings. Such functors are called prolongation functors. Later various
prolongation functors on subcategories of M were studied.
Main problem of the theory of natural differential operators is to give a complete clas-
sification of them for concrete underlying geometric structures. Such classification is based
on the one-to-one correspondence of natural differential operators and equivariant maps be-
tween standard fibres. To classify equivariant maps we can use several methods. Formerly
the method of Lie equations was used, Krupka and Janyska [33], recently we use the al-
gebraic method by Kolar, Michor and Slovak [24]. In literature it is possible to find many
examples and applications of natural operations used in geometry and physics. For wide
list of references we recommend to see Fatibene and Francaviglia [9], Kolar, Michor and
Slovak [24] and Krupka and Janyska [33].
2.1. Natural Bundles

Natural bundles were introduced by Nijenhuis [40] over the category Mm . We use more
general definition by Kolar, Michor and Slovak [24].
Definition 2.1. A natural bundle functor on a subcategory C of M is a covariant functor

F from C to the category F M satisfying
(i) (prolongation) for each manifold M Ob C , pM : F M M is a fibred manifold
over M ,
(ii) (localization) for each f Mor C , F f is a fibred manifold map covering f such
that F (U ) = (F U ) for any open subset : U M .
A natural bundle functor on the subcategory Mm of M , for a certain m, is a natural

lift functor, Nijenhuis [40]. In literature natural bundle functors on M are also called
prolongation functors, Kolar [22].
In original definitions of natural lift and prolongation functors there is the regularity
condition saying that a smoothly parameterized family of diffeomorphisms is prolonged
146 Josef Janyska
into a smoothly parameterized family of diffeomorphisms. But this condition turns out to
be a consequence of remaining prolongation and localization properties. This was proved
by Epstein and Thurston [7] for natural lift functors and by Kolar and Slovak [25] for natural
prolongation functors.
A natural bundle is then a triplet (F M, pM , M ).
Later (Theorem 2.2) we will see that pM : F M M is indeed a bundle.
2.2. Geometrical Object

A geometrical object on a manifold M is now an element from F M , where F is a natural
bundle functor. A section : M F M is a field of geometrical objects on M .
2.3. Order of Natural Bundle Functors

We say that a natural lift functor F is of finite order r if r is the smallest number such that
jxr f = jxr g F f |Fx M = F g|Fx M
for any (f, g : M M ) Mor Mm and any x M .

Palais and Terng [42] proved that the order r of a natural lift functor is finite r < 2n+1
where n is the dimension of the fiber F0 Rm . Later Epstein and Thurston [7] gave much
better bound. They proved r 2n + 1 and that this bound is sharp for m = 1. Finally Zajtz
n n
[54] proved r max{ n1 ;m + 1}. Mikulski [38] has shown that a natural prolongation
functor with infinite order exists.
2.4. Differential Group

Let us denote by Grm the Lie group
Grm = invJ0r (Rm , Rm )0
of invertible r-jets (with source and target 0) of diffeomorphisms of Rm which preserve 0.

The group multiplication is given by the jet composition. The canonical coordinates on Grm
will be denoted by (a , . . . , a1 ...r ) and tilde will refer to the inverse element.
2.5. Standard Fiber

Let F be an r-order natural lift functor and let F0 = F0 Rm . Because of (ii) of Definition
2.1 F0 is diffeomorphic with Fx M for any x M , M ObMm . F0 will be called the
standard fiber of F . Applying F on origin-preserving diffeomorphisms of Rm we get a left
action of Grm on F0 which defines on F0 a structure of a left smooth Grm -manifold, Krupka
[29] and Terng [48].
2.6. Natural Fibred Coordinate Chart

Local coordinate charts (x ) on M and (y p ) on F0 induce a fibred coordinate chart (x , y p )
on F M , which is said to be the natural fibred coordinate chart.
2.7. Examples
1. The tangent functor T is a natural bundle functor of order one on the category M
with values in the category V B. In dimension m the corresponding standard fiber is Rm on
which G1m = Gl(m, R) acts in the standard way by the matrix multiplication. The natural
fibred coordinate chart on T M will be denoted by (x , x ).
2. The cotangent functor T is a natural lift functor of order one with values in the
category V B m . The standard fiber is Rm with the standard action of G1m .
3. The functor p T of pforms is a natural lift functor of order one with values in
the category V B m . The standard fiber is p Rm on which G1m acts in the standard tensor
way. The natural fibred coordinate chart on p T M will be denoted by (x , 1 ...p ),
1 1 < < p m. Especially, for p = m, we obtain the natural lift functor of
volume forms.
4. The functor of pseudoRiemannian metrics pRm is a natural lift functor of order
one such that pRm(M ) are subbundles of bundles from the category V B m . The standard
fiber (pRm)0 is the subspace in 2 Rm of non-degenerate symmetric matrices with the
tensor action of G1m . The natural fibred coordinate chart on pRm(M ) will be denoted by
(x , g ), g = g , det(g ) 6= 0.
5. The functor of k r -velocities Tkr is a natural bundle functor of order r on the category
M . For any M Ob M , we define Tkr M = J0r (Rk , M ) and, for any f Mor M ,
f : M M , we define Tkr f (j0r ) = j0r (f ), where j0r Tkr M . The standard fiber of
Tkr , in dimension m, is J0r (Rk , Rm )0 and the action of Grm on the standard fiber is given by
the jet composition.
6. The functor of r-order frames F r is a natural lift functor of order r. For any M
Ob Mm , we define F r M = invJ0r (Rm , M ) and, for any f Mor Mm , F r f is defined as
in Example 2.7..5. The values of the functor F r are in the category PB m (Grm ).
7. The functor Cla of classical (linear) connections on a given manifold is a natural
lift functor of order two with values in the category A B m . Its standard fiber is Rm
Rm Rm on which G2m acts via the well known transformation relations of the Christoffel
symbols, Christoffel [3],
= a ( a a a ) .
The natural fibred coordinate chart on Cla M will be denoted by (x , ).

By Cla will be denoted the functor of torsion free linear connections. In natural fibred
coordinates Cla is characterized by = .
8. Let F be a natural lift functor of order r and J s be the functor of s-jet prolongation,
Saunders [43]. Then J s F J s F is a natural lift functor of order (r + s). If F0 is the
standard fiber of F , then the standard fiber of J s F is (J s F )0 = Tns F0 and the action of
Gr+s s r
m on (J F )0 is obtained by the jet prolongation of the action of Gm on F0 .
2.8. The Bundle Structure

In the theory of natural lift functors the functor of r-order frames, defined in Example 2.7..6,
plays a fundamental role. Namely, we have the following theorem, Krupka [29] and
Terng [48].
148 Josef Janyska
Theorem 2.2. Any natural lift functor F of order r, with the standard fiber F0 , is canoni-
cally represented by
F M = [F r M, F0 ], F f = [F r f, idF0 ],
where M Ob Mm , f Mor Mm , and [F r M, F0 ] = (F r M, F0 )/Grm is the bundle

associated with F r M .
This theorem implies that there is the one-to-one correspondence between r-order nat-
ural lift functors and left Grm -manifolds.
2.9. Natural Differential Operators

Let F be a natural lift functor, f : M M be a mapping in MorMm and : M F M
be a section. Then we define the section f : M F M by f = F f f 1 .
Definition 2.3. A natural differential operator D from a natural lift functor F1 to a natural
lift functor F2 is a family of differential operators
{D(M ) : C (F1 M ) C (F2 M )}M ObMm
such that
(i) D(M )(f ) = f D(M )() for every section C (F1 M ) and every f : M
M in MorMm ,
(ii) DU (|U ) = (DM )|U for every section C (F1 M ) and every open submani-
fold U M ,
(iii) every smoothly parameterized family of sections of F1 M is transformed into a
smoothly parametrized family of sections of F2 M .
2.10. Order of Natural Differential Operator

A natural differential operator is of order k, 0 k , if all D(M ), M ObMm , are of
order k. Thus, a k-order natural differential operator D from F1 to F2 is characterized by
the associated fibred manifold morphisms D(M ) : J k F1 M F2 M , over M , according
to the formula D(M )(jxk ) = D(M )()(x). The family D = {D(M )}M ObMm defines
a natural transformation of the functors J k F1 and F2 . In what follows we will identify k-
order natural differential operators with the corresponding natural transformations and use
the same symbol.
2.11. Equivariant Mappings Given by Natural Operators

Coordinate independent geometrical constructions are in fact natural differential operators
between natural lift functors. The study of natural differential operators is based on rela-
tions between natural differential operators and equivariant mappings. The basic tool is the
following theorem, Terng [48].
Theorem 2.4. There is a bijective correspondence between the set of k-order natural dif-
ferential operators from a natural lift functor F1 to a natural lift functor F2 and equivariant
mappings from the standard fiber of J k F1 to the standard fiber of F2 .
2.12. Examples
1. The exterior differential d is a first order natural differential operator from p T , p
0, to p+1 T . The corresponding G2n -equivariant mapping from J 1 (p T )0 = Tn1 (p Rm )
to (p+1 T )0 = p+1 Rm is given, in the canonical coordinate chart (1 ...p ), 1 1 <
< p m, on (p Rm ), by
1 ...p+1 d = [1 ...p ,p+1 ] ,
where [...] denotes the anti-symmetrization. For p 1 the naturality determines d up to

a constant multiple while in classical proofs the linearity was supposed, see for instance
Kolar [23], Krupka and Janyska [33] and Krupka and Mikolasova [34].
2. The Levi-Civita connection is a first order natural differential operator from pRm to
Cla . The corresponding G2m -equivariant mapping from J 1 (pRm)0 to Cla0 is given by the
formal Christoffel symbols, Christoffel [3],
= 12 g (g, + g, g, ),
where (g ) is the inverse matrix of (g ). The uniqueness of the LeviCivita connection is

the classical geometrical problem. The proof of the uniqueness by using natural techniques
can be found in Krupka and Janyska [33], Krupka and Mikolasova [34] and Slovak [47].
3. The curvature tensor of a classical connection is a first order natural differential
operator from Cla to T T (2 T ). The corresponding G3m -equivariant mapping from
J 1 Cla0 to (T T (2 T ))0 = Rm Rm (2 Rm ) is given by
w = , , + .
The curvature tensor is not unique operator of this type and plays an important role in
classification of natural operators defined on classical connections, see for instance Kolar
[23], Kolar, Michor and Slovak [24] and Schouten [44].
2.13. Infinitesimal Properties of Natural Lift Functors

The regularity property of lift functors allows us to lift vector fields on a manifold M to
projectable vector fields on a natural bundle F M by using flows. Namely if exp(t) is the
flow of a vector field on M then
F (exp(t)) = exp(tF)
is the flow of the vector field F on F M which is said the flow lift of . Moreover, if F is
of order r, then F depends on r-jets of .
For instance the flow lift of a vector field = x with respect to the tangent functor
is the vector field

T = + x
x x x
150 Josef Janyska
on T M and the flow lift of with respect to the natural lift functor of classical connections,
see Example 2.4..7, is the vector field

Cla = + (2.1)
x x

2
+
x x x x
on Cla M .
2.14. Infinitesimal Properties of Natural Operators

If : M F M is a section of an r-th order natural bundle (a field of geometrical objects)
then we can define the Lie derivative of with respect to a vector field by the formula
d
L = |0 {exp(t) } . (2.2)
dt
L is a section of V F M . Natural differential operators D from a natural lift functor F1 to
a natural lift functor F2 are infinitesimally characterized by the commutativity with the Lie
derivatives, Kirillov [21] and Janyska and Modugno [19], in the following sense.
Theorem 2.5. A k-order differential operator D(M ) from a natural bundle F1 M to a
natural bundle F2 M is natural if and only if
L D(M )() = T D(M )(L ) (2.3)
for all C (F1 M ) and all vector fields on M .
If we identify D with the corresponding natural transformation D, then the above con-
dition (2.3) is equivalent with
T D(M )(J k F1 )(j k ) = F2 (D(M )(j k )) . (2.4)
3. Natural Lagrangian Structures on Natural Bundles

If we assume in (2.4) the natural bundle of volume forms as the target bundle, i.e. F2 =
m T , the condition (2.4) means that natural Lagrangians on the natural bundle F1 M are
just invariant Lagrangians on F1 M (in the sense of (1.2)) and infinitesimal symmetries of
such Lagrangians are vector fields F1 , for all vector fields on M .
So a k-order Lagrangian L : J k F1 M m T M is invariant (natural) if and only if
LJ k F1 L = 0 (3.1)
for any vector field on M .
The above infinitesimal characterization of invariant Lagrangians on natural bundles
leads in natural fibered local coordinates to a system of partial differential equations which
is generally difficult to solve. Much more simple is to use algebraical method of classi-
fication of equivariant mappings corresponding to natural operators. Special situation we
obtain in the case of invariant Lagrangians on the natural bundle of classical symmetric
connections. In this case we can use the reduction theorems by Schouten [44].
3.1. The First and the Second Reduction Theorems

Let us assume the natural bundle of classical symmetric connections on manifolds given
by the lift functor Cla , see Example 2.7..7. Schouten [44] proved that all finite order
polynomial concomitants on Cla M and a bundle of geometrical object of order one with
values in an other bundle of geometrical objects of order one factorize through the curvature
tensor, a given field of geometrical objects and their covariant differentials. By using the
theory of natural differential operators on natural bundles the results of Schouten were
generalized by Kolar, Michor and Slovak [24] and we obtain the first reduction theorem in
the form.
Theorem 3.1. All natural differential operators of order r with values in a natural bundle
of order one of a classical symmetric connection are natural differential operators of order
zero of the curvature tensor and its covariant differentials up to the order (r 1), i.e.
D(j r ) = D(
e (r1) R[]) ,
where (r1) = (id, , . . . , r1 ).

The second reduction theorem can be formulated as follows.
Theorem 3.2. Let be a field of geometrical objects of order one and s r 1. All
natural differential operators with values in a natural bundle of order one of a classical
symmetric connection (in order s) and (in order r) are natural differential operators of
the curvature tensor, and their covariant differentials up to the orders (s 1) and r,
respectively, i.e.
D(j s , j r ) = D(
e (s1) R[], (r) ) .
Remark 3.3. If is a classical non-symmetric connection on M , then there exists its unique
splitting =
e + T , where e is the classical symmetric connection obtained by the sym-
metrization of and T is the torsion tensor of . Then all natural operators of and are
of the form
D(j s , j r ) = D(j s ,
e j s T, j r ) = D(
e e (s1) R[],
e e (s) T,
e (r) ) ,
where
e refers to the connection .
e
3.2. Invariant Lagrangians of Classical Symmetric Connections

Now, as a simple consequence of Theorem 3.1, we have
Theorem 3.4. All invariant (natural) Lagrangians of order r on the natural bundle of
classical symmetric connections are of the type
L(j r ) = L(
e (r1) R[]) .
Remark 3.5. r-order invariant Lagrangians on classical symmetric connections are infinites-
imally given as solutions of the system of partial differential equations given by
LJ r Cla L = 0 (3.2)
152 Josef Janyska
for all vector fields on M and Cla given by (2.1). From Theorem 3.4 it follows that all
solutions of (3.2) factorize through the curvature tensor and its covariant differentials up to
the order (r 1).
As a consequence of Theorem 3.2, we have
Theorem 3.6. All invariant Lagrangians on the natural bundle of classical symmetric con-
nections (in order s) and a natural bundle of order one (in order r) are of the type
L(j s , j r ) = L(
e (s1) R[], (r) ) .
3.3. Invariant Lagrangians of Metric Fields

In physical theories M is usually an oriented spacetime with a metric g. In this case we
have the Levi Civita connection (g) and the canonical volume form
p
(g) = |g| dx1 . . . dxm .
Any invariant Lagrangian is of the form L(j r g, j r ) = l(j r g, j r ) (g) where l(j r g, j r )
is an invariant Lagrangian function. Then, as a consequence of Theorem 3.6, invariant
Lagrangian functions are of the type
l(j r g, j r ) = e
l(g, (r2) R[(g)], (r) ) .
Invariant Lagrangians on the natural bundle of metrics were studied for instance by
Krupka [27, 28] and Novotny [41].
4. Gauge-Natural Bundles
Natural differential operators on natural bundles describe the invariance of geometrical or
physical theories with respect to changes of local coordinates. But in physical theories an-
other sort of invariance plays an important role, the so called gauge invariance. Invariant
gauge theory has been introduced in the book by H. Weyl [52] in 1918 as a generalization
of the Einsteins general relativity. Weyl considered operators on a spacetime invariant not
only with respect to isomorphisms of spacetime but also with respect to gauge transforma-
tions g 7 e g (the term gauge was used for the first time by H. Weyl). The original
invariant physical gauge theories was related with the gauge group U (1) acting on wave
functions and electromagnetic potentials. In early 1950s the concept of gauge invariance
was generalized first for the spin group, see for instance Yang and Mills [55], and then for
any Lie group G playing the role of the gauge group. The first geometrical interpretation
of gauge invariance with respect to a general gauge group can be found in the famous pa-
per by Utiyama [51]. Gauge invariant theories can be described geometrically by using the
concepts of gauge-natural bundle functors and natural differential operators between gauge-
natural bundles. So in Section 4. we recall basic definitions and properties of gauge-natural
bundle functors, see Eck [6], Fatibene and Francaviglia [9], Kolar [23] and Kolar, Michor
and Slovak [24].
4.1. Gauge-natural Bundle Functors

Gauge-natural bundle functors was introduced by Eck [6]. We recall here the definition of
Kolar, Michor and Slovak [24]. Let us recall that B is the base functor from the category
F M to the category M .
Definition 4.1. A gauge-natural bundle functor over m-dimensional manifolds is a functor
F : PB m (G) F M m such that
(i) every PB m (G)-object : P BP is transformed into a fibered manifold qP :
F P BP over BP ,
(ii) every PB m (G)-morphism f : P P is transformed into a fibered morphism
F f : F P F P over Bf ,
(iii) for every open subset U BP , the inclusion : 1 (U ) P is transformed into
the inclusion F : qP1 (U ) F P .
A gauge-natural bundle is then a quadruple (F P, P , M, : P BP ).
Later (Theorem 3.3.) we will see that F P is actually a bundle.
In the original definition by Eck [6] there is one more regularity (continuity) condition
which says that a smoothly parametrized family of diffeomorphisms of P is transformed
into a smoothly parameterized family of isomorphisms of F P . But this condition is a
consequence of (i), (ii) and (iii), Kolar, Michor and Slovak [24].
4.2. Functor W r
Let ( : P M ) Ob PB m (G), let W r P be the space of all r-jets j(0,e) r , where
: R G P is in Mor PB m (G), 0 R and e is the unity in G. The space W r P is
m m
a principal fibre bundle over M with the structure group Wm r G = Jr m m

(0,e) (R G, R G)
of all r-jets of principal fibre bundle isomorphisms : Rm G Rm G covering the
diffeomorphisms : Rm Rm such that (0) = 0. The group Wnr G is the semidirect
product Grm Tm r G of Gr and T r G with respect to the action of Gr on T r G given by
m m m m
the jet composition. Let ( : P P ) Mor PB m (G), then we can define the principal
bundle morphism W r : W r P W r P by the jet composition. The rule transforming
any P Ob PB m (G) into W r P Ob PB m (Wm r G) and any Mor PB (G) into
m
r r
W Mor PB m (Wm G) is a gauge-natural bundle functor, Kolar [23].
4.3. Bundle Structure

The gauge-natural bundle functor W r described in Subsection 4.2. plays a fundamental role
in the theory of gauge-natural bundle functors. We have, Eck [6],
Theorem 4.2. Every gauge-natural bundle F P is a fibred bundle associated with the
gauge-natural bundle W r P for a certain order r.
4.4. Order of Gauge-natural Bundle Functors

The number r from Theorem 4.2 is called order of the gauge-natural bundle functor F . So
if F is an r-order gauge-natural bundle functor then
F P = [W r P, F0 ], F = [W r , idF0 ],
154 Josef Janyska
r G-manifold called the standard fibre of F .
where F0 is a left Wm
4.5. Gauge and Total Order of Gauge-natural Functors

Let F be an s-order gauge-natural bundle functor and let r s be the minimal number
such that the action of Wm s G = Gs T s G on F can be factorized through the canonical
m m 0
projection rs : Tms G T r G, s r. Then s is said to be the total order of F , r is
m
the gauge order and we say that F is of order (s, r). In what follows we shall denote by
(s,r)
Wm G = Gsm Tm r G and by W (s,r) P the corresponding principal bundle.
4.6. Gauge-natural Fibred Coordinate Chart

A local fibred coordinate chart (x , p ) on P and a coordinate chart (y p ) on F0 induce a
fibred coordinate chart (x , y p ) on F P which is said to be the gauge-natural fibred coordi-
nate chart.
4.7. Examples
1. Any r-order natural lift functor in the sense of Definition 2.1 is the (r,0)-order gauge-
natural bundle functor with the trivial gauge action, i.e. the action (Grm G) F0 F0
does not depend on G.
2. Let ( : P M ) Ob PB m (G) and let us denote by Pri P M the bundle
of principal connections on P . Then Pri is a (1,1)-order gauge-natural bundle functor with
the standard fibre G Rm and with the action of Wm 1 G given by, Kolar [23],
(X, g, Z)(Y ) = ad(g)(Y + Z)X 1 .

In particular, let G = Grn , then Pri P can be viewed as the bundle Lin E of linear
connections on an associated vector bundle E M with n-dimensional fibres. The stan-
dard fibre of Lin is Lin0 = Rn Rn Rm with coordinates (Kj i ), i, j = 1, ..., n,
(1,1)
= 1, ..., m, and the action of Wm G1n = G1m Tm 1 G1 on Lin is given, in the canonical
n 0
i i 1 1 1
coordinates (a , aj , aj ) on Gm Tm Gn , by
Kj i = aip Kq p aqj a + aip apj a ,
where tilde refers to the inverse element.
3. Let F0 be a left Gmanifold. The associated gauge-natural bundle functor is defined
by
assF0 (P ) = [P, F0 ], assF0 () = [, idF0 ] ,
where P ObPB m (G), MorPB m (G). assF0 (P ) is a 0-order gauge-natural bun-
dle. Especially the adjoint bundle ad P is the 0-order gauge-natural bundle given by the
adjoint action of G on its Lie algebra G .
4. If F is a gauge-natural bundle functor of order (s, r) then J k F is a gauge-natural
bundle functor of order at most (s + k, r + k). The number (s + k) is exact, but (r + k)
may be too big. For instance if F is an s-order natural lift functor, i.e. an (s,0)-order gauge-
natural bundle functor, then J k F is an (s+k)-order natural lift functor, i.e. an (s+k,0)-order
gauge-natural bundle functor.
5. ad P (p T M ) is a (1,0)-order gauge-natural vector bundle.
4.8. Natural Operators

Let (, f ) Mor PB m (G), : P P over f : M M , F be a gauge-natural bundle
functor and : M F P be a section. Then we define the section : M F P by
= F f 1 .
Definition 4.3. A natural differential operator D from a gauge-natural bundle functor F1

to a gauge-natural bundle functor F2 is a family of differential operators
{D(P ) : C (F1 P ) C (F2 P )}P ObPBm (G)
such that
(i) D(P )( ) = D(P )() for every sec-
tion
C (F1 P ) and every (, f ) MorP
Bm (G), : P P over f : M M ,
(ii) D( 1 (U ))(|U ) = (D(P )())|U for every section C (F1 P ) and every
open subset U M ,
(iii) every smoothly parameterized family of sections of F1 P is transformed into a
smoothly parameterized family of sections of F2 P .
4.9. Order of Natural Differential Operators

A natural differential operator D from F1 to F2 is of a finite order k if all D(P ), ( : P
M ) ObPB m (G), depend on k-order jets of sections of F1 P . Thus, a k-order natural
differential operator from F1 to F2 is characterized by the associated fibred manifold mor-
phism D(P ) : J k F1 P F2 P , over M , such that the family D = {D(P )}P ObPBm (G)
is a natural transformation of J k F1 to F2 . In what follows we will identify k-order natu-
ral differential operators with the corresponding natural transformations and use the same
symbol.
Theorem 4.4. Let F1 and F2 be gauge-natural bundle functors of order r. Then we have
a one-to-one correspondence between natural differential operators of order k from F1 to
F2 and Wm r+k G-equivariant mappings from (J k F ) to (F ) .
1 0 2 0
This theorem is due to Eck [6], see also Kolar, Michor and Slovak [24].
4.10. Curvature Operator

The curvature operator of principal connections is a 1-order natural differential operator
(2,2)
from Pri to ad (2 T ) with the associated Wm G-equivariant morphism
(ua ) R = a , a , + cabd b d ,
where c = (cabd ) are the structure constants of G.

156 Josef Janyska
4.11. Infinitesimal Properties of Gauge-natural Bundle Functors

The regularity (continuity) property of gauge-natural bundle functors allows to transform
right G-invariant vector fields on a principal G-bundle P to projectable vector fields on
a gauge-natural bundle F P by using flows. Namely if exp(t) is the flow of a right G-
invariant vector field on P , projectable on the vector field on M , then F (exp(t)) =
exp(tF) is the flow of the vector field F on F P which is said the flow transformation
of . Moreover, if F is of order r, then F depends on r-jets of .
For instance if (Bea ) is the base of vertical right G-invariant vector fields on P given by
a base (Ba ) of the Lie algebra G and is a right G-invariant vector field = (x) x +
a (x)Bea on P . Then the flow transformation of with respect to the gauge-natural bundle
functor of principal connections, see Example 4.7..2, is the vector field
a
P ri = + cabd b d a + (4.1)
x x x a
on Pri P .
4.12. Infinitesimal Properties of Natural Differential Operators

If : M F P is a section of an r-th order gauge-natural bundle then we can define the
Lie derivative of with respect to a right G-invariant vector field on P , over the vector
field on M , by the formula
d
L = |0 {exp(t) } .
dt
L is a section of V F P . Natural differential operators D from a gauge-natural bundle
functor F1 to a gauge-natural functor functor F2 can be infinitesimally characterized by the
commutativity with the Lie derivatives, Janyska and Modugno [19], in the following sense.
Theorem 4.5. A k-order differential operator D(P ) from a gauge-natural bundle F1 P to
a gauge-natural bundle F2 P is natural if and only if
L D(P )() = T D(P )(L ) , (4.2)
for all right G-invariant vector fields on P and all sections C (F1 P ).
If we identify D with the corresponding natural transformation D, then the above con-
dition (4.2) is equivalent with
T D(P )(J k F1 )(j k ) = F2 (D(P )(j k )) . (4.3)
5. Natural Lagrangians on Gauge-Natural Bundles

Gauge invariant theories can be very efficiently formulated by using the theory of gauge-
natural bundles and natural differential operators and have wide applications in gauge field
theories, see Fatibene and Francaviglia [9]. As applications of natural differential operators
on gauge-natural bundles we will generalize the Utiyamas reduction method for the linear
group Gl(n, R) as a gauge group, Janyska [15], and for a general Lie group G as a gauge
group, Janyska [17, 18].
5.1. Gauge Invariant Lagrangians on Electromagnetic Potentials

Originally gauge invariant Lagrangians were studied on a complex wave function (a par-
ticle field) and a gauge field A = (A (x)). The gauge invariance was considered with
respect to the gauge transformation of given in the form (x) 7 ei(x) (x). In order
to obtain a gauge invariant Lagrangian of of order one it is necessary to consider also a

gauge field A with the gauge transformation A 7 (A + ) x x
. Then a gauge invariant
Lagrangian of order one is of the form
L(j 1 , A) = L(,
e ) , (5.1)
where = i A . In physical theories A is an electromagnetic potential and

the above result is known as the minimal coupling principle. Further, a gauge invariant
Lagrangian of order one of the gauge field A only factorizes through the 2-form (electro-
magnetic 2-form) F = A A , i.e.
L(j 1 A) = L(F
e ). (5.2)
The above classical gauge invariance corresponds to the invariance with respect to the
gauge group U (1). The above results can be interpreted geometrically as follows: let Q
M be a complex line bundle (a quantum bundle) with a Hermitian product h and local
fibered coordinates (x , z). Then : M Q is a section and

A = dx (
+ iA (x) )
x z
is a linear connection such that h = 0 (a Hermitian connection). Then is the standard
covariant differential and the 2-form F is given by the curvature tensor of A.
5.2. Utiyamas Reduction Theorems for Principal Connections

In [51] Utiyama generalized the results of Subsection 5.1. for a general Lie group G as
a gauge group. A particle field is now a section of a vector bundle associated with
a principle G-bundle P and a gauge field is a principal connection on P . Then any
gauge invariant first order Lagrangian on is given by a gauge invariant Lagrangian of the
curvature tensor R[], i.e.
L(j 1 ) = L(R[])
e . (5.3)
This result is in literature cited as the Utiyamas theorem.

Further, Utiyama generalized the minimal coupling principle as the invariant interac-
tion of the particle field and the gauge field in the form
L(, j 1 ) = L(,
e ) . (5.4)
In his original paper [51] Utiyama considered his theorems only locally with gauge
transformations described in coordinates. Later the Utiyamas theorem was reproved by
many authors also globally, see for instance Castrillon, Munoz and Ratiu [2], Eck [6] and
Mangiarotti and Modugno [36]. The Utiyamas results can be very simply generalized
158 Josef Janyska
for operators with values in a gauge-natural bundle of order (1, 0). In this case we shall
use the term Utiyama-like theorem instead of the Utiyamas theorem. The Utiyama-like
theorem was proved (in order 1) in Kolar, Michor and Slovak [24]. A generalization of the
invariant interaction (5.4) was proved globally by Betounes [1] who proved L(j 1 , j 1 ) =
L(R[],
e , ).
5.3. First Reduction Theorem for General Linear Connections

In Janyska and Modugno [20], see also Janyska [12, 13, 14], we have studied second or-
der natural quantum Lagrangians and second order natural Schrodinger operators on the
quantum bundle, i.e. we studied second order operators on a gravitational field (a classical
symmetric connection on spacetime), an electromagnetic potential (a quantum connection,
i.e. a general linear connection on the quantum bundle preserving the Hermitian product)
and a wave function (a section of the quantum bundle). In both situations such operators are
factorized through the covariant differentials of sections of the quantum bundle where the
first order covariant differentials are given by the quantum connection only but the second
order covariant differentials are given by both quantum and spacetime connections. This
fact was a motivation how to generalize the reduction theorems, see Subsection 3.1., for
general linear connections on vector bundles.
Let E M be a vector bundle with a m-dimensional base and n-dimensional fibres.
Local linear fiber coordinate charts on E will be denoted by (x , y i ).
We define a linear connection on E to be a linear splitting K : E J 1 E . Considering
the contact morphism J 1 E T M T E over the identity of T M , a linear connection
can be regarded as a T E-valued 1-form K : E T M T E projecting onto the identity
of T M . The coordinate expression of a linear connection K is of the type
K = d + Kj i y j i , Kj i C (M, R) .

with
Linear connections can be regarded as sections of a (1,1)-order G = Gl(n, R)-gauge-
natural bundle Lin E M described in Example 4.7..2.
The curvature of a linear connection K on E turns out to be the vertical valued 2form
R[K] = [K, K] : E V E 2 T M , where [, ] is the Froelicher-Nijenhuis bracket. If
we consider the identification V E = E E and linearity of R[K], the curvature R[K] can
M
be considered as the curvature tensor field R[K] : M E E 2 T M and
R[K] : C (LinE) C (E E 2 T M )
is a natural differential operator which is of order one.
p,r
Let us set Eq,s = p E q E r T M s T M . Then a classical connection on
M and a linear connection K on E induce the linear tensor product connection Kqp rs =
p,r
p K q K r s on Eq,s
Kqp rs : Eq,s
p,r
T M T Eq,s
p,r
M
p p,r p,r
which can be considered as a linear splitting Kq rs : Eq,s J 1 Eq,s . Let
p,r
C (Eq,s ). We define the covariant differential of with respect to the pair of connec-
p,r
tions (K, ) as a section of Eq,s T M defined by, Janyska [16],
(K,) = j 1 (Kqp rs ) .
The iterated rth order covariant differential applied on the curvature tensor of the linear
connection K is a natural differential operator which is of order (r 1) with respect to the
classical connection and of order (r + 1) with respect to the linear connection K. Let
(s) (r)
us denote by CLr E the image of this operator and by CC M M CL E the (s, r)-order
curvature bundle of classical and linear connections given as the image of the pair of the
operators ((s+1) R[], (r+1) R[K]), s r 2, (s) = (id, , . . . , s ), defined on
Cla M Lin E. Let us assume a (1, 0)-order Gl(n, R)-gauge natural bundle F E, then the
first reduction theorem for linear and classical connections can be formulated as follows,
Janyska [15].
Theorem 5.1. Let s r 2, r 0. All natural differential operators
D : C (Cla M Lin E) C (F E)
M
which are of order s with respect to classical symmetric connections and of order r with
respect to general linear connections are of the form
D(j s , j r K) = D(
e (s1) R[], (r1) R[K])
where D
e is a zero order natural operator
e : C (C (s1) M C (r1) E) C (F E) .
D C L
M
5.4. Second Reduction Theorem for General Linear Connections

Let us assume the kth order covariant differential of sections of Eqp11,q,p22 (particle fields). It
is a natural operator of order k with respect to sections of Eqp11,q,p22 and of order (k 1) with
respect to classical and linear connections. Let us define the k-th order Ricci bundle Z (k) E
as the image of the triplet of the operators ((k2) R[], (k2) R[K], (k) ) defined on
Cla M Lin E Eqp11,q,p22 . Then the second reduction theorem for linear and classical
connections can be formulated as follows, Janyska [15].
Theorem 5.2. Let s, r k 1, s r 2. All natural differential operators
D : C (Cla M Lin E Eqp11,q,p22 ) C (F E)

M M
which are of order s with respect to classical symmetric connections, of order r with respect
to general linear connections and of order k with respect to sections of Eqp11,q,p22 are of the
form
D(j s , j r K, j k ) = D(
e (s1) R[], (r1) R[K], (k) )
where D
e : C ((C (s1) M C (r1) E)

D Z (k) E) C (F E) .
C L
M (k2)
CC
(k2)
M CL E
M
160 Josef Janyska
5.5. Higher Order Utiyamas Theorem

Higher order local version of the Utiyama-like theorem was studied by Horndeski [10] who
generalized the replacement theorem of Thomas and his collaborators, [49, 50], for gauge
fields. The results obtained in [10] are local and not complete since only concomitants
obtained from the covariant differentials of the curvature tensor of the gauge field are as-
sumed, while concomitants obtained from classical connections on the base only are not
considered. By using the methods of gauge-natural bundles we obtain complete and global
coordinate free description of higher order Utiyama-like theorem.
Let G be an n-dimensional Lie group, P Ob PBm (G), be a principal connection
on P and ad P is the adjoint vector bundle associated with the principal bundle P . Then
we have the induced adjoint linear connection ad() on ad P . If has the coordinate
expression

= d ( + a (x) B
ea ) , (5.5)
x
then ad() has, in the induced fibered coordinates (x , ua ) on ad P , the coordinate expres-
sion

a b d
ad() = d + cbd (x) u . (5.6)
x ua
The curvature tensors of principal connections are given by a 1-order natural operator
from Pri P into ad P 2 T M . The covariant differential of the curvature tensor R[]
with respect to and a classical connection on the base M is then defined as the covariant
differential with respect to ad() and , see Subsection 5.2. and Janyska [17]. Then the
iterated rth order covariant differential r R[] is a natural operator on Cla M Pri P
which is of order (r 1) with respect to classical connections and of order (r + 1) with
(s) (r)
respect to principal connections. Let us denote by CC M CP P , s r 2, (s, r)-
order curvature bundle for classical and principal connections obtained as the image of the
pair of the operators ((s) R[], (r) R[]) defined on Cla Pri P . Then higher order
Utiyama-like theorem for principal and classical connections can be formulated as follows,
Janyska [17].
Theorem 5.3. Let s r 2, r 0, and let F be a (1, 0)-order G-gauge-natural bundle
functor. All natural differential operators
D : C (Cla M Pri P ) C (F P )
M
which are of order s with respect to classical symmetric connections and of order r with
respect to principal connections are of the form
D(j s , j r ) = D(
e (s1) R[], (r1) R[])
where D
e : C (C (s1) M C (r1) P ) C (F P ) .
D C P
M
Remark 5.4. The curvature bundle of classical symmetric and principal connections is given
by identities depending on the structure constants of the group G. So all natural operators
defined on the curvature bundle depend also on the structure constants, i.e.
D(j s , j r ) = D(c,
e (s1) R[], (r1) R[]) .
For instance cbab r . . . 1 Ra is an example of a natural tensor field of the type (0, r+2)
on M given by (in order (r 1)) and (in order (r + 1)). In the case of (general) linear
connections the structure constants are given by the Kronecker deltas and they contract with
the curvature tensor fields, i.e. they are not visible.
5.6. Higher Order Utiyamas Invariant Interaction

Let E M be a vector bundle obtained as the bundle associated with P with respect to
a linear representation : G Gl(n, R), = (ij (g)), and let us denote by the set
i
of constants ija = gja |e, g a beeing the local coordinates on G and e G is the unit
element. Then any principal connection on P induces the linear connection () on E by
()j i = ijb b .
Let us assume Eqp11,q,p22 defined as in Subsection 5.3., then the covariant differential of
sections of Eqp11,q,p22 (particle fields) with respect to the pair of connection (, ) is defined as
the covariant differential with respect to the pair of connections ((), ), see Subsection
5.3.. The k-order covariant differential is then a natural differential operator of order k with
respect to sections of Eqp11,q,p22 and of order (k 1) with respect to classical and principal con-
nections. Let us define the k-th order Ricci bundle Z (k) P as the image of the triplet of the
operators ((k2) R[], (k2) R[K], (k) ) defined on Cla M Lin E Eqp11,q,p22 . Then
the higher order invariant interaction of particle fields, principal and classical symmetric
connections can be formulated as follows, Janyska [18].
Theorem 5.5. Let s, r k 1, s r 2. All natural differential operators
D : C (Cla M Pri P Eqp11,q,p22 ) C (F P )

M M
which are of order s with respect to classical symmetric connections, of order r with respect
to principal connections and of order k with respect to sections of Eqp11,q,p22 are of the form
D(j s , j r , j k ) = D(
e (s1) R[], (r1) R[], (k) )
where D
e : C ((C (s1) M C (r1) P )

D Z (k) P ) C (F P ) .
C P
M (k2)
CC
(k2)
M CP P
M
Remark 5.6. The Ricci identities and the identities on the curvature bundle of classical
symmetric and principal connections depend on the structure constants of the group G and
the constants . So all natural differential operators defined on the Ricci bundle depend
also on the structure constants and , i.e.
e , (s1) R[], (r1) R[]) .
D(j s , j r ) = D(c,
162 Josef Janyska
5.7. Gauge Invariant Lagrangians

As direct consequences of Theorems 5.3 and 5.5 we obtain higher order gauge invariant
Lagrangians in the form.
Theorem 5.7. All natural Lagrangians of orders s in , r in and k in are of the form
L(j s , j r ) = L(c,
e (s1) R[], (r1) R[]) ,
e , c, (s1) R[], (r1) R[], (k) ) ,
L(j s , j r , j k ) = L(
where Le is a unique zero order gauge invariant Lagrangian.
Remark 5.8. Gauge invariant Lagrangians on classical symmetric connections (in order s)
and principal connections (in order r) are infinitesimally given as solutions of the system of
partial differential equations given by
LJ s Cla +J r P ri L = 0 (5.7)
for all right G-invariant vector fields on P over on M , where Cla is given by (2.1)
and P ri is given by (4.1). From Theorem 5.3 it follows that all solutions of (5.7) factorize
through the curvature tensors of both connections and their covariant differentials up to the
orders (s 1) and (r 1).
Example 5.9. Let (M, g) be a (pseudo-)Riemannian oriented manifold. Any natural La-
grangian of g, and is given by an gauge invariant Lagrangian function
l( , c, (s2) R[(g)], (r1) R[], g, (k) ) .

e
All natural Lagrangians used in fields theories are of this type (for r, s 1), Fatibene and
Francaviglia [9]. It is easy to see that the invariant Lagrangian function
l = g 1 1 g 2 2 g 1 1 . . . g r r cdae cebd Ra 1 2 ;1 ;...;r Rb 1 2 ;1 ;...;r

e
defines a higher order Yang-Mills Lagrangian which is of order (r+1) with respect to and
of order r with respect to g. For r = 0 we obtain just the classical Yang-Mills Lagrangian.
5.8. Final Remarks

Let us note that in literature as gauge invariant operators are sometimes considered operators
invariant with respect to principal bundle morphisms over the identity of base manifolds,
see for instance Etayo, Garca, Munoz and Perez [8] and Manno, Pohjanpelto and Vitolo
[37]. In this case the corresponding mapping between standard fibres is equivariant with
respect to the subgroup {e} Tm r G Gs T r G. Then for r-order (r 2) operators
m m
on the gauge-natural bundle of principal connections it is not necessary to use auxiliary
classical connections on base manifolds and all r-order natural differential operators (with
values in a gauge-natural bundle of order (r + 1, 0)) factorize through gauge covariant
differentials of the curvature tensor up to the order r. The gauge covariant differentiation
was defined intrinsically by Eck [6], for local formulas see also normal gauge concomitants
by Horndeski [10]. For instance the gauge covariant differential of the curvature tensor R[]
is given in coordinates by

R[]a = R[]a ; = R[]a cabd b R[]a
and it is a (2, 0)-order field.

Then we have
Theorem 5.10. All Lagrangians of orders r in and k in which are invariant with respect
to principal bundle morphisms over the identity of the base manifold are of the form
L(j r ) = L(c,
e (r1) R[]) ,
e , c, (r1) R[], (k) ) ,
L(j r , j k ) = L(
where Le is a unique zero order Lagrangian.
For instance, if in Example 5.9 we replace covariant differentials with respect to (, )

by gauge covariant differentials with respect to only, then the corresponding higher-order
Yang-Mills Lagrangian is invariant with respect to principal bundle morphisms over the
identity of the base manifold.
Acknowledgement
This research has been supported by the Ministry of Education of the Czech Republic under
the project MSM0021622409, by the Grant agency of the Czech Republic under the project
GA201/05/0523.
References
[1] D. Betounes, The geometry of gauge-particle field interaction: a generalization of
Utiyamas theorem, J. Geom. Phys. 6 (1989) 107125.
[2] M. Castrillon Lopez, J. Munoz Masque and T. Ratiu, Gauge invariance and variational
trivial problems on the bundle of connection, Diff. Geom. Appl. 19 (2003) 127145.
[3] E. B. Christoffel, Ueber die Transformation der homogenen Differentialausdrucke

zweiten Grades, Journal fur die reine und angewandte Mathematik, Crelless Jour-
nals 70 (1869) 4670.
[4] J. A. Dieudonne and J. B. Carrell, Invariant Theory, Old and New (Academic Press,
New-York-London, 1971).
[5] W. Drechsler and M. E. Mayer, Fibre Bundle Techniques in Gauge Theories (Springer-
Verlag, New York, 1977).
[6] D. E. Eck, Gauge-natural bundles and generalized gauge theories, Mem. Amer. Math.
Soc. 33 (247) (1981).
164 Josef Janyska
[7] D. B. A. Epstein and W. P. Thurston, Transformation groups and natural bundles, Proc.
London Math. Soc. (3) 38 (1979) 219236.
[8] F. Etayo Gordejuela, P. L. Garca Perez, J. Munoz Masque and J. Perez Alvarez,
Higher-order Utiyama-Yang-Mills Lagrangians, J. Geom. Phys. 57 (2007) 10891097.
[9] L. Fatibene and M. Francaviglia, Natural and Gauge Natural Formalism for Classi-
cal Field Theories, a Geometric Perspective including Spinors and Gauge Theories
(Kluwer Academic Pub. 2003).
[10] G. W. Horndeski, Replacement Theorems for Concomitants of Gauge Fields, Utilitas

Math. 19 (1981) 215146.
[11] J. Janyska, Geometrical properties of prolongation functors, Cas. pest. mat. 110
(1985) 7786.
[12] J. Janyska, Natural quantum Lagrangians in Galilei quantum mechanics, Rendiconti

di Matematica (Roma), Serie VII 15 (1995) 457468.
[13] J. Janyska, Natural Lagrangians for quantum structures over 4-dimensional space-
time, Rendiconti di Matematica (Roma), Serie VII 18 (1998) 623648.
[14] J. Janyska, A remark on natural quantum Lagrangians and natural generalized

Schrodinger operators in Galilei quantum mechanics, In: The Proceedings of the
Winter School Geometry and Topology (Srn, 2000, Supplemento ai Rendiconti del
Circolo Matematico di Palermo, Serie II No. 66, 2001) 117128.
[15] J. Janyska, Reduction theorems for general linear connections, Diff. Geom. Appl. 20
(2004) 177196.
[16] J. Janyska, On the curvature of tensor product connections and covariant differentials,
In: The Proceedings of the 23rd Winter School Geometry and Physics (Srn, 2003,
Supplemento ai Rendiconti del Circolo Matematico di Palermo, Serie II - Numero 72,
2004) 135143.
[17] J. Janyska, Higher order Utiyama-like theorem, Rep. Math. Phys. 58 (2006) 93118.
[18] J. Janyska, Higher order Utiyama invariant interaction, Rep. Math. Phys. 59 (2007)
6381.
[19] J. Janyska and M. Modugno, Infinitesimal natural and gauge-natural lifts, Diff. Geom.
and its Appl. 2 (1992) 99121.
[20] J. Janyska and M. Modugno, Covariant Schroedinger operator, J. Phys. A: Math. Gen.
35 (2002) 84078434.
[21] A. A. Kirillov, Invariant operators over geometric quantities, (Russian), Current prob-
lems in mathematics 16, VINITI, Moscow (1980) 329.
[22] I. Kolar, Prolongations of generalized connections, In: Diff. Geom. (Math. Soc. Janos
Bolyai, 31, Budapest Hungary, North Holland, 1979) 317325.
[23] I. Kolar, Some natural operators in differential geometry, In: Proc. Conf. Diff. Geom.
and Its Appl. (Brno 1986, D. Reidel, 1987) 91110.
[24] I. Kolar, P. W. Michor and J. Slovak, Natural Operations in Differential Geometry
(SpringerVerlag, 1993).
[25] I. Kolar and J. Slovak, On the geometric functors on manifolds, Suppl. Rend. Circ.
Mat. Palermo 21 (1989) 223233.
[26] D. Krupka, A setting for generally invariant Lagrangian structures in tensor bundles,
Bull. Acad. Polon. Sci., Ser. Math. Astronom. Phys. 22 (1974) 967972.
[27] D. Krupka, A theory of generally invariant Lagrangians for the metric fields I, Internat.
J. Theoret. Phys. 17 (1978) 359368.
[28] D. Krupka, A theory of generally invariant Lagrangians for the metric fields II, Inter-
nat. J. Theoret. Phys. 15 (1976) 949959.
[29] D. Krupka, Elementary theory of differential invariants, Arch. Math. 4, Scripta Fac.
Sci. Nat. UJEP Brunensis, XIV, (1978) 207214.
[30] D. Krupka, Prirozene Lagrangeovy struktury, DrScthesis, Prrodovedecka fakulta
UJEP, Brno, 1978.
[31] D. Krupka, Natural Lagrangian structures, Banach Center Publications 12 (1984)
185210.
[32] D. Krupka, Local invariants of a linear connection, In: Diff. Geom. (Coll. Math. Soc.
Janos Bolyai 31, Budapest, Hungary, 1979, North Holland, 1982) 349369.
[33] D. Krupka and J. Janyska, Lectures on Differential Invariants (Folia Fac. Sci. Nat.
Univ. Purkynianae Brunensis, Brno 1990).
[34] D. Krupka and V. Mikolasova, On the uniqueness of some differential invariants: d,
{ , }, , Czechoslovak Math. J. 34 (1984) 588597.
[35] D. Krupka and A. Trautman, General invariance of Lagrangian structures, Bull. Acad.
Polon. Sci., Ser. Math. Astronom. Phys. 22 (1974) 207211.
[36] L. Mangiarotti and M. Modugno, On the geometric structure of gauge theories, J.
Math. Phys. 26 (1985) 13731379.
[37] G. Manno, J. Pohjanpelto and R. Vitolo, Gauge invariance, charge conservation, and
variational principles, preprint 2007.
[38] W. Mikulski, There exists a prolongation functor of infinite order, Cas. pest. mat. 114
(1989) 5759.
[39] A. Nijenhuis, Theory of the geometric object, Thesis, University of Amsterdam, 1952.
[40] A. Nijenhuis, Natural bundles and their general properties, In: Diff. Geom. (in honour
of K. Yano, Kinokuniya, Tokyo, 1972) 317334.
166 Josef Janyska
[41] J. Novotny, On the generally invariant Lagrangians for the metric field and other tensor
fields, Internat. J. Theoret. Phys. 17 (1978) 677684.
[42] R. S. Palais and C. L. Terng, Natural bundles have finite order, Topology 16
(1977) 271277.
[43] D. J. Saunders, The Geometry of Jet Bundles (London Math. Soc., Lecture Note Series
142, Cambridge University Press, 1986).
[44] J. A. Schouten, Ricci Calculus (Berlin-Gottingen, 1954).
[45] J. A. Schounten and J. Haantjes, On the theory of the geometric object, Proc. London
Math. Soc. 42 (1937) 356376.
[46] J. Slovak, On finite order of some operators, In: Proc. Conf. Diff. Geom. and Its Appl.
(Communications, Brno 1986) J. E. Purkyne University, 1987 283294.
[47] J. Slovak, On natural connections on Riemannian manifolds, Comment. Math. Univ.

Carolinae 30 (1989) 389393.
[48] C. L. Terng, Natural vector bundles and natural differential operators, Am. J. Math.
100 (1978) 775828.
[49] T. Y. Thomas, Differential Invariants of Generalized Spaces (Cambridge University

Press, Cambridge, 1934).
[50] T. Y. Thomas and A. D. Michal, Differential invariants of affinely connected mani-

folds, Ann. Math. 28 (1927) 196236.
[51] R. Utiyama, Invariant theoretical interpretation of interaction, Phys. Rev. 101

(1956) 15971607.
[52] H. Weyl, Raum - Zeit - Materie (Space - Time - Matter) (1918, translated from the 4th
German Edition, London: Methuen 1922).
[53] A. Zajtz, Foundations of Differential Geometry on Natural Bundles (Caracas, 1983).
[54] A. Zajtz, The sharp upper bound on the order of natural bundles of given dimensions,
Bull. Soc. Math. Belgique 3 (1987) 347357.
[55] C. N. Yang and R. L. Mills, Conservation of isotopic spin and isotopic gauge invari-
ance, Physical Rev. (2) 96 (1954) 191195.
Chapter 10
C ONNECTIONS ON H IGHER O RDER F RAME

B UNDLES AND T HEIR G AUGE A NALOGIES
Ivan Kolar
Institute of Mathematics and Statistics
Faculty of Science, Masaryk University
Janackovo nam. 2a, 602 00 Brno
Czech Republic
Abstract
In the first part of the paper, we present a survey of the basic properties of con-
nections on the r-th order frame bundle of a manifold. Special attention is paid to the
torsion and torsion-free connections. In the second part, connections on the r-th prin-
cipal prolongation of a principal bundle are treated from similar points of view. The
case of the first principal prolongation is discussed in detail.
2000 Mathematics Subject Classification. 58A20, 53C05, 58A32

Key words and phrases. r-th order frame bundle, connection, torsion, natural operator,
semiholonomic 2-jet, r-th principal prolongation of principal bundle, gauge-natural opera-
tor
In the present paper, connection means a principal connection on a principal bundle

unless otherwise specified.
Several properties of connections on the r-th order frame bundle P r M of a manifold
M appear in the framework of the general theory of natural bundles and operators. This is
described in the book by D. Krupka and J. Janyska, [21], and in the monograph [18]. So the
first part of the present paper is devoted to a survey of some more specific properties of con-
nections on P r M that are mostly related with the idea of torsion. In Section 1 we underline
that the Lie algebroid version of a connection on the principal bundle P r M (M, Grm ) is a
linear r-th order connection on T M . So we have two different approaches to the concept
of torsion. Proposition 2 reads that both approaches are equivalent.

To Demeter Krupka, on the occasion of his 65th birthday

E-mail address: kolar@math.muni.cz
168 Ivan Kolar
In Section 3 we clarify that the torsion-free connections on P r M are in bijection with

the reductions of P r+1 M to the canonical injection of G1m into Gr+1 m . This enables us to
define the r-th exponential operator transforming every torsion-free connection on P 1 M
into a torsion-free connection on P r M . In particular, this implies that determines a gen-
eral connection on every natural bundle over m-manifolds. In Section 5, the Lie algebroid
construction of the exponential operator is based on two interesting lemmas concerning the
r-jet of the commutator of vector fields on M . Then we deduce that the torsion-free con-
nections on P r M are in bijection with the splittings from the cotangent bundle T M into
the bundle of (1, r + 1)-covelocities on M .
In Section 7 we discuss a connection on P r M from the viewpoint of the theory of higher
order G-structures and we characterize its integrability in this sense. In Section 8 we present
a recent result by W. Mikulski, [27], who determined all natural operators transforming a
torsion-free connection on P 1 M into a connection on P r M . Section 9 is devoted to the
basic properties of semiholonomic 2-jets, that represent a useful tool for several problems
of the present paper.
The principal prolongation W r P of an arbitrary principal bundle P (M, G) is defined
in Section 10 in a formally slightly different way to [18]. We hope this could be useful
in applications. Then we summarize the basic properties of connections on W r P and their
torsions. In Section 11 we clarify that every connection on W 1 P is canonically identified
with the triple (, , D) of a connection on P , a connection on P 1 M and a section D
2
of L0 P T M , where L0 P is the adjoint bundle of P . In Section 12 we present the list
N
of all gauge-natural operators transforming every pair (, ) of a connection on P and

a torsion-free connection on P 1 M into a connection on W 1 P . Finally we outline how
the semiholonomic 2-jets can be used in the theory of connections on W 1 P . In particular,
we introduce the conjugate connection to every connection on W 1 P by using the
canonical involution of semiholonomic 2-jets.
All manifolds and maps are assumed to be infinitely differentable. Unless otherwise
specified, we use the terminology and notations from the book [18].
1. The Algebroid form of Connections on P r M

First we recall that the Lie algebroid LP M of an arbitrary principal bundle P (M, G) is
defined by LP = T P/G. So the elements of LP are the right invariant families of tangent
vectors along the individual fibers of P , every section : M LP is identified with a
right invariant vector field : P T P and the bracket [ 1 , 2 ] of right invariant vector
fields on P induces the bracket [[1 , 2 ]] of LP . The canonical projection q : LP T M is
called the anchor map. Clearly, a connection on P can be interpreted as a linear morphism
: T M LP satisfying q = id T M , [26].
We write P r M for the r-th order frame bundle of an m-dimensional manifold M .
This is a principal bundle over M with structure group Grm = inv J0r (Rm , Rm )0 . Every
local diffeomorphism f : M1 M2 induces a principal bundle morphism P r f : P r M1
P r M2 , so that P r is a bundle functor on the category Mfm of m-dimensional manifolds
and their local diffeomorphisms, [18]. For every vector field X : M T M , its flow
Connections on Higher Order Frame Bundles and Their Gauge Analogies 169
prolongation
r
P r X := P (F ltX ) (1)
t 0
is a right invariant vector field on P r M . This follows directly from the fact that the values of
P r are in the category PBm (Grm ) of principal Grm -bundles over m-manifolds and their local
principal bundle isomorphisms. Since P r is an r-th order bundle functor, the restriction
P r X | Pxr M depends on jxr X only, x M .
Proposition 1. The rule
r
IM (jxr X) = P r X | Pxr M (2)
identifies J r T M with LP r M .
Proof. We have to prove that IMr is a diffeomorphism. But P r M = reg T r M is an open
m
subset of the bundle Tmr M of all (m, r)-velocities on M and P r X is the restriction of
the flow prolongation Tmr X to this subset. Hence the bijectivity of IM

r follows from the
existence of an exchange isomorphism M : Tm T M T Tm M such that Tmr X = M

r r
Tmr M , [16], [18].
A linear r-th order connection on T M is a linear morphism T M J r T M that splits

the target jet projection. According to Proposition 1, every connection : P r M J 1 P r M
is identified with a linear splitting : T M J r T M . We say that is the algebroid form
of .
2. Two Approaches to the Torsion on P r M

The canonical (Rm gr1 r m
m )-valued 1-form r on P M is defined as follows. We have R
r1 r1 m r1 r m
gm = Ter1 P R , where er1 = j0 id Rm . Every u = j0 f , f : R M , induces
P r1 f : P r1 Rm P r1 M . The tangent map u := Ter1 P r1 f : Ter1 P r1 Rm
Tur1 P r1 M , ur1 = r1
r (u), depends on u only. Then one defines
r (A) = u1 T r1
r
A Tu P r M .

(A) ,
P. C. Yuen introduced the torsion of a connection on P r M as the exterior covariant

differential D r of r , [33]. Since D r is a horizontal 2-form on P r M , it can be
interpreted as a map P r M (Rm gr1 2
m ) T M . Taking into account the identification
m r
u1 : R Tx M , u1 = 1 (u), we construct
D r : P r M (Rm gr1 2 m
m ) R . (3)
On the other hand, the (r 1)-jet at x M of the bracket [, ] of two vector fields ,
on M depends on the r-jets jxr and jxr . This defines a map
[ , ]r1 : J r T M M J r T M J r1 T M .
Let : T M J r T M be the algebroid form of . According to A. Zajtz, [28], the torsion

of is a map : T M M T M J r1 T M defined by

(A, B) = (A), (B) r1
, A, B Tx M . (4)
170 Ivan Kolar
Clearly, can be interpreted as a section of J r1 T M 2 T M . This is a fiber bundle

associated to P r M with standard fiber (Rm gr1m ) R
2 m . So the frame form of is
a map
: P r M (Rm gr1m ) R
2 m
.
In [14], we deduced
Proposition 2. We have D r = 12 .
Let and be two connections on P r M over the same connection on P r1 M . Con-
sider their algebroid forms , : T M J r T M . Since the kernel of r1 r : J rT M
J r1 T M is T M S r T M , [18], the difference of and is a section
: M T M Sr T M T M . (5)
Proposition 3. If both and are torsion-free, then the values of lie in T M

S r+1 T M .
Proof. If xi are some local coordinates on M , X i = dxi are the induced coordinates on
T M and Xi are the jet coordinates on J r T M , then the equations of or are
Xi = ij (x)X j or Xi = ij (x)X j , 1 || r ,
where is a multi-index of range m. The difference can be interpreted as a map

T M M T M T M S r1 T M of the form
ijk ijk j k ,

|| = r 1 .
The only r-th order terms in j r1 [, ] are
r i r i
j
j , || = r 1 . (6)
xj x xj x
If is torsion-free, then (6) yields ijk = ikj . If is also torsion-free, (5) is symmetric
in the last two subscripts.
From the proof one sees directly that is an arbitrary section of T M S r+1 T M .
3. Torsion-Free Connections on P r M as Reductions of P r+1 M

Every a G1m is a matrix, which defines a linear map l(a) : Rm Rm . This induces a
group homomorphism
lr1 : G1m Grm , lr1 (a) = j0r l(a) .
S. Kobayashi proved, [8], that the torsion-free connections on P 1 M are in bijection with
the reductions of P 2 M to the subgroup l1 (G1m ) G2m . We deduce an analogous result for
arbitrary order r. This is based on the following injection irM : P r+1 M J 1 P r M . Every
u = j0r+1 f P r+1 M determines a local section of P r M M
(y) = j0r f tf 1 (y) ,

(7)
where y lies in a neighbourhood of f (0) M and tf 1 (y) : Rm Rm is the translation

x 7 x + f 1 (y). Then we set irM (u) = jf1(0) . If xi , xij , . . . , xiji ...jr are the standard coor-
dinates on P r Rm , xij,k , . . . , xij1 ...jr ,k are the induced coordinates on J 1 P r Rm and xij1 ...jr+1
are the additional coordinates on P r+1 Rm , then (7) implies directly the following coordi-
nate form of ir
xij,k = xijl xlk , . . . , xij1 ...jr ,k = xij1 ...jr l xlk , (8)
where xij is the inverse matrix to xij .
Every X J 1 P r M over X P r M is identified with an m-plane in the tangent
space TX P r M , which will be denoted by the same symbol X. Hence we can consider the
restriction dr | X of the exterior differential of r to X. Denote by X1 J 1 P r1 M the
underlying element of X. The following lemma from [14] is close to a result by Yuen, [33].
Lemma 1. Let X J 1 P r M satisfy X1 = ir1 r
M (X). Then X iM (P
r+1 M ) if and only
if dr | X = 0.
For r = 1 we have no X1 and the claim d1 | X = 0 if and only if X i1M (P 2 M )
was used in [8].
For every torsion-free connection on P r M we define a map () : P 1 M P r+1 M
by the following induction. Consider a connection : P r M J 1 P r M such that the
underlying connection 1 : P r1 M J 1 P r1 M is torsion-free, so that 1 determines a
map (1 ) : P 1 M P r M be the induction hypothesis.
Proposition 4. is torsion-free, if and only if the values of (1 ) lie in irM (P r+1 M ).
Then we define () = (irM )1 (1 ) : P 1 M P r+1 M .
Proof. By Lemma 1, we have dr | X = 0 for all X (1 )(P 1 M ) . But r is a

pseudotensorial form, [18], so that dr | A = 0 holds for every A (P r M ). This is

equivalent to D r = 0.
For every principal bundle P (M, G), we have an induced right action of G on J 1 P ,
jx1 s(y), g 7 jx1 s(y)g , where s is a local section of P on a neighbourhood of x M

and g G. This action will be denoted by (X, g) 7 X(g).

Lemma 2. For every v P r+1 M and a G1m , we have
irM vlr (a) = irM (v) lr1 (a) .

