3319621807

UNIPA Springer Series
GiovanniFalcone Editor
Lie Groups,
Differential
Equations, and
Geometry
Advances and Surveys
Editor-in-chief
Carlo Amenta, Universit di Palermo, Palermo, Italy
Series editors
Sebastiano Bavetta, Universit di Palermo, Palermo, Italy
Calogero Caruso, Universit di Palermo, Palermo, Italy
Gioacchino Lavanco, Universit di Palermo, Palermo, Italy
Bruno Maresca, Universit di Salerno, Fisciano, Italy
Andreas chsner, Griffith School of Engineering, Southport Queensland, Australia
Mariacristina Piva, Universit Cattolica Sacro Cuore, Piacenza, Italy
Roberto Pozzi Mucelli, Policlinico G.B. Rossi, Verona, Italy
Antonio Restivo, Universit di Palermo, Palermo, Italy
Norbert M. Seel, University of Freiburg, Germany, Germany
Gaspare Viviani, Universit di Palermo, Palermo, Italy
More information about this series at http://www.springer.com/series/13175
Giovanni Falcone
Editor
Lie Groups, Differential

Equations, and Geometry
Advances and Surveys
123
Editor
Giovanni Falcone
Dipartimento di Matematica e Informatica
University of Palermo
Palermo, Italy
ISSN 2366-7516 ISSN 2366-7524 (electronic)

ISBN 978-3-319-62180-7 ISBN 978-3-319-62181-4 (eBook)
DOI 10.1007/978-3-319-62181-4
Library of Congress Control Number: 2017952063
Springer International Publishing AG 2017

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
Printed on acid-free paper
This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
This volume collects articles from a number of authors who, mostly, met in the
context of the scientific project Lie Groups, Differential Equations, and Geometry,
which was supported by the European Unions Seventh Framework Programme
(FP7/20072013) under grant agreement no. 317721.
Besides the many secondments, it behooves me to mention the exciting atmo-
sphere of the conferences in Batumi (1722 June 2013), Palermo (30 June5 July
2014), and Chongqing (1218 June 2015). The Georgian and Chinese hospitality
was undoubtedly unrivaled, and while the forthcoming conference in Modica (814
June 2016) will be a pale shadow of those in Batumi and Chongqing, it will still be
an occasion for fruitful discussions.
Last but not least, I would like to pay credit to the contributions of mathemati-
cians of excellent caliber who kindly and ardently participated in this task, even if
not supported by the project. It is an honor for me to introduce such a collection and
a pleasure to thank the authors for granting me the privilege of their friendship.
Palermo, Italy, February 11, 2016

Giovanni Falcone
A few days before this volume went to the editor, my dearest friend and mentor,
Karl Strambach, passed away. I cannot but dedicate to Karl, who was undoubtedly
a main actor in the project, the small efforts I made in editing this volume.
v
Introduction
Lie theory takes its name from the Norwegian mathematician Marius Sophus Lie
who, in 1871, at the age of 29, moved to Erlangen and visited Felix Klein to start
a fruitful collaboration that unfortunately ended in reciprocal accusations. Today,
Lie theory finds a major place in many branches of mathematics, both abstract and
applied. It is current not only important in algebra and geometry, but also plays a
significant role in some fields of physics, for example quantum theory, and in control
theory (cf., e.g., [1]). The reason for the wide presence of Lie theory probably lies
in the fact that it links several basic questions in mathematics, that is, linearization
and approximation, exponentiation, differentiation, and nonassociativity.
In Chap. 1, we begin from the very first definitions, giving short surveys of
basic topics in the theory of Lie algebras and in Representation theory, two
essays of which are given in Chaps. 2 and 3. In the former, the author deals
with the representation of the Heisenberg Lie algebra in an infinite dimensional
Hilbert space, which, in Quantum Mechanics, describes the commutation relation
between the position Q and the momentum P. In the latter, the author, led by
the most influential KazhdanLusztig conjecture posed in 1979, aims to describe
representation theory from a categorical point of view and shows how this approach
comes to determine irreducible characters, that is, the traces of the images of an
irreducible representation, which are a basic tool that summarizes much information
of the representation itself.
In Chap. 1, Sect. 1.1.3, the connection between Lie algebras and Lie groups
is sketched. After showing that the exponential map transforms the Heisenberg
Lie algebra into the Heisenberg group, we give the first few terms of the Baker
CampbellHausdorff formula, which is applied, in Chap. 4, to the problem of the
computation of order conditions for exponential splitting schemes, together with an
alternative approach for finding the order conditions by inspecting the coefficients
of leading LyndonShirshov words in an exponential function of a sum of Lie
elements. LyndonShirshov words produce a basis of a free Lie algebra (cf. [2, 3]),
and Chibrikov [4] generalized the Shirshov construction of a free basis for a Sabinin
algebra. On the other hand, in the words of Sabinin and Mikheev [5], a more
natural context for the BakerCampbellHausdorff formula is the completion of the
vii
viii Introduction
universal enveloping algebra of a relatively free Sabinin algebra on two generators,

thus indicating a connection with the theory of loops. Roughly speaking, a loop
is a group where associativity is, at best, replaced by some weaker algebraic law.
Extending previous results of Malcev and Kuzmin, Sabinin and Mikheev (see also
[6]) showed that the correspondence between Lie algebras and local Lie groups is
valid in the nonassociative case, as well, by defining an algebraic structure on the
tangent space of any analytic loop, which is called, in fact, Sabinin algebra.
In Chap. 1, Sect. 1.2.1, we give a short recap of the theory of loops and its
relation to the foundations of Geometry, a connection almost forgotten today.
In Chap. 6, the authors study half-automorphisms of Moufang loops and, more
generally, of dissociative 2-loops, that is, such that any two elements generate a
group, and introduce the well-known CayleyDickson construction of a sequence
of algebras of dimension 2n, whose multiplicative loops in dimensions 8 and 16
are the Moufang loop of octonions and the sedenion loop, respectively. Besides Lie
algebras and Cayley (or octonion) algebras, among nonassociative algebras we also
find Jordan algebras and Gerstenhaber algebras, which play an important role in
Mathematical Physics. Gerstenhaber algebras are a main object of Chap. 5, and in
Chap. 1, Sect. 1.2.2, we give their definition.
As among Lie groups one finds all closed subgroups of the Euclidean group of
motions, it is not surprising that Lie theory has found applications in control theory.
Chapter 7 conducts a survey of classic and recent results on left-invariant affine
control systems and the associated Hamilton-Poisson systems, extensively in the
cases of the three-dimensional Lie groups, the (2n C 1)-dimensional Heisenberg
groups, and the six-dimensional orthogonal group. Chapter 8, on the other hand, is
devoted to optimal control problems whose behavior is described by quasilinear
differential equations of the first order on the plane with nonlocal boundary
conditions, proving an existence and uniqueness theorem of a generalized solution,
and obtaining an a priori estimate.
Chapters 9 and 10 deal with differential equations, a field which has been
motivating for Lie theory since its birth. It is well known that Lies original aim was
to develop a theory for ordinary differential equations, equivalent to Galois theory
for algebraic equations, and it must be acknowledged that his approach unifies many
solving techniques and also provides new ones which apply to both ordinary and
partial differential equations. The book by Olver [7] is a recommended source for
all beginners.
Indeed, the composition of differential operators is a nonassociative product; thus
they often produce a Lie algebra of operators. In Chap. 1, Sect. 1.1.3, we give the
definition of the Weyl algebra of differential operators, showing its connections to
the Heisenberg algebra and to the irreducible representations of sl(2;k).
The Lie group symmetry is often also applied to Finsler geometry, where the
computations are usually more difficult than in Riemannian geometry. In Chap. 1,
Sect. 1.3, we give an account of Finsler geometry. By constructing tangent Lie
algebras to the holonomy group of a non-Riemannian Finsler manifold, in Chap. 12
the authors introduce the reader to a method for the investigation of its holonomy
properties, giving a unified treatment of their previous results. Some advances in
Introduction ix
the smooth solutions of Hilberts fourth problem in the regular case are presented
in Chap. 11, where the authors give a classification of some families of projectively
flat Finsler metrics.
As a natural environment for Lepage manifolds, Finsler geometry also appears
in Chap. 13, where the author illustrates how the geometric properties of Lepage
manifolds mirror geometrical, topological, and dynamical properties of differential
equations, connecting differential equations with geometry and physics.
Palermo, Italy Giovanni Falcone
References
1. R. Gilmore, Lie Groups, Physics, and Geometry: An Introduction for Physicists, Engineers and
Chemists (Cambridge University Press, Cambridge, 2008)
2. A.I. Shirshov, On free Lie rings. Mat. Sb. 45, 113122 (1958, in Russian)
3. A.I. Shirshov, On bases of a free Lie algebra. Algebra Logika 1, 1419 (1962, in Russian)
4. E. Chibrikov, On free Sabinin algebras. Commun. Algebra 39, 40144035 (2011)
5. L.V. Sabinin, P.O. Mikheev, Infinitesimal theory of local analytic loops. Sov. Math. Dokl. 36(3),
545548 (1988)
6. L.V. Sabinin, Smooth Quasigroups and Loops (Kluwer Academic Publishers, Dordrecht, 1999)
7. P.J. Olver, Applications of Lie Groups to Differential Equations. Graduate Texts in Mathematics,
vol. 107 (Springer, New York, 1986)
Contents
1 A Short Survey on Lie Theory and Finsler Geometry.. . . . . . . . . . . . . . . . . 1

Giovanni Falcone, Karl Strambach, and Ming Xu
2 Remarks on Infinite-Dimensional Representations
of the Heisenberg Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 23
Camillo Trapani
3 Character, Multiplicity, and Decomposition Problems
in the Representation Theory of Complex Lie Algebras . . . . . . . . . . . . . . . 41
Peter Fiebig
4 The BCH-Formula and Order Conditions for Splitting Methods . . . . . 71
Winfried Auzinger, Wolfgang Herfort, Othmar Koch,
and Mechthild Thalhammer
5 Cohomology Operations Defining Cohomology Algebra
of the Loop Space .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 85
Tornike Kadeishvili
6 Half-Automorphisms of CayleyDickson Loops . . . .. . . . . . . . . . . . . . . . . . . . 109
Maria de Lourdes Merlini-Giuliani, Peter Plaumann,
and Liudmila Sabinina
7 Invariant Control Systems on Lie Groups . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 127
Rory Biggs and Claudiu C. Remsing
8 An Optimal Control Problem for a Nonlocal Problem
on the Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 183
D. Devadze, Ts. Sarajishvili, and M. Abashidze
9 On the Geometry of the Domain of the Solution of Nonlinear
Cauchy Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 205
. Figula and M.Z. Menteshashvili
xi
xii Contents
10 Reduction of Some Semi-discrete Schemes

for an Evolutionary Equation to Two-Layer Schemes
and Estimates for the Approximate Solution Error . . . . . . . . . . . . . . . . . . . . 223
Jemal Rogava, David Gulua, and Romeo Galdava
11 Hilberts Fourth Problem and Projectively Flat Finsler Metrics. . . . . . 247
Xinyue Cheng, Xiaoyu Ma, Yuling Shen, and Shuhua Liu
12 Holonomy Theory of Finsler Manifolds . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 265
Zoltn Muzsnay and Pter T. Nagy
13 Lepage Manifolds .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 321
Olga Rossi
List of Contributors
M. Abashidze Batumi Shota Rustaveli State University, Batumi, Georgia

Winfried Auzinger Technische Universitt Wien, Vienna, Austria
Rory Biggs Department of Mathematics, Rhodes University, Grahamstown, South
Africa
Xinyue Cheng School of Mathematics and Statistics, Chongqing University of
Technology, Chongqing, Peoples Republic of China
D. Devadze Batumi Shota Rustaveli State University, Batumi, Georgia
Giovanni Falcone Dipartimento di Matematica e Informatica, Palermo, Italy
Peter Fiebig Department Mathematik, FAU Erlangen-Nrnberg, Erlangen,
Germany
. Figula Institute of Mathematics, University of Debrecen, Debrecen, Hungary
Romeo Galdava Sokhumi State University, Tbilisi, Georgia
David Gulua Georgian Technical University, Tbilisi, Georgia
Wolfgang Herfort Technische Universitt Wien, Vienna, Austria
Tornike Kadeishvili A. Razmadze Mathematical Institute of Tbilisi State Univer-
sity, Tbilsi, Georgia
Othmar Koch Fakultt fr Mathematik, Universitt Wien, Vienna, Austria
Shuhua Liu School of Mathematics and Statistics, Chongqing University of
Xiaoyu Ma School of Mathematics and Statistics, Chongqing University of Tech-
nology, Chongqing, Peoples Republic of China
xiii
xiv List of Contributors
M.Z. Menteshashvili Muskhelishvili Institute of Computational Mathematics of

the Georgian Technical University, Tbilisi, Georgia
Sokhumi State University, Tbilisi, Georgia
Maria de Lourdes Merlini-Giuliani Universidade Federal do ABC, Santo Andr,
Brazil
Zoltn Muzsnay Institute of Mathematics, University of Debrecen, Debrecen,
Hungary
Pter T. Nagy Institute of Applied Mathematics, buda University, Budapest,
Hungary
Peter Plaumann Universidad Autnoma Benito Jurez de Oaxaca, Oaxaca de
Jurez, Mexico
Claudiu C. Remsing Department of Mathematics, Rhodes University, Graham-
stown, South Africa
Jemal Rogava I. Vekua Institute of Applied Mathematics, Iv. Javakhishvili Tbilisi
State University, Tbilisi, Georgia
Olga Rossi Department of Mathematics, Faculty of Science, The University of
Ostrava, Ostrava, Czech Republic
Liudmila Sabinina Universidad Autnoma del Estado de Morelos, Cuernavaca,
Mexico
Ts. Sarajishvili Batumi Shota Rustaveli State University, Batumi, Georgia
Yuling Shen School of Mathematics and Statistics, Chongqing University of
Karl Strambach Department Mathematik, FAU Erlangen-Nrnberg, Erlangen,
Germany
Mechthild Thalhammer Institut fr Mathematik, Universitt Innsbruck, Inns-
bruck, Austria
Camillo Trapani Dipartimento di Matematica e Informatica, Palermo, Italy
Ming Xu College of Mathematics, Tianjin Normal University, Tianjin, Peoples
Republic of China
Chapter 1
A Short Survey on Lie Theory and Finsler
Geometry
Giovanni Falcone, Karl Strambach, and Ming Xu
The aim of this chapter is to provide readers with a common thread to the many
topics dealt with in this book, which is intermediate and which aims at postgraduate
students and young researchers, wishing to intrigue those who are not experts in
some of the topics. We also hope that this book will contribute in encouraging new
collaborations among disciplines which have stemmed from related problems. In
this context, we take the opportunity to recommend the book by Hawkins [18]. A
basic bibliography is outlined between the lines.
1.1 Generalities
A Lie algebra over a field k is a k-vector space g, endowed with a non-associative

binary operation ; W gg ! g, called Lie bracket, which is bilinear on each factor
(that is, axCby; z D ax; zCby; z, and z; axCby D az; xCbz; y), alternating
(that is, x; x D 0) and fulfilling the Jacobi identity: x; y; z C z; x; y C
y; z; x D 0. If the characteristic of k is not 2, then x; x D 0 x; y D
y; x. The cases where k D R; C; Qp are of particular interest. A very classic
introductory book is the one by Jacobson [21].
G. Falcone ()
Dipartimento di Matematica e Informatica, via Archirafi 34, 90123 Palermo, Italy
e-mail: giovanni.falcone@unipa.it
K. Strambach
Department Mathematik, FAU Erlangen-Nrnberg, Cauerstr. 11, 91058 Erlangen, Germany
e-mail: strambach@mi.uni-erlangen.de
M. Xu
College of Mathematics, Tianjin Normal University, 300387 Tianjin, Peoples Republic of China
e-mail: mgmgmgxu@163.com
Springer International Publishing AG 2017 1

G. Falcone (ed.), Lie Groups, Differential Equations, and Geometry,
UNIPA Springer Series, DOI 10.1007/978-3-319-62181-4_1
2 G. Falcone et al.
Any associative algebra over a field k, with the Lie bracket defined by
x; y D xy yx;
is a Lie algebra. In particular, the algebra of endomorphisms of a k-vector space

V, isomorphic to the algebra of n n matrices with entries in k, is a prototypical
example, denoted by gl.V/, or gl.n; k/. Another basic example is its Lie subalgebra
of traceless endomorphisms, denoted by sl.V/, or sl.n; k/ (notice that it is closed
under Lie bracket, but not closed under addition, nor multiplication). Finally, we
want to mention the Heisenberg algebra h3 of strictly upper triangular 33 matrices,
which is generated by three elements
0 1 0 1 0 1
010 000 001
x D @0 0 0A; y D @0 0 1A; z D @0 0 0A (1.1)
000 000 000
fulfilling the Lie brackets
x; y D z;
and which can be generalized, more abstractly, to the Lie algebra h2nC1 , generated
by 2n C 1 elements fx1 ; : : : ; xn ; y1 ; : : : ; yn ; zg and defined by the Lie brackets
xh ; yk D hk z; (1.2)
where hk is the Kronecker symbol. We notice, in fact, that, in order to compute the
Lie brackets between vectors of a Lie algebra, it is sufficient to know the nonzero
Lie brackets between the vectors of a basis.
Similarly to the case of associative algebras, one defines and distinguishes the
concept of a Lie subalgebra and the one of a Lie ideal: a Lie subalgebra I is an ideal
of g, if
x; y 2 I; for all x 2 I and y 2 g:
For instance, sl.n; k/ is an ideal of gl.n; k/.