Proof. If v = j0r+1 f , then irM (v) lr1 (a) = jx1 j0r f tf 1 (y) l(a) . On the other

hand, irM vlr (a) = jx1 j0r f l(a) tl(a)1 (f 1 (y)) . But tz l(a) = l(a) tl(a)1 (z) ,

z Rm , is a well known relation from the affine geometry.
By Lemma 2, ()(P 1 M ) is a reduction of P r+1 M to the subgroup lr (G1m ) Gr+1

m .
Indeed, using induction we obtain
()(ua) = (irM )1 (1 )(u) lr1 (a)

= ()(u)lr (a) .
On the other hand, every reduction Q P r+1 M to the subgroup lr (G1m ) induces a map
(denoted by the same symbol) Q : P 1 M P r+1 M as follows. For every v Q we
172 Ivan Kolar
construct u = 1r+1 (v) and we set Q(u) = v. Any other v in the same fiber of Q M
is of the form v = vlr (a), a G1m . This implies 1r+1 (v) = ua, so that our definition is
correct.
Proposition 5. Proposition 4 establishes a bijection between torsion-free connections on

P r M and reductions of P r+1 M to lr (G1m ).
Proof. First we deduce that () : P 1 M P r+1 M is a reduction to lr (G1m ). For every

u P 1 M and a G1m we have
()(ua) = (irM )1 (1 )(ua) = (irM )1 (1 )(u)lr1 (a)

= (irM )1 (1 )(u) lr1 (a) = (irM )1 (irM ) ()(u) lr (a)

by definition, by the induction hypothesis, by right-invariance of and by Lemma 2. Con-

versely, if Q : P 1 M P r+1 M is a reduction to lr (G1m ), then Q1 = rr+1 Q : P 1 M
P r M is a reduction to lr1 (G1m ). We define : Q1 (P 1 M ) J 1 P r M by Q1 (u) =
irM Q(u). By Lemma 2, it holds Q1 (ua) = irM Q(ua) = irM Q(u)lr (a) =
r

iM Q(u) lr1 (a) = Q1 (u) lr1 (a) . Hence is a right-invariant map, which
is canonically extended into a connection on P r M .
4. The r-th Order Exponential Prolongation

The following construction represents an interesting application of Proposition 5. Consider
a torsion-free connection on P 1 M . For every x M , determines the exponential map
expx : Ux M , where Ux Tx M is a neighbourhood of the origin. Then we define a
map Er () : P 1 M P r+1 M by
Er ()(u) = j0r+1 (exp

x u) , u Px1 M , (9)
where u is interpreted as a map Rm Tx M .
Proposition 6. Er ()(P r M ) is a reduction of P r+1 M to lr (G1m ).
Proof. For all u P 1 M and a G1m , we have Er ()(ua) = j0r+1 exp

x u l(a) =
Er ()(u)lr (a).
By Proposition 5, Er () is a torsion-free connection on P r M , that is called the r-th

exponential prolongation of . The rule 7 Er () is said to be the r-th order exponential
operator on the bundle Q P 1 M of torsion-free connections on P 1 M .
W. Mikulski invented another construction of the exponential prolongation, [27]. Every
X Tx M is extended into a vector field X on Tx M by means of translations. The expo-
nential map exp
x transforms X locally into a vector field (expx ) (X) on M . Then we can
construct
r ()(X) = j0r (exp r

x ) (X) Jx T M . (10)
In Section 5 we deduce that r () is the algebroid form of Er ().
This result enables us to describe another geometrically interesting construction of

Er (). The flow prolongation P r (exp r
x ) (X) is a vector field on P M . By Section 1,
r r
the lifting map P M M T M T P M of Er () is
Er ()(u, X) = P r (exp u Pxr M .

x ) (X) (u) ,
Further, consider an r-th order natural bundle F over m-manifolds, [18]. So F M is a

fiber bundle associated to P r M with standard fiber F0 Rm . Every principal connection
on P r M induces a general connection F on P r M . We shall use the construction of F
by means of lifting vector fields. In general, every right-invariant vector field Z on P r M
induces a vector field ZF on F M as follows. If Z(u) = dc(0) r
dt , c : R P M , u = c(0),
then
d
a F0 Rm .

ZF {u, a} = 0 c(t), a ,
dt
Since Z is right-invariant, this definition is correct. Then the F -lift of a vector field X on
M is prescribed by F (X) = (X)F .
Clearly, the flow prolongation FX of a vector field X on M with respect to F satisfies
FX = (P r X)F . Thus we have deduced
Proposition 7. For every r-th order natural bundle F , the rule
Er ()F (v, X) = F (exp

x ) (X) (v) , v Fx M, X Tx M
transforms every torsion-free connection on P 1 M into general connection Er ()F on

FM.
5. The Exponential Prolongation in the Algebroid Form

Consider an arbitrary linear splitting : T M J r T M . For a linear frame u Px1 M ,
u = (A1 , . . . , Am ), Ai Tx M , we take vector fields Xi satisfying jxr Xi = (Ai ), i =
1, . . . , m. Then
F ltX1 1 F ltXmm (x)

is a local map Rm M and we define
()(u) = j0r+1 F ltX1 1 F ltXmm (x) Pxr+1 M .

(11)
One verifies easily that ()(u) depends on u and only.

Proposition 8. If is torsion-free, then ()(P 1 M ) is a reduction of P r+1 M to lr (G1m ).
Proof is based, in a very instructive way, on the definition (4) of . We shall use the
following two lemmas from [15].
Consider two vector fields X and Y on M . Then F ltX F lY (x) is a local map

r+1
R2 M , so that j0,0 F ltX F lY (x) (T2r+1 M ) is a (2, r + 1)-velocity on M .
Lemma 3. If jxr1 [X, Y ] = 0, then

r+1 r+1
F ltX F lY (x) = j0,0 F lY F ltX (x) .

j0,0 (12)
174 Ivan Kolar
Further, F ltX F lt (x) is a local map R M , so that j0r+1 F ltX F ltY (x)
Y

(T1r+1 M )x is a (1, r + 1)-velocity on M .
Lemma 4. If jxr1 [X, Y ] = 0, then
j0r+1 F ltX F ltY (x) = j0r+1 F ltX+Y (x) .

(13)
We shall also apply the well known formula

X
F lct = F ltcX , c R. (14)
Take a = (aij ) G1m and consider ua = (aji Aj ). Since is torsion-free, by (13), (14) and
(12) we obtain gradually
a1 X1 ++am
1 Xm a1 X1 ++am
m Xm
()(ua) = j0r+1 F lt11

F ltmm
a 1 X1 a m Xm a 1 X1 a m Xm
= j0r+1 F lt11 F lt11 F ltmm F ltmm
= j0r+1 F laX11t1 F laXmmt1 F laX11 tm F laXm
m

m
1 1 m mt
j0r+1 F laX11t1 ++a1 tm F laXmmt1 ++am tm .

=
1 m 1 m
This proves Proposition 8.

To clarify the relation of () to the reduction () from Section 3, we need the fol-
lowing form of the injection irM : P r+1 M J 1 P r M . We have P r M Tm r M . Clearly,
r r m
j0 f Tm M , f : R M , can be expressed in the form
j0r f = (Tm
r
f )(er ) , er = j0r id Rm . (15)
r i r Rm , where i : Rm Rm is the translation t1 =

Write Ei = t j Ter Tm
0 0 t t
t1 , . . . , ti = ti + t, . . . , tm = tm . If we consider j0r+1 P r+1 M , then
r
(T Tm )(Ei ) (16)
is an m-tuple of tangent vectors at j0r P r M . The linear span of these vectors defines
r+1
iM j0 J 1 P r M .
r
Proposition 9. If is torsion-free and is the corresponding connection on P r M , then

() = ().
Proof. We proceed by introduction. If 1 and 1 are the underlying connections in the order
r 1, then (1 ) = (1 ) by the induction hypothesis. Consider u = (A1 , . . . , Am )
Px1 M and write
v = (1 )(u) = (1 )(u) .
By (16), irM j0r+1 F ltX1 1 F ltXmm (x) is the linear span of the vectors

r
F ltX1 1 F ltXmm (Ei ) ,

T Tm i = 1, . . . , m . (17)
Using the basic properties of flows, Lemma 3 and (15), we deduce that (17) is equal to
r Xi
0
Tm F ltX1 1 F lt+t Xm
i F ltm (er )
t
T rX T rX T rX
= 0 F lt m i F lt1m 1 F ltmm m (er )
t
= Tmr Xi Tm r
(F ltX1 1 F ltXmm )(er ) = Tmr Xi (v) .

By (2) and by the induction hypothesis, this m-tuple spans ()(v).
From the proof of Proposition 8 we obtain easily that the construction (10) of r ()
by W. Mikulski is the algebroid form of the exponential prolongation Er () introduced in
Section 4.
6. Splittings T M T r+1 M
The space T r+1 M = J r+1 (M, R)0 of all (1, r + 1)-covelocities on M is a vector bundle,
[18]. By a splitting s : T M T r+1 M we mean a linear morphism satisfying 1r+1
s = id T M . We remark that such splittings play an interesting role in the construction of
Poincare-Cartan morphisms in the higher order variational calculus, [11].
Proposition 10. There is a canonical bijection between reductions Q P r+1 M to sub-

group lr (G1m ) and splittings s : T M T r+1 M .
Proof. Every b Tx M determines a linear map (b) : Tx M R. Let v = j0r+1 f Qx ,

so that u = 1r+1 (v) Px1 M can be interpreted as a map u : Rm Tx M . Then we set
s(b) = jxr+1 (b) u f 1 Txr+1 M .

(18)
Since Q is a reduction to lr (G1m ), (18) does not depend on the choice of v Qx , The fact
that s is a splitting follows directly from (18). Conversely, let s : T M T r+1 M be
a splitting. A frame u Px1 M is a basis (e1 , . . . , em ) of Tx M . Consider the dual basis
u = (e1 , . . . , em ) of Tx M . Then s(e1 ), . . . s(em ) are the components of an (r + 1)-jet
1
s(u ) Jxr+1 (M, R)0 . Write Q(u) = s(u ) Pxr+1 M for the inverse jet. If we
j
take ua = (ai ej ), then (ua) = (aj e ), where aj is the inverse matrix to aij . Hence
i j i
s (ua) = aij s(ej ) = lr (a1 ) s(u ), which implies Q(ua) = Q(u)lr (a). Finally, one

verifies easily that the maps Q 7 s and s 7 Q are inverse each other.
Taking into account Proposition 5, we obtain a canonical bijection between torsion-free

connections on P r M and splittings T M T r+1 M .
We remark that Proposition 10 yields another proof of Proposition 3.
7. The Viewpoint of Higher Order G-Structures

A connection on P r M can be viewed as a kind of higher order G-structure on M . We recall
that a k-th order G-structure on M is said to be integrable, if it is locally isomorphic to the
176 Ivan Kolar
product Rm G, where G Gkm is the structure group. We are going to apply this approach
to the algebroid form : T M J r T M . (We remark that this kind of integrability plays
an important role in our theory of the flow prolongation of some tangent valued forms, [1].)
On Rm , there is a distinguished connection Cr : T Rm J r T Rm defined by
Cr (X) = jxr X , X Tx Rm ,
where X is the constant vector field on Rm constructed from X by means of translations.

Definition 1. We say that : T M J r T M is integrable, if for every x M there exists a
neighbourhood U and a diffeomorphism f : U Rm satisfying Cr T f = J r T f ( | U ).
Clearly, every integrable connection is torsion-free. According to a classical result,
a connection on P 1 M is integrable, iff it is both torsion-free and curvature-free. Then
every exponential prolongation k () is also integrable.
Thus, the torsion of is the first obstruction to its integrability. Consider the underlying
connections k = kr , k = 1, . . . , r. Clearly, if is torsion-free or integrable, then each
k is so. Assume that is torsion-free. Then the curvature of 1 is the second obstruction
to the integrability of . If this curvature vanishes, each connection k (1 ) is integrable.
The difference
2 2 (1 )
is a tensor field of type T M S 3 T M that is the third obstruction to the integrability of
. Assume by induction that the first up to (k + 1)-st obstruction to the integrability of
vanish. Then k = k (1 ) and the tensor field of type T M S k+2 T M
k+1 k+1 (1 )
is the next obstruction to the integrability of . If all these r + 1 obstructions vanish, then
= r (1 ) is integrable. Thus, we have proved
Proposition 11. is integrable if and only if all the following conditions are satisfied
a) is torsion-free,
b) 1 is curvature-free,
c) all the gradually defined tensor fields k k (1 ), k = 2, . . . , r, vanish.
8. Natural Operators C Q P 1 M C QP r M
We write QP for the connection bundle of an arbitrary principal bundle P (M, G), [18].
The connections on P form the space C QP of all sections of QP M . Further,
we write Q P r M for the bundle of all torsion-free connections on P r M , [14]. So the
r-th exponential operator on M is a natural operator C Q P 1 M C Q P r M . Us-
ing Er , W. Mikulski solved a rather sophisticated problem of finding all natural operators
C Q P 1 M C QP r M and C Q P 1 M C Q P r M , [27].
Every torsion-free connection on P 1 M defines a vector bundle isomorphism
r
: J r T M T M Sk T M
M
(19)
k=0
as follows. Write
r
I : J0r T Rm T0 Rm S k T0 Rm
M
k=0
for the standard identification. Let be a -normal coordinate system on M with center x
and B Jxr T M . We define
r
(T0 1 S k T01 1 ) I J r T (B)
M
(B) = . (20)
k=0
Since the identification I is G1m -equivariant, (20) is a correct definition.
Proposition 12. Let D : C Q P 1 M C QP r M be a natural operator. Then there

exist uniquely determined natural operators
Ak : C Q P 1 M C (T M S k T M T M ) ,
k = 0, . . . , r, such that A0 = 0, A1 = 0 and

D() = Er () + 0, 0, A2 (), . . . , Ar ()
in the sense of the identification (19).
Proof. The difference D() Er () is decomposed into r + 1 natural operators by (19).

The natural operators A0 and A1 vanish according to 25.3 and Lemma 33.4 in [18].
Now Proposition 3 yields directly
Proposition 13. Let D : C Q P 1 M C Q P r M be a natural operator. Then there

exist uniquely determined natural operators
Ak : C Q P 1 M C T M S k+1 T M ,

k = 0, . . . , r, such that A0 = 0, A1 = 0 and

D() = Er () + 0, 0, A2 (), . . . , Ar () .
The natural operators A2 , . . . , Ar in Proposition 12 or Proposition 13 can be prescribed

arbitrarily.
According to Lemma 33.4 of [18], all natural operators C Q P 1M C T M
Nk
T M are R-linearly generated by
the curvature tensor and its covariant derivatives,
constructing tensor products (including tensor products with invariant tensors) and
contractions.
In the case of some prescribed symmetries in the covariant part we add the corresponding
symmetrizations of the operators in question.
178 Ivan Kolar
Example 1. Write Rjkl i = Rjlki for the curvature tensor of . If we look for all natu-
ral operators D : C Q P M C Q P 2 M , we have to determine all natural operators
1
C Q P 1 M C (T M S 3 T M ). By the above mentioned procedure, all these oper-

m Ri
ators are the constant multiples of (j kl)m . Hence all Ds form the one-parameter family
m i
D() = E2 () + c((j Rkl)m ) , c R.
Example 2. J. Janyska and the author determined all natural operators C Q P 1 M

C QP 2 M by using a direct approach, [7], [12]. The list of them is rather long. Using
Proposition 12, we can interpret that list in a very clear geometric way.
9. Semiholonomic 2-Jets
Some aspects of our problems are properly related with the theory of semiholonomic 2-jets.
First we describe the general ideas, [3], [23].
Consider a fibered manifold p : Y M . Its second nonholonomic prolongation J2 Y
is defined by the iteration
J2 Y = J 1 (J 1 Y M ) .
If xi , y p are some fiber coordinates on Y , the induced coordinates on J 1 Y are yrp = i y p (x)
and the coordinates further induced on J2 Y are
p p
y0i = i y p (x) and yij = j yip (x) .
There are two canonical projections J2 Y J 1 Y , namely the target jet projection
1 : J2 Y J 1 Y and the jet prolongation J 1 : J2 Y J 1 Y of the target jet projec-
tion : J 1 Y Y . The second semiholonomic prolongation J2 Y is the set of all A J2 Y
satisfying
1 (A) = (J 1 )(A) .
In coordinates, this condition means
p
y0i = yip . (21)
The injection J 2 Y J2 Y is defined by
jx2 s 7 jx1 (j 1 s) .
So the subset J 2 Y J2 Y is characterized by
yip = y0i
p p
and yij p
= yji . (22)
Hence J 2 Y J2 Y . According to the general theory, both 1 : J2 Y J 1 Y and

J 1 : J2 Y J 1 Y are affine bundles.
For two manifolds M and N , the space J2 (M, N ) or J2 (M, N ) of nonholonomic
or semiholonomic 2-jets of M into N is the second nonholonomic or semiholonomic
prolongation of the product fibered manifold M N M , respectively. C. Ehres-

mann introduced the composition of nonholonomic jets, [3]. Consider another mani-
fold Q and A Jx2 (M, N )y , B Jy2 (N, Q)z . So A = jx1 and B = jy1 , where
: M J 1 (M, N ) and : N J 1 (N, Q) are sections of the source jet projection

. Hence (u) = (u), u M , so that the composition of 1-jets (u) and
(u) is defined. Then we set
B A = jx1 (u) (u) Jx2 (M, Q)z .

(23)
a , z a be the induced coordinates on J2 (N, Q)

Let z a be some local coordinates on Q, zpa , z0p pq
and wi , w0i , wij be the induced coordinates on J2 (M, Q). Evaluating (23), we obtain the
a a a
coordinate formula for the composition of nonholonomic 2-jets
wia = zpa yip , a

w0i a p
= z0p y0i , a
wij a p q
= zpq p
yi y0j + zpa yij . (24)
Clearly, the composition of two semiholonomic or holonomic 2-jets is semiholonomic or

holonomic as well.
A jet A Jx2 (M, N )y is called regular, if there exists B Jy2 (N, M )x such that
B A = jx2 id M . By (24), A is regular iff both 1 (A) and (J 1 )(A) are regular. In
coordinates this means that both yip and y0i
p
are regular matrices. If dim M = dim N , then
B A = jx2 id M implies A B = jy2 id N . In this case, regular is equivalent to invertible.
Every (u) defines a linear map Tu M T N , p = yip (u) i . This yields a local
map T M T N , whose tangent map at each point of Tx M is determined by jx1 . So the
nonholonomic 2-jet A = jx1 (u) Jx2 (M, N )y can be interpreted as a map (T T M )x
(T T N )y of the form
p = yip i , p
dy p = y0i dxi , p i
d p = yij dxj + yip d i . (25)
Consider the canonical involution of the iterated tangent functor, M ( i , dxi , d i ) =

p
(dxi , i , d i ), and A Jx2 (M, N )y in the form (25) with y0i = yip . Then the map N AM
is of the form
p = yip i , dy p = yip dxi , p i

d p = yji dxj + yip d i . (26)
This map corresponds to another semiholonomic 2-jet
(A) Jx2 (M, N )y , (yip , yij

p
) = (yip , yji
p
). (27)
Definition 2. The map is called the canonical involution of semiholonomic 2-jets.
Since J2 (M, N ) J 1 (M, N ) is an affine bundle, A and (A) determine a tensor
(A) = A (A) Ty (N ) 2 Tx M (28)
called the difference tensor of semiholonomic 2-jet A. (J. Pradines uses the name dis-
symetrie, [30].) Clearly, A is holonomic, iff (A) = 0.
180 Ivan Kolar
Example 3. We present the first remarkable application of this concept. Consider a general
connection : Y J 1 Y on an arbitrary fibered manifold Y M , [18]. If is viewed
as a morphism over M , we can construct J 1 : J 1 Y J2 Y . Clearly, the values of the
composition = J 1 lie in J2 Y . The difference tensor
: Y V Y 2 T M
coincides with the curvature of .

The second order semiholonomic frame bundle P 2 M of M is defined by P 2 M =
reg J02 (Rm , M ). This is a principal bundle P 2 M (M, G2m ), where G2m =
inv J0 (R , R )0 is the second order semiholonomic jet group in dimension m. The in-
2 m m
clusion G2m G2m defines an injection (denoted by the same symbol) l1 : G1m G2m . One
verifies easily that (8) with r = 1 defines an identification i1M : J 1 P 1 M P 2 M . Consider
a principal connection : P 1 M J 1 P 1 M with the coordinate expression
xij,k = ilk xlj . (29)
Using (8), we verify directly the following result by P. Libermann, [23].

Proposition 14. The rule 7 i1M defines a bijection between the connections on P 1 M
and the reductions of P 2 M to the subgroup l1 (G1m ) G2m .
It is remarkable that the canonical involution yields a simple construction of the
connection conjugate to . Indeed, we have
i1M = (i1M ) , (30)
i.e. transforms the reduction determined by into the reduction determined by .

Example 4. Another interesting application of the concept of difference tensor is in the
theory of G-structures. We recall that a (first order) G-structure on M is a reduction P of
P 1 M to a subgroup G G1m . Then J 1 P J 1 P 1 M P 2 M and suggests a very
conceptual way to the construction of the structure function of P , [19], [23]. In particular,
this approach clarifies, in an instructive way, the difference between the prolongability and
the flatness of P , [19].
Finally we remark that the theory of the covariant differentation with respect to connec-
tions on P 2 M is systematically developed in [6].
10. W r P as a Generalization of P r M
Consider a principal bundle : P M with structure group G, dim M = m. Its r-
th order principal prolongation W r P is the bundle of all r-jets j(0,e)
r of local principal
bundle isomorphisms
: Rm G P , 0 Rm , e = the unit of G. (31)
This is a principal bundle over M with structure group Wm r G := W r (Rm G), whose
0
action on W r P is given by the jet composition, [2], [18].
r P , we write
Given A = j0r f Tm
A = j0r ( f ) Tm
r
M.
Further, we introduce
r r r
reg Tm P = {A Tm P; A reg Tm M} .
Clearly, the local PB-isomorphism (31) is determined by its restriction | Rm

{e} : Rm P . Hence
W r P = reg Tm
r
P. (32)
Let P (M , G) be another principal G-bundle, m = dim M . For every local principal bun-
dle isomorphism f : P P , Tm r f : T r P T r P restricts and corestricts into a map
m m
reg Tm P reg Tm P . This defines W r f : W r P W r P .
r r
If A reg Tm r P , then A T r M is invertible, so that A (A)1 satisfies A

1
m
(A) = jx id M , x = f (0) . This implies A (A)1 J r P . Hence
r
W r P = P r M M J r P . (33)
Clearly, W r f is identified with P r f f J r f , where f : M M is the base map of f . In

particular, the structure group
W0r (Rm G) =: Wm
r
G = Grm Tm
r
G (34)
is the group semidirect product with the group composition

(g1 , C1 )(g2 , C2 ) = g1 g2 , (C1 g2 ) C2 ,
where denotes the induced group composition in Tm r G, [18]. The first product projection
W r P P r M is a principal bundle morphism with the associated group homomorphism

Wm r G Gr determined by (34).
m
W r P is a fundamental structure for the gauge theories of mathematical physics, [5]. In
differential geometry, the main role of W r P is based on the fact that forevery associated
bundle P [S], where S is a left G-space, the r-th jet prolongation J r P [S] is a fiber bundle
associated to W r P , [18]. Further, we have a canonical injection P r M W 1 P r1 M ,
j0r f 7 j(0,e
1
r1 )
P r1 f , f : Rm M . So W 1 P can play the role of a suitable recurrence
model for several geometric problems, [18]. In particular, the reductions of the principal
bundle W 1 P are called generalized G-structures, [10]. Several properties of higher order
G-structures are well reflected in the framework of this more general theory.
If G = {e} is the one-element group, then M {e} is identified with M and W r (M
{e}) = P r M . Hence many properties of W r P can be viewed as a generalization of the case
of P r M . In particular, we have the canonical one-form r : T P r M Rm gr1 r
m on P M .
On W r P , we introduce analogously a canonical one-form r : T W r P Rm wr1 m G=
T(0,Er1 ) W r1 (Rm G), Er1 = the unit of Wm r1 G, wr1 G = Lie(W r1 G). Consider
m m
u = j(0,e) W r P and write u1 = r1 r (u) W r1 P , where r
r1 is the jet projection.
The tangent map
u = T(0,Er1 ) W r1 : Rm wr1
m G Tu 1 W
r1
P
182 Ivan Kolar
is a linear isomorphism depending on u only. For every Z Tu W r P , we define
r (Z) = u1 T r1 (Z) .

Clearly, the following diagram commutes
r
T W rP Rm wr1
m G
r
T P rM Rm gr1
m
Analogously to Section 2, we introduced in [9]

Definition 3. The torsion of a connection on W r P is the covariant exterior differential
D r .
The Lie algebroid LW r P of W r P coincides with the r-th jet prolongation J r (LP
M ), [17], [22]. Let : T M J r LP be the algebroid form of . Analogously to the case
of J r T M , we have the truncated bracket
[[ , ]]r1 : J r LP M J r LP J r1 LP .
The torsion of can be introduced as a morphism
: T M M T M J r1 LP
defined by
( )(Z1 , Z2 ) = [[(Z1 ), (Z2 )]]r1 , (Z1 , Z2 ) T M M T M .
In [17] it is deduced that D r and are naturally equivalent.

The r-jets j0r g, g G, of the constant maps g : Rm G, x 7 g, define an injection
r
m : G Tm r G. Clearly, the direct group product l 1 r
r1 (Gm ) m (G) is a subgroup of
r
Wm G. The following assertion, that is an analogy of Proposition 5, is proved in [17].
Proposition 15. The torsion-free connections on W r P are in bijection with the reductions
of W r+1 P to the subgroup lr (G1m ) m
r+1 (G) W r+1 G.
m
11. Connections on W 1 P
We are going to discuss the connections on W 1 P in more details. By (33),
W 1 P = P 1 M M J 1 P . (35)
Write p1 : W 1 P P 1 M and p2 : W 1 P J 1 P for the product projections. Since

p1 : W 1 P P 1 M and the target jet projection : W 1 P P are principal bundle mor-
phisms, every connection
: W 1 P J 1 (W 1 P ) = J 1 P 1 M M J2 P (36)
induces a pair of connections p1 on P 1 M and on P .

Conversely, consider two connections : P J 1 P and : P 1 M J 1 P 1 M . Define
W 1 P R() = u, (v) ; (u, v) P 1 M M P .

(37)
Using the action of G on P from Section 3, one finds easily that R() is a reduction of
W 1 P to the subgroup G1m m 1 (G) W 1 G. Since (37) identifies R() with P 1 M P ,
m M
the product connection on P 1 M M P is identified with a connection on R() and
the latter connection is uniquely extended into a connection p(, ) on W 1 P . Clearly,
p(, ) = and p1 p(, ) = .
Write L0 P for the adjoint bundle of P . (Our notation is motivated by the fact that
L0 P is the subset of the Lie algebroid LP of all elements A satisfying q(A) = 0.) The
projections and p1 give rise to projections L0 W 1 P T M L0 P T M and L0 W 1 P
2
T M L0 P 1 M T M . The common kernel of these projections is L0 P T M ,
N
[12].
Proposition 16. Connections on W 1 P are in bijection with triples (, , D), where

2
C (QP ), C (QP 1 M ) and D C (L0 P T M ).
N
Proof. For C (QW 1 P ) we set = , = p1 and D = p( , p1 ).
2
The factor T M T M gives rise to an exchange map ex : L0 P T M L0 P
N
2
T M . Thus, if we replace by the classical conjugate connection and D by exD, we
N
obtain a connection said to be conjugate to . In Section 13 we present a more geometric

construction of by using the canonical involution of semiholonomic 2-jets.
There is another construction transforming the pair (, ) into a connection on W 1 P .
It is based on the general idea of flow prolongation of connections, [18]. Consider in the
lifting form
: P M T M T P .
For every vector field X on M , we first construct its -lift X : P T P and then the flow
prolongation W 1 (X) : W 1 P T W 1 P . This defines a map
W 1 : W 1 P M J 1 T M T W 1 P .
If we add in its algebroid form T M J 1 T M , we obtain the lifting map
W 1 (, ) : W 1 P M T M T W 1 P
of a principal connection on W 1 P . In [20] we deduced W 1 (, ) = , p1 W 1 (, ) =

2
. So the difference p(, ) W 1 (, ) is a section of L0 P T M . We recall that the
N
curvature C() of is a section of L0 P 2 T M . In [20], we proved
Proposition 17. We have p(, ) W 1 (, ) = C().

184 Ivan Kolar
12. Gauge-Natural Operators on Connections

Analogously to Section 8 one can pose the question of finding all geometric operators trans-
forming a pair (, ) of a connection on P and a connection on P 1 M into a connec-
tion on W 1 P . The precise meaning of geometric is gauge-natural in the sense of [2].
Roughly speaking, when passing from the classical natural operators to the gauge-natural
ones, we meet the higher order principal prolongations W r P in the former role of the higher
order frame bundles P r M , see [18] for a complete theory.
In [20] we deduced the following list of all gauge-natural operators A : C (QP )
C (Q P 1 M ) C (QW 1 P ). (The assumption is torsion-free is of technical char-
acter. If we replace Q P 1 M by QP 1 M , the list will be much longer with many further
terms of less geometric interest.) First of all one proves that the underlying connections of
A(, ) are A(, ) = and p1 A(, ) = . So the difference A(, ) p(, ) is
2
a section of L0 P T M .
N
We already know that the curvature C() is a section of that bundle. Let Z Lin(g, g)
be the subspace of all linear maps commuting with the adjoint action of G. Since ev-
ery z Z is an equivariant map between the standard fibers, it induces a vector bundle
morphism zP : L0 P L0 P . Hence one can construct the modified curvature operator
C()(z) = (zP id ) C(). On the other hand, by Example 28.7 of [18] all natural
operators C (Q P 1 M ) C (T M T M ) are linearly generated by the contractions
R1 () = (Rkijk ) and R () = (Rk ) of the curvature tensor (Ri ) of . Let S g be
2 ikj jkl
the subspace of all vectors invariant with respect to the adjoint action. Since every B S
is an invariant element of the standard fiber, it determines a section BP of L0 P . Our result
from [20] reads
Proposition 18. All gauge-natural operators C (QP ) C (Q P 1 M ) C (QW 1 P )

are of the form
p(, ) + C()(z) + B1P R1 () + B2P R2 ()
for all z Z and all B1 , B2 S.
We underline that there exist many interesting open problems concerning the gauge-
natural operators related with connections on W r P .
13. Connections on W 1 P as Reductions of W 2 P

Finally we outline how the semiholonomic 2-jets can be used in the theory of connections
on W 1 P . In general, the bundle of nonholonomic (n, 2)-velocities on a manifold M is
defined by
Tn2 M = J02 (Rn , M ) .
We shall frequently use a natural identification
Tn2 M Tn1 (Tn1 M ) . (38)

Write tu : Rn Rn for the translation x 7 x + u. If : Rn J 1 (Rn , M ) is a section,

then u 7 (u) j01 tu is a map Rn Tn1 M . Passing to 1-jets defines (38). One verifies
easily that (38) identifies reg Tn2 M with reg Tn1 (reg Tn1 M ).
The second order nonholonomic frame bundle of M is defined by
P 2 M = inv J02 (Rm , M ) .
This is a principal bundle over M with structure group
G2m = inv J02 (Rm , Rm )0 ,
the right action of which on P 2 M is determined by the composition of nonholonomic 2-jets.

On the other hand,
W 1 (P 1 M ) = P 1 M M J 1 P 1 M .
Using (32) and (38), we obtain
P 2 M = W 1 (P 1 M ) . (39)
In particular, G2m = G1m Tm

1 G1 .
m
Further we introduce the second nonholonomic prolongation of P (M, G) by the itera-
tion
W 2 P = W 1 (W 1 P ) . (40)
2 G := W 1 (W 1 G). We have
This is a principal bundle over M with structure group Wm m m
W 1 (W 1 P ) = P 1 M M J 1 P 1 M M J2 P = P 2 M M J2 P
and
1 1
Wm (Wm G) = G1m Tm
1 1 1 1
G m Tm Tm G G2m Tm
2
G,
where the group semidirect product has an analogous meaning to (34). In the semiholo-
nomic case, we have
W 2 P W 2 P := P 2 M M J2 P . (41)
The structure group of W 2 P is Wm2 G = G2 T 2 G.
m m
Consider a connection : W P J 1 (W 1 P ) = J 1 P 1 M M J2 P . Using the iden-
1
tification J 1 P 1 M P 2 M from Section 9 and the inclusion J2 P J2 P , we obtain an

inclusion W 2 P J 1 W 1 P . Let and be the underlying connections of . In Sec-
tion 11 we constructed the reduction R() W 1 P to the subgroup G1m m 1 (G). One
verifies easily that R() W 2 P is a reduction to the subgroup l1 (G1m ) m 2 (G)

2 2 2 2
Gm Tm G Gm Tm G. The following assertion generalizes the result by P. Libermann
mentioned in Section 9.
Proposition 19. The connections on W 1 P are in bijection with the reductions of W 2 P to
2 (G) G2 T 2 G.
the subgroup l1 (G1m ) m m m
2 (G). Hence its projection Q into
Proof. Let Q be a reduction of W 2 P to l1 (G1m ) m 1
1 1 1
W P is a reduction to the subgroup Gm m (G). Then Q can be interpreted as a map
(denoted by the same symbol) Q : Q1 J 1 W 1 P . This map is equivariant, so that Q can
be uniquely extended into a connection on W 1 P .
186 Ivan Kolar
Now we can construct the connection conjugate to by using the canonical in-
volution of semiholonomic 2-jets. Using Proposition 3 of [20], one proves that if
red() W 2 P is the reduction corresponding to , then
red() = red() (42)
is the reduction corresponding to .
Acknowledgement
The author was supported by the Ministry of Education of the Czech Republic under the
project MSM 0021622409 and the grant GACR No. 201/05/0523.
References
[1] A. Cabras and I. Kolar, Flow prolongations of some tangent valued forms, to appear
in Czechoslovak Math. J.
[2] D. Eck, Gauge-natural bundles and generalized gauge theories, Mem. Amer. Math.
Soc. 247 (1981).
[3] C. Ehresmann, Extension du calcul des jets aux jets non holonomes, CRAS Paris 239
(1954).
[4] M. Elzanowski and S. Prishepionok, Connections on higher order frame bundles, In:
New Developments in Differential Geometry (Proceedings, Kluwer, 1996) 131142.
[5] L. Fatibene and M. Francaviglia, Natural and Gauge Natural Formalism for Classical
Fields Theories (Kluwer, 2003).
[6] M. Ferraris, M. Francaviglia and F. Tubiello, Connections Over the Bundles of

Second-Order Frames, In: Proceedings of Conference DGA (Brno 1989, World-
Scientific, Singapore, 1990) 3346.
[7] J. Janyska, On natural operations with linear connections, Czechoslovak Math. J. 35

(1985) 106115.
[8] S. Kobayashi, Canonical forms on frame bundles of higher order contact, Proc. of
Symposia in Pure Math. 3 (1961) 186193.
[9] I. Kolar, A generalization of the torsion form, Cas. pest. mat. 100 (1975) 284290.
[10] I. Kolar, Generalized G-structures and G-structures of higher order, Boll. Un. Math.
Ital., Suppl. fasc. 3 (1975) 249256.
[11] I. Kolar, A geometric version of the higher order Hamilton formalism in fibered man-
ifolds, J. of Geometry and Physics 1 (1984) 127137.
[12] I. Kolar, Some natural operators in differential geometry, In: Proc. Conf. Diff. Geom.
and Its Applications (Brno, 1986, Dordrecht, 1987) 91110.
[13] I. Kolar, Torsion free connections on higher order frame bundles, In: New Develop-
ments in Differential Geometry (Proceedings, Kluwer, 1996) 233241.
[14] I. Kolar, On the torsion of linear higher order connections, Central European Journal
of Mathematics 3 (2003) 360366.
[15] I. Kolar, On the torsion-free connections on higher order frame bundles, Publ. Math.
Debrecen 67 (2005) 373379.
[16] I. Kolar, Weil Bundles as Generalized Jet Spaces, In: Handbook of Global Analysis
(Elsevier, 2008) 625664.
[17] I. Kolar, Connections on principal prolongations of principal bundles, In: Proc. 10th
Int. Conf. on Diff. Geom. and Appl. (Olomouc 2007, (O. Kowalski, D. Krupka, O.
Krupkova and J. Slovak, Eds.) World Scientific, Singapore, 2008) 279291.

(Springer Verlag, 1993).
[19] I. Kolar and I. Vadovicova, On the structure function of a G-structure, Math. Slovaca
35 (1985) 277282.
[20] I. Kolar and G. Virsik, Connections in first principal prolongations, In: Suppl. ai Ren-
diconti del Circolo Matematico di Palermo (Serie II, 43, 1996) 163171.
[21] D. Krupka and J. Janyska, Lectures on Differential Invariants (Univerzita JEP, Brno,
1990).
[22] A. Kumpera and D. Spencer, Lie Equations I (Princeton University Press, 1972).
[23] P. Libermann, Introduction to the theory of semi-holonomic jets, Arch. Math. (Brno)
33 (1997) 173189.
[24] P. Libermann, Sur les prolongements des fibres principaux et des groupoides
differentiables banachiques, In: Analyse globale (Sem. Math. Superieures, No. 42,
Presses Univ. Montreal, 1971) 7108.
[25] P. Libermann, Charles Ehresmann Concepts in Differential Geometry, In: Geometry

and Topology of Manifolds (Banach Center Publications, Vol. 76, Institute of Mathe-
matics PAN, Warszawa 2007) 3550.
[26] K. Mackenzie, General Theory of Lie Groupoids and Lie Algebroids (Cambridge Uni-
versity Press, Cambridge 2005).
[27] W. M. Mikulski, Higher order linear connections from first order ones, to appear in
Arch. Math. (Brno).
[28] M. Paluszny and A. Zajtz, Foundations of Differential Geometry on Natural Bundles

(Lecture Notes Univ. Caracas, 1984).
188 Ivan Kolar
[29] F. W. Pohl, Connections in differential geometry of higher order, Trans. Amer. Math.
Soc. 125 (1966) 169211.
[30] J. Pradines, Fibres vectoriels double symmetriques et jets holonomes dordre 2, CRAS
Paris, serie A 278 (1974) 15571560.
[31] N. Que, Du prolongements des espaces fibres et des structures infinitesimales, Ann.
Inst. Fourier, Grenoble 17 (1967) 157223.
[32] D. J. Saunders, The Geometry of Jet Bundles (London Math. Society Lecture Notes
Series 142, Cambridge, 1989).
[33] P. C. Yuen, Higher order frames and linear connections, Cahiers Topol. Geom. Diff.
12 (1971) 333371.
Chapter 11
N ATURAL L IFTS IN R IEMANNIAN G EOMETRY

Oldrich Kowalski1and Masami Sekizawa2
1
Faculty of Mathematics and Physics
Charles University in Prague, Sokolovska 83, 186 75 Praha 8
Czech Republic
2
Department of Mathematics, Tokyo Gakugei University
Koganei-shi Nukuikita-machi 4-1-1,
Tokyo 184-8501, Japan
Abstract
Using the concepts and the methods developed by D. Krupka, J. Janyska and
V. Mikolasova, we fully classified (during 1986-88) naturally lifted metrics to tan-
gent bundles, linear frame bundles and cotangent bundles. All classical constructions
of metrics on such bundles are special examples of our lifted metrics. We shall survey
our own earlier work and also the later development in the geometrical study of natural
metrics.
2000 Mathematics Subject Classification. 53A55, 53C20, 53C25

Key words and phrases. Riemannian metric, natural lift, tangent bundle, linear frame
bundle, cotangent bundle, sectional curvature, scalar curvature.
Introduction
There are well-known classical examples of lifted metrics on the tangent bundle T M and
on the linear frame bundle LM over a Riemannian manifold (M, g), and also of a lifted
metric on the the cotangent bundle T M over an affine manifold (M, ). Namely, they are
the Sasaki metric, the horizontal lift and the vertical lift on T M ; then the diagonal lift, the
horizontal lift and the vertical lift on LM ; and finally the Riemann extension on T M . All
these constructions have been studied extensively (see e.g. [12, 38, 30, 11, 14, 37, 43, 44,
47].) As we shall see, the classical constructions are examples of natural transformations

E-mail address: kowalski@karlin.mff.cuni.cz

E-mail address: sekizawa@u-gakugei.ac.jp
190 Oldrich Kowalski and Masami Sekizawa
of the second order. We shall survey our work [22, 21, 39] on the full classifications of
(possibly degenerate) naturally lifted metrics on T M , LM and T M . We have proved that
the complete family of naturally lifted metrics on T M and LM (for a fixed base metric) is
a module over real functions generated by some generalizations of known classical lifts. In
the case of T M , just two arbitrary parameters appear.
Our idea of naturality is closely related to that of A. Nijenhuis [33], D. B. A. Epstein
[13], P. Stredder [42] and others. Yet, we shall use for our purposes the concepts and
methods developed by D. Krupka [26, 27] and D. Krupka and V. Mikolasova [28]. See
also I. Kolar, P. W. Michor and J. Slovak [18, pp. 227280], and D. Krupka and J. Janyska
[29, pp. 160166] for other presentations of our study and for the concept of naturality in
general.
Rigidity and heredity of general natural metrics on tangent bundles and linear frame
bundles have been extensively studied. This study is usually a hard work. We shall survey
our own work from the 1980s, and the later development in the geometrical study of natural
metrics.
1. Natural Transformations
Let us recall the general theory of natural transformations due to D. Krupka. We refer to
[26, 27, 28, 29] for more details.
Let r be any non-negative integer. Then the r-th order differential group Lrn of the
n-dimensional Euclidean space Rn , n 2, is the Lie group of all r-jets of local diffeo-
morphisms of Rn with source and target at the origin o Rn . Let P and Q be smooth
manifolds on which the group Lrn acts to the left. Then an r-th order differential invariant
f : P Q is an Lrn -equivariant map of the left Lrn -space P to the left Lrn -space Q.
Further, let F r M denote the bundle of all frames of r-th order over M , which carries a
natural structure of a principal Lrn -bundle F r (M, Lrn , nr ). We get a natural functor from
the category Dn of smooth n-manifolds and injective immersions into the category of prin-
cipal Lrn -bundles and Lrn -bundle morphisms. For a left Lrn -space P , let FPr M denote the
fiber bundle with fiber P , associated to the principal Lrn -bundle F r M . We obtain a natural
functor FPr from the category Dn into the category of fiber bundles and their morphisms.
For each manifold M and each differential invariant f : P Q, we can define a
morphism fM : FPr M FQr M over the identity map id : M M by fM ([y, p]) =
[y, f (p)] for all [y, p] FPr M . This morphism fM is called the realization of a differential
invariant f on the manifold M .
An r-th order natural transformation T of the functor Fpr into the functor FQr is a
collection of bundle morphisms TM : FPr M FQr M over the identity map id : M
M , M Dn , such that FQr TM1 = TM2 FPr holds for every morphism : M1 M2
of Dn .
The following Theorem by D. Krupka in [26] says that any concrete classification of all
r-th order natural transformations of FPr to FQr can be reduced to a classification of all r-th
order differential invariants f from P to Q.
Theorem 1.1 ([26]). Let f : P Q be an r-th order differential invariant. Then the
correspondence Tf : M fM , where M is an object of Dn , is a natural transformation
Natural Lifts in Riemannian Geometry 191
of the functor FPr to the functor FQr . Moreover, the correspondence f Tf is a bijection
between the set of all r-th order differential invariants from P to Q and the set of all r-th
order natural transformations of FPr to FQr .
2. Tangent Bundle
After recalling classical examples of lifts of a given metric on the base manifold to its
tangent bundle, we shall present our full classification of naturally lifted metrics and related
results. Then, we shall survey some geometric properties of such lifted metrics.
2.1. Classical Examples of Metrics on the Tangent Bundle

The tangent bundle T M over an n-dimensional smooth manifold M , n 2, consists of all
pairs (x, u), where x is a point of M and u is a vector from the tangent space Mx of M at
x. We denote by p the natural projection of T M to M defined by p(x, u) = x.
Let g be a Riemannian metric on the manifold M and its Levi-Civita connection.
Then the tangent space (T M )(x,u) of T M at (x, u) T M splits into the horizontal and
vertical subspace H(x,u) and V(x,u) with respect to :
(T M )(x,u) = H(x,u) V(x,u) .
If a point (x, u) T M and a vector X Mx are given, then there exists a unique vector
X h H(x,u) such that p (X h ) = X. We call X h the horizontal lift of X to T M at (x, u).
The vertical lift of X to (x, u) is a unique vector X v V(x,u) such that X v (df ) = Xf
for all smooth functions f on M . Here we consider a one-form df on M as a function
on T M defined by (df )(x, u) = uf for all (x, u) T M . The map X 7 X h is an
isomorphism between Mx and H(x,u) , and the map X 7 X v is an isomorphism between
Mx and V(x,u) . In an obvious way we can define horizontal and vertical lifts of vector fields
on M . These are uniquely defined vector fields on T M .
The three classical constructions of metrics on tangent bundles T M which are derived
from a Riemannian metric g on M are given as follows:
(a) The metric g s constructed by Sasaki [38] is a (positive definite) Riemannian metric
on T M given by
s
g(x,u) (X h , Y h ) = gx (X, Y ), s
g(x,u) (X h , Y v ) = 0,
s
g(x,u) (X v , Y h ) = 0, s
g(x,u) (X v , Y v ) = gx (X, Y )
for all X, Y Mx .
(b) The horizontal lift g h of g is a pseudo-Riemannian metric on T M with signature
(n, n) which is given by
h
g(x,u) (X h , Y h ) = 0, h
g(x,u) (X h , Y v ) = gx (X, Y ),
h
g(x,u) (X v , Y h ) = gx (X, Y ), h
g(x,u) (X v , Y v ) = 0
for all X, Y Mx .
(c) The vertical lift g v of g is a degenerate metric of rank n on T M given by

v
g(x,u) (X h , Y h ) = gx (X, Y ), v
g(x,u) (X h , Y v ) = 0,
v
g(x,u) (X v , Y h ) = 0, v
g(x,u) (X v , Y v ) = 0
for all X, Y Mx .
2.2. Naturally Lifted Metrics on the Tangent Bundle

Let us consider the symmetric tensor product E = Rn Rn , i.e., the vector space of
symmetric bilinear forms on Rn and let E+ E denote the open subset of all positive
inner products. Put P = Rn Tn1 E+ and denote by (uk , gij , gij,k ), 1 i j n; k =
1, 2, . . . , n, the system of canonical coordinates of P . Also, put Q = Rn (R2n R2n )
and denote by (v k , GAB ), k = 1, 2, . . . , n; A, B = 1, 2, . . . , 2n, the system of canonical
coordinates on Q. This can be also written in the form (v k , Gij , Gik , Gij ), 1 i j
n; k = 1, 2, . . . , n, where i stands for n + i. We define the actions of L2n on P and Q,
respectively, having in mind the transformation rules on Riemannian objects under changes
of systems of natural local coordinates.
We see easily that the corresponding associated L2n -bundles FP2 M and FQ2 M over a
manifold M always have canonical bundle projections
FP2 M T M, FQ2 M T M. (2.1)
We define the problem to find all second order natural transformations of a Riemannian
metric on a manifold to a metric on its tangent bundle as the problem to find all those natural
transformations of FP2 M to FQ2 M which, for each M and via the projections (2.1), induce
the identity map id : T M T M . Hence, by Theorem 1.1, this reduces to the problem to
find all second order differential invariants f : (uk , gij , gij,k ) (v k , GAB (uk , gij , gij,k ))
from P to Q such that v k f = uk , k = 1, 2, . . . , n. We use the computational method
proposed by Krupka and Mikolasova in [28].
We obtain
Theorem 2.1 ([22]). Let ij , ij and ij be functions on P which are solutions of the
following system of differential equations:
ij
= 0, (2.2)
gpq,r
n
ij ij
uq p = ip jq + pj iq
X
2 gap (2.3)
a=1
gaq u
for i, j, p, q, r = 1, 2, . . . , n. Then a necessary condition for a map f : P Q such that
v k f = uk , k = 1, 2, . . . , n, to be a differential invariant is that its representation by the
canonical coordinates is of the form
n
X n n
ub uc sbi tcj st +
X
ub sbi sj +
X
Gij = ub sbj si + ij , (2.4)
b,c,s,t=1 b,s=1 b,s=1
n
X
Gik = ub sbi sk + ki , (2.5)
b,s=1
Gij = ij (2.6)
for 1 i j n and k = 1, 2, . . . , n, where kij s are the formal Christoffel symbols
derived from gij and gij,k .
We have given in [22] a geometric meaning to the system of differential equations (2.3)
and its solution. We have shown first which type of differential invariants of the form
ij = ij (uk , gkl ) (and which type of natural transformations) belongs to the system of
differential equations. Consider the left L1n -spaces V = Rn E+ and W = Rn Rn with
the natural actions of L1n defined, again, having in mind transformation rules under changes
of systems of natural local coordinates. Let (uk , gij ), 1 i j n; k = 1, 2, . . . , n, and
(ij ), i, j = 1, 2, . . . , n, be the systems of canonical coordinates of V and W , respectively.
We can check easily that the system of differential equations (2.3) gives a necessary
condition for a map ij = ij (uk , gkl ) to be (first order) differential invariant from V to
W . Long but routine calculations show that the formulas (2.4)(2.6) provide a differential
invariant f from P to Q if and only if the functions ij , ij and ij defined on V describe
some differential invariants , and from V to W . Thus our problem reduces to the
study of the first order natural transformations of FV1 into FW 1 .
We say that a bundle morphism of the form : T M T M T M M R is an

F-metric on M if it is linear in the second and the third argument (and smooth in the first
argument). We also say that is symmetric or skew-symmetric if it is symmetric or skew-
symmetric with respect to the second and third argument, respectively. (We use here the
letter F to recall the Finsler geometry.) Any Riemannian metric g on M is a symmetric
F-metric which is independent on u. In our special case, letting g be a given Riemannian
metric on M , we speak about natural F-metrics derived from g which are F-metrics ,
for a fixed u T M , whose components (u)ij = (u, /xi , /xj ) with respect to a
system of local coordinates (x1 , x2 , . . . , xn ) in M are solutions of the system of differential
equations (2.3).
We obtain that
Theorem 2.2 ([22]). Let (M, g) be an n-dimensional oriented Riemannian manifold. Then
all natural F-metrics on M derived from g are given as follows:
(1) For n = 2, all symmetric natural F-metrics are of the form
(u; X, Y ) = (kuk2 )g(X, Y ) + (kuk2 )g(X, u)g(Y, u)
+ (kuk2 ){g(X, u)g(Y, Ju) + g(X, Ju)g(Y, u)},
and all skew-symmetric natural F-metrics are of the form
(u; X, Y ) = (kuk2 ){g(X, u)g(Y, Ju) g(X, Ju)g(Y, u)},
where , , and are arbitrary functions of kuk2 = g(u, u) and J is one of the two
canonical almost complex structures of (M, g) (for which (M, g, J) is a Kahler manifold).
(2) For n = 3, all symmetric natural F-metrics are of the form

(u; X, Y ) = (kuk2 )g(X, Y ) + (kuk2 )g(X, u)g(Y, u) (2.7)
and all skew-symmetric natural F-metrics are of the form
(u; X, Y ) = (kuk2 )g(X Y, u),
where , and are arbitrary functions of kuk2 = g(u, u), and X Y is the usual vector
product of X and Y .
(3) For n > 3, all natural F-metrics are symmetric and of the form (2.7).
I. Kolar, P. W. Michor and J. Slovak have given in [18, Proposition 33.22] a new and
elegant proof of the classification of natural F-metrics on non-oriented Riemannian mani-
folds.
Theorem 2.3 ([18]). Let (M, g) be an n-dimensional non-oriented Riemannian manifold,
n 1. Then all natural F-metrics on M derived from g are symmetric and given by (2.7),
where and are arbitrary smooth functions defined on the interval (0, ). In particular,
= 0 if n = 1.
M. T. K. Abbassi has proved in [1] explicitly that all basic functions from Theorems 2.2
and 2.3 can be prolonged to smooth functions on the set of all non-negative real numbers.
This result found many applications in the techniques used for the thorough investigation
of g-natural metrics by M. T. K. Abbassi and M. Sarih.
For a given Riemannian metric g on M , we define the classical lifts of F-metrics from
M to T M , with respect to g, as follows:
(a) The Sasaki lift s,g of a symmetric F-metric with respect to g is defined by
s,g s,g
(x,u) (X h , Y h ) = x (u; X, Y ), (x,u) (X h , Y v ) = 0,
s,g s,g
(x,u) (X v , Y h ) = 0, (x,u) (X v , Y v ) = x (u; X, Y )
for all X, Y Mx .
(b) The horizontal lift h,g of an arbitrary F-metric with respect to g is defined by
h,g h,g
(x,u) (X h , Y h ) = 0, (x,u) (X h , Y v ) = x (u; Y, X),
h,g h,g
(x,u) (X v , Y h ) = x (u; X, Y ), (x,u) (X v , Y v ) = 0
for all X, Y Mx .
(c) The vertical lift v of a symmetric F-metric with respect to g is defined by
v
(x,u) (X h , Y h ) = x (u; X, Y ), v
(x,u) (X h , Y v ) = 0,
v
v
(x,u) (X v , Y h ) = 0, (x,u) (X v , Y v ) = 0
for all X, Y Mx .
Obviously, the vertical lift does not depend on the choice of g. We note that s,g , h,g
and v are (not necessarily regular) pseudo-Riemannian metrics on T M . If we take = g,
then s,g , h,g and v are just the classical lifts g s , g h and g v , respectively.
Thus we have all metrics on T M which come from a second order natural transforma-
tion of a given Riemannian metric on M .
Theorem 2.4 ([22]). Let g be a Riemannian metric on an n-dimensional smooth manifold

M , n 2, and let G be a (possibly degenerate) pseudo-Riemannian metric on the tangent
bundle T M which comes from a second order natural transformation of g. Then there are
natural F-metrics 1 , 2 and 3 derived from g, where 1 and 3 are symmetric, such that
G = 1 s,g + 2 h,g + 3 v .
Moreover, all natural F-metrics derived from g are given by Theorem 2.2.
If n = 2, then the family of all natural metrics G on T M depends on 10 arbitrary
functions of one variable, for n = 3 it depends on seven arbitrary functions of one variable,
and for n > 3 on six arbitrary functions of one variable.
Our concept of F-metric coincides with that of M-tensor of type (0, 2) introduced by
Y. C. Wong and K. P. Mok in [46]. Further, our Theorem 2.2 is strongly related to Theo-
rem 5.1 of the paper by K. P. Mok, E. M. Patterson and Y. C. Wong in [31]. Nevertheless, the
authors above do not mention the naturality problem in their investigations, and they also
do not give any nontrivial examples of M-tensor. In fact, Theorem 5.1 mentioned above
only says that each metric on the tangent bundle T M is a sum of three classical lifts of
three independent M-tensors of type (0, 2) with respect to a generalized connection (called
M-connection).
The metrics given by Theorem 2.4 are called g-natural metrics by M. T. K. Abbassi in
[1]. He has formulated these metrics in the form:
Theorem 2.5 ([1, 4]). Let (M, g) be an n-dimensional Riemannian manifold and G be a
g-natural metric on the tangent bundle T M . Then there are real valued functions i and
i , i = 1, 2, 3, defined on [0, ) such that
G(x,u) (X h , Y h ) = (1 + 3 )(r2 )gx (X, Y )
+ (1 + 3 )(r2 )gx (X, u)gx (Y, u),
G(x,u) (X h , Y v ) = 2 (r2 )gx (X, Y ) + 2 (r2 )gx (X, u)gx (Y, u),
G(x,u) (X v , Y h ) = 2 (r2 )gx (X, Y ) + 2 (r2 )gx (X, u)gx (Y, u),
G(x,u) (X v , Y v ) = 1 (r2 )gx (X, Y ) + 1 (r2 )gx (X, u)gx (Y, u)

hold at each point (x, u) T M for all u, X, Y Mx , where r2 = gx (u, u). For n = 1,
the same holds with i = 0, i = 1, 2, 3.
2.3. Riemannian Geometry of the Tangent Bundle

Let be a (local) transformation of a manifold M . Then we define a transformation of
T M by
(x, u) = (x, x u)
for all (x, u) T M . If is a (local) affine transformation with respect to the Levi-Civita
connection of (M, g), then we have
(X h ) = ( X)h , (X v ) = ( X)v
for all X X(M ). Using this fact we can easily see that all g-natural metrics are invariant:
Theorem 2.6 ([24]). Let be a (local) isometry of a Riemannian manifold (M, g). Then
every g-natural metric G on the tangent bundle T M over (M, g) is invariant by the lift
of . In other words, is a local isometry of (T M, G) whose projection on (M, g) is .
Riemannian geometry of the tangent bundle T M with the metric g s defined by S. Sasaki
in [38] has been studied by many authors (see for example, [12, 19, 32, 47]). The first author
of the present paper has proved in [19] that (T M, g s ) is never locally symmetric unless the
base metric g on M is locally Euclidean. Further, E. Musso and F. Tricerri have proved in
[32] that g s is extremely rigid in the following sense:
Theorem 2.7 ([32]). The tangent bundle (T M, g s ) with the Sasaki metric has constant
scalar curvature if and only if the base manifold (M, g) is locally Euclidean.
It seems that another metric nicely fitted to the tangent bundle is the so-called Cheeger-
Gromoll metric. Its construction has been suggested by J. Cheeger and D. Gromoll in [8]
and expressed more explicitly by E. Musso and F. Tricerri in [32]. It is a nonclassical
natural metric on the tangent bundle T M over a Riemannian manifold (M, g) defined at
each point (x, u) T M by
cg 1
kuk2 gxv kxv + gxs,g + kxs,g ,

g(x,u) = 2
1 + kuk
where k is the natural F-metric given by k(u; X, Y ) = g(X, u)g(Y, u). In the other form,
g cg is given by
cg cg cg
g(x,u) (X h , Y h ) = gx (X, Y ), g(x,u) (X h , Y v ) = g(x,u) (X v , Y h ) = 0,
cg 1
(X v , Y v ) =

g(x,u) gx (X, Y ) + gx (X, u)gx (Y, u)
1 + kuk2
for all X, Y Mx . The curvatures of g cg have been studied in detail by the second author
of the present paper in [41]. In particular, it has been proved that the scalar curvature
of g cg is never constant if g has constant sectional curvature. (Unfortunately there are
computational errors in the paper [41], which have been pointed and corrected precisely by
S. Gudmandsson and E. Kappose in [16]).
It should be interesting to find non-rigid metrics on T M over a Riemannian manifold
(M, g). The first example of such metrics has been given by V. Oproiu in [36]. The family
of Oproiu metrics depends on two arbitrary functions of one variable, and belong to a family
of metrics on T M which come from a second order natural transformation of g given by
Theorem 2.4. In terms of Theorem 2.5, The Oproiu metrics are g-natural metrics on T M
such that
(1 + 3 )(t) = v(t/2), (1 + 3 )(t) = w(t/2),
1 w(t/2)
1 (t) = , 1 (t) = ,
v(t/2) v(t/2) v(t/2) + tw(t/2)
2 = 2 = 0,
where v and w are real valued smooth functions defined on (0, ) such that v > 0 and
v(t) + 2tw(t) > 0 for all t (0, ). (see [5]). The main result of [36] is the following
Theorem 2.8. If (M, g) is an n-dimensional space of negative constant sectional curvature,

n > 2, then T M equipped with any Oproiu metric is a Kahler Einstein manifold with
positive constant scalar curvature.
M. T. K. Abbassi and M. Sarih have proved in [6] the following (highly nontrivial)
heredity theorem:
Theorem 2.9. Let G be any Riemannian g-natural metric on T M , where (M, g) is an ar-
bitrary Riemannian manifold. Then, if (T M, G) is flat, or locally symmetric, or of constant
sectional curvature, or of constant scalar curvature, or an Einstein manifold, respectively,
then (M, g) possesses the same property, respectively.
As an application of study on the Oproiu metric, they have proved that
Theorem 2.10 ([6]). Let (M, g) be an n-dimensional space of negative constant sectional
curvature, n 3. Then there is a one-parameter family F of g-natural metrics on T M
with nonconstant defining functions i and i , i = 1, 2, 3, such that, for every G F,
(T M, G) is a space of positive constant scalar curvature. Moreover, for each (M, g) as
above, and each prescribed positive constant S, there is a metric G F with the constant
scalar curvature S.
Moreover they have obtained examples of Riemannian g-natural metrics with constant
defining functions i and i , i = 1, 2, 3, which can have an arbitrary constant scalar curva-
ture (not necessarily positive as in the above Theorem).
Theorem 2.11 ([6]). Let (M, g) be a space of constant sectional curvature K 6= 0, and
let G = a g s + b g h + c g v a metric on the
tangent bundle T M , where a, b and c are
constant such that a > 0 and b2 = (1 + 13 )a(a + c)/6. Then (T M, G) is a space of
nonzero constant scalar curvature with the same sign as K. Furthermore, we can choose
the constants a, b and c so that the scalar curvature of G has any prescribed nonzero
constant value with the same sign as K.
3. Linear Frame Bundle

After recalling classical examples of lifts of a given metric on the base manifold to its linear
frame bundle, we shall present our full classification of naturally lifted metrics. Then, we
shall remind some rigidity results for special cases of these metrics.
3.1. Classical Examples of Metrics on the Linear Frame Bundle

The linear frame bundle LM over an n-dimensional smooth manifold M , n 2, consists
of all pairs (x, u), where x is a point of M and u is a basis for the tangent space Mx of M
at x. We denote by p the natural projection of LM to M defined by p(x, u) = x.
Let g be a Riemannian metric on the manifold M and its Levi-Civita connection.
Then the tangent space (LM )(x,u) of LM at (x, u) LM splits into the horizontal and
vertical subspace H(x,u) and V(x,u) with respect to :
(LM )(x,u) = H(x,u) V(x,u) .

If a point (x, u) LM and a vector X Mx are given, then there exists a unique vector
X h H(x,u) such that p (X h ) = X. We call X h the horizontal lift of X to LM at
(x, u). We define naturally n different vertical lifts of X Mx . If is a one-form on M ,
then , = 1, 2, . . . , n, are functions on LM defined by ( )(x, u) = (u ) for all
(x, u) = (x, u1 , u2 , . . . , un ) LM . The vertical lifts X v, , = 1, 2, . . . , n, of X Mx
to LM at (x, u) are the n vectors such that X v, ( ) = (X) , , = 1, 2, . . . , n, hold
for all one-forms on M , where denotes the Kroneckers delta. The n vertical lifts are
always uniquely determined, and they are linearly independent if X 6= 0. In an obvious
way we can define horizontal and vertical lifts of vector fields on M . These are uniquely
defined vector fields on T M .
The three classical constructions of metrics on linear frame bundles LM which are
derived from a Riemannian metric g on M are given as follows:
(a) The diagonal lift g d of g defined by K. P. Mok in [30] is a (positive definite) Rieman-
nian metric on LM given by
d
g(x,u) (X h , Y h ) = gx (X, Y ), d
g(x,u) (X h , Y v, ) = 0,
d
g(x,u) (X v, , Y h ) = 0, d
g(x,u) (X v, , Y v, ) = gx (X, Y )
for all X, Y Mx and , = 1, 2, . . . , n. This metric is called also the Sasaki-Mok metric
because it resembles the Sasaki metric of the tangent bundle over a Riemannian manifold.
(b) The horizontal lift g h of g is a degenerate metric on LM of rank 2n and signature
(n, n) which is given by
h
g(x,u) (X h , Y h ) = 0, h
g(x,u) (X h , Y v, ) = gx (X, Y ),
h
g(x,u) (X v, , Y h ) = gx (X, Y ), h
g(x,u) (X v, , Y v, ) = 0
for all X, Y Mx and , = 1, 2, . . . , n.

(c) The vertical lift g v of g is a degenerate metric of rank n on LM given by
v
g(x,u) (X h , Y h ) = gx (X, Y ), v
g(x,u) (X h , Y v, ) = 0,
v
g(x,u) (X v, , Y h ) = 0, v
g(x,u) (X v, , Y v, ) = 0
for all X, Y Mx and , = 1, 2, . . . , n.
3.2. Naturally Lifted Metrics on the Linear Frame Bundle

Put P = GL(n, R) Tn1 E+ and denote by (uk , gij , gij,k ), 1 i j n; k = 1, 2, . . . , n,
the system of canonical coordinates of P . Also, put Q = GL(n, R) {(Rn Rn )
(Rn Rn Rn ) [(Rn Rn ) (Rn Rn )]} and denote by (vk , Gij , Gi k , G ij )
the system of canonical coordinates on Q. Here 1 i j n; k, = 1, 2, . . . n and
(1, 1) (i, ) (j, ) (n, n) in the lexicographic arrangement. We define the actions
of L2n on P and Q, respectively, having in mind the transformation rules on Riemannian
objects under changes of systems of natural local coordinates.
Discussing as in the previous section, we have

Theorem 3.1 ([21]). Let ij , i j and ij be functions on P which are solutions of the
following system of differential equations:
ij
= 0, (3.1)
gpq,r
n n
ij ij
uq p = ip jq + pj iq
X X
2 gap (3.2)
a=1
gaq =1 u
for i, j, p, q, r = 1, 2, . . . , n. Then a necessary condition for a map f : P Q such that

v k f = uk , k = 1, 2, . . . , n, to be a differential invariant is that its representation by the
canonical coordinates is of the form
n
X n n
ua ub sai tbj st

X
ub sbi s j +
X
Gij = + ub sbj s i + ij , (3.3)
a,b,s,t,,=1 b,s,=1 b,s,=1
n

Gi k =
X
ua sai sk + i k , (3.4)
a,s,=1
G
ij = ij (3.5)
for 1 i j n; k, = 1, 2, . . . n and (1, 1) (i, ) (j, ) (n, n) in the
lexicographic arrangement, where kij s are the formal Christoffel symbols derived from
gij and gij,k .
We have given in [21] a geometric meaning to the system of differential equations (3.2)
and its solution. Consider the left L1n -spaces V = GL(n, R) E+ and W = Rn Rn
with the natural actions of L1n defined, again, having in mind transformation rules under
changes of systems of natural local coordinates. Let (uk , gij ), 1 i j n; k, =
1, 2, . . . , n, and (ij ), i, j = 1, 2, . . . , n, be the systems of canonical coordinates of V and
W , respectively.
We can check easily that the system of differential equations (3.2) gives a necessary
condition for a map ij = ij (uk , gkl ) to be (the first order) differential invariant from V to
W . Long but routine calculations show that the formulas (3.3)(3.5) provide a differential

invariant f from P to Q if and only if the functions ij , i j and ij defined on V describe
some differential invariants , and from V to W .
We say that a bundle morphism of the form : LM T M T M M R
is an L-metric on M if it is linear in the second and the third argument (and smooth in
the first argument). We also say that is symmetric or skew-symmetric if it is sym-
metric or skew-symmetric with respect to the second and third argument, respectively.
Any Riemannian metric g on M is a symmetric L-metric which is independent on u. In
our special case, letting g be a given Riemannian metric on M , we speak about natural
L-metrics derived from g which are L-metrics , for a fixed u LM , whose components
(u)ij = (u, /xi , /xj ) with respect to a system of local coordinates (x1 , x2 , . . . , xn )
in M are solutions of the system of differential equations (3.2). Solving this system of dif-
ferential equations, we obtain
Theorem 3.2 ([21]). Let (M, g) be an n-dimensional Riemannian manifold. Then all nat-
ural L-metrics on M derived from g are given by
n
X
(u; X, Y ) = (X) (Y ) (3.6)
,=1
where { 1 , 2 , . . . , n } are the dual frame to a linear frame u = {u1 , u2 , . . . , un } and

, , = 1, 2, . . . , n, are arbitrary smooth functions of n(n + 1)/2 variables w =
g(u , u ), 1 n.
For a given Riemannian metric g on M , we define the classical lifts of L-metrics from
M to LM with respect to g as symmetric (0, 2)-tensor fields on LM which are constructed
as follows:
(a) Let be a symmetric L-metric and ( ), 1 n, a family of arbitrary
L-metrics. The diagonal lift d,g of the family = (, ) with with respect to g is
defined by
d,g
(x,u) (X h , Y h ) = x (u; X, Y ),
d,g
(x,u) (X h , Y v, ) = 0, = 1, 2, . . . , n,
d,g
(x,u) (X v, , Y v, ) = x (u; X, Y ), 1 n,
for all X, Y Mx .
(b) The horizontal lift h,g of an n-tuple = ( ) of L-metrics with respect to g is
defined by
h,g
(x,u) (X h , Y h ) = 0,
h,g
(x,u) (X h , Y v, ) = x (u; Y, X), = 1, 2, . . . , n,
h,g
(x,u) (X v, , Y v, ) = 0, , = 1, 2, . . . , n,
for all X, Y Mx .
If we take = g and = g in (a), and = g in (b), then d,g and h,g are just
the classical lifts g d and g h , respectively. Also, if we take = g and = 0 in (a), then
d,g is the vertical lift g v of g.
Thus we have all metrics on LM which come from a second order natural transforma-
tion of a given Riemannian metric on M :
Theorem 3.3 ([21]). Let g be a Riemannian metric on an n-dimensional smooth manifold
M , n 2, and let G be a (possibly degenerate) pseudo-Riemannian metric on the linear
frame bundle LM which comes from a second order natural transformation of g. Then
there are families 1 = (, ) and 2 = ( ) of natural L-metrics derived from g, where
1 n, = 1, 2, . . . , n and is symmetric, such that
G = 1 d,g + 2 h,g .
Moreover, all natural L-metrics derived from g are given by Theorem 3.2.
The family of all natural metrics G on LM over an n-dimensional Riemannian manifold

(M, g) depends on n(n3 + 3n2 + n + 1)/2 arbitrary functions of n(n + 1)/2 variables.
Equivalently, the metrics G in Theorem 3.3 can be expressed in the following form:
n
G(x,u) (X h , Y h ) =
X
x (X)x (Y ),
,=1
n
G(x,u) (X h , Y v, ) = x (X)x (Y ),
X
(3.7)
,=1
n
G(x,u) (X v, , Y v, ) =
X

x (X)x (Y )
,=1
for all X, Y Mx , where , and

, , , , = 1, 2, . . . , n, are arbitrary smooth
functions of n(n + 1)/2 variables w = g(u , u ), 1 n. In the following we
shall call G a g-natural metric on LM .
Remark. The case of symmetric affine connection given on the base manifold gives a com-
pletely analogous classifications of metrics on LM as described for the metric case in The-
orems 3.2 and 3.3, where just two lifts of metric type of L-metrics are replaced by the
lifts of affine type. (See [40].)
3.3. Riemannian Geometry of the Linear Frame Bundle

Let be a (local) transformation of a manifold M . Then we define a transformation of
LM by
(x, u) = (x, x u1 , x u2 , . . . , x un )
for all (x, u) = (x, u1 , u2 , . . . , un ) LM . If is a (local) affine transformation with
respect to the Levi-Civita connection of (M, g), then we have
(X h ) = ( X)h , (X v, ) = ( X)v,
for all X X(M ) and = 1, 2, . . . , n. Using this fact we can easily see that all g-natural
metrics are invariant:
Theorem 3.4 ([25]). Let be a (local) isometry of a Riemannian manifold (M, g). Then
every g-natural metric G on the linear frame bundle LM over (M, g) is invariant by the
lift of . In other words, is a local isometry of (LM, G) whose projection on (M, g) is
.
Let (M, g) be an n-dimensional Riemannian manifold, n 2, and LM its linear frame

bundle. It seems that phenomenon similar to geometry of T M with the Sasaki metric g s
happens about geometry of LM with the diagonal lift g d . L. A. Cordero and M. de Leon
have shown in [10] that g d is also rigid in some sense. Namely, they have proved
Theorem 3.5 ([10]). The Riemannian manifold (LM, g d ) is never locally symmetric unless
(M, g) is locally Euclidean.
Theorem 3.6 ([10]). If (LM, g d ) is an Einstein manifold, then (M, g) is flat.

Theorem 3.7 ([10]). The Riemannian manifolds (LM, g d ) and (M, g) have the same con-
stant scalar curvature if and only if both are flat.
It should be interesting to find non-rigid metrics on LM . We study in this section spe-
cific Riemannian metrics from our list given by Theorem 3.3, which are a bit more general
ones than the diagonal lift g d . The horizontal and vertical distributions are no more orthog-
onal in general with respect to the metrics g treated in this section. Usually, calculations
for getting geometric objects of LM are not short. They are complicated sometimes. We
have given in [23] some simpler procedure in order to calculate them. The metrics which
we pick in the following are still rigid when the base manifold is of constant curvature.
Let g be a metric on LM defined by taking the coefficient functions in (3.7) as
= w , = c w and
= c w ,
where c and c = c are constants. That is, g is given by
g(x,u) (X h , Y h ) = gx (X, Y ),
g(x,u) (X h , Y v, ) = c gx (X, Y ), (3.8)
g(x,u) (X v, , Y v, ) = c gx (X, Y )
for all X, Y Mx and , = 1, 2, . . . , n. The metric g is positive definite if and only if all
principal minor determinants of
1 c1 c2 ... cn

1
c c11 c12 . . . c1n

2
c21 22 2n

c c ... c

... ... ...

cn cn1 cn2 . . . cnn
are positive. In particular, the matrix [c ] is positive definite. If c = 0 and c = for

, = 1, 2, . . . , n, then the metric g is the diagonal lift g d of g.
We have shown in [23] that the metric g is rigid in the following sense:
Theorem 3.8 ([23]). Let (M, g) be a space of constant sectional curvature K, and g the
positive definite metric on LM defined by (3.8). If the scalar curvature of the frame bundle
(LM, g) is constant then both manifolds are flat.
This can be applied to the diagonal metric and hence we obtain a new result in the spirit
of Theorems 3.53.7.
4. Cotangent Bundle
If the base manifold M has a Riemannian metric g, then its cotangent bundle T M is dual
to the tangent bundle T M , and hence all natural lifts of g to T M are settled from the those
of g to T M through this duality. But there is a situation when only T M comes in the
game, namely if the base manifold is an affine manifold (M, ). Here a new operation,
so-called Riemann extension is known [37]. After recalling this notion, we shall present the
full classification of naturally lifted (pseudo-Riemannian) metrics from a given symmetric
affine connection on M to T M .
4.1. Classical Examples of Metrics on the Cotangent Bundle

The cotangent bundle T M over an n-dimensional smooth manifold M , n 2, consists of
all pairs (x, w), where x is a point of M and w is a covector from the cotangent space Mx
of M at x. We denote by p the natural projection of T M to M defined by p(x, w) = x. Let
p : T (T M ) T M be the differential of the natural projection p, and q : T (T M )
T M be the natural projection of the tangent bundle T (T M ) (over T M ) to T M . The
canonical one-form on T M is a one-form defined by (X) = q(X)(p X) for all X
T (T M ), and the canonical two-form on T M is the exterior derivative d of .
Let be an affine connection on the base manifold M . Then the tangent space
(T M )(x,w) of T M at (x, w) T M splits into the horizontal and vertical subspace

H(x,w) and V(x,w) with respect to :

(T M )(x,w) = H(x,w) V(x,w) .
Let X = hX + vX be the decomposition of a vector filed X on T M into the horizontal
and vertical part. The Riemann extension g of the affine connection on M to T M is a
pseudo-Riemannian metric of the signature (n, n) defined by
g(X, Y) = (d)(vX, hY) + (d)(vY, hX)
for all X, Y T (T M ). This metric does not depend on the skew-symmetric part of .
4.2. Naturally Lifted Metrics on the Cotangent Bundle

Let us consider the space P = Rn Rn (Rn Rn ) , and denote by (wh , hij ),

h = 1, 2, . . . , n; 1 i j n, the system of canonical coordinates of P . Also, put

Q = Rn (Rn Rn ) (Rn Rn ) and denote by (zh , Gij , Ghi , Gij ), h = 1, 2, . . . , n;
1 i j n, the system of canonical coordinates on Q. We define the actions of L2n
on P and Q, respectively, having in mind the transformation rules on Riemannian objects
under changes of systems of natural local coordinates.
Discussing as in the previous section, we have
Theorem 4.1 ([39]). All differential invariants f : P Q such that, in the canonical
coordinates, zh f = wh , h = 1, 2, . . . , n, are given by
n
X
Gij = 2a ws sij + b wi wj ,
s=1
Ghi = a ih ,
Gij = 0
for 1 i j n and h = 1, 2, . . . , n, where a and b are constants.
Thus we have
Theorem 4.2 ([39]). Let be a symmetric affine connection on an n-dimensional manifold

M , n 2. Then a (possibly degenerate) pseudo-Riemannian metric G on the cotangent
bundle T M over M comes from a second order natural transformation of if and only
if G = a g + b 2 , where g is the Riemann extension of , 2 is the tensor square of the
canonical one-form of T M , and a, b are constants.
This metric G is non-degenerate (of signature (n, n)) if and only if a 6= 0.
Acknowledgement
The first author was supported by the grant GA CR 201/05/2707 and by the project
MSM 0021620839.
References
[1] M. T. K. Abbassi, Note on the classification theorems of g-natural metrics on the
tangent bundle of a Riemannian manifolds (M, g), Comment. Math. Univ. Carolinae
45 (4) (2004) 591596.
[2] M. T. K. Abbassi and G. Calvaruso, On the curvature of g-natural contact metric

structures on unit tangent sphere bundles, preprint.
[3] M. T. K. Abbassi and O. Kowalski, On g-natural metrics with constant scalar curvature
on unit tangent sphere bundles, In: Topics in almost Hermitian geometry and related
fields (World Sci. Publ., Hackensack, NJ, 2005) 129.
[4] M. T. K. Abbassi and M. Sarih, On some hereditary properties of Riemannian g-natu-

ral metrics on tangent bundles of Riemannian manifolds, Diff. Geom. Appl. 22 (2005)
1947.
[5] M. T. K. Abbassi and M. Sarih, On natural metrics on tangent bundles of Riemannian

manifolds, Arch. Math. (Brno) 41 (2005) 7192.
[6] M. T. K. Abbassi and M. Sarih, On Riemannian g-natural metrics of the form a.g s +
b.g h + c.g v on the tangent bundle of a Riemannian manifold (M, g), Mediter. J. Math.
2 (2005) 1943.
[7] A. A. Borisenko and A. L. Yampolsky, Riemannian geometry of bundles, (Russian)

Uspekhi Mat. Nauk 46 (6), (282) (1991) 5195; translation in Russian Math. Surveys
46 (1991) 55106.
[8] J. Cheeger and D. Gromoll, On the structure of complete manifolds of nonnegative

curvature, Ann. Math. 96 (1972) 413443.
[9] L. A. Cordero and M. de Leon, Lifts of tensor fields to the frame bundle, Rend. Circ.
Mat. Palermo 32 (1983) 236271.
[10] L. A. Cordero and M. de Leon, On the curvature of the induced Riemannian metric on
the frame bundle of a Riemannian manifold, J. Math. Pures Appl. 65 (1986) 8191.
[11] L. A. Cordero, C. T. J. Dodson and M. de Leon, Differential Geometry of Frame

Bundles (Mathematics and its Applications, 47, Kluwer Academic Publishers Group,
Dordrecht, 1989).
[12] P. Dombrowski, On the geometry of the tangent bundles, J. Reine Angew. Math. 210
(1962) 7388.
[13] D. B. A. Epstein, Natural tensors on Riemannian manifolds, J. Differential Geometry

10 (1975) 631645.
[14] E. Garca-Ro, D. N. Kupeli, M. E. Vazquez-Abal and R. Vazquez-Lorenzo, Affine

Osserman connections and their Riemann extensions, Diff. Geom. Appl. 11 (1999)
145153.
[15] A. Gray, Pseudo-Riemannian almost product manifolds and submersions, J. Math.

Mech. 16 (1967) 715737.
[16] S. Gudmandsson and E. Kappose, On the geometry of the tangent bundle with the
Cheeger-Gromoll metric, Tokyo J. Math. 25 (2002) 7583.
[17] S. Kobayashi and K. Nomizu, Foundations of Differential Geometry II (Interscience

Publishers, New York-London-Sydney, 1969).
[18] I. Kolar, P. W. Michor and J.Slovak, Natural Operations in Differential Geometry

(Springer-Verlag, Berlin-Heidelberg-New York, 1993).
[19] O. Kowalski, Curvature of the induced Riemannian metric of the tangent bundle of a
Riemannian manifold, J. Reine Angew. Math. 250 (1971) 124129.
[20] O. Kowalski and M. Belger, Riemannian metrics with the prescribed curvature tensor
and all its covariant derivatives at one point, Math. Nachr. 168 (1994) 209225.
[21] O. Kowalski and M. Sekizawa, Natural transformations of Riemannian metrics on

manifolds to metrics on linear frame bundlesa classification, In: Differential Ge-
ometry and Its Applications (Brno, 1986, Math. Appl., East European Ser. 27, Reidel,
Dordrecht, 1987) 149178.
[22] O. Kowalski and M. Sekizawa, Natural transformations of Riemannian metrics on

manifolds to metrics on tangent bundlesA classification, Bull. Tokyo Gakugei
Univ. (4) 40 (1988) 129; http://ir.u-gakugei.ac.jp/handle/2309/36051.
[23] O. Kowalski and M. Sekizawa, On curvatures of linear frame bundles with naturally
lifted metrics, Rend. Sem. Mat. Univ. Pol. Torino. 63 (2005) 283295.
[24] O. Kowalski and M. Sekizawa, Invariance of g-natural metrics on tangent bundles, in:
Proc. 10th International Conference on Differential Geometry and Its Applications
(World Scientific, Singapore, 2008) 171181.
[25] O. Kowalski and M. Sekizawa, Invariance of g-natural metrics on linear frame bun-
dles, to appear in Arch. Math. (Brno).
[26] D. Krupka, Elementary theory of differential invariants, Arch. Math. (Brno) 4 (1978)
207214.
[27] D. Krupka, Differential invariants (Lecture Notes, Faculty of Science, Purkyne Uni-
versity, Brno, 1979).
[28] D. Krupka and V. Mikolasova, On the uniqueness of some differential invariants: d,

[ , ], , Czechoslovak Math. J. 34 (1984) 588597.
[29] D. Krupka and J. Janyska, Lectures on Differential Invariants (University J. E.

Purkyne in Brno, 1990).
[30] K. P. Mok, On the differential geometry of frame bundles of Riemannian manifolds,

J. Reine Angew Math. 302 (1978) 1631.
[31] K. P. Mok, E. M. Patterson and Y. C. Wong, Structure of symmetric tensors of type

(0, 2) and tensors of type (1, 1) on the tangent bundle, Trans. Amer. Math. Soc. 234
(1977) 253278.
[32] E. Musso and F. Tricerri, Riemannian metrics on tangent bundles, Ann. Mat. Pura
Appl. (4) 150 (1988) 120.
[33] A. Nijenhuis, Natural bundles and their general propertiesGeometric objects

revisited, In: Differential Geometry (in honor of K. Yano (S. Kobayashi, M. Obata
and T. Takahashi, Eds.) Kinokuniya Book-store, Tokyo, 1972) 317334.
[34] B. ONeill, The fundamental equations of a submersion, Michigan Math. J. 13 (1966)

459469.
[35] B. ONeill, Semi-Riemannian Geometry with Applications to Relativity (Academic

Press, New YorkLondon, 1983).
[36] V. Oproiu, Some new geometric structures on the tangent bundle, Math. Publ. Debre-
cen 55 (1999) 261281.
[37] E. M. Patterson and A. G. Walker, Riemann extensions, Quart. J. Math., Oxford (2)
Ser. 3 (1952) 1928.
[38] S. Sasaki, On the differential geometry of tangent bundles, Tohoku Math. J. 10 (1958)
338354.
[39] M. Sekizawa, Natural transformations of affine connections on manifolds to metrics

on cotangent bundles, In: Proceedings of the 14th winter school on abstract analysis
(Srn, 1986, Rend. Circ. Mat. Palermo (2) Suppl. 14, 1987) 129142.
[40] M. Sekizawa, Natural transformations of symmetric affine connections on manifolds

to metrics on linear frame bundles: a classification, Monatshefte fur Math. 105 (1988)
229243.
[41] M. Sekizawa, Curvatures of tangent bundles with Cheeger-Gromoll metric, Tokyo J.

Math. 14 (1991) 407417.
[42] P. Stredder, Natural differential operators on Riemannian manifolds, Tohoku Math. J.

10 (1958) 338354.
[43] M. Toomanian, Killing vectorfields and infinitesimal affine transformations on a gen-

eralised Riemann extension, Tensor (N.S.) 32 (1978) 335338.
[44] L. Vanhecke and T. J. Willmore, Riemann extensions of DAtri spaces, Tensor (N.S.)
38 (1982) 154158.
[45] J. A. Wolf, Elliptic spaces in Grassmann manifolds, Illinois J. Math. 7 (1963) 447
462.
[46] Y. C. Wong and K. P. Mok, Connections and M-tensors on the tangent bundle T M , In:
Topics in Differential Geometry ((H. Rund and W. F. Forbes, Eds.) Academic Press,
New York, 1976) 157172.
[47] K. Yano and S. Ishihara, Tangent and Cotangent Bundles (Marcel Dekker, Inc., New
York, 1973).
Chapter 12
I NVARIANT VARIATIONAL P ROBLEMS

AND I NVARIANT F LOWS VIA M OVING F RAMES
Peter J. Olver
School of Mathematics, University of Minnesota
Minneapolis 55455, USA
Abstract
This paper reviews the moving frame approach to the construction of the invariant
variational bicomplex. Applications include explicit formulae for the Euler-Lagrange
equations of an invariant variational problem, and for the equations governing the
evolution of differential invariants under invariant submanifold flows.
1. Introduction
This survey paper describes some aspects of the authors recent research, done partly in
collaboration with Irina Kogan, [31, 53], into moving frames, the invariant variational bi-
complex, and invariant submanifold flows. These results are based on combining two pow-
erful ideas in the modern, geometric approach to differential equations and the variational
calculus. The first is the variational bicomplex, which is of fundamental importance in the
study of the geometry of jet bundles, differential equations and the calculus of variations.
Its origins can be found in the work of Dedecker, [15], then developed in full detail by Tul-
czyjew, [67], and Vinogradov, [68, 69]. Later contributions of Tsujishita, [66], Anderson,
[2, 3], and Krupka and Janyska, [32, 33], have amply demonstrated the power of the bicom-
plex formalism for both local and global problems in the geometric theory of differential
equations and the calculus of variations.
The second ingredient is a reformulation of Cartans method of moving frames, [17, 51].
For a general finite-dimensional transformation group G, a moving frame is defined as an
equivariant map from an open subset of jet space to the Lie group G. Moving frames are
constructed by the process of normalization based on the choice of cross-section to the

E-mail address: olver@umn.edu,www.math.umn.edu/ olver
210 Peter J. Olver
group orbits. The moving frame then provides a canonical mechanism, called invarianti-
zation, that allows us to systematically construct the invariant counterparts of all objects of
interest in the usual variational bicomplex, including differential invariants, invariant dif-
ferential forms, invariant differential operators, etc. The key recurrence formulae relate the
differentials of ordinary functions and forms to the invariant differentials of invariant func-
tions and forms, and thereby lead to the complete structure of the algebra1 of differential
invariants, including the syzygies and commutation formulae, [26, 27, 52]. The equivari-
ant moving frame method has impacted a remarkable range of subjects, including symme-
try methods for partial differential equations, the calculus of variations, classical invari-
ant theory, computer vision, numerical analysis, Hamiltonian systems, integrable soliton
equations, materials and micromagnetics, joint invariants, relativity, quantum mechanics,
invariants of Lie algebras, Lie pseudo-groups, symbolic methods, and (non-commutative)
differential algebra; see [50, 51] for recent surveys of developments in the field.
A key application of the invariant variational bicomplex is the general solution to an
outstanding problem in the calculus of variations. Every group-invariant variational prob-
lem can be written in terms of the differential invariants. The associated Euler-Lagrange
equations inherit the symmetry group, and so can also be written in terms of the differential
invariants. The problem is to directly construct the invariant form of the EulerLagrange
equations from the invariant form of the variational problem. Before the general solution to
this problem appeared in [31], only a few specific examples were known, [3, 22]. A striking
recent application of these techniques is the work of Starostin and van der Heijden, [65], on
equilibrium configurations of flexible Mobius bands.
A second application is to the evolution of differential invariants under invariant sub-
manifold flows. Invariant curve flows and surface flows arise in an impressive range
of applications, including geometric optics, [7], elastodynamics, [37], computer vision,
[55, 56, 60, 62, 64], visual tracking and control, [45], vortex dynamics, [25, 36], interface
motions, [64], thermal grooving, [9], and elsewhere. A celebrated example is the Euclidean
invariant curve shortening flow, [18, 20], in which a plane curve moves in its normal di-
rection in proportion to its curvature. In computer vision, Euclidean curve shortening and
its equi-affine counterpart have been successfully applied to image denoising and segmen-
tation, [55, 61, 62]. In three dimensional space, Euclidean-invariant curve flows include
the integrable vortex filament flow, [25, 36], while mean curvature and Willmore flows of
surfaces have been the subject of extensive analysis and applications, [6, 14].
Given an invariant submanifold flow, a key issue is to track the induced evolution of
its basic geometric invariants curvature, torsion and the like. While a number of partic-
ular examples have been worked out by direct computation, e.g., in [18, 43], many cases
of interest have yet to appear in the literature, owing to their computational complexity.
Therefore, it is worth developing general, practical tools to ameliorate this often tedious
task. Mansfield and van der Kamp, [39], have developed a method based on the differential
invariant syzygies. Here we present a direct approach, applying the invariant variational
bicomplex calculus discussed above. As we will see, the same basic invariant differential
operators appearing in the construction of invariant EulerLagrange equations also play a
1
Technically, because differential invariants may only be locally defined, we should speak of the sheaf of
differential invariants. However, as we work locally on suitable open subsets, this extra level of abstraction is
not required; moreover, experts can readily translate our constructions into sheaf-theoretic language, [70].
Invariant Variational Problems and Invariant Flows via Moving Frames 211
key role in this context.
2. The Invariant Variational Bicomplex

In this section, we review the basics of prolonged group actions on submanifold jets, moving
frames, and the induced invariant variational bicomplex. Basic references include [48, 49]
for jets, contact forms, and prolonged Lie group actions, [3, 66] for the variational bicom-
plex, [17, 51, 52] for the equivariant approach to moving frames, and [31] for the moving
frame construction of the invariant variational bicomplex. For simplicity, we will only deal
with finite-dimensional Lie group actions in this paper, although the general ideas can be
straightforwardly adapted to infinite-dimensional pseudo-group actions using more recent
extensions of the moving frame technology, [54].
Let G be an r-dimensional Lie group, acting smoothly on a m-dimensional manifold
M . We will study the induced action on p-dimensional submanifolds S M . For 0 n
, let Jn = Jn (M, p) denote the n-th order (extended) jet bundle for such submanifolds,
[49]. The action of G on M naturally prolongs to an action on Jn . Since the prolonged
group actions are all mutually compatible under projection Jn Jk , we will avoid explicit
reference to the order of prolongation, and just use g z (n) for the action of g G on the
jet z (n) Jn , rather than the more traditional notation g (n) z (n) .
By definition, a moving frame is right-equivariant2 map3 : Jn G, meaning that
(g z (n) ) = (z (n) ) g 1 for all g G and all z (n) Jn where defined. The existence
of a moving frame requires that the prolonged group action be free, meaning the isotropy
subgroups of each individual jet are trivial, and regular, meaning the prolonged group orbits
form a regular foliation, on an open subset V Jn . Under these conditions, a moving
frame can be algorithmically constructed by a normalization process based on the choice
of a compatible cross-section K n Jn to the group orbits. Specifically, given z (n) Jn ,
we set g = (z (n) ) to be the unique group element such that g z (n) K n , when defined.
Compatibility of moving frames under the jet space projections allows us to also suppress
the order in the notation of . We use to denote the invariantization process induced
by the moving frame. The invariantization of a differential form is the unique invariant
differential form () that agrees with when restricted to the cross-section. In particular,
if is an invariant differential form or function, then () = . Invariantization defines an
(exterior) algebra morphism that projects differential functions and forms on Jn to invariant
differential functions and forms.
Let (x, u) = (x1 , . . . , xp , u1 , . . . , uq ) be local coordinates on M . Viewing the xs
as independent variables and the us as dependent variables, we let uJ = #J u/xJ be
the usual induced local coordinates on Jn . Separating the local coordinates (x, u) on M
into independent and dependent variables naturally splits the differential one-forms on J
into horizontal forms, spanned by dx1 , . . . , dxp , and vertical forms, spanned by the basic
2
All classical moving frames, [23], are left-equivariant, and can be obtained by composing with the group
inversion g 7 g 1 . We choose to concentrate on the right-equivariant version to (slightly) simplify some of
the calculations.
3
All maps, differential forms, differential functions, etc., need only be locally defined; thus, the domain of
is typically a suitable open subset of Jn .
212 Peter J. Olver
contact one-forms
p
X
J = duJ uJ,i dxi , = 1, . . . , q, #J 0. (1)
i=1
Let H and V denote the projections mapping one-forms on J to their horizontal and
vertical (contact) components, respectively. The induced splitting d = dH + dV of the
differential into horizontal and vertical components results in the variational bicomplex4 . In
particular, if F (x, u(n) ) is any differential function, its horizontal and vertical differentials
are
p
X X F X F
dH F = (Di F ) dxi , dV F = DF () = DJ = , (2)

uJ uJ J
i=1 ,J ,J
in which Di = Dxi denote the total derivative operators with respect to the independent
variables, DJ = Dj1 Djk are the higher order total derivatives, = (1 , . . . , q )T is
the column vector containing the order zero contact forms, while DF = (DF,1 , . . . , DF,q )
is the Frechet derivative or formal linearization of the differential function F .
We will employ our moving frame to invariantize the variational bicomplex as follows.
First, invariantization of the jet coordinate functions produces the fundamental differential
invariants:
H i = (xi ), IJ = (uJ ), = 1, . . . , q, #J 0. (3)
These naturally split into two classes: The r = dim G combinations defining the cross-
section equations will be constant, and are known as the phantom differential invariants.
The remainder, called the basic differential invariants, form a complete system of function-
ally independent differential invariants. Next, let
i = i + i = (dxi ), where i = H (i ), i = V (i ), (4)
denote the invariantized horizontal one-forms. Their horizontal components 1 , . . . , p

form, in the language of [49], a contact-invariant coframe for the prolonged group action,
while 1 , . . . , p supply contact corrections that make the one-forms 1 , . . . , p fully
G-invariant. The corresponding dual invariant total differential operators D1 , . . . , Dp are
defined so that
H i = (xi ), IJ = (uJ ), = 1, . . . , q, #J 0. (5)
for any differential function F and, more generally, differential form , on which the Di
act via Lie differentiation. Finally, let
J = (J ), = 1, . . . , q, #J 0. (6)
4
Since the splitting depends on a choice of independent variables on M , the variational bicomplex is not
intrinsic. A more refined version of this construction, known as the C spectral sequence, [68, 69], relies on
the contact filtration of the algebra of differential forms. However, since all our calculations take place in local
coordinates, we will avoid all the extra complications inherent in this more sophisticated machinery. Experts
will be able to readily translate our results as desired.
be the invariantized basis contact forms.

As in the usual, non-invariant bicomplex construction, the decomposition of invariant
one-forms on J into invariant horizontal and invariant contact components induces a de-
composition of the differential. However, now d = dH + dV + dW splits into three
constituents, where dH adds an invariant horizontal form, dV adds a invariant contact
form, while dW replaces an invariant horizontal one-form with a combination of wedge
e r,s denote the space
products of two invariant contact forms. In other words, if we let
of differential forms of degree r + s spanned by wedge products of r invariant horizontal
one-forms (4) and s invariant contact one-forms (6), then
e r,s
dH : e r+1,s , e r,s
dV : e r,s+1 , e r,s
dW : e r1,s+2 . (7)
The resulting invariant variational quasi-tricomplex is characterized by the formulae
d2H = 0, dH dV + dV dH = 0,
d2V + dH dW + dW dH = 0. (8)
d2W = 0, dV dW + dW dV = 0,
Fortunately, the third, anomalous component dW plays no role in the applications; in par-
ticular, dW F = 0 for any differential function F .
The most important fact underlying the moving frame construction is that the in-
variantization map does not respect the exterior derivative operator. Thus, in general,
d () 6= (d). The recurrence formulae, [17, 31], which we now review, provide the
missing correction terms d() (d). Remarkably, these formulas can be explicitly
and algorithmically constructed using only linear differential algebra without knowing
the explicit formulas for either the differential invariants or invariant differential forms, the
invariant differential operators, or even the moving frame! The only required ingredients
are the cross-section equations and the formulae for the prolonged infinitesimal generators
of the group action.
Let v1 , . . . , vr be a basis for the infinitesimal generators of our transformation group.
We prolong each infinitesimal generator to Jn . For conciseness, we will retain the same
notation v for the prolonged vector fields on any Jn which, in local coordinates, take the
form
p q n
X
i X X
v = (x, u) i + J, (x, u(j) ) , = 1, . . . , r. (9)
x uJ
i=1 =1 j=#J=0
The coefficients J, = v (uJ ) can be successively constructed by Lies recursive prolon-

gation formula, [48, 49]:
p
X
Ji, = Di J, uJj Di j . (10)
j=1
A straightforward induction establishes the explicit prolongation formula, first written down
by the author in [47]:
p
X p
X
J, = DJ Q + i uJ,i , where Q = i ui (11)
i=1 i=1
are the components of the characteristic of v .

214 Peter J. Olver
Strikingly, all the recurrence relations are consequences of a single universal recurrence
formula that prescribes the differential of an invariantized differential function or form.
Theorem 1. If is any differential form on J , then
r
X
d () = (d ) + [v ()], (12)
=1
where 1 , . . . , r are the invariantized MaurerCartan forms dual to the infinitesimal gener-
ators v1 , . . . , vr , while v () denotes the Lie derivative of with respect to the prolonged
infinitesimal generator v .
The invariantized MaurerCartan forms 1 , . . . , r are obtained by pulling back the
usual dual MaurerCartan forms 1 , . . . , r on G by the moving frame map: = .
Details would take us too far afield, [31], but, fortunately, are not required thanks to the
following marvelous result that allows us to compute them directly without reference to
their underlying definition:
Lemma 2. Let I1 = (z1 ), . . . , Ir = (zr ) be the phantom differential invariants stemming
from our cross-section. Then the corresponding phantom recurrence formulae
r
X
0 = dI = d (z ) = (dz ) + [v (z )], = 1, . . . , r, (13)
=1
can be uniquely solved for the invariantized MaurerCartan forms 1 , . . . , r .

Having solved the linear system (13) for 1 , . . . , r , we then decompose the resulting
invariantized MaurerCartan forms into their invariant horizontal and contact components:
p
X X

= + , where = Ri i , = S,J J , (14)
i=1 ,J
where Ri , S,J are certain differential invariants. The Ri will be called the MaurerCartan
invariants, [26, 27, 52]. In the case of curves, the Ri appear as the entries of the Frenet
Serret matrix D(x, u(n) ) (x, u(n) )1 , in the case G GL(N ) is a matrix Lie group,
[23]. Substituting (14) back into the universal formula (12) produces a complete system
of explicit recurrence relations for all the differentiated invariants and invariant differential
forms.
In particular, taking to be any one of the individual jet coordinate functions xi , uJ ,
results in the recurrence formulae for the fundamental differential invariants (3):
r
X r
X
dH i = (dxi ) + [v (xi )] = i + (i ) ,
=1 =1
r p r
!
X X X
dIJ = (duJ ) +
[v (uJ )] = uJi dxi + J + (J, ) (15)
=1 i=1 =1
p
X r
X

= IJi i + J + (J, ) .
i=1 =1
In view of (14), the coefficient of i in (15) yields the recurrence relations

r r
ij
X X
j
Di H = + Ri (i ), Di IJ =
IJi + Ri (J, ), (16)
=1 =1
where ij is the usual Kronecker delta. Owing to the functional independence of the basic
(non-phantom) differential invariants, these formulae, in fact, serve to completely char-
acterize the structure of the non-commutative differential invariant algebra, [17, 26, 52].
Similarly, the contact components in (15) yield the vertical recurrence formulae
r
X r
X
dV H i = (i ) , dV IJ = J + (K ) , (17)
=1 =1
while, as noted earlier, dW H i = dW IJ = 0.

The recurrence formulae (12) for the derivatives of the invariant horizontal forms are
r
X
i i 2 i
d = d[(dx )] = (d x ) + [v (dxi )]
=1
r p q
!
X X X i

= Dk i dxk + (18)
u
=1 k=1 =1
r X p r Xq i
X
i
k
X
= Dk + .
u
=1 k=1 =1 =1
The resulting two-form can be decomposed into three basic constituents, belonging, re-
spectively, to the invariant summands e 2,0
e 1,1
e 0,2 . In view of (14), the terms in (18)
involving wedge products of two horizontal forms, i.e., in e 2,0 , yield
X
dH i = Yjki
j k ,
j<k
where
p
r X
X
i
Yjk = Rj (Dj i ) Rk (Dk i ) (19)
=1 j=1
are called the commutator invariants, since combining (19) with (5) produces the commu-
tation formulae for the invariant differential operators:
p
X p
X
i i
[ Dj , Dk ] = Yjk Di = Ykj Di . (20)
i=1 i=1
Next, the terms in (18) involving wedge products of a horizontal and a contact form yield
r
" q p
#
i
X X X
i i k
dV = + (Dk ) . (21)
u
=1 =1 k=1
216 Peter J. Olver
Finally, the remaining terms, involving wedge products of two contact forms, provide the
formulas for the anomalous third component of the differential:
r Xq i
i
X
dW = . (22)
u
=1 =1
In a similar fashion, we derive the recurrence formulae (12) for the differentiated invariant
contact forms:
r p r
!
X X X
i
dJ = d[(J )] = (dJ ) + [v (J )] = dx Ji + (J

),
=1 i=1 =1
(23)
where
p
X p
X

v (J ) dJ Ji dxi uJi di dV J uJi dV i

J = = + = (24)
i=1 i=1
is known as the vertical prolongation coefficient of the vector field v . For our purposes,
we only require the component of (23) that involves invariant horizontal forms:
p
X r
X
dH J = i
Ji + (J

). (25)
i=1 =1
Since5
p
X
dH = i Di (26)
i=1
for any contact form , we deduce the recurrence formulae
r
X
Di J = Ji + Ri (J

) (27)
=1
for the invariant (Lie) derivatives of the invariant contact forms. The latter can inductively
be solved to express the higher order invariantized contact forms as certain invariant deriva-
tives of those of order 0:
Xq

J = EJ, ( ) = EJ (), (28)
=1
in which = (1 , . . . , q )T denotes the column vector containing the order zero invari-
antized contact forms, while EJ = (EJ , . . . , EJ ) are certain invariant differential operators.
In view of (17, 28), if K = K(. . . H i . . . IJ . . .) is any differential invariant, we can
write its invariant vertical derivative in the form
p q
X K i
X K

X
dV K = d V H + d V JI = A K () = AK, ( ), (29)
H i IJ
i=1 ,J =1
in which AK = (AK,1 , . . . , AK,q ) is a row vector of invariant differential operators. We

view (29) as the invariant version of the vertical differentiation formula dV F = DF (),
cf. (2), which motivates the following terminology.
5
Warning: The analogous formula is not valid for horizontal forms.
Definition 3. The invariant linearization of a differential invariant K is the invariant dif-

ferential operator AK that satisfies dV K = AK ().
Remark. In [31], AK was called the Eulerian operator associated with K owing to its ap-
pearance in the differential invariant form of the EulerLagrange equations for an invariant
variational problem; see Theorem 5 below.
Similarly, we combine (14), (21), and (28), to produce formulae
p X
X q p
X
dV i = i
Bj ( ) j = Bji () j (30)
j=1 =1 j=1
for the vertical differentials of the invariant horizontal forms, in which Bji = (Bj1
i , . . . , Bi )
jq
2
is a family of p row-vector-valued invariant differential operators, known, collectively, as
the invariant Hamiltonian operator complex, cf. [58], again stemming from its role in the
invariant calculus of variations.
Example 4. The Euclidean geometry of plane curves is governed by the standard action
y = x cos u sin + a, v = x sin + u cos + b, of the proper Euclidean group
g = (, a, b) SE(2) on M = R 2 . The prolonged group transformations are constructed
by applying the implicit differentiation operator Dy = (cos ux sin )1 Dx to v, and so
sin + ux cos uxx
vy = , vyy = , etc.
cos ux sin (cos ux sin )3
Solving the normalization equations y = v = vy = 0 for the group parameters produces
the right moving frame
x + uux xux u
= tan1 ux , a=p , b= p . (31)
1 + u2x 1 + u2x
(The classical moving frame, [23], is the left counterpart obtained by inverting the group
element given in (31).) Invariantization of the coordinate functions, which is done by sub-
stituting the moving frame formulae into the prolonged group transformations, produces
the fundamental normalized differential invariants
(x) = H = 0, (u) = I0 = 0, (ux ) = I1 = 0,
(uxx ) = I2 = , (uxxx ) = I3 = s , (uxxxx ) = I4 = ss + 33 ,
and so on. The first three, arising from the normalizations, are called phantom invariants.
The lowest order non-trivial differential invariant is the Euclidean curvature I2 = =
uxx (1 + u2xp
)3/2 , while s , ss , . . . denote the derivatives of with respect to the arc-length
form = 1 + u2x dx. The invariant horizontal one-form
dx + ux du p ux
= (dx) = p = 1 + u2x dx + p (32)
2
1 + ux 1 + u2x
is a sum of the contact-invariant arc length form along with a contact correction. In the
same manner we obtain the basis invariant contact forms
(1 + u2x ) x ux uxx
= () = p , 1 = (x ) = , ... . (33)
1 + u2x (1 + u2x )2
218 Peter J. Olver
To obtain the explicit recurrence formulae, we begin with the prolonged infinitesimal
generators of SE(2):
v1 = x , v2 = u , v3 = u x + x u + (1 + u2x ) ux + 3ux uxx uxx + .
The one-forms , governing the correction terms are found by applying the recurrence
formulae (12) to the phantom invariants. From the first equation in (12), we obtain
0 = dH H = (dH x) + (v1 (x)) 1 + (v2 (x)) 2 + (v3 (x)) 3 = + 1 ,

0 = dH I0 = (dH u) + (v1 (u)) 1 + (v2 (u)) 2 + (v3 (u)) 3 = 2 ,
0 = dH I1 = (dH ux ) + (v1 (ux )) 1 + (v2 (ux )) 2 + (v3 (ux )) 3 = + 3 ,
and hence 1 = , 2 = 0, 3 = . Similarly, applying dV to the phantom invari-

ants and using the second equation in (12) yields 1 = 0, 2 = , 3 = 1 . We are
now ready to substitute the non-phantom invariants into (12). The horizontal differentials
dH Ik of the normalized differential invariants In = (un ) are used to produce the explicit
recurrence formulae
= I2 , s = DI2 = I3 , ss = DI3 = I4 3I23 , ...
relating them to the differentiated invariants Dm . Similarly, the second equation in (12)
gives the vertical differential
dV I2 = dV = (2 ) + (v3 (uxx )) 3 = 2 = (D2 + 2 ) , (34)
where the final equation follows from the invariant contact form recurrence formulae D =
1 , D1 = 2 2 , which are found by applying dH to the invariant contact forms
and using the first equation in (12). Thus, we deduce the following invariant linearization
operators:
A = D2 + 2 , As = D3 + 2 D + 3 s ,
(35)
Ass = D4 + 2 D2 + 5 s D + 4 ss + 3 2s ,
etc. In fact, one can recursively construct the higher order operators starting with A via
An = D An1 + n , (36)
where n = Dn . Finally, applying the second formula in (12) to yields
dV = ,
and hence the invariant Hamiltonian operator is
B = . (37)
3. Invariant Variational Problems

We now apply our construction to derive the formulae for the Euler-Lagrange equations
associated with an invariant variational problem. Let us recall the variational bicomplex
construction of the Euler-Lagrange equations.
R
A variational problem I[ u ] = L[ u ] dx is determined by the Lagrangian form =
L[ u ] dx p,0 . Its differential d = dV p,1 defines a form of type (p, 1). We
introduce an equivalence relation on such forms, so that if and only if = + dH
for some p1,1 . The quotient space F 1 = p,1 / is known as the space of source
forms.
Pq Integration by parts proves that every source form has a canonical representative
(n) ) dx, and so can be identified with a q-tuple of differential functions
=1 (x, u
= (1 , . . . , q ). In applications, a source form is regarded as defining a system of q
differential equations 1 = = q = 0 for the q dependent variables u = (u1 , . . . , uq ).
Composing the differential d : p,0 p,1 with the projection : p,1 F 1 produces
the variational differential = d that takes a Lagrangian form = L[ u ] dx to its
variational derivative source form
q
X X L
E (L) dx, where E (L) = (D)J (38)
uJ
=1 J
are the classical Euler-Lagrange expressions for the Lagrangian L.

According to Lie, [38, 49], as long as we work on the open subset V Jn where
G acts regularly and freely, any G-invariant variational problem is given by an invariant
Lagrangian form = L e , where = 1 p is the contact-invariant volume form,
and the invariant Lagrangian L e is an arbitrary differential invariant, and hence a function
of the fundamental differential invariants I 1 , . . . , I l and their invariant derivatives DJ I .
The associated Euler-Lagrange equations E(L) = 0 admit G as a symmetry group, and so,
under suitable nondegeneracy hypotheses, [49, Theorem 6.25], can themselves be written in
terms of the differential invariants. The main problem is to go directly from the differential
invariant formula for the variational problem to the differential invariant formula for the
Euler-Lagrange equations.
Not surprisingly, the required calculations rely on an invariant version of integration by
parts. For this purpose, given invariant differential forms , , let us write whenever
= + dH . If F is a differential function and is a contact one-form, then6
p
X p
X
i
dH F = (Di F ) dH = i Di . (39)
i=1 i=1
To effect the computation, we begin by replacing our Lagrangian = L e by the

fully invariant version = L , noting that, since they differ by contact forms, they have
e e
identical EulerLagrange equations. We then compute
X Le

dV
e=
dV I,K + L dV ,
e (40)
I,K
,K
Introduce the (p 1)forms
(i) = Di e p1,0 .
= (1)i1 1 i1 i+1 p
6
Warning: The second identity is not true for a general one-form.
220 Peter J. Olver
If F is any differential function and any contact one-form, then
dH (F (i) ) = dH F (i) + F dH (i) F dH (i) . (41)
Since dH (i) e p,0 , it must be a multiple of the invariant volume form, and we write
dH (i) = Zi , where Z1 , . . . , Zp are certain differential invariants, which we will call
the twist invariants. Using (39) we can rewrite (41) as
F dH (i) = F (Di ) (Di + Zi )F = (Di F ) ,

(42)
where Di = (Di + Zi ) is called the twisted invariant adjoint of the invariant differential
operator Di . If we choose = dV H where H is a differential function, then (42) results
in the multivariate invariant integration by parts formula
p
F d(Di H) = (Di F ) dV H
X
F (Dj H) dV j (i) . (43)
j=1
We use (43) repeatedly to integrate the first term of (40) by parts, leading to
q
X p
X
dV
e e dV I
E (L) Hji (L)
e dV j (i) , (44)
=1 i=1
where
q X
L L
DK DK
X e X e
E (L)
e =
, Hji (L) e ji +
e = L
I,J,j , (45)
I,K I,J,i,K
K =1 J,K
are, respectively, the invariant Eulerian and invariant Hamiltonian tensor of the invariant
Lagrangian L.e In (45), we use the twisted adjoints
DK = Dk1 Dkm = (1)m (Dk1 + Zk1 ) (Dkm + Zkm ), K = (k1 , . . . , km ),
of the repeated invariant differential operators.

The second phase of the computation requires the vertical differentiation formulae
q q
j
X X

dV I = A ( ), dV = j
Bi, ( ) i , (46)
=1 =1
where A = A denotes the Eulerian operator, which is an m q matrix of invariant

differential operators whose rows are the invariant linearizations of the fundamental differ-
1 l 2 j j
ential invariants I , . . . , I , while the p row vectors Bi = Bi, of invariant differential
operators form the invariant Hamiltonian operator complex. This allows us to write (44) in
the vectorial form
p
e B j () .
X
dV
e E(L)
e A() Hji (L) i
i,j=1
We now apply (42) to further integrate both terms by parts. The final result is written in
terms of twisted adjoints of the Eulerlian and Hamiltonian operators,

p
(Bij ) Hji (L)
X
dV
e e = A E(L) e e .
i,j=1
Theorem 5. The Euler-Lagrange equations of the invariant Lagrangian form

e =
L(I
e (n) ) have the following invariant form:
p
(Bij ) Hji (L)
X

A E(L)
e e = 0. (47)
i,j=1
In the case of curves, when p = 1, there are no twist invariants, and so the general
formula (47) reduces to
A E(L)
e B H(L) e = 0, (48)
where A and B are the ordinary formal adjoints of the invariant Eulerian and Hamiltonian
operators, respectively.
Example 6. In the context of the Euclidean group acting on plane curves in Example 4,
any Euclidean-invariant variational problem corresponds to a contact invariant Lagrangian
= L(,
e s , ss , . . .) . Both the Eulerian operator (35) and the Hamiltonian operator
(37) are invariantly self-adjoint: A = A and B = B . Thus, the invariant Euler-Lagrange
formula (48) reduces to the known formula, [3, 22],
(D2 + 2 ) E(L)
e + H(L)
e =0
for the Euclidean-invariant Euler-Lagrange equation.
Example 7. Consider the standard action of the Euclidean group SE(3) on surfaces S
R 3 . We assume that the surface is parametrized by z = (x, y, u(x, y)), noting that the
final formulae are, in fact, parameter-independent. The classical (local) left moving frame
(x, u(2) ) = (R, a) SE(3) consists of the point on the curve defining the translation
component a = z, while the columns of the rotation matrix R contain the unit tangent
vectors forming the Frenet frame along with the unit normal to the surface. The fundamental
differential invariants are the principal curvatures 1 = (uxx ), 2 = (uyy ). The mean and
Gaussian curvature invariants H = 21 (1 + 2 ), K = 1 2 , are often used as convenient
alternatives, since they eliminate some of the residual discrete ambiguities in the moving
frame. Higher order differential invariants are obtained by repeatedly applying the dual
invariant differential operators D1 , D2 associated with the diagonalizing Frenet coframe
1 = (dx1 ), 2 = (dx2 ). The resulting differentiated invariants are not functionally
independent, owing to the Codazzi identity
1,1 2,1 + 1,2 2,2 2(2,1 )2 2(1,2 )2

1,22 2,11 + 1 2 (1 2 ) = 0. (49)
1 2
222 Peter J. Olver
The Codazzi syzygy can, in fact, be directly deduced from our infinitesimal moving frame
computations by comparing the recurrence formulae for 1,22 and 2,11 with the normalized
invariant (uxxyy ).
Any Euclidean-invariant variational problem has the form
Z
e (n) ) 1 2 ,
L( where 1 2 = 2,0 (1 2 )
is the usual intrinsic surface area 2-form. The invariant Lagrangian L e is an arbitrary differ-
ential invariant, and so can be rewritten in terms of the principal curvature invariants and
their derivatives, or, equivalently, in terms of the Gaussian and mean curvatures. The former
representation leads to simpler formulae and will be retained. Since
2,1
2 1
1,2
dH (1) = dH = 1 , dH (2) = dH = 2 ,
2 1
the twist invariants are
2,1 1,2
Z1 = , Z2 = .
1 2 2 1
These invariants appear in Guggenheimers proof of the fundamental existence theorem for
Euclidean surfaces, [23, p. 234]. The denominator vanishes at umbilic points on the surface,
where the moving frame is not valid. The Codazzi syzygy (49) can be written compactly as
K = 1 2 = D1 (Z1 ) + D2 (Z2 ) = (D1 + Z1 )Z1 (D2 + Z2 )Z2 ,
which expresses the Gaussian curvature K as an invariant divergence. This fact lies at
the heart of the GaussBonnet Theorem. The invariant vertical derivatives of the principal
curvatures are straightforwardly determined from (12),
dV 1 = (xx ) = D12 + Z2 D2 + (1 )2 ,

dV 2 = (yy ) = D22 + Z1 D1 + (2 )2 ,

where = () = (du ux dx u y dy) is the fundamental

invariant contact form.
D12 + Z2 D2 + (1 )2
Therefore, the Eulerian operator is A = . Further,
D22 + Z1 D1 + (2 )2
D1 D2 Z2 D1 2

1 1 1
dV = + ,
1 2
D2 D1 Z1 D2 1

dV 2 = 2 2 ,
2 1
which yields the Hamiltonian operator complex
B11 = 1 , B22 = 2 ,
1 1
B21 = D2 D1 Z1 D2 = B12 .

D1 D2 Z2 D1 = 1
1 2 2
Therefore, according to our formula (47), the Euler-Lagrange equation for a Euclidean-
invariant variational problem is
0 = (D1 + Z1 )2 (D2 + Z2 ) Z2 + (1 )2 E1 (L)

e
+ (D2 + Z2 )2 (D1 + Z1 ) Z1 + (2 )2 E2 (L)
e + 1 H11 (L)
e + 2 H22 (L)

e
!
H21 (L)
e H2 (L)
1
e
+ (D2 + Z2 )(D1 + Z1 ) + (D1 + Z1 ) Z2 1 2
.

e are the invariant Eulerians with respect to the principal curvatures ,

As before, E (L)
i
while Hj (L) e 1 , 2 ) does not depend on
e are the invariant Hamiltonians. In particular, if L(
any differentiated invariants, the Euler-Lagrange equation reduces to
L 2 2 L
(D1 )2 + D2 Z2 + (1 )2
e 2 e
(1 + 2 )L

+ (D2 ) + D 1 Z1 + ( ) e = 0.
1 2
For example, the problem of minimizing surface area has invariant Lagrangian L
e = 1, and
1 2
so has the well-known Euler-Lagrange equation E(L) = ( + ) = 2H = 0, and
hence minimal surfaces have vanishing mean curvature. The mean curvature Lagrangian
e = H = 1 (1 + 2 ) has Euler-Lagrange equation
L 2
1 2 2 2 1 2 2
1
= 1 2 = K = 0.

2 ( ) + ( ) ( + )
e = 1 (1 )2 + 1 (2 )2 , [3, 8], the Euler-Lagrange equation is

For the Willmore Lagrangian L 2 2
0 = E(L) = (1 + 2 ) + 12 (1 + 2 )(1 2 )2 = 2 H + 4 (H 2 K)H,
where = (D1 + Z1 )D1 + (D2 + Z2 )D2 = D1 D1 D2 D2 is the LaplaceBeltrami

operator on our surface.
4. Invariant Submanifold Flows

In this section, we shift our attention to invariant submanifold flows. Let us single out the
m = p + q invariant one-forms
1 , . . . , p , 1 , . . . , q (50)
consisting of the invariant horizontal forms i = (dxi ) and the order 0 invariant contact
forms = ( ). Each is a linear combination of the coordinate one-forms dx1 , . . . , dxp ,
du1 , . . . , duq on M , whose coefficients are certain n-th order differential functions, where
n is the order of the underlying moving frame.
Let S M be a p-dimensional submanifold. Evaluating the coefficients of (50) on the
submanifold jet (x, u(n) ) = jn S|z produces a basis for the cotangent space T M |z of the
ambient manifold at z = (x, u) S, which we continue to denote by (50). By construction,
the resulting cotangent space basis is equivariant under the action of G on S M .
Let t1 , . . . , tp , n1 , . . . , nq , denote the corresponding dual tangent vectors, which form
a Gequivariant basis of the bundle T M S, or frame on S. Since the contact forms
224 Peter J. Olver
annihilate the tangent space to S, the vectors t1 , . . . , tp form a basis for the tangent bundle
T S, while n1 , . . . , nq form a basis for the complementary Gequivariant normal bundle
N S S induced by the moving frame. In classical geometrical situations, [23], they can
be identified with the classical moving frame vectors.
Example 8. Let us return to the case of planar Euclidean curves C M = R 2 . According

to Example 4, the invariant coframe is given by the invariant horizontal form (32) and the
order 0 invariant contact form in (33). The corresponding dual frame vectors are the usual
(right-handed) Euclidean frame vectors the unit tangent and unit normal:

1 1
t= p + ux , n= p ux + . (51)
1 + u2x x u 1 + u2x x u
In general, let
p
X q
X
j
V = V |S = VT + VN = I tj + J n (52)
j=1 =1
be a section of the bundle T M S, where VT , VN denote, respectively, its tangential and

normal components, while I j , J are differential functions, depending on the submanifold
jets. We will, somewhat imprecisely, refer to V as a vector field, even though it is only
defined on S. Any such vector field generates a submanifold flow:
S
= V|S(t) , (53)
t
which forms an n-th order system of partial differential equations, where n refers to the
larger of the order of our moving frame and the coefficients I j , J . Assuming local ex-
istence and uniqueness, a solution S(t) to the submanifold flow equations (53) defines a
smoothly varying family of p-dimensional submanifolds of M . On the other hand, one
typically expects singularities to appear if the flow is continued for a sufficiently long time.
The submanifold flow (53) is called G-invariant if G is a symmetry group of the partial
differential equation, which requires that its coefficients I j = h V ; j i, J = h V ; i,
be differential invariants.
The tangential components VT do not affect the extrinsic geometry of the submanifold,
but only its internal parametrization. Thus, if we are only interested in the images of S(t)
under the flow, and not their underlying parametrizations, we can set VT = 0 without loss
of generality. Therefore, the normal component
q
X
VN = J n (54)
=1
serves to characterize the same invariant submanifold flow as V, modulo reparametrization.

We will say that the vector field VN generates a normal flow, since it only moves the
submanifold in its G-equivariant normal direction as prescribed by the moving frame.
Example 9. The most well-studied are the Euclidean-invariant curve and surface flows. A
plane curve flow is generated by a vector field of the form
V=It+Jn or, equivalently, VN = J n, (55)
if we are not concerned about the tangential components effect of the parametrization of
the curve. In this case, n denotes (one of the two) Euclidean normals to the curve; by
convention, we use the inwards normal n when the curve is closed. Particular cases include:
i. V = n: this induces the geometric optics or grassfire flow, [7, 61];
ii. V = n: this generates the celebrated curve shortening flow, [18, 20], used to great
effect in image processing, [55, 61];
iii. V = 1/3 n: the induced flow is equivalent, modulo reparametrization, to the equi-
affine invariant curve shortening flow, also effective in image processing, [4, 55, 61];
iv. V = s n: this flow induces the modified KortewegdeVries equation for the curva-
ture evolution, and is the simplest of a large number of soliton equations arising in
geometric curve flows, [13, 19, 42];
v. V = ss n: this flow models thermal grooving of metals, [9].