Two basic ideals are the commutator ideal g; g, generated by all elements x; y,
and the center z, consisting of all elements x such that x; g D 0. One says that g is
Abelian, if g; g D 0 (or, equivalently, if z D g).
As for associative rings, ideals are precisely the kernels of homomorphisms.
Given a Lie algebra h and an ideal I in it, one constructs the factor Lie algebra
h=I: its elements are the cosets x C I.
This is probably the right place to mention the following theorem, which was
proved in characteristic zero in 1935 by Ado and in arbitrary characteristic in
1948/1949 by Iwasawa and Harish-Chandra:
1 A Short Survey on Lie Theory and Finsler Geometry 3
Every finite-dimensional Lie algebra h over a field k is isomorphic to a Lie

subalgebra of gl.n; k/ for some integer n, that is, it has a faithful representation,
where a representation of a Lie algebra h on kn is a Lie algebra homomorphism
W h ! gl.n; k/;
and a representation is said to be faithful if its kernel is trivial, i.e., it is injective.

In a nutshell, the structure of finite-dimensional Lie algebras can be summarized
as follows. The commutator ideal g; g brings on two series of ideals: the derived
series
g.k/ D g.k1/ ; g.k1/ ;
and the descending central series
gk D g; gk1 ;
where we put g.0/ D g0 D g. A Lie algebra is called solvable if the derived series
eventually ends to zero, and nilpotent if the descending central series eventually
ends to zero. Inductively, one sees that g.k/ 6 gk , hence nilpotent Lie algebras are
solvable. Moreover, it turns out that g is solvable if and only if g; g is nilpotent. In
1876, Sophus Lie proved that a finite-dimensional complex solvable Lie algebra
has a common eigenvector, and, by a recursive argument, this yields a matrix
representation in upper triangular form. Every Lie algebra has a unique maximal
solvable ideal, called the radical.
If the only ideals of a non-Abelian Lie algebra g are zero and g itself, then g
is called simple. The classification of simple complex Lie algebras was essentially
carried out by Wilhelm Killing in the late 1880s and sensibly improved by lie
Cartan who extended it to the real case in 1914. If the radical is zero, the Lie algebra
is called semisimple. Every semisimple Lie algebra turns to decompose as the direct
sum of simple Lie algebras. Finally, in 1905 Eugenio Elia Levi proved that every
real Lie algebra decomposes as the semidirect sum of its radical with a semisimple
subalgebra, called thereafter a Levi subalgebra of g (which, in general, is not an
ideal of g).
1.1.1 Representations, Derivations, Killing Form
The reader can easily imagine that the same Lie algebra can appear in very different
ways. A representative case is the one of the three-dimensional, simple, Lie algebra
sl.2; k/ of traceless 2 2 matrices, which is generated by the matrices

1 0 01 00
X0 D I X1 D I X2 D
0 1 00 10
4 G. Falcone et al.
fulfilling
X0 ; X1 D 2X1 ; X0 ; X2 D 2X2 ; X1 ; X2 D X0 ;
and which has, for every integer n > 2, a unique (up to conjugation) faithful
representation
n W sl.2; k/ ! sl.n; k/ 6 gl.n; k/ (1.3)
defined, for 0 6 k 6 n, by the following action on the vectors fe0 ; : : : ; en1 g of a

basis:
n .X0 /ek D .n 2k 1/ek I n .X1 /ek D kek1 I n .X2 /ek D .n k 1/ekC1 :
Notice that each n is irreducible, that is, it has no nontrivial invariant subspace.
We mention here a fundamental theorem of Hermann Weyl in the theory
of representations of Lie algebras: any finite-dimensional representation of a
semisimple Lie algebra g is completely reducible, that is, every g-invariant subspace
has a g-invariant complementary subspace. On the contrary, the upper triangular
representation of a solvable Lie algebra g determines a flag of g-invariant subspaces
having no g-invariant complementary subspace.
Physical observables can be represented by operators, acting on the Hilbert space
of the states of the system, whose commutating relations are defined through Lie
brackets. Two examples: first, if Q represents the position of a particle and P its
momentum, then the Heisenberg principle states that P; Q D i1, where 1 is
the identity operator; second, if the differential operators Lx , Ly , Lz represent the
components of the angular momentum, then they satisfy the following commutating
relations:
Lx ; Ly D iLz ; Ly ; Lz D iLx ; Lz ; Lx D iLy : (1.4)
The three-dimensional real Lie algebra generated by Lx , Ly , Lz is isomorphic

to the Lie algebra denoted by su.2/, and is a subalgebra of the complex Lie
algebra sl.2; C/ described in (1.3), but in Chap. 2 the author looks at these
commutation relations within an infinite-dimensional representation in a Hilbert
space of unbounded operators.
A fundamental, direct representation is obtained when one considers, for a given
x 2 h, the linear map adx W h ! h defined by
adx .y/ D x; y;
which is called the adjoint map of x. Notice that the Jacobi identity itself can be
re-written in terms of the adjoint map as
adx .y; z/ D adx .y/; z C y; adx .z/:

More generally, a map W h ! h such that y; z D y; z C y; z (and,

recursively, such that Leibniz rule
!
X
n
n k nk
y; z D
n
y; z
0
k
is fulfilled) is called a derivation, and the derivations of the form adx are called inner
derivations.
A corroboration of the self-consistency of Lie theory comes from the fact that,
whereas the product of two derivations 1 and 2 is not necessarily a derivation, it is
immediate to see that the Lie bracket
1 ; 2
is still a derivation, that is, Der.g/ is a Lie subalgebra of the Lie algebra gl.g/ of
endomorphisms of the vector space h. Moreover, if is a derivation of g, then the
vector space generated by g and becomes a Lie algebra putting ; y D .y/.
The map ad W g ! Der.g/; x 7! adx , is called the adjoint representation of
g. Notice that the kernel of ad is precisely the center z of g, therefore, if z D 0,
the adjoint representation is faithful and represents concretely the Lie algebra g as
a subalgebra of gl.g/. This can be seen, but only for Lie algebras with trivial center,
as an alternative proof of Ados theorem.
The adjoint representation provides a fundamental tool in Representation theory:
the Killing form
.x; y/ D Tr.adx ady /;
which has the following properties: it is a symmetric bilinear form, which is zero if
and only if g is nilpotent, and which is non-degenerate if and only if g is semisimple
(Cartan criterion). In the latter case, if k D R, then the Killing form will discriminate
the case where the semisimple Lie algebra g is compact (see p. 9). The -orthogonal
of any ideal I of g is also an ideal, and two ideals with trivial intersection are -
orthogonal.
For a minimal bibliographical support, we address the reader to one of the most
popular introductory books in Representation theory, that is, the one by Humphreys
[20].
1.1.2 The Weyl Algebra of Differential Operators
As mentioned in the opening, the composition of differential operators is a non-

associative product, thus they often produce a Lie algebra of operators. A good
example to be mentioned here is the Weyl algebra of differential operators with
6 G. Falcone et al.
polynomial coefficients in n variables. Its generators xk (acting as f .x/ 7! xk f .x/)

and @=@xk satisfy the Lie bracket
@=@xh ; xk D hk ;
because
@ @ @ @f
xk xk f .x/ D xk f .x/ xk .x/ D
@xh @xh @xh @xh
@f @f
hk f .x/ C xk .x/ xk .x/ D hk f .x/:
@xh @xh
It was introduced by Hermann Weyl in his mathematical formulation of quantum

mechanics, and it is very closed related to the (generalized) Heisenberg algebra
h2nC1 seen in (1.2). More precisely, the Weyl algebra is a quotient of the universal
enveloping algebra of h2nC1 (that is, the free algebra on h2nC1 modulo the ideal
generated by the relations xy yx x; y), obtained by factoring the ideal generated
by z 1.
Notice, in passing, that also each irreducible representation n of sl.2; k/ in (1.3)
can be seen as the action of the differential operators
@ @ @ @
0 D x2 x1 ; 1 D x2 ; 2 D x1
@x2 @x1 @x1 @x2
on the algebra of homogeneous polynomial of degree n 1 in two variables x1 and

x2 , because they act on the vectors of the basis of the form ek D xk1 xnk1
2 mapping
0 .ek / D .n 2k 1/ek ; 1 .ek / D kek1 ; 2 .ek / D .n k 1/ekC1 :
1.1.3 Lie Groups, Exponential Map
A Lie group is a group defined over a finite-dimensional smooth manifold, such

that the group multiplication and the inversion are smooth maps. The general linear
group GL.n; R/ of n n invertible matrices over R is a n2 -dimensional smooth
manifold where the group operation is differentiable, hence it is a Lie group. This
holds also for its topologically closed subgroups (Cartans theorem), also called
linear Lie groups.
Similarly to what one does by taking the linear part of the Maclaurin series of a
function, one can associate to each connected Lie group G, a Lie algebra g whose
underlying vector space is the tangent space to the group in the neutral element
[27]. This Lie algebra g completely captures the local structure of the group G.
There can be several Lie groups with the same Lie algebra, so this is not one-to-one
correspondence, and one says that two Lie groups are locally isomorphic, if their
Lie algebras are isomorphic. All of them, however, are a quotient of the largest one,
the so-called (simply connected) universal covering group associated with g. In this
correspondence, subalgebras (resp. ideals) correspond to closed subgroups (resp.
closed normal subgroups).
Two smallest, prototypical examples: first, consider the group SO2 .R/ of rota-
tions
! ! ! ! !
cos sin 10 0 12 2 0 0 16 3
XD D C C C 1 3 C ;
sin cos 01 0 0 12 2 6
0
with 2 R, which, by the way, are written in complex form as
1 1
1 C i 2 3 C D exp.i/:
2 6
In a neighborhood of the identity, that is, considering the first order approximation,
we see that

1 1 1 2 1 1 2 1 C 2
D ;
1 1 2 1 1 2 1 1 2
and, disregarding infinitesimal elements of the second order, this says that the
multiplication of infinitesimal rotations behaves like the addition of arcs in the
tangent space to the circle at the identity, which is

@ 0
IC Xj D0 D I C 2 R:
@ 0
This explains also why Sophus Lie called infinitesimal groups what we now call Lie
algebras. Moreover, Lie theory gives a deeper meaning to the concept of exponential
map: the usual one, x 7! exp.x/, transforms the additive group of the reals onto the
multiplicative group of the positive reals. Similarly, the above example of SO2 .R/
shows that the exponential map x 7! exp.ix/ transforms the additive group of the
reals onto the group of rotations. Hence, the Lie groups R (non-connected), RC
(simply connected), and SO2 .R/ (not simply connected) are locally isomorphic,
with SO2 .R/ being in fact a homomorphic image of the simply connected group
RC modulo the lattice of periods generated by 2, and the connected component of
R being isomorphic to RC via the exponential map.
The second example is the Heisenberg group
8 0 1 9
< 1 x1 x3 =
H3 D X D @ 0 1 x 2 A W x 1 ; x 2 ; x 3 2 R :
: ;
0 0 1
8 G. Falcone et al.
Since for k D 1; 2; 3 the three derivatives @x@k XjXDI are, respectively, the three
elements x, y, and z given in (1.1), the tangent space at the identity element I is h3 .
On the other hand, as the cube of any element x 2 h3 is zero, we see that exp x D
I C x C 12 x2 , that is,
0 1 0 1
0ab 1 a b C 12 ac
1
x D @0 0 cA
7 ! exp x D I C x C x2 D @ 0 1 c A;
2
000 00 1
and

1
exp x exp y D exp x C y C x; y : (1.5)
2
Just in the same way as the exponential map x 7! exp.ix/ transforms the additive
group of the reals onto the group of rotations, the above Eq. (1.5) transforms the
additive group of Lie algebra h3 onto the non-commutative Lie group H3 .
Finally, for a given Lie algebra g and for its Lie algebra of derivations Der g,
one can consider the Lie group exp.Der g/ and find out that this is a group of
automorphisms of g, such that
exp..x// D exp./.x/;
confirming, once again, the self-consistency of Lie theory.