A second important class are the invariant curve flows that preserve arc length. Remark-
ably, in many classical geometries, certain basic intrinsic curve flows induce integrable,
soliton evolutions for the differential invariants. The prototypical example is the Euclidean
invariant vortex filament flow studied by Hasimoto, [25, 35, 36]. The curvature and torsion
invariants of the evolving filament satisfy an integrable dynamical system, which can be
mapped to the completely integrable nonlinear Schrodinger equation, [1]. This led Lamb,
[34], to draw attention to the surprisingly common, but still poorly understood connection
between invariant curve flows and integrable soliton dynamics; since then, many other ex-
amples have been found, [5, 12, 13, 16, 19, 24, 28, 40, 41, 42, 44, 57, 59]. By integrable,
we shall mean that the evolution equation possesses a recursion operator, [46], inducing an
infinite hierarchy of higher order symmetries. However, not all induced differential invari-
ant evolutions are integrable, and, at present, we do not understand the general conditions
on the group action and invariant curve flow needed to guarantee integrability.
When p = 1, there is only one independent invariant horizontal one-form
= + = ds + , (56)
whose horizontal component = ds can be identified with the G-invariant arc length
element. Invariance requires that the Lie derivative V() vanishes on the submanifold,
which (because Lie derivatives preserve the contact ideal) implies the following:
Lemma 10. The curve flow induced by
q
X
V=It+ J n , where I = h V ; i, J = h V ; i, (57)
=1
preserves arc length if and only if the Lie derivative V() is a contact form.
226 Peter J. Olver
Submanifolds of dimension p 2 do not have distinguished parametrizations to play

the role of the arc length parameter; this is because the invariant horizontal forms are almost
never exact on the submanifold. On the other hand, the Lie derivative condition can be
straightforwardly mimicked.
Definition 11. The invariant submanifold flow induced by V is called intrinsic if V(i )
0 for all i = 1, . . . , p.
Lemma 12. If the vectorfield V defines an intrinsic flow, then it commutes with the invariant
differentiations: V, Di = 0 for i = 1, . . . , p. This holds if and only if
p
X q
X
i i k i
Dj I + Yjk I + Bj (J ) = 0. (58)
j,k=1 =1
In particular, for curve flows generated by (57), the condition (58) guaranteeing arc
length preservation reduces to
q
X
DI = B(J) = B (J ), (59)
=1
where D is the arc length derivative, while B = (B1 , . . . , Bq ) is the invariant Hamiltonian
operator, defined by (30).
Example 13. For the Euclidean group action on plane curves, in view of (30), the condition
that a curve flow generated by the vector field V = I t + J n be intrinsic is that
DI = J. (60)
Most of the curve flows listed in Example 9 have non-local intrinsic counterparts owing
to the non-invertibility of the arc length derivative operator on J. An exception is the
modified Korteweg-deVries flow, where J = s , and so I = 12 2 . In general, the normal
flow induced by VN = J n has a local intrinsic version if and only if E( J) = 0, where E
is the invariantized EulerLagrange operator, [31].
The next result prescribes the evolution of differential invariants under general intrinsic
and normal invariant submanifold flows. See [53] for the proof.
Theorem 14. Let K be any differential invariant. If the submanifold flow (53) is intrinsic,
then
p
K X
= V(K) = AK (J) + I i Di K. (61)
t
i=1
If the submanifold flow (53) is normal, then
K
= V(K) = AK (J). (62)
t
Example 15. For any of the Euclidean invariant normal plane curve flows Ct = J n listed
in Example 9, we have, according to Example 4,
s
= (D2 + 2 ) J, = (D3 + 2 D + 3 s ) J,
t t (63)
ss
= (D4 + 2 D2 + 5 s D + 4 ss + 3 2s ) J.
t
For instance, for the grassfire flow J = 1, and so
s ss
= 2 , = 3 s , = 4 ss + 3 2s . (64)
t t t
The first equation immediately implies finite time blow-up at a caustic for a convex initial
curve segment, where > 0. For the curve shortening flow, J = , and
s ss
= ss + 3 , = sss + 4 2 s , = ssss + 5 2 ss + 8 2s , (65)
t t t
thereby recovering formulas used in Gage and Hamiltons analysis, [18]; see also Mikula
and Sevcovic, [43]. Finally, for the mKdV flow, J = s ,
s
= sss + 2 s , = ssss + 2 ss + 3 2s ,
t t (66)
ss
= sssss + 2 sss + 9 s ss + 3 3s .
t
Warning: Normal flows do not preserve arc length, and so the arc length parameter s will
vary in time. Or, to phrase it another way, time differentiation t and arc length differenti-
ation D = Ds do not commute as can easily be seen in the preceding examples. Thus,
one must be very careful not to interpret the resulting evolutions (6466) as partial differ-
ential equations in the usual sense. Rather, one should regard the differential invariants
, s , ss , . . . as satisfying an infinite dimensional dynamical system of coupled ordinary
differential equations.
Turning our attention to the intrinsic, arc length preserving curve flow, the complication
alluded to in the preceding paragraph does not arise because, by Lemma 12, time differen-
tiation now commutes with arc length differentiation. Substituting (59) in the formula (61):
Theorem 16. Under an arc-length preserving flow,
t = R (J) where R = A s D1 B (67)
is the characteristic operator associated with . More generally, the time evolution of n =
Dn is given by arc length differentiation: n /t = Dn R (J).
In this case arc length is preserved, and hence the arc length and time derivatives com-
mute. Thus, unlike (62), the arc-length preserving flow (67) is of a more usual analytical
form. However, there is a complication in that the term
Z
1
s D B(J) = s B(J) ds (68)
228 Peter J. Olver
may very well be nonlocal, and so (67) is, in general, an integro-differential equation.
Note that any integration constant appearing in (68) just adds in a multiple of s , which
represents the arc length preserving tangential flow t = s that just serves to translate the
arc length parameter: s 7 s + c and so can be effectively ignored. Also, on a closed curve,
the integral in (68) need not be periodic in s, and so one may not be able to continuously
assign a uniquely determined evolution along the entire curve although, by the preceding
remarks, all such evolutions only differ by an overall translation, by an integer multiple of
the total length of the curve, of the arc length parameter.
In certain situations, (67) turns out to be a well-known local integrable evolution equa-
tion, and the characteristic operator R is its recursion operator!
Example 17. In the case of Euclidean plane curves, the evolution of the curvature is
given by
t = R (J), (69)
where
R = A s D1 B = D2 + 2 + s D1 = Ds2 + 2 + s Ds1 (70)

is the modified Korteweg-deVries recursion operator, [48]. In particular, for the mKdV
flow, J = s , and (69) becomes
t = R (s ) = sss + 32 2 s ,
which is the modified Korteweg-deVries equation, and R is its recursion operator, [48]. On
the other hand, for the grassfire flow, J = 1, and so
t = R (1) = 2 + s Ds1 .
For the curve shortening flow, J = , and so
t = R () = ss + 3 + s Ds1 2 .
Finally, for the thermal grooving flow, J = ss and so
t = R (ss ) = ssss + 2 ss + s Ds1 ss .
As noted above, the induced curvature flow (69) is local if and only if E( J) = 0, where E
is the invariantized Euler operator or variational derivative, [48]. Clearly not all these local
curvature flows will be integrable.
Example 18. As another example, consider the action
(x, u) 7 ( x + u + a, x + u + b), = 1, (71)
of the equi-affine group SA(2) = SL(2) R2 on plane curves C R2 . Applications to
computer vision can be found, for instance, in [4, 10, 55, 60]. According to [17, 23, 31],
the classical equi-affine moving frame arises from the choice of coordinate cross-section
x = u = ux = 0, uxx = 1, uxxx = 0. The fundamental differential invariant is the
equi-affine curvature
uxx uxxxx 35 u2xxx
= (uxxxx ) = . (72)
u8/3
xx
All higher order differential invariants are obtained by invariant differentiation with
respect to the invariant arc length form
uxxx
= (dx) = + , where = ds = u1/3
xx dx, = , (73)
3 u5/3
xx
with dual invariant differential operator D = u1/3

xx Dx being the equi-affine arc length
derivative. Applying our computational algorithm, but suppressing the details, we obtain
dV = A (), dV = B() ,
where
A = D4 + 53 D2 + 53 s D + 13 ss + 49 2 , B= 1
3 D2 29 .
The characteristic operator is
R = A s D1 B = D4 + 53 D2 + 43 s D + 13 ss + 94 2 + 29 s Ds1 . (74)
As in the Euclidean action, both the Eulerian and Hamiltonian operators are invariantly self-
adjoint: A = A and B = B . Therefore, the Euler-Lagrange equation for an equi-affine
invariant Lagrangian L(,
e s , . . .) ds takes the invariant form (48), namely,
D4 + 53 D2 + 53 s D + 31 ss + 94 2 E(L) 1
D2 92 H(L)

e e = 0.
3
R
The equi-affine arc-length functional ds with L
e = 1 has E(L)
e = 0, H(L)
e = 1, and
hence the Euler-Lagrange equation is
A (0) B (1) = 92 = 0.
We conclude that the minimal equi-affine curves are those with zero
R equi-affine curvature
the conic sections. As another example, the variational problem ds has Euler-Lagrange
equation
A (1) B () = 23 ss + 29 2 = 0,
the solution to which, [30], gives as an elliptic function of s.
A general equi-affine invariant curve flow takes the form
Ct = I t + J n, (75)
where t, n are, respectively, the equi-affine tangent and normal directions, [23]. The equi-
affine curve shortening flow, [4, 61], is the normal flow with I = 0, J = 1. Under this flow,
the equi-affine curvature and its derivative evolves according to

= A (1) = 13 ss + 49 2 ,
t (76)
s
= As (1) = D A (1) s B(1) = 13 sss + 10
9 s .
t
230 Peter J. Olver
A second example is the intrinsic (arc-length preserving) flow with J = s . In this case,
the curvature evolution arises from the characteristic operator:
t = R(s ) = 5s + 53 sss + 53 s ss + 59 2 s ,
which is the integrable SawadaKotera equation, [63]. In this case, the characteristic op-
erator R is closely related to, but not the same as the SawadaKotera recursion operator,
which is given by the following formula, [12]:
b = R (D2 + 1 + 1 s D1 ).
R (77)
3 3
Example 19. In the case of space curves C R 3 , under the Euclidean group G = SE(3) =
SO(3) R 3 , there are two generating differential invariants, the curvature and torsion .
According to [31], the relevant moving frame formulae are
dV = A (), dV = A (), dV = B() ,
where = (1 , 2 )T is the column vector containing the order 0 invariant contact forms,
while the characteristic and Hamiltonian operators are:
A = Ds2 + (2 2 ), 2 Ds s ,

ss s s + 23

2 2 3s 2s
A = Ds + D s + ,
2 2 B = , 0 .
1 3 s 2 2 2 s 2 2 s

Ds 2 Ds + Ds + ,
2
Thus, under an intrinsic flow with normal component VN = J n1 + K t2 , the curvature and
torsion evolve via

t R A s
, where R= = D1 B
t R A s
is the recursion operator for the integrable vortex filament flow, with J = s , K = s . This
flow can be mapped to the nonlinear Schrodinger equation via the Hasimoto transformation,
[25, 36].
Acknowledgement
It is a pleasure to thank Evelyne Hubert, Irina Kogan, Liz Mansfield, Gloria Mar Beffa and
Jing Ping Wang for advice and inspiration. Also, thanks to Demeter Krupka and Olga Krup-
kova for their friendship, hospitality and help during my stay in Bratislava and Olomuoc in
August, 2007. This research was supported in part by NSF Grant DMS 0505293.
References
[1] M. J. Ablowitz and P. A. Clarkson, Solitons, Nonlinear Evolution Equations and the
Inverse Scattering Transform (L.M.S. Lecture Notes in Math., vol. 149, Cambridge
University Press, Cambridge, 1991).
[2] I. M. Anderson, Introduction to the variational bicomplex, Contemp. Math. 132 (1992)
5173.
[3] I. M. Anderson, The Variational Bicomplex (Technical Report, Utah State University,
1989).
[4] S. Angenent, G. Sapiro and A. Tannenbaum, On the affine heat equation for non-
convex curves, J. Amer. Math. Soc. 11 (1998) 601634.
[5] M. Antonowicz and A. Sym, New integrable nonlinearities from affine geometry,
Phys. Lett. A 112 (1985) 12.
[6] A. I. Bobenko and P. Schoder, Discrete Willmore flow, In: International Conference
on Computer Graphics and Interactive Techniques ((J. Fujii, Ed.) Assoc. Comput.
Mach., New York, NY, 2005).
[7] M. Born and E. Wolf, Principles of Optics (Fourth Edition, Pergamon Press, New
York, 1970).
[8] R. L. Bryant, A duality theorem for Willmore surfaces, J. Diff. Geom. 20 (1984) 23
53.
[9] P. Broadbridge and P. Tritscher, An integrable fourth-order nonlinear evolution equa-

tion applied to thermal grooving of metal surfaces, IMA J. Appl. Math. 53 (1994)
249265.
[10] E. Calabi, P. J. Olver, C. Shakiban, A. Tannenbaum and S. Haker, Differential and

numerically invariant signature curves applied to object recognition, Int. J. Computer
Vision 26 (1998) 107135.
[11] E. Cartan, La Methode du Repere Mobile, la Theorie des Groupes Continus, et les
Espaces Generalises (Exposes de Geometrie No. 5, Hermann, Paris, 1935).
[12] K.-S. Chou and C. Qu, Integrable equations arising from motions of plane curves,
Physica D 162 (2002) 933.
[13] K.-S. Chou and C.-Z. Qu, Integrable equations arising from motions of plane curves
II, J. Nonlinear Sci. 13 (2003) 487517.
[14] K. Deckelnick, G. Dziuk and C. M. Elliott, Computation of geometric partial differ-

ential equations and mean curvature flow, Acta Numer. 14 (2005) 139232.
[15] P. Dedecker, Calcul des variations et topologie algebrique, Mem. Soc. Roy. Sci. Liege
19 (1957) 1216.
[16] A. Doliwa and P. M. Santini, An elementary geometric characterization of the inte-

grable motions of a curve, Phys. Lett. A 185 (1994) 373384.
[17] M. Fels and P. J. Olver, Moving coframes. II. Regularization and theoretical founda-
tions, Acta Appl. Math. 55 (1999) 127208.
232 Peter J. Olver
[18] M. Gage and R. S. Hamilton, The heat equation shrinking convex plane curves, J. Diff.
Geom. 23 (1986) 6996.
[19] R. E. Goldstein and D. M. Petrich, The KortewegdeVries equation hierarchy as dy-

namics of closed curves in the plane, Phys. Rev. Lett. 67 (1991) 32033206.
[20] M. Grayson, The heat equation shrinks embedded plane curves to round points, J. Diff.
Geom. 26 (1987) 285314.
[21] P. A. Griffiths, On Cartans method of Lie groups and moving frames as applied to
uniqueness and existence questions in differential geometry, Duke Math. J. 41 (1974)
775814.
[22] P. A. Griffiths, Exterior Differential Systems and the Calculus of Variations (Progress
in Math. vol. 25, Birkhauser, Boston, 1983).
[23] H. W. Guggenheimer, Differential Geometry (McGrawHill, New York, 1963).
[24] M. Gurses, Motion of curves on two-dimensional surfaces and soliton equations, Phys.
Lett. A 241 (1998) 329334.
[25] H. Hasimoto, A soliton on a vortex filament, J. Fluid Mech. 51 (1972) 477485.
[26] E. Hubert, Differential invariants of a Lie group action: syzygies on a generating set,
INRIA, 2007.
[27] E. Hubert and P. J. Olver, Differential invariants of conformal and projective surfaces,
SIGMA 3 (2007) 097.
[28] T. Ivey, Integrable geometric evolution equations for curves, Contemp. Math. 285
(2001) 7184.
[29] G. R. Jensen, Higher order contact of submanifolds of homogeneous spaces (Lecture

Notes in Math., No. 610, SpringerVerlag, New York, 1977).
[30] E. Kamke, Differentialgleichungen Losungsmethoden und Losungen (vol. 1, Chelsea,

New York, 1971).
[31] I. A. Kogan and P. J. Olver, Invariant Euler-Lagrange equations and the invariant vari-
ational bicomplex, Acta Appl. Math. 76 (2003) 137193.
[32] D. Krupka, Variational sequences and variational bicomplexes, In: Differential Geom-
etry and Applications (Brno, 1998, (I. Kolar, O. Kowalski, D. Krupka and J. Slovak,
Eds.) Masaryk Univ., Brno, 1999) 525531.
[33] D. Krupka and J. Janyska, Lectures on Differential Invariants (University J. E.

Purkyne, Brno, 1990).
[34] G. L. Lamb, Solitons on moving space curves, J. Math. Phys. 18 (1977) 16541661.
[35] J. Langer, Recursion in curve geometry, New York J. Math. 5 (1999) 2551.
[36] J. Langer and R. Perline, Poisson geometry of the filament equation, J. Nonlin. Sci. 1
(1991) 7193.
[37] J. Langer and D. A. Singer, Lagrangian aspects of the Kirchhoff elastic rod, SIAM Rev.
38 (1996) 605618.
[38] S. Lie, Uber Integralinvarianten und ihre Verwertung fur die Theorie der Differential-
gleichungen, Leipz. Berichte 49 (1897) 369410; also Gesammelte Abhandlungen 6
B.G. Teubner, Leipzig (1927) 664701.
[39] E. L. Mansfield and P. E. van der Kamp, Evolution of curvature invariants and lifting
integrability, J. Geom. Phys. 56 (2006) 12941325.
[40] G. Mar Beffa, The theory of differential invariants and KdV Hamiltonian evolutions,
Bull. Soc. Math. France 127 (1999) 363391.
[41] G. Mar Beffa, Projective-type differential invariants and geometric curve evolutions
of KdV-type in flat homogeneous manifolds, Ann. Institut Fourier, to appear.
[42] G. Mar Beffa, J. A. Sanders and J. P. Wang, Integrable systems in three-dimensional

Riemannian geometry, J. Nonlinear Sci. 12 (2002) 143167.
[43] K. Mikula and D. Sevcovic, Evolution of plane curves driven by a nonlinear function
of curvature and anisotropy, SIAM J. Appl. Math. 61 (2001) 14731501.
[44] K. Nakayama, H. Segur and M. Wadati, Integrability and motion of curves, Phys. Rev.
Lett. 69 (1992) 26032606.
[45] M. Niethammer, A. Tannenbaum and S. Angenent, Dynamic active contours for visual
tracking, IEEE Trans. Auto. Control 51 (2006) 562579.
[46] P. J. Olver, Evolution equations possessing infinitely many symmetries, J. Math. Phys.
18 (1977) 12121215.
[47] P. J. Olver, Symmetry groups and group invariant solutions of partial differential equa-
tions, J. Diff. Geom. 14 (1979) 497542.
[48] P. J. Olver, Applications of Lie Groups to Differential Equations (Second Edition,

Graduate Texts in Mathematics, vol. 107, SpringerVerlag, New York, 1993).
[49] P. J. Olver, Equivalence, Invariants, and Symmetry (Cambridge University Press,

Cambridge, 1995).
[50] P. J. Olver, An introduction to moving frames, In: Geometry, Integrability and Quan-
tization (vol. 5, (I. M. Mladenov and A. C. Hirschfeld, Eds.) Softex, Sofia, Bulgaria,
2004) 6780.
[51] P. J. Olver, A survey of moving frames, In: Computer Algebra and Geometric Algebra
with Applications ((H. Li, P. J. Olver and G. Sommer, Eds.) Lecture Notes in Computer
Science, vol. 3519, SpringerVerlag, New York, 2005) 105138.
234 Peter J. Olver
[52] P. J. Olver, Generating differential invariants, J. Math. Anal. Appl. 333 (2007) 450
471.
[53] P. J. Olver, Invariant submanifold flows, preprint, 2007.
[54] P. J. Olver and J. Pohjanpelto, Moving frames for Lie pseudogroups, Canadian J.
Math., to appear.
[55] P. J. Olver, G. Sapiro and A. Tannenbaum, Differential invariant signatures and flows
in computer vision: a symmetry group approach, In: GeometryDriven Diffusion in
Computer Vision; ((B. M. Ter Haar Romeny, Ed.) Kluwer Acad. Publ., Dordrecht, the
Netherlands, 1994) 255306.
[56] S. J. Osher and J. A. Sethian, Front propagation with curvature dependent speed: Al-
gorithms based on HamiltonJacobi formulations, J. Comp. Phys. 79 (1988) 1249.
[57] C. Qu and S. Zhang, Motion of curves and surfaces in affine geometry, Chaos Solitons
Fractals 20 (2004) 10131019.
[58] H. Rund, The Hamilton-Jacobi Theory in the Calculus of Variations (D. Van Nostrand
Co. Ltd., Princeton, N.J., 1966).
[59] J. A. Sanders and J. P. Wang, Integrable systems in n-dimensional Riemannian geom-

etry, Moscow Math. J. 3 (2003) 13691393.
[60] G. Sapiro, Geometric Partial Differential Equations and Image Analysis (Cambridge
University Press, Cambridge, 2001).
[61] G. Sapiro and A. Tannenbaum, On affine plane curve evolution, J. Func. Anal. 119
(1994) 79120.
[62] G. Sapiro and A. Tannenbaum, On invariant curve evolution and image analysis, Indi-
ana J. Math. 42 (1993) 9851009.
[63] K. Sawada and T. Kotera, A method for finding N -soliton solutions of the K.d.V.
equation and K.d.V.-like equation, Prog. Theor. Physics 51 (1974) 13551367.
[64] J. A. Sethian, Level Set Methods: Evolving Interfaces in Geometry, Fluid Mechanics,
Computer Vision, and Materials Science (Cambridge University Press Cambridge,
1996).
[65] E. L. Starostin and G. H. M. Van der Heijden, The shape of a Mobius strip, Nature
Materials Lett. 6 (2007) 563567.
[66] T. Tsujishita, On variational bicomplexes associated to differential equations, Osaka

J. Math. 19 (1982) 311363.
[67] W. M. Tulczyjew, The Lagrange complex, Bull. Soc. Math. France 105 (1977) 419
431.
[68] A. M. Vinogradov, The Cspectral sequence, Lagrangian formalism and conservation

laws. I. The linear theory, J. Math. Anal. Appl. 100 (1984) 140.
[69] A. M. Vinogradov, The Cspectral sequence, Lagrangian formalism and conservation

laws. II. The nonlinear theory, J. Math. Anal. Appl. 100 (1984) 41129.
[70] R. O. Wells, Jr., Differential Analysis on Complex Manifolds (Graduate Texts in Math-
ematics, vol. 65, SpringerVerlag, New York, 1980).
Chapter 13
D IFFERENTIAL I NVARIANTS OF THE M OTION

G ROUP ACTIONS
Boris Kruglikov1and Valentin Lychagin2
Institute of Mathematics and Statistics, University of Troms
Troms 90-37, Norway
Abstract
Differential invariants of a (pseudo)group action can vary when restricted to in-

variant submanifolds (differential equations). The algebra is still governed by the Lie-
Tresse theorem, but may change a lot. We describe in details the case of the motion
group O(n) Rn acting on the full (unconstraint) jet-space as well as on some invari-
ant equations.
2000 Mathematics Subject Classification. 35N10, 58A20, 58H10

Key words and phrases. differential invariants, invariant differentiations, Tresse deriva-
tives, PDEs.
Introduction
Let G be a pseudogroup acting on a manifold M or a bundle : E M . This action can
be prolonged to the higher jet-spaces J k () (one can also start with an action in some PDE
system E J k () and prolong it).
The natural projection k,k1 : J k () J k1 () maps the orbits in the former space
to the orbits in the latter. If the pseudogroup is of finite type (i.e. a Lie group), this bundle
(restricted to orbits) is occasionally a covering outside the singularity set. Otherwise it
will become a sequence of bundles for k 1. Ranks of these bundles varies but it is
occasionally given by the Hilbert-Poincare polynomial of the pseudogroup action.

E-mail address: kruglikov@math.uit.no

E-mail address: lychagin@math.uit.no
238 Boris Kruglikov and Valentin Lychagin
The orbits can be described via differential invariants, i.e. invariants of the action on
some jet level k. Existence and stability of the above mentioned Hilbert-Poincare polyno-
mial is a consequence of the Lie-Tresse theorem, which claims that the algebra of differ-
ential invariants is finitely generated via the algebraic-functional operations and invariant
derivations.
This theorem in the ascending degree of generality was proved in different sources
[11, 15, 14, 10, 6]. In particular, the latter reference contains the full generality statement,
when the pseudogroup acts on a system of differential equations E J l () (the standard
regularity assumption is imposed, which is an open condition in finite jets).
In the case the pseudogroup G acts on the jet space, E must be invariant and so consist
of the orbits, or equivalently it has an invariant representation E = {J1 = 0, . . . , Jr = 0},
where Jl are (relative) differential invariants. Now the following dichotomy is possible.
If the orbits forming E are regular, the structure of the algebra of differential invariants
on E can be read off from that one of the pure jet-space.
On the other hand if E consists of singular orbits1 (which is often the case when the
system is overdetermined, so that differential syzygy should be calculated, which is an
invariant count of compatibility conditions), then the structure of the algebra of differential
invariants is essentially invisible from the corresponding algebra I of the pure jet-space,
because E is the singular locus for differential invariants I I (if these exist, as was just
remarked1 ).
In this note we demonstrate this effect on the example of motion group G acting natu-
rally on the Euclidean space Rn . The group is finite dimensional, but even in this case the
described effect is visible. For infinite pseudogroups this follow the same route (see, for
instance, the pseudogroup of all local diffeomorphisms acting on the bundle of Riemannian
metrics in [5]).
We lift the action of G to the jets of functions on Rn and describe in details the structure
of algebra of scalar differential invariants in the unconstrained (J Rn ) and constrained
(system of PDEs) cases. This motion group was a classical object of investigations (see
e.g. the foundational work [12]), but we have never seen the complete description of the
differential invariants algebra.
1. Differential Invariants and Lie-Tresse Theorem

We refer to the basics on pseudogroup actions to [10, 7], but recall the relevant theory about
differential invariants (see also [15, 14, 9]). Since well be concerned with a Lie group in
this paper, it will be denoted by one symbol G (in infinite case G should be co-filtered as
the equations in formal theory).
A function I C (J ) (this means that I is a function on a finite jet space J k for
some k > 1) is called a differential invariant if it is constant along the orbits of the lift of
the action of G to J k . For connected groups G we have an equivalent formulation: The
Lie derivative vanishes LX (I) = 0 for all vector fields X from the lifted action of the Lie
algebra.
1
In this case E can be defined via vanishing of an invariant tensor J, with components Ji , though in general
the latter cannot be chosen as scalar differential invariants.
Differential Invariants of the Motion Group Actions 239
Note that often functions I are defined only locally near families of orbits. Alterna-
tively we should allow I to have meromorphic behavior over smooth functions (but well
be writing though about local functions in what follows, which is a kind of micro-locality,
i.e. locality in finite jet-spaces).
The space I = {I} forms an algebra with respect to usual algebraic operations of
linear combinations over R and multiplication and also the composition I1 , . . . , Is 7 I =
F (I1 , . . . , Is ) for any F Cloc (Rs , R), s = 1, 2, . . . any finite number. However even
with these operations the algebra I is usually not locally finitely generated. Indeed, the
subalgebras Ik I of order k differential invariants are finitely generated on non-singular
strata with respect to the above operations, but their injective limit I is not.
To cure this difficulty S.Lie and later his French student A.Tresse introduced invariant
derivatives, i.e. such differentiations that belong to the centralizer of the Lie algebra
g = Lie(G) lifted as the space of vector fields on J (). To be more precise we consider
the derivations C (J ) C (M ) D(M ) (C -vector fields on ), which commute
with the G-action. These operators map differential invariants to differential invariants
: Ik Ik+1 .
We can associate invariant differentiations to a collection of differential invariants
I1 , . . . ,
In (n = dim M ) in general position, meaning dI 1 . . . dI n 6= 0. Moreover the
whole theory discussed above transforms to the action on equations2 E J ().
Namely, given n functionally independent invariants I 1 , . . . , I n we assume their restric-
tions IE1 , . . . , IEn are functionally independent3 (in fact we can have the latter invariants only
without the former), so that they can be considered as local coordinates.
Then one can introduce the horizontal basic forms (coframe) i = dI i . Its dual frame
E
consists of invariant differentiations / I
i = P
[D a (I b )]1 D . The invariant derivative
j
E j E ij
of a differential invariant I are just the coefficients of the decomposition of the horizontal
differential by the coframe:
n
=
X I
dI i
i
I
i=1 E
and they are called Tresse derivatives.

All invariant tensors and operators can be expressed through the given frame and co-
frame and this is the base for the solution of the equivalence problem.
Lie-Tresse theorem claims that the algebra of differential invariants I is finitely gener-
ated with respect to algebraic-functional operations and invariant derivatives.
2. Motion Group Action

Consider the motion group O(n) Rn . It is disconnected and for the purposes of further
study of differential invariants we restrict to the component of unity G = SO(n) Rn . The
two Lie groups have the same Lie algebra g = o(n) Rn and the differential invariants of
the latter become the differential invariants of the second via squaring.
2
At this point we do not need to require even formal integrability of the system E [6], but this as well as
regularity issues will not be discussed here.
3
Here and in what follows one can assume (higher micro-)local treatment.
Since the latter is inevitable even for the group G, the difference between two algebras
of invariants is by an extension via finite group and will be ignored.
Below we will make use of the action of G on the space of codimension m affine
subspaces of Rn :
AGr(m, n) { + c} {(, c) : Gr(n m, n), c }.
The action of G is x 7 Ax + b, x Rn , it is transitive on AGr(m, n) and the stabilizer

equals
St( + c) = {(A, b) G : A = , b (1 A)c + } SO() SO( ) .

n(n + 1)
We have dim G = , dim AGr(m, n) = m(n m + 1) and
2
AGr(k, n) G/(SO(m) SO(n m) Rnm )
(note that this implies AGr(m, n) 6= AGr(n m, n) except for n = 2m contrary to the
space Gr(m, n)).
We can extend the action of G on Rn to the space Rn Rm by letting g G act
g (x, u) = (g x, u).
We can prolong the action to the space J k (n, m).

For k = 1 the action commutes with the natural Gl(m)-action in fibers of the bundle
10 : J 1 (n, m) J 0 (n, m) and the action descends on the projectivization, which can be
identified with the open subset in Rn AGr(k, n) by associating the space Ker(dx f ) to a
(surjective at x if we assume n > m) function f : Rn Rm .
Thus u is indeed an invariant of the G-action (scalar invariants are its components ui ,
so that we can assume the fiber Rm being equipped
P iwithj coordinates), and the scalar differ-
4 i j
ential invariants of order 1 are hu , u i = uxs uxs .
These form the generators of scalar differential invariants of order5 1.
Remark 1. Sophus Lie investigated the vertical actions of G in J 0 (m, n) = Rm Rn and
the invariants of its lift to J (m, n) [12] (actually in this paper for m = 1, n = 3). This
case is easier since the total derivatives D1 , . . . , Dm are obvious invariant derivations.
In what follows we restrict to the case m = 1 and investigate invariants of the G-action
in J (n, 1) = J (Rn ). Partially the results extend to the case of general m, though the
theory of vector-valued symmetric forms S k (Rn ) Rm is more complicated.
3. Differential Invariants: Space J (Rn )

Denote V = T0 Rn . Our affine space Rn (as well as the vector space V ) is equipped with
the Euclidean scalar product h, i and G is the symmetry group of it. In what follows we will
identify the tangent space Tx Rn with V via translations (using the affine structure on Rn ).
Recall that the base space Rn is equipped with the Euclidean metric preserved by G.
4
5
This claim holds at an open dense subset of J 1 (n, m). However if we restrict to the set of singular orbits
with rank(dx u) = r < m, the basic set of invariants will be quite different.
The space J (Rn ), which is the projective limit of the finite-dimensional manifolds
J k (Rn ), has coordinates (xi , u, p ), where = (i1 , . . . , in ) Zn0 is a multiindex with
length || = i1 + + in .
The only scalar differential invariants6 of order 1 are
I0 = u and I1 = |u|2 .
For each x1 J 1 (Rn ) the group G has a large stabilizer. Provided x1 is non-singular
the dimension of the stabilizer St1 is dim G 2n + 1 = 21 (n 1)(n 2).
However the stabilizer completely evolves upon the next prolongation: the action of
G on an open dense subset of J k (Rn ) for any k 2 is free. Note that due to the trivial
connection in J 0 (Rn ) = Rn R we can decompose
J k (Rn ) = Rn R V S 2 V S k V . (1)
Thus we can represent a point xk J k (Rn ) as the base projection x Rn and a sequence
of pure jets Qt = dt u S t V , t = 0, . . . , k.
Covector Q1 can be identified with the vector v = u.
Consider the quadric Q2 S 2 V . Due to the metric we can identify it with a linear
operator A V V , which has spectrum
Sp(A) = {1 , . . . , n }
and the normalized eigenbasis e1 , . . . , en (each element defined up to a sign!), provided Q2
is semi-simple. Since Q2 is symmetric, the basis is orthonormal.
In what follows we assume to work over the open dense subset U J 2 (Rn ), where A is
simple, so that the basis is defined (almost) uniquely (this can be relaxed to semi-simplicity,
but then the stabilizer is non-trivial and the number of scalar invariants drops a bit).
There are precisely (2n 1) = dim J 2 (Rn ) dim St1 differential invariants of order

Pn is2 to take I2,i = i and I2,(i) = hei , vi, i = 1, . . . , n. There is an obvious
2. One choice
relation i=1 I2,(i) = 1, so that we can restrict to the first (n 1) invariants in this group,
but beside this the invariants are functionally independent.
Another choice of invariants is provided by the restriction Q of Q2 to = v , which
has spectrum (again by converting quadric to an operator) Sp(Q ) = {1 , . . . , n1 } and
normalized eigenvectors ei . So the following invariants can be chosen: I2,i = i , I2,n =
Q2 (v, v) and I2,(i) = Q2 (v, ei ).
Both choices have disadvantages of using transcendental functions (solutions to alge-
braic equations), but we can overcome this with the following choice:
I2,i = Tr(Ai ), I2,(i) = hAi v, vi, i = 1, . . . , n.
Here the number of invariants is 2n, but they are dependent7 due to Newton-Girard
P formu-
las, which relate the elementary symmetric polynomials Ek (A) = i1 <<ik i1 ik
and power sums Sk (A) = Tr(Ak ) = ki (these are I2,k ):
P
k
X
kEk (A) = (1)i1 Si (A)Eki (A),
i=1
6
From now on by this we mean the minimal set of generators.
7
The first (2n 1) invariants are however independent and algebraic in the jets.
which together with E0 = 1 gives an infinite chain of formulas
E1 = S1 , 2E2 = S12 S2 , 6E3 = S13 3S1 S2 + 2S3 , . . .
Now with the help of Cayley-Hamilton formula
An = E1 (A)An1 E2 (A)An2 + + (1)n En1 (A)A (1)n En (A)
we can express
I2,(n) = E1 (A)I2,(n1) E2 (A)I2,(n2) + (1)n det A
through our invariants since Ei (A) are functions of I2,i .

Remark 2. We could restrict only to invariants I2,(i) , i = 1, . . . , 2n 1. This is helpful as
we shall see. But when we restrict to singular (from the orbits point of view) PDEs these
differential invariants may turn to be non-optimal, and this will be precisely the case in the
example we investigate.
Now there are precisely n+2 = dim S 3 V differential invariants of order 3, n+3

3 4 =
4 n+k1 k

dim S V differential invariants of order 4, . . . , k = dim S V differential invari-
ants of order k.
The third order invariants are the following:
I3, = Q3 (ei , ej , el ), where = (ijl) S 3 {1, . . . , n}.
Generating invariants of orders 4 and higher are obtained from the similar formulae, namely
as the coefficients q of the decomposition
X
Qk = q , where = i1 ik , 1 i1 ik n.
=(i1 ,...,ik )
They are again transcendental functions. To get algebraic expressions one can use the
third order functions
I3, = Q3 (Ai v, Aj v, Al v), = (ijk) with 1 i j l n
and similar expressions for the higher order.
Theorem 1. The invariants Ii, with i 3 is the base of differential invariants for the Lie
group G action in J (Rn ) via algebraic-functional operations and Tresse derivatives.
This statement is an easy dimensional count8 together with examination of indepen-

dency condition. To get Tresse derivatives n invariants (for instance of order 2) should
be chosen.
However this is not necessary, if one does not care about transcendental functions.
Indeed, the vector fields e1 , . . . , en are invariant differentiations (they can be expressed
through the total derivatives D1 , . . . , Dn with coefficients of the second order).
8
In fact for n 4 the same arguments imply that the base can formed only by the invariants Ii, with i 2.
Remark 3. Notice that the moving frame
e1 , . . . , en C (U, 2 T Rn )
uniquely fixes an element g G, which transforms it to the standard orthonormal frame at

1
0 Rn . This leads to the equivariant map defined on the open dense set ,2 (U ):
J (Rn ) J 2 (Rn ) U G.
Such map is called the moving frame in the approach of Fells and Olver [2].
4. Relations in the Algebra I

Since the commutatorP k of invariant 1differentiations is an invariant differentiation, decompo-
sition [ei , ej ] = cij ek yields 2 n2 (n 1) (in general precisely this number) 3rd order
differential invariants ckij . The number of pure 3rd order invariants obtained via invariant
differentiations of the 2nd order invariants is n(2n 1). So since
n(n + 1)(n + 2) n2 (n 1) n(n + 4)(1 n)

n(2n 1) = 0
6 2 3
we can conclude that differential invariants Ii, with i 2 and invariant differentiations
{ei }ni=1 generate the whole algebra I on an open set U J (Rn ).
Thus we are lead to the question on relations in this algebra. They can be all deduced
from the expressions for pure jets of u
2 , Q4 = Q
Q3 = Q 3 etc
using the structural equations. Here

: C (i S i V ) C (i+1

S i+1 V )
is the symmetric covariant derivative induced by the flat connection in the trivial bundle
J 0 (Rn ) = Rn R, V = T Rn (the map is the composition of the horizontal differential d
and symmetrization).
However for the sake of algebraic formulations we change invariant differentiations ei
to the following ones:
X
v1 =v = v Dx = u i Di
= Av Dx =
X
v2 =Av ui uij Dj
X
v3 =Ad 2 v = A2 v D = ui uij ujk Dk
x
... ... ...

X
\
vn =An1 v = An1 v D =
x ui1 ui1 i2 . . . uin1 in Din .
Now we are going to change the basis of differential invariants in Ik to describe the
relations in the simplest way.
Namely for the basis of invariants of order 2 we can take I2,(ij) = Q2 (Ai v, Aj v),
0 i j < n. However since Q2 (v, w) = hAv, wi and A is self-adjoint we get
I2,(ij) = hAi+1 v, Aj vi = hAi+j+1 v, vi = I2,(i+j+1) ,
so that the new invariants are precisely the old ones I2,(i) , just with the larger index range
i = 1, . . . , 2n 1 (we can allow arbitrary index i, but the corresponding invariants are
expressed via these ones, see Remark 2 and before).
Basic higher order invariants are introduced in the same fashion:
Is,(i1 ...is ) = Qs (Ai1 v, . . . , Ais v), 0 i1 is < n.
Suppose now that our set of generic (regular) points U J 2 (Rn ) is given by not only
the constraint that Sp(A) is simple, but also the claim that the n n matrix kij k0i,j<n
with entries ij = hAi v, Aj vi = I2,(i+j) is non-degenerate. Let
1
1 I2,(1) I2,(n1)
I2,(1) I2,(2) I2,(n)
[ ij ] = .

.. .. ..
..

. . .
I2,(n1) I2,(n) I2,(2n2)
be the inverse matrix. Note that all its entries are invariants. Now
(Ai0 v Dx ) Qs (Ai1 v, . . . , Ais v) = Qs+1 (Ai0 v, Ai1 v, . . . , Ais v)

Xs
+ Qs (Ai1 v, . . . , Aij 1 v, i0 ij , Aij +1 v, . . . , Ais v),
j=1
where i0 ij = Ai0 v (Aij v) is the vector which, due to metric duality, is dual to the covector
i0
P
+=ij 1 Q3 (A v, A v, A ). Thus we obtain
Theorem 2. The algebra I is generated by the invariants Is, and invariants derivatives
v1 , . . . , vn , which are related by the formulae (s 2):
s n1
X X X
vi0 Is,(i1 ...is ) = Is+1,(i0 i1 ...is ) + Is,(i1 ...ij1 ,a,ij+1 ...is ) ab I3,(i0 ,,b+) .
j=1 a,b=0 +=ij 1
In this case we can choose Is, , s 3 and vi as the generators.
This representation for I via generators and relations is not minimal, as clear from the
first part of the section. However the relations are algebraic, explicit and quite simple.
To explain how to achieve minimality let us again change the set of generators (basic
differential invariants). For the second order we return to I2,i , I2,(i) , 1 i n. For the
third order we add the invariants
I3,[ij]l = Tr(Q3 (Ai , Aj , Al v)).

They can indeed be expressed algebraically through the invariants I3,(ijk) together with the
lower order invariants.
For higher order we have more possibilities of inventing new invariants (which can be
described via graphs of the type (k, 1)-tree), but they are again algebraically dependent with
already known differential invariants.
The relations are as follows (0 k < n and we show only top of the list):
v1 I0 = I1 , v2 I0 = I2,(1) , . . . , vn I0 = I2,(n1) ,
v1 I1 = 2I2,(1) , v2 I1 = 2I2,(2) , . . . , vn I1 = 2I2,(n) ,
X X
vk+1 I2,l = I3,[]k , vk+1 I2,(l) = I3,(k) + 2I2,(k+l+1) etc.
+=l1 +=l1
Elaborate work with these shows that all P the invariants can be obtained from I0 and
structural constants cij of the frame [vi , vj ] = ckij vk .
k
Corollary 1. By shrinking U J (Rn ) further (but leaving it open dense) we can arrange
that the algebra I of differential invariants is generated only by I0 and the derivations
v1 , . . . , vn .
5. Algebra of Differential Invariants: Equation E

Consider the PDE E = {kuk = 1}. By the standard arguments it determines a cofil-
tered manifold in J (Rn ) and we identify E with it, so that it consists of the sequence of
prolongations Ek J k (Rn ) and projections k,k1 : Ek Ek1 .
Since the prolongation of the defining equation for E to the second jets is Q2 (v, ) = 0
or v Ker(A) we conclude that most of the invariants, introduced on the previously defined
subset U , vanish: the equation is singular. Indeed, 0 Sp(A), so that det A = 0, the matrix
[ij ] is not invertible etc.
In particular, I2,(i) = 0, Is,(i1 ...is ) = 0 if at least one it 6= 0, v2 = = vn = 0. Thus
the algebra I description from the previous section does not induce any description of the
algebra IE of differential invariants of the group G action on E: the notion of regularity and
basic invariants are changed completely!
Again the group acts freely on the second jets. So there is 1 invariant of order 0
I0 = u,
no invariants of order 1 and (n 1) invariants of order 2:
I2,1 , . . . , I2,n1 or equivalently E1 (A), . . . , En1 (A).
The number of invariants of pure order k > 2 coincides with the ranks of the projections:

1 n+k2
dim k,k1 () = .
k
The principal axes of Q2 (or normalized eigenbasis of A) are now e1 = v, e2 , . . . , en .

These are still the invariant derivations and the invariants of order k > 2 are the coefficients9
of the decomposition by basis in S k Ann(v) S k V :
X
Qk |E = q , q = Qk (vi1 , . . . , vik ).
=(i1 ...ik ):it >1

Theorem 3. The invariants I0 , I2,i and I3, 1 i < n, = (i1 , i2 , i3 ), it 6= 1 form
a base of differential invariants of the algebra IE via algebraic-functional operations and
Tresse derivatives.
Algebra of differential invariants can again be represented in a simpler form via differ-
ential invariants and invariant derivatives. If we choose ei for the latter the relations can be
read off from the algebra I, though this again involves transcendental functions.
Denote the Christoffel symbols of in the basis e by k (these are differential invari-
ij
ants of order 3):
X j
e ej =
X
k j
i ij ek ei = ik k .
Notice that since the connection is torsionless, T = 0, these invariants determine the
structure functions ckij = kij kji .
Let us now substitute the formulas (eigenvalues i can be expressed through the invari-
ants I2,i , however in a transcendental way; 1 = 0 corresponds to e1 )
X X
Q2 = i ( i )2 , Q3 = qijk i j k
1<in 1<ijkn
2 = Q3 :
into the identity Q
i )( i )2 + 2 i
X X X
i ( i )2 = ( i i
X X
= ek (i ) i i k 2 i ijk i j k .
We get for 1 < i j k n:

(i)
X
qijk = ek (i )ij + ei (k )jk ek (i )ik 2 (i) (j) (k)
S3
Since in addition, in general position the invariants i can be expressed through the invari-
ants ei I0 (1 < i n)10 , then by adding decomposition of the covariant derivatives by the
frame into the set of operations, we obtain the following
Corollary 2. By shrinking U E further (but leaving it open dense) we can arrange

that the algebra IE of differential invariants is generated only by I0 and the derivations
e1 , . . . , en .
9
Note that these invariants are defined up to and so should be squared to become genuine invariants;
alternatively certain products/ratios of them define absolute invariants.
10
We have e1 I0 = 1 on E.
6. Algebra of Differential Invariants: Equation E

Completely new picture for the algebra of differential invariants emerges, when we add one
more invariant PDE: the system becomes overdetermined and compatibility conditions (or
differential syzygies) come into the play.
We will study the following system11 , which comes from application to relativity [1]
(when Laplacian is changed to Dalambertian ):
{kuk = 1, u = f (u)} E.
This equation is a non-empty submanifold in J 2 (Rn ), but when we carry the prolongation-
projection scheme, it becomes much smaller.
It turns out that for most functions f (u) the resulting submanifold E is just empty. We
are going to decompose it into the strata
E = 1 (E) n (E),
where i (E) = {x E : #[Sp(AE )] = i} for the operator AE corresponding to the 2-jet

Q2 |E .
It is possible to show that the spectrum of A on E depends on u (and some constants)
only. This was done in [3] via the Cayley-Hamilton theorem, though they used the Dalam-
bertian instead of the Laplace operator. In the next section we prove it for the Laplace
operator via a different approach.
More detailed investigation leads to the following claim:
Conjecture: The strata n (E), . . . , 3 (E) are empty, while 2 (E), 1 (E) are
not and they are finite-dimensional manifolds.
Let us indicate the idea of the proof for the stratum n (E) because on other strata
the eigenbasis ei is not defined (but the arguments can be modified). It turns out that the
compatibility is related to dramatic collapse of the algebra IE of differential invariants.
Indeed, as follows from the discussion above and the next section, there is only one
invariant u of order 2 for the G-action on E. Since the coefficients of the invariant
derivations have the second order, we obtain the following statement:
Theorem 4. All differential invariants of the Lie group G-action on the PDE system E can
be obtained from the function I0 = u and invariant derivations.
Now relations in the algebra IE are differential syzygies for E and they boil down to a
system of ODEs on f (u), which completely determines it.
The details of this program will be however realized elsewhere.
11
This interesting system was communicated to the first author by Elizabeth Mansfield.
7. Geometry of the System

In this section we justify the claim from 6. and prove that the spectrum of the operator
A = AE , obtained from the pure 2-jet Q2 |E via the metric, depends on u only. To do this
we reformulate the problem with nonlinear differential equations in the geometric language
from contact geometry [13].
The first equation E we represent as a level surface H = 12 (1 ni=1 p2i ) = 0 in the
P
jet-space J 1 (Rn ). The second equation from E can be represented as Monge-Ampere type
via n-form
n
X
1 = dx1 . . . dxi1 dpi . . . dxn f (u)dx1 . . . dxn .
i=1
Namely a solution to the system is a Lagrangian submanifold Ln {H = 0} such that

1 |Ln = 0. Representing Ln = graph[j 1 (u)] we obtain the standard description.
The contact Hamiltonian vector field XH preserves the contact structure and being
restricted to the surface
P H =P 0 it coincides with the field of Cauchy characteristic
YH = XH |H=0 = pi Dxi = pi xi + u .
Since Cauchy characteristics are always tangent to any solution, the forms 1+i =
(LXH )i 1 also vanish on any solution of the system E. We simplify them modulo the form
1 and get:
2 = LXH 1 + f (u)1
X
=2 dx1 . . . dpi dxi+1 . . . dpj dxj+1 . . . dxn (f + f 2 ) dx1 . . . dxn ,
3 = LXH 2 + (f (u) + f 2 (u))1
X
= 3! dx1 . . . dpi dxi+1 . . . dpj dxj+1 . . . dpk dxk+1 . . . dxn
(D + f )2 (f ) dx1 . . . dxn ,
... ... ... ...
n = n! dp1 . . . dpn (D + f )n1 (f ) dx1 . . . dxn ,
n+1 = (D + f )n (f ) dx1 . . . dxn ,
where D is the operator of differentiation by u and f is the operator of multiplication by

f (u). Thus a necessary condition for solvability is the following non-linear ODE:
(D + f )n+1 (1) = 0. (2)
This equation can be solved via conjugation D + f = eg Deg with g(u) = f (u) du [4],
R
which reduces the ODE to the form Dn+1 eg = 0, so that g = Log Pn (u), where Pn (u) is a
polynomial of degree n, whence12
n
X 1
f (u) = , i = const . (3)
u i
i=1
12
Here we can assume we are working over C, though this turns out to be inessential.
However there are more compatibility conditions, which produce further constraints on
numbers i . The above relations i = 0 can be used to find Sp(A). Namely let us rewrite
them as follows:
X X
E1 (A) = i = f, E2 (A) = i j = 21 (D + f )2 (1),
i<j
X
E3 (A) = i j k = 1
3! (D + f )3 (1), . . . , En (A) = 1 n = 1
n! (D + 1)n (1).
i<j<k
These, due to Newton-Girard formulas, imply the equivalent identities:

X X
I2,1 = i = f (u), I2,2 = 2i = f (u),
X X
I2,3 = 3i = 12 f (u), I2,4 = 4i = 3!1 f (u), ...
In particular we get i = (u i )1 and so

1 1
A Diag ,..., .
u 1 u n
The fact that det(A) = 0 on E implies that n = and using symmetry u (shift along u)
we can arrange 1 = 0 (we use freedom of renumbering the spectral values).
The conjecture from the previous section is equivalent to the claim that other i equal
either 0 or . But this will be handled in a separate paper.
8. Integrating the System Along Characteristics

Let us now consider the quotient of the submanifold {H = 0} J 1 (Rn ) by the Cauchy
characteristics. We can identify it with the transversal section 2n1 = {H = 0, u =
const}. The solutions will be (n 1)-dimensional manifolds of the induced exterior dif-
ferential system.
pi dxi and
P
Note that we should
P iaugment the system with the contact form = du
its differential 0 = dx dpi . Note that if we choose f (u) to be the solution of the
1
ODE (2), then n! n = dp1 . . . dpn = 0 on solutions.
Let us start investigation from the case n = 2. In this case the induced differential
system is given by two 1-forms:
1
= iXH 1 | = p1 dp2 p2 dp1 (p1 dx2 p2 dx1 )
u
and 0 = iXH 0 | = p1 dp1 + p2 dp2 , but it vanishes on . The form is contact:
d 6= 0, so solutions of E are represented by all Legendrian curves on (3 , ).
1 1
Consider now n = 3. In this case we know that Sp(A) = {0, u , u+ } (in fact,
= 0, but let us pretend we do not know it yet).
We have: f = u22u 2
2
, f + f 2 = u2 2.
Again 0 = iXH 0 vanishes on 5 , so the exteriour differential system is generated by

two 2-forms:
1 = iXH 1 = (p1 dp2 p2 dp1 ) dx3 + (p2 dp3 p3 dp2 ) dx1

2u
+(p3 dp1 p1 dp3 ) dx2 (p dx2
u2 2 1
dx3 + p2 dx3 dx1 + p3 dx1 dx2 );
2 = 21 iXH 2 = p1 dp2 dp3 + p2 dp3 dp1 + p3 dp1 dp2
1
u2 2 (p1 dx2 dx3 + p2 dx3 dx1 + p3 dx1 dx2 ).
The integral surfaces of this system integrate to solutions of E.

Digression. Let us choose another section for J 1 (R3 ): since the Cauchy charac-
p the system {xi = pi , u = 1}, we can take in the domain p3 > 0:
teristics are given by
x3 = const, p3 = 1 p21 p22 . Then the forms giving the differential system are given
by (being multiplied by p3 ):
1 = (1 p22 ) dp1 + p1 p2 dp2 dx2 + dx1 p1 p2 dp1 + (1 p21 ) dp2

u22u
2
(1 p21 p22 ) dx1 dx2 ;
1 p21 p22
2 = dp1 dp2 dx1 dx2 .
u2 2
If we identify J 1 (R2 ) with the contact form = du p1 dx1 p2 dx2 , the above
2-forms become represented by the following Monge-Ampere equations:
(1 u2x )uxx + 2ux uy uxy + (1 u2x )uyy = 2u

u2 2
(1 u2x u2y ),
uxx uyy u2xy = 1
u2 2
(1 2 2
ux uy ).
Compatibility of this pair yields = 0.

Remark 4. The above system is of the kind investigated in [8]: when the surface 2 =
graph{u : R2 R1 } R3 has prescribed Gaussian and mean curvatures, K and H
respectively (this leads to a complicated overdetermined system). In fact the PDEs of the
above system can be written in the form H = F1 (u, u), K = F2 (u, u).
References
[1] C. B. Collins, Complex potential equations. I. A technique for solution, Math. Proc.
Cambridge Philos. Soc. 80 (1) (1976) 165187.
[2] M. Fels and P. Olver, Moving frames and coframes, In: Algebraic methods in physics
(Montreal, 1997, CRM Ser. Meth. Phys., Springer, 2001) 4764.
[3] W. I. Fushchich, R. Z. Zhdanov and I. A. Yegorchenko, On the reduction of the nonlin-

ear multi-dimensional wave equations and compatibility of the DAlembert-Hamilton
system, J. Math. Anal. Appl. 161 (2) (1991) 352360.
[4] M. Kontsevich, private communication.

[5] B. Kruglikov, Invariant characterization of Liouville metrics and polynomial inte-

grals, 2007; arXiv:0709.0423.
[6] B. S. Kruglikov and V. V. Lychagin, Invariants of pseudogroup actions: Homological

methods and Finiteness theorem, Int. J. Geomet. Meth. Mod. Phys. 3 (5 & 6) (2006)
11311165.
[7] B. S. Kruglikov and V. V. Lychagin, Geometry of Differential equations, prepr.

IHES/M/07/04; In: Handbook of Global Analysis ((D. Krupka and D. Saunders, Eds.)
Elsevier, 2008) 725772.
[8] B. S. Kruglikov and V. V. Lychagin, Compatibility, multi-brackets and integrability of

systems of PDEs, prepr. Univ. Troms 2006-49; ArXive: math.DG/0610930.
[9] D. Krupka, J. Janyska, Lectures on Differential Invariants (Folia Facultatis Scien-

tiarum Naturalium Universitatis Purkynianae Brunensis. Mathematica 1, University
J.E. Purkyne, Brno, 1990).
[10] A. Kumpera, Invariants differentiels dun pseudogroupe de Lie. I-II., J. Differential

Geometry 10 (2) (1975) 289345; 10 (3) (1975) 347416.
[11] S. Lie, Ueber Differentialinvarianten, Math. Ann. 24 (4) (1884) 537578.
[12] S. Lie, Zur Invariantenteorie der Gruppe der Bewgungen, Leipzig Ber. 48 (1896) 466
477; Gesam. Abh. Bd. VI 639648.
[13] V. V. Lychagin, Contact geometry and nonlinear second order differential equations,
Uspekhi Mat. Nauk 34 (1) (1979) 137165 (in Russian); English transl.: Russian
Math. Surveys 34 (1979) 149180.
[14] L. V. Ovsiannikov, Group analysis of differential equations (Russian: Nauka, Moscow,

1978; Engl. transl.: Academic Press, New York, 1982).
[15] A. Tresse, Sur les invariants differentiels des groupes continus de transformations,
Acta Math. 18 (1894) 188.
Part III
D IFFERENTIAL E QUATIONS AND

G EOMETRICAL S TRUCTURES
253
Chapter 14
R EMARKS ON THE H ISTORY OF THE N OTION

OF L IE D IFFERENTIATION
Andrzej Trautman
Instytut Fizyki Teoretycznej, Uniwersytet Warszawski
Hoza 69, Warszawa, Poland
1. The derivative X(f ) of a function f , defined on a smooth manifold, in the direction of the
vector field X and the bracket of two vector fields, introduced by Sophus Lie himself, are
the first examples of what is now called the Lie derivative. Another early example comes
from the Killing equation. David Hilbert [1], in his derivation of the Einstein equations,
used the expression
X g g X g X
and stated that it is a tensor field for every tensor field g and vector field X. Around 1920,
Elie Cartan defined a natural differential operator L(X) acting on fields of exterior forms.
He noted that it commutes with the exterior derivative d and gave, in equation (5) on p. 84
in [2], the formula1
L(X) = d i(X) + i(X) d, (1)
where i(X) is the contraction with X.
2. Wadysaw Slebodzinski, in his article of 1931 [5], wrote an explicit formula for the Lie
derivative (without using that name) in the direction of X of a tensor field A of arbitrary
valence. He gave also an equation equivalent to
L(X)(A B) = (L(X)A) B + A L(X)B
and noted that L(X) commutes with contractions over pairs of tensorial indices. He then
applied his results to Hamiltons canonical equations of motion. For a function H(p, q),

E-mail address: andrzej.trautman@fuw.edu.pl
1
In this note, I transcribe all equations from the form given by their authors to the notation in current
usage. All manifolds and maps among them are assumed to be smooth. Good references for my notation and
terminology are [3] and [4].
256 Andrzej Trautman
p = (p ), q = (q ), = 1, . . . , n, Slebodzinski defined the vector field
H H
XH =
,
p q q p
introduced the symplectic form A = dq dp , the Poisson bivector B = /q /p

and showed that L(XH )A = 0 and L(XH )B = 0.
This allowed him to generalize some results of Theophile de Donder in the theory of
invariants [6].
The priority of Slebodzinski in defining the Lie derivative in the general case was recog-
nized by David van Dantzig who wrote, in footnote on p. 536 of [7], Der Operator [the Lie
derivative] wurde zum ersten Mal von W. Slebodzinski eingefuhrt. It was van Dantzig who
introduced, in the same paper, the name Liesche Ableitung. Also Jan Arnoldus Schouten,
in footnote 1 on p. 102 of [8], lists the 1931 paper by Slebodzinski as the first reference for
the notion of Lie differentiation. Van Dantzig complemented the approach of Slebodzinski
by pointing out that the Lie derivative can be defined as the difference between the value of
a geometric object A at a point and the value of that object at the same point obtained by an
infinitesimal dragging along a vector field. In contemporary notation this is expressed by
the formula
d
L(X)A = t A|t=0 , (2)
dt
where t A is the pull-back of A by the flow (t , t R) generated by X. In view of the
equation
d
A = t L(X)A,
dt t
the vanishing of L(X)A is equivalent to the invariance of A with respect to the flow gener-
ated by X; see, e.g., 24 in [9].
3. For quite some time, physicists had been using Lie derivatives, without reference to the
work of mathematicians. Leon Rosenfeld [10] introduced what he called a local variation
A of a geometric object A induced by an infinitesimal transformation of coordinates
generated by X. He noted that commutes with differentiation. It is easily seen that
his A is L(X)A; see, e.g., [11]. Assuming that A is a tensor of type determined by
a representation of GL(4, R) in the vector space RN and denoting by End RN
the matrices of the corresponding representation of the Lie algebra of GL(4, R), one can
deduce from Rosenfelds equations the following formula for the Lie derivative
L(X)A = X A X A.
In particular, assuming that L is a Lagrange function depending on the components of A

and on their first derivatives and such that Ld4 x is an invariant, Rosenfeld showed that
R
L(X)L = (LX )
and used the formula

L L
L(X)L = L(X)A + (L(X)A)
A ( A)
Remarks on the History of the Notion of Lie Differentiation 257
to derive a set of identities of the Noether type, and the conservation laws of energy-
momentum and of angular momentum. One of the main results of that paper was the
symmetrization of the canonical energy-momentum tensor t achieved by adding to it an
expression linear in the derivatives of the spin tensor s.
Incidentally, it is remarkable that this symmetrization, derived independently also by
F. J. Belinfante, is a natural consequence of the EinsteinCartan theory of gravitation. In
that theory, based on a metric tensor g and a linear connection = dx which is
metric, but may have torsion, there are field equations relating curvature and torsion to t
and s, respectively; see [12] and the references given there. If these SciamaKibble field
equations are satisfied and X is a vector field generating a symmetry of space-time so that
L(X)g = 0 and L(X) = 0
then, denoting by t and s the 3-forms (densities) of energy-momentum and spin, and the
covariant derivative with respect to the transposed connection = dx by ,
e one has
the conservation law dj = 0, where
j = X t + 12
e X s .
In the limit of special relativity, if X generates a translation, then j reduces to the cor-
responding component of the density of energy-momentum; for X generating a Lorentz
transformation, one obtains a component of the density of total angular momentum.
4. The Lie derivative defines a homomorphism of the Lie algebra V(M ) of all vector fields
on an n-dimensional manifold M into the Lie algebra of derivations of the algebra of all
tensor fields on M ,
L([X, Y ]) = [L(X), L(Y )].
The Cartan algebra C(M ) = np=0 C p (M ) of all exterior forms on M is Z-graded by the
L
degree p of the forms. A derivation D of degree q Z maps linearly C p (M ) to C p+q (M )

and satisfies the graded Leibniz rule,
D( ) = (D) + (1)pq D for every C p (M ).
Derivations of odd degree are often called antiderivations. The vector space Der C(M ) of
all derivations of C(M ) is a super Lie algebra with respect to the bracket

[D, D ] = D D (1)deg D deg D D D. (3)
The degree of [D, D ] is the sum of the degrees of D and D and there holds a super
Jacobi identity; see [13] for an early review of super Lie algebras, written for physicists.
In particular, d is a derivation of degree +1 and, if X V(M ), then L(X) and i(X) are
derivations of degrees 0 and 1, respectively. The Cartan formula (1) represents L(X) as
a bracket, as defined in (3), of d and i(X).
The contraction i(X) generalizes to fields of vector-valued exterior forms. Let X
V(M ), C p (M ), p = 0, . . . , n, and Y = X , then Y is a vector-valued p-form and
i(Y ) is a derivation of the Cartan algebra, of degree p 1, defined by
i(Y ) = i(X), C(M ).

258 Andrzej Trautman
By linearity one extends i(Y ) to arbitrary vector-valued p-forms. The bracket [d, i(Y )] is
now a derivation of degree p; by the super Jacobi identity its bracket with d is zero and every
derivation (super) commuting with d is of this form. If Y and Z are vector-valued forms of
degrees p and q, respectively, then the bracket [[d, i(Y )], [d, i(Z)]] super commutes with d
and, therefore, there exists a vector-valued (p + q)-form [Y, Z] such that
[d, i([Y, Z])] = [[d, i(Y )], [d, i(Z)]]. (4)
The FrohlicherNijenhuis [14] bracket [Y, Z], defined by (4), generalizes the Lie bracket of
vector fields; it is super anticommutative,
[Z, Y ] = (1)pq [Y, Z],
and makes the vector space of all vector-valued forms into a super Lie algebra. For example,
an almost complex structure J on an even-dimensional manifold is a vector-valued 1-form
and [J, J] is its Nijenhuis torsion.
5. A convenient framework to generalize the definition (2) of Lie derivatives is provided by
natural bundles. A natural bundle is a functor F from the category of manifolds to that of
bundles such that M : F (M ) M is a bundle and if : M N is a diffeomorphism,
then F () : F (M ) F (N ) is an isomorphism of bundles covering . If A is a section of
N : F (N ) N , i.e. a field on N of geometric objects of type F , then A = F (1 )
A is its pull-back by to M . All tensor bundles are natural, but spinor bundles are not.
The vertical bundle V F (M ) is the subbundle of the tangent bundle T F (M ) consisting of
all vertical vectors, i.e. vectors that are annihilated by T M . Let (t , t R) be the flow
generated by X V(M ) and let A be a section of M . The curve t 7 (t A)(x) is vertical
for every x M and the Lie derivative L(X)A is now defined as the section of the vector
bundle V F (M ) M such that (L(X)A)(x) is the vector tangent to t 7 (t A)(x) at
t = 0. The monograph by Kolar, Michor and Slovak [15] contains a full account of this
approach and, in Ch. XI, an even more general definition of Lie differentiation.
References
[1] D. Hilbert, Die Grundlagen der Physik (Erste Mitteilung) (Nachr. Gottingen, 1915)
395407.
[2] E. Cartan, Lecons sur les invariants integraux (based on lectures given in 1920-21 in
Paris, Hermann, Paris 1922, reprinted in 1958).
[3] Y. Choquet-Bruhat, C. DeWitt-Morette and M. Dillard-Bleick, Analysis, Manifolds

and Physics (2nd ed., North-Holland, Amsterdam 1982).
[4] I. Agricola and Th. Friedrich, Global analysis: Differential Forms in Analysis, Geom-
etry and Physics (transl. from the 2001 German edition, Graduate Studies in Mathe-
matics, vol. 52, American Mathematical Society, Providence, RI, 2002).
[5] W. Slebodzinski, Sur les equations de Hamilton, Bull. Acad. Roy. de Belg. 17 (1931)
864870.
Remarks on the History of the Notion of Lie Differentiation 259
[6] Th. de Donder, Theorie des invariants integraux (GauthierVillars, Paris 1927).
[7] D. van Dantzig, Zur allgemeinen projektiven Differentialgeometrie, Proc. Roy. Acad.
Amsterdam 35 (1932) Part I: 524534; Part II: 535542.
[8] J. A. Schouten, Ricci-Calculus (2nd ed., Springer-Verlag, Berlin 1954).
[9] A. Lichnerowicz, Geometrie des groupes de transformations (Dunod, Paris 1958).
[10] L. Rosenfeld, Sur le tenseur dimpulsion-energie, Memoires Acad. Roy. Belg., Classe
des Sciences 18 Fasc. 6 (1940) 130.
[11] A. Trautman, Sur les lois de conservation dans les espaces de Riemann, In: Les
Theories Relativistes de la Gravitation (Royaumont 1959, Ed. du CNRS, Paris 1962)
113116.
[12] A. Trautman, EinsteinCartan theory, In: Encycl Math. Phys. ((J.-P. Francoise, G. L.
Naber and S. T. Tsou, Eds.) Elsevier, Oxford, vol. 2, 2006) 189195.
[13] L. Corwin, S. Sternberg and Y. Neeman, Graded Lie algebras in mathematics and
physics, Rev. Mod. Phys. 47 (1975) 573603.
[14] A. Frohlicher and A. Nijenhuis, Theory of vector-valued differential forms Part I,

Indag. Math. 18 (1956) 338359.