The question which arises after these examples is the following: does the
exponential map always transform the addition of a suitable, but linear, Lie algebra
structure on the tangent space to a Lie group G into the multiplicative structure of the
group G itself? Surprisingly, within a series of circumstances which give an account
of the richness of the theory, the answer is mostly yes. In order to be applied to the
most general case where the group is not commutative, Eq. (1.5) must be completed
to the BakerCampbellHausdorff formula, whose first terms are
1 1
exp.x/exp.y/ D exp.xCyC x; yC x; x; y C y; y; x C: : :/: (1.6)
2 12
Three-dimensional Lie groups, .2n C 1/-dimensional Heisenberg groups, and

six-dimensional orthogonal groups are extensively investigated in Chap. 7 as the
manifolds on which a left-invariant affine control system is defined. A smooth
control system may be viewed as a family of dynamical systems (or vector fields)
u on a manifold, (smoothly) parametrized by control vectors u 2 Rl , and fulfilling
@
g D u .g/:
@t
An integral curve of such a vector field is called a trajectory of the system, and the
main question is whether any two points can be connected by a trajectory, possibly
minimizing some cost function.
A significant subclass of control systems, first considered in 1972 by R. W.
Brockett and by V. Jurdjevic and H.J. Sussmann, is that of left-invariant, affine
control systems, where the manifolds are real, finite-dimensional, connected, linear
Lie groups G, and the dynamics are invariant under left translations, i.e.,
u .g/ D gu .1/, and affinely parametrized by control vectors u 2 Rl , i.e.,
u .1/ D .ACu1 B1 C Cul Bl /, where A; B1 ; : : : ; Bl are elements of the Lie algebra
g of G, with B1 ; : : : ; Bl linearly independent. Such a system fulfills therefore
@
g D g.A C u1 B1 C C ul Bl /;
@t
and for such a system the left translation of any trajectory is a trajectory. Addition-
ally, the subclass of drift-free systems, i.e., systems with A D 0, is reinterpreted in
Chap. 7 as invariant sub-Riemannian structures, and some examples are revisited.
1.1.4 Compact Lie Algebras
If the Lie group G is compact, one says that its Lie algebra g is compact. But a
useful intrinsic criterion for the compactness of g comes from the Killing form :
the Lie algebra g is compact if and only if is definite negative , as we illustrate with
the following example, which is the three-dimensional compact real group SU.2/ of
special unitary matrices, represented by the compact, simply connected group of
quaternions having the norm equal to one:
0 1
x1 x2 x3 x4
B x2 x1 x4 x3 C
XDB
@ x3
C; x21 C x22 C x23 C x24 D 1:
x4 x1 x2 A
x4 x3 x2 x1
As such, it is the real form of the group of complex matrices

z w
; z D x1 C ix2 ; w D x3 C ix4 ; jzj2 C jwj2 D 1;
w z
but, the conjugation being not complex analytic, this is not a complex Lie group
(also, its real dimension is odd!).
10 G. Falcone et al.
Putting alternatively x3 D x4 D 0, x2 D x4 D 0, and x2 D x3 D 0, we obtain

three one-dimensional subgroups of matrices of the form
0 1 0 1
cos sin 0 0 cos 0 sin 0
B sin cos 0 0 C B 0 cos 0 sin C
@ 0 0 cos sin
A; @ sin 0 cos 0 A ;
0 0 sin cos 0 sin 0 cos
0 1
cos 0 0 sin
B 0 cos sin 0 C
@ 0 sin cos 0 A;
sin 0 0 cos
whence one immediately sees that the tangent space at the identity element I is
generated by
0 1 0 1 0 1
0 1 0 0 0 0 10 0 0 0 1
B 1 0 0 0 C B 0 0 0 1C B 0 0 1 0C
B C; B C; B C:
@ 0 0 0 1 A @ 1 0 0 0A @ 0 1 0 0A
0 0 1 0 0 1 00 1 0 0 0
These matrices generate the real Lie algebra su.2/ given at p. 4, and are the real
forms of the matrices

i 0 0 1 0 i
; ; ;
0 i 1 0 i 0
which in turn generate the complex Lie algebra sl2 .C/ of traceless matrices. Notice
that the same complex Lie algebra is obtained as the complexification sl2 .C/ D
sl2 .R/R C of the real Lie algebra sl2 .R/. This shows that the Lie algebras g of non-
isomorphic real Lie groups may have isomorphic complexifications gC D g R C.
More generally, starting from a real Lie algebra g, one can consider its complex-
ification gC D g R C, thus non-isomorphic real Lie groups may have isomorphic
complexifications. Notice that the Killing form
.x; y/ D Tr.adx ady /
of a semisimple real Lie algebra g determines, by Sylvesters inertia, an index, which

is the number of positive entries of the diagonal form of . It turns out that the index
is zero, that is, is definite negative, if and only if g is the Lie algebra of a compact
real Lie group, and the real Lie algebra g is said a compact real form of gC . In
the opposite case, where is definite positive, the real Lie algebra g is said a split
real form of gC . The intermediate cases correspond to the Cartan decomposition
g D k p in two -orthogonal subspaces, the former being a maximal compact
subalgebra, the latter being only a complementary subspace. For instance, in the
(not infrequent) case that g has a representation as a real Lie algebra of matrices that
is closed under transposition, the Cartan decomposition is just the decomposition

into the subalgebra of skew-symmetric part and the complementary subspace of
symmetric matrices, thus gC D .kip/.ikp/. Now, kip is a subalgebra which
consists of Hermitian matrices, is a real Lie subalgebra of gC , whose Killing form
is negative definite, hence it is compact. Finally, its complexification is clearly gC .
1.2 Non-associative Structures
1.2.1 Loops and Geometry
As one already could see, the non-associative structures are recharged from many
sources. One of them which today is almost forgotten is the theory of projective
planes. If an affine plane E is non-desarguesian, then E can be coordinatized
only by ternary fields, and the addition and multiplication of them are loops.
In particular, translation planes are coordinatized by quasifields, whose addition
defines an Abelian group but whose multiplication defines a loop such that these two
operations are connected by a distributive law. The whole theory of non-associative
algebras grew out of this basement. Today a loop is defined as a set L with a binary
operation .x; y/ 7! x y, such that there exists an element e 2 L with e x D x e D x
for all x 2 L, and the equations a y D b and x a D b have precisely one
solution. The calculations with loops based on this definition are tedious. For this
reason one tried to formulate the theory of loops completely within the theory of
groups. The possibility for this is delivered, for a loop L, by the left translations
a W y 7! a y W L ! L, which are bijections of L for any a 2 L. Following Cayley,
one can see the loop L as a set of permutations of L acting sharply transitively on L.
If G is the group generated by the left translations of L, and H is the stabilizer of the
identity 1 of L, then the left translations a W y 7! a y form a section W G=H ! G.
If W G ! G=H is the natural mapping x 7! xH, then W G=H ! G=H is the
identity. The image .G=H/ forms a system of representatives for the left cosets of
H in G. Conversely, let G be a group and H a subgroup of G containing no nontrivial
normal subgroup of G. If W G=H ! G is a section with .H=H/ D 1 2 G such
that .G=H/ generates G and operates sharply transitively on G=H (which means
that to any xH and yH there exists precisely one z 2 .G=H/ with zxH D yH),
then the multiplication on the factor space G=H given by xH yH D .xH/yH,
respectively the multiplication on the set .G=H/ given by x y D .xyH/, yields
a loop having G as the group generated by the left translations xH 7! .xH/,
respectively x 7! .xH/. The advantage of this procedure lies in the enormous
power of group theory. Choosing in particular for G a finite group, a Lie group, an
algebraic group, or a p-adic Lie group such that H is a closed subgroup and is a
map in the corresponding category, then many remarkable classification results can
be obtained, since the simple groups in these categories are known.
Special classes of loops may be defined by algebraic laws which can be seen as
weakenings of the associativity. For loops which are near to groups the algebraic
expressions for weak associativity and the validity of the configurations in the
corresponding geometries show the same complexity. As an example for this, we
can quote the classes of Moufang loops and Bol loops.
But, in general, nicely algebraically defined loops are rare from a geometrical
point of view. For instance, the multiplicative loops of quasifields, in which only
one distributive law holds, and which coordinatize translation planes, cannot be
described in a simple algebraic way.
The real analytic loops possess tangential objects, Malcev algebras, Bol algebras,
Sabinin algebras, which describe them near to the identity. Using Malcev algebras
it was possible to create, for Moufang loops, a theory having the same level as the
theory of Lie groups and to get a complete classification of analytic Moufang loops.
Any connected analytic Moufang loop is an almost direct product of a Lie group
and of the Moufang loops arising from the multiplicative loop of octonions. The
real octonion algebra satisfies Moufang identities and is the only real composition
non-associative division algebra arising by the CayleyDickson process, which, by
the way, is introduced in Chap. 6 to define a sequence of algebras, among which one
finds in particular the octonion and the sedenion loops as the multiplicative loop.
Extensive majority of non-commutative Lie groups can be seen as groups of
left translations of topological loops. For instance, among the three-dimensional
Lie groups, only the group of homotheties of the Euclidean plane R2 cannot be
the group topologically generated by the left translations of a three-dimensional
topological loop. The situation changes drastically if one considers Lie groups
G which are generated by left translations together by right translations of loops
homeomorphic to a Lie group. Already any loop L defined on a connected one-
dimensional manifold such that its group generated by the left and right translations
is locally compact must be a group isomorphic either to R or to SO2 .R/. In the case
of the two-dimensional Lie loop corresponding to the hyperbolic plane and having
the three-dimensional group PSL2 .R/ as its group generated by the left translations,
it is the group generated by left and right translations of an infinite-dimensional Lie
group. For loops corresponding to higher dimensional hyperbolic spaces one meets
the same situation. This indicates that the simplicity of a Lie group is an obstruction
for being a group generated by the left and right translations of a topological loop.
The compactness of a Lie group G is a serious obstruction even for the case that
G is a group generated by the left translations of a topological loop homeomorphic
to a Lie group. For instance, any 1-dimensional connected topological loop having
a compact Lie group as the group topologically generated by its left translations is
the orthogonal group SO2 .R/. Another example showing the restricting power of
compact Lie groups is given by topological loops L homeomorphic to the 7-sphere
or to the 7-dimensional real projective space. Any such loop L having a compact
connected Lie group G as the group topologically generated by the left translations
of L is one of the two 7-dimensional compact Moufang loops, G is locally
isomorphic to PSO8 .R/ and the stabilizer H of the identity e 2 L is isomorphic
to SO7 .R/. Moreover, if the compact Lie group G is topologically generated by
the left translations of a proper topological loop, then G is homeomorphic to a

connected semisimple Lie group. In a sense, conversely there does not exist any
proper topological loop L which is homeomorphic to a connected quasi-simple
Lie group and has a compact Lie group as the group topologically generated by
its left translations. More generally, if L is homeomorphic to a product of quasi-
simple simply connected compact Lie groups and has a compact Lie group G as
the group topologically generated by its left translations, then G is at least 14-
dimensional. In the case where dim G D 14, the group G is locally isomorphic to
Spin3 .R/ SU3 .C/ Spin3 .R/ and L is homeomorphic to a group which is locally
isomorphic to Spin3 .R/ SU3 .C/:
1.2.2 Gerstenhaber Algebras
As mentioned above, among non-associative algebras, that is, k-vector spaces with
a non-associative, k-bilinear product, we also find Jordan algebras and Gerstenhaber
algebras.
In particular, here we limit ourselves to give the definition of a Gerstenhaber
algebra, because they are a main object of Chap. 5, and to address the reader to the
excellent introduction on this topic given in [4].
A k-vector space A with a k-bilinear product W A A ! A and associator
.x; y; z/ D .xy/zx.yz/ is a Gerstenhaber algebra (or right-symmetric algebra), if
.x; y; z/ D .x; z; y/:
Whilst the exterior algebra of a Lie algebra is a Gerstenhaber algebra, the commu-
tator
x; y D x y y x
endows a given Gerstenhaber algebra with the structure of a Lie algebra, because
the Jacobi identity can be formally written as
x; y; z C z; x; y C y; z; x D
.x; y; z/ C .y; z; x/ C .z; x; y/ .y; x; z/ .x; z; y/ .z; y; x/:
The latter identity, holding in any two-sided distributive, not necessarily associative,
algebra, shows that Gerstenhaber algebras and left-symmetric algebras deserve the
alternate name of pre-Lie algebras.
Gerstenhaber algebras, and their opposite left-symmetric algebras, appeared,
not surprisingly, in 1896 in a paper by Arthur Cayley. They were revived in the
early 1960s by Vinberg [29], Koszul [24], and Gerstenhaber [17]. More recently,
homotopy Gerstenhaber algebras found a place in the investigations of Kontsevich

[22], and of Connes and Kreimer [6], among many others.
As pointed out in [23], it was Koszul [25] who realized that an odd Laplacian,
S operator of degree 1 on an associative
that is, a second-order differential
commutative graded algebra A D k Ak such that 2 D 0 and .1/ D 0, provides
A with the structure of a Gerstenhaber algebra, by putting:

x; y D .1/k .a b/ .a/ b .1/k a .b/ ;
where x 2 Ak , y 2 A. A slight generalization of the operator defines Batalin

Vilkovisky algebras, a useful tool in the description of certain Lagrangian gauge
theories [30].
1.3 Finsler Geometry
1.3.1 Finsler Metric and Minkowski Norm
Finsler geometry is more general than Riemannian geometry in the sense that it
does not require the metric to be quadratic for each tangent space. Its definition
was proposed by Riemann in his talk in 1854, and the systematical study on Finsler
geometry was started by the thesis of Finsler in 1918 [16]. This research field has
been quiet for many decades, until its arising in the 1990s, by the full advocation of
S.S. Chern. He, together with D. Bao, Z. Shen, and many other geometers, rebuilt
the foundations for modern Finsler geometry [2]. Nowadays, Finsler geometry has
many applications in biological statistics, electrical science, and general relativity. It
also provides viewpoints for us to understand this geometric world beyond intuition
and imagination.
We start with the concept of Finsler metric, which defines the length of each
tangent vector. A Finsler metric on a smooth manifold M is a continuous function
F W TM ! 0; C1/, satisfying the following conditions:
1. Regularity: F is a positive smooth function on the slit tangent bundle TMn0.
2. Positive homogeneity: F.y/ D F.y/ for any tangent vector y and > 0.
3. Strong convexity: For any standard local coordinates on TM, i.e., x D .xi / 2 M
and y D yi @xi 2 Tx M, the Hessian matrix

1 2
.gij .x; y// D F .x; y/yi y j
2
is positive definite.
We call .M; F/ a Finsler manifold or Finsler space.
In each tangent space, the Finsler metric defines a Minkowski norm. Minkowski
norm can be more abstractly defined on any real vector space, satisfying similar
conditions as above (1)(3).
Well-known examples of Finsler metrics include the following. A Finsler metric
is Riemannian when the Hessian matrix is irrelevant to the y-coordinates for
any standard local coordinates. In this case, it defines a smooth global section
gij .x/dxi dx j of Sym2 .T M/ which is usually referred to as the Riemannian metric.
The most simple and important class of non-Riemannian metrics are Randers
metrics, which are of the form F D C where is a Riemannian metric, and
is a 1-form. The .; /-metrics are generalizations of Randers metrics, which can
be presented as F D
.=/, where and are similar as for Randers metrics,
and
is a smooth function. Notice for a Randers metric (or an .; /-metric) F,
and (or , , and
resp.) must satisfy certain conditions to satisfy the convexity
condition (3) [5, 26].
1.3.2 Geodesic and Curvature
The arc lengths of piecewisely smooth curves can be similarly defined in Finsler
geometry. Their infimum for all curves from one fixed point to another provides a
(possibly irreversible) distance dF .; /. Using variational method, geodesics can be
defined which satisfies the locally minimizing principle for dF .; /.
There is another way to describe the geodesics on a Finsler space. The geodesic
spray G is a global smooth tangent vector field on TMn0, which can be presented
as G D yi @xi 2Gi @yi , with Gi D 14 gil .F 2 xk yl yk F 2 xl /, for any standard local
coordinates on TM. Then a smooth curve c.t/ with a nonzero constant speed (i.e.,
F.Pc.t// const > 0) is a geodesic iff its lifting .c.t/; cP .t// in TM is an integration
curve of G.
The geodesic spray G and its coefficients Gi are very useful for presenting other
geometric quantities (curvatures) in Finsler geometry. Curvature is the core concept
of Finsler geometry. Some curvatures are generalized from Riemannian geometry,
which are called Riemannian curvatures. They can help us to detect how a Finsler
space curves. Some others vanish in Riemannian geometry, i.e., they only appear for
non-Riemannian metrics and can be used to measure how these metrics differ with
the Riemannian ones. We call them non-Riemannian curvatures [28].
Let us see some curvatures in Finsler geometry. First we look at the Riemannian
ones.
In the Jacobi field equation for a smooth family of constant speed geodesics, we
can find a linear operator RFy W Tx M ! Tx M for any nonzero y 2 Tx M. We call it the
Riemann curvature. For any standard local coordinates on TM, it can be presented
as RFy D Rik .y/@xi dxk , where
Rik .y/ D 2@xk Gi y j @2x j yk Gi C 2G j @2y j yk Gi @y j Gi @yk G j :

Using Riemann curvature, we can easily define flag curvature, Ricci curvature,
scalar curvature, etc. For example, flag curvature is a natural generalization of
sectional curvature in Riemannian geometry. For any x 2 M, nonzero y 2 Tx M
(the pole), and tangent plane P containing y (the flag), linearly spanned by y and v,
the flag curvature
hRy v; viFy
K F .x; y; y ^ v/ D K F .x; y; P/ D ;
hy; yiFy hv; viFy .hy; viFy /2
where h; iFy is the inner product on Tx M defined by the Hessian matrix .gij .x; y//.
The flag curvature does not depend on the choice of v but only on y and P.
Next we look at the non-Riemannian curvatures.
The first degree derivatives of the Hessian matrix .gij .x; y//, i.e., the third
degree derivatives of F 2 .x; y/, define the Cartan tensor, the most fundamental non-
Riemannian curvature in Finsler geometry. In a global fashion, it can be presented as
1 @3
CyF .u; v; w/ D F 2 .x; y C ru C sv C tw/jrDsDtD0 ;
4 @r@s@t
where y; u; v; w 2 Tx M and y 0.
Notice you may further differentiate the Cartan tensor with respect to the
y-coordinates as many times as you like. Then you get infinitely many non-
Riemannian curvatures from this thought. They provide obstacles for a (non-
compact) Finsler space to be isometrically imbedded in Minkowski spaces. But they
are not very useful most of the time.
S-curvature invented by Z. Shen is another important non-Riemannian curvature,
with a totally different style. It can be described as the derivative of the distortion
function
p
det.gij .x; y//
.x; y/ D ln
.x/
in the direction of the geodesic spray G. Here .x/ is the coefficient function in
some canonical volume form. An important case is to take the Busemann-Hausdorff
volume form dVBH D .x/dx1 dxn , where
!n
.x/ D :
Volf.yi / 2 Rn jF.x; yi @xi / < 1g
Though the definition of .x; y/ involves standard local coordinates, it is in fact a

globally defined smooth function on TMn0. It is crucial that S-curvature has many
important properties which relate it to other curvatures in Finsler geometry.
Why do we need the Lie symmetry for studying Finsler geometry

1.3.3 Local Homogeneity and Global Homogeneity
There exists a main obstacle for the study of general Finsler geometry. The
calculations and formulas are much more complicated than those in Riemannian
geometry. Thus we need to call Lie theory for help. The Lie group symmetries could
be applied to Finsler geometry in both the local and the global senses.
First look at the Lie symmetries in the local sense. Sometimes we restrict
our interest to some types of Finsler metrics with special importance. For exam-
ple, Riemannian metrics are the most well-known Finsler metrics. In the last
century, Riemannian geometry has become one of the most important mathe-
matical branches. Generally speaking, we concern non-Riemannian metrics and
spaces in Finsler geometry. Randers and .; /-metrics are the most simple and
important non-Riemannian Finsler spaces. The majority of research works on
Finsler geometry are related to Randers metrics and .; /-metrics. The reason
behind this phenomenon is a good local homogeneity, which means for each
tangent space Tx M, the connected linear isometry group L0 .Tx M; F/ (preserving the
Minkowski norm F.x; / on Tx M) is relatively big. For example, an n-dimensional
Riemannian manifold .M; F/, L0 .Tx M; F/ is isomorphic to SO.n/, which is the
maximum among all possible ones. For a non-Riemannian Randers space or an
.; /-space, L0 .Tx M; F/ is isomorphic to SO.n 1/, which is a maximum among
all non-Riemannian ones. These local homogeneities can significantly reduce the
complexity of calculations in Finsler geometry, i.e., make the research subject much
easier.
Here are two remarks on the local homogeneity.
First, there is another class of Finsler metrics, generalizing Randers metrics and
.; /-metrics [15]. Let be a Riemannian metric on M, with respect to which TM
can be orthogonally decomposed as the sum of two sub-bundles TM D V1 V2 .
The restrictions i D jVi can be naturally regarded as functions on TM. Then a
Finsler metric F of the form F D f .1 ; 2 / for some positive smooth function f
(which must be positively homogeneous of degree 1) is call an .1 ; 2 /-metric.
For an .1 ; 2 /-space .M; F/, its connected linear isometry group L0 .Tx M; F/ is
isomorphic to SO.n1 / SO.n2 /, where ni is the real dimension of the fibers of Vi . It
reaches another maximum among all non-Riemannian metrics [15]. We guess, the
most that works on .; /-metrics can be generalized to .1 ; 2 /-metrics.
Second, we may improve the efficiency of the local homogeneities by consider-
ing the whole linear isometry group instead of its identity component. For example,
reversible metrics have slightly better local homogeneities than irreversible metrics,
because of the extra Z2 -isometries for the antipodal map.
Next we look at the Lie symmetries in the global sense. That means a Lie
group G acts nontrivially and isometrically on the Finsler space .M; F/. We can
assume G is a closed connected subgroup of the isometry group I.M; F/. Notice the
isometry group I.M; F/ itself is a Lie transformation group [9]. One of the most
typical cases is that G acts transitively on the Finsler space .M; F/. Then we call
.M; F/ a homogeneous Finsler space [8]. Mostly we assume M is connected. Denote
M D G=H a homogeneous Finsler space where G is a closed connected subgroup

of I0 .M; F/ and H is the compact isotropy subgroup at o D eH. Then we have an
Ad.H/-invariant decomposition g D hCm, where g D Lie.G/ and h D Lie.H/. The
subspace m can be canonically identified with To .G=H/. A G-homogeneous Finsler
metric F on G=H is one-to-one determined by an Ad.H/-invariant Minkowski norm
on m. The study on homogeneous Finsler spaces is called homogeneous Finsler
geometry.
In homogeneous Finsler geometry, curvatures can be reduced to some tensors
on m, and differential equations are reduced to linear equations. So we may avoid
some extremely complicated calculations, and find the intrinsic nature of curvatures
from the Lie algebra structures. The most successful examples for this approach
include the following. We can use Killing frames (i.e., local frames provided by
Killing vector fields) to present the S-curvature, and use the Finslerian submersion
technique to deduce the flag curvature formula, for a homogeneous Finsler space
[33, 38]. Notice in Finsler geometry, curvatures are mostly defined with standard
local coordinates fx D .xi / 2 M; y D yi @xi 2 Tx Mg. But in the homogeneous
context, generally speaking, the local coordinates are not compatible with the
homogeneous structure, and invariant frames or Killing frames seem more suitable.
Applying invariant frames, L. Huang provides explicit formulas for all curvatures in
homogeneous Finsler geometry [19].
1.3.4 Some Recent Progress
With the local and global view points, we see an expecting research field in Finsler
geometry for which we have both a good local homogeneity and a good global
homogeneity. For a homogeneous Finsler space .G=H; F/, the isotropy subgroup
H is contained in the linear isometry group of To .G=H/. So .G=H; F/ has a
good local homogeneity when H is relatively big. Here are some recent progress
in homogeneous Finsler geometry which uses the local homogeneity directly or
indirectly.
1. The study on the curvatures of homogeneous Randers spaces and homogeneous
.; /-spaces [1, 7, 12, 13].
2. The study on the affine Berwald symmetric Finsler spaces [11]. Notice the
Berwald condition implies many properties of these spaces coincide with those
in Riemannian geometry. The Berwald condition can be replaced by assuming
the homogeneous space is symmetric [10], i.e., in the Ad.H/-invariant decompo-
sition g D h C m, we have m; m h. On one hand the symmetric condition
implies the Lie algebra h and thus the isotropy group H be big enough. On the
other hand, using Killing frames, its geodesic spray can be explicitly presented,
which coincides with that for a Riemannian symmetric metric, and proves the
symmetric homogeneous Finsler space G=H is Berwald.
3. The classification of Clifford-Wolf homogeneous Randers spaces [32], which

generalizes the corresponding work of Berestovskii and Nikonorov in the
Riemannian context [3], and some partial results for left-invariant Clifford-
Wolf homogeneous .; /-metrics and .1 ; 2 /-metrics on compact Lie groups
[14, 15].
4. The classification of positively curved homogeneous Finsler spaces [34, 35, 37,
38]. The main tool for this project is the homogeneous flag curvature formula
[19]. To efficiently reduce the classification problem to a solvable algebraic
problem, we need to use the local homogeneity encoded in the rank inequality for
a positively curved homogeneous Finsler space G=H [38], i.e., when assuming
G is compact, rkG 6 rkH C 1. There exists a sufficiently big torus in H, which
is important for this classification work. In particular, when dim G=H is odd, we
need a slightly better local homogeneity which requires the positively curved
homogeneous metric F on G=H to be reversible.
5. Define the flag-wise positively curved condition and find interesting examples.
In Finsler geometry, we can define a special version of non-negatively curved
condition. First, the metric is non-negatively curved, i.e., the flag curvature is
non-negative everywhere. Second, the metric is flag-wise positively curved, i.e.,
for any tangent plane P Tx M, we can find a suitable pole y 2 P, such that
K F .x; y; P/ > 0. Using the Killing navigation process, we can find many compact
coset spaces which admit non-negatively and flag-wise positively curved normal
homogeneous Randers metrics but no positively curved homogeneous metrics
[36]. The flag-wise positively curved condition alone seems to be a much weaker
condition. Perturbing a non-negatively curved homogeneous Riemannian metric
generically, we have a big chance to finding flag-wise positively curved metrics
[31].
Finally, I would like to summarize with our purpose of studying homogeneous
Finsler geometry. The curvatures are still in our main consideration. But now
they can be described by algebraic data and structures for the Lie algebras. So
geometric conditions are translated to algebraic conditions as well. In many cases,
those algebraic conditions can be corresponded with explicit classification lists of
homogeneous Finsler spaces, which provide precise examples for us to understand
the intrinsic nature of Finsler geometry in general.
Acknowledgements G. Falcone and K. Strambach were supported by the European Unions

Seventh Framework Programme (FP7/20072013) under grant agreement no. 317721.
References
1. H. An, S. Deng, Invariant (,)-metrics on homogeneous manifolds. Monatsh. Math. 154, 89

102 (2008)
2. D. Bao, S.S. Chern, Z. Shen, An Introduction to Riemannian-Finsler Geometry (Springer,
New York, 2000)
3. V.N. Berestovskii, Y.G. Nikonorov, Killing vector fields of constant length on locally symmet-
ric Riemannian manifolds. Transform. Groups 13, 2545 (2008)
4. D. Burde, Left-symmetric algebras, or pre-Lie algebras in geometry and physics. Cen. Eur. J.
Math. 4, 323357 (2006)
5. S.S. Chern, Z. Shen, Riemann-Finsler Geometry (World Scientific Publishers, River Edge, NJ,
2004)
6. A. Connes, D. Kreimer, Hopf algebras, renormalization and noncommutative geometry.
Commun. Math. Phys. 199, 203242 (1998)
7. S. Deng, The S-curvature of homogeneous Randers spaces. Differ. Geom. Appl. 27, 7584
(2009)
8. S. Deng, Homogeneous Finsler Spaces (Springer, New York, 2012)
9. S. Deng, Z. Hou, The group of isometries of a Finsler space. Pac. J. Math. 207, 149155 (2002)
10. S. Deng, Z. Hou, Minkowski symmetric Lie algebras and symmetric Berwald spaces. Geom.
Dedicata 113, 95105 (2005)
11. S. Deng, Z. Hou, On symmetric Finsler spaces. Isr. J. Math. 166, 197219 (2007)
12. S. Deng, Z. Hou, Homogeneous Einstein-Randers spaces of negative Ricci curvature. C.R.
Math. Acad. Sci. Paris 347, 11691172 (2009)
13. S. Deng, Z. Hu, Curvature of homogeneous Randers spaces. Adv. Math. 240, 194226 (2013)
14. S. Deng, M. Xu, Left invariant Clifford-Wolf homogeneous (,)-metrics on compact
semisimple Lie groups. Transform. Groups 20(2), 395416 (2015)
15. S. Deng, M. Xu, (a1;a2)-metrics and Clifford-Wolf homogeneity. J. Geom. Anal. 26, 2282
2321 (2016)
16. P. Finsler, ber Kurven und Flchen in allgemeinen Rumen (Dissertation, Gttingen, 1918)
(Birkhuser Verlag, Basel, 1951)
17. M. Gerstenhaber, The cohomology structure of an associative ring. Ann. Math. 78, 267288
(1963)
18. Th. Hawkins, Emergence of the Theory of Lie Groups: An Essay in the History of Mathematics
(Springer, New York, 2000), pp. 18691926
19. L. Huang, On the fundamental equations of homogeneous Finsler spaces. Differ. Geom. Appl.
40, 187208 (2015)
20. J.E. Humphreys, Introduction to Lie Algebras and Representation Theory. Graduate Texts in
Mathematics, vol. 9 (Springer, New York, 1978)
21. N. Jacobson, Lie Algebras (Dover Publications, New York, 1979)
22. M. Kontsevich, Deformation quantization of Poisson manifolds. Lett. Math. Phys. 66, 157216
(2003)
23. Y. Kosmann-Schwarzbach, Exact Gerstenhaber algebras and Lie bialgebroids. Acta Appl.
Math. 41, 153165 (1995)
24. J.-L. Koszul, Domaines borns homognes et orbites de groupes de transformationes affines.
Bull. Soc. Math. France 89, 515533 (1961)
25. J.-L. Koszul, Crochet de Schouten-Nijenhuis et cohomologie in Elie Cartan et les mathma-
tiques daujourdhui (Astrisque, hors srie, 1985), pp. 257271
26. M. Matsumoto, Foundations of Finsler Geometry and Special Finsler Spaces (Kaiseisha Press,
Otsushi, 1986)
27. J.-P. Serre, Lie Algebras and Lie Groups (Springer, Berlin, 2006)
28. Z. Shen, Lectures on Finsler Geometry (World Scientific, Singapore, 2001)
29. E.B. Vinberg, Convex homogeneous cones. Transl. Moskow Math. Soc. 12, 340403 (1963)
30. S. Weinberg, The Quantum Theory of Fields, vol. II (Cambridge University Press, Cambridge,
2005)
31. M. Xu, Examples of flag-wise positively curved spaces. Diff. Geom. Appl. 52, 4250 (2017)
32. M. Xu, S. Deng, Clifford-Wolf homogeneous Randers spaces. J. Lie Theory 23, 837845
(2013)
33. M. Xu, S. Deng, Killing frames and S-curvature of homogeneous Finsler spaces. Glasg. Math.
J. 57(2), 457464 (2015)
34. M. Xu, S. Deng, Normal homogeneous Finsler spaces. Transformation Groups (2017).
arXiv:1411.3053. https://doi.org/10.1007/s00031-017-9428-7
35. M. Xu, S. Deng, Towards the classification of odd-dimensional homogeneous reversible Finsler
spaces with positive flag curvature. Ann. Mat. Pura Appl. 196(4), 14591488 (1923)
36. M. Xu, S. Deng, Homogeneous Finsler spaces and the flag-wise positively curved condition.
arXiv:1604.07695 (2016)
37. M. Xu, W. Ziller, Reversible homogeneous Finsler metrics with positive flag curvature. Forum
Math. arXiv:1606.02474. https://doi.org/10.1515/forum-2016-0173
38. M. Xu, S. Deng, L. Huang, Z. Hu, Even-dimensional homogeneous Finsler spaces with positive
flag curvature, Indiana Univ. Math. J. 66(3), 949972 (2017)
Chapter 2
Remarks on Infinite-Dimensional
Representations of the Heisenberg Algebra
Camillo Trapani
Abstract Infinite-dimensional representations of Lie algebras necessarily invoke

the theory of unbounded operator algebras. Starting with the familiar example
of the Heisenberg Lie algebra, we sketch the essential features of this inter-
action, distinguishing in particular the cases of integrable and nonintegrable
representations. While integrable representations are well understood, nonintegrable
representations are quite mysterious objects. We present here a short and didactical-
minded overview of the subject.
2.1 Introduction
The theory of representations of Lie algebras by linear operators in infinite-

dimensional Hilbert spaces is one of the most fascinating branches of modern
mathematics; it involves algebra, topology, functional analysis, operator theory, and
differential manifolds and this fact produces at once its charm and its difficulty.
Motivations for these studies come from several mathematical subjects and also
from physical applications. In analysis, for instance, the theory of representations of
Lie algebras is a fundamental tool for the theory of partial differential equations
(in certain approaches). On the other hand, it provides a precise mathematical
framework for several problems in quantum theory and particle physics. In fact,
physical observables (i.e., measurable quantities) are represented, in general, by
operators acting on the Hilbert space of the states of the system and observables have
a natural structure as Lie algebra whose brackets are realized through commutation
relations of the corresponding operators. In fact, the operators A; B representing two
observables either commute (i.e., A; B D 0) or satisfy prescribed commutation
relations.
For instance, if Q represents the position of a particle and P its momentum, the
Heisenberg principle implies that P; Q D i1, where 1 stands for the identity
operator.
C. Trapani ()
Dipartimento di Matematica e Informatica, via Archirafi 34, 90123 Palermo, Italy
e-mail: camillo.trapani@unipa.it