(Springer-Verlag, Berlin, 1993).
Chapter 15
S ECOND -O RDER D IFFERENTIAL E QUATION F IELDS

WITH S YMMETRY
M. Crampin1 and T. Mestdag2

1
Department of Mathematical Physics and Astronomy
Ghent University, Krijgslaan 281,
B-9000 Ghent, Belgium
2
Department of Mathematics, University of Michigan
530 Church Street, Ann Arbor, MI 48109, USA
Abstract
We examine the reduction of a system of second-order ordinary differential equa-
tions which is invariant under the action of a symmetry group. We describe the reduced
system, and show how the integral curves of the original system can be reconstructed
from the reduced dynamics. We then specialize to invariant Lagrangian systems. We
compare and contrast two approaches to reduction in this case. The first leads to the
so-called Lagrange-Poincare equations. The second involves an extension of Rouths
reduction procedure to an arbitrary Lagrangian system (that is, one whose Lagrangian
is not necessarily the difference of kinetic and potential energies) with a symmetry
group which is not necessarily Abelian. Throughout we use a new method of analysis
based on adapted frames and associated quasi-velocities.
2000 Mathematics Subject Classification. 34A26, 37J15, 53C05, 70H03

Key words and phrases. Dynamical system, Lagrangian system, Lagrange-Poincare equa-
tions, symmetry, Routhian, reduction, reconstruction.
1. Introduction
The concept of symmetry plays an important role in a great number of applications in
dynamics. Symmetry properties of dynamical systems have been studied intensively in

For Demo Krupka on his 65th birthday

E-mail address: crampin@btinternet.com

E-mail address: Tom.Mestdag@UGent.be
262 M. Crampin and T. Mestdag
recent years: see for example the survey in the recent monograph [6] by Marsden et al.,
as well as the more long-established reference [7]. Perhaps the most important aspect of
symmetry is its use in reduction. When a dynamical system has a Lie group of symmetries,
which is to say that considered as a vector field on some manifold it is invariant under the
action of the group on the manifold, then the corresponding equations of motion can be
reduced to a new set of equations with fewer unknowns. The working assumption is that
the reduced system will be simpler to deal with than the original one.
The bulk of the literature on symmetry in dynamics concentrates on the Hamiltonian
description of dynamical systems with symmetry, in which the theory of Poisson manifolds
plays the main role. Less well-known is symmetry reduction for Lagrangian systems. It is
the latter that is at the core of the present paper.
There are in fact accounts of several different Lagrangian reduction theories to be found
in the literature. For example, one distinctive reduction method applies when the configu-
ration space is itself a Lie group; it is called Euler-Poincare reduction. The particular issue
that we will be concerned with, however, is the following. In rough terminology, the in-
variance of the Lagrangian leads via the Noether theorem to a set of conserved quantities
(the components of momentum). There are two alternative broad types of Lagrangian re-
duction theory, which differ in whether or not the existence of these conserved quantities
is explicitly taken into account in the reduction process. The more direct approach, which
effectively ignores conservation laws, is called Lagrange-Poincare reduction (and includes
Euler-Poincare reduction as a special case). Taking account of momentum conservation
leads to Rouths procedure. For more details and some comments on the history of these
reduction theories, see e.g. [6] and [8].
One main purpose of our paper is to compare and contrast Lagrange-Poincare reduction
and Rouths procedure. Over the last couple of years we have been developing our own
techniques for analysing symmetry and reduction of dynamical systems. These techniques,
while being well adapted to the discussion of Lagrangian systems, are not restricted to
them; in this respect they are different from the techniques usually found in the literature.
In fact our techniques are designed to apply to any dynamical system, that is, any vector
field, which is invariant under a Lie group.
The basic ideas, which we exploit throughout the paper, are most succinctly explained
in the simplest context, that of a first-order dynamical system or plain and unadorned vector
field. We discuss this case in the following section. The transition to Lagrangian sys-
tems, that is, to dynamical systems of Euler-Lagrange type, is made via the consideration
of second-order systems. By a second-order system we mean a second-order differential
equation field, that is to say, a vector field on a tangent bundle belonging to that special
class whose integral curves satisfy a system of second-order differential equations. We de-
scribe the general second-order theory in Section 3. Those particular second-order systems
defined by Lagrangians are discussed in Section 4, where the two approaches, Lagrange-
Poincare and Routh, are explained. We end with an example of a second-order system with
symmetry, reduced by all three methods.
This paper is in effect a survey and summary of work which has been presented in
greater detail in a number of other articles; we draw the readers attention in particular to
[4], [5] and [9].
Throughout the paper, symmetry groups are supposed to act as follows. We have a
Second-Order Differential Equation Fields with Symmetry 263
differentiable manifold M and a connected Lie group G which acts freely and properly
to the left on M , so that M is a principal G-bundle. We denote the base by M/G, the
projection by M : M M/G, and the action by (x, g) 7 gM x. The Lie algebra of G is
denoted by g, and for g, is the fundamental vector field corresponding to (the vector
M
field whose flow is t 7 exp(t) ).
2. The First-Order Case

Suppose we have a first-order dynamical system, represented by a vector field Z on the man-
ifold M , which is invariant under a symmetry group G acting on M as described above. We
can of course express the condition of invariance in terms of the action M ; alternatively,
Z] = 0 for all g, and conversely when this
we can note that if Z is invariant then [,
differential condition holds and G is connected (as we assume) then Z is invariant. We will
find it convenient always to use the differential version of the condition for invariance.
The invariance of Z implies that it passes to the quotient. That is to say, there is a
well-defined vector field on M/G, say Z, which is M -related to Z: Z = M Z. We call Z
the reduced dynamical system. Two questions arise:
reduction: how to describe the reduced system Z explicitly and conveniently;
reconstruction: how, given an integral curve of Z, to obtain an integral curve of Z in a

systematic fashion.
2.1. Reduction
In order to formulate a simple description of the reduced dynamics we introduce and work
with a local frame {Ea , Xi }, where the Ea , a = 1, 2, . . . , dim G, are tangent to the fibres
of M : M M/G, the Xi , i = 1, 2, . . . , dim(M/G) are transverse to the fibres, and all
of the members of the basis are G-invariant.
We define the Ea as follows. Let {ea } be a basis for g, ea the corresponding funda-
c e , where the C c are the structure constants of
mental vector fields. Then [ea , eb ] = Cab c ab
g with respect to the given basis. It is clear that in general the ea are not invariant. We set
Ea = Aba eb , and enquire under what conditions on the coefficients Aba the Ea are invariant.
We have
[ea , Eb ] = ea (Acb ) Cad
c
Adb ec ,
so the Ea are invariant if and only if
ea (Acb ) = Cad
c
Adb .
The integrability conditions for these equations are satisfied by virtue of the Jacobi identity.
There are therefore local solutions, for which the matrix A = (Aba ) is non-singular, and for
which A is the identity matrix on some specified local section of M . Such a local section
determines a local trivialization M GM/G of M ; identifying the fibres with G, we see
that each Ea corresponds to a left-invariant vector field on G. Each ea , on the other hand,
corresponds to a right-invariant vector field on G (which explains the sign in the expression
for the bracket); A(g) is the matrix of ad(g) with respect to the basis {ea }. In the literature,
{ea } is sometimes referred to as the moving frame, {Ea } as the body-fixed frame (see for
example [1]).
To define the part of the frame transverse to the fibres, we assume that we have at
our disposal a principal connection on the principal G-bundle M ; we take Xi to be the
horizontal lift with respect to the connection of a member of some local basis of vector
fields on M/G. In particular, we may (and generally will) take this to be a coordinate basis.
We may now write Z = Z a Ea + Z i Xi . Since Z, Ea and Xi are all invariant, so also
are the coefficients Z a and Z i . We may therefore regard the Z i (in particular) as functions
on M/G, and we have

M Z = Z = Z i i ,
x
where the xi are coordinates on M/G. The reduced equations are simply
xi = Z i (x).
2.2. Reconstruction
Suppose given an integral curve of Z, say t 7 z(t) (a curve in M/G). We have to find an
integral curve of Z, t 7 z(t) (a curve in M ), over z (so that M z = z).
We proceed as follows. Take any lift of z to M , t 7 (t) (so that M = z). Then
there is a curve t 7 g(t) G such that z(t) = g(t) M (t). The next questions therefore are
how to lift z(t) to (t) in a systematic fashion, and having done so, how to find g(t).
Assume as before that we have a principal connection on M , with connection form (a
g-valued 1-form on M ). Then we can take (t) to be a horizontal lift of z(t). We can now
derive a differential equation for g(t). First, differentiate the equation for z(t):

z(t) = M
g(t) ^
(t) + (g(t))|(t) ,
where is the Maurer-Cartan form of G (i.e. g 1 g for a matrix group). We want z(t) to be
an integral curve of Z, so
M
z(t) = Zz(t) = g(t) Z(t)
by invariance. Thus
^
Z(t) = (t) + (g(t))|(t) .
This formula expresses Z(t) in terms of its horizontal and vertical components. We pick
out the vertical component, or in other words apply :
(g(t)) = (t) (Z).
The right-hand side is a curve in g, so this is an equation in g, and it has a unique solution
for g(t) with g(0) = id. (This is evident for a matrix group, for which the equation is
g = g (Z).)
M (t) is an integral curve of Z. It is the integral curve through (0): to
Then z(t) = g(t)
find the integral curve over z through some other point in the fibre over z(0), say gM (0),
we merely have to left translate z(t), that is, take gM z(t).
3. The Second-Order Case

A second-order dynamical system determines and is determined by a vector field on a
tangent bundle T M , which has the form

= u
+ i (x, u)
x u
when expressed in terms of natural coordinates (x , u ); such a vector field is called a
second-order differential equation field.
In order to consider symmetries of a second-order differential equation field we must
extend the group action from M to T M . Suppose G acts on M as before; then the induced
action of G on T M is given by gT M (x, u) = (gM x, g M u). (Transformations of T M of
this form are sometimes called point transformations.) The fundamental vector fields of
the induced action are the complete lifts of the fundamental vector fields of the action on
M , which we denote by C . Moreover, T M is a principal G-bundle, and we denote by
T M : T M T M/G the projection (which is not to be confused with the projection
T M M , which we denote by .)
We assume now that the second-order differential equation field is invariant under the
induced action of G:
[C , ] = 0 for all g.
3.1. Reduction
We will make extensive use of the complete and vertical lifts of vector fields on M to T M :
we denote the vertical lift of a vector field X on M by X V (and its complete lift by X C as
above). We recall the following formulae for the brackets of such lifts:
[X C , Y C ] = [X, Y ]C , [X V , Y C ] = [X, Y ]V , [X V , Y V ] = 0.
From these formulae it is clear that the complete and vertical lifts of a G-invariant vector
field on M are both invariant under the induced action of G on T M . So if we take an
invariant local basis {Ea , Xi } on M as before, then {EaC , XiC , EaV , XiV } is an invariant
local basis of vector fields on T M .
We now introduce new fibre coordinates with respect to , adapted to the invariant basis,
which we call quasi-velocities. For any vector field basis {Z } on M we denote by v the
components of u Tx M with respect to {Z |x }: so u = v Z |x . Considered as functions
on T M the v are fibre coordinates; these are the quasi-velocities corresponding to the
basis {Z }. Alternatively, let { } be the 1-form basis dual to {Z }; each defines a
fibre-linear function on T M , ; then v = . We denote by (v a , v i ) the quasi-velocities
corresponding to {Ea , Xi }.
We need expressions for the derivatives of the quasi-velocities with respect to the mem-
bers of the invariant basis {EaC , XiC , EaV , XiV }. To find them, the following two fomulae are
indispensible:
Z C () = Ld Z , Z V () = (Z).
For example, we have eCa (v i ) = L[ i
ea = 0 (since the basis dual to an invariant basis is also
invariant).
We also need expressions for the pairwise brackets of {Ea , Xi }: we have [Ea , Eb ] =
c E ,
Cab and we set
c
a
[Xi , Xj ] = Kij Ea , [Xi , Ea ] = Xi (Aba )eb = bia Eb .
It is worth noting that since the vector fields Xi and Ea are G-invariant, so are their brackets,
and so are the coefficients Kij a and b .
ia
i
Let (x ) be coordinates on M/G, as before. Then we find that
eCa (xi ) = 0, eCa (v i ) = 0, eCa (v b ) = 0,

EaC (xi ) = 0, EaC (v i ) = 0, EaC (v b ) = bia v i + Cac
b vc,
EaV (xi ) = 0, EaV (v i ) = 0, b

Ea (v ) = a ,
V b
XiC (xj ) = ij , XiC (v j ) = 0, XiC (v a ) = Kij a v j a v b ,

ib
XiV (xj ) = 0, XiV (v j ) = ij , a
Xi (v ) = 0.
V
From the first line, (xi , v i , v a ) define coordinates on T M/G. The invariant vector fields
of the basis project onto T M/G, and we can read off the coordinate expressions for their
projections from the formulae above:

T M EaC = bia v i + Cac
b c
v , T M EaV = ,
v b v a

T M XiC = K a j
ij v + a b
ib v , T M XiV = .
xi v b v i
Since is a second-order differential equation field,
= v a EaC + v i XiC + a EaV + i XiV .
Each term is invariant, so a and i define functions on T M/G. We have

T M = = v a (bia v i + Cac
b c
v ) + vi i
v b x

v i Kij
a j
v + aib v b b
+ a a + i i
v v v
i i a
= v + + .
xi v i v a
The reduced equations are xi = v i , v i = i (xj , v j , v b ), v a = a (xj , v j , v b ), or
xi = i (xj , xj , v b ), v a = a (xj , xj , v b );
they are of mixed first- and second-order type.

So far as we are aware, the study of the reduction of general second-order dynamical
systems with symmetry by methods similar to ours has been attempted by other authors
only for single symmetries (that is, 1-dimensional symmetry groups), in [2]. For a more
detailed account of our approach, see [4].
3.2. Reconstruction
In order to carry out reconstruction using the method described in Section 2 we need a
principal connection on the bundle T M : T M T M/G. We have already assumed
that we have at our disposal a principal connection on M : M M/G. There is in
fact a simple method of lifting such a connection to one on T M . The initial connection
is specified by its connection form . We show that the pull-back of to T M by
the tangent bundle projection is the connection form of a principal connection on the
principal G-bundle T M . Clearly, is a g-valued 1-form on T M . The action of G
on T M is -related to the action on M . Likewise, for any g the fundamental vector
field C corresponding to the action on T M is -related to ,
the fundamental vector field
corresponding to the action on M . Thus
(C ) = ( C ) = ()
= ,
while
gT M = gM = ad(g 1 ) ,
as required. The connection defined by is called the vertical lift of the original connec-
tion, and its connection 1-form is denoted by V .
When we use V in the reconstruction process, the right-hand side of the reconstruction
equation is V (). The special natures of (that it is a second-order differential equation
field) and V (that it is a vertical lift connection) now come into play. For at any point
u T M , uV () = (u) ( u ) = (u) (u); that is to say, uV () is just the vertical
part of u (considered as an element of g), and in particular is the same for all G-invariant
second-order differential equation fields on T M .
4. Lagrangian Systems
We now suppose that we are dealing with a second-order dynamical system defined by
a regular Lagrangian L on T M . Thus is the Euler-Lagrange field of L, and satisfies the
Euler-Lagrange equations, which in terms of coordinates (x , u ) can be written
L L

= 0.
u x
We assume that L is regular, which is to say that its Hessian with respect to the fibre coor-
dinates, the symmetric matrix with entries
2L
,
u u
is non-singular. Then is uniquely determined by the Euler-Lagrange equations (and the
fact that it is a second-order differential equation field). In order to use the methods de-
scribed in the previous sections we have to express the Euler-Lagrange equations in terms
of a vector field basis on M which is not of coordinate type. With respect to the basis {Z }
they take the form
(ZV (L)) ZC (L) = 0.
Assume that the regular Lagrangian L is G-invariant: C (L) = 0. Then the Euler-
Lagrange field is also G-invariant, as one would expect. We wish to carry out a reduction,
and to express the reduced equations in terms of an appropriate reduced version of the
Lagrangian. As we mentioned in the Introduction, there are in fact two different ways of
proceeding.
In the first, which is called Lagrange-Poincare reduction, we work with the invariant
basis {Ea , Xi }, as before. The Euler-Lagrange equations become
(XiV (L)) XiC (L) = 0, (EaV (L)) EaC (L) = 0,
and the reduced equations determine a vector field on T M/G.

The second approach could be characterized as making more direct use of the particular
properties of the Euler-Lagrange formalism. This time we use a mixed basis {ea , Xi }
(mixed in the sense that only part of it is invariant); since eCa (L) = 0, the Euler-Lagrange
equations are
(XiV (L)) XiC (L) = 0, (eVa (L)) = 0.
Thus the momentum, whose components are eVa (L), is conserved. The first step in the re-
duction process in this case (if reduction is the right word) just consists in restriction to
a level set of momentum. The process as a whole is called Rouths procedure; it general-
izes the elimination of the momentum conjugate to a cyclic coordinate which was Rouths
original version of the procedure [11].
4.1. Lagrange-Poincare Reduction

Since it is invariant, L defines a function L on T M/G. The Euler-Lagrange equations
reduce directly to
(XiV (L)) XiC (L) = 0, (EaV (L)) EaC (L) = 0
where XiC = T M XiC etc. Using the formulae from the previous section we obtain
!
L L a j L
= (Kij v + aib v b ) a
v i x i v
!
L L
= (bia v i + Cac
b c
v ) ;
v a v b
and as before,

= v i + i i + a a .
xi v v
These are the Lagrange-Poincare equations [3], though they are usually written with d/dt
in place of ; see also [9].
4.2. Rouths Procedure

We set pa = eCa (L). Considered as a vector, (pa ) takes its values in g , the dual of the Lie
algebra: it is the (generalized) momentum. Since (pa ) = 0, the vector field is tangent
to the level sets of momentum; we will concentrate on its restriction to one level set, say
N : pa = a .
We work now with the mixed basis {ea , Xi }. The quasi-velocities are (v a , v i ), where
v a = Aab v b ; the v a are not invariant. The pairwise brackets of elements of the basis are
a a
[ea , Xi ] = 0, [Xi , Xj ] = Rij ea , Rij = Aba Kij
b
.
a as the components of curvature of the con-
(The expression for [Xi , Xj ] identifies the Rij
nection on M , regarded as a g-valued 2-form on M/G.)
The derivatives of the quasi-velocities are
XiC (v j ) = 0, XiC (v a ) = Rij a vj ,
XiV (v j ) = ij , XiV (v a ) = 0,
eCa (v i ) = 0, eCa (v b ) = Cac
b v c ,
eVa (v i ) = 0, b
ea (v ) = a .
V b
Set gab = eVa (pb ) = eVa (eVb (L)). Since vertical lifts commute, gba = gab . We assume
that the symmetric matrix (gab ) is everywhere non-singular. Then eVa is transverse to N ,
and in principle we can solve the equations pa = a for v a . Thus restricting to a level set
of momentum is a form of reduction, in the sense that by doing so we reduce the number of
variables, and presumably thereby the difficulty of the problem. It is however a somewhat
different form of reduction from those discussed so far: reduction by restriction rather than
projection.
The gab are in fact components of the Hessian of L. The Hessian of L can be defined in
a coordinate-independent way as the symmetric covariant 2-tensor g along given by
g(u, v) = uV (v V (L)),
for any vectors u, v at the same point of M . We have gab = g(ea , eb ).

We may use gab to define vector fields tangent to N . Denote by (g ab ) the matrix inverse
to (gab ). For any vector field Y on T M , the vector field Y g ab Y (pa )eVb annihilates pa
and is therefore tangent to each level set of momentum. In particular, set
XiC = XiC g ab XiC (pb )eVa = XiC Pia eVa

XiV = XiV g ab XiV (pb )eVa = XiV Qai eVa .
Then XiC , XiV are tangent to N . We have
XiC (v a ) = Rij
a j
v Pia , XiV (v a ) = Qai .
We now derive some Euler-Lagrange-like equations which determine the restriction of

to the level set of momentum N . These equations involve, not the Lagrangian itself, but
a modification of it called the Routhian, which is given by R = L pa v a . Now
a j
XiC (L) = XiC (L) + Pia pa = XiC (L) pa (XiC (v a ) + Rij v )
a j
= XiC (R) pa Rij v ;
XiV (L) = XiV (L) + Qai pa = XiV (L) pa XiV (v a )
= XiV (R).
But (XiV (L)) XiC (L) = 0, and is tangent to N . So if R is the restriction of the
Routhian to N we have
(XiV (R )) XiC (R ) = a Rij

a j
v .
These are the required equations; we call them the generalized Routh equations.
The generalized Routh equations may appear to be straightforwardly second-order dif-
ferential equations, unlike the other reduced equations for second-order differential equa-
tion fields, which are mixed first- and second-order equations. This appearance is deceptive.
In the first place, the generalized Routh equations (when expressed explicitly as differential
equations) are equations on N , not T M/G as is the case for the other reduced equations.
Now N can be locally identified with M M/G T (M/G). For local coordinates on N
we may take (xi , a , v i ), where (a ) are fibre coordinates on M , so that (xi , a ) are coor-
dinates on M and (xi , v i ) coordinates on T (M/G). The quasi-coordinates (v i , v a ) on T M
are linear combinations of xi and a , and in fact v i = xi . So we can express v a in terms
of xi and a . On T M the resulting expression is an identity; but on restricting to N , the
equations pa = a , when expressed in this way in terms of xi and a , become additional im-
plicit first-order differential equations, which we may regard as equations for the a (since
the equations v i = xi are already subsumed in the representation of the generalized Routh
equations as second-order equations).
The level set N is not in general G-invariant: C is not in general tangent to N . In
fact
C (pa ) = b eCb (pa ) = b [eCb , eVa ](L) = b Cab
c
pc .
Thus C will be tangent to N if and only if b Cab
c = 0. The set of g which satisfy
c
this condition forms a subalgebra g of g. It is in fact the algebra of G , the isotropy group
of g under the coadjoint action of G in g . Now |N , R , and the generalized
Routh equations, are all invariant by G . We can therefore carry out a further reduction, by
G , in the manner described earlier, to obtain a reduced system on N /G . The resulting
reduced equations have been called the Lagrange-Routh equations [8]. We do not give
the derivation here, but refer the reader to [5], as well as [8], for the details. In fact [8]
contains an extensive discussion of the background to Rouths procedure and its modern
generalization. The methods used in this paper are quite different from ours, however, and
it deals only with so-called simple mechanical systems. For a more detailed account of all
aspects of our approach see [5].
4.3. Reconstruction
The same method of reconstruction as was described for second-order differential equations
in the previous section, namely using the vertical lift connection, can be used for Lagrange-
Poincare reduction. For Rouths procedure it is necessary to carry out reconstruction only
for the final stage of reduction by G : an integral curve of the restriction of to N is, after
all, an integral curve of . It is not so obvious how to adapt the vertical lift connection to this
situation, though it can be done. We will now describe an alternative way of constructing a
connection, which is based more closely on the fact that we are dealing with a Lagrangian
system, and applies more-or-less directly to both reconstruction problems.
We consider first the case of a simple mechanical system, which is one in which the
Lagrangian takes the simple form L = T V where T is a kinetic energy function derived
from a Riemannian metric g, and V a function on M defining the potential energy. The
symmetry group G consists of all isometries of the metric g leaving V invariant. Then the
distribution on M consisting of all vectors orthogonal (with respect to g) to the fibres of
M : M M/G is G-invariant, and defines a principal connection (of which it is the
horizontal distribution). This is the so-called mechanical connection on M . It can be lifted
to a principal connection on T M : T M T M/G, as before. For the vertical lift of the
= 0 for all g.
mechanical connection, v Tu T M is horizontal just when g (u) ( v, )
This connection can be used for reconstruction in the Lagrange-Poincare case. For Rouths
procedure we define the required connection by saying that v Tu N is horizontal just
= 0 for all g .
when g (u) ( v, )
In general, we can use the Hessian of L in place of the Riemannian metric. This doesnt
give a connection on M , but does give connections on T M T M/G and N N /G .
Indeed, since we have (wittingly) used the same symbol, g, for both the metric in the case of
a simple mechanical system and the Hessian in general, the definitions are almost identical:
the only difference is that in general g is not projectable. For the Lagrange-Poincare case,
we say that v Tu T M is horizontal just when gu ( v, ) = 0 for all g. For the
= 0 for all g .
Routhian case, we say that v Tu N is horizontal just when gu ( v, )
Both of these specifications define principal connections, which we call collectively the
generalized mechanical connection. A fuller account of this construction can be found
in [9].
5. An Example: Wongs Equations

In this final section we determine the reduced equations for an interesting second-order
differential equation field, namely the geodesic field for a Riemannian manifold on which a
group G acts freely and properly to the left as isometries. We make the further stipulation
that the vertical part of the metric (that is, its restriction to the fibres of M : M M/G)
comes from a bi-invariant metric on G. The reduced equations in such a case are known
as Wongs equations [3, 10]. We will derive the reduced equations by each of the three
methods discussed above.
We will denote the metric by g. The fact that the symmetry group acts as isometries
means that the fundamental vector fields are Killing fields: L g = 0. It follows that the
components of g with respect to the members of an invariant basis {Ea , Xi } are themselves
invariant. We have a small notational problem to deal with here: we will need to distinguish
between the components of g with respect to the fundamental vector fields ea and those
with respect to the Ea . We will set g(ea , eb ) = gab , as before. For g(Ea , Eb ) we will write
hab . We set g(Xi , Xj ) = gij . We will use the mechanical connection, which means that
g(ea , Xi ) = 0. Since both hab and gij are G-invariant functions, they pass to the quotient;
in particular, the gij are the components with respect to the coordinate fields of a metric on
M/G, the reduced metric.
The further assumption about the vertical part of the metric has the following implica-
tions. It means in the first place that LEc g(Ea , Eb ) = 0 (as well as Lec g(Ea , Eb ) = 0),
and secondly that the hab must be independent of the coordinates xi on M/G, which is to
say that they must be constants. From the first condition, taking into account the bracket
c E , we easily find that the h must satisfy h C d + h C d = 0.
relations [Ea , Eb ] = Cab c ab ad bc bd ac
It is implicit in our choice of an invariant basis that we are working in a local trivialization
of M M/G. Then ea , Ea and Aba are all objects defined on the G factor, and so are
independent of the xi . We may write

Xi = ia Ea
xi
for some coefficients ia which are clearly G-invariant; moreover
[Xi , Ea ] = ic Cac
b
Eb = bia Eb .
b , and therefore h c + h c = 0.
Thus bia = ic Cac ac ib bc ib
The second-order differential equation field of interest is the geodesic field of the
Riemannian metric g. To find the reduced equations by the direct method we have to express
in terms of the invariant basis, and for this purpose we need the connection coefficients
of the Levi-Civita connection in terms of this basis. Using the data above in the standard
Koszul formulae for the Levi-Civita connection coefficients of g with respect to the basis
i , which are just the connection
{Ea , Xi } we find that the only non-zero coefficients are jk
coefficients of the Levi-Civita connection of the reduced metric gij , and
a a a
bc = 21 Cbc , jb = ajb , a
jk a
= 12 Kjk , jbi = 21 g ik hbc Kjk
c
= bji .
It follows that
= v i XiC + v a EaC

i j k
jk v v + (jbi + bji )v j v b + bci v b v c XiV

a j k a a j b a b c
jk v v + (jb + bj )v v + bc v v EaV

= v i XiC + v a EaC jk
i j k
v v g ik hbc Kjk
c j b
v v XiV ajb v j v b EaV .
The reduced vector field on T M/G is therefore

= v i i
i j k
jk v v g ik
hbc K c j b
jk v v i
ajb v j v b a
x v v
and the reduced equations are
xi + jk
i j k
x x = g ik hbc Kjk
c j b
x v
va + ajb xj v b = 0.
These are Wongs equations. (The form of the second of these equations suggests that the
ajb should be regarded as connection coefficients. It is indeed the case that they are: the
connection in question is that induced by on the adjoint bundle, that is, the vector bundle
associated with the principal G-bundle M by the adjoint action of G on g.)
The geodesic equations may also be derived from the Lagrangian
L = 21 g u u = 12 gij v i v j + 12 hab v a v b .
It is of course G-invariant. We may therefore apply Lagrange-Poincare reduction, which

gives the reduced equations
d gjk j k
(gij v j ) 12 v v = (Kij a j
v + aib v b )hac v c
dt xi
d
(hab v b ) = (bia v i + Cac
b c
v )hbd v d .
dt
Now aib hac is skew-symmetric in b and c, and Cac b h is skew-symmetric in c and d, so the
bd
final terms in each equation vanish identically, and we may write the equations in the form

j k l
gij xj + kl x x c j b
= hbc Kij x v

hab v b + bic xi v c = 0,
using the skew-symmetry of cib hac again in the second equation. These equations are
equivalent to the ones obtained by the direct method (Kij c is of course skew-symmetric in
its lower indices).

In order to use Rouths procedure we must rewrite the Lagrangian in terms of the quasi-
velocities associated with the mixed basis: it is given by
L = 21 gij v i v j + 12 gab v a v b .
The momentum is given by pa = gab v b , and the Routhian by
R = L pa v a = 12 gij v i v j 12 g ab pa pb .
The next problem is to calculate XiV (R) and XiC (R). In fact, it is easy to see that XiV (R) =
gij v j . The calculation of XiC (R) reduces to the calculation of Xi (gij ) and Xi (g ab ). The
first is straightforward. For the second, we note that gab = Aca Adb hcd , where (Aba ) is the
matrix inverse to (Aba ); since the right-hand side is independent of the xi , so is gab , and so
equally is g ab . It follows that
gjk j k 1 c
XiC (R) = 1
2 v v 2 i Ec (g ab )pa pb .
xi
Now Ec (g ab ) = Adc (g ae Cde
b + g be C a ), from Killings equations. Using the relation
de
between gab and hab , and the fact that ad is a Lie algebra homomorphism, we find that
Ec (g ab ) = Aad Abe (hdf Ccf

e
+ hef Ccf
d
).
The expression in the brackets vanishes, as follows easily from the properties of hab . Thus
the generalized Routh equation is
d gjk j k
j k l

(gij v j ) 21 v v = gij v j
+ kl v v a j
= a Rij v .
dt xi
a = Aa K b , and = g v b = Ac h v b , so Ra = h K c v b . The generalized
But Rij b ij a ab a bc a ij bc ij
Routh equation is therefore equivalent to

j k l
gij xj + kl c j b
x x = hbc Kij x v
again. On the other hand, the constancy of a gives
d c b
hbc (A v ) = 0.
dt a
If we are to understand this equation in the present context, we evidently need to calculate
Aba . Now
Aba = v i Xi (Aba ) + v c ec (Aba ) = v i cia Abc + v c Ccd
b
Ada .
It follows that
d c
hbc (A ) = hbc Ada Ace Aed = hbc Ada Ace (v i fid Aef + v f Cfeg Agd )
dt a
= hbc Ada (v i cid + v e Ced
c
),
where in the last step we have again used the fact that ad is a Lie algebra homomorphism.
Now from the skew-symmetry properties of hab we obtain
d c
hbc (A ) = hcd Ada (v i cib + v e Ceb
c
),
dt a
and therefore
d c b
(A v ) = hcd Ada (v c + cib v i v b ).
hbc
dt a
The first-order part of Wongs equations is thus equivalent to the constancy of momentum.
Acknowledgement
The first author is a Guest Professor at Ghent University: he is grateful to the Department
of Mathematical Physics and Astronomy at Ghent for its hospitality.
The second author is currently a Research Fellow at The University of Michigan through
a Marie Curie Fellowship. He is grateful to the Department of Mathematics for its hospi-
tality. He also acknowledges a research grant (Krediet aan Navorsers) from the Fund for
Scientific Research - Flanders (FWO-Vlaanderen), where he is an Honorary Postdoctoral
Fellow.
References
[1] A. M. Bloch, Nonholonomic Mechanics and Control (Interdisciplinary Applied Math-
ematics 24, Springer 2003).
[2] F. Bullo and A. D. Lewis, Geometric Control of Mechanical Systems (Texts in Applied
Mathematics 49, Springer 2004).
[3] H. Cendra, J. E. Marsden and T. S. Ratiu, Lagrangian Reduction by Stages (Memoirs

of the American Mathematical Society 152, AMS 2001).
[4] M. Crampin and T. Mestdag, Reduction and reconstruction aspects of second-order

dynamical systems with symmetry, preprint (2006); available at maphyast.ugent.be.
[5] M. Crampin and T. Mestdag, Rouths procedure for non-Abelian gymmetry groups,
to appear in J. Math. Phys.; arXiv:0802.0528.
[6] J. E. Marsden, G. Misiolek, J-P. Ortega, M. Perlmutter and T. S. Ratiu, Hamiltonian

Reduction by Stages (Lecture Notes in Mathematics 1913, Springer 2007).
[7] J. E. Marsden and T. Ratiu, Introduction to Mechanics and Symmetry (Texts in Applied
Mathematics 17, Springer 1999).
[8] J. E. Marsden, T. Ratiu and J. Scheurle, Reduction theory and the Lagrange-Routh
equations, J. Math. Phys. 41 (2000) 33793429.
[9] T. Mestdag and M. Crampin, Invariant Lagrangians, mechanical connections and the
Lagrange-Poincare equations, to appear in J Phys. A: Math. Theor.; arXiv:0802.0146.
[10] R. Montgomery, Canonical formulations of a classical particle in a Yang-Mills field

and Wongs equations, Lett. Math. Phys. 8 (1984) 5967.
[11] E. J. Routh, A Treatise on the Stability of a Given State of Motion (MacMillan 1877,
available on google.books.com).
Chapter 16
D IMENSIONAL R EDUCTION OF
C URVATURE -D EPENDENT C ENTRAL P OTENTIALS
ON S PACES OF C ONSTANT C URVATURE
Jose F. Carinena1, Manuel F. Ranada1 and Mariano Santander2

1
Departamento de Fsica Teorica, Facultad de Ciencias
Universidad de Zaragoza, 50009 Zaragoza, Spain
2
Departamento de Fsica Teorica, Facultad de Ciencias
Universidad de Valladolid, 47011 Valladolid, Spain
Abstract
The motion of a particle on a curvature-dependent central potential in a config-
uration spaces of constant curvature will be described, as well as the corresponding
reduction process.
2000 Mathematics Subject Classification. 37J35, 70H06, 37J15, 70G45

Key words and phrases. Spaces of constant curvature, central potentials, spheres and
hyperbolic planes.
1. Introduction and Motivation

A very well-known result of classical mechanics in three-dimensional Euclidean space is
that the motion of a particle in a central force problem takes place in a plane through the
centre of force. One can also ask what happens in spaces of constant curvature. Actually,
in the usual Euclidean case the reasoning turns out to be very simple because both the
position vector and the linear momentum vector of the particle lie always in a plane that
is orthogonal to the angular momentum vector, which is a fixed vector in space [1, 2, 3].
This fact however does not remain true anymore in the general case of a nonzero constant

E-mail address: jfc@unizar.es

E-mail address: mfran@unizar.es

E-mail address: msn@fta.uva.es
278 Jose F. Carinena, Manuel F. Ranada and Mariano Santande
curvature Riemannian space. It is therefore necessary to use a more geometric approach to

mechanics [4, 5] and to handle geometric tools for dealing with such a problem.
The theory of central potentials in constant curvature spaces has been developed by
many authors using different approaches (see Refs. [6]-[22]) but most cases are focused on
the Kepler problem, the hydrogen atom or the harmonic oscillator, with some extensions
to other particular HamiltonJacobi separable (or super-separable) potentials. We have re-
cently studied in Ref. [23] the theory of central potentials in two-dimensional spaces with
a constant curvature, the sphere S 2 and the hyperbolic plane H 2 , using a formalism intro-
duced for the theory of super-integrable systems [19], and applied to the harmonic oscillator
in previous articles [24, 25]. The corresponding quantum case has also been studied in [26].
The surprising fact is that even if apparently the three manifolds look different from a ge-
ometric perspective, they share some dynamical properties. For instance, it was proved in
[23] that Binet equation for a central potential is not only valid in the Euclidean space but
in all the spaces with a nonzero constant curvature too.
Once the well-known reduction from motion in a three-dimensional Euclidean space to
a motion in a two-dimensional Euclidean plane has been recalled, we set as our aim here to
study the curvature dependent generalization of this reduction, which should lead from the
motion in a three dimensional constant curvature space to the motion in a two-dimensional
constant curvature submanifold of the former. We present all the mathematical expressions
using the curvature as a parameter. They reduce to the appropriate property for the system
on the sphere S 3 , or on the hyperbolic plane H 3 , when particularised for > 0, or < 0,
respectively. The Euclidean case arises as the particular case = 0.
The use of not only Riemannian but also pseudo-Riemannian metrics is also possible
and we shortly describe how to develop a theory in 2d constant curvature spaces and in
particular we summarize some results on the curved Kepler problem recently published
in [27].
2. Tagged Trigonometric Functions

In order to deal simultaneously with all possible values of , it is convenient to use the
following tagged trigonometric functions

1 sin x if > 0,

cos x if > 0,
S (x) = x if = 0, C (x) = 1 if = 0,
1 sinh x if < 0. cosh x if < 0,

Note that for = 1 we have the usual trigonometrical functions,
S1 (x) = sin x, C1 (x) = cos x ,

while for = 0 one gets the parabolic sine and cosine,
S0 (x) = x, C0 (x) = 1 ,
and finally, for = 1, they are the hyperbolic functions:
S1 (x) = sinh x, C1 (x) = cosh x .

Dimensional Reduction of Curvature-Dependent Central Potentials 279
These tagged trigonometric functions have slightly modified -dependent relations but
similar to those of the classical trigonometric ones, namely,
2 2 d d
C (x) + S (x) = 1 , S (x) = C (x) , C (x) = S (x) ,
dx dx
and
2 2
C (2x) = C (x) S (x) , S (2x) = 2 S (x) C (x) .
Note that in the flat case = 0 all C (x) are replaced by 1, while all S (x) are replaced
by its variable x. In this sense C (x) may be looked at as a kind of curved deformation
of the function 1, while S (x) is a deformation of the linear function x.
3. Central Potentials in 3d Constant Curvature Manifolds

The three-dimensional sphere S 3 of radius one can be seen as a submanifold embedded in
the ambient space R4 = (x0 , x1 , x2 , x3 ) through the constraint equation x20 +x21 +x22 +x32 =
1.
We can consider a local coordinate system with three orthogonal polar coordinates
(r, , ), 0 < r < , 0 < < , 0 < < 2, given by
x0 = cos r , x1 = sin r sin cos , x2 = sin r sin sin , x3 = sin r cos , (1)
i.e. r is defined by x0 = cos r and and are the usual coordinates in a sphere of radius
sin r, x21 + x22 + x32 = sin2 r.
The North pole (1, 0, 0, 0) is outside the domain of this chart but it is obtained as the
limit when r goes to zero. It can be considered as the origin of these new coordinates. The
expression of the induced Riemannian metric in S 3 is given by
ds2 = dr2 + sin2 r d2 + sin2 r sin2 d2 , (2)
from which the following expression for the kinetic energy T in S 3 , the Lagrangian for the
free system, is obtained
1 2
vr + sin2 r v2 + sin2 r sin2 v2 .

T = (3)
2
The dynamical vector field is given by

X T = vr + v + v + fr + f + f , (4)
r vr v v
where the functions fr , f and f are given by
= sin r cos r (v2 + sin2 v2 ) ,

fr

cos r

v vr + sin cos v2 ,

f = 2 (5)

sin r

cos r

f

= 2 v vr .
sin r
The trajectories of this free system are then given by the solutions of the system of second-
order differential equations
= sin r cos r (2 + sin2 2 ) ,

r

cos r

r + sin cos 2 ,

= 2 (6)

sin r

= 2 cos r r ,

sin r
which coincide with the geodesics of the metric. Note that the curves = 0 , = 0 ,
r = 1 are geodesics through the North pole and the distance from this point to (r0 , 0 , 0 )
is r0 .
A natural Lagrangian (kinetic term minus a potential) is then of the following form:
1 2
vr + sin2 r v2 + sin2 r sin2 v2 k U (r, , ) ,

L= (7)
2
and the corresponding dynamics is given in the velocity phase space by the associated
Euler-Lagrange vector field

X L = vr + v + v + Fr + F + F , (8)
r vr v v
where (Fr , F , F ) denote the forces

Fr = fr k Ur ,
k

U ,

F = f (9)
sin2 r

k
F = f U .

sin r sin2
2
The Euler-Lagrange equations of motion are given by

r
= Fr (r, , , r, , ) ,
= F (r, , , r, , ) , (10)

= F (r, , , r, , ) .
Note that XL is a second-order differential equation vector field, because the Lagrangian is
regular.
Suppose now that instead of considering R3 we consider a family of three-dimensional
submanifolds Q3 of R4 defined by
x20 + (x21 + x22 + x23 ) = 1,
where R.
We can use the above mentioned tagged functions to parametrise such -dependent
submanifold Q3 with local coordinates (r, , ), 0 < r < , 0 < < , 0 < < 2, as
follows
x1 = S (r) sin cos , x2 = S (r) sin sin , x3 = S (r) cos , x0 = C (r) ,
(11)
i.e. we are using the usual spherical coordinates (, ) on the sphere of radius S (r) with the
ambient metric. Note however that in the induced metric the curvature is . In this way, if
we use the notation of Q3 for representing the three three-dimensional spaces with constant
curvature , then the following -dependent expression for the induced Riemannian metric
on the manifold Q3 , that is, in S3 , E3 and H3 , respectively, is obtained:
ds2 = dr2 + S2 (r) d2 + S2 (r) sin2 d2 , (12)
which reduces respectively to
ds21 = dr2 + (sin2 r) d2 + (sin2 r sin2 ) d2 ,

ds20 = dr2 + r2 d2 + (r2 sin2 ) d2 ,
ds21 = dr2 + (sinh2 r) + (sinh2 r sin2 ) d2 ,
and we can study simultaneously the three-dimensional sphere S3 , the Euclidean space E3 ,
and the hyperbolic space H3 , just particularising the parameter for the values > 0,
= 0 or < 0, respectively.
In the mechanical setting, the free motion on such manifolds is described by the La-
grangian L0 corresponding to the metric (12) which is given by
1 2
vr + S2 (r) v2 + S2 (r) sin2 v2 .

L0 () = (13)
2
The dynamical vector field is

XL0 = vr + v + v + fr + f + f , (14)
r vr v v
where (fr , f , f ) represent the -forces characterising the free geodesic motion on the
configuration space Q3 , explicitly given by

fr = S (r) C (r) (v2 + sin2 v2 ) ,
C (r)

v vr + sin cos v2 ,

f = 2

S (r) (15)
C (r) cos

f = 2 v vr 2 v v ,

S (r) sin
and the trajectories of the free motion are the solutions of the system of second-order dif-
ferential equations

r = S (r) C (r) (2 + sin2 2 ) ,
C (r)

r + sin cos 2 ,

= 2

S (r) (16)
C (r) cos

= 2 r 2 ,

S (r) sin
which coincide with the equations for the geodesics of the metric. Note that the curves
starting from the North pole = 0 , = 0 and r = 1, are geodesics and the distance from
this point to (r0 , 0 , 0 ) is r0 .
A more general natural Lagrangian (-dependent kinetic term minus a potential) for a
system in the configuration space Q3 , has the following form
1 2
vr + S2 (r) v2 + S2 (r) sin2 v2 k U (r, , ; ) ,

L() = (17)
2
in such a way that for = 0 we recover a standard Euclidean system
1 2
lim L() = (v + r2 v2 + r2 sin2 v2 ) k V (r, , ) , V (r, , ) = U (r, , ; 0) .
0 2 r
The dynamics is then represented by the -dependent second-order differential equation
vector field

X L = vr + v + v + Fr + F + F , (18)
r vr v v
where the three forces (Fr , F , F ) are respectively given by

Fr = S (r) C (r) (v2 + sin2 v2 ) k Ur ,
C (r) k

v vr + sin cos v2 2 U ,

F = 2

S (r) S (r) (19)
C (r)

v vr 2 cos k

F = 2 sin v v S2 (r) sin2 U ,

S (r)
and therefore the trajectories of the motion under the action of a potential k U (r, , ; )
are the solutions of the system of second-order differential equations

r = S (r) C (r) (2 + sin2 2 ) k Ur ,
C (r) k

r + sin cos 2 2 U ,

= 2

S (r) S (r) (20)
C (r) cos k k

= 2 r 2 2 U 2 U .

2
S (r) sin S (r) sin S (r) sin2
Now, having in mind the meaning of the coordinate r as distance from the point to
the North pole, we call central a potential which as a function on the configuration space
depends only on the distance r. Our aim is to prove that the dynamical vector field XL
arising from a central potential U (r; ), i.e. described by a Lagrangian
1 2
vr + S2 (r) v2 + S2 (r) sin2 v2 k U (r; ) ,

L() = (21)
2
with forces
Fr
= fr k Ur ,
F
= f , (22)
F

= f ,

is tangent to a -dependent foliation M () of the phase space T Q3 whose leaves are

tangent bundles of a family M () of submanifolds of the configuration space Q3 and that
this property is true no matter the possible values of .
We recall that a vector field X X(P ) in a manifold P is tangent to a submanifold N

of P when X|N takes values in T N . If N is determined by a rank one function : P R
as N = 1 (0), this is characterized by the weak (in Diracs sense) equality
X 0 (X)(n) = 0 , n N . (23)
e ,~ : Q3 R
The family M () is defined as follows: Consider first the function
given by
~
e , (r, , ) C (r) + S (r) (1 sin cos + 2 sin sin + 3 cos ) ,
(24)
and its zero value level set, which is the intersection of Q3 with the hyperplane x0 +
1 x1 + 2 x2 + 3 x3 = 0. Varying the parameters (, 1 , 2 , 3 ), we obtain a family M f
2 3
of either two-dimensional spheres S inside S when > 0, or Euclidean planes E inside 2
E3 when = 0, or hyperbolic planes H2 inside H3 when < 0. Furthermore, in the three

cases the centre of the potential is placed at the particular point characterised by r = 0, that
in the spherical > 0 case, corresponds to the North pole N of S3 .
Consequently, we can restrict ourselves to the subfamily M () of M f() given by all
those submanifolds in M () passing through this particular point, i.e. we put = 0, and
f
consider the intersection with the zero level set of the function.
~
S (r) (1 sin cos + 2 sin sin + 3 cos ) . (25)
So, in the following we restrict our study to this two-parameter family M (). The
tangency conditions for the second-order vector field XL is
~ ~ ~ ~
XL ( ) = 0 , XL ( ) 0 . (26)
The first tangency condition has a geometric or kinematic meaning, independent of the
potential function U (r; k),
vr r + v + v = 0, (27)
while the second condition for the vector field XL leads to
vr2 rr + v2 + v2 + 2vr v r + 2vr v r + 2v v +

+(fr k Ur )r + f + f 0 ,
which splits into a system of seven -dependent equations. Three of these equations,
C (r) C (r) cos

r = 0 , r = 0 , = 0, (28)
S (r) S (r) sin
are identically satisfied by the own definition of the function , and the other four equations
turn out to be
rr = 0 ,
+ S (r) C (r) r = S2 (r) 0 ,
+ S (r) C (r) sin2 r + sin cos = S2 (r) sin2 0 ,
C (r)
r Ur = Ur 0 .
S (r)
We see from 0 that the conditions are satisfied in the weak sense. Therefore, the
trajectories of a particle on Q3 under the action of a central potential U (r; ) are always on
a two-dimensional submanifold Q2 inside the initial configuration space Q3 in such a way
that:
(1) If the configuration space is the sphere S3 then the trajectories of the particle are
curves on a two-dimensional sphere S 2 passing through the center of forces.
(2) If the configuration space is the Euclidean space E3 then the trajectories are curves
on a Euclidean plane E2 passing through the center of forces.
(3) If the configuration space is the hyperbolic space H3 then the trajectories of the
particle are curves on a hyperbolic plane H 2 passing through the center of forces.
Hence this property is not a specific or special characteristic of the Euclidean world,
but it also holds in all the three spaces of constant curvature; therefore, the three mentioned
situations are but three particular instances of a more general property. Also we note that,
in all the three cases, the value of the curvature of the two-dimensional submanifold, S 2 ,
E2 or H 2 , is the same as in the original three-dimensional space, i.e. there is a reduction
of the dimension but not a change of the curvature.
4. Motions on 2d Constant Curvature Submanifolds

Instead of considering 2d constant curvature surfaces in a 3d Euclidean space we can study
the more general case of the CayleyKlein (hereafter shortened as CK) 2d spaces S21[2 ]
[28] with constant curvature 1 and metric of a signature type (1, 2 ). Each one of the
two parameters 1 and 2 can be brought to a standard value 1, 0 or 1, and the nine
combinations of these values correspond to the so called standard CK spaces.
We display these nine spaces in a Table. The three rows accommodate spaces with either
a Riemannian, or a degenerate, or a pseudo-Riemannian (Lorentzian) metric, respectively,
according to the sign of 2 . The three instances along each row are the spaces with positive,
zero or negative constant curvature (for more details see [29]-[31].
The symmetric homogeneous space S21[2 ] admits a maximal three-dimensional isom-
etry group denoted SO1 ,2 (3), whose three dimensional Lie algebra is so1 ,2 (3) with
generators P1 , P2 and J given in the matrix realization by:

0 1 0 0 0 1 2 0 0 0
P1 = 1 0 0 P2 = 0 0 0 J = 0 0 2 ,

0 0 0 1 0 0 0 1 0
with the following commutation relations among the generators J, P1 and P2 :
[J, P1 ] = P2 , [J, P2 ] = 2 P1 , [P1 , P2 ] = 1 J .

Table 1. The nine standard two-dimensional CK spaces S21[2 ] .
Measure of distance & Sign of 1

Measure of angle Elliptic Parabolic Hyperbolic
& Sign of 2 1 = 1 1 = 0 1 = 1
Elliptic Euclidean Hyperbolic
Elliptic 2 = 1 S2 E2 H2
Co-Euclidean Galilean Co-Minkowskian
Oscillating NH Expanding NH
Parabolic 2 = 0 ANH1+1 G1+1 NH1+1
Co-Hyperbolic Minkowskian Doubly Hyperbolic
Anti-de Sitter De Sitter
Hyperbolic 2 = 1 AdS1+1 M1+1 dS1+1
The group SO1 ,2 (3) acts by matrix multiplication on a R3 ambient space by isome-
tries of the ambient space metric:
dl2 = (ds0 )2 + 1 (ds1 )2 + 1 2 (ds2 )2 ,
and the space S21[2 ] is the homogeneous space S21[2 ] SO1 ,2 (3)/SO2 (2), where
SO2 (2) is the subgroup generated by J.
The relation among ambient coordinates and polar coordinates is given by:

s0 C1 (r)
1
s = S1 (r) C2 () ,

s2 S1 (r) C2 ()
and therefore the induced metric is given by:

2
dl|S 2 = dr2 + 2 S21 (r) d2 .
1[2 ]
The Killing vector fields are the first-order differential operators in polar coordinates:
S ()
X P1 = C2 () 2 ,
r T1 (r)
C2 ()
X P2 = 2 S2 () + ,
r T1 (r)

XJ = .

and close on a so1 ,2 (3) Lie algebra. The associated momenta are:

P1 C2 () vr 2 C1 (r) S1 (r) S2 ()v
2 2 S2 ()vr + 2 C1 (r) S1 (r) C2 ()v ,
P =

J 2 S21 (r)v
but it is more convenient to introduce new variables, to be called CK Noether momenta:
P2 J
P1 := P1 , P2 := , J := ,
2 2
which are essentially equivalent to the previous ones when 2 6= 0 but admit a limit when
2 0. The three CK momenta are linked, for all values of 1 and 2 by a fundamental
relation:
s2 P1 s1 P2 + s0 J = 0,
which reduces in the standard Euclidean case to the well known (Euclidean) relation be-
tween angular and linear momentum J = xP2 yP1 .
The Kepler potential in a S21[2 ] , defined to be [6, 9]:
k
VK = . (29)
T1 (r)
has recently been studied in [27] and it has been shown to be super-integrable. A first
integral is linked to the invariance of VK under rotations around the potential center, and
leads to the constancy of angular momentum J , and to the quadratic constant IJ 2 = J 2 .
In any S21[2 ] , the potential (29) allows for two additional constants of motion of types
IJ P1 , IJ P2 , which are associated to the separability of the Kepler potential in two equipara-
bolic 01 and 20 coordinate systems (see [32]). These constants of motion are:
IJ P1 = J P1 + W01 , W01 = k S2 ()
(30)
IJ P2 = J P2 + W02 , W02 = k V2 () ,
where V2 () is the CK version of the versed sine, V2 () = (1C2 ())/2 ; for 2 = 0

this reduces to the function 2 /2. There is a relation among the functions W01 , W02 as a
consequence of C22 () + 2 S22 () = 1:
(W01 )2 + W02 (2 W02 2) = 0 . (31)
The two constants IJ P1 , IJ P2 , together with the energy and the angular momentum:
1 2 k
IE = (P1 + 2 P22 + 1 2 J 2 ) ,
2 T1 (r) (32)
IJ 2 = J 2,
provide a set of four constants of motion. As the maximal number of functionally indepen-
dent constants of motion for this system is three, a single relation among the four I above
should exist. This relation, which is quadratic in the Is can be checked using (31):
(IJ P1 )2 + IJ P2 (2 IJ P2 2k) = (2IE 1 2 IJ 2 ) IJ 2 . (33)

The two constants IJ P1 , IJ P2 can be seen as the components of a single Keplerian

(conserved) vector under the (sub)group SO2(2) of rotations around the origin in the space
S21[2 ] . This vector will be called here the CK eccentricity vector E ; along any evolution
under the Kepler potential, the (constant) values of the components of E will be denoted
E01 , E02 : ! ! !
E01 IJ P1 J P1 + W01
= = , (34)
E02 IJ P2 J P2 + W02
and in terms of E01 , E02 and of the values of energy and angular momentum, (33) is written

2
E01 + E02 (2 E02 2k) = 2E 1 2 J 2 J 2 , (35)
which is well defined in all CK spaces S21[2 ] . The vector E is related to the ordinary
LRL vector in the Euclidean case; notice also the appearance in the r.h.s. of the specific
combination 2E 1 2 J 2 , which for the Euclidean space reduces to 2E.
Hence, the existence of a Keplerian additional conserved vector is not a specifically
Euclidean property, but still holds even if the configuration space is the more general space
S21[2 ] , with any constant curvature and any signature. The Riemannian part of this
statement has been known since a long time for the Kepler problem in S21 and in H21 ;
these cases appear in our approach when 2 = 1. This CK formalism also covers the
cases where 2 < 0, i.e. the Kepler problem in a locally Minkowskian constant curvature
configuration space. For the motion of a particle under the action of the curved Kepler
potential in a curved 3-d configuration space of CayleyKlein type, there also exists a
curved form for the LaplaceRungeLenz vector, which has been obtained by Herranz
and Ballesteros [29].
We also point out that the super-integrability of the problem can be used to find the
orbits in this curved Kepler potential, as it has been shown in [27], by introducing the CK
LaplaceRungeLenz vector a vector A :
! ! !
A1 2 E02 k 2 (J P2 + W02 ) k
:= = . (36)
A2 E01 J P1 W01
In the Euclidean plane this vector reduces precisely to the LaplaceRungeLenz vector. In
intrinsic terms, the CK LaplaceRungeLenz vector is the Hodge 2 -* of the eccentricity
vector shifted by the constant vector (k, 0). Moreover, it has also been proved that the
momentum hodographs for this curved Kepler problem are also cycles. The details can be
found in the recent paper [27].
5. Final Comments
The Euclidean space E3 can be considered as a very particular or limiting case of the con-
stant curvature spaces. The curvature modifies some details but preserves the fundamental
structures.
There is a sound geometrical reason for this situation. If we consider a particular po-
tential in the flat Euclidean space that is invariant under reflection in a given 2-plane, then
the reflected of any possible motion is a possible motion too, the one determined by the re-
flected initial conditions. If we choose the initial conditions to be a position in the plane and
a velocity tangent to the plane, such conditions are invariant under reflection, and therefore
the reflected motion has the same initial conditions as the original one, hence both motions
coincide, i.e. for these special initial conditions the whole trajectory is contained in that
plane. A central potential is invariant under any reflection in a 2-plane through the origin.
Thus any generic set of initial conditions (position not on the centre, nonzero velocity) de-
termines a particular plane passing through the centre of forces and the initial point, and
containing the initial velocity. The previous reasoning when applied to this plane implies
that the motion is contained in such a plane.
Assume now that the 3d configuration space is not R3 , but it has a constant curvature .
Then a similar reasoning can be made, with totally geodesic 2-dimensional submanifolds
playing the role of Euclidean 2-planes. This follows directly from the Weierstrass ambient
space model, where the totally geodesic 2-dimensional submanifolds are the intersection of
the sphere x20 + (x21 +x22 +x23 ) = 1 with a 3-plane (just in the same way as geodesics S 1
S 2 are the intersection of S 2 R3 with 2-planes through the origin in R3 ). Reflections in
that 3-plane in the ambient space represent reflections on the totally geodesic S 2 S 3 or
H 2 H 3 . Hence motion on a space S 3 or H 3 , in a potential invariant under geodesic
reflection on a fixed totally geodesic S 2 or H 2 with initial position on that submanifold and
initial velocity tangent to it, is always fully contained in that submanifold. As an arbitrary
central potential is invariant under reflection in any totally geodesic 2d submanifold, either
S 2 or H 2 respectively, through the centre of forces, the previous reasoning, when applied
to these submanifolds, implies that the motion stays in such submanifold.
In one of our previous articles we quote a statement by Fronsdal [33]: A physical
theory that treats space-time as Minkowskian flat must be obtainable as a well-defined
limit of a more general physical theory, for which the assumption of flatness is not es-
sential. Our study is not a relativistic one, but, in a sense, this statement looks rather
similar (Minkowskian must be changed to Euclidean) to our idea of using the concept of
deformation as an approach with as the parameter of the deformation. The flatness as-
sumption = 0 is not necessary for proving that the motion of a particle in a central force
problem is always a motion in an analogous two-dimensional submanifold with the same
curvature as in the total space. A similar situation was found in [23] for the Kepler problem
where it was proved that even if 6= 0 the particle moves in conics. Also in this case the
property of = 0 flatness was not a necessary assumption for arriving to such an important
characteristic.
We end this paper with two comments. In the first place let us recall that gravita-
tion introduces curvature not only in the four-dimensional space-time but also in the three-
dimensional space-like surfaces. So, although this study has been a non-relativistic one,
we think that the results obtained in this article could also be of use in the study of rela-
tivistic gravitation models. Secondly, we have presented an approach based on tangency
conditions that has proved to be successful; nevertheless it is natural to look also at this ge-
ometric problem from the more traditional viewpoint of Noethers theorem and the theory
of symmetries (in this case -dependent symmetries). This alternative approach remains as
an interesting matter to be studied.
Acknowledgement
Support of projects E24/1 (DGA), MTM-2005-09183, MTM-2006-10531, and VA-013C05
is acknowledged.
References
[1] E. T. Whittaker, A Treatise on the Analytical Dynamics of Particles and Rigid Bodies
(4th ed., Cambridge University Press, 1965).
[2] H. Goldstein, Classical Mechanics (2nd ed., Addison-Wesley, Reading, Mass., 1980).
[3] J. V. Jose and E. J. Saletan, Classical Dynamics. A contemporary approach (Cam-

bridge University Press, Cambridge, 1998).
[4] R. Abraham and J. E. Marsden, Foundations of Mechanics (2nd ed., Benjamin/Cum-

mings, Reading, Mass., 1978).
[5] G. Marmo, E. Saletan, A. Simoni and B. Vitale, Dynamical Systems. A Differential

Geometric Approach to Symmetry and Reduction (John Wiley, Chichester, 1985).
[6] E. Schrodinger, A method of determining quantum mechanical eigenvalues and eigen-

functions, Proc. R. Ir. Acad. A 46 (1940) 916.
[7] L. Infeld, On a new treatment of some eigenvalue problems, Phys. Rev. 59 (1941)
737747.
[8] A. F. Stevenson, Note on the Kepler Problem in a spherical space, and the factoriza-
tion method of solving eigenvalue problems, Phys. Rev. 59 (1941) 842843.
[9] L. Infeld and A. Schild, A note on the Kepler problem in a space of constant negative
curvature, Phys. Rev. 67 (1945) 121122.
[10] P. W. Higgs, Dynamical symmetries in a spherical geometry I, J. Phys. A: Math. Gen.

12 (1979) 309323.
[11] H. I. Leemon, Dynamical symmetries in a spherical geometry II, J. Phys. A: Math.

Gen. 12 (1979) 489501.
[12] E. G. Kalnins, W. Miller and P. Winternitz, The group O(4), separation of variables
and the hydrogen atom, SIAM J. Appl. Math. 30 (1976) 630664.
[13] A. O. Barut and R. Wilson, On the dynamical group of the Kepler problem in a curved
space of constant curvature, Phys. Lett. A 110 (1985) 351354.
[14] A. O. Barut, A. Inomata and G. Junker, Path integral treatment of the hydrogen atom
in a curved space of constant curvature, J. Phys. A: Math. Gen. 20 (1985) 62716280.
[15] A. O. Barut, A. Inomata and G. Junker, Path integral treatment of the hydrogen atom in
a curved space of constant curvature: II, J. Phys. A: Math. Gen. 23 (1990) 11791190.
[16] C. Grosche, The path integral for the Kepler problem on the pseudosphere, Ann. Phys.
204 (1990) 208222.
[17] C. Grosche, On the path integral in imaginary Lobachevsky space, J. Phys. A: Math.
Gen. 27 (1994) 34753489.
[18] C. Grosche, G. S. Pogosyan and A. N. Sissakian, Path integral discussion for

Smorodinsky-Winternitz potentials II, Fortschr. Phys. 43 (1995) 523563.
[19] M. F. Ranada and M. Santander, Superintegrable systems on the two-dimensional

sphere S 2 and the hyperbolic plane H 2 , J. Math. Phys. 40 (1999) 50265057.
[20] J. J. Slawianowski, Bertrand systems on spaces of constant sectional curvature, Rep.