24 C. Trapani
Similarly, the differential operators Lx ; Ly ; Lz representing the components of the

angular momentum satisfy the following commutation relations:
Lx ; Ly D iLz ; Ly ; Lz D iLx ; Lz ; Lx D iLy :
Of course, one can look at these commutation relation as to infinite-dimensional

representations of some Lie algebra.
A representation of a Lie algebra L is a linear map from L into an operator
algebra A acting on some vector space V.
Thus if V is taken to be a Hilbert space H , the most natural choice would be
taking as A the C*-algebra B.H / of all bounded operators in H .
We recall that a linear operator A in Hilbert space H is bounded (equivalently,
continuous) if there exists C > 0 such that
kAk 6 Ckk; 8 2 H :
If the Hilbert space is finite-dimensional, then every linear operator is bounded. If

dim.H / D 1, unbounded (and hence everywhere discontinuous) operators are
called on the stage.
As we shall see below, the operators involved in representations of Lie algebras
are often unbounded (and for some algebra they need to be unbounded!); the
drawback of involving unbounded operators is that the theory becomes more
complicated since one has to deal with sometimes difficult problems of domains,
extension of operators, self-adjointness problems, and so on.
But this is not the end of the story. In fact, if we want to represent Lie algebras by
unbounded operators we need a precise definition of commutator for two unbounded
operators and this is quite a difficult task!
In this paper, which has mostly didactical purposes, we quickly review the main
features of unbounded representations of Lie algebras, focusing our attentions to the
different (and, unfortunately, inequivalent) notions of commutators for unbounded
operators. In particular we will consider representations of the Heisenberg Lie
algebra h, which describes the commutation rules of the position Q and of the
momentum P in Quantum Mechanics. By the WienerWielandtvon Neumann
theorem (Theorem 2.2.1), the representations of this Lie algebra are necessarily
unbounded and the celebrated Stonevon Neumann theorem (Theorem 2.2.2)
characterizes the so-called integrable representations of h, i.e., representations
obtained by differentiating a unitary representation of a Lie group: they are all
unitarily equivalent to the Schrdinger representation that can be described, as we
shall see below, by the elegant formalism of annihilation and creation operators (see,
e.g., [16]). But, nonintegrable representations of h can also be easily constructed
and often occur in applications to Physics. Thus, it is worth discussing them and
try to understand what is preserved of the known theory under these very different
circumstances.
2 Remarks on Infinite-Dimensional Representations of the Heisenberg Algebra 25
2.2 Representations of the Heisenberg Algebra
As we mentioned in Sect. 2.1, some Lie algebras do not allow bounded represen-
tations in infinite-dimensional Hilbert space. A typical example comes from the
so-called canonical commutation relations (CCR, for short) that are nothing but
a special representation of the Heisenberg Lie Algebra h generated by three
elements a; b; c 2 h whose Lie brackets are defined by
a; b D c a; c D b; c D 0
The carrier space of is a Hilbert space H .

The algebra B.H / of bounded operators in H can easily be made into a Lie
algebra by defining A; B WD AB BA, A; B 2 B.H /.
Thus, as a first step, it is natural to suppose that takes its values in B.H /. So
we should have
W h ! B.H /;
.x; y/ D .x/.y/ .y/.x/; x; y 2 fa; b; cg:
The structure itself of the Heisenberg algebra h suggests that .a; b/ must be a
central element of B.H /, which consists of multiples of the identity 1 only. Hence
we require that
.a; b/ D .a/.b/ .b/.a/ D 1:
But, unfortunately,
Theorem 2.2.1 (Wiener, Wielandt, von Neumann) There exists no bounded
representation of the Heisenberg algebra satisfying .a; b/ D 1.
A very quick proof of this statement can be made by looking at the spectra of the
pretended bounded operators. We recall that, if A 2 B.H /, the spectrum .A/ of
A is defined as
.A/ D f 2 C W A 1 62 GL.H /g ;
where GL.H / denotes the group of invertible elements of B.H /. For every A 2
B.H /, .A/ is a nonempty compact subset of the complex plane (see, e.g., [14]).
The familiar rules of the spectra give
..a/.b// [ f0g D ..a/.b// [ f0g
..a/.b// D 1 C ..b/.a//;
contradicting the compactness of the spectra.

26 C. Trapani
This is not a case, since as shown by Doebner and Melsheimer [10], representa-
tions of every noncompact Lie algebra are unbounded.
2.2.1 An Unbounded Representation of the Heisenberg

Algebra
We now define the so-called Schrdinger representation of the Heisenberg algebra.

The lesson of Theorem 2.2.1 is that we need to introduce unbounded representations
and, consequently, unbounded operator algebras.
For readers convenience we recall some features of unbounded operators. The
first one is that they are not, in general, defined on the whole Hilbert space H : a
linear operator A is in fact a linear map defined on a subspace D.A/ of H , called
the domain of A, which one supposes to be dense in H . An unbounded operator
is necessarily everywhere discontinuous in its domain. A regularity requirement
for an unbounded operator is its closedness: this means that the graph f.; A/ 2
D.A/ H g is a closed subspace of H H . An operator is called closable if it has
a closed extension.
Now we turn to unbounded operator algebras.
Let D be a dense domain in Hilbert space. Then, the family of closable operators
L .D/ WD fA W D ! D W A D Dg
is a *-algebra of (in general) unbounded operators, with the usual operations of

addition, multiplication by scalars, multiplication (i.e., composition) of operators
and involution defined by A WD A D, where A denotes the hilbertian adjoint of
A defined as
8
< D.A / D f 2 H W 9 2 H such that hA j i D h j i ; 8 2 Dg
:
A D ; 2 D.A /
For readers convenience, we recall that an operator A, with domain D.A/ is

symmetric if hA j i D h j Ai for every ; 2 D.A/. This means that the adjoint
A of A extends A. The operator A is called self-adjoint if A D A and essentially
self-adjoint if its minimal closed extension A is self-adjoint. For more details we
refer to [14, 16].
For the Schrdinger representation of h, we take as domain the Schwartz space
S .R/ . L2 .R/, clearly) of C1 -functions f that are rapidly decreasing; i.e.,
sup jxk .Dm

x f /.x/j < 1; 8k; m 2 N;
x2R
where Dx denotes the derivation operator with respect to x. Of course, S .R/

L2 .R/.
We then define the annihilation operator A and the creation operator A as

follows:
1 1
Af D p .xf C Dx f / and A f D p .xf Dx f /; f 2 S .R/:
2 2
Then, both A and A are elements of L .S .R// and
AA f A Af D f ; 8f 2 S .R/:
For a; b; c 2 h, as above, we put
.a/ D A .b/ D A
Then .a/; .b/ D 1. The operator A A, the so-called number operator, is essen-
tially self-adjoint on S .R/ and has discrete simple spectrum; in fact, .A A/ D N.
Eigenvectors of A A are obtained by successive application of A to the vacuum
x2
vector 0 .x/ D e 2 , which is annihilated by A; inpother words, an eigenvector
associated to the eigenvalue k 2 N is given by k D . 2A /k 0 .
The Schrdinger representation can be, equivalently, defined through the follow-
ing operators:
Qf D xf Pf D if 0 P; Q f D if ; f 2 S .R/:
Then, one has
1 1
Af D p .Q f C iPf /; A f D p .Q f iPf /
2 2
and D 12 .Q2 C P2 C 1/ is essentially self-adjoint on S .R/.

The Schrdinger representation of h is of fundamental importance, since it is
essentially the unique integrable and irreducible representation of the Heisenberg
algebra.
Definition 2.2.1 A representation of a Lie algebra L is called integrable if there
exists a connected and simply connected Lie group G and a unitary representation
U of G such that is the differential of U; i.e. D dU.
The integrability of the Schrdinger representation depends clearly from h being
the Lie algebra of a Lie group G. Let us consider, in fact, as G the set of 3 3-
matrices of the form
0 1
1
g.; ; / D @ 0 1 A ; ; 2 R
00 1
Checking that G is a Lie group is a straightforward exercise. Then, if we call

a; b; c the generators of the one-dimensional subgroups, the Lie bracket of a; b; c
28 C. Trapani
are nothing but a; b D cI a; c D b; c D 0; that is, the Lie algebra associated to

G is exactly h.
For every 2 R n f0g a unitary representation of G in L2 .R/ is obtained by
putting, for every f 2 L2 .R/,
.U .g.; ; //f /.x/ WD exp.i.x C //f .x C /; x 2 R:
By differentiating one finds

d
.a/ D ; .b/ D ix; .c/ D i:
dx
It is not difficult to check that this representation is equivalent to that defined above.
2
The operators .a/ and .b/ leave S .R/ invariant and the operator D dtd C
2 t2 C 2 is essentially self-adjoint on S .R/.
A fundamental result about representations of h is the following theorem due to
M.H. Stone and J. von Neumann.
Theorem 2.2.2 (Stone, von Neumann) Every integrable irreducible representa-
tion of the Heisenberg Lie algebra is unitarily equivalent to the Schrdinger
representation.
As a consequence of this theorem, an integrable representation of h decomposes
into the direct sum of (possibly, infinitely many) representations each one of
them being unitarily equivalent to the Schrdinger representation, since every
representation can always be decomposed into irreducible ones.
The example discussed above can be cast into the more general situation
described by the Grding theory. This approach to the infinite-dimensional repre-
sentation theory of Lie algebras starts essentially with a work of L. Grding of 1947
[12]. The cornerstone of the theory is the following theorem [9, Chap. 11].
Theorem 2.2.3 Let U be a unitary representation of a d-dimensional Lie group G
in Hilbert space H and D dU the corresponding representation of the associated
Lie algebra LG . Then,
(i) there exists a common invariant dense domain DG for the operators .Xi /,
where X1 ; : : : ; Xd 2 LG are the infinitesimal generators of U;
(ii) the Nelson operator X12 C X22 C C Xd2 is essentially self-adjoint on DG .
It is clear that is integrable by definition. The proof of Theorem 2.2.3 is
constructive: thus once a unitary representation of the Lie Group G is given, we
can explicitly construct the so-called Grding domain. The converse of the previous
theorem also holds, provided that DG consists of analytic vectors f for . A vector
f 2 H is analytic for if, for some s > 0,
X1 X
1
k.Xi1 / .Xid /f ksn < 1:
nD0
n 16i ;:::;i 6d
1 n
The notion of analytic vector (that appears first in HarishChandra) was intro-
duced, in a slight different context by Nelson [15], but it is with Grding theory that
it develops all its potentiality [13].
2.3 Nonintegrable Representations of the Heisenberg

Algebra
Integrable representations of Lie algebras are, as we have seen, quite well under-
stood. But this kind of well-behavior is not so frequent in practice and it is not
difficult to construct examples where integrability fails (sometimes unexpectedly).
Let us consider the following representation of h in the Hilbert space H WD
L2 .0; 1/. Similarly, to what we have done before, we define
.Q f /.x/ D xf .x/; f 2H:
D.P/ D f f 2 C1 .0; 1/ W f .0/ D f .1/ D 0; f 0 2 L2 .0; 1/g
.Pf /.x/ D if 0 .x/
Then Q is bounded and self-adjoint, P is neither bounded nor self-self-adjoint.

Performing the same construction as before we put
1 1
A D p .Q iP/ and B D p .Q C iP/
2 2
Of course A; B D 1; B D A but B A , the hilbertian adjoint of

A. By Grding theorem the integrability of the representation depends on the
existence of a common invariant dense domain D where the Nelson operator
D 12 .P2 C Q2 / is essentially self-adjoint. The largest domain invariant for P
and Q is clearly C01 .0; 1/; but is not essentially self-adjoint on C01 .0; 1/, since
P is not. Thus this representation is not integrable. This is, in a sense surprising,
since the representation defined in this way can be taken for describing position and
momentum of a particle confined to move in an interval of finite length.
On the other hand, it should be mentioned that for several physical models it
is not possible to find a common invariant dense domain D with the prescribed
properties and this fact provides a good motivation for considering nonintegrable
representations.
30 C. Trapani
2.3.1 Commutation Relations of Operators: Usual Approach
At this point, a discussion about the mathematical meaning to give to the commu-
tation relations of two unbounded operators (or even to the fact that they commute)
is in order. In fact, studying operators A; B such that B A and A; B D 1 (in a
sense to be precise) is an unavoidable step (but just the first one) for the analysis
of nonintegrable representations. Of course, the significant cases are D 0 or
D 1. The latter has been analyzed in [7, 8] moving from the idea of pseudo-bosons
discussed at length by Bagarello in several papers [46].
Let A; B be two closed operators with dense domains, D.A/ and D.B/, respec-
tively, in Hilbert space H . In order to give a meaning to the formal commutation
relation A; B D 1, we require that the identity AB BA D 1 holds, at least, on a
dense domain D of Hilbert space H . In other words, we assume that there exists a
dense subspace D of H such that
(D.1) D D.AB/ \ D.BA/;
(D.2) AB BA D , 8 2 D, where, as usual, D.AB/ D f 2 D.B/ W B 2
D.A/g.
As in [6], we will suppose that
(D.3) D D.A / \ D.B /.
2.3.1.1 The Case D 0
Let A, B be linear operators in Hilbert space H . What we mean when we say that
A and B commute? It is clear that, if A; B are both bounded operators, the equation
AB BA D 0 has a precise meaning; but the same algebraic relation can be very
poor in information if both A and B are unbounded. So, many different mathematical
approaches have been developed for handling this case.
Before going forth, we recall some basic facts about self-adjoint operators.
A spectral family on the real line R is a family fE./g of projection operators,
depending on the real parameter , satisfying the following conditions:
(a) E./ 6 E./ or, equivalently, E./E./ D E./, if 6 ;
(b) lim E. C / D E./x, for every 2 H ;
!0C
(c) lim E./ D 0 e lim E./ D , for every 2 H .
!1 !C1
A crucial result of operator theory is the spectral theorem for self-adjoint operators:
Theorem 2.3.1 To every self-adjoint operator A there corresponds a unique spec-
tral family such that
Z
A D dE./;
R
for every 2 D.A/.

Moreover one has

Z
2
D.A/ D 2 H W d hE./ j i < 1 :
R
A relevant consequence of this theorem is the functional calculus for self-adjoint

operators which allows to define functions of an operator. For instance, if A is self-
adjoint and fE./g is its spectral family, then
Z
eitA D eit dE./; 2H
R
defines a one-parameter group UA .t/ WD eitA of unitary operators in H . The operator

A plays the role of infinitesimal generator of UA .t/; that is
UA ./ 1
iAf D lim f; f 2 D.A/:
!0
Let now A and B be two self-adjoint operators in H . The notion of strong

commutation of A and B can be expressed by anyone of the following equivalent
statements:
1. the associated spectral families fEA ./g, fEB ./g, ; 2 R commute;
2. the associated unitary groups UA .t/ WD eiAt , UB .s/ WD eiBs commute, for every
s; t 2 R;
3. the resolvent functions .AI/1 ; .BI/1 , commute, for every 2 .A/; 2
.B/,
.X/ denotes the resolvent set of the operator X; i.e. the complement of the
spectrum. We refer again to [14] or [16] for the basic definitions and for the above
mentioned results.
Also in the case where the operators A; B 2 L .D/ the equation AB D BA is
well defined, since, as we have seen, L .D/ is a *-algebra. But Nelsons example
[16] shows that there exist two operators A; B and a dense domain D such that
1. A; B 2 L .D/:
2. A; B essentially self-adjoint in H ;
3. AB D BA, for every 2 D
but the spectral families do not commute; in contrast with what happens for bounded
self-adjoint operators.
On the other hand, strong commutation of self-adjoint operators has a precise
meaning when they are supposed to represent physical observables: the two
observables are, in this case, compatible (i.e., they can be measured simultaneously)
and a joint probability distribution for these observables does exist.
32 C. Trapani
2.3.1.2 The Case D 1
Parallel to strong commutation is the formulation given by Weyl for the formal
commutation relation A; B D 1 through the groups of unitaries they generate (when
they do!).
Thus, if A and B are self-adjoint operators, the equation A; B D 1 can be
translated by the celebrated Weyl commutation relations:
UA ./UB ./ D e UB ./UA ./; 8; 2 R:
But for nonself-adjoint operators (excluding trivial cases) this approach is mostly
impossible.
2.3.2 Commutation Relations of Operators: A Different

Approach
Let A, B be two operators satisfying (D.1)(D.3). Then, the operators S WD A D

and T WD B D are elements of the partial *-algebra L .D; H / and satisfy the
equality

T j S S j T D h j i ; 8; 2 D: (2.1)
We recall that L .D; H / denotes the set of all (closable) linear operators X such
that D.X/ D D; D.X*/ D: The set L .D; H / is a partial *-algebra with respect
to the usual sum X1 C X2 , the scalar multiplication X, the involution X 7! X WD
X*D and the (weak) partial multiplication X1 X2 D X1 *X2 , defined whenever
X2 is a weak right multiplier of X1 (we shall write X2 2 Rw .X1 / or X1 2 Lw .X2 /),
that is, whenever X2 D D.X1 */ and X1 *D D.X2 */, [2].
The commutation relation (2.1) is the main matter of what follows.
Let us first consider the case where S and/or T are generators of some weakly
continuous semigroup V.t/ of bounded operators.
More precisely,
Definition 2.3.1 Let V.t/, t > 0, be a semigroup of bounded operators. We say that
X0 2 L .D; H / is the D-generator of V.t/ if

V.t/ 1
lim j D hX0 j i ; 8; 2 D:
t!0 t
In [7] the following definitions were given.

Definition 2.3.2 Let S; T 2 L .D; H /. We say that

(CR.1) the commutation relation S; T D 1 is satisfied (in L .D; H /) if,
whenever S T is well defined, T S is well-defined too and S T T S D
1.
(CR.2) the commutation relation S; T D 1 is satisfied in weak sense if

T j S S j T D h j i ; 8; 2 D:
(CR.3) the commutation relation S; T D 1 is satisfied in quasi-strong sense if S is

the D-generator of a weakly continuous semigroups of bounded operators
VS ./ and

hVS ./T j i VS ./ j T D hVS ./ j i ; 8; 2 DI > 0:
(CR.4) the commutation relation S; T D 1 is satisfied in strong sense if S and T

are D-generators of weakly continuous semigroups of bounded operators
VS ./; VT ./, respectively, satisfying the generalized Weyl commutation
relation
VS ./VT ./ D e VT ./VS ./; 8; > 0:
The following implications hold [7]:
(CR.4) ) .CR:3/ ) .CR:2/ ) .CR:1/:
Example 1 The implications in the opposite direction do not hold. For instance,
as shown by Fulgede and Schmdgen [11, 17, 18], there exist two essentially self-
adjoint operators P; Q with common invariant dense domain D such that PQ
QP D i, for 2 D, but the unitary groups UP .t/; UQ .s/ generated by P, Q do
not satisfy the Weyl commutation relation UP .t/UQ .s/ D eits UQ .s/UP .t/, s; t 2 R.
2.4 Commutation Relations and Spectral Behavior
It is well known that the traditional commutation relation A; A D 1 gives rise

(under some additional conditions) to a rich series of spectral results that have
plenty of applications [Sect. 2.1]. Here we will shortly discuss analogous results
(mostly presented in [7, 8]) for two operators S; T satisfying (in one sense or the
other described in Definition 2.3.2) the commutation relation S; T D 1.
34 C. Trapani
2.4.1 Existence of Eigenvectors
We begin with assuming that S; T 2 L .D; H / satisfy the commutation relation

S; T D 1 in weak sense. Assume that there exists a vector 0 0 2 D such that
S0 D 0. Then

T0 j S D h0 j i ; 8 2 D: (2.2)
This implies that T0 0 (otherwise, 0 D 0) and
T0 2 D.S /; S T0 D 0 2 D:
Thus T can be applied once more and we get
.TS /T0 D T.S T0 / D T0 : (2.3)
Hence T0 is an eigenvector of TS with eigenvalue 1.

Let us now assume that T0 2 D. Then it is easily seen that
2
T 0 j S D 2 hT0 j i ; 8 2 D: (2.4)
This, in turn, implies that T 2 0 2 D.S / and S T 2 0 D 2T0 . and
.TS /T 2 0 D 2T 2 0 :
Iterating this procedure we conclude that

Proposition 2.4.1 Let S; T 2 L .D; H / satisfy the commutation relation S; T D
1 in weak sense. Assume that there exists a nonzero vector 0 2 D such that S0 D 0
and the vectors T0 ; T 2 0 ; : : : T n1 0 all belong to D. Then
(i) T n 0 is an eigenvector of TS with eigenvalue n;
(ii) T n1 is eigenvector of S T with eigenvalue n.
0
Proposition 2.4.2 Let S; T 2 L .D; H / satisfy the commutation relation S; T D

1 in weak sense. Let 2 D and assume that T k 2 D for k 6 n, n 2 N [ f1g. Then
S 2 D..T /k /, k 6 n and
ST k .T /k S D kT k1 ; k 6 n:
Remark 1 In particular if S D 0 then ST k D kT k1 .

If the assumptions of Proposition 2.4.1 are satisfied one may have that the largest
n for which T n 0 2 D is finite or infinite. As we have seen the point spectrum
p .TS / contains all natural numbers up to n. Let us denote by N0 the subspace of
D spanned by fo ; T0 ; : : : T n 0 g and by N WD N0 its closure in H . Clearly TS
leaves N0 invariant. The restriction of TS to N0 , denoted by .TS /0 , behaves in

quite regular way. We have indeed
Proposition 2.4.3 In the assumptions of Proposition 2.4.1, the point spectrum

p .TS0 / the operator .TS /0 , restriction of TS to N0 , consists exactly of the
set f0; 1; : : : ; ng, where n 2 N [ f1g is the largest natural number such that the
vectors T0 , T 2 0 ; : : :, T n1 0 all belong to D. Each eigenvalue is simple (in N0 ).
Example 2 Let us consider the Hilbert space L2 .R; wdx/ where the weight w is a
positive continuously differentiable function with the properties
1. lim w.x/ D 0;
jxj!C1
R
2. R w.x/dx < 1.
Let

Z x
2 2
D.p/ D f 2 L .R; wdx/ W 9g 2 L .R; wdx/; f .x/ D g.t/dt :
1
For shortness, we adopt the notation f 0 .x/ D g.x/, for f 2 D.p/.
D.q/ D ff 2 L2 .R; wdx/ W xf .x/ 2 L2 .R; wdx/g:
Put D D D.q/ \ D.p/. Then both the operators S; T defined by
.Sf /.x/ D f 0 .x/; .Tf /.x/ D xf .x/; f 2D
map D into L2 .R; wdx/.

The operator T is symmetric in D. As for S, we have g 2 D.S / if, and only if
0
g 2 D.p/ and g ww 2 L2 .R; wdx/. In this case
w0 .x/
.S g/.x/ D g0 .x/ g.x/ :
w.x/
Hence S 2 L .D; H /, with H D L2 .R; wdx/ if, and only if,

Z
jw0 .x/j2
jg.x/j2 dx < 1:
R w.x/
This is certainly satisfied if, for instance, w0 =w is a bounded function on R. An easy

computation shows that the commutation relation S; T D 1 is satisfied in weak
sense.
The function u0 .x/ D 1, for every x 2 R, is clearly in the kernel of S for every
function w satisfying the assumptions made so far.
36 C. Trapani
Now we make some particular choice of w.

1. Let us consider w.x/ D w .x/ D .1 C x4 / , > 34 . It is easily seen that w .x/
satisfies all the conditions we have required (for instance, w0 =w is bounded).
The function u0 .x/ D 1, which belongs to L2 .R; w dx/ for any > 34 , satisfies
Su D 0 and the largest n for which T n u0 belongs to D satisfies n < 2 32 .

Hence, the dimension of the corresponding subspace N0 is 2 32 C 1.
2
2. Let us now take w.x/ D ex =2 and D the subspace consisting of all polynomials
in x. In this case the functions uk .x/ D xk , k D 1; 2; : : : ; belong to D and they
satisfy TS uk D kuk for every k 2 N. The subspace N0 coincides in this case
with D. One can readily check that every complex number with < > 12 is
an eigenvalue of TS ; but the corresponding eigenvector is in D if and only if
< is a natural number.
As we have seen with the previous examples the subspace N0 spanned by fT k 0 ; k 2
Ng can be finite dimensional. Thus N WD .TS /0 is a bounded symmetric operator
on N0 D N Cn , having the numbers 0; 1; : : : ; n as eigenvalues. Hence N
is positive and thus there exists an operator C 2 B.N / such that N D C C.
None of the possible solutions of this operator equation can, however, satisfy the
commutation relation C; C D 1, due to the WienerWielandtvon Neumann
theorem. If N0 is infinite dimensional, then N may fail to be symmetric, as the
last case in Example 2 shows.
2.4.2 Existence of Intertwining Operators
Assume now that the assumptions of Proposition 2.4.1 hold not only for the pair
S; T but also for the pair T ; S . This means that we assume that there exists also
a nonzero vector 0 2 D such that T 0 D 0 and the vectors S 0 , .S /2 0 ; : : :
.S /m1 0 all belong to D, then Proposition 2.4.1 gives that
(i) .S /m 0 is an eigenvector of S T with eigenvalue m;
(ii) .S /m1 0 is eigenvector of T S with eigenvalue m.
Let m 2 N [ f1g be the largest natural number such that the vectors S 0 ,
.S / 0 ; : : : ; .S /m1 0 all belong to D and M0 the linear span of these vectors.
2
Then the point spectrum p ..S T /0 /, of the operator .S T /0 WD S T M0 ,

consists, as before, of the numbers f0; 1; : : : ; mg.
One may wonder if any relation between the two numbers n and m can be
established. The answer is negative, in general. Indeed, the operators S; T considered
in the second case of Example 2 provide an instance where n D 1 and m D 0.
Thus it is apparently impossible to find a relationship between N0 and M0 without
additional assumptions.
Let us now call k WD p1k T k 0 , k D 1; : : : ; n and r WD p1 .S /r 0 , r D 1; : : : ; m.
r
It is always possible to choose the normalization of 0 and 0 in such a way that
h0 j 0 i D 1.We put F WD fk I k D 1; : : : ; ng F WD fr I r D 1; : : : ; mg. Then, the

sets F and F are biorthogonal; i.e.,

i j j D i;j ;
for all i D 1; : : : ; n and j D 1; : : : ; m.

Assume now that n D m D 1. Thus the subspaces N0 and M0 are both infinite
dimensional. Then we can define two operators which obey interesting intertwining
relations. More in detail, let us define K via its action on the basis F : K .j / D j ,
j 2 N. We can also introduce a second operator K via its action on the second basis
constructed above, F : K .j / D j , j 2 N. Both the operators K and K are then
extended by linearity to M0 and N0 , respectively. It is clear that one is the inverse
of the other: K D K1 , but in general neither K nor K are bounded. A direct
computation shows that they obey the following intertwining relations:

K TS
D S T K
; 8
2 M0 I

K S T D TS K ; 8 2 N0 :
In particular, if N0 D M0 D H and K and K are bounded, it is possible to show

that F and F are Riesz bases of H and an orthonormal basis e D fej g can be
1=2
defined by, for instance, ej D K j (see [4]).
It is worth mentioning here that the properties discussed in Sects. 2.4.1 and 2.4.2
depend in an essential way on the nonzero vectors 0 and 0 satisfying, respectively,
S0 D 0 and T 0 D 0: choosing different elements in the kernels of S and T may
give drastically different results.
2.4.3 Time Operators and Weyl Extensions
In [18] Schmdgen studied pairs of operators .T; H/ with T symmetric and H self-
adjoint such that.
1. eitH D.T/ D.T/;
2. TeitH D eitH .T C t/
The operators T; H are regarded as members of L .D; H / with D D D.T/ \
D.H/. and the two conditions above are equivalent to say that the operators T; S WD
iH satisfy S; T D 1 in quasi-strong sense.
Definition 2.4.1 T is time operator for H if .T; H/ satisfies the two conditions
above.
In many examples, mostly taken from physics, H is a semibounded operator (this
often happens, for instance, when H is the Hamiltonian of some physical system).
38 C. Trapani
Theorem 2.4.4 (Arai) Let H be self-adjoint and semibounded. Then no time

operator T of H can be essentially self-adjoint.
Indeed, as shown in [3], the spectrum .T/ is one of the following sets
1. C, if H is bounded
2. C or C D fz 2 C W =z > 0g if H is bounded below
3. C or D fz 2 C W =z 6 0g if H is bounded above.
The assumption eitH D.T/ D.T/ is quite strong and weaker conditions should
be analyzed.
Definition 2.4.2 Let S; T be symmetric operators of L .D; H /. We say that fS; Tg
satisfy the weak Weyl commutation relation if there exists a self-adjoint extension H
of S such that
(ww1 ) D.T/
D.H/;

(ww2 ) eitH j T D .T C t/ j eitH ; 8; 2 DI t 2 R.
Then H is called the weak Weyl extension of S (with respect to T).
Proposition 2.4.5 Let fS; Tg satisfy the weak Weyl commutation relation and let H
be the weak Weyl extension of S. The following statements hold.
(i) Suppose that T is essentially self-adjoint. Then fH; Tg satisfy the Weyl commu-
tation relation, that is,
eitH eisT D eits eisT eitH ; 8s; t 2 R (2.5)
(ii) If H is semibounded, then T is not essentially self-adjoint.