Math. Phys. 46 (2000) 429460.
[21] E. G. Kalnins, J. M. Kress, G. S. Pogosyan and W. Miller, Completeness of superin-

tegrability in two-dimensional constant-curvature spaces, J. Phys. A: Math. Gen. 34
(2001) 47054720.
[22] E. G. Kalnins, W. Miller and G. S. Pogosyan, The Coulomb-oscillator relation on n-

dimensional spheres and hyperboloids, Phys. of Atomic Nuclei 65 (2002) 11191127.
[23] J. F. Carinena, M. F. Ranada and M. Santander, Central potentials on spaces of con-

stant curvature: The Kepler problem on the two-dimensional sphere S 2 and the hyper-
bolic plane H 2 , J. Math. Phys. 46 (2005) 052702, 125.
[24] M. F. Ranada and M. Santander, On the Harmonic Oscillator on the two-dimensional

sphere S 2 and the hyperbolic plane H 2 , J. Math. Phys. 43 (2002) 431451.
[25] M. F. Ranada and M. Santander, On the Harmonic Oscillator on the two-dimensional

sphere S 2 and the hyperbolic plane H 2 II, J. Math. Phys. 44 (2003) 21492167.
[26] J. F. Carinena, M. F. Ranada and M. Santander, The quantum harmonic oscillator on

the sphere and the hyperbolic plane: -dependent formalism, polar coordinates, and
hypergeometric functions, J. Math. Phys. 48 (2007) 102106.
[27] J. F. Carinena, M. F. Ranada and M. Santander, Superintegrability on curved spaces,

orbits and momentum hodographs: revisiting a classical result by Hamilton, J. Phys.
A: Math. Theor. 40 (2007) 1364513666.
[28] J. F. Carinena, M. A. del Olmo and M. Santander, A new look at Dimensional Analysis
from a group-theoretical viewpoint, J. Phys. A: Math. Gen. 18 (1985) 18551872.
[29] F. J. Herranz and A. Ballesteros, Superintegrability on three-dimensional Riemannian

and relativistic spaces of constant curvature, SIGMA 2 (2006) 010 pp. 22.
[30] F. J. Herranz, R. Ortega and M. Santander, Trigonometry of space-times: a new self-

dual approach to a curvature/signature (in)dependent trigonometry, J. Phys. A: Math.
Gen. 33 (2000) 45254551.
[31] F. J. Herranz, M. Santander, Casimir Invariants for the complete family of quasi-
simple orthogonal algebras, J. Phys. A: Math. Gen. 30 (1997) 54115426.
[32] J. F. Carinena, M. F. Ranada, M. Santander and T. Sanz-Gil, Separable potentials and

triality in 2d-spaces of constant curvature, J. Nonlinear Math. Phys. 12 (2005) 230
252.
[33] C. Fronsdal, Elementary particles in a curved space, Rev. Mod. Phys. 37 (1965) 221
224.
Chapter 17
D IRECT G EOMETRICAL M ETHOD

IN F INSLER G EOMETRY
L. Tamassy
Debrecen University
I. Minkowski Angle and Related Problems
Introduction
Finsler geometry is a very natural generalization of Riemannian geometry. Both are built
on the arc length of curves.
A Finsler space F n = (M, F) is given by an n-dimensional manifold M and a funda-
mental (metric) function
F : T M R+ , (p, y) 7 F(p, y), p M, y Tp M,
where R+ means the non-negative reals. F must satisfy the following requirements [BSC]:
(i) regularity: F C on the slit tangent bundle T M : T M without the null section
{(p, 0)} T M , and F C on the null section
(ii) positive homogeneity: F(p, y) = F(p, y), R+

2F 2
(iii) strong convexity: y i y j
(p, y)v i v j 12 gij (p, y)v i v j 0, v Tp M .
An important, more restrictive version of homogeneity is the
(iv) absolute homogeneity: F(p, y) = ||F(p, y), R.
The Finsler norm of a vector y Tp M is
kyk := F(p, y). (1)

E-mail address: tamassy@math.klte.hu
294 L. Tamassy
The Finsler arc length of a curve c : [a, b] M is defined by the integral of the norm
kc(t)k of the tangent vector of the curve c(t):
Z b
s := F(c(t), c(t))dt. (2)
a
(1) means that a Finsler space endows the vectors y Tp M with a norm with the properties:
a) kyk 0, and kyk = 0 y = 0; b) kyk = kyk, R+ ; c) ky1 + y2 k
ky1 k + ky2 k, y1 , y2 Tp M ; and d) a differentiability (continuity) property in p and y
following from (i) and (1). Conversely, if Tp M are endowed with such a norm, then M and
F given by (1) is a Finsler space. A Riemann space V n = (M, g) is that special case where
F 2 is quadratic in c. Then the Riemann arc length has the form s = ab (hc, cig )1/2 dt.
R
An important notion is the indicatrix hypersurface
I(p0 ) := {y Tp0 M | F(p0 , y) = 1}.
It plays the role of the Euclidean unit sphere S n1 E n . In consequence of (iii) I(p)
is strictly convex. (iv) is equivalent to the symmetry of the indicatrix, or to F(p, y) =
F(p, y), p, y. (ii) is equivalent to the invariance of the arc length s with respect to the
orientation preserving parameter transformations. (iv) is equivalent to the invariance of the
arc length s with respect to any parameter transformation. In case of (iv) the norm induced
by F on Tp M has the property: eb) kyk = || kyk, R, and thus in this case Tp M
becomes a Banach space. In the case of a Riemann space the indicatrices are ellipsoids
varying with the point p, and the tangent spaces become Euclidean spaces. So we can say
with S. S. Chern [7] that Finsler geometry is nothing but Riemannian geometry without the
quadratic restriction.
Finsler geometry originsted from a variational problem. The first steps were done by P .
Finsler in his thesis written under the guidance of C. Charatheodory in Gottingen in 1918,
and thus Finsler geometry has close relation to the field of Demeter Krupka to whom this
volume is dedicated. Later Finsler geometry was developed on the analogy of Riemann
geometry. Thus the existence of a linear metrical connection between the tangent vectors
of the base manifold is of basic importance in this theory. Such a connection is metrical if
the length (the norm) of the parallel translated tangent vectors remains unaltered. It means
that by a metrical connection indicatrices (the ensemble of the unit Finsler vectors) are taken
into indicatrices. However in a Finsler space in the general case this can not be satisfied by a
linear connection, for indicatrices may be arbitrary, and thus they cannot be taken into each
other by linear transformations. Indeed, it is easy to see that by the homogeneity condition
(ii) indicatrix bundle and fundamental function determine each other. Thus, if we give a
Finsler space by such indicatrices which are not in affine (linear) relation to each other,
then in this Finsler space there exists no linear metrical connection between the tangent
vectors.
This problem was eliminated by E. Cartan by introducing the line elements (p, y). (p, y)
is equivalent to (p, y), R, 6= 0. Then the n-dimensional vectors sitting on the
line elements (according to R. Ingarden the Finsler vectors) form a vector bundle =
(E, , T M, V n ) with the 2n-dimensional base space T M , and with n-dimensional vector
spaces as fibers: 1 (p, y) = V(p,y) T M T T M . The square of the Finsler norm of such a
Direct Geometrical Method in Finsler Geometry 295
vector is k(p, y)k2 = gij (p, y) i (p, y) j (p, y), gij (p, y) taken from to (ii). With this norm
one can construct metrical linear connections in VT M = {V(p,y) T M | (p, y) T M }.
On the other hand, introduction of line elements makes the apparatus and the theory
of Finsler geometry a little more complicated. This inconvenience does not arise if we
consider purely metrical questions, only, as arc length, area, angle, geodesic, isometry, etc.,
or if we investigate Finsler spaces which admit metrical linear connections between the
tangent vectors (p) of the base manifold M . There are a number of such Finsler spaces,
and just they are the ones most near to Riemannian spaces, and that gives them a certain
significance.
In this paper we investigate questions which do not involve line elements (or involve
them very rarely), i.e. we consider purely metrical questions, and such Finsler spaces,
which admit metrical linear connections in the tangent bundle T M . We often use direct
geometrical considerations in place of analytical calculations. In spite of the fact that the
analytical method is in general more powerful than the geometrical method, the last one
turns out in several cases to be surprisingly effective.
1. Angle in Minkowski and Finsler Spaces

In a Finsler space F n = (M, F) angle usually means the angle of two vectors (p, y),
(p, y) of V(p,y) T M T T M . The cosine of = (, ) is given by
gij (p, y) i (p, y) j (p, y)

cos := , ()
kk.kk
where gij (p, y) is given by (iii). This gij is a Riemannian metric in V(p,y) T M .
In this paragraph we consider the angle of two vectors (p), (p) of a tangent space
Tp M of the base manifold of a Finsler space. This Minkowski angle has attracted less
interest. Since the Finsler space makes its tangent space into a Minkowski space, measuring
of such angles in a Finsler space reduces to that in a Minkowski space. In Section 2 we prove
that a diffeomorphism between two Finsler spaces is an isometry iff it keeps angle (in the
above sense) and area, similary to the well-known result of Riemannian geometry. We also
show (Section 3) that this angle is applicable in measuring the deviation of a Finsler space
from being Riemannian.
Given a Finsler space F n = (M, F) we consider an angle = (a, b) between two
rays a, b Tp0 M emanating from the origin 0 = p0 of Tp0 M . Tp0 M is an n-dimensional
vector space V n . a and b span a two-dimensional linear subspace of Tp0 M , provided a is
not parallel to b : a b. The convex domain of bounded by a and b will be denoted by
A. This is unambigous if a b. If a = b, then A = . If a, b g, a 6= b, then the straight
line g cuts into + and . Then A = + or A = .
Minkowski space is a special Finsler space. If the fundamental function F of the Finsler
space F n = (Rn , F) (where Rn is the n-dimensional number space with the cannonical
topology) has the property that in a coordinate system (x) F(x, y) is independent of the
points x, then it is a Minkowski space Mn = (Rn , F), and the coordinate system (x) is
called adapted. F n makes each Tx0 M into a Minkowski space Mnx0 with indicatrix body
Bxn0 (1) := {y | F(x0 , y) 1} Tx0 M and with the Minkowski functional F(y) =
296 L. Tamassy
F(x0 , y) : Tx0 M R+ . Bxn (1) is then a Minkowski ball of radius 1, and Bxn (1) = I is
the indicatrix (hyper) surface. By Bx2 (1) = Bxn (1) , Mnx (or F n ) induces on Tx M
a two-dimensional Minkowski metric and thus an M2x . Bx2 (1) A = D is a segment of the
indicatrix body of M2x belonging to (a, b).
We denote the Minkowski area in M2x by k kM , and the Euclidean area in equipped
with a Euclidean metric by k kE . Then the 2-dimensional Minkowski area of D in Mnx is
the Minkowski area of D in M2 M:

Z
kDkM = dy 1 dy 2 , = , (3)
D kB 2 kE
(Z. Shen [15, 1.3], or H. Busemann [6], H. Rund [14, Chap. I, 8], D. Bao
S. S. Chern Z. Shen [2, 1.4], and many other places.) Since D dy 1 dy 2 is the Euclidean
R
area of D, (3) is equivalent to

kDkE
kDkM = . (3 )
kB 2 kE
(3) and (3 ) are true for any domain G in place of D.
The Minkowski measure of the angle M (a, b) can be defined as follows:
Definition.
M (a, b) := 2kDkM . = 1 or 1. (4)
The sign depends on the orientation of the angle. (4) is basicaly the Landsberg an-
gle [14, p. 33]. M can be expressed by the Minkowski functional F and the data of the
two legs a and b [5, (3.a) and (3.b)].
If Mnx is a Euclidean space E n , then (4) reduces to the Euclidean measure E of the
angle . Indeed, if Mnx = E n , then B 2 is the Euclidean unit ball. Thus M = 2kDkM
is a generalization of the Euclidean measure of .
kDkM has a sign, and because of the additivity of the area, M is additive:
M 1 + M 2 = M (1 + 2 ). Also M is symmetric in the sense that |M (a, b)| =
|M (b, a)|.
The case of the straight angle: In this case a b = g is a line through 0 Tx M . Let
(a, b) = + be the straight angle with the domain + +
g = A , and (b, a) = the

straight angle with the domain g = A . Because of the additivity
M + + M = 2 g.
Therefore the equality M + = M of the Minkowski measure of the two straight

angles + and implies kB 2 A+ kE = k+ D2 kE = k D2 kE = kB 2 A kE , and
conversely. In other words: M + = M iff g bisects B 2 .
If B 2 is symmetric, then every line g through O bisects B 2 . We show that also con-
versely, if every g through O bisects B 2 , then B 2 is symmetric. Suppose that B 2 is
non-symmetric. Then there exists a 0 , such that in a Euclidean polar coordinate system
r(0 ) > r(0 + ), where (r(), ) B 2 , . A g is fixed by its direction . Then for
every g()
1 + 2 1
Z
r ()d = kB 2 kE , 0 < .
2 2
Especially Z 0 + Z 0 ++
r2 ()d = r2 ()d.
0 0 +
Hence Z 0 + Z 0 ++
r2 ()d = r2 ()d.
0 0 +
By the integral mean value theorem
0 1 0 +
2r2 (1 ) = 2r2 (2 ),
0 + 2 0 + + ,
and because of the continuity of r(), 0 yields r(0 ) = r(0 + ) in contradiction to

our assumption. Therefore B 2 is symmetric. This is equivalent to the absolute homogeneity
of F.
These statements are summed up in
Theorem 1. M = 2kDkM is an additive, symmetric measure of the angles in
Minkowski or Finsler spaces. In a Euclidean space this reduces to the Euclidean measure
of . M
e = for every straight angle e if and only if the Finsler metric is absolute
homogeneous.
2. Diffeomorphisms which Keep Angle and Area

Using the notations of the previous paragraph, let F n = (M, F) and F n = (M , F) be
two connected paracompact Finsler spaces, : M M a diffeomorphism, I(p0 ) :=
{y Tp0 M | F(p0 , y) = 1} and I(p0 ) := {y Tp0 M | F(p0 , y) = 1}, p0 = (p0 )
are indicatrix hypersurfaces (indicatrices) of F n and F n resp. I(p) = I 2 (p) is the
2 2
indicatrix of M2p , and I(p) = I (p), p = (p), = () is the indicatrix of Mp .
is an isometry iff
(d)I(p) = I(p), p M. (5)
Theorem 2. The diffeomorphism : M M is an isometry between the Finsler spaces
F n and F n iff keeps angle (in the sense of paragraph 1) and (2-dimensional) area.
A) Suppose that is an isometry. By (5)
2
(d)I 2 (p) = (d)I(p) (d) = I(p) = I (p). (6)
2
and equipped with Euclidean metrics are E 2 and E resp. Then by (3 )
kDkE
kDkM = kDkM2 = ,
kB 2 kE
and, since d is a linear mapping which keeps the ratio of areas,
k(d)DkE
kDkM = .
k(d)B 2 kE
298 L. Tamassy
Finally, in consequence of (6), we obtain

kDkE
kDkM = 2 = kDkM , D = (d)D.
kB kE
This means that keeps (2-dimensional) area. (It is easy to see that an isometry keeps also
the k-dimensional (1 k n) area.)
According to (4) M is defined by area. Thus, if keeps area, then keeps angle
too. Indeed, we know that M = 2kDkM and M = 2kDkM , where = (d).
Then, from kDkM = k(d)DkM = kDkM ( keeps area) we obtain M = M , that
is keeps angle too.
B) Suppose that keeps area (r) and angle (n). Let us denote (d)I 2 (p) =: Ie2 (p).
2 2
F n determines the indicatrix I (p) = I(p) . We denote (d)1 I (p) =: Ib2 (p) and
Bb 2 A =: D,
b where B b 2 means I b2 and the points in its inside, and A is the domain of the
angle (a, b).
d maps a, b into a b, and the domain A into A.
e, e e Furthermore D e := A eB e 2 and
2 2
e 2 (resp. B ) means I 2
e2 (resp. I ) and the points in its inside.
D := Ae B , where B
1
Moreover (d) takes a e e 2
e, b into a, b, and I (p) into I 2 (p). (Figure 1)
b c
A
I2 C
b
D
C D
b
a
p eb
Ie2
Ib2 D
e
A
e
d
D
a
e
I2 (p) = p
Figure 1.
By our assumption (r) and (n) we obtain

(r) (n) (a)
kDkM = kDk
e
M and kDkM = kDkM = kDk
b M.
Thus
kDkE b M = kDkE = kDE k = kDk
b
= kDkM = kDk b E (7)
kB 2 (p)kE kB 2 (p)kE
for any (a, b).

Let c be a ray in , c I 2 (p) = C, and c Ib = C. b Suppose that there exist p and
c such that C 6= C,b and let us say that C 2
b is outside B . Then, because of the continuity,
p
there exists a ray h(6= c), such that the whole arc Cb H
b (H b is outside B 2 . Now
b := h I)
p
the segment D(c, h) of Bp2 is a proper part of the segment D(c,b h) bounded by c, h and
I.
b Then kD(c, h)kE < kD(c, b h)kE , what contradicts (7). Therefore C = C,b c, p. Thus
2
Bp2 = I 2 (p) = Ib2 . Consequently (d)I 2 (p) = Ie = (d)Ib = I (p), p M . This
yields (5), and so is an isometry.
In Theorem 2 we used the Minkowski angle M (, ) of the tangent vectors
(p), (p)
Tp M . Nevertheless in the case of a Finsler space the Finsler angle F (, ) of
(p, y), (p, y) V(p,y) T M is used in general. Its cos is given in () (at the beginning of
Section 1). It is easy to see that if a diffeomorphism keeps area and d keeps Minkowski
angle, then d keeps Finsler angle too. Namely in this case is an isometry (Theorem 2).
Then the metric tensors g(p, y) of F n and the metric tensor g(p, y), p = (p), y = (d)y
of F n are related by (d)g(p, y) = g(p, y), and this implies that d keeps the Finsler angle
F ((p, y), (p, y)).
Also conversely, if d keeps area and Finsler angle, then it keeps Minkowski angle
too. We know that F n induces by the metric tensor g(p0 , y) on V(p0 ,y) T M a Riemannian
space V n (p0 ), and F n induces by g(p0 , y), p0 = (p0 ), y = (d)y on V(p0 ,y) T M a
V n (p0 ). If d keeps Finsler angle, then d : V n (p0 ) V n (p0 ) is a conformal mapping,
and thus g(p0 , y) = c(d)g(p0 , y), y Tp0 M . But d keeps area. Thus the ellipses
Q(p0 , y) and Q(p0 , y) corresponding to g(p0 , y) and g(p0 , y) have the same Minkowski
area. Hence c = 1. This implies (d)I(p) = I(p), that is is an isometry, and then d
keeps Minkowski angle.
3. Deviation of Finsler Spaces from Riemannian Spaces

There are known several conditions which imply the reduction of a Finsler space F n to
a Riemannian space V n . Such a condition is the vanishing of the Cartan tensor Cijk
or the constantness of the distortion (x, y) [17]. Many other quantities, such as the S-
curvature [16], Landsberg curvature, Cartan torsion, etc. can be coupled with this problem.
Also, a Finsler space is a Riemann space iff the indicatrices are ellipsoids. We want to
present conditions expressed by the Minkowskian angle which imply the reduction of the
indicatrices to ellipsoids and thus the F n to a V n .
We consider a Finsler space F n = (M, F) and its tangent space as a Minkowski space
n
M = (Tp0 M, F(p0 , y)) and a 2-dimensional linear subspace of Tp0 M . Tp0 M can be
indentified with a vector space V n or the coordinate space Rn (x) which can be equipped
with a Euclidean metric, yielding E n (x). B n is the indicatrix body of Mn , B n = I the
indicatrix surface, and I = I 2 is the indicatrix of the M2 induced by Mn on . If F n
is a Riemannian space V n , then M2 is a Euclidean space, and I reduces to an ellipse. In
this case the Minkowskian and the Euclidean angle are the same, M (a, b) = E (a, b),
and it equals iff is a straight angle: its two legs a, b are two half lines of a straight line
300 L. Tamassy
g trough the origin: a b = g. As we have shown in Section 1
M (a, b) = if a b = g, g Tp0 M, p M
is necessary for an F n to be a V n .
Given an arbitrary ray a , let a be the other ray, such that a a is a line g, and let
b be such a ray in that M (a, b) = . b depends on a, and |M (b, a)| =: f (a) 0
is a function of a . f (a) = 0, a is necessary for F n = V n . Let in d0 be
an initial ray, y a point, 0y the ray trough y, r = F(p0 , y), and = M (0y, d0 ). Then
(r, ) is a Minkowskian polar coordinate system in . The arbitrary a is a function of .
Denoting f (a()) by f (), we obtain that
Z 2
G(p, ) := f ()d = 0, f () f (a()), Tp0 M, p M (8)
=0
is necessary for F n = V n . This and Section 1 of this Chapter yield

Proposition 1. (8) is equivalent to: 1) b = a, a, 2) M (a, a) = , a, 3) g bysects
I 2 , 4) I(p) is symmetric, 5) F n is absolutely homogeneous.
All these are necessary for a Finsler space to be Riemannian. G(p, ) 0 measures the
deviation of an F n from being absolutely homogeneous in at p.
We want to obtain sufficient conditions for F n = V n . Our tool for this will be the dif-
ference between Minkowski orthogonality and transversality. Since the properties listed in
the Proposition 1 are necessary, we suppose that the indicatrices of F n = (M, F) are sym-
metric. Let g = aa, h = bb be lines and rays in Tp M , M2p = (Tp M, F(p, y)). Our
considerations will be restricted to . Because of the symmetry of I 2 (p) the Minkowskian
perpendicularities aM b, i.e. M (a, b) = 2 , aM b, aM b, aM b are equiva-
lent. They mean gM h. So, in the case of the symmetry of I 2 (p) we can speak of the
perpendicularity of lines in place of rays. Denoting by g a line perpendicular to g, we
obtain (g ) k g. Another notion is transversality. Let g I 2 (p) = G, G . Then
the tangent TG I 2 (p) =: g is called transversal to g. Because of the symmetry of I 2 (p),
TG I 2 (p) =: (g ) is parallel to g . Also, any line parallel to g is said to be transversal to g.
So we can speak of transversality of a direction to another direction. Nevertheless, this rela-
tion is not symmetrical, that is the direction transversal to g is in general not g : (g ) g.
The relation
(g ) k g, g (9)
means that in M2p the transversality operation is involutive.
A strictly convex, closed, differentiable curve with O in its interior and with the property
(9) is called a Radon curve. Every ellipse is a Radon curve, but not conversely. This
shows that if the indicatrices of an F 2 = (M, F) satisfy (9) at every point p M , then
these indicatrices need not to be ellipses, and thus F 2 needs not to be a Riemannian space
V 2 = (M, g).
We claim that if n > 2 and (9) is satisfied in every I 2 (p), then F n is a V n . Indeed,
under these conditions every I 2 (p) = I(p) , p, is a Radon curve. Then in Tp M
every cylinder osculating to Bpn osculates along a planar curve [19]. In this case, according
to W. Blaschke [4, pp. 157159], every I(p) is an ellipsoid, and thus F n = V n .
If F n = V n , then I 2 (p) is an ellipse, and (9) is satisfied. But (g ) k g is always

true if I 2 (p) is symmetric. Hence in this case g k g . g.
If M (g , g ) = 0, i.e. if g = g , then (9) is realized, for (g ) k g is true. Hence
Z
K(p, ) := |M (g (), g ())|d = 0,
0 (10)
g , Tp0 M, p M
is sufficient for F n = V n . Conversely, (10) is always satisfied in a V n . Thus we obtain
Theorem 3. An absolutely homogeneous Finsler space F n , n > 2 reduces to a Riemann

space if and only if K(p, ) = 0, Tp0 M , p M .
The deviation of an absolutely homogeneous F n from being a Riemannian space on

Tp0 M can be measured by K(). Thus K() can be considered as a kind of sectional
curvature. The deviation at a point p0 M can be measured by the integral
1
Z
G(p0 ) = R K()d 0,
Gn,2 d Gn,2
where Gn,2 is the Grassmann manifold of theR2-dimensional linear subspaces of Tp0 M , and
d is a positive measure on Gn,2 , such that Gn,2 d is finite and invariant with respect to
linear transformations in Tp0 M . The deviation of F n from being Riemannian on M (the
global case) can be measured by the integral
1
Z
H(M ) = R G(x)d 0,
M d M
R
where d is the Finsler volume element, and M d is supposed to be finite.
II. Isometry between Finsler Spaces
Introduction
We consider two Finsler spaces F n = (M, F) and F n = (M , F) with base manifolds M
and M , and with fundamental functions (Finsler metrics) F and F resp. Since fundamental
function F and indicatrices I(p), p M determine each other, we write F n = (M, I) and
F n = (M , I). Let : M M be a diffeomorphism. In this article we want to investigate
the case when is a (global) isometry. Since global isometry induces local isometries, and
in our case ( is a diffeomorphism) local isometry implies global isometry, we will consider
often the local case only.
takes the point P (x) M into the point P (x) M . This is a point transformation.
However, can be considered also as a coordinate transformation (x) (x). After this
coordinate transformation F n = (M (x), I(x)) appears in the form (M (x), I(x)), where
x
I(x) = x I(x). But this coordinate transformation does not change the metric of the
302 L. Tamassy
Finsler space. Now, if we consider x = (x) as the coordinates of a point of another space
M (M P (x) 7 P (x) M ), then we get F n = (M , I), which is clearly isometric to
F n by . This clearly can be reversed. If : M M is an isometry between F n and F n ,
then F n can be considered as F n (on 1 (M ) = M ) in another coordinate system (x).
This gives the following:
Any pair of Finsler spaces F n = (M, I) and F n = (M , I) isometric by : M M
can be represented by a single Finsler space F n in two different coordinate systems: F n =
(M (x), I(x)) and F n = (M (x), I(x)). Conversely, if a Finsler space is represented in
two different coordinate systems, then these representations are isometric.
Nevertheless with this the problem is not settled. Isometries of Finsler spaces have
been investigated from different points of view by a number of geometers. Here we men-
tion only a few of them, as Shaoquiang Deng and Zixin Hou [8] proved that diffeomor-
phisms of a Finsler space F n = (M, F) onto itself which preserve F coincide with the
diffeomorphisms which preserve distance (p, q), p, q M . Other interesting problems of
Finsler isometries are investigated among others by M. Matsumoto [13], L. Kozma and P. I.
Radu [11], Z. I. Szabo [18], Lovas [12], etc.
In this chapter we investigate diffeomorphisms : M M , which yield isometries
between Finsler spaces F and F . We use indicatrices I in place of fundamental functions
F, and prefer direct geometric considerations rather than analytic calculations.
First we point out that if : M M yields an isometry between F n = (M, I) and
n
F = (M , I), then the corresponding indicatrices I(x) and I((x)) are affine equivalent:
I(x) ae I((x)) (i.e. there exists an affine transformation taking I(x) into I((x)). This
is a necessary condition in order that generates an isometric mapping. Fulfilment of this
condition will nearly always be presupposed. In Section 4 we show that the affine equiva-
lence yields a decomposition of M into disjunct subsets M . In Section 5 we show that any
Finsler space over each M is the affine deformation of a locally Minkowski space. (This is
the decomposition of F ). Also we show that at an isometry each M is mapped into a subset
of M with similar property. If M = M , then we obtain motions. Section 6 is devoted to the
study of several motions of Finsler spaces. We consider a connected geodesically complete
F 2 = (M, I), which admits a non-trivial 1-parameter group of continuous motions t , and
in the decomposition of M appear two subsets M1 and M2 consisting of a single point only.
We show that this F 2 is diffeomorphic to the sphere S 2 (and t are rotations). In Sec-
tion 7 we construct to F n and F n two Riemannian spaces V n and V n (V n is arbitrary), and
we show that F n is isometric to F n iff V n is isometric to V n . This reduces the isometry
of Finsler spaces to that of Riemannian spaces. In the final Section 8 we obtain analytical
conditions for the considered isometries, and in case if F n is an affine deformation of a
locally Minkowski space (in this case the assumption I(x) ae I((x)) is satisfied automat-
ically) we investigate the existence of a 1-parameter group t of continuous motions, and
we obtain Killing type equations.
4. Decomposition of the Base Manifold

Let F n = (M F) and F n = (M , F) be two n-dimensional Finsler spaces with fundamental
functions F and F, and base manifold M and M resp. In this chapter Finsler spaces will
be supposed to be regular and positively homogeneous. One can suppose strict convexity
too, but this will not be exploited. We suppose that : M M is a diffeomorphism.
First we give a necessary condition in order that be an isometry. Locally let be
given by xi xi (x), i = 1, 2, . . . , n, where xi and xi are local coordinates on U M
and (U ) = U M resp. Then
F(x, y) = F((x), (d)y) F(x, y), y Tx M, y Tx M . (11)
The indicatrix I(x) of F n at x consists of those y Tx M for which F(x, y) = 1. Then

from (11) we obtain
(d)I(x) = I((x)) = I(x). (12)
d is a (centro) affine transformation a(x, x) taking Tx M into Tx M , i.e. d is a regular
linear transformation, which takes the origo O x of Tx M into the origo O x of Tx M .
The attribute centro relates to this property. In the following affine transformations a
always mean centro-affine transformations. Thus at any isometry
I((x)) = a(x, x)I(x). (13)
If for two configurations C Tx M and C Tx M the relation
a(x, x)y = y C and a1 (x, x)y = y C
holds, for y C and y C, then we say that C and C are affine equivalent: C ae C.
In a Finsler space F n = (M, F) fundamental function F and indicatrices I(p), p M
reciprocally determine each other. So we may write (M, I) in place of (M, F). So we
obtain
Theorem 4. If the diffeormorphism : M M is an isometry between the Finsler spaces
F n = (M, I) and F n = (M , I), then I(p) and I((p)) are affine equivalent:
I(p) ae I((p)). (14)
(14) is a simple, but essential necessary condition of the isometry between F n and
F n . (It means the linearization of the problem.) Nevertheless (14) is far from sufficient
for the isometry. Indicatrices of a Riemannian space V n = (M, g) are ellipsoids (i.e.
special quadrics) Q(x) Tx M given by gik (x)v i v k = 1, v Tx M . So, given two
Riemannian spaces V n = (M, Q) and V n = (M , Q), and a diffeomorphism : M M ;
Q(x) ae Q((x)) is always satisfied, nevertheless two diffeomorphic Riemannian spaces
(over the same base manifold) are not isometric in general.
On the base manifold M of a Finsler space F n = (M, I) (14) is an equivalence relation.
(14) splits M into equivalence classes. Let p1 be an arbitrary point of M . Then we define
M1 := {p M | I(p) ae I(p1 )},

M2 := {p M | I(p) ae I(p2 )}, p2 M, p2
/ M1
M3 := {p M | I(p) ae I(p3 )}, p3 M, p3
/ (M1 M2 )
..
.
304 L. Tamassy
Theorem 5. [
M= M (15)
L
is a decomposition of M . This decomposition is unique, and M L are equivalence

classes.
5. Affine Deformation of a Finsler Space

Tx M can be identified with a real vector space V n (y) consisting of the n-dimensional
vectors y. Let (x) = {y } Tx M be a subset (a configuration) of Tx M V n (y), and
a(x) : Tx M Tx M an affine transformation. Then
a(x) : (x) (x) = {y = a(x)y }.
Let A be a section of the bundle M fibered by GL(n, R). The elements of A can be con-
sidered as affine transformations: a(p) : Tx M Tx M . The components aij (p) of a(p)
are continuous. If the (previous) configurations are the indicatrices I(p) of a Finsler space
F n = (M, I), then
a(p)I(p) =: I(p) (16)
yields new indicatrices, and the Finsler space F n = (M, I) is called the affine deformation
of F n and denoted by AF n .

A Finsler space F n = (M, I) is a locally Minkowski space Mn = (M, I) if for each
point p M there exists a chart U p with a coordinate system (x), such that F(x, y)
is independent of x, or equivalently, the indicatrices I(x) are parallel translates of each
other, (x) being considered as an affine coordinate system. Such a local coordinate system
is called adapted. F n is a Minkowski space, if M = U is Rn or homeomorphic to it.
Lemma. If M admits a locally Minkowski structure, then F n = (M, I) restricted to M

is the affine deformation of a locally Minkowski space Mn (M , I) on M :

F n M = A (Mn (M , I)). (17)
Proof. Given F n = (M, I) and M M , first we construct a locally Minkowski space

Mn (M , I), and then an affine deformation A. We supposed that M admits a locally
Minkowski structure. In this case M is an affine manifold [20, Chap. III. Section 1]. An
affine manifold has a system of local charts U (x ), A covering M , such that the
transition x1 x2 between any two coordinate systems (on U1 U2 ) is linear. Let
us consider such a system of local charts {U (x )}, and let U (x) be one of them. We

endow Tx0 M , x0 U with an indicatrix I(x0 ), which should be the indicatrix I(x0 ) of
F n at x0 . Considering the coordinate system (x) on U as an affine coordinate system, we

define I(x), x U by parallel translation of I(x0 ). Thus we obtain on U a Minkowski

space Mn (U, I), for which (x) is an adapted coordinate system. The fact that the transition

between the different coordinate systems (x ) is linear, makes it possible to extend I from

U to the whole M in such a way that I is independent of x on U , A (see [20,

Chap. III., Theorem 8]). Thus we obtain a locally Minkowski space Mn (M , I), where
every (x ) is an adapted coordinate system.
The affine deformation A is easier to obtain. According to the construction of M
the indicatrices of F n at p0 and at an arbitrary point p M are affine equivalent:
I(p) = m1 (p0 , p)I(p0 ), where m1 (p0 , p) : Tp0 M Tp M is an affine transformation.
Furthermore, since I(p) C on M , also m1 (p0 , p) must be continuous: m1 C .

I(p) arises from I(p0 ) by (finitely many) successive linear transitions from an adapted

coordinate system to the next one. Thus I(p) = m2 (p0 , p)I(p0 ), and thus I(p) =

m1 (p0 , p) m1
2 (p, p0 )I(p) a(p)I(p).
a(p) := m1 (p0 , p) m1
2 (p, p0 )
yields A, which satisfies (17).
A relation similar to (17) is true even if M admits no locally Minkowski structure

(e.g. if M is no manifold), but in this case the consideration is a little more intricate. If
M can be covered by a manifold M f which admits a locally Minkowski structure, then

we construct a locally Minkowski space Mn (M
f , I) starting with I(p0 ) = I(p0 ), p0
M M , and we obtain
f

F n M = [A Mn (M
f , I)] M .
It may happen that M cannot be covered by a manifold M f admitting a locally Min-

kowski structure (e.g. a sphere admits no locally Minkowski structure (see [1, p. 250]).
Then M can be split into (possibly few) parts: M : K M = M , such that M1
M2 = , 1 , 2 K, and each M can be covered by a manifold M f admitting a
locally Minkowski structure. This is certainly so, for M is covered by the union of local
charts U , and any chart admits a locally Minkowski structure. Then M
f = U M .
Denoting by A, we can write
[[ [ [
M
f M
f M = M,

and by the Lemma we obtain:
Theorem 6. Any Finsler space is the union of affine deformations of locally Minkowski
spaces

Fn =
[
A (Mn (M

f , I) M . (18)
A
Proposition 2. Under the condition (C) given below, M of the decomposition (15) is
closed.
Proof. Let us consider a sequence pn of points in M convergent in the topology of M to

a point q M . According to the construction of the M the indicatrices at p0 M and
306 L. Tamassy
at an arbitrary point p M are affine equivalent: I(p) = m1 (p0 , p)I(p0 ). Furthermore,

since I(p) C on M , m1 (p0 , p) must be continuous: m1 C on M . Now we suppose
a little more:
(C) m1 (p0 , p) can be extended continuously to the closure of M .
Hence there exists
lim m1 (p0 , pn ) = m1 (p0 , q).
n
Then from h i
lim I(pn ) = lim m1 (p0 , pn ) I(p0 )
n n
we obtain
I(q) = m1 (p0 , q)I(p0 ),
which means that q M . Then M is closed. This means a certain restriction on the
possible M in the decomposition (15).
According to (15) for the base manifolds of two Finsler spaces F n = (M, I) and
F n = (N, I) we have the decompositions
[ [
a) M = M and b) N = N . (19)
L R
M is determined by a point p M , and N is determined by a point p N .
Theorem 7. If : M N is an isometry and hence a diffeomorphism between F n and

F n , then (M ) = N is a bijective relation between {M } and {N }, and consequently
L = R. Also the Finsler areas kM kF and kN kF equal.
Proof. We show that
(M ) = {(p) | p M } N (20.a)
and
1 (N ) = {1 (p) | p N } M . (20.b)
First we prove (20.a). Let be an isometry, p an arbitrary and p a fixed point of M . Then
(d)I(p) = I((p)) = I(p) and (d)I(p ) = I((p )) = I(p ).
Thus
a) I(p) ae I(p) and b) I(p ) ae I(p ). (21)
According to the construction of M
I(p) ae I(p ). (21.c)
Then
(21.a) (21.c) (21.b)
I((p)) = I(p) ae I(p) ae I(p ) ae I(p ) = I((p )). (21.d)
This yields (20.a).

Conversely, if p, p N , then according to the construction of N
I(p) ae I(p ). With respect to this and (21.a,b), from (21.d) we obtain (21.c), which yields
(20.b).
Since M in (20.a), and later N in (20.b) were arbitrarily chosen sets of the decompo-
sition (19), the above considerations prove the 1 : 1 relation between {M } and {N }.
So (M ) = N is a necessary condition for : M N to be an isometry between
F n = (M, I) and F n = (N, I), but it is not sufficient.
kM kF = kN kF is obvious from (M ) = M , since is an isometry.
Corollary 1. If : M M is a motion on F n = (M, I), then any M from (15) must be

mapped on itself:
(M ) = M , L. (22)
So (22) is a necessary condition for to be a motion in F n .
Corollary 2. If an M consists of a single point p only, then p must be a fix point of any
motion of F n .
The role of the fix points of a motion of a Finsler space is investigated also by L. Kozma
and P. I. Radu [11], and in a most recent work of Shaoquiang Deng.
6. 1-Parameter Continuous Group of Motions

In this section we present a result concerning the 1-parameter continuous group of motions
in Finsler spaces. Theorem 7 has a consequence also in this case
Corollary 3. If t is a 1-parameter continuous group of motions of F n , then any orbit lies

in an M which contains a subset diffeomorphic to the line R1 or to the circle S 1 .
Theorem 8. If t 6= id is a 1-parameter continuous group of motions on a geodesically

complete connected 2-dimensional Finsler space F 2 = (M, I), and in the decomposition
of M according to (15) M1 and M2 consist of p1 resp. p2 only, and the sum r1+ + r2 , resp.
r2+ + r1 of the injectively radii at p1 and p2 is greater than the distance (p1 , p2 ) resp.
(p2 , p1 ) then M is diffeomorphic to the sphere S 2 , and t consists of rotations around
p1 and p2 .
Proof. Let M1 = p1 , M2 = p2 and g(p1 , p2 ) = g(s) be a short geodesic of F 2 in arc

length parameter connecting p1 and p2 . Let g(0) = p1 , g(D) = p2 , and q0 = g(a),
such that a < r1+ and D a < r2 . Then according to our assumption on the injectivity
radii there exist the following two geodesic circles through q0 . The first one is Sp+1 (a) :=
{q M | (p1 , q) = a} the geodesic (forward) circle centered at p1 with radius a, where
(p1 , q) designates the Finsler distance from p1 to q. Sp+1 (a) and its interior is a geodesic
ball Bp+1 (a). The second one is Sp2 (b) := {q M | (q, p2 ) = b}, b = D a the geodesic
(backward) circle centered at p2 with radius b, and Bp2 (b) is the corresponding geodesic
ball. Then q0 is a common point of the two geodesic balls, and these balls have no common
inner point.
308 L. Tamassy
We claim that q0 is no fix point of t . We show that the assumption that q0 is a fix
point of t leads to a contradiction. We know that t (Sp+1 (a)) = Sp+1 (a) globally, for if
q1 Sp+1 (a), then (p1 , q1 ) = (p1 , t (q1 )) implies t (q1 ) Sp+1 (a). If we suppose
that there exist q1 Sp+1 (a) and t such that t (q1 ) = q2 6= q1 , then the arcs q0 q1 and
q0 q2 = t (q0 ) t (q1 ) have different arc length. But this is impossible for a motion t 6= id.
Therefore if q0 is fix, then Sp+1 (a) is pointwise fix at any t . But then t = id on the whole
M too. Let namely p be an arbitrary point of M . Then the (prolonged) geodesic g(p1 , p)
joining p1 and p intersects Sp+1 (a) in a point q3 . p1 and q3 are fix at t . Consequently also
p is fix. Nevertheless we supposed that t 6= id. Thus q0 cannot be a fix point.
Now t (q0 ) = q Sp+1 (a), q 6= q0 and for an appropriate t t (q0 ) may be any point
of the arc q0 q . With an appropriate choice of t and by several repetitions of t any point
of Sp+1 (a) can be reached. But any t keeps the image of q0 also on Sp2 (b). Therefore
Sp+1 (a) = Sp2 (b), which are diffeomorphic to S 1 . Also the geodesic circles Sp+1 (a), a < a,
and Sp2 (b), b < b are diffeomorphic to S 1 , and they fill Bp+1 (a) and Bp2 (b). These geodesic
balls are diffeomorphic to a hemisphere. Their union is diffeomorphic to S 2 , and the orbits
of t on Bp+1 (a) Bp2 (b) B are geodesic circles, that is t is a rotation on B.
Finally we show that
Bp+1 (a) Bp2 (b) = M. (23)
Let again p be an arbitrary point of M , and g(s) in arc length parameter s the short
geodesic connecting p1 = g(0) with p = g(s0 ). If s0 a, then p Bp+1 (a). If s0 > a,
then g(s) crosses Sp+1 (a) = Sp2 (b) at g(a), and goes over to Bp2 (b). If s0 is big enough,
then g(s) crosses again Sp+1 (a) = Sp2 (b), and goes over to Bp+1 (a), and so on. This gives
that any p M is a point of Bp+1 (a) or of Bp2 (b), and this yields (23).
Corollary 4. If in an F 2 = (M, I) in the decomposition of M according to (15) there

are three sets consisting of one point: M1 = p1 , M2 = p2 , M3 = p3 , then F 2 admits no
1-parameter continuous group of motions t (except the trivial case t = id).
It is rather easy to see this. The short proof is omitted.

One can see also that the cut locus of p1 is p2 , and conversely.
rem In case of a Riemannian space V 2 the conditions M1 = p1 and M2 = p2 cannot
be satisfied, for in a V n any two indicatrices Q(p1 ), Q(p2 ) are affine equivalent. However,
Theorem 8 and Corollary 4 remain valid if M1 = p1 , M2 = p2 are replaced by the condition
that t has two fix points p1 and p2 . Also the proof remains unaltered.
In case of n = 3 the situation is different. In an E 3 which is a special F 3 , a rotation
around a line L is a 1-parameter continuous group of motions, which leaves fix the line L,
and thus three of its points p1 , p2 , p3 . rem
In the case of an absolutely homogeneous F 2 Theorem 8 can be related to some other
results. Any geodesic g(s) emanating from p1 is a forward geodesic with respect to p1 until
g(a) = q Sp+1 (a) = Sp2 (b), and it is a backward geodesic with respect to p2 from a until
g(D) = p2 . It is easy to see that the cut locus of p1 is p2 . If we assume additionally that
also the sum of the injectivity radii r2+ + r1 is greater than (p2 , p1 ), then after a consid-
eration analoguos to the previous one we find that the cut locus of p2 is p1 , that is the cut
loci both of p1 and p2 are a unique point. Manifolds with this property were investigated by
several authors. L. Green [9] showed that a compact 2-dimensional Riemannian manifold
V 2 for which the cut locus of any point consists of a unique point only (called wiedersehen
manifold) is isometric to the sphere. In Theorem 8 (beside of the existence of a 1-parameter
continuous group of motions) the one point cut locus property holds for the points p1 , p2
only, and also the consequence is weaker: F 2 is only diffeomorphic to the sphere. Greens
theorem was later generalized to compact V n of even dimension by M. Berger and J. Kaz-
dan [3] and to similar manifolds of odd dimension by C. T. Yang [21]. A survey on these
problems can be found in J. Kazdans paper [10].
7. Isometries between Finsler Spaces

In Theorem 4 we have shown that if a diffeomorphism : M M is an isometry between
two Finsler spaces F n = (M, I) and F n = (M , I), then there exists an A, such that
I((x)) = a(x, (x))I(x), a A or I((x)) ae I(x), (13)
where A : M GL(n, R), M x 7 a(x), and a(x) is a regular n n matrix, which can
be considered as an affine transformation
a(x, (x)) : Tx M T(x) M .
As we pointed out in Section 4, the relation (13) or (14) is a necessary, but far from sufficient
condition in order that be an isometry between F n and F n .
We remark that a may not be unique in (13). Let k(x) : Tx M Tx M be an affine
transformation taking I Tx M globally into itself: kI = I (e.g. if I is the unit sphere
S of the Euclidean space E n , and k is a rotation, then kS = S). Such a k is called affine

automorphism of I. Then for a k =: a

aI = I and {k} =: K GL(n, R)
is a subgroup of the group of all affine transformations. So in place of (13) the more
comprehensive and more appropriate relation is

I((x)) = a(x, (x))I(x). (24)
Now we want to complete (24) to a sufficient condition. Clearly,

a(x, (x)) = (d)(x) (25)
together with (24) is sufficient for to be an isometry between F n and F n . This must be

understood in such a way that there exists a k(x), such that a(x, (x)) = a(x, (x)) k(x)
satisfies (25).
Let V n = (M, g) be an arbitrary Riemannian space on M . Indicatrices of V n are the
ellipsoids gij (x)v i v j = 1, v Tx M . They will be denoted by Q(x). So V n = (M, Q).
We define V n = (M, Q) by

Q((x)) := a(x, (x))Q(x). (26)
310 L. Tamassy

d takes Q(x) into Q((x)) = (d)x Q(x). Thus V n = (M , Q) is isometric to V n by .
If is an isometry between V n and V n too, then also
Q((x)) = (d)x Q(x) (27)
must hold. Then from (26) and (27) we obtain (25). In this case (24) yields the isometry of
F n and F n . Thus the isometry between V n and V n implies the isometry between F n and
n
F .
If M admits a locally Minkowski structure [1, 19] then V n can be chosen
as a locally Euclidean space E n (a space which can be covered by local charts
U (x), U (y), U (z), . . . , such that any transition function y i = y i (x) on U U etc.,
is linear and the metric on each chart is Euclidean). Then V n = E n has vanishing curva-
ture R = 0, and if is an isometry between V n and V n , then also R = 0 on M . Thus in
this case the condition for the isometry between F n and F n (beside of (24)) is R = 0.
The converse is rather simple. We show that if is a Finsler isometry between F n and
F n , then is an isometry between V n and the constructed V n , respectively if V n = E n ,
then R = 0. If is a Finsler isometry between F n and F n , then I(x) = (d)I(x).

This d is an affine transformation a. Thus I = aI. From these we obtain (25): a = d.

V n was arbitrary, and V n was constructed by Q = aQ. From this and (25) we obtain
Q = (d)Q, and this means that V n and V n are isometric by . In the special case of

V n = E n we obtain R = R = 0. We remark that in the reversed considerations I = aI
was no condition. This follows from the assumption that is a Finsler isometry between
F n and F n .
Now we summarize our result. We know that for the isometry of the Finsler spaces
F = (M, I) and F n = (M , I) by the diffeomorphism : M M (24) is a necessary
n
condition. Let V n = (M, Q) be an arbitrary Riemannian space, and V n = (M , Q) given

by (26). Then we obtain
Theorem 9. Under the above conditions F n is isometric to F n by if and only if there

exists an a such that V n is isometric to V n .
This theorem reduces the isometry of Finsler spaces to isometry of Riemannian spaces.

In the only if case (24) is automatially satisfied with a = a (k = id).
If V n is chosen as a locally Euclidean space V n = E n , then the isometry between V n
and V n is equivalent to R = 0, the vanishing of the curvature tensor of V n .
We remark that V n can be replaced by any other Finsler space (M, J). Then V n is

replaced by the Finsler space (M , J), where J := aJ. Such a choice is favorable if the
isometry of (M, J) and (M , J) can easily be decided.
8. Analytical Condition
As we have seen in the previous Section, : M M is an isometry between the Finsler
spaces F n = (M, I) and F n = (M , I) iff (24) and (25) hold. (24) assures the linear
relation between I and I, and (25) has the form
xi
aij (x) = (x). (28)
xj

Here is written in the form xi = xi (x), and aij (x) are the components of a(x, (x)). (28)

has a solution for xi (x) iff aij (x) is a gradient vector of a function f i (x) for any fix i. Thus
we obtain
Theorem 10. : M M is an isometry between F n = (M, I) and F n = (M , I) if and

only if there exists an aij (x) satisfying (24) and (25). To this aij must be a gradient vector
for each i.
In the case if F n is the affine deformation of a locally Minkowski spaces we can say a
little more:
Proposition 3. A) If F n = (M, I) is the affine deformation of a locally Minkowski space:

F n = A[Mn (M, J)], then it can be mapped isometrically only on another Finsler space
F n = (M , I), which is also affine deformation of a locally Minkowski space.
n
B) In order that two Finsler spaces A[ Mn (M, J)] and A [ Mn (M , J)] admit an
isometric mapping : M M it is necessary that any pair J(p) and J(q) of their
indicatrices be affine equivalent.
Proof. A) According to the Introduction of this Chapter to F n = (M, I) =

A[Mn (M, J)] are those spaces isometric which can be represented (locally) as F n =
(M (x), I(x)) in other coordinate systems (x). Mn (M, J) is a locally Minkowski
space in any other coordinate system. Then the isometric images of our F n are again
affine deformations of a locally Minkowski space in the form A(x) Mn (M, (x), J(x))
e x
A( e) = xe
x
A(x).
B) Suppose that : M M is an isometry between F n = A[Mn (M, J)] and
F n = A[Mn [M , J)]. First we remark that in F n the indicatrices J(p), J(q), are affine
equivalent: J(p) ae J(q), p, q M . Namely let us connect p and q by an arc. This
can be covered by finitely many adapted coordinate systems U (), A. Within one
adapted coordinate system, considered as an affine coordinate system, the equation of the
indicatrices is the same. On transition from one adapted coordinate system to another the
transition is linear. The number of the transitions (linear transformations), is finite. De-
noting the product of these linear transformations by L, we obtain J(q) = LJ(p), that is
I(p) ae I(q). The affine deformation A does not hurt this relation. The same is true for
F n = A[Mn (M , J)] too, that is J(p)aeJ(q), p, q M . Since : M M is
an isometry between F n and F n , J(p) ae J((p)). Combination of this relation with the
previous two ones yields that J(p) ae J(q) with arbitrary p M and q M .
From these it follows that if two indicatrices I(p) and J(q) of the above F n and F n are
not affine equivalent, then these affine deformations of locally Minkowski spaces cannot be
isometric.
Now we want to obtain a sufficient analytical condition for the isometry between F n =
A[Mn (M, J)] and F n = A[Mn (M , J)].
Let U (x) and U (y) be local charts on M and M resp, and let the necessary condition
(13) of the isometry between F n = A[Mn (U, J)] and F n = A[ Mn (U , J)] be satisfied,
that is let there exist points x0 U and y0 U such that J(x0 ) ae J(y0 ) or J(y0 ) =
a0 J(x0 ). We give a sufficient condition for the isometry of F n U and F n U . Since
312 L. Tamassy
in a locally Minkowski space indicatrices of any two points are affine equivalent, we have:
J(x0 ) ae J(x) with arbitrary x U , and also J(y) ae J(y0 ) with arbitrary y U . Then
there exist matrices a(x) and a(y) such that J(x) = a(x)J(x0 ) and J(y) = a(y)J(y0 ).
Then
a(y) a0 a1 (x)J(x) = J(y) or m(x, y)J(x) = J(y)
(29)
m(x, y) = a(y) a0 a1 (x).
y
If y = y(x) is an isometry between A[Mn (U, J)] and A[Mn (U , J)], then m = x ,
y
and conversely, from m = x follows the isometry. Hence the sufficient condition for the
isometry (beside the necessary condition (13)) is the solvability of the ODE system
y 0
mij (x, y(x)) = (x) (30)
xj
for y i = y i (x), where mij is given by (29).
Using this result and Theorems 6 and 7 we can say that F n = (M, I) is isometric to
F n = (N, I) iff in the decompositions (19) to each M there exists an N , and for a pair
of points x M , y N I(x)aeI(y) holds, and the PDE system (30) is solvable for
y = y(x), and the solutions match differentiably.
In this result the condition of the isometry is explicitely expressed by PDE systems.
Thus we obtain
Theorem 11. Two Finsler spaces F n = (M, I) and F n = (N, I) are isometric iff for the
corresponding pairs M and N of the decompositions M = M and N = N (see
(19)) the PDE systems (30) are solvable for y = y(x), and the solutions match differen-
tiably.
This result has some application to the group t of 1-parameter continuous motions on
a Finsler space F n = (M, I).
Let t : M R M be a 1-parameter group of transformations on M . t takes a point
x M into another point y M which is determined by x and t : t : x 7 y = y(t, x).
dt takes Tx M into Ty(t,x) M . If F n is an affine deformation of a locally Minkowski space,
which is a special Finsler space: F n = A[Mn (M, I)], then the indicatrices of F n are
affine equivalent to each other. Then I(y(t, x)) = a(x, y(t, x))I(x). If t is a motion, then
dt takes indicatrix into indicatrix. In this case
y y i
a(x, y(t, x)) = (t, x) or aik (x, y(t, x)) = (t, x) (31)
x xk
(c.f. (13)). Also conversely, if (31) holds, then t is a 1-parameter continuous group of
motions. (31) can be considered as a PDE system for y(t, x).
d i
y (t, x) =: v i (t, x) (32)
dt
are the components of the velocity vector v of the motion at y(t, x). Thus
Z t
i
y (t, x) = v i (, x) d,
0
where y i (t, x) is the solution of (31).

Proposition 4. (31) and (32) are a kind of Killing equations for F n = A Mn .
For an arbitrary F n = (M, I) we have the decomposition (18). In order that t be a 1-

parameter continuous group of motions on F n (31) must be satisfied on each subset M of
the decomposition (15). The different solutions on M (resp on M ) must fit differentiably.
References
[1] D. Bao and S. S. Chern, A note on the Gauss-Bonnet therorem for Finsler spaces, Ann.
Math. 143 (1946) 233252.
[2] D. Bao, S. S. Chern and Z. Shen, An Introduction to Riemann-Finsler Geometry

(Springer, New York, 2000).
[3] M. Berger and J. L. Kazdan, A Strum-Liouville inequality with application to an

isoperimetric inequality for volume in terms of injectivity radius, and to wiedersehen
manifolds, In: General Inequalities 2 (Birkhauser, 1980) 367377.
[4] W. Blaschke, Kreis und Kugel (Leipzig, 1916).
[5] T. Q. Binh and L. Tamassy, Deviation of a Finsler space from a Riemannian one,
Tensor N. S. 68 (2007) 4450.
[6] H. Busemann, Intrinsic area, Ann. Math. 48 (1947) 234267.
[7] S. S. Chern, Finsler geometry is just Riemannian geometry without the quadratic re-
striction, Notices Amer. Math. Soc. 43 (1996) 953959.
[8] S. Deng and Z. Hou, The group of isometries of a Finsler space, Pacific J. Math. 207
(2002) 149155.
[9] L. W. Green, Auf Wiedersehensflachen, Ann. of Math. 78 (1963) 289299.
[10] J. L. Kazdan, An isoperimetric inequality and wiedersehen manifolds, In: Seminar on

Diff. Geom. ((S. T. Yau, Ed.) Annals of Math. Studies 102, 1982) 143157.
[11] L. Kozma and P. I. Radu, Weinsteins theorem for Finsler manifolds, J. Math. Kyoto
Univ. 42 (2006) 377382.
[12] R. Lovas, On the Killing vector fields of generalized metrics, SUT J. Math. 40 (2004)
133156.
[13] M. Matsumoto, V -transformations of Finsler spaces I., Definitions, infinitesimal trans-

formations and isometries, J. Math. Kyoto Univ. 12 (1978) 479512.
[14] H. Rund, The Differential Geometry of Finsler Spaces (Springer, 1959).
[15] Z. Shen, Lecture notes on Finsler geometry (Indiana Univ., 1998).
[16] Z. Shen, Differential Geometry of Spray and Finsler Spaces (Kluwer, 2001).
314 L. Tamassy
[17] Z. Shen, Lectures on Finsler Geometry (Singapore, 2001).
[18] Z. I. Szabo, Generalized spaces with many isometries, Geom. Dedicata 11 (1981)
369383.
[19] L. Tamassy, Ein Problem der zweidimensionalen Minkowskischen Geometrie, Ann.

Polon. Math. 9 (1960) 3948.
[20] L. Tamassy, Finsler geometry in the tangent bundle, In: Finsler Geometry (Advanced
Studies in Pure Math., Japan, 48, Sapporo 2005) 168194.
[21] C. T. Yang, Odd-dimensional wiedersehen manifolds are spheres, J. Diff. Geom. 15

(1980) 9196.
Chapter 18
L INEAR C ONNECTIONS A LONG THE TANGENT

B UNDLE P ROJECTION
W. Sarlet
Department of Mathematical Physics and Astronomy
Ghent University, Krijgslaan 281
B-9000 Ghent, Belgium
Abstract
An intrinsic characterization is given of the concept of linear connection along the
tangent bundle projection : T M M . The main observation thereby is that every
such connection D gives rise to a horizontal lift, which is needed to extend the action
of the associated covariant derivative operator to tensor fields along in a meaningful
way. The interplay is discussed between the given D and various related connections,
such as the canonical non-linear connection of the geodesic equations and certain lin-
ear connections on the pullback bundle . This is particularly relevant to understand
similarities and differences between various notions of torsion and curvature. I further
discuss aspects of variationality and metrizability of a linear D along and let me
guide for the selected topics by a very short, old paper of Krupka and Sattarov.
1. Introduction
On the occasion of celebrating a scientists 65th birthday, it is respectable to look back at
the history of the persons involvement in science and it is definitely a good sign if one can
easily detect older work which still raises interesting questions or challenges. I recently laid
my hands on what is in fact a rather minor contribution of Demeter Krupka [10], also one of
the first reprints he gave me personally, and I was astonished to see that, looking at it now,
it confronts me with questions I had not thought of before, even though they are directly
related to my own research of the past 10 to 15 years.
Section 2 in [10] carries connections on the tangent bundle in the title, but is about
maps from a tangent bundle T M into the fibre bundle M M of linear connections on
M , which make the following diagram commutative:

E-mail address: Willy.Sarlet@UGent.be
316 W. Sarlet
M
>

D

?
TM - M

In my opinion, the concept of a connection on a manifold has an unambiguous mean-
ing in the literature, and that is not what the above diagram is about. Instead, a much better
name for the D under consideration here is linear connection along the tangent bundle pro-
jection. The surprising observation for me, however, is that, having been involved in the
development of a comprehensive theory of derivations of forms along the tangent bundle
projection in [15, 16], and having made use of the calculus along in many applications
since then, the idea of such a linear connection along never came up. Needless to say,
other types of connections frequently play a role in my use of the calculus along , such
as what are called linear connections on the pullback bundle : T M T M , so
it becomes intriguing to understand the difference or interplay between all such related,
but different concepts. What is more, it turns out that (to the best of my knowledge) not
much can be found in the literature about linear connections along and what is available
all seems based on (sometimes rather untidy) ad hoc coordinate constructions, i.e. seems
to lack a proper coordinate-free foundation. For example, going back to section 2 of [10]
i (q, v) are connection co-
again, if (q, v) is taken as notation for coordinates on T M and jk
efficients of D, the authors state, as though it should be common knowledge, that geodesics
are curves in M , satisfying the equations
q k + ij
k
(q, q)q i q j = 0, (1)
and that there is a covariant derivative operator for tensor fields along , defined in a
standard manner, which in the case of a metric tensor g along is given by
gij gij s r m m
gij;k = q gim jk gjm ik . (2)
q k q s rk
But is that common knowledge? It seems to me that the term geodesic should be used
only if there is a clear notion of parallel transport first, leading subsequently to geodesics
as auto-parallel curves. This, plus a coordinate-free backing for the standard manner in
which gij;k should be defined, I was unable to find in the literature on which the statements
in [10] must have been based.
Linear connections along were probably introduced for the first time by Hanno Rund,
who called them direction-dependent connections [21]. Unfortunately, what is probably
a comprehensive account of Runds involvement in this theory, seems to have appeared
only in extra chapters of the Russian translation (by Asanov) of his book on Finsler spaces
(see the Math. Review MR0641695 (83i:53097)). So probably, the best source now (for
illiterates in Russian) is Appendix A of Asanovs own book [4]. There, one will find corre-
sponding concepts of torsion, curvature, Bianchi identities, etcetera explained. The torsion
tensor, for example, is defined as having components ij k k , as one might expect, and
ji
Linear Connections Along the Tangent Bundle Projection 317
the components of the curvature tensor are given by

k k
! !
ij ij ilk ilk m r
mvr m v k m
+ ml k
ij mj ilm . (3)
q l v m rl q j v rj
But although an attempt is made to construct such tensors in an intrinsic way, the result is
rather unsatisfactory for several reasons: to begin with, objects which should be regarded
as living along the tangent bundle projection are often subjected to operations (such as an
exterior derivative) which act on the full tangent bundle; the result then is usually not a
tensor along ; this in turn prompts the author to add corrective terms in a rather ad hoc
manner, in order to arrive at a quantity with a proper tensorial meaning. Observe for later
that such corrections always involve derivatives with respect to the fibre coordinates v i
on T M .
My aim is to shed a refreshing light on all such concepts by making use of the cal-
culus along in a systematic way. This will lead to new questions which cannot all be
exhaustively discussed in the course of the present paper. In selecting topics for discussion,
therefore, I will let the further aspects treated in Krupkas paper [10] be my guidance.
It is unfortunate that the literature is full of rather strange terminology for things which
are (closely or not) related to the topics under consideration here. The point is that in all
such cases, the issue is about tensor fields along and operations on them, which have not
been properly identified or recognized as such. In [2], for example, it is mentioned that what
I would call objects along are sometimes called d-objects, or M -objects, or even Finsler
objects (though they have nothing whatsoever to do with Finsler spaces). I dare hope that
the present paper can inspire to more unification in this terminology as well.
2. Elements of the Calculus of Forms Along

Let X ( ) denote the C (T M )-module of vector fields along , which are maps fitting
in a similar commutative diagram as above, with T M replacing M , or equivalently are
sections of the pullback bundle : T M T M . Likewise, k ( ) and V k ( ) will
V
refer to scalar and vector-valued k-forms along respectively.

In coordinates, elements X X ( ) and 1 ( ) are of the form,
V

X = X i (q, v) , = i (q, v) dq i , (4)
q i
while, more generally, an element L V ( ) is of the form

V
L = i with i = ij1 j dq j1 dq j ( ), (5)
q i
where the ij1 j again are functions on T M .

V V
Definition. D : ( ) ( ) is a derivation of degree r if
Vp Vp+r
1. D( ( )) ( )
2. D( + ) = D + D, IR
318 W. Sarlet
Vp
3. D( ) = D + (1)pr D, ( ).
V
A derivation D of degree r of V ( ) has an associated derivation of ( ), also denoted by
D, such that in addition to the above rules: for L V ( ) and p ( ),
V
D( L) = D L + (1)pr DL.
V
For practical purposes, it is of interest to know that every D of ( ) is completely
determined by its action on functions on T M and on basic 1-forms, i.e. 1-forms on M
regarded as 1-forms along by composition with . For an extension to V ( ), it suffices to
specify a consistent action on basic vector fields (vector fields on M ).
The commutator of D1 and D2 (of degree r1 and r2 respectively) is the degree r1 + r2
derivation, defined by
[D1 , D2 ] = D1 D2 (1)r1 r2 D2 D1 , (6)
and satisfies a graded Jacobi identity.