Example 3 Let I be an interval of the real line. Denote by q the multiplication
operator on L2 .I/ by the variable x 2 I
For every choice of I, q is a self-adjoint operator.
Let p be the operator on L2 .I/ defined as follows:
D. p/ WD C01 .I/
. pg/.x/ WD ig0 .x/; g 2 D.p/
Case 1 I D 0; 1/. In this situation, q is positive (but unbounded); p is a time
operator of q and .p/ D C :
Case 2 I D .`=2; `=2/, ` > 0. q is a bounded self-adjoint operator and p is a
time operator of q. But, in this case, .p/ D C:
2.5 Conclusions
The main lesson one can draw from the previous discussion is that infinite-
dimensional representations of Lie algebras can be reasonably handled only by
invoking the whole apparatus of unbounded operator algebras theory [19]. As
it was clear from the very first steps of the theory of unbounded algebras, in
the early 1970s, the theory of representations of Lie algebras (or enveloping Lie
algebras) was the natural arena for its applications also for the large number of
applications in Quantum Physics, where the relevance of invariance under Lie
groups of transformations of the dynamical equations was clear all along. This case
(integrable representations) was completely under control by the Grding theory.
But, even if someone says that Nature knows groups rather than algebras,
nonintegrable representations of Lie algebras arise in some recent developments of
quantum theories and this is a good reason for them to deserve a more detailed
study. The weak form (2.1) of the commutation relations involves, as we have
discussed so far, more general structures for operators families: partial *-algebras
provide, in our opinion, the natural framework where studying a generalized form
of the representation theory. The example of nonintegrable representations of the
Heisenberg Lie algebra h that we have discussed in some detail shows that very
few features are preserved in comparison with integrable representations (from the
spectral point of view, for instance). Thus, there is a lot of work to be done. We hope
to have ignited the interest of some young mathematician to this promising research
field.
References
1. J.-P. Antoine, C. Trapani, Partial Inner Product Spaces: Theory and Applications. Springer
Lecture Notes in Mathematics, vol. 1986 (Berlin, Heidelberg, 2009)
2. J.-P. Antoine, A. Inoue, C. Trapani, Partial *-Algebras and Their Operator Realizations
(Kluwer, Dordrecht, 2002)
3. A. Arai, Generalized weak Weyl relation and decay of quantum dynamics. Rep. Math. Phys.
17, 10711109 (2005)
4. F. Bagarello, Pseudo-bosons, Riesz bases and coherent states. J. Math. Phys. 50, 023531 (2010)
(10 pp.). doi:10.1063/1.3300804
5. F. Bagarello, Examples of Pseudo-bosons in quantum mechanics. Phys. Lett. A 374, 3823
3827 (2010)
6. F. Bagarello, Pseudo-bosons, so far. Rep. Math. Phys. 68, 175210 (2011)
7. F. Bagarello, A. Inoue, C. Trapani, Weak commutation relations of unbounded operators and
applications. J. Math. Phys. 52, 113508 (2011)
8. F. Bagarello, A. Inoue, C. Trapani, Weak commutation relations of unbounded operators:
nonlinear extensions. J. Math. Phys. 53, 123510 (2012)
9. A.O. Barut, R. Racza, Theory of Group Representations and Applications (PWN, Warszawa,
1980)
10. H.D. Doebner, O. Melsheimer, Limitable dynamical groups in quantum mechanics. J. Math.
Phys. 9, 16381656 (1968)
11. B. Fuglede, On the relation PQ PQ D iI. Math. Scand. 20, 7988 (1967)
12. L. Grding, Note on continuous representations of Lie groups. Proc. Natl. Acad. Sci. U.S.A.
33, 331332 (1947)
13. L. Grding, Vecteurs analytiques dans les rpresentations des groupes de Lie. Bull. Soc. Math.
France 88, 7393 (1970)
14. R.V. Kadison, J.R. Ringrose, Fundamentals of the Theory of Operator Algebras, vol. I
(American Mathematical Society, Providence, 1997)
40 C. Trapani
15. E. Nelson, Analytic vectors. Ann. Math. 70, 572615 (1959)

16. M. Reed, B. Simon, Methods of Modern Mathematical Physics. Functional Analysis, vol. I
(Academic, San Diego, 1980)
17. K. Schmdgen, On the Heisenberg commutation relations I. J. Funct. Anal. 50, 849 (1983)
18. K. Schmdgen, On the Heisenberg commutation relations II. Publ. RIMS, Kyoto Univ. 19,
601671 (1983)
19. K. Schmdgen, Unbounded Operator Algebras and Representation Theory (Birkhuser-
Verlag, Basel, 1990)
Chapter 3
Character, Multiplicity, and Decomposition
Problems in the Representation Theory
of Complex Lie Algebras
Peter Fiebig
Abstract This essay is meant as an introduction to a very successful approach

towards understanding the structure of certain highest weight categories appearing
in algebraic Lie theory. For simplicity, the focus lies on the case of the category
O of representations of a simple complex Lie algebra. We show how the approach
yields a proof of the classical KazhdanLusztig conjectures that avoids the theory
of D-modules on flag varieties.
3.1 Introduction
This expository article is meant to give an overview on a certain approach towards

the representation theory of Lie algebraic structures. This approach is due to
W. Soergel and pieces together a couple of ideas that, in one form or the other,
make sense in various situations. Its main goal is to understand representation theory
from a categorical point of view, i.e. to understand the category of representations
rather than the structure of an individual object. Yet sometimes it is strong enough
to solve questions that can be formulated categorically. An example is the problem
to determine irreducible characters.
For the sake of simplicity only the most classical example of complex semisimple
Lie algebras and their highest weight modules will be considered. The leitmotif
is the famous and most influential KazhdanLusztig conjecture from 1979 on the
irreducible highest weight characters. The category that surrounds this conjecture
is the not less famous category O that was introduced in the early 1970s by J.
Bernstein, I.M. Gelfand, and S.I. Gelfand. We show how one can translate the
character problem into the contexts of Soergel bimodules, of parity sheaves or of
moment graph sheaves, where strong additional methods and results to work with
are present.
P. Fiebig ()
Department Mathematik, FAU ErlangenNrnberg, Cauerstrae 11, 91058 Erlangen, Germany
e-mail: fiebig@math.fau.de

42 P. Fiebig
Here is a short summary of this article. After describing the Character Problem
and the KazhdanLusztig Conjecture the first problem is translated into the
Multiplicity Problem I on the JordanHlder multiplicities of Verma modules.
Section 3.3 introduces the category O. The BGG-reciprocity relates JordanHlder
multiplicities of Verma modules to Verma module multiplicities in Verma flags of
projective objects in O. Section 3.3 ends with introducing the translation functors
which, together with the BGG-reciprocity, allow to translate the Multiplicity Prob-
lem I into the Decomposition Problem I for Bott-Samelson modules. Section 3.4
describes in more detail the deformed category O and provides proofs of the block
decomposition and the BGG-reciprocity. Section 3.5 is to encounter W. Soergels
theory that allows to translate the Decomposition Problem I into the Decomposition
Problem II for a certain iterated tensor product of a commutative algebra Z . In
Sect. 3.6 sketches are given of translations of the Decomposition Problem II into
the language of Soergel bimodules, moment graph sheaves, or parity sheaves. The
article ends with mentioning how one can proceed further: the theory of intersection
cohomology complexes or, alternatively, the work of B. Elias and G. Williamson on
the Hodge theory of Soergel bimodules can be used to solve the remaining problem
and hence to prove the KazhdanLusztig conjecture. In the short epilogue a few
words are said about the modular analogue of Soergels theory. As prerequisites we
assume basic knowledge on Lie algebras and, in particular, the structure theory of
semisimple complex Lie algebras. The combinatorial data of the root system, the
Weyl group, etc. will be used throughout the article. Due to limitation of space full
details cannot be given. However, hopefully the reader will find enough citations to
original or review papers and books to be able to delve more deeply into the theory.
Surely, as we move along, the arguments become sketchier and sketchier.
This overview has some overlap with Humphreys book [13] on the category
O, but the treatment of the categorical structure deviates at some places from the
theory developed there. In particular, no reference is made to HarishChandras
work on the center of the universal enveloping algebra. Instead a deformation theory
that goes back at least to Jantzens book [14] is used. The advantage is that almost
everything that is developed here can be more or less immediately generalized to
the case of symmetrizable KacMoody algebras. There are analogous approaches
for modular Lie algebras or quantum groups as well (cf. [1]). As mentioned before,
the main ideas first appeared in W. Soergels fundamental paper [19]. However,
again this paper deviates a little from this classical paper, and the expert will at
certain places see the moment graph theory appear in disguise. There is also some
overlap with the paper [11], yet the present article does not focus on the moment
graph theory or the theory of parity sheaves. One might view it as a significantly
expanded version of Sect. 3 of [11].
3 Character, Multiplicity and Decomposition Problems 43
3.2 The Character Problem and the KazhdanLusztig

Conjecture
Our main motivation is to understand the representation theory of a Lie algebra g,

and a decisive first step is to understand the structure of its irreducible modules. For
this, we would first like to have a list of all irreducible modules up to isomorphism.
In general, this already is a quite hopeless task, which forces us to focus on special
cases. For example, we might want to specify the class of Lie algebras that we
consider, or we could restrict our attention to modules that are not only irreducible,
but also satisfy some additional assumptions. We will do both: assume that g is a
simple Lie algebra over the field C of complex numbers, and that the irreducible
modules have a highest weight.
3.2.1 Highest Weights
The notion of highest weight depends on some choices. Fix a Cartan subalgebra
h g and a Borel subalgebra b g that contains h. A Cartan subalgebra is a
maximal abelian subalgebra of g such that the adjoint action adW h ! EndC .g/,
H 7! .X 7! H; X/ is semisimple. Then the Lie algebra g decomposes into weight
spaces under this action, i.e.
M
gD g ;
2h?
where we define for each 2 h? D HomC .h; C/ the weight space g WD fX 2 g j

H; X D .H/X 8H 2 hg. The maximality of h implies g0 D h, and one calls the
set R WD f 2 h? j a 0; g f0gg the set of roots of g with respect to h. Hence
M
gDh g :
2R
A Borel subalgebra is a maximal solvable subalgebra of g. There always exist

Borel subalgebra containing a given Cartan subalgebra h. Then h acts on b via the
adjoint action as well, and the corresponding weights form a subset RC of R. Then
R h? satisfies the axioms of a root system, and RC R is a choice of a system of
positive roots. Moreover, our choices yield a partial order on the set h? . It is defined
by 6 if can be written as a sum of elements in RC .
44 P. Fiebig
3.2.2 Modules with Highest Weight
We call an h-module M a weight module if it is semisimple. As the simple h-modules

are one-dimensional (and given by a character 2 h? ), semisimplicity means that
M decomposes into a direct sum of its weight spaces, i.e.
M
MD M ;
2h?
where for each 2 h? we define M D fm 2 M j H:m D .H/m 8H 2 hg. Then

is called a weight of M if M f0g and we call each element of M a weight
vector of weight . Let 2 h? .
Definition 3.2.1 A g-module M is called a module of highest weight if the
following holds:
1. M is a weight module,
2. all weights of M are smaller or equal to , i.e. M f0g implies 6 ,
3. M is generated by a non-zero vector of weight .
(One can show that (3) implies (1).) Here is the classification theorem for the
irreducible highest weight modules:
Theorem 3.2.2 (Sect. 1.3 [13]) Let 2 h? . There exists an up to isomorphism
unique irreducible module L./ of g of highest weight . Moreover, L./ L./
implies D .P
Set WD 12 2RC 2 h? .
Definition 3.2.3 An element of h? is called
1. integral, if h; _ i 2 Z for all 2 R,
2. dominant, if h C ; _ i 6< 0 for all 2 RC ,
3. antidominant, if h C ; _ i 6> 0 for all 2 RC ,
4. regular, if h C ; _ i 0 for all 2 RC .
3.2.3 The Character
How can we describe the structure of L./? A precise description could be, at least
in the case that L./ is finite dimensional, to explicitly write down the representing
matrices for generators of g in terms of a basis of L./. This might not be very
instructional, though. A first interesting piece of information is the dimension of
L./. Here one has the following result:
Theorem 3.2.4 ([13, Sect. 1.6]) The g-module L./ is finite dimensional if and only
if is integral and dominant.
A better invariant than the dimension is the character of L./. We will define the
character for the class of modules M of g that satisfy the following assumptions:
M is a weight module.
There exists 1 ,. . . ,n 2 h? such that for any weight of M there is an i 2
f1; : : : ; ng with 6 i .
Obviously each module with highest weight satisfies these assumptions, and so does
each finite direct sum of highest weight modules.
Denote by Zh? the group algebra of the abelian group h? . Note that it is the free
abelian group with basis fe g2h? and multiplication determined by e e D eC .
Then Zh? can be viewed as the space of maps h? ! Z with finite support, in
such a way that the e are the -functions at the point . Now we can define the
1
subspace Zh? of Map.h? ; Z/ of maps with locally bounded support. It contains
all f 2 Map.h? ; Z/ with the property that there exist 1 ; : : : ; n 2 h? , depending
on f , such that f ./ 0 implies that there is i 2 f1; : : : ; ng with 6 i . It
is not difficult to see that the multiplication on Zh? has a smooth extension to a
1
multiplication on Zh? .
1
Definition 3.2.5 Suppose that M satisfies the two assumptions above. Denote by
ch M the element in Zh? that is given by 7! dimC M .
Here is the main problem that we are concerned with:
Character Problem Calculate the character of L./.
This is a difficult problem.
3.2.4 The Character Formula of Weyl
In the integral dominant case (i.e., for finite dimensional L./), the characters have
been known for some time now. In order to state the theorem, we introduce the
Weyl group of g. For each 2 R there is a unique element _ 2 g ; g , called
the coroot of , with the property . _ / D 2. The homomorphism s W h? ! h? ,
7! h; _ i is called the reflection associated with . Note that it is indeed
a reflection, i.e. s2 D idh? and the fixed points of s form a complex hyperplane.
The group W of linear automorphisms of h? that is generated by the s with 2 R
is called the Weyl group of g. Let S W be the set of simple reflections, i.e. the
set of s , where 2 RC is a simple root (i.e., cannot be written as the sum of
at least two positive roots). It turns out that W is already generated by the simple
reflections, and .W ; S / is a Coxeter system (cf. [12]). Hence it comes with a length
function lW W ! N and a Bruhat order 6 on W .
Theorem 3.2.6 ([13, Sect. 2.4]) Let be an integral and dominant element
in h? .
46 P. Fiebig
P P
The element w2W .1/l.w/ ew.C/ is divisible by w2W .1/l.w/ ew./ in Zh? .
As Zh? contains no zero-divisors, the element
P
w2W .1/
l.w/ w.C/
e
./ D P
w2W .1/
l.w/ ew./
is hence well defined.