There is a canonically defined vertical exterior derivative dV on V ( ). In the light of
what was said above, it suffices to know that

dVF = Vi (F ) dq i , with Vi = , F C (T M ), (7)
v i

V1
dV = 0 for (M ), dV = 0. (8)
q i
The classification of derivations of forms along and many related issues were discussed
in great detail in [15, 16]. I shall limit myself here to recalling the essentials of this theory
which will be needed further on. The first point to observe is that a classification requires
the availability of a (non-linear or Ehresmann) connection on : T M M . As a matter
of fact, any choice of a connection, giving rise to a local basis of horizontal vector fields

Hi = i
ji (q, v) j , (9)
q v
allows to construct a horizontal exterior derivative dH as follows:
dHF = Hi (F ) dq i , F C (T M ), (10)

V1
dH = d for (M ), d H
i
= Vi (kj ) dq j k . (11)
q q
Inspired by the standard Frolicher and Nijenhuis theory of derivations of (scalar) forms [8],
one is then led to distinguish four types of derivations.
Type i derivations are those which vanish on functions; they are determined by some
L V ( ), written as iL , and defined exactly as in the standard theory. That is to say,
for L V r ( ), 1 ( ),
V
iL (X1 , . . . , Xr ) = (L(X1 , . . . , Xr )),
which extends to a derivation of degree r 1 on scalar forms, and is taken to be zero

also on basic vector fields.
Type dV derivations are those of the form dVL = [iL , dV] for some L.
Likewise, type dH H H
derivations are those of the form dL = [iL , d ].
Finally, the extension to vector-valued forms requires an extra class of derivations,

V
said to be of type a . By definition, these vanish on ( ), they are denoted as aQ
for some Q r ( ) V 1 ( ) and (in view of what has been said before) are further
V
completely determined by the following action on X X ( ):
aQ X(X1 , . . . , Xr ) = Q(X1 , . . . , Xr )(X). (12)
The classification theorem proved in [15] states that every derivation of V ( ) has a unique
representation as the sum of one of each of the above four types of derivations.
The torsion T and curvature R of the non-linear connection we started from, make their
appearance within this theory as vector-valued 2-forms along (as opposed to vertical-
vector-valued semi-basic forms on T M in other approaches, for example). In fact, T and
V
R are uniquely determined by the following commutators on ( ) (extra terms of type a
come in when the same commutators are regarded as derivations on V ( )):
1 H H
[dH, dV] = dVT , 2 [d , d ] = idV R + dVR . (13)
In coordinates,
1 k
T = 2 Tij dq i dq j
, Tijk = Vj (ki ) Vi (kj ), (14)
q k
1 k i j k
R = 2 Rij dq dq q k , Rij = Hj (ki ) Hi (kj ). (15)
A concise formulation of Bianchi identities then is obtained as follows:
dHR = 0, dHT + dVR = 0. (16)
The connection we are using to develop these ideas of course also provides ways to pass
from objects along to objects on the full tangent bundle and vice versa. This really works
both ways: for example, if X = X i (q, v)/q i is any vector field along , we have vertical
and horizontal lifts to vector fields on T M , which in coordinates are given by
X V = X i Vi , X H = X i Hi , (17)
but conversely, every vector field on T M has a unique decomposition into a horizontal and
vertical part and this may reveal new interesting objects along . To see this interplay at
work, consider the brackets of horizontal and vertical lifts on T M . We have
[X V , Y V ] = ([X, Y ]V )V , (18)
H V H V V H
[X , Y ] = (DX Y ) (DY X) , (19)
H H H V
[X , Y ] = ([X, Y ]H ) + R(X, Y ) . (20)
We see, for example, that the decomposition of the bracket of a horizontal and a vertical
lift inevitably leads to the identification of two important derivations of degree zero, the
320 W. Sarlet
horizontal and vertical covariant derivative : DVX and DH

X depend linearly on their vector
field argument, are determined by the following action on functions F C (T M ) and
coordinate vector fields

DVX F = X V (F ), DVX = 0, (21)
q i

DH H
X F = X (F ), DH
X i
= X j Vi (kj ) k , (22)
q q
extend to 1-forms along by the duality rule
DhX, i = hDX, i + hX, Di, (23)
and then further to arbitrary tensor fields along in the usual way. The other two brackets
above identify the curvature tensor R again, plus horizontal and vertical brackets of vector
fields along , which are given by

[X, Y ]V = (X V (Y i ) Y V (X i )) , [X, Y ]H = (X H (Y i ) Y H (X i )) . (24)
q i q i
Note that the vertical bracket satisfies a Jacobi identity, but the horizontal one doesnt,
unless R is zero.
The covariant derivative operators DV and DH in turn define a linear connection on the
pull-back bundle T M T M , as follows: every X (T M ) has its unique decompo-
sition in the form = X H + Y V , with X, Y X ( ), define : X ( ) X ( ) by (see
[14])
= DH V
X + DY . (25)
is said to be a connection of Berwald type. For more insight in the geometric features
of such connections, see [7].
Since acts on vector fields along , one may raise the question: should this be called
a linear connection along ? As explained in the introduction, however, it is only after
looking back at Krupkas old paper [10], that I realized that there is something else which
corresponds better to this terminology, although there should be links with what has just
been recalled. Before entering into the subtleties of this discussion, I need to say a few
words about the special case that the non-linear connection which has been used so far,
is the canonical one associated to a second-order equation field (S ODE). To keep it well
distinguished from the general case, I will do this in a separate section.
3. The Case of a Bsode Connection

As is well known, the tangent bundle T M of a manifold comes equipped with an intrinsi-
cally defined type (1,1) tensor field S, usually called the vertical endomorphism, which has
the coordinate expression S = dq i /v i . A S ODE is characterized by the fact that
S() is the dilation vector field = v i /v i . It has the property (L S)2 = I (the identity
tensor), from which it follows that
1 1
PH = (I L S), PV = (I + L S)
2 2
are complementary projection operators and thus define a non-linear connection. If has
the coordinate representation

= vi + f i (q, v) i , (26)
q i v
the connection coefficients of the S ODE connection are given by
1 f j
ji = . (27)
2 v i
A S ODE connection is characterized by the fact that it has zero torsion T .
Note that there exists a canonical vector field along , namely the identity map on T M ,
which will be denoted by

T = vi i . (28)
q
It is of some interest to point out that for any non-linear connection, TH is a S ODE, let
us call it the associated S ODE, but if the connection we start from is a S ODE connection,
its associated TH will in general not coincide with the original , as is obvious from the
coordinate expressions.
In addition to the machinery developed for arbitrary connections, the case of a S ODE
connection has two very important extra tools to offer: one is the dynamical covariant
derivative , which is a degree 0 derivation, the other is a (1,1) tensor V 1 ( ), called
the Jacobi endomorphism. They are forced upon us, for example, via the same sort of
interplay between the calculus along and standard calculus on T M , by looking at the
decomposition of L X H . Indeed, it turns out that the vertical part in this decomposition
depends tensorially on X, while the horizontal part identifies a derivation. So, we can write,
L X H = (X)H + (X)V , (29)
which defines and on X ( ). is further determined by the duality rule (23) and the
fact that F = (F ) on functions F C (T M ). For computational purposes:

= ki , (dq i ) = ik dq k , (30)
q i q k
and
f i
= ij dq j , with ij = kj ik (ij ). (31)
q i q j
The relevance of and is already obvious from the properties:
dV = 3R, dH = R. (32)
Since variationality will be one of the topics under discussion later on, I conclude this
section with the very concise formulation of the so-called Helmholtz conditions within this
geometric approach: the necessary and sufficient conditions for the existence of a (regular)
322 W. Sarlet
Lagrangian formulation of a given S ODE are the existence of a non-degenerate, symmetric

type (0,2) tensor field g along , satisfying the requirements [16]:
g = 0, (33)
V V
DX g(Y, Z) = DZ g(Y, X), (34)
g(X, Y ) = g(Y, X). (35)
To close the sections about the main ingredients of the calculus along and its relevance
in the study of second-order dynamics, I should say that the calculus along is being used
systematically also in Szilasis magnum opus [23].
4. Linear Connections Along

Let us go back now to Runds direction-dependent connections, i.e. maps D : T M M
fitting in the commutative diagram of the introduction. My first aim is to find a coordinate-
free justification for the equations (1), as geodesic equations, and for (2) as defining relation
of the covariant derivative of a metric along .
The fibre of M at m M consist of maps
D : Tm M Xm Tm M,
where Xm denotes the module of vector fields on M defined in a neighbourhood of m,

which have the properties
Dvm Y = Dvm Y, IR
Dvm (f Y ) = f (m)Dvm Y + vm (f )Y (m), f C (M )
plus linearity with respect to the sum in both arguments. So for each wm Tm M , D(wm )
k on T M , such that
is such a map, and is locally defined by functions ij

k
D(wm ) = ij (wm ) k . (36)
q i
q j q m
m
For an alternative view, given a D in the above sense, define for all X X ( ), and
Y X (M ), a map
D : X ( ) X (M ) X ( ),
by
(DX Y )(wm ) = D(wm )Xwm Y. (37)
By construction, DX Y will be IR-linear in both arguments and further satisfies
DF X Y = F DX Y, F C (T M ) (38)

DX (f Y ) = f DX Y + X(f ) Y, f C (M ). (39)
Proposition 1. A linear connection D along is a map D : X ( ) X (M ) X ( ) which

is IR-linear in both arguments and has the properties (38, 39).
Proof. Indeed, conversely, using the standard trick with a bump function, the above prop-
erties imply that the value of DX Y at a point wm Tm M only depends on the value of X
at wm . As a result, it makes sense to define a map D : T M M by
D(wm )vm Y = (DX Y )(wm ),
where X is any vector field along such that Xwm = vm , and this provides a linear
connection along , in the sense of the commutative diagram we started from.
To extend D further as covariant derivative operator, we need to extend the second

argument to X ( ), which in turn requires some notion of horizontal lift for the action on
functions on T M .
Given a curve : t 7 q i (t) in M , take a vector field along , i.e. a curve : t 7
(q (t), i (t)) in T M which projects onto , and define another vector field D along by
i
(D )(t) = D((t))(t) Y. (40)
Here, Y is any vector field, defined in a neighbourhood of (t), such that Y ((t)) = (t).
In coordinates:

k k
D (t) = (t) + ij ((t))q i (t) j (t) , (41)
q k (t)
where
Y k
(t) = (q(t), q(t)) and k (t) = (q(t))q i (t).
q i
We can now come to a notion of parallel transport in the usual way: is said to be parallel
along if D (t) = 0 for all t. As in the standard theory, for a given curve in M and an
arbitrary point v in the fibre of (t0 ) say, there is a unique along , which passes through
v and is parallel. This is called the horizontal lift of (through v): = h . Note,
however, that the differential equations to be solved for the i are non-linear here.
Definition. A curve in M is said to be a geodesic of the linear connection D along if

h = , i.e.
D = 0. (42)
From (41), it is clear that in coordinates, t 7 q i (t) is a geodesic if it satisfies the S ODE
equations
q k + ij
k
(q, q)q i q j = 0,
which are indeed the equations (1). This resolves our first query.
In order to obtain a covariant derivative operator, acting on tensorial objects along , it
suffices now to extend the horizontal lift construction to vector fields.
Definition. The horizontal lift of X X (M ) is the vector field X h on T M , which projects

onto X and is further determined by the requirement that its integral curves are horizontal
lifts of integral curves of X.
324 W. Sarlet
It follows that, in coordinates:

put
X h = X i (q) i
k
ij (q, v)v j k = X i hi , (43)
q v
and this horizontal lift naturally extends to X ( ).
We can then, in the first place, extend the action of D to an operator
D : X ( ) X ( ) X ( ), (44)
by putting
DX F = X h (F ) X X ( ), F C (T M ), (45)
and
DX (F Y ) = F DX Y + X h (F )Y Y X (M ), F C (T M ). (46)
Finally, DX further extends to 1-forms along by duality, and subsequently to arbitrary
tensor fields along . In particular, for g T20 ( ), an intrinsic definition of DX g becomes:
(DX g)(Y, Z) = X h (g(Y, Z)) g(DX Y, Z) g(Y, DX Z). (47)
This justifies the covariant derivative formula (2) of Rund (as found in [4] and [10], for
example), except for a difference in convention! Indeed, the coordinate expression of (47)
reads
gij gij s r m m
gij|k = (D/qk g)ij = v ki gjm kj gim , (48)
q k v s kr
and has a different order for the bottom indices of the connection coefficients, an issue of
course which depends on the convention adopted at the very start, namely with the defining
relations (36).
5. A Variety of Related Connections

We have seen in the previous section that a linear connection D along comes with an as-
sociated notion of horizontal lift, as it should, and defines a S ODE, namely the equations for
geodesics. But that inevitably means that there are a number of other connections around.
Studying the relationship between those connections is a somewhat slippery domain, be-
cause it is easy to get dragged away into identifying all sorts of tensors, which are perhaps
only marginally interesting. I shall try to limit myself here to what seem to me the bare
essentials of such a discussion.
The horizontal lift (43) was indispensable to arrive at the covariant derivative ope-
rator DX . It defines at the same time a non-linear connection, however, with connection
coefficients
h k k j
i = ij v , (49)
and that in turn, in agreement with the general formulas (22), defines a horizontal covariant
derivative DhX . Since DX and DhX agree on functions, their difference, when acting on
some Y X ( ), depends tensorially on Y . Hence, every linear connection D along has
an associated type (1,2) tensor field along , which I shall denote by K, and seems not to
have been noticed or highlighted before in the literature.
Definition. The fundamental tensor field K of a linear connection D along is the type
(1, 2) tensor along determined by
K(X, Y ) = DX Y DhX Y. (50)
In coordinates
ilk l h k
K = Kkij dq i dq j , Kkij = v = k
i + ij . (51)
q k v j v j
In terms of the different types of derivations, discussed in section 2, we can write that DX =
DhX + aX K for the action on X ( ) (with X K(Y ) = K(X, Y )). By the duality rule (23),
this implies that for the action on 1-forms 1 ( ), we will have DX = DhX iX K .
V
It follows that for the action on arbitrary tensor fields,
DX = DhX + X K , (52)
where for an arbitrary (1,1) tensor A, A = aA iA (see [16]).

There is, of course, also the S ODE connection associated with the geodesic equations
(1) coming from D. Its connection coefficients, according to (27), are given by
k
lm
H
ki = 21 (ilk + lik )v l + 1
2 vl vm, (53)
v i
and it has its own horizontal covariant derivative DHX . From now on, I shall systematically
use superscripts h (or subscripts h ) for everything that relates to the derivative Dh , while
H
(or H ) will refer to the canonical connection of the geodesic S ODE . The difference
between the two horizontal distributions determines a type (1,1) tensor, for which I will
not introduce a separate notation because it is derived from more fundamental tensors; its
components are given by
k
lm
H
ki h ki = 21 (lik ilk )v l + 1
2 vl vm. (54)
v i
The second term on the right is easily seen to be 21 (T K)ki , while the first term comes
from the torsion tensor of D. Indeed, as will be seen in the next section, one can give an
intrinsic definition of the concept of torsion of D, which then is found to have the coordinate
representation
1D k
D
T = 2 T ij dq i dq j , D
T kij = ij
k k
ji . (55)
q k
So the first term on the right in (54) is 12 (iT D T )ki .

The subtle interplay between the different connections which are around can be seen
from the following properties. From the general discussion in section 3, it should not come
as a surprise that TH is not the geodesic S ODE from which the horizontal lift H is derived.
But we do have that Th = . Recall further that an important degree zero derivation
associated to a S ODE is the dynamical covariant derivative . But by choosing the vector
field X in the different covariant derivative operators we have so far considered to be the
326 W. Sarlet
canonical T, there are, so to speak, three more degree zero derivations available which
should have some similarity to , namely DT , DhT and DH T . The last one is quite different
from the others, in general. Since Th = , , DT and DhT on the other hand all coincide
on functions. Yet, they are not the same since, for example,

i
= H ki , DT = v l lik k , (56)
q q k q i q
while, in accordance with (52), the difference between DT and DhT is determined by T K.
An interesting property of the linear D along is that
DX T = 0, X X ( ). (57)
This can easily be verified in coordinates. It follows that
DhX T = K(X, T). (58)
In particular, we have DT T = 0, a property which is in general not shared by the related

derivations , DhT and DH T.
A different aspect which needs some inspection in this sec-
tion is the relationship between linear connections along
and linear connections on the pullback bundle : T M
T M . This relationship is not a very strict one a priori as, again, there are many
different elements one can bring into the picture. But I shall try to develop arguments
which bring us to a kind of natural correspondence in the end. At the start of such a
discussion, however, one has to make a clear distinction between the defining requirements
of a linear connection along , as expressed for example by Proposition 1, and the extension
to a map D : X ( ) X ( ) X ( ) which was needed to arrive at a covariant derivative
operator on tensors along .
The most immediate association between both concepts one can think of, goes as fol-
lows. Let D be a linear connection along , locally determined by
k
D/qi = ij (q, v) k .
q j q
Define for each X (T M ) a map : X ( ) X ( ) by
k
/qi j
= ij , /vi = 0, F = F , (59)
q q k q j
(F X) = F X + (F ) X, F C (T M ), X X ( ). (60)
It is easy to see that satisfies the requirements for a linear connection on , but this
association calls for more intrinsic procedures and insights.
It seems to me, however, that in view of what precedes, it is appropriate to bring first the
availability of an extra horizontal distribution in the discussion. This way, we can make a
link also with yet another construction in the literature. Indeed, a pair ( , PH ), consisting
of a linear connection on and a horizontal projector on T M is essentially (possibly
after identification of the pullback bundle with the bundle of vertical tangent vectors to
T M ) what is called a Finsler connection by Matsumoto [17] and in Bejancus book [5],
for example, and indeed in many other sources.
Suppose we have such a pair ( , PH ), where PH is general here (i.e. not necessarily
the S ODE connection of the geodesics), so that every has its decomposition = X H +Y V
for some X, Y X ( ). Then, for each X X ( ) we can define DX : X (M ) X ( ) as
DX = X H |X (M ) .
It is clear that DX has all the right properties (Proposition 1), and if we put

/qi j
= kij k ,
q q
k = k . Observe, however, that the associated h -lift
the connection coefficients of D are ij ij
of D is not the H we started from. Note also that X V |X (M ) defines a tensorial object
which we disregard in this construction.
Conversely, suppose the data are a linear D and an arbitrary horizontal lift H (not nec-
essarily the h associated to D). Then, a corresponding can be constructed as follows:
for X X ( ), Z X (M ) and F C (T M ), put
X H Z = DX Z, X V Z = 0,
(F Z) = F Z + (F ) Z.
The property X V Z = 0 expresses that acts on any fibre of T M by so-called com-

plete parallelism (see [7]). The element of caution to mention here is that X H (F Z) 6=
DX (F Z).
Now, it is clear that in both of the directions of the above construction, one of the data
in fact can imply the availability of the other, so let us finally reduce the data again in
that sense. This means that in the first construction, we now assume that only a horizontal
projector is given. Then, there is a naturally associated linear connection on , namely the
Berwald type connection (25). The linear D along which then follows is determined by
DX = DH
X |X (M ) ,
and if ki are the connection coefficients of the given non-linear connection, we have: ij
k =
ki /v j . Again, one has to be careful, because the extension of this DX to X ( ), as

discussed in the previous section is not the DH X we started from.
For the converse construction, it suffices to have D as only data, because D comes
with its own horizontal lift h as discussed before. The above general construction did not
depend on the choice of a horizontal lift anyway, so we can carry it out just as well with h .
The connection coefficients of are given by: kij = ij k (and zero), i.e. we recover the
direct association we mentioned at the beginning, but in a more elegant way. This time,
we do have that X h (F Z) = DX (F Z) for F C (T M ). However, the resulting
is generally not of Berwald type, which is essentially due to the fact that DX 6= DhX , i.e.
to the fundamental tensor K. Nevertheless, this association between D and is the most
relevant point in our discussion, and we formalize it, therefore, in the following statement.
328 W. Sarlet
Proposition 2. Let D be a linear connection along , then there is a linear connection

on , which is uniquely determined by the following prescriptions: for X X ( ),
Z X (M ) and F C (T M ),
X h Z = DX Z, X V Z = 0, (61)
(F Z) = F Z + (F ) Z, (62)
where h is the horizontal lift determined by D.
We can now give an interesting alternative interpretation of the fundamental tensor K

of D. The point is the following: there are five tensors along which can be associated to
a pair like ( , Ph ). They can be regarded as torsions of and were called A, R, B, P
and S in [7]. The term torsion can be justified by the fact that any pair of a linear on
and a horizontal distribution on T M induces a linear connection on T (T M ) also, whose
torsion has five components which (as by now familiar) are lifts of tensor fields along (see
also [22] and [5] for a discussion about these five torsion tensors). Now, B = S = 0 here
(as a result of the second condition in (61)), and P, in particular is defined by
P(X, Y ) = X h Y DhX Y. (63)
It follows from the first of (61) that P = K. The two other tensors A and R will be
encountered in the next section. Incidentally, the association between DX and X h in
(61) is a kind of generalization of what are called h-basic covariant derivative operators by
Szilasi [23].
The most obvious conclusion one can draw at the end of this section is that one should
be extremely careful in comparing or using different types of connections which are around
in this area!
6. Torsion and Curvature of the Linear D Along

There are different ways of approaching the concept of torsion of a connection. For a direct
definition of the torsion tensor, one needs a bracket of vector fields. In the case of a linear
connection D along , since D induces a horizontal distribution, it looks natural to think of
the associated horizontal bracket, as defined in (24).
Definition. The torsion D T of a linear connection D along is the vector-valued 2-form

along , defined by
D
T (X, Y ) = DX Y DY X [X, Y ]h , X, Y X ( ). (64)
It is easy to verify that D T is indeed a tensor. In fact, we have (see [16]),
[X, Y ]h = DhX Y DhY X h T (X, Y ), (65)
where h T is the torsion of the non-linear connection h , which according to (14, 49) has
components
h k
Tij = Vj (h ki ) Vi (h kj ), h k
i = ilk v l . (66)
This implies that

D
T (X, Y ) = K(X, Y ) K(Y, X) + h T (X, Y ). (67)
Evaluating this expression in coordinates, all derivatives of the ij k cancel out, and one
indeed obtains the previously cited formula (55), which agrees with the expression given
by Asanov [4]. Note further that, in terms of the association expressed by Proposition 2,
D
T is in fact the A-torsion of the linear connection on .
As for curvature, it is best first of all to make a notational distinction: I shall use the
notation curv when talking about curvature of any sort of linear connection and R as in
section 2 for the curvature tensor of a non-linear connection. The natural definition of
curvature of a linear D along would seem to be

D
curv(X, Y )Z = DX DY DY DX D[X,Y ]h Z. (68)
Observe, however, that this is tensorial in X, Y , but not in Z, unless the Z-argument is
restricted to X (M ). With F C (T M ), it follows from (20) that
V
curv(X, Y )(F Z) = F (Dcurv)(X, Y )Z + (hR(X, Y )) (F ) Z,
D
(69)
where hR is the curvature of the non-linear connection associated to the h -lift. In coordi-
nates, with hi referring to the local basis of horizontal vector fields in (43), one readily finds
that
(Dcurv)m m m l m l m
ijk = hi (jk ) hj (ik ) + jk il ik jl . (70)
These are effectively (with a transposition of the lower indices in the connection coeffi-
cients, as reported before) the curvature components (3) mentioned in [4]. One should
keep in mind, however, that they are, for the moment at least, components of a map
D
curv : X ( ) X ( ) X (M ) X ( ). An interesting side observation is that
h i
Rjk = hk (h ij ) hj (h ik ) = (Dcurv)ikjl v l . (71)
Note also that hR is, up to a sign, the so-called R-torsion of the associated linear connection
on , determined by Proposition 2.
At this point, it is worth referring to the comments in the introduction which follow
equation (3), and to illustrate again now that there is a marked advantage in conceiving
all objects and operations of interest as living along (as opposed to mixing calculations
along with calculations on T M ). Indeed, the property (69) provides the clue to remedy
the non-tensorial aspect of Dcurv.
Definition. The curvature of a linear connection D along is the type (1, 1) tensor-valued
2-form Dc]
urv along , defined by

D
urv(X, Y )Z = DX DY DY DX D[X,Y ]h Z DV(hR)(X,Y ) Z.
c] (72)
The components of this tensor, which are obtained by taking coordinate vector fields
for the arguments X, Y, Z, are the same as (70), since the extra term does not contribute
to this computation. Note again that, as with the torsion (or in fact the covariant derivative
330 W. Sarlet
DX itself) the non-linear h -connection is needed to define the curvature of D. It is clear

then that one can define a related tensor field of curvature type as follows:

h
curv(X, Y )Z = DhX DhY DhY DhX Dh[X,Y ]h Z DV(hR)(X,Y ) Z. (73)
One can verify that the relation between both curvature tensors is given by
D
urv(X, Y )Z = hcurv(X, Y )Z + DhX K (Y, Z) DhY K (X, Z)
c]
+ K(h T (X, Y ), Z) + K(X, K(Y, Z)) K(Y, K(X, Z)), (74)
or equivalently
D
urv(X, Y )Z = hcurv(X, Y )Z + DX K (Y, Z) DY K (X, Z)
c]
+ K(D T (X, Y ), Z) K(X, K(Y, Z)) + K(Y, K(X, Z)). (75)
Concerning the geometrical interpretation of the tensor hcurv, it is worth observing that this
is in fact the horizontal component of the curvature of the Berwald-type connection on ,
associated to the h -lift. Indeed, taking the general formula (20) for brackets of horizontal
lifts into account, and denoting the Berwald-type covariant derivative by as in (25), it is
easy to see that

h
curv(X, Y )Z = X h Y h Y h X h [X h ,Y h ] Z. (76)
Note finally that the property (71) has the following intrinsic content:
D
urv(X, Y )T = hR(X, Y ).
c] (77)
I now briefly turn to the issue of Bianchi identities. In fact, with the insights we have
gained now, we should not expect new features to appear here, because torsion and curvature
of D are after all closely related to h T and hR, through (67) and (71) for example, and
Bianchi identities for these tensors are known to be compactly represented by properties of
the form (16). What looks appealing in this respect, however, is to explore the meaning of
an exterior derivative associated to D.
For the horizontal exterior derivative associated to the h -lift, we know [16] that on 1-
forms 1 ( ),
V
dh(X, Y ) = DhX (Y ) DhY (X) + (h T (X, Y )). (78)
By analogy, therefore, it looks recommended to define the exterior derivative dD on func-

tions F C (T M ) by dDF (X) = DX F = X h (F ) and on 1-forms along by
dD(X, Y ) = DX (Y ) DY (X) + (D T (X, Y )). (79)
But then, in view of (50) and (67), it follows that dD = dh on scalar forms. For the action on
vector fields along we have dhX(Y ) = DhY X, hence by analogy put dDX(Y ) = DY X.
This implies that (dDX dhX)(Y ) = K(Y, X), meaning that for the full action on V ( ),
we have
dD = dh + aK , (80)
V1
where K, in agreement with (12) is regarded here as a tensor in ( ) V 1 ( ). However,
since there is no difference between dD and dh on scalar forms, commutator properties of
dD of type (13), which could be regarded as defining torsion and curvature, will actually
reproduce h T and hR. As a result, the Bianchi identities essentially remain (see (16))
dh (hR) = 0, dh (h T )) + dV(hR) = 0, (81)
and can be re-formulated, using (80), (77) and (67) in terms of corresponding objects related
to D if needed.
7. Variationality versus Metrizability

The two questions to be addressed here in fact have very little in common, but that is not
always so clear in the literature. In my opinion, metrizability is a property that can be
attributed to a connection, while variationality, in the context of a connection, can only
be attributed to its geodesic equations, and as such is common to an equivalence class of
connections, namely those which have the same geodesics.
A linear D along is variational if its geodesic equations (1) are variational. The
point to make is that studying this problem subsequently has very little to do with D any-
more; instead, the geometric tools of the S ODE connection of the geodesics now enter the
scene. And the most comprehensive formulation of the problem is simply the existence of
a (non-singular) symmetric g along , satisfying the Helmholtz conditions (33, 34, 35). All
operations of interest in this problem, such as the dynamical covariant derivative and ,
have coordinate expressions in terms of the forces f i of the S ODE , which are given by
f i (q, v) = jk
i
(q, v) v j v k . (82)
Hence, as observed in [12], nothing will change if we consider different D, i.e. different
i which produce the same f i . In fact, variationality of a given D was defined in [10] also
jk
i , possibly different from the given ones but giving rise to
as the existence of some set of jk
the same f i , such that the Helmholtz conditions are satisfied.
Reference [12] contains another statement which is worth situating within our present
analysis. As we discussed in section 5, there is a certain similarity between the dynamical
covariant derivative of the geodesic S ODE and the operator DT (see (56) for example),
so it is of some interest to investigate to what extent g = 0 differs from DT g = 0.
The answer to this question is the result (8.14) in [12] which, translated into our present
notations, states that, provided the linear D along is taken to be torsion free,
!
1 k k
g = 0 DT g = gik lm j
+ gjk lm vl vm. (83)
2 v v i
Expressed differently, and in more intrinsic terms, we can say that for a torsion-free D,
(g = 0 DT g = 0) if and only if T K g = 0. (84)
In fact, in view of (52), we then also have DhT g = 0.

332 W. Sarlet
A discussion of metrizability is more delicate, because here I differ in opinion with

some of the references cited so far. In my view, a linear D along is metrizable if there
exists a (non-singular) symmetric g along , such that DX g = 0 for all X X ( ) (without
any further assumptions on g). The requirements on g here are directly related to the given
connection and are quite different from g = 0, for example. In other words, generally
speaking, it doesnt make much sense to expect that metrizability might imply variationality
or vice versa, unless of course you modify your definition until that works.
There are roughly two aspects one can consider: one is the strict and hard question of
studying for a given fixed D the existence of a suitable g; the other and much more tangible
question is to start from a given g and try to modify the connection to make it metric with
respect to this g. I will argue, however, that even the second question does not have an
entirely compatible solution in the case of a linear connection along .
Inspired by the concept of Cartan tensor which one can find, for example, in the work
of Miron and co-workers (see e.g. [20] and [2]), the following looks like a natural concept
to introduce here.
Definition. The Cartan tensor associated to a given non-singular, symmetric g and a linear
connection D along , is the symmetric type (1, 2) tensor CD along , determined by
g(CD (X, Y ), Z) = DX g (Y, Z) + DY g (X, Z) DZ g (X, Y ), X, Y, Z X ( ). (85)
The coordinate expression of CD is found to be
CD kij = g kl hi (gjl ) + hj (gil ) hl (gij ) (ij

k k

+ ji )
+ g kl gjm (lim ilm ) + gim (ljm jl
m

), (86)
and simplifies somewhat for a torsion-free connection.
Proposition
3. We have
(i) DX + 12 X CD g = 0, X X ( ),
(ii) DX g = 0, X X ( ) CD = 0.
Proof. The first property is a straightforward computation: we have

DX + 12 X 1

CD g (Y, Z) = DX g (Y, Z) 2 g(CD (X, Y ), Z) + g(CD (X, Z), Y ) ,
from which the result immediately follows by using (85). Obviously then, if CD = 0 it
follows that DX g = 0, X, while the converse trivially follows from the definition of
CD .
So, metrizability of a given D is (as usual) a matter of a vanishing Cartan tensor, in other
words, the hard question then is to study under what circumstances, with given ij k (q, v),
the equations glk CD kij = 0 can have a solution for g.

It would seem that the statement (i) in the above proposition contains the (expected)
answer about how to modify the given connection to make it metric with respect to g.
There is a technical problem, however, which makes that this is not quite true! To see this,
lets go back to the original concept of a linear connection along , in the interpretation of
Proposition 1. This shows that the difference between two such connections is a type (1,2)
tensorial object indeed, C say, but in the following sense:
C : X ( ) X (M ) X ( ).
Hence, starting from a DX , putting DX := D +

X X C defines a new linear connection
along , but as soon as one wants to extend the action of DX to X ( ) by the rules (45, 46),

there is a certain incompatibility, because the horizontal lift h induced by D is different
from the original h -lift, while the defining relation of DX , without modification, would
h
imply that on functions F C (T M ): DX F = X F . As was the case in discussing the

notion of curvature (see (72)), one has to correct with a vertical derivative term (which does
not modify the connection coefficients) to remedy this deficiency, and the clue on how to
do this here comes from the property (57) which every properly extended linear connection
along should have.
Proposition 4. If C is an arbitrary type (1, 2) tensor along , then the following modifi-
cation of a given D defines a new linear connection along which is compatible with its
induced horizontal lift:

DX = DX + X C DVC(X,T) . (87)
Proof. Recall that for Y X ( ), we have: X C Y = C(X, Y ) and DVY T = Y . It easily

follows that DX T = 0 implies DX T = 0, and one can verify in coordinates that this is the
same as saying that for F C (T M ), DX

F = X h (F ) where k = k + C k .
ij ij ij
Coming back to the subject of Proposition 3 now, we see that the modified covariant
derivative operator of the first statement necessarily has to be taken in its extended sense,
since it acts on g (not just on basic vector fields). Therefore, the genuine modified linear
connection D e which is at stake here, reads
e X = DX + 1
D 21 DVCD (X,T) . (88)
2 X CD
Unfortunately, however, the conclusion then is that we dont have D

e X g = 0, X, but rather
e X g + 1 DV
D 2 CD (X,T) g = 0. (89)
One might consider to cover this technicality by defining a linear D along to be metrical
with respect to some g, if for all X X ( ), DX g = 0 modulo vertical derivatives of g.
It is interesting to look at the preceding technical problem still from a different per-
spective. By Proposition 2, we know how to associate with D a linear connection on the
pullback bundle . In turn, as was done (in a time-dependent set-up) in [18], for example,
one can then define a horizontal Cartan tensor in that context by
g(Ch (X, Y ), Z) = X h g (Y, Z)+Y h g (X, Z)Z h g (X, Y ), X, Y, Z X ( ). (90)
Actually Ch = CD , but we can pose the problem of constructing a modified metrical con-
nection at this level without any complications. Indeed,
e h = h+1
(91)
X X 2 X Ch
334 W. Sarlet
will have the property that e h g = 0, for all X. Since it is only the horizontal component
X
of which becomes metrical in this process, one could call this a connection of Chern-
e
Rund type . The problem we have encountered before comes from the fact (explained in
detail in section 5) that in going back from e h to a linear D
X
e X along , the horizontal lift
induced by this D is not the one we started from.
e
For completeness, coming back to the point I made at the very beginning of the section,
I should mention the special case of a standard linear connection on M , where the con-
nection coefficients do not depend on the v i and the geodesics come from a spray. Then,
the requirement that for some (quasi-Riemannian) metric g we have gij|k = 0 (for all k) ,
obviously is equivalent to requiring that gij|k v k = 0, in other words, in that situation we
have
DX g = 0, X DT g = g = 0.
That is why variationality (where g = 0 is a key condition) and metrizability (which is
about DX g = 0, X) are closely related then. By the way, it looks like an interesting ques-
tion to investigate, for the general case of a linear D along , is under what circumstances
DT g = 0 will imply DX g = 0, X X ( ) (some form of homogeneity is probably
indispensable for that).
There is a final link I should explain to conclude this discussion. A very recent paper
[6] carries the title Metric nonlinear connections. So what is this about? The author takes
a S ODE , plus a non-linear connection with connection coefficients Nji say, to build a
covariant derivative operator, say, by putting (I am identifying vertical tangent vectors to
T M with vectors along ):

F = (F ), F C (T M ), = Nij j . (92)
q i q
Here, the non-linear connection may or may not be the canonical one coming from (al-
though I dont see the point really in taking a different one to start with). Anyhow, it is clear
that this operator is of the type of a dynamical covariant derivative, and it can be of inter-
est, of course, to study compatibility of with some metric g, in the sense that g = 0.
Whether it is appropriate to classify this question under metrizability problems is perhaps
debatable here, if a dynamical covariant derivative is all one has. The main problem which
is addressed in section 2 of [6] is to construct from a new such that g = 0, where
g is a given metric along . The solution to this problem is in fact quite simple: taking
and to be identical on functions, they must be related by a formula of the form:
= + A ,
for some (1, 1) tensor A, which must be chosen in such a way that
g(AX, Y ) + g(X, AY ) = g(X, Y ), X, Y.
It is clear that a solution for A is given by
Aij = 21 g il (g)lj ,
which means that the modified Nji are determined by
gli Nji = 12 (g)lj + gli Nji = 12 (glj ) + 21 (gli Nji gji Nli ).
In the case that the Nji we start from are the canonical ij (see (27)), so that is precisely
the dynamical covariant derivative of the S ODE , the above relation is exactly equation
(12) in [6], and thus explains Theorem 2.2 of that paper.
8. Hessian Metrics Along and Finsler Spaces

Lets go back to the general question of metrizability of a linear D along , meaning that
we want a g such that DX g = 0, X. Since solving the vanishing of (48) for g, for
example, is a hard problem, one will naturally be led to look at the effect of imposing extra
restrictions on g. It seems to me that a good way to proceed would be to work in stages, as
follows: first study the effect of imposing that gij be the Hessian matrix of some function
(with respect to the fibre coordinates v i ), and secondly assume that g is homogeneous in
the v i , for example (but not exclusively) of degree zero. I will not enter into this study in
any depth here. Though I think this has not been carried out yet in a systematic way, it is
true of course that many aspects of this idea frequently turn up in the literature (sometimes
rather related to one of the other types of connections discussed in section 5 though). In the
terminology of the Miron-school, for example, the first step would correspond to passing
from a generalized Lagrange space to a Lagrange space.
In [11], a linear D along is said to be metrizable if there exists a Finsler metric such
that DX g = 0, so homogeneity of g is taken to be part of the definition of metrizability
(and in fact also of the definition of variationality of D). A general S ODE on the other
hand is said to be metrizable if there exists a variational metric g, meaning it is a Hessian,
such that g = 0. In other words, two of the three Helmholtz conditions (33, 34, 35) are
incorporated in the definition of metrizability here, which of course makes life easier. Inci-
dentally, an interesting question which emerges in this context is: under what circumstances
is the remaining Helmholtz condition redundant? It is well known that this is the case for
Finsler metrics. I claim it is true also as soon as we have homogeneity of any order. It
would take me too far away from the subject of this paper, however, to prove this statement
here.
Instead, to finish, lets go back to the master we are celebrating in this volume. Specif-
ically, let me return to the paper [10] I started from, because it contains a definition of
metrizability, which is very surprising if you see it for the first time and leads to an equally
surprising conclusion which is worth explaining in more intrinsic terms. Let me recall first
the definition of a Hessian metric in an intrinsic way.
Definition. The Hessian of a function L on T M is the symmetric type (0, 2) tensor field g
along , defined by
g(X, Y ) = DVY DVX L DVDV X L, X, Y X ( ). (93)

Y
Now in [10], a linear D along is said to be metrizable if there exists a non-singular,

symmetric g along , such that
gij j
DX g = 0, X, and v = 0. (94)
v k
336 W. Sarlet
In intrinsic terms, the surprise extra condition means that
T DVX g = 0, X. (95)
Lemma. If a symmetric g along has the property (95), then it is a Hessian and is homo-
geneous of degree 0 in the fibre coordinates.
Proof. Put L = 12 g(T, T). Then, it follows from (95) and the general property DVX T = X
that DVX L = g(X, T). Taking a further vertical derivative with respect to Y and using the
intermediate result plus (95) again, we obtain that g satisfies (93), i.e. is the Hessian of L.
But a Hessian is characterized by the property (34), from which it follows, taking T as
one of the arguments and using (95) again, that DVT g = 0. This precisely means that g is
homogeneous of degree zero.
This result is far from new: (95) in one form or another is the defining relation for
a g along to be what is called normal; it is well known as the necessary and sufficient
condition for g to be a Finslerian metric (see e.g. [19], [13]) and as such is attributed to
Hashiguchi [9]. A side remark: now that Finsler structures come into the picture, I will
omit the technicalities about having to pass from the tangent bundle to the slit tangent
bundle, and also not go into requirements about positive definiteness.
The main theorem in [10] states that under the conditions (94), the connection D is
variational and more precisely is the Cartan connection of a Finsler structure. The proof
of this theorem leaves the reader a bit startled, however. First of all, it is clear that the
calculations in the proof take for granted that the connection coefficients ijk of the given
D are symmetric: it is indeed a somewhat hidden assumption throughout the paper that
the connection is torsion free. Secondly, what is explicitly shown is an equivalence of
connections in the sense of variationality, that is to say: the geodesic S ODE of D is shown
to be the same as the one coming from the Euler-Lagrange equations of the Finslerian g.
There is no explicit verification, however, that the assumptions imply that the given ijk are
effectively those of what is called (at least by some) the Cartan connection in that context.
I shall finish by presenting a slight generalization of this theorem which consists in
obtaining roughly the same results from somewhat weaker conditions. This will give me a
chance to illustrate some of the features discussed in the previous sections, while the details
of the Krupka-Sattarov theorem will follow as a special case.
Theorem. Let D be a torsion-free, linear connection along , for which there exists a
non-singular, symmetric g along , such that
CD (X, T) = 0 and T DVX g = 0, X. (96)
Then the following assertions hold true:

(i) D is variational: its geodesic S ODE in fact is the canonical spray of g which is a
Finsler metric.
(ii) There exists a variationally equivalent D
e along which is metric with respect to g:
DX g = 0, X.
e
k of D
(iii) The connection coefficients eij e have an explicit expression in terms of g only
and as such are those of the Cartan connection of g.
Proof. (i) From the lemma we know that g is Finslerian: it is the Hessian of L = 21 g(T, T),
and L is homogeneous of degree 2 in the v i . From the coordinate expression (86), taking
the symmetry of the connection into account, it is clear that CD (T, T) = 0 implies that
k i j 1
g kl hi (gjl ) + hj (gil ) hl (gij ) v i v j

ij vv = 2
1 kl gjl gil gij

set
= 2g + j v i v j = (g )kij v i v j , (97)
q i q q l
where the reduction from the first to the second line follows from the property (95) again.
This second line clearly reveals the force terms of the Euler-Lagrange equations of L (and
should be read also as defining relations of the functions (g )kij (q, v)).
(ii) Since CD (X, T) = 0, the defining equation (88) of D e reduces to
e X = DX + 1
D 2 X CD ,
and it follows from Proposition 3 that D e X g = 0, X. Moreover, the new k are given
eij
k = k + 1 C k , so that C (T, T) = 0 implies that
by eij k v i v j = k v i v j , proving the
ij 2 D ij D eij ij
variational equivalence of both connections.
(iii) We can now compute the eij k from the vanishing of the Cartan tensor of D. e For
simplicity in notations, let us omit the tildes, i.e. assume we are back in the Krupka-Sattarov
situation now, so that DX g = 0. Then CD = 0 implies that
k 1
g kl hi (gjl ) + hj (gil ) hl (gij ) ,

ij = 2 (98)
but the right-hand sides in this expression contain in the vertical derivatives of the g-
components factors of the form ijk v j . Multiplying (98) with v j (in other words, using
CD (X, T) = 0) and making use of (95) again, we find that
k j gil
ij v = (g )kij v j 21 g kl js
r j s
v v . (99)
v r
Again, we still have s in the right-hand side, but we can eliminate them now by using
(97). Substituting these intermediate results back into the expression (98), we finally get
gjl gil gij

k
ij = (g )kij 12 g kl r
(g )ris + r (g )rjs (g )rls v s
v v v r
gjl git gil gjt gij glt

+ 14 g kl g rt + (g )usp v s v p . (100)
v r v u v r v u v r v u
This provides an explicit expression of the ijk in terms of the metric g, which is the same
as equation (4.5) in [10], and is called the Cartan connection there.
Remark. the implicit specification of the ij k , from which the final result can be deduced,
is also (for a symmetric connection) equation (A.27) in the previously cited Appendix of
[4], where it is referred to as the Cartan connection too. The type of computation in part
(iii) of the proof in fact is similar also to the way the Cartan connection is set up in [1]. We
have seen in section 5, however, that when D is mapped onto a linear connection on the
pullback bundle according to Proposition 2, the metric nature of D corresponds to being
338 W. Sarlet
horizontally metric, so that the more common terminology in that context would be that we
are talking about a connection of Chern-Rund type.
There are some interesting corollaries of the above theorem. Once we know that the
geodesic S ODE of D is the canonical spray of a Finsler metric g, it follows that g = 0.
Hence, we are in the situation (84) and know that T K g = 0. In fact, we can do a bit better
and show that T K = 0.
Proposition 5. Under the assumptions of the above theorem, the fundamental tensor K of
the linear connection D has the property K(T, X) = 0, X X ( ). It follows that the
horizontal lift h associated to D, and the horizontal lift H of its geodesic S ODE coincide.
Proof. For the components of T K, we can write
k
ij
r v i v j = r (ij
k i j k j
v v ) + 2 rj v .
v v
But we know that CD (X, T) = 0, so that we can use the expressions (97) and (99) to
compute the two terms in the right-hand side. It is straightforward to verify that making
these substitutions, and using (95), all terms cancel out. The last statement then immediately
follows from (54).
Finally, taking (52) and (56) into account, plus the fact that Th = always, it follows
that under the assumptions of the theorem: DT = DhT = DH T = .
9. Concluding Remarks
Having arrived in Finsler country at the end of our journey, where there is a vast literature,
and where every household seems to foster its own notations, it is possible that there are still
other results to which I could or should have compared aspects of what has been discussed
in this paper. But I hope the reader will find the road to these results using [3], though this
is not an easy navigating system .
The general construct of linear connections along , in the sense of Runds direction-
dependent connections, has been used strictu senso only occasionally in the literature, but
a number of aspects about such operations remained unclear, specifically with respect to
the intrinsic foundations of the theory. I hope I have managed to clarify such aspects in
this paper. New, potentially interesting questions have come up in my analysis and there
are undoubtedly many more one can think of. However, having identified also a number of
technicalities and dangers for confusion with related concepts, there is one major question I
would like to put forward, namely: Do we actually need linear connections along ?. Isnt
it possible for example that, with the geometrical calculus offered by linear connections on
the pullback bundle , we have sufficient tools in our hand to analyse all theoretical
questions one might wish to study with linear connections along ?
References
[1] M. Abate and G. Patrizio, Finsler metrics A Global Approach (Lecture Notes in
Math. 1591, Springer-Verlag, Berlin, 1994).
[2] M. Anastasiei, Finsler connections in generalized Lagrange spaces, Balkan J. of

Geom. and its Appl. 1 (1996) 19.
[3] P. L. Antonelli (Ed.), Handbook of Finsler Geometry, Vol.1 and Vol.2 (Kluwer Aca-
demic Publishers, Dordrecht, 2003).
[4] G. S. Asanov, Finsler Geometry, Relativity and Gauge Theories (D. Reidel Publ.
Comp. Dordrecht, 1985).
[5] A. Bejancu, Finsler Geometry and Applications (Ellis Horwood, 1990).
[6] I. Bucataru, Metric nonlinear connections, Diff. Geom. Appl. 25 (2007) 335343.
[7] M. Crampin, Connections of Berwald type, Publ. Math. (Debrecen) 57 (2000) 455
473.
[8] A. Frolicher and A. Nijenhuis, Theory of vector-valued differential forms, Proc. Ned.
Acad. Wetensch. Ser. A 59 (1956) 338359.
[9] M. Hashiguchi, On parallel displacements in Finsler spaces, J. Math. Soc. Japan 10

(1958) 365379.
[10] D. Krupka and A. E. Sattarov, The inverse problem of the calculus of variations for
Finsler structures, Math. Slovaca 35 (1985) 217222.
[11] O. Krupkova, Variational metrics on R T M and the geometry of nonconservative

mechanics, Math. Slovaca 44 (1994) 315335.
[12] O. Krupkova, Variational metric structures, Publ. Math. Debrecen 62 (2003) 461495.
[13] R. L. Lovas, J. Pek and J. Szilasi, Ehresmann connections, metrics and good metric
derivatives, In: Finsler Geometry (Sapporo 2005, Advanced Studies in Pure Mathe-
matics, Math. Soc. Japan 48, 2007) 263-308.
[14] E. Martnez and J. F. Carinena, Geometric characterization of linearizable second-

order differential equations, Math. Proc. Camb. Phil. Soc. 119 (1996) 373381.
[15] E. Martnez, J. F. Carinena and W. Sarlet, Derivations of differential forms along the
tangent bundle projection, Diff. Geom. Appl. 2 (1992) 1743.
[16] E. Martnez, J. F. Carinena and W. Sarlet, Derivations of differential forms along the
tangent bundle projection II, Diff. Geom. Appl. 3 (1993) 129.
[17] M. Matsumoto, Foundations of Finsler geometry and special Finsler spaces (Kai-
seisha Press, Japan, 1986).
[18] T. Mestdag and W. Sarlet, The Berwald-type connection associated to time-dependent

second-order differential equations, Houston J. of Math. 27 (2001) 763797.
[19] T. Mestdag, J. Szilasi and V. Toth, On the geometry of generalized metrics, Publ.
Math. (Debrecen) 62 (2003) 511545.
340 W. Sarlet
[20] R. Miron and M. Anastasiei, The Geometry of Lagrange Spaces: Theory and Ap-
plications (Fundamental Theories of Physics Vol. 59, Kluwer Academic Publishers,
Dordrecht, 1994).
[21] H. Rund, Direction-dependent connections and curvature forms, Abh. Math. Sem.
Univ. Hamburg 50 (1980) 188209.
[22] J. Szilasi, Notable Finsler connections on a Finsler manifold, Lect. Matematicas 19

(1998) 734.
[23] J. Szilasi, A setting for spray and Finsler geometry, In: Handbook of Finsler Geometry
(vol. 2, (P. L. Antonelli, Ed.) Kluwer Academic Publishers, Dordrecht, 2003) 1183
1426.
Chapter 19
O N THE I NVERSE P ROBLEM

FOR AUTOPARALLELS
G.E. Prince
Department of Mathematics and Statistics,
La Trobe University, Australia
Abstract
The inverse problem in the calculus of variations for a given set of second order
ordinary differential equations consists of deciding whether their solutions are those
of EulerLagrange equations and exhibiting the non-uniqueness of the resulting La-
grangians when they occur. In this paper we use the techniques of the inverse problem
to examine the conditions under which such a system of equations are the geodesic
equations of a Finsler metric.
Key words and phrases. Inverse problem of the calculus of variations, EulerLagrange
equations, Helmholtz conditions, linear connection, Finsler metrizability.
1. The Inverse Problem in the Calculus of Variations

and Finsler Geometry
This paper addresses the question (loosely described) of when the autoparallels of a linear
connection are the geodesics of a Finsler metric. I thank Demeter Krupka for introducing
me to this problem through his paper with A.E. Sattarov [14]. This is quite a deep question
with many aspects and the work here focuses on rather preliminary stages of an investigation
based upon the existing studies of a broader problem known as the inverse problem in the
calculus of variations which I now describe.

This paper is dedicated to Demeter Krupka on the occasion of his 65th birthday. It has been my pleasure to
work with Professor Krupka over the last few years and to have enjoyed his hospitality. His energy, generosity
and mathematical virtuosity are an inspiration to us all.

E-mail address: G.Prince@latrobe.edu.au
342 G.E. Prince
The inverse problem in the calculus of variations involves deciding whether the solu-
tions of a given system of second-order ordinary differential equations
xa = F a (t, xb xb ), a, b = 1, . . . , n (1)
are the solutions of a set of Euler-Lagrange equations
2L b 2L b 2L L
a b
x + b a
x + a
=
x x x x t x xa
for some Lagrangian function L(t, xb , xb ). (Note that the curve parameter for both systems
2
is t.) Clearly the Hessian matrix xa Lxb should be invertible on some domain. When a
Lagrangian exists the equations (1) are said to be variational. The problem dates to the end
of the 19th century and it still has importance for mathematics and mathematical physics
(see [16, 22]).
Because the Euler-Lagrange equations are not generally in normal form, the problem is
to find a so-called multiplier matrix gab (t, xc , xc ) which is invertible on some domain and
such that
d L L
gab (xb F b ) a.
dt xa x
The most commonly used set of necessary and sufficient conditions for the existence of
the gab are the socalled Helmholtz conditions due to Douglas [10] and put in the following
form by Sarlet [23]:
gab gac
gab = gba , (gab ) = gac cb + gbc ca , gac cb = gbc ca , c
= ,
x xb
where
1 F a F a
ab := , ab := cb ac (ab ),
2 xb xb
and where

:= + ua a + F a a .
t x u
2
When a solution gab exists a corresponding Lagrangian is recovered from xa Lxb = gab .
A full review of this inverse problem can be found in the recent article by Krupkova and
Prince [16]. Some key papers in the literature are [3, 5, 6, 8, 20, 19, 24, 25, 26].
There are a number of inverse problems in Finsler geometry (see, for example, section
12.4 of [28]). In [14] Krupka and Sattarov address two questions: when are the autopar-
allel equations of a linear connection (depending on both xa and xa ) variational?, and
when does there exist a Finsler metric tensor covariantly constant under the connection?
A connection having this latter property is called metrisable, and Krupka and Sattarov prove
that every metrisable connection is variational using the Cartan connection of the Finsler
structure.
However, the Berwald connection is arguably a better point of contact between second
order differential equations and Finsler geometry (see [9, 12, 18, 19, 28]), so we will use the
Berwald connection to answer the question When are the solutions of (1) the geodesics of
a Finsler structure, in the given parametrisation? Since the Finsler geodesic equations are
xa + bc
a
(x, x)xb xc = 0, a = 1, . . . , n, (2)
On the Inverse Problem for Autoparallels 343
2 2

a := 1 g ad gbd + gcd gbc ,
where gab := 12 xaF xb for the Finsler function F and bc 2 x c x b x d
this entails applying the inverse problem described above to the case where the functions
F a are autonomous and positively homogeneous of degree 2 in the xa (this is explained
later). This means that the equations (1) describe the autoparallels of a certain symmetric
linear connection whose coefficients are homogeneous of degree zero in the xa , that is, the
equations are the autoparallel equations of their own Berwald connection. The multiplier
we seek will be the Finsler gab and so of necessity it must be autonomous and positively
homogeneous of degree 0 in the xa . The Helmholtz conditions for this Finsler inverse
problem will therefore yield identities in Finsler geometry which at best may be new but
which at least will provide a new angle on known Finsler facts. In stating this inverse
problem we immediately face the dilemma of notation. The inverse problem world and the
Finsler world share much common content but not too much notation, but since I will be
approaching the problem from the inverse problem perspective I will uncompromisingly
use that notation. Occasionally I will identify the corresponding Finsler equivalent using
[28] as a standard.
There has recently been renewed interest in this and related inverse problems in Finsler
geometry and I refer the interested reader to [4, 15, 21, 29].
2. The Geometry of SODEs

We will provide only enough of the geometric setting of the inverse problem to make the
later discussion viable; more complete descriptions and further references can be found in
[2, 16].
Suppose that M is some differentiable manifold with generic local co-ordinates (xa ).
The slit evolution space is defined as E := R T M , where T M is the tangent bundle
without the zero section, with projection onto the first factor being denoted by t : E
R and bundle projection : E R M . E has adapted co-ordinates (t, xa , ua )
associated with t and (xa ). We use the slit evolution space to ensure compatibility with our
Finsler inverse problem.
A system of n second order differential equations (SODEs) with local expression
xa = F a (t, xb , xb )
is associated with a smooth vector field on E given in the same co-ordinates by

:= + ua a + F a a .
t x u
is called a semispray. It can be thought of as the total derivative operator associated
with the differential equations. The integral curves of are just the parametrised and lifted
solution curves of the differential equations. When the system admits a Lagrangian as
described in section 1, is called the Euler-Lagrange field.
The evolution space E is equipped with the vertical endomorphism S, defined locally
by S := Va a (see [7] for an intrinsic characterisation). S combines the contact structure
and vertical subbundle, V (E ), of E , a being the local contact forms a := dxa ua dt
344 G.E. Prince
and Va := u a forming a basis for vector fields tangent to the fibres of : E R M

(the vertical subbundle).
It is natural to study the deformation of S produced by the flow of , L S. The
eigenspaces of this (1, 1) tensor field produce a direct sum decomposition of each tan-
gent space of E . It is shown in [7] that L S (acting on vectors) has eigenvalues 0, +1
and 1. The eigenspace at a point of E corresponding to the eigenvalue 0 is spanned by
, while the eigenspace corresponding to +1 is the vertical subspace of the tangent space.
The remaining eigenspace (of dimension n) is called the horizontal subspace. Unlike the
vertical subspaces these eigenspaces are not integrable; their failure to be so is due to the
curvature of this nonlinear connection (induced by ) which itself has components
1 F b
ab := .
2 ua
(In [28] these components are denoted Nba and F a are replaced by 2Ga .) The most useful
basis for the horizontal eigenspaces has elements with local expression

Ha = a
ba b
x u
so that a local basis of vector fields for the direct sum decomposition of the tangent spaces
of E is {, Ha , Va } with corresponding dual basis {dt, a , a } where
a = dua F a dt + ab b .
(In [28] the horizontal fields are denoted x a .) The components of the curvature manifest
themselves in the expression for the commutators of the horizontal fields:
d
[Ha , Hb ] = Rab Vd
where the curvature of the connection is defined by

2 d
2F d 1 F c 2 F d F c 2 F d

d 1 F
Rab := + .
2 xa ub xb ua 2 ua uc ub ub uc ua
It will be useful to have some other commutators:
1 2F c
[Ha , Vb ] = ( a b )Vc = Vb (ca )Vc = Va (cb )Vc = [Hb , Va ],
2 u u
[, Ha ] = ba Hb + ba Vb , [, Va ] = Ha + ba Vb ,
and, of course, [Va , Vb ] = 0.
We wont go on to describe the canonical linear connection of Berwald type associated
to the semispray, the reader is referred to [12, 17, 19].
Finally, the following will be important
ab = Hb (F a ) + cb ac (ab ), (3)
Ha (cb ) Hb (ca ) = Rab
c
, (4)
c c c
Va (b ) Vb (a ) = 3Rab . (5)
In the context of our Finsler inverse problem we begin with equations (1) with
F a (t, xb , ub )
autonomous and positively homogeneous of degree 2 in the ub . (See [15]
for more general, time-dependent considerations.) We immediately make the usual obser-
a
vation (see, for example, [28]) that ab is homogeneous of degree 1 in the ua , abc := ubc is
homogeneous of degree 0 in the ua so that F a = abc ub uc and (1) becomes
xa + abc (x, x)xb xc = 0. (6)
At this point we make the important remark due to Berwald and so well explained in the
introduction to Mestdags thesis [18], namely that the abc transform as the components of
a linear connection while the bc a of the Finsler geodesic equation (2) in general do not. In
fact, for the Finsler forces F a := bca ub uc we have

e Cbdc e Cbdc
abc = a
bc +g ad
u +F e e
2Cbde c 2Cdce b , (7)
xe ue
where Cabc := 12 g ab
uc are the coefficients of the Cartan torsion (see [28]: Shen also uses the
same notation as us for the abc ).
Returning to the Finsler inverse problem whose starting point is equation (6), equations
(3), (4), and (5) become
ab = Rdbc
a
uc ud , (8)
c
Rdab ud = Rab c
, (9)
c c
(Va (Rebd ) Vb (Read ))ue ud = 0, (10)
a := H (a ) H (a ) + a e a e are homogeneous of degree zero in

where Rdbc b cd c db eb cd ec bd
a a = Ra and Ra + Ra + Ra = 0 (compare with [9]).
the u and satisfy Rdbc dcb dbc cdb bcd
3. The Helmholtz Conditions and the Finsler Inverse Problem

We begin with some general observations about the Helmholtz conditions given in section
1.. They are the necessary and sufficient conditions that a two form gab a b be closed
and of maximal rank on some domain. This can be given an even more geometric phrasing
in the following theorem from [7]:
Theorem 3.1. Given a semispray , the necessary and sufficient conditions for there to be
Lagrangian for which is the EulerLagrange field is that there should exist a 2form
such that
(V1 , V2 ) = 0, V1 , V2 V (E),
= 0, d = 0,
is of maximal rank.
346 G.E. Prince
Following Aldridge [1] the simplest way to see how the Helmholtz conditions arise
from Theorem 3.1 is to put := gab a b and compute d:
d = ((gab ) gcb ca gac cb )dt a b

+ (Hd (gab ) gcb Va (cd )) a b d
+ Vc (gab ) c a b + gab a b dt
+ gca cb a b dt + gca Hb (cd )a b d .
The four Helmholtz conditions are
d(, Va , Vb ) = 0, d(, Va , Hb ) = 0,
d(, Ha , Hb ) = 0, d(Ha , Vb , Vc ) = 0.
The remaining conditions arising from d = 0, namely
d(Ha , Hb , Vc ) = 0 and d(Ha , Hb , Hc ) = 0,
can be shown to be derivable from the first four (notice that this last condition is void in
dimension 2).
In the Finsler inverse problem we assume that the multiplier gab , if it exists, does not
depend on t, is homogeneous of degree zero in the ua and is symmetric and non-degenerate.
The consequences of the remaining Helmholtz conditions are given in following lemmas:
Lemma 3.2 (Compare with theorem 2 of [14]). The Helmholtz conditions (gab ) = gac cb +
gbc ca and Vc (gab ) = Vb (gac ) imply that
!
a a ad e Cbdc e Cbdc e e
bc = bc + g u +F 2Cbde c 2Cdce b (11)
xe ue
and

1 gab
gad dbc gbd dac 2Cabd dc
2 xc
Cabc Cabc
= F d d
+ Cdbc da + Cadc db + Cabd dc ud (12)
u xd

a := 1 g ad gbd + gcd gbc , and C
where F a = 21 abc ub uc , bc 1 gab
2 x c xb xd abc := 2 uc .
Proof. The proof entails using the expression for in the first of the given Helmholtz con-
ditions and then differentiating this condition with respect to uk . Generating a further two
conditions by permuting the indices a, b, k and taking the appropriate linear combination
gives the result (11) upon using the conditions Vc (gab ) = Vb (gac ) and the homogeneity of
the abc . The result (12) comes from (11) by permutation of indices and addition.
Lemma 3.3. The Helmholtz conditions (gab ) = gac cb + gbc ca and Vc (gab ) = Vb (gac )
imply that
Ha (gbc ub uc ) = 0 and (gbc ub uc ) = 0. (13)
Proof. Following the proof given in section 4.2 of [28] for the Finsler case, we contract
(12) on ub uc and use Vc (gab ) = Vb (gac ) and the homogeneity of gab to remove all the C
terms giving
gbc b c
u u = 2gbe ub ea .
xa
Hence
(gbc ub uc ) b c
e (gbc u u ) gbc b c
Ha (gbc ub uc ) = a = u u 2gbe ub ea = 0
xa ue xa
where Vc (gab ) = Vb (gac ) and the homogeneity of gab has been used to establish the last
equality. Finally, because gab is assumed independent of t,
(gbc ub uc ) = ua Ha (gbc ub uc ) = 0.
Lemma 3.4 (Compare with proposition 8.2.1 of [28]). The Helmholtz conditions (gab ) =
gac cb + gbc ca and Vc (gab ) = Vb (gac ) imply that
gac cb = gbc ca d
and gad Rbc d
+ gcd Rab d
+ gbd Rca = 0. (14)
Equivalently,
c
gac Rdbe ud ue = gbc Rdae
c
ud ue and d
(gad Rebc d
+ gcd Reab d
+ gbd Reca )ue = 0. (15)
Proof. To see that the g condition is a consequence of the given conditions, consider
the commutator [Ha , Hb ] acting on gcd uc ud and use lemma 3.3, the condition Vc (gab ) =
Vb (gac ) and the homogeneity of gab :
0 = [Ha , Hb ](gcd uc ud ) = Rab

e
Ve (gcd uc ud )
e
= 2Rab gce uc = 2Rdab
e
gce ud uc (16)
Similarly for [, Ha ]:
0 = [, Ha ]((gcd uc ud ) = 2ba gbc uc .
Differentiating this equality vertically gives
Vk (ba )gbc uc = ba gbk
Interchanging a and k, taking the difference and using (5) gives

b
3Rdka gbc ud uc = bk gba ba gbk .
Applying (16) gives the result.

Deriving the gR condition from the g condition (and hence from the given conditions)
is a straightforward matter of vertical differentiation of the g condition and the use of
(5).
348 G.E. Prince
We point out that for arbitrary SODEs the g condition is generally independent of the
differential Helmholtz conditions, although the gR condition is a differential consequence
of the g condition.
As a result of these lemmas and theorem 3.1 we have
Theorem 3.5. Necessary and sufficient conditions for equations (6) to be the geodesic
equations of a Finsler metric are that there exist functions gab on E , symmetric, non-
degenerate and homogeneous of degree 0 in the uc satisfying Va (gbc ) = Vb (gac ) and equa-
tion (11).
The theorem indicates the paramount importance of the Berwald connection in the study
of this Finsler inverse problem. However, it does not provide necessary and sufficient condi-
tions on the connection alone for the existence of gab . Such conditions usually follow from
the restrictions to specific dimensions or classes of Finsler spaces. Nonetheless the formal-
ism of the inverse problem in the calculus of variations does provide a powerful means of
deriving results in Finsler geometry.
4. Remarks
Since the Riemannian situation is a special case of the Finsler one (once we restore the zero
section of T M ) the results so far all apply and they can be refined using the independence of
the metric and the connection on the velocities. Our inverse problem in Riemannian geom-
etry is the question of whether a symmetric linear (affine) connection on an manifold M is
the metric connection of a Riemannian manifold. This is a known problem of long-standing
(see, for example, [11, 27]) but only recently have the techniques of the inverse problem
been applied to it: see [4], and we might expect that examination of the Helmholtz con-
ditions will shed light on the extent to which Riemannian curvature determines the metric
structure, Riemannian or Finslerian.
Following the analysis so far we can immediately state the following theorem
Theorem 4.1. Let abc be the connection coefficients of a symmetric linear connection
on a manifold M with coordinates xa . If the autoparallels of are the solutions of Euler-
Lagrange equations with corresponding multiplier gab (xc ), then g is a metric on M with
Christoffel symbols abc .
This result is not useful from a constructive point of view, but the general study of
the Helmholtz conditions involves the construction of a maximal rank system of algebraic
conditions for the multiplier gab based on the Jordan normal form of the tensor field ([8]).
Following Edgar [11], it is fruitful to examine the algebraic conditions (h)T = h for
nondegenerate (0, 2) tensor field solutions h. Before stating a minor result for dimension
2 in this regard, I pose the following question: is it possible to derive consequences of the
first of equations (14) (equivalently (15)) which are positively homogeneous of order less
than 2 in the ua ?
Theorem 4.2. Let be a symmetric linear connection on M with dim(M ) = 2, spray ,

connection coefficients abc , non-degenerate Ricci tensor Rab and ab := abc uc . is metric
if and only if the Ricci tensor is recurrent with

1 cd
R (Rcd ) 2dd Rab = (Rab ) Rac cd Rbc ca .
2
and there exists a function on M satisfying
1 ab
() = R (Rab ) 2aa .
2
The metric is gab := 2e Rab .
a
This result, derived from the Helmholtz conditions and using the fact that Rcdb =
a a
b Rcd d Rcb in dimension 2, appears to provide a bridge to the holonomy results of
Schmidt [27] (see [13] on recurrence and holonomy).
Acknowledgements
I thank the Department of Algebra and Geometry, Palacky University, Olomouc in the
Czech Republic for their generous hospitality and support during my sabbatical in 2006.
References
[1] J. E. Aldridge, Aspects of the inverse problem in the calculus of variations. Ph.D. the-
sis, Mathematics Dept., LaTrobe University, Melbourne, Australia. Submitted March
2003.
[2] J. Aldridge, G. Prince, W. Sarlet and G. Thompson, The EDS approach to the inverse
problem in the calculus of variations, J. Math. Phys. 47 (2006) 103508 pp. 22.
[3] I. Anderson and G. Thompson, The inverse problem of the calculus of variations for
ordinary differential equations, Memoirs Amer. Math. Soc. 98 (473) (1992).
[4] M. Crampin, On the inverse problem for sprays, Publ. Math. Debrecen 70 (2007)
319335.
[5] M. Crampin, E. Martnez and W. Sarlet, Linear connections for systems of second
order ordinary differential equations, Ann. Inst. H. Poincare Phys. Theor. 65 (1996)
223249.
[6] M. Crampin, G. E. Prince, W. Sarlet and G. Thompson, The inverse problem of the
calculus of variations: separable systems, Acta Appl. Math. 57 (1999) 239254.
[7] M. Crampin, G. E. Prince and G. Thompson, A geometric version of the Helmholtz

conditions in time dependent Lagrangian dynamics, J. Phys. A: Math. Gen. 17 (1984)
14371447.
[8] M. Crampin, W. Sarlet, E. Martnez, G. B. Byrnes and G. E. Prince, Toward a geo-

metrical understanding of Douglass solution of the inverse problem in the calculus of
variations, Inverse Problems 10 (1994) 245260.
350 G.E. Prince
[9] M. Crampin and D. J. Saunders, Projective connections, J. Geom. Phys. 57 (2007)

691727.
[10] J. Douglas, Solution of the inverse problem of the calculus of variations, Trans. Am.
Math. Soc. 50 (1941) 71128.
[11] S. B. Edgar, Conditions for a symmetric connection to be a metric connection, J. Math.