It holds that ch L./ D ./.
But what about more general ?
3.2.5 The Hecke Algebra and KazhdanLusztig Polynomials
Before KazhdanLusztigs generalization of Weyls character formula can be stated,

we have to introduce another player, the Hecke algebra associated with g. Let L D
Zv; v 1 be the ring of integral Laurent polynomials in the variable v.
Definition 3.2.7 The Hecke algebra associated with .W ; S / is the L -algebra with
generators fTw gw2W and relations
Tx Ty D Txy for x; y 2 W with l.xy/ D l.x/ C l.y/;

Ts2 D .v 2 1/Ts C v 2 Te for s 2 S :
It turns out that fTw gw2W is an L -basis of H and Te is the multiplicative unit in
H . Moreover, each Tx is invertible in H . D. Kazhdan and G. Lusztig defined in
[16] a Z-linear involution on H as follows: v D v 1 and T x D Tx1 1 and they
proved the following result:
Theorem 3.2.8 For each w 2 W there exists a unique element Cw0 2 H with the
following properties:
1. Cw0 is self-dual, 0
P i.e. Cw D Cw .
0
0 l.w/
2. Cw 2 v x6w Px;w Tx , where Px;w is a polynomial in v of degree 6 l.w/
l.x/ 1.
The polynomials Px;w are the famous KazhdanLusztig polynomials that are
ubiquitous in the representation theory and geometry of Lie algebras and Lie groups.
3.2.6 A Generalization of Weyls Character Formula:

The Conjecture of Kazhdan and Lusztig
For all w 2 W set Lw WD L.w./ / and w WD .w./ /. The dot-action

of W on h? is defined by setting w: D w. C / . Then w./ D w:.2/.
As w0 :0 D 2, where w0 2 W is the longest element, it holds that Lw D L.ww0 :0/

and w D .ww0 :0/.
In [16] one finds the following conjecture:
Conjecture 3.2.9 For w 2 W the character of Lw is given by
X
ch Lw D .1/l.y/Cl.w/Py;w .1/ ch y :
y6w
This astounding conjecture is most influential and is the starting point of what
sometimes is called KazhdanLusztig theory. From the theory of translation functors
(cf. Sect. 3.3.4) one can deduce that the above formula is equivalent to the analogous
formula where the weight 0 is replaced by an arbitrary integral, dominant, and
regular . There are similar formulas for dominant and regular, but not necessarily
integral weights . For those on should replace W by the so-called integral Weyl
group W of (cf. Definition 3.3.3).
Conjecture 3.2.9 has been proved by Beilinson and Bernstein in [2] and,
independently and in the same year, by Brylinski and Kashiwara in [4] using
the theory of D-modules. In this paper we sketch a different proof of the above
conjecture that is originally due to Soergel. The first part of this proof is to translate
the conjecture into a conjecture involving Soergel bimodules (or, equivalently,
moment graph sheaves). Already this translation is quite a journey! We start with
rephrasing the character conjecture of Kazhdan and Lusztig as a conjecture on
JordanHlder multiplicities of Verma modules.
3.2.7 The Multiplicity Version of the Character Problem
For each 2 h? there is a universal module with highest weight. It is defined

as follows. Let U.g/ and U.b/ be the universal enveloping algebras of g and b,
respectively. Let C be the one-dimensional module of b with weight , and obtain
by induction (cf. [13, Sect. 1.3]) the Verma module
./ WD U.g/ U.b/ C :

L
The subalgebra n WD 2RC g of g is nilpotent, and the PoincarBirkhoff
Witt Theorem implies that U.g/ D U.n / C U.b/ (as a left U.n /and right
U.b/-module). Hence ./ is free over U.n / of rank 1. Another application of the
PoincarBirkhoffWitt Theorem yields the following character formula:
Proposition 3.2.10 ([13, Sect. 1.16]) The character of a Verma module with high-
est weight 2 h? is
Y
ch ./ D e .1 C e C e2 C /:
2RC
48 P. Fiebig
An easy argument shows that the Character Problem of Sect. 3.2.3 is equivalent
to the following:
Multiplicity Problem I Calculate the JordanHlder multiplicities ./ W L./
for all ; 2 h? .
The inversion formula in [16] implies that Conjecture 3.2.9 above is equivalent
to the following statement (cf. [13, Sect. 8.4]):
Conjecture 3.2.11 For 2 b h? all subquotients of w are of the form Lx for some
x 2 W , and the multiplicities are
w W Lx D Pw0 w;w0 x .1/:
3.3 Category O
As a next step we introduce an abelian category that contains all the objects that
were relevant for us up to this point.
Definition 3.3.1 Denote by O the full subcategory of the category of all g-modules
that contains all objects that are semisimple for the action of h and locally finite for
the action of b.
Recall that semisimplicity for the action of h is the same as being a weight
module. A g-module M is locally finite for the action of b if every element of
M is contained in a sub-b-module of M of finite dimension. Sometimes one also
requires an object of O to be finitely generated as a g-module. For us, this is
not necessary and it is not a good assumption in the KacMoody case, because
the universal enveloping algebra of a KacMoody algebra is not Noetherian. All
highest weight modules (in particular, the simple highest weight modules L./ and
the Verma modules ./) belong to category O.
3.3.1 The Block Decomposition of O
We now determine the decomposition of category O into indecomposable block

and the corresponding linkage principle which determines the set of pairs .; /
with ./ W L./ 0. We start with an equivalence relation on the set h? .
Definition 3.3.2 Let
be the equivalence relation on h? that is generated by

if there is 2 RC and n 2 Z such that h C ; _ i D n and D n.
We can describe the equivalence classes much more explicitly using the action
of the Weyl group on h? . Let 2 h? .
Definition 3.3.3 Define
R WD f 2 R j h; _ i 2 Zg;

T WD fs j 2 R g
W WD ht j t 2 T i:
The subgroup W of W is called the integral Weyl group with respect to .
It is now not difficult to prove the following.
Lemma 3.3.4 Let ; 2 h? .
1. If
, then R D R and hence T D T and W D W .
2. We have
if and only if 2 W :.
Hence the equivalence classes are orbits in h? under certain subgroups of the
Weyl group. We now discuss the block decomposition of O. For any equivalence
class with respect to
let O be the full subcategory of O that contains all
objects M such that M W L./ 0 implies 2 . Note that even though not all
objects in M have a finite JordanHlder series, the statement M W L./ 0 makes
sense and means that there is a submodule N of M that has L./ as a quotient.
Theorem 3.3.5 The functor
Y
O ! O;
2h? =
M
fM g 7! M ;
2h? =
is an equivalence of categories.
Theorem 3.3.5 is a special case of the more general Theorem 3.4.21 that we will
prove using deformation theory.
3.3.2 Objects Admitting a Verma Flag
Here is an important concept for category O.

Definition 3.3.6 Let M be an object in O. We say that M admits a Verma flag if
there exists a finite filtration
0 D M0 M1 Mn D M
50 P. Fiebig
such that Mi =Mi1 is isomorphic to a Verma module for i D 1; : : : n.

It is not difficult to show that if M admits a Verma flag, the corresponding
multiplicities
.M W .// D #fi 2 f1; : : : ; ng j Mi =Mi1 ./g
are independent of the chosen filtration.
3.3.3 Projectives in O
The advantage of the categorical framework that O provides is that we can rephrase
the multiplicity problem of Sect. 3.2.7 in terms of projective objects in O. In the
following we give an account of the theory of projectives in O that is different from
the one presented in [13], but has the advantage that it can be generalized more
easily to other situations such as the KacMoody case.
Definition 3.3.7 A subset J of h? is called open if 2 J and 2 h? with
6 implies 2 J . An open subset J of h? is called locally bounded if for all
2 J the set f 2 J j 6 g is finite.
It is easy to show that this defines indeed a topology on the set h? . It has the
additional property that arbitrary intersections of open subsets are open again. For
any open subset J h? denote by O J the full subcategory of O that contains all
objects M with the property that M 0 implies 2 J . Note that ./ and L./
are contained in O J if and only if 2 J .
Theorem 3.3.8 Suppose that J h? is open and locally bounded and let 2 J .
Then there exists a projective cover PJ ./ ! L./ of L./ in O J . It has the
1. PJ ./ admits a Verma flag.
2. It holds that .PJ ./ W .// D ./ W L./ (BGG-reciprocity) for all 2
J.
Recall that we call an object P a projective cover of a simple object L in an abelian
category A if P is projective and if there is an epimorphism f W P ! L in A that
has the property that for any morphism gW M ! P in A with the property that
f gW M ! L is an epimorphism, g is an epimorphism.
We will prove the above theorem in Sect. 3.4 in a more general situation, i.e. in
the context of the deformed category O. Using the BGG-reciprocity result we can
rephrase the Multiplicity Problem I as follows:
Multiplicity Problem II Calculate .PJ ./ W .// for all locally bounded open
subsets J and all ; 2 J .
3.3.4 Translation Functors
This section introduces Jantzens translation functors. To simplify matters, assume

that is regular and integral, i.e. W D W and StabW D feg. Suppose also that
is anti-dominant. Set D .
Theorem 3.3.9 ([14]) Let s 2 S . There exists a functor Ts W O ! O with the
1. Ts is exact and self-adjoint.
2. For w 2 W it holds that Ts .w:/ Ts .ws:/, and if ws > w, then there is a
short exact sequence
0 ! .ws:/ ! Ts .w:/ ! .w:/ ! 0:
Lemma 3.3.10
1. If M admits a Verma flag, then so does Ts M and for the multiplicities holds
.Ts M W .w:// D .M W .w:// C .M W .ws://:
2. If P is projective in O , then Ts P is also projective in O .

Proof Theorem 3.3.9 implies that Ts maps each Verma module into a module that
admits a Verma flag. As Ts is also exact, it follows that Ts preserves the subcategory
in O of modules that admit a Verma flag. The formula in (1) is another consequence
of the exactness of Ts and the short exact sequence in Theorem 3.3.9, (2).
Note that there is a natural isomorphism
HomO .Ts P; / HomO .P; Ts .//:
for any object P in O . If P is projective, then HomO .P; / is an exact functor, hence
Hom.Ts P; / is isomorphic to the composition of the exact functors HomO .P; / and
Ts , i.e. Ts P is projective as well.
3.3.5 The Decomposition Problem I
Let s D .s1 ; : : : ; sn / be a finite sequence of simple reflections. Let 0 2 be

dominant, integral, and regular. Then .0 / is projective in O . As the translation
functors preserve projectivity, the object BS.s/ WD Ts1 Tsn .0 / is projective
in O . Hence there are numbers aws 2 N such that
M L
aw
BS.s/ P.w:0 / s :
w2W
52 P. Fiebig
The theory of translation functors can be used to prove that the numbers aws do not
depend on the choice dominant integral regular element .
Decomposition Problem I Determine all numbers aws for all finite sequences s and
all w 2 W .
Our next problem is to translate Conjecture 3.2.11 into a conjecture about the
numbers as .w/. Recall from Theorem 3.2.8 the self-dual basis fCw0 gw2W of the
Hecke algebra H . For any sequence s D .s1 ; : : : ; sn / we consider the element
Cs0 n Cs0 1 . There are unique elements bws 2 Zv; v 1 such that
X
Cs0 n Cs0 1 D bws Cw0 :
w2W
It is now not difficult to show that Conjecture 3.2.11 is equivalent to the following
Decomposition conjecture:
Conjecture 3.3.11 For each finite sequence s in S and all w 2 W we have aws D
bws .1/.
3.4 Deforming the Category O
In this section the proofs of Theorems 3.3.5 and 3.3.8 are given. In contrast to the
proofs in [13], a deformation theory is used. Let S D S.h/ D U.h/ be the symmetric
algebra of the vector space h.
Definition 3.4.1 A deformation algebra is a commutative, local, Noetherian, and
unital S-algebra.
Remark 3.4.2
1. If A is a deformation algebra with (unique) maximal ideal m, then the residue
field A=m (with the induced S-algebra structure) is a deformation algebra as well.
2. Let eS be the localization of S at the maximal ideal generated by h. In geometric
terms, S is the C-algebra of regular functions on the complex affine space h? , and
e
S is the localization at the closed point 0 2 h? . Then e
S is a deformation algebra.
Its residue field is C with the S-algebra structure that is given by evaluating a
local function at the closed point 0.
3. If p e S is any prime ideal, then the localization e
Sp of eS at p is a deformation
algebra as well. In this way we obtain, for example, the quotient field Quot S D
e
S.0/ of e
S.
For any deformation algebra A we will now introduce an A-linear version of
category O. A more detailed account of the deformed category O can be found in
[7]. Let l be a complex Lie algebra. Let M be a l-A-bimodule. Note that this is an A-
module that is endowed with a homomorphism W l ! EndA .M/ of C-Lie algebras.
Denote by W h ! A the structural map, i.e. the composition h S ! A, where the

last homomorphism sends f to f 1A .
Definition 3.4.3
Let M be an h-A-bimodule and 2 h? . Define
M WD fm 2 M j H:m D . C /.H/m 8H 2 hg:
Here we consider . C /.H/ D .H/ 1A C .H/ as an Lelement in A.

An h-A-bimodule M is called a weight module if M D 2h? M .
A b-A-bimodule M is called locally finite if for any m 2 M the set U.b/m M
generates a finitely generated A-module.
The category OA is the full subcategory of the category of g-A-bimodules that
contains all M that are weight modules for the action of h and that are locally
finite for the action of b.
Remark 3.4.4
1. As A is supposed to be Noetherian one can show that OA is an abelian category.
2. Note that for A D C with the S-algebra structure that comes from evaluating
at 0 2 h? , we have D 0 and OC coincides with the classical category O of
Definition 3.3.1.
3. More generally, if A D K is a field, the deformed category OK is a full
subcategory of the usual category O over the Lie algebra gK D g C K. It
contains all objects M with weights contained in the subset C h? of h?K D
HomK .hK ; K/.
4. Let f W A ! B be a homomorphism of deformation algebras. Then the change of
scalars functor A A0 induces a functor OA ! OA0 .
3.4.1 Deformed Verma Modules and Verma Flags
Let 2 h? . Let A be the b-A-bimodule that is free of rank 1 as an A-module and

C
on which b acts via the character b ! h ! A.
Definition 3.4.5 The deformed Verma module with highest weight is defined as
A ./ WD U.g/ U.b/ A :
It is not difficult to check that A ./ is contained in OA . For a homomorphism

f W A ! A0 of deformation algebras there is an isomorphism A ./ A A0 A0 ./.
The definition of objects that admit a Verma flag in OA is the same as
Definition 3.3.6 in the case of O. Again, the multiplicity #fi 2 f1; : : : ; ng j i D g
of a subquotient isomorphic to A ./ is independent of the particular filtration, and
54 P. Fiebig
is denoted by .M W A .//. Define
supp M WD f 2 h? j .M W A .// 0g;
the support of the Verma flag object M.

For 2 h? let us denote by v D 1 1 2 A ./ a generator of the highest
weight space. The following statement is an immediate consequence of the universal
property of induction and the definition of the b-A-bimodule A :
Lemma 3.4.6 For all 2 h? and all objects M 2 OA the A-linear homomorphism
HomO J .A ./; M/ ! fm 2 M j nC m D 0g;

A
f 7! f .v /
is an isomorphism. L
Here, by nC we denote the nilpotent subalgebra 2RC g .
Remark 3.4.7 One can use the above universal property to reorder a Verma flag
0 D M0 M1 Mn D M with MiC1 =Mi A ./ in such a way that
i j implies i 6 j.
3.4.2 Simple Objects in OA
We can also classify the simple objects in OA :

Proposition 3.4.8 ([9, Proposition 2.1]) Let A be a local deformation