Phys. 33 (1992) 37163722.
[12] M. Jerie and G. E. Prince, Jacobi fields and linear connections for arbitrary second
order ODEs, J. Geom. Phys. 43 (2002) 351370.
[13] S. Kobayashi and K. Nomizu, Foundations of Differential Geometry. Vol I. (John Wi-
ley, New York-London, 1963).
[14] D. Krupka and A. E. Sattarov, The inverse problem of the calculus of variations for
Finsler structures, Math. Slovaca 35 (1985) 217222.
[15] O. Krupkova, Variational metric structures, Publ. Math. Debrecen 62 (2003) 461495.
[16] O. Krupkova and G.E. Prince, Second Order Ordinary Differential Equations in Jet
Bundles and the Inverse Problem of the Calculus of Variations, In: Handbook of
Global Analysis ((D. Krupka and D. Saunders, Eds.) Elsevier, 2008) 841908.
[17] E. Massa and E. Pagani, Jet bundle geometry, dynamical connections, and the inverse
problem of Lagrangian mechanics, Ann. Inst. Henri Poincare, Phys. Theor. 61 (1994)
1762.
[18] T. Mestdag, Berwald-type connections in time-dependent mechanics and dynamics on

affine Lie algebroids Ph.D. thesis, Universiteit Gent, Gent, 2003.
[19] T. Mestdag and W. Sarlet, The Berwald-type connection associated to time-dependent

second-order differential equations, Houston J. Math. 27 (2001) 763797.
[20] G. Morandi, C. Ferrario, G. Lo Vecchio, G. Marmo and C. Rubano, The inverse prob-
lem in the calculus of variations and the geometry of the tangent bundle, Phys. Rep.
188 (1990) 147284.
[21] Z. Muzsnay, The Euler-Lagrange PDE and Finsler metrizability, Houston J. Math. 32
(2006) 7998.
[22] G. E. Prince, The inverse problem in the calculus of variations and its ramifications,
In: Geometric Approaches to Differential Equations ((P. Vassiliou and I. Lisle, Eds.)
Lecture Notes of the Australian Mathematical Society, CUP, 2000).
[23] W. Sarlet, The Helmholtz conditions revisited. A new approach to the inverse problem
of Lagrangian dynamics, J. Phys. A: Math. Gen. 15 (1982) 15031517.
[24] W. Sarlet and M. Crampin, Addendum to: The integrability conditions in the inverse
problem of the calculus of variations for second-order ordinary differential equations,
Acta Appl. Math. 60 (2000) 213224.
[25] W. Sarlet, M. Crampin and E. Martnez, The integrability conditions in the inverse
problem of the calculus of variations for second-order ordinary differential equations,
Acta Appl. Math. 54 (1998) 233273.
[26] W. Sarlet, G. Thompson and G. E. Prince, The inverse problem of the calculus of
variations: the use of geometrical calculus in Douglass analysis, Trans. Am. Math.
Soc. 354 (2002) 28972919.
[27] B. G. Schmidt, Conditions on a connection to be a metric connection, Comm. Math.

Phys. 29 (1973) 5559.
[28] Z. Shen, Differential Geometry of Spray and Finsler Spaces (Kluwer, Dordrecht,
2001).
[29] J. Szilasi and Sz. Vattamany, On the Finsler-metrizabilities of spray manifolds, Period.
Math. Hungar. 44 (1) (2002) 81100.
Part IV
A PPENDIX
353
D EMETER K RUPKA
List of publications, to December 2007

Research articles and memoirs
1. Lagrange theory in fibered manifolds, Rep. Math. Phys. 2 (1971) 121133.
2. Some Geometric Aspects of Variational Problems in Fibered Manifolds, Folia Fac.

Sci. Nat. Univ. Purk. Brunensis, Physica 14, Brno, Czechoslovakia (1973) pp. 65;
arXiv:math-ph/0110005.
3. On the structure of the Euler mapping, Arch. Math. (Brno) 10 (1974) 5561.
4. On generalized invariant transformations, Rep. Math. Phys. (Torun) 5 (1974) 353

358.
5. (with A. Trautman) General invariance of Lagrangian structures, Bull. Acad. Polon.

Sci., Ser. Sci. Math. Astronom. Phys. 22 (1974) 207211.
6. A setting for generally invariant Lagrangian structures in tensor bundles, Bull. Acad.
Polon. Sci., Ser. Sci. Math. Astronom. Phys. 22 (1974) 967972.
7. Lagrangians and topology, Scripta Fac. Sci. Nat. UJEP Brunensis, Physica 34
(1975) 265270.
8. A geometric theory of ordinary first order variational problems in fibered manifolds,

I. Critical sections, J. Math. Anal. Appl. 49 (1975) 180206.
9. A geometric theory of ordinary first order variational problems in fibered manifolds,

II. Invariance, J. Math. Anal. Appl. 49 (1975) 469476.
10. A theory of generally invariant Lagrangians for the metric fields II, Internat. J. The-
oret. Phys. 15 (1976) 949959.
356 Demeter Krupka
11. On a class of variational problems defined by polynomial lagrangians, Arch. Math.

(Brno) 12 (1976) 99106.
12. A map associated to the Lepagean forms of the calculus of variations in fibered man-
ifolds, Czech. Math. J. 27 (1977) 114118.
13. A theory of generally invariant Lagrangians for the metric fields I, Internat. J. Theo-
ret. Phys. 17 (1978) 359368.
14. Elementary theory of differential invariants, Arch. Math. (Brno) 14 (1978) 207214.
15. (with M. Horak) On the first order invariant Einstein-Cartan variational structures,
Internat. J. Theoret. Phys. 17 (1978) 573584.
16. Mathematical theory of invariant interaction lagrangians, Czech. J. Phys. B 29 (1979)
300303.
17. Fundamental vector fields on the type fibres of jet prolongations of tensor bundles,
Math. Slovaca 29 (1979) 159167.
18. Reducibility theorems for differentiable liftings in fibre bundles, Arch. Math. (Brno)
15 (1979) 93106.
19. On the Lie algebras of higher differential groups, Bull. Acad. Polon. Sci., Ser. Sci.
Math. Astronom. Phys. 27 (1979) 235239.
20. A remark on algebraic identities for the covariant derivatives of the curvature tensor,
Arch. Math. (Brno) 16 (1980) 205211.
21. Finite order liftings in principal fiber bundles, Beitrage zur Algebra und Geometrie
11 (1981) 2127.
22. On the local structure of the Euler-Lagrange mapping of the calculus of variations,
In: Proc. Conf. on Diff. Geom. Appl (Nove Mesto na Morave, Czechoslovakia, Sept.
1980, Charles Univ., Prague, 1982) 181188.
23. (with M. Francaviglia) The Hamiltonian formalism in higher order variational prob-
lems, Ann. Inst. H. Poincare, Sec. A 37 (1982) 295315.
24. Local invariants of a linear connection, In: Differential Geometry (Colloq. Math.
Soc. Janos Bolyai 31, North Holland, 1982) 349369.
25. Lepagean forms in higher order variational theory, In: Modern Developments in An-
alytical Mechanics (Proc. IUTAM-ISIMM Sympos., Turin, June 1982, Academy of
Sciences of Turin, 1983) 197238.
26. (with J. Musilova) Integrals of motion in higher order mechanics and field theory,
Pokroky mat. fyz. astronom. 28 (1983) 259266 (in Czech).
27. (with O. Stepankova) On the Hamilton form in second order calculus of variations,
In: Proc. Internat. Meeting Geometry and Physics (Florence, October 1982,
Pitagora, Bologna, 1983) 85101.
Demeter Krupka 357
28. (with J. Musilova) Calculus of Odd Base Forms on Differential Manifolds, Folia Fac.
Sci. Nat. UJEP Brunensis 24 (1983) pp. 65.
29. Natural Lagrangian structures, In: Differential Geometry (Banach Center Publica-
tions 12, Diff. Geom. Semester, Warsaw, Sept.Dec. 1979, Polish Scientific Publish-
ers, Warsaw, 1984) 185210.
30. (with J. Musilova) Hamilton extremals in higher order mechanics, Arch. Math.
(Brno) 20 (1984) 2130.
31. On the higher order Hamilton theory in fibered spaces, In: Proc. Conf. on Diff.
Geom. Appl. (Vol. 2, Nove Mesto na Morave, Czechoslovakia, (D. Krupka, Ed.)
Sept. 1983, J. E. Purkyne Univ., Brno, 1984) 167184.
32. (with V. Mikolasova) On the uniqueness of some differential invariants, Czech. Math.
J. 34 (1984) 588597.
33. (with A. Sattarov) The inverse problem of the calculus of variations for Finsler struc-
tures, Math. Slovaca 35 (1985) 217222.
34. Geometry of Lagrangean structures, 1. Odd base forms, Arch. Math. (Brno) 22
(1986) 159174.
35. Geometry of Lagrangean structures, 2. Differential forms on jet prolongations of

fibered manifolds, Arch. Math. (Brno) 22 (1986) 211228.
36. Geometry of Lagrangean structures, 3. Lepagean forms and the first variation, In:
Proc. 14th Winter School on Abstract Analysis (Srn, Czechoslovakia, Jan. 1986
Suppl. Rend. del. Circolo Mat. di Palermo, Ser. II, 1987) 187224.
37. Regular Lagrangians and Lepagean forms, In: Differential Geometry and its Appli-
cations (Proc. Conf., Brno, Czechoslovakia, (D. Krupka and A. Svec, Eds.) August
1986, Kluwer Academic Publishers, 1987) 111148.
38. Variational sequences on finite order jet spaces, In: Differential Geometry and its
Applications (Proc. Conf., Brno, Czechoslovakia, (J. Janyska and D. Krupka, Eds.)
August 1989, World Scientific, Singapore, 1990) 236254.
39. Topics in the calculus of variations: Finite order variational sequences, In: Differ-
ential Geometry and its Applications (Proc. 5th Internat. Conf., Opava, Czechoslo-
vakia, (O. Kowalski and D. Krupka, Eds.) Aug. 1992, Silesian Univ. in Opava, 1993)
473495.
40. The contact ideal, Diff. Geom. Appl. 5 (1995) 257276.
41. Lectures on Variational Sequences, Advanced Texts in Mathematics, Open Ed. &
Sci., Opava, Czech Republic, 1995, pp. 94.
42. The trace decomposition problem, Beitrage zur Algebra und Geometrie 36 (1995)
303315.
358 Demeter Krupka
43. The trace decomposition of tensors of type (1,2) and (1,3), In: New Developments
in Differential Geometry (Proc. Colloq. on Diff. Geom., July 1994, Debrecen, (J.
Szenthe and L. Tamassy, Eds.) Kluwer Academic Publishers, Dordrecht, 1996) 243
253.
44. Variational sequences in mechanics, Calc. Var. 5 (1997) 557583.
45. (with D. R. Grigore) Invariants of velocities and higher order Grassman bundles, J.
Geom. Phys. 24 (1998) 244264.
46. (with J. Musilova) Trivial lagrangians in field theory, Diff. Geom. Appl. 9 (1998)
293305.
47. (with J. Musilova and O. Krupkova) The variational sequence in physical theories,
Cs. cas. pro fyziku A48 (1998) 330341 (in Czech).
48. (with Dao Qui Chau) 3rd order differential invariants of coframes, Math. Slovaca 49
(1999) 563576.
49. Variational sequences and variational bicomplexes, In: Differential Geometry and
Applications (Proc. Conf., (I. Kolar, O. Kowalski, D. Krupka and J. Slovak, Eds.)
Brno, August 1998, Masaryk Univ., Brno, Czech Republic, 1999) 525531.
50. (with J. Musilova), Erratum: Trivial lagrangians in field theory, Diff. Geom. Appl. 9
(1998) 293305; Diff. Geom. Appl. 10 (1999) 303.
51. (with J. Musilova) The variational sequence and trivial lagrangians, In: Proc. 13th
Conf. of Czech and Slovak Physicists (Zvolen, Slovakia, August 23 - 26, 1999, (M.
Reiffers and L. Just, Eds.) The Zvolen Technical Univ., 2000) 328331 (in Czech).
52. (with M. Krupka) Jets and Contact Elements, In: Proceedings of the Seminar on
Differential Geometry ((D. Krupka, Ed.) Mathematical Publications, Silesian Univ.
in Opava, Opava, Czechia, 2000) 3985.
53. (with J. Musilova) Recent results in variational sequence theory, In: Steps in Differ-
ential Geometry (Proc. Colloq., (L. Kozma, P. T. Nagy and L. Tamassy, Eds.) July
2000, Debrecen, Hungary, University of Debrecen, 2001) 161186.
54. Global variational functionals on fibered spaces, Nonlinear Analysis 47 (2001) 2633
2642.
55. Natural projectors in tensor spaces, Beitrage zur Algebra und Geometrie 43 (2002)
217231.
56. Variational principles for energy-momentum tensors, Rep. Math. Phys. 49 (2002)
259268.
57. The Weyl tensors, Publ. Math. Debrecen 62 (2003) 447460.
58. (with P. Musilova) Differential invariants of immersions of manifolds with metric

fields, Rep. Math. Phys. 51 (2003) 307313.
Demeter Krupka 359
59. Global variational principles: Foundations and current problems, In: Global Analysis
and Applied Mathematics (AIP Conference Proceedings 729, American Institute of
Physics, 2004) 318.
60. (with J. Brajerck) Variational principles for locally variational forms, J. Math. Phys.
46, 052903 (2005) 115.
61. (with J. Brajerck) Variational principles on frame bundles, In: Differential Geometry
and Its Applications (Proc. Conf., (J. Bures, O. Kowalski, D. Krupka and J. Slovak,
Eds.) Prague, August 2004, Charles University in Prague, Czech Republic, 2005)
559569.
62. (with J. Sedenkova) Variational sequences and Lepage forms, In: Differential Ge-
ometry and Its Applications (Proc. Conf., (J. Bures, O. Kowalski, D. Krupka and J.
Slovak, Eds.) Prague, August 2004, Charles University in Prague, Czech Republic,
2005) 617627.
63. (with O. Krupkova, G. Prince and W. Sarlet) Contact symmetries and variational
sequences, In: Differential Geometry and Its Applications (Proc. Conf., (J. Bures,
O. Kowalski, D. Krupka and J. Slovak) Prague, August 2004, Charles University in
Prague, Czech Republic, 2005) 605615.
64. (with J. Brajerck) Noether currents and order reduction on frame bundles, In: Proc.
40th Sympos. on Finsler Geometry In the Memory of our Teachers (Hokkaido
Tokai University, Sapporo, Society of Finsler Geometry, 2006) 3437.
65. Trace decompositions of tensor spaces, Linear and Multilinear Algebra 54 (2006)
235263.
66. The total divergence equation, Lobachevskii Journal of Mathematics 23 (2006) 71

93.
67. (with O. Krupkova, G. Prince and W. Sarlet) Contact symmetries of the Helmholtz
form, Differential Geometry and Applications 25 (2007) 518542.
68. (with J. Brajerck) Cohomology and local variational principles, Proc. of the XV In-
ternational Workshop on Geometry and Physics (Puerto de la Cruz, Tenerife, Canary
Islands, Spain, September 1116, 2006, Publ. de la RSME, 2007) 119124.
69. The Vainberg-Tonti Lagrangian and the Euler-Lagrange mapping, In: Differential
Geometric Methods in Mechanics and Field Theory, Volume in Honour of W. Sarlet
((F. Cantrijn and B. Langerock, Eds.) Gent, Academia Press, 2007) 8190.
70. Natural variational principles, In: Symmetry and Perturbation Theory (SPT 2007, (G.
Gaeta, R. Vitolo and S. Walcher, Eds.) Proc. Conf., Otranto, Italy, June 2-9, 2007,
World Scientific, 2007) 116123.
71. Global variational theory in fibred spaces, In: Handbook of Global Analysis (Elsevier,
2008) 773836.
360 Demeter Krupka
72. The structure of the Euler-Lagrange mapping, paper in honour of N.I. Lobachevskii,
Russian Mathematics (Iz. VUZ) 51 (12) (2007), N.I. Lobachevskii Anniversary Vol-
ume, 5270.
73. (with A. Patak) Geometric structure of the Hilbert-Yang-Mills functional, Int. J.

Geom. Met. Mod. Phys. 5 (2008) 387405.
74. (with O. Krupkova) Contact symmetries and variational PDEs, Acta Appl. Math.,
101 (13) (2008), A special issue dedicated to the 60th anniversary of Valentin Ly-
chagin, 163176.
75. (with Z. Urban) Differential invariants and higher order Grassmann bundles, In: Dif-
ferential Geometry and its Applications (Proc. 10th Int. Conf. on Diff. Geom. and
Appl., Olomouc 2007, World Scientific, Singapore, 2008, 463473).
Books
1. D. Krupka, J. Janyska, Lectures on Differential Invariants (J. E. Purkyne Univ., Fac-
ulty of Science, Brno, Czechoslovakia, 1990, pp. 193).
2. D. Krupka, D. Saunders (Eds.), Handbook of Global Analysis (Elsevier, 2008, pp.

1229)
Dissertations
1. A contribution to the theory of weak interactions of strongly interacting particles (in
Slovak), MSc thesis, J. E. Purkyne University, Brno, 1965.
2. Geometric Aspects of the Theory of Invariant Lagrange Structures, PhD (CSc.) Dis-
sertation, Faculty of Mathematics and Physics, Charles University, Prague, 1976.
3. Geometric Problems of the Calculus of Variations on Fibered Spaces, Assoc. Prof.

(Doc.) Dissertation, J. E. Purkyne University, Brno, 1980.
4. Natural Lagrangean structures, DrSc. Dissertation, Czechoslovak Academy of Sci-

ences, Prague, 1981.
Edited proceedings
1. Differential Geometry and Applications (Proc. Conf. Vol. 2, Nove Mesto na Morave,
Czechoslovakia, Sept. 1983 (D. Krupka, Ed.) J. E. Purkyne Univ., Brno, 1984).
2. Differential Geometry and Its Applications (Proc. Conf., Brno, Czechoslovakia, Au-
gust 1986 (D. Krupka and A. Svec, Eds.) Kluwer Academic Publishers, 1987).
3. Differential Geometry and Its Applications (Proc. Conf., Brno, Czechoslovakia, Au-
gust 1989 (J. Janyska and D. Krupka, Eds.) World Scientific, Singapore, 1990).
Demeter Krupka 361
4. Differential Geometry and Its Applications (Proc. 5th Internat. Conf., Opava,
Czechoslovakia, August 1992 (O. Kowalski and D. Krupka, Eds.) Silesian Univ.
Opava, 1993).
5. Differential Geometry and Its Applications (Proc. 7th Internat. Conf., Brno, Czech
Republic, August 1998, (I. Kolar, O. Kowalski, D. Krupka, and J. Slovak, Eds.)
Masaryk Univ., Brno, 1999).
6. Proceedings of the Seminar on Differential Geometry ((D. Krupka, Ed.) Silesian

University at Opava, Opava, 2000).
7. Differential Geometry and Its Applications (Proc. 8th International Conference on

Diff. Geom. and Appl., Opava, Czech Republic, August 2731, 2001, (O. Kowalski,
D. Krupka, and J. Slovak, Eds.) Diff. Geom. Appl. 17, 2002, special issue).
8. Differential Geometry and Its Applications (Proc. Conf, August 2001, Opava, Czech
Republic, (O. Kowalski, D. Krupka, and J. Slovak, Eds.) Silesian University in
Opava, Opava, 2003).
9. Global Analysis and Applied Mathematics (AIP Conference Proceedings 729,

(K. Tas, D. Krupka, O. Krupkova and D. Baleanu, Eds.) American Institute of
Physics, 2004).
10. Differential Geometry and Its Applications (Proc. Conf., Prague, August 2004
(J. Bures, O. Kowalski, D. Krupka and J. Slovak, Eds.) Charles University in Prague,
Czech Republic, 2005).
11. Differential Geometry and its Applications (Proceedings of the 10th Int. Conf., Olo-
mouc, August 2007 (O. Kowalski, D. Krupka, O. Krupkova and J. Slovak, Eds.)
World Scientific, Singapore, 2008).
University textbooks
1. (with F. Klvana) Exercises and Solved Problems in Quantum Mechanics (in Czech),
Faculty of Science, Brno University, 1973, pp. 189.
2. Mathematical Foundations of the General Relativity Theory (in Czech), Faculty of

Science, Brno University, 1979, pp. 130.
3. (with J. Musilova) Integration on Euclidean Spaces and Manifolds (in Czech), SPN
Praha, 1982, pp. 320.
4. Introduction to Analysis on Manifolds (in Czech), SPN Praha, 1986, pp. 96.
5. (with J. Musilova) Linear and Multilinear Algebra (in Czech), SPN Praha, 1989, pp.
281.
6. (with O. Krupkova) Topology and Geometry, Lectures and Solved Problems, I. Gen-
eral Topology (in Czech), SPN Praha, 1990, pp. 404.
362 Demeter Krupka
Translations
1. (with. I. Horova) V. M. Alexejev, V. M. Tichomirov, S. V. Fomin, Mathematical
Theory of Optimal Processes (translation from Russian to Czech), Academia, Praha,
1991.
2. (with I. Horova) V. M. Alexejev, E. M. Galejev, V. M. Tichomirov, Problems in Op-

timization Theory (translation from Russian to Czech), 1988, unpublished.
Unpublished notes, preprints

1. CP-noninvariant weak interactions of strongly interacting particles (in Slovak),
Czechoslovak Students Scientific Activity 1965, Bratislava (Slovakia) (paper
awarded the first prize in theoretical physics).
2. Differential Invariants (Lecture Notes) (Department of Algebra and Geometry, Fac-

ulty of Science, J.E. Purkyne University, Brno, 1979, pp. 67).
3. (with L. Klapka and J. Musilova) Geometric methods in the Calculus of Variations

(Lecture Notes), Poprad 1984, J. E. Purkyne University, Brno, 1984 (in Czech).
4. The geometry of Lagrange Structures, Preprint Series in Global Analysis GA7/1997,

Silesian Univ. Opava, Czech Republic, pp. 82.
5. Elementary sheaf theory, Preprint Series in Global Analysis GA2/1998, Silesian

Univ. Opava, Czech Republic, pp. 64.
6. Smooth Manifolds, Preprint Series in Global Analysis GA14/2000, Silesian Univ.

Opava, Czech Republic, pp. 31.
7. (with M. Lenc) The Hilbert variational principle, Preprint 3/2002 GACR

201/00/0724, Masaryk University, Brno, 2002, pp. 75.
8. (with A. Patak) The Hilbert-Yang-Mills functional: Global theory, Preprint Series in

Global Analysis, UP Olomouc, 2008.
INDEX
boundary conditions, 14
brane, 3, 5
1 Brno, vii, 3, 24, 25, 26, 27, 50, 51, 52, 53, 54, 75,
113, 143, 165, 166, 167, 186, 187, 204, 205, 206,
1G, 182 232, 251, 355, 356, 357, 358, 360, 361, 362
Bulgaria, 233
A
C
A, 105, 184, 220, 264, 311
aab, 344 calculus, vii, 3, 4, 5, 6, 10, 19, 22, 23, 24, 26, 27, 28,
Abelian, 48, 261 29, 35, 50, 51, 52, 53, 54, 78, 84, 96, 97, 99, 102,
achievement, 102 112, 113, 114, 175, 209, 210, 217, 316, 317, 321,
ad hoc, 120, 316, 317 322, 338, 339, 341, 342, 348, 349, 350, 351, 356,
age, 356, 357, 358 357
AIP, 54, 359, 361 CAM, 138
Algebra of differential invariants, 246 Cartan form, 28, 32, 35, 49, 51, 54, 58, 79, 80, 82,
Algebraic method, 250 84, 88, 90, 91, 257
Algebraic methods, 250 Cartan tensor, 299, 332, 333, 337
algorithm, 110, 139 casting, vii
alternative, vii, 6, 69, 137, 262, 270, 288, 322, 328 CCC, 266
alternatives, 221 Central Europe, 187
AMS, 274 charge density, 126
Amsterdam, 25, 26, 50, 51, 74, 139, 165, 258, 259 charged particle, 117, 118, 119, 125, 127
analog, 6 Charged string, 125
angular momentum, 257, 277, 286, 287 classes, 7, 22, 48, 49, 212, 303, 304, 348
anisotropy, 233 classical, 3, 5, 14, 23, 26, 28, 35, 42, 49, 51, 57, 58,
annihilation, 65 59, 78, 83, 84, 85, 86, 117, 120, 124, 128, 129,
anomalous, 213, 216 130, 137, 139, 144, 147, 149, 150, 151, 152, 157,
ants, 246 158, 159, 160, 161, 162, 176, 183, 184, 189, 190,
application, 3, 5, 6, 14, 18, 29, 32, 37, 87, 172, 180, 191, 194, 195, 197, 198, 200, 210, 211, 217, 219,
197, 210, 247, 312, 313 221, 224, 225, 228, 238, 275, 277, 279, 290
argument, 193, 199, 320, 323 classical mechanics, 35, 277
assumptions, 332, 336, 338 classification, 47, 111, 145, 149, 150, 190, 191, 194,
Australia, 27, 49, 341, 349 197, 203, 204, 205, 206, 318, 319
availability, 318, 326, 327 closure, 37, 306
Co, 139, 140, 234
Codazzi syzygy, 222
B collaboration, 209
communication, 114, 250
behavior, 239 community, viii
Belgium, 261, 315 commutativity, 22, 108, 150, 156
Bianchi identities, 316, 319, 330, 331 compatibility, 238, 247, 249, 250, 334, 343
boson, 5 complement, 101
Boston, 51, 73, 232
364 Index
complexity, 210 172, 173, 177, 211, 214, 258, 283, 319, 324, 325,
complications, 212, 333 328, 329, 332, 335
components, 9, 10, 28, 33, 35, 38, 39, 43, 45, 46, 48, deformation, 31, 279, 288, 302, 304, 305, 311, 312,
59, 60, 66, 121, 125, 127, 134, 175, 193, 199, 212, 344
213, 214, 215, 224, 238, 240, 256, 262, 264, 265, degenerate, 190, 192, 195, 198, 200, 204, 284, 348
268, 269, 271, 287, 304, 311, 312, 316, 317, 325, degrees of freedom, 57, 58, 59
328, 329, 337, 338, 344, 345 denoising, 210
composition, 30, 60, 61, 146, 147, 153, 179, 180, density, 117, 119, 121, 122, 123, 124, 125, 126, 127,
181, 185, 239, 243, 318 257
computation, 68, 95, 119, 134, 135, 210, 219, 220, dependent variable, 77, 99, 211, 219
329, 332, 337 derivatives, 19, 41, 49, 59, 60, 64, 66, 71, 73, 77, 81,
computing, 110 83, 84, 102, 104, 108, 109, 110, 121, 143, 150,
concentrates, 262 156, 177, 205, 212, 215, 216, 217, 219, 222, 225,
concrete, 3, 18, 145, 190 227, 237, 239, 240, 242, 244, 246, 256, 257, 258,
configuration, 86, 87, 118, 120, 121, 122, 123, 124, 265, 269, 317, 329, 333, 337, 339, 356
126, 130, 132, 262, 277, 281, 282, 284, 287, 288, deviation, 295, 300, 301
304 dichotomy, 238
confusion, 60, 338 differential equations, 27, 38, 42, 43, 44, 46, 54, 110,
conjecture, 69, 73, 100, 249 111, 192, 193, 199, 209, 219, 227, 234, 237, 238,
conjugation, 89, 248 248, 251, 262, 270, 280, 281, 282, 323, 339, 342,
Connection, 320 343, 350
conservation, 5, 13, 14, 15, 17, 18, 23, 26, 27, 29, 32, Differential invariant, 114, 166, 206, 232, 234, 237,
43, 47, 111, 165, 235, 257, 259, 262 358, 360
Conservation law, 18 differentiation, 120, 123, 162, 212, 216, 217, 220,
constraints, 25, 26, 41, 42, 55, 64, 115, 249 227, 229, 243, 248, 256, 258, 347
construction, 19, 58, 59, 103, 110, 120, 168, 172, dilation, 320
173, 175, 180, 183, 196, 209, 210, 211, 212, 213, Discrete mechanics, 97
218, 223, 271, 305, 306, 307, 322, 323, 326, 327, discrete variable, 94
348 discretization, 86, 94, 95
continuity, 112, 124, 145, 153, 156, 294, 297, 299 distribution, 38, 47, 89, 90, 92, 96, 100, 101, 111,
contractions, 29, 103, 177, 184, 255 271, 326, 328
control, 210 divergence, 43, 113, 222, 359
convex, 227, 231, 232, 294, 295, 300 duality, 86, 203, 231, 244, 320, 321, 324, 325
Copenhagen, 24 dynamical properties, 278
cosine, 278, 295 dynamical system, 5, 57, 225, 227, 261, 262, 263,
couples, 63, 66, 70 265, 267, 274
coupling, 157 dynamical systems, 261, 262, 274
covering, viii, 144, 145, 153, 237, 258, 304
CRM, 250
C-spectral sequence, 26, 100, 106, 111, 112 E
curiosity, 6
Curvature, 155, 205, 279, 284, 328 Education, 113
cycles, 287 electromagnetic, 118, 125, 126, 127, 152, 157, 158
Czech Republic, 3, 25, 26, 27, 53, 77, 113, 114, 143, electromagnetism, 127
163, 167, 186, 189, 349, 357, 358, 359, 361, 362 energy, 14, 124, 125, 257, 271, 279, 286, 287, 341
energy-momentum, 128, 257, 358
EP-1, 20, 106
D equality, 66, 67, 107, 110, 283, 296, 347
equilibrium, 210
danger, 60 Euclidean space, 190, 238, 277, 278, 281, 284, 287,
De DonderHamilton equations, 39, 40, 41 294, 296, 297, 299, 309, 310
decomposition, viii, 5, 10, 21, 22, 28, 30, 31, 32, 33, Euclidean-invariant variational problem, 221, 222
65, 103, 104, 203, 213, 239, 242, 246, 302, 304, Eulerian, 217, 220, 221, 222, 229
305, 306, 307, 308, 313, 319, 320, 321, 327, 344, Eulerian operator, 217, 220, 221, 222
357, 358 Euler-Lagrange equations, 38, 58, 69, 78, 87, 88, 92,
deficiency, 333 93, 94, 95, 218, 219, 221, 232, 267, 268, 280, 336,
definition, 4, 5, 6, 8, 11, 13, 22, 23, 32, 37, 45, 58, 337, 342
59, 60, 64, 69, 84, 101, 102, 105, 135, 145, 153, evolution, 15, 209, 210, 225, 226, 227, 228, 230,
231, 232, 234, 287, 343
Index 365
g-natural metric, 194, 195, 196, 197, 201, 204, 205,

F 206
google, 275
failure, 344 grants, 49
family, 21, 28, 31, 35, 41, 44, 47, 92, 111, 145, 146, graph, 248, 250
148, 153, 155, 178, 190, 195, 196, 197, 200, 201, Grassmannian bundle, 61
217, 224, 280, 282, 283, 291 gravitation, 257, 259, 288
fiber, 14, 15, 57, 58, 59, 62, 85, 118, 120, 121, 123, gravitational field, 158
125, 146, 147, 148, 158, 170, 172, 173, 178, 181, gravitational stress, 128
184, 190, 240, 356 gravity, 41
fiber bundles, 85, 190, 356 groups, viii, 30, 45, 48, 164, 232, 233, 238, 239, 262,
fibers, 9, 168, 184, 240, 294 266, 275, 356
field theory, 5, 6, 8, 14, 15, 26, 27, 35, 36, 41, 43, 49, G-structure, 175, 180, 187
51, 53, 54, 55, 57, 58, 59, 76, 115, 118, 120, 121, guidance, 294, 317
124, 128, 129, 130, 137, 139, 356, 358
filament, 210, 225, 230, 232, 233
filtration, 212 H
Finsler arc length, 294
Finsler geometry, 81, 193, 293, 294, 295, 313, 314, H1, 223
339, 340, 342, 343, 348 H2, 108, 223, 278, 284, 285, 287, 288, 290
Finsler inverse problem, 343, 345, 346, 348 Hamilton equations, 27, 29, 38, 39, 42, 47, 130, 131,
Finsler norm, 293, 294 134, 135, 136, 137
Finsler space, 293, 294, 295, 297, 299, 300, 301, Hamilton extremal, 38, 39, 40, 41, 47, 52, 357
302, 303, 304, 305, 306, 307, 309, 310, 311, 312, Hamiltonian, 25, 28, 29, 38, 39, 41, 46, 47, 48, 49,
313, 316, 317, 339, 348 50, 51, 53, 54, 58, 74, 75, 76, 96, 130, 133, 139,
first principles, 120, 124 210, 217, 218, 220, 221, 222, 226, 229, 230, 233,
flatness, 180, 288 248, 262, 275, 356
flow, 13, 17, 30, 101, 144, 149, 150, 156, 168, 169, Hamiltonian system, 38, 39, 41, 46, 47, 210
173, 176, 183, 210, 224, 225, 226, 227, 228, 229, Hamilton-Jacobi, 84, 129, 130, 131, 133, 134, 135,
230, 231, 256, 258, 263, 344 137, 138, 139, 140, 234
fluid, 96, 124, 127 hands, 315
F-metric, 193, 194, 195, 196 heart, 222
Fourier, 51, 74, 75, 97, 188, 233 heat, 231, 232
France, 23, 26, 114, 233, 234 Helmholtz conditions, 99, 321, 331, 335, 341, 342,
freedom, 14, 57, 58, 59, 121, 249 343, 345, 346, 347, 348, 349, 350
friendship, 230 Helmholtz equation, 65, 69, 72, 73
Helmholtz form, 27, 48
hemisphere, 308
G heredity, 190, 197
Hermitian connection, 157
gas, 200 Hessian matrix, 335, 342
gauge, 121, 143, 144, 152, 154, 156, 157, 160, 162, Higgs, 289
163, 165, 166, 181, 184, 186 Hilbert, 41, 82, 124, 125, 127, 255, 258, 362
gauge group, 144, 152, 156, 157 Hilbert form, 82, 124, 125, 127
gauge invariant, 143, 144, 157, 162 Holland, 51, 164, 165, 356
Gauge invariant Lagrangian, 162 homeomorphic, 304
gauge theory, 152 homogeneity, 82, 293, 294, 297, 334, 335, 346, 347
Gauge-natural bundle, 153, 163, 186 homomorphism, 170, 181, 257, 273, 274
Gauge-natural bundle functor, 153 household, 338
Gauss-Bonnet, 313 Hungary, 24, 25, 164, 165, 358
Gaussian, 221, 222, 250 hydrodynamics, 96
gene, 11, 57, 58, 190 hydrogen, 278, 289
General Relativity, 41, 361 hyperbolic, 277, 278, 281, 283, 284, 290
generalization, 18, 22, 50, 57, 58, 85, 152, 158, 163, hypothesis, 58, 92, 171, 172, 174, 175
181, 186, 270, 278, 293, 296, 328, 336
generalizations, 11, 57, 58, 190
generators, 213, 214, 218, 240, 241, 244, 284 I
Geodesic, 84, 97
globalization, 37, 85 IDEA, 3
366 Index
identification, 85, 121, 123, 137, 158, 169, 177, 180, Jordan, 348
184, 319, 326 justification, 322
identity, 21, 66, 100, 123, 158, 162, 163, 190, 192,
219, 221, 246, 257, 258, 263, 270, 318, 320, 321
Illinois, 207 K
IMA, 231
image analysis, 234 kernel, 89, 102, 104, 170, 183
images, 81, 111, 224, 311 kinetic energy, 271, 279
immersion, 61, 62 Kirchhoff, 233
inclusion, 102, 103, 106, 107, 108, 180, 185 Krupka theorem, 32
incompatibility, 333
independence, 71, 215, 348
independent variable, 7, 211, 212 L
Indiana, 234, 313
indices, 29, 33, 35, 36, 40, 59, 60, 63, 66, 67, 70, L1, 14, 86, 110, 118, 193, 199
100, 255, 273, 324, 329, 346 L2, 14, 110, 192, 198
induction, 65, 69, 72, 171, 172, 174, 175, 176, 213 Lagrange structure, 10, 14, 16
inequality, 67, 313 Lagrangian, 4, 5, 7, 8, 10, 11, 12, 13, 14, 15, 16, 17,
infinite, 5, 100, 105, 107, 111, 115, 124, 146, 165, 18, 24, 26, 27, 28, 29, 31, 32, 33, 34, 35, 36, 37,
225, 227, 238, 242 38, 39, 40, 41, 42, 43, 44, 45, 47, 48, 49, 50, 51,
injection, 168, 170, 174, 180, 181, 182 54, 55, 57, 58, 65, 66, 67, 68, 69, 70, 71, 73, 74,
insight, 123, 320 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87,
inspection, 326 88, 90, 91, 92, 93, 94, 95, 96, 99, 109, 110, 113,
inspiration, 230, 341 115, 117, 118, 119, 120, 122, 123, 124, 125, 127,
integration, 19, 29, 46, 47, 119, 219, 220, 228 143, 144, 145, 147, 149, 150, 151, 152, 153, 155,
interaction, 118, 119, 120, 122, 125, 126, 157, 158, 157, 159, 161, 162, 163, 165, 219, 220, 221, 222,
161, 163, 164, 166, 356 223, 229, 233, 235, 248, 261, 262, 267, 268, 269,
interactions, 360, 362 270, 271, 272, 273, 274, 279, 280, 281, 282, 342,
interface, 210 343, 345, 349, 350, 355, 357, 359
interpretation, 152, 166, 328, 330, 332 Lagrangian density, 86, 117, 118, 119, 123, 124, 125
interval, 194 Lagrangian formalism, 26, 65, 69, 115, 235
intrinsic, vii, 28, 32, 34, 38, 86, 100, 108, 212, 222, Lagrangian formulation, 26
225, 226, 227, 230, 287, 315, 317, 324, 325, 326, language, 18, 78, 210, 212, 248
330, 331, 335, 336, 338, 343 law, 13, 27, 43, 123, 257
Invariant Lagrangian, 151, 152, 157, 162, 275 laws, 5, 14, 15, 17, 18, 23, 26, 29, 32, 47, 111, 235,
Invariant Theory, 163 257, 262
Invariantization, 211, 217 lead, 32, 100, 111, 122, 210, 243, 278, 317
invariants, vii, viii, 49, 114, 165, 166, 190, 191, 192, Legendre coordinates, 39, 40, 41
193, 199, 203, 206, 209, 210, 212, 213, 214, 215, Legendre transformation, 39, 41, 53
217, 218, 219, 220, 221, 222, 223, 224, 225, 226, Leibniz, 257
227, 229, 230, 232, 233, 234, 237, 238, 239, 240, Lepage congruences, 85, 86
241, 242, 243, 244, 245, 246, 247, 251, 256, 258, Lepage equivalent of a Lagrangian, 4, 5, 8, 11, 12,
259, 356, 357, 358, 360 18, 28, 45, 57, 77
Inverse problem of the calculus of variations, 341 Lepage form, vii, 3, 4, 5, 6, 8, 11, 12, 13, 18, 22, 23,
inversion, 211 25, 27, 28, 29, 32, 35, 37, 41, 42, 49, 53, 64, 79,
involution, 168, 179, 180, 183, 186 114, 359
isomorphism, 30, 42, 107, 108, 169, 176, 181, 182, Lie algebra, 154, 156, 210, 239, 256, 257, 258, 259,
191, 258 263, 273, 274, 284, 285, 356
isotropy, 211, 270 Lie algebroid, 167, 168, 182, 183, 350
Italy, 50, 52, 99, 359 Lie group, 61, 144, 146, 152, 156, 157, 160, 190,
iteration, 178, 185 209, 211, 214, 232, 237, 238, 239, 247, 262, 263
Lie-Tresse theorem, 238, 239
linear, vii, 14, 40, 62, 64, 65, 66, 70, 73, 78, 81, 94,
J 108, 144, 147, 154, 156, 157, 158, 159, 160, 161,
164, 165, 167, 168, 169, 170, 173, 174, 175, 179,
Jacobi endomorphism, 321 182, 184, 186, 187, 188, 189, 190, 193, 197, 198,
Jacobian, 121, 128 199, 200, 201, 205, 206, 213, 214, 223, 235, 239,
January, vii, viii 241, 257, 270, 277, 279, 286, 294, 295, 297, 299,
Japan, 189, 314, 339 301, 303, 304, 305, 310, 311, 315, 316, 320, 322,
Index 367
323, 324, 325, 326, 327, 328, 329, 331, 332, 333, mixing, 329
334, 335, 336, 337, 338, 341, 342, 343, 344, 345, models, 225, 288
346, 348, 350, 356 modules, 99
Linear connection, 158, 316, 349 momentum, 97, 125, 257, 262, 268, 269, 273, 274,
linear function, 279 277, 286, 287, 290
links, 320 monograph, 167, 258, 262
L-metric, 199, 200 Moscow, 54, 164, 234, 251
localization, 145, 146 motion, 4, 5, 10, 14, 43, 119, 127, 233, 237, 238,
locus, 238, 308, 309 239, 255, 262, 277, 278, 280, 281, 282, 286, 287,
London, 54, 84, 96, 140, 164, 166, 188 288, 307, 308, 312, 356
motivation, 29, 45, 49, 105, 158
Moving frame, 209, 234, 250
M multiples, 97, 178
multiplication, 60, 146, 147, 239, 248, 285
M1, 168, 285, 302, 303, 307, 308 multiplier, 342, 343, 346, 348
machinery, 118, 212, 321 multivariate, 220
manifold, 3, 4, 5, 6, 7, 8, 9, 11, 14, 15, 16, 23, 24, 25,
27, 28, 29, 30, 31, 42, 52, 53, 57, 58, 59, 60, 61,
63, 70, 71, 73, 77, 78, 79, 80, 81, 82, 85, 86, 89, N
97, 100, 106, 110, 111, 112, 113, 127, 130, 132,
133, 134, 137, 139, 143, 144, 145, 146, 147, 148, natural, vii, 3, 5, 23, 47, 48, 58, 61, 71, 85, 86, 89,
149, 151, 153, 155, 162, 163, 165, 166, 167, 168, 92, 100, 101, 102, 103, 105, 106, 107, 108, 111,
178, 179, 180, 184, 186, 189, 190, 191, 192, 193, 143, 144, 145, 146, 147, 148, 149, 150, 151, 152,
194, 195, 196, 197, 198, 200, 201, 202, 203, 204, 154, 155, 156, 158, 159, 160, 161, 162, 164, 165,
205, 206, 207, 210, 211, 223, 233, 237, 241, 245, 166, 167, 168, 173, 176, 177, 178, 184, 186, 189,
247, 249, 255, 257, 258, 262, 263, 271, 278, 281, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199,
283, 293, 294, 295, 301, 302, 303, 304, 305, 306, 200, 201, 202, 203, 204, 237, 240, 255, 257, 258,
308, 309, 313, 314, 316, 320, 340, 343, 348, 351, 265, 280, 282, 288, 293, 326, 328, 329, 332, 344
355, 356, 357, 358 Natural bundle, viii, 143, 145, 165, 166, 206
manifolds, 3, 4, 5, 8, 11, 16, 23, 24, 25, 27, 28, 29, Natural differential operator, 145, 150, 152, 156, 207
52, 53, 59, 61, 77, 78, 85, 97, 100, 111, 112, 113, Natural Lagrangian, 24, 145, 147, 149, 150, 151,
137, 139, 144, 145, 151, 153, 162, 165, 166, 168, 153, 155, 156, 157, 159, 161, 163, 164, 165, 357
178, 186, 190, 194, 202, 204, 205, 206, 207, 233, Natural transformation, 205, 206
241, 247, 249, 255, 258, 262, 278, 281, 301, 306, Netherlands, 234
309, 313, 314, 351, 355, 356, 357, 358 New York, 24, 54, 113, 128, 163, 205, 206, 207,
manners, 125 231, 232, 233, 235, 251, 313, 350
mapping, 4, 5, 9, 10, 11, 12, 16, 19, 22, 24, 29, 30, Newtonian, 23, 118
39, 40, 41, 48, 52, 148, 149, 162, 212, 297, 299, Noether equation, 42
302, 311, 355, 356, 359, 360 Noether theorem, vii, 27, 28, 262
Mathematical Methods, 138 non-Abelian, 275
mathematicians, 256 nonlinear, 26, 54, 55, 225, 230, 231, 233, 235, 248,
mathematics, vii, viii, 164, 259, 342 250, 251, 315, 318, 319, 320, 321, 323, 324, 327,
matrix, 61, 95, 96, 147, 149, 170, 171, 175, 202, 214, 328, 329, 330, 334, 339, 344
220, 221, 244, 245, 263, 264, 267, 269, 273, 284, nonlinearities, 231
285, 309, 335, 342 normal, 162, 210, 221, 224, 225, 226, 227, 229, 230,
metals, 225 336, 342, 348
metric, 14, 16, 18, 120, 121, 122, 123, 124, 125, 126, Normal flow, 227
127, 128, 152, 165, 166, 189, 190, 191, 192, 193, normalization, 209, 211, 217
194, 195, 196, 197, 198, 199, 200, 201, 202, 203, Norway, 237
204, 205, 207, 240, 241, 244, 248, 257, 271, 272, numerical analysis, 210
279, 280, 281, 284, 285, 293, 295, 296, 297, 299,
301, 310, 316, 322, 332, 334, 335, 336, 337, 338,
339, 341, 342, 348, 349, 350, 351, 355, 356, 358 O
mines, 288
Ministry of Education, 23, 49, 163, 186 object recognition, 231
Minkowski space, 5, 295, 299, 302, 304, 305, 311, observations, 345
312 obstruction, 99, 176
Minkowski tensor, 124, 125, 127 operator, 4, 5, 6, 19, 21, 44, 48, 71, 86, 99, 107, 108,
Minnesota, 209 109, 110, 111, 143, 148, 149, 150, 155, 156, 158,
368 Index
159, 160, 161, 164, 167, 168, 172, 176, 177, 184, private, 114, 250
213, 217, 218, 220, 221, 222, 223, 225, 226, 227, program, 247
228, 229, 230, 241, 247, 248, 255, 315, 316, 323, projector, 60, 89, 133, 134, 326, 327
324, 326, 331, 333, 334, 343 propagation, 234
Operators, 145, 148, 150, 155, 156, 176, 184 property, 6, 11, 19, 23, 32, 36, 39, 41, 42, 45, 47, 59,
Oproiu metric, 196, 197 78, 79, 105, 107, 108, 149, 156, 197, 278, 282,
optics, 210, 225 284, 287, 288, 294, 295, 300, 302, 303, 308, 309,
orbit, 307 320, 326, 327, 329, 330, 331, 332, 333, 334, 336,
ordinary differential equations, 44, 45, 53, 261, 341, 337, 338, 342
342, 349, 350, 351 proposition, 66, 67, 68, 69, 103, 104, 108, 332, 347
orientation, 81, 294, 296 pseudo, 162, 237
orthogonality, 300
oscillator, 278, 290
Q
P quantum, 157, 158, 164, 210, 278, 289, 290
quantum mechanics, 164, 210
Pacific, 313 quantum structure, 164
packaging, 120, 125 query, 323
pairing, 86
PAN, 187
paper, vii, 3, 5, 6, 8, 11, 13, 23, 27, 28, 29, 40, 41, R
53, 58, 70, 73, 77, 84, 88, 99, 100, 127, 129, 130,
135, 144, 145, 152, 157, 167, 168, 195, 196, 209, radius, 279, 281, 296, 307, 313
211, 238, 240, 249, 256, 257, 262, 270, 287, 288, range, 130, 170, 210, 244
295, 309, 315, 317, 320, 334, 335, 336, 338, 341, real numbers, viii, 194
360, 362 reasoning, 109, 277, 288
parabolic, 278 recall, 27, 32, 38, 77, 100, 110, 117, 144, 145, 152,
parallelism, 327 153, 168, 175, 180, 183, 190, 193, 218, 238, 265,
parameter, 14, 15, 31, 92, 226, 227, 228, 278, 281, 283, 288, 335
288, 294, 307, 308, 342 recalling, 28, 191, 197, 203, 318
Paris, 49, 50, 51, 76, 84, 96, 112, 114, 186, 188, 231, recognition, 231
258, 259 reconcile, 120
partial differential equations, vii, 45, 47, 51, 75, 113, reconstruction, 261, 263, 264, 267, 270, 271, 274
150, 151, 162, 210, 224, 227, 231, 233, 237, 238, recurrence, 181, 210, 213, 214, 215, 216, 218, 222,
242, 250, 251, 360 349
particles, 10, 14, 118, 124, 125, 127, 291, 360, 362 recursion, 225, 228, 230
periodic, 228 Reduction, 69, 86, 96, 144, 150, 151, 156, 157, 158,
pH, 212 159, 164, 172, 173, 174, 175, 180, 183, 185, 186,
phase space, 58, 280, 282 250, 261, 262, 263, 265, 266, 268, 269, 270, 273,
physicists, 256, 257 274, 275, 278, 279, 281, 283, 284, 285, 287, 289,
physics, vii, viii, 3, 5, 10, 18, 112, 113, 128, 139, 291, 299, 337, 359
145, 181, 250, 259, 342, 362 Reduction theorems, 164
planar, 224, 300 reflection, 287, 288
plasma, 127, 128 regular, vii, 27, 29, 38, 40, 41, 47, 51, 54, 61, 82, 94,
play, 4, 27, 28, 29, 42, 46, 132, 175, 181, 210, 226, 96, 105, 179, 194, 211, 238, 244, 267, 268, 280,
247, 267, 316 303, 309, 321
pleasure, 111, 230, 341 Regular Lagrangian, 38, 52, 357
Poisson, 233, 256, 262 Regularization, 231
Poland, 255 relationship, 59, 111, 324, 326
polar coordinates, 279, 285, 290 Relativistic particle, 14
Polyakov action, 18 relativity, 152, 210, 247
polynomial, 104, 108, 110, 151, 237, 238, 248, 251, relevance, 70, 321, 322
356 repetitions, 308
polynomials, 110, 241 Representation of the variational sequence, 24, 52,
pond, 284 113
potential energy, 14, 271 research, vii, viii, 23, 99, 100, 111, 112, 163, 209,
power, 102, 124, 209, 241 230, 274, 315
powers, 102 resolution, 19, 26, 48, 106
Index 369
Ricci bundle, 159, 161 symbolic, 7, 210

rigidity, 197 symbols, 147, 149, 193, 199, 246, 348
Rome, 96 symmetry, 5, 48, 59, 66, 70, 73, 101, 143, 210, 219,
rotations, 286, 287, 302, 307 224, 234, 240, 249, 257, 261, 262, 263, 266, 271,
Routhian, 261, 269, 270, 271, 273 274, 294, 300, 337
Russian, 53, 54, 112, 113, 164, 204, 251, 316, 360, symplectic, 6, 50, 58, 77, 130, 256
362 systems, 4, 5, 10, 14, 25, 29, 36, 38, 41, 47, 49, 51,
53, 58, 73, 76, 85, 86, 117, 118, 126, 192, 193,
198, 199, 203, 210, 233, 234, 251, 261, 262, 266,
S 270, 278, 286, 290, 302, 304, 311, 312, 342, 349
Sasaki lift, 194

scalar, 41, 118, 119, 121, 122, 124, 125, 127, 189, T
196, 197, 202, 204, 238, 240, 241, 317, 318, 330,
331 tangible, 332
scalar field, 118 technology, 211
school, 5, 6, 206 TEM, 124, 127
searching, 4 tension, 126
segmentation, 210 tensor field, 158, 161, 176, 204, 255, 257, 315, 316,
selecting, 317 317, 320, 322, 324, 325, 328, 330, 335, 344, 348
SEM, 118, 119, 120, 123, 124, 125, 126, 127 tensor products, 177
separation, 289 textbooks, 361
September 11, 359 theory, vii, 3, 4, 5, 6, 7, 8, 11, 18, 23, 25, 27, 28, 29,
shape, 234 37, 39, 40, 41, 42, 46, 47, 48, 49, 52, 54, 57, 79,
SIGMA, 232, 290 82, 83, 84, 97, 106, 118, 124, 128, 129, 130, 137,
sign, 60, 197, 241, 263, 284, 296, 315, 329 139, 140, 143, 145, 147, 151, 153, 156, 165, 166,
similarity, 326, 331 167, 168, 176, 178, 180, 181, 184, 187, 190, 206,
sine, 278, 286 209, 210, 233, 235, 238, 239, 240, 256, 257, 259,
Singapore, 25, 51, 52, 53, 54, 75, 114, 186, 187, 205, 262, 275, 278, 288, 294, 295, 316, 318, 319, 323,
314, 357, 360, 361 338, 355, 356, 357, 358, 359, 360, 362
singular, 42, 139, 238, 240, 242, 245 third order, 242, 244
singularities, 224 three-dimensional, 233, 277, 278, 279, 280, 281,
Slovakia, vii, 358, 362 284, 290
soliton, 210, 225, 232 three-dimensional space, 281, 284
solutions, 29, 47, 94, 95, 128, 137, 151, 152, 162, time, 11, 14, 15, 28, 86, 94, 95, 96, 118, 120, 123,
192, 193, 199, 233, 234, 241, 249, 250, 263, 280, 124, 130, 152, 164, 224, 227, 256, 268, 287, 316,
281, 282, 312, 313, 341, 342, 348 324, 327, 335, 349
Source form, 48 title, 315, 334
spacetime, 58, 118, 120, 121, 123, 124, 125, 126, Tokyo, 165, 189, 205, 206, 207
127, 152, 158, 257, 288 topology, vii, 83, 295, 305, 355
Spain, 117, 129, 138, 277, 359 Torsion, 169, 187, 328
special relativity, 257 tracking, 210, 233
spectrum, 241, 247, 248 trajectory, 118, 120, 288
speed, 234 trans, 304
spheres, 277, 283, 290, 314 transcription, 27
spin, 152, 166, 257 transfer, 29, 30
SPT, 359 transformation, 14, 17, 39, 41, 42, 53, 62, 63, 147,
stability, 238 148, 150, 155, 156, 157, 190, 192, 193, 194, 195,
stages, 88, 335, 341 196, 198, 199, 200, 201, 203, 204, 209, 213, 230,
stress, 33, 38, 47, 48, 64, 125, 128 256, 257, 294, 301, 302, 303, 304, 305, 309, 310
Stress-energy-momentum tensor, 128 transformations, 5, 13, 14, 15, 18, 36, 42, 43, 148,
students, 6 152, 155, 157, 189, 190, 191, 192, 193, 205, 206,
subgroups, 211 207, 217, 251, 259, 265, 294, 301, 303, 304, 309,
supersymmetric, 23 311, 312, 313, 355
supervision, vii transition, 262, 304, 310, 311
supply, 212 transitions, 305, 311
surface area, 16, 222, 223 translation, 14, 112, 113, 171, 174, 185, 204, 221,
surprise, 325, 336 228, 257, 304, 316, 362
swarm, 118 transport, 316, 323
370 Index
Tresse derivative, 237, 239, 242, 246 89, 90, 95, 100, 101, 111, 130, 131, 132, 133, 134,
TTM, 179, 294, 295 135, 137, 138, 143, 144, 149, 150, 152, 154, 156,
Turkey, 54 157, 158, 160, 161, 162, 166, 168, 169, 172, 173,
two-dimensional, 6, 278, 283, 284, 285, 288, 290, 175, 176, 183, 184, 191, 192, 194, 198, 203, 212,
295, 296 213, 216, 224, 225, 226, 230, 238, 239, 240, 241,
two-dimensional space, 278 242, 244, 248, 255, 256, 257, 258, 262, 263, 264,
265, 266, 267, 268, 269, 271, 272, 277, 279, 280,
281, 282, 283, 285, 287, 293, 294, 295, 299, 304,
U 311, 312, 313, 317, 318, 319, 320, 321, 322, 323,
325, 328, 329, 330, 333, 343, 344, 356
unification, 317 velocity, 60, 280, 288, 312
uniform, 95 Victoria, 27
university education, 28 visible, 161, 238
Utah, 23, 49, 73, 112, 231 vision, 210, 228, 234
Utiyama-like theorem, 158, 160, 164 vortex, 210, 225, 230, 232
V W
valence, 255 Warsaw, vii, 24, 357
values, 78, 80, 83, 86, 90, 119, 123, 126, 147, 151, wave equations, 250
158, 162, 169, 170, 171, 180, 249, 268, 278, 281, weak interaction, 360, 362
282, 283, 284, 286, 287 web, 58
variable, 121, 123, 195, 196, 279 Weierstrass excess function, 77, 78, 83
variables, 58, 63, 66, 70, 71, 73, 93, 121, 128, 200, Weierstrass necessary condition, 83
201, 212, 269, 286, 289 Weyl tensor, 358
variation, 5, 6, 8, 10, 25, 27, 28, 31, 32, 34, 37, 42, winter, 206
79, 85, 86, 96, 256, 357 writing, 62, 122, 239
Variation, 31, 32, 73, 75, 76
Variational equations, 53
Variational sequence, 25, 26, 52, 53, 99, 113, 114, Y
232, 357, 358, 359
vector, 4, 8, 9, 10, 11, 12, 13, 19, 21, 28, 29, 30, 31, Yang-Mills, 162, 163, 275
32, 34, 37, 38, 39, 42, 43, 46, 47, 63, 76, 79, 86, yield, 100, 102, 104, 105, 123, 215, 300, 302, 343

Variations, Geometry and Physics in Honour of Demeter Krupka's Sixty-Fifth Birthday

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Variations, Geometry and Physics in Honour of Demeter Krupka's Sixty-Fifth Birthday

Uploaded by

Copyright:

Available Formats

VARIATIONS , G EOMETRY AND P HYSICS

Nova Science Publishers, Inc.

NOTICE TO THE READER

Library of Congress Cataloging-in-Publication Data

Published by Nova Science Publishers, Inc. New York

PART I VARIATIONAL P RINCIPLES ON J ET B UNDLES 1

Chapter 1 Lepage Forms in Variational Theories: From Lepages Idea

Chapter 2 Lepage Forms in the Calculus of Variations 27

Chapter 3 On a Generalization of the Poincare-Cartan Form

Chapter 4 Krupkas Fundamental Lepage Equivalent and

Chapter 5 Lepage Congruences in Discrete Mechanics 85

Chapter 6 Finite Order Variational Sequences: a Short Review 99

Chapter 7 Concatenating Variational Principles and

Chapter 8 A Geometric Hamilton-Jacobi Theory for Classical

PART II N ATURAL B UNDLES AND D IFFERENTIAL I NVARIANTS 141

Chapter 9 Natural Lagrangian Structures 143

Chapter 10 Connections on Higher Order Frame Bundles

Chapter 11 Natural Lifts in Riemannian Geometry 189

Chapter 12 Invariant Variational Problems and Invariant

Chapter 13 Differential Invariants of the Motion Group Actions 237

PART III D IFFERENTIAL E QUATIONS AND G EOMETRICAL

Chapter 14 Remarks on the History of the Notion

Chapter 15 Second-Order Differential Equation Fields with Symmetry 261

Chapter 16 Dimensional Reduction of Curvature-Dependent

Chapter 17 Direct Geometrical Method in Finsler Geometry 293

Chapter 18 Linear Connections Along the Tangent Bundle Projection 315

Chapter 19 On the Inverse Problem for Autoparallels 341

PART IV A PPENDIX 353

Demeter Krupka: List of Publications 355

This book is a commemorative volume to celebrate the

Olga Krupkova, David Saunders (Editors)

L EPAGE F ORMS IN VARIATIONAL T HEORIES :

physical theories. Theories developed on fibered manifolds with one-dimensional bases

the variational equivalence problem leading to the concept of Lepage equivalents of

2. Lepages Idea a Brief Historical Review

defined by the formula

= f (x, y; z1 , . . . , zn ; p1 , . . . , pn ; q1 , . . . , qn ) dx dy, (2.2)

and our aim is to find its stationary paths . The forms

i = dzi pi dx qi dy, 1 i n, (2.3)

i = Xi dx + Yi dy + Aij j , Aij + Aji = 0, 1 i, j n,

Xi , Yi , Aij being differentiable functions of (x, y, zj , pj , qj ). The relation

The result is as follows:

v : (x, y, z1 , . . . , zn ) (pi (x, y, z1 , . . . , zn ), qi (x, y, z1 , . . . , zn ))

are defined by the condition d[] = 0, where [] = v and [i ] = i v.)

3. Lepage Forms on Fibered Manifolds

3.1. Lagrange Structures, Variational Functionals

1 k r, 1 i1 , . . . , ik n, is the associated fibered chart on (J r Y, r , X).

where the components of are functions on J s Y , in general. Let be a -projectable

is the total derivative of a function f on J s Y with respect to xj .

vector field . A form on J s Y is called contact if J s = 0 for every section of .

= dy yi dxi , i1 = dyi1 yi1 j dxj ,

where 1 m, 1 i1 , . . . , is n, form the basis of 1-forms called the basis adapted

= L(xi , y , yi1 , . . . , yi1 ...ir ) 0 , 0 = dx1 . . . dxn .

The pair (, ) is so called Lagrange structure. Let U = (W ). Let U be a compact

is called the variational functional or variational integral over . In physics it is often

3.2. Lepage Equivalents of Lagrangians

for r-jets of all sections of at x, and every collection of vector arguments 1 , . . . , p

nX (J 1 Y ) E() 1Y (J 2 Y ), h(d) = E(h()) ,

Theorem. Let W Y be an open set. A form sn W is a Lepage form if and only

where f0 is a function, defined in coordinates by h = f0 0 , is a contact (n 1)-form,

The Lepage equivalent of a Lagrangian = L 0 is defined as such a Lepage form for

For mechanics (n=1) and of order r we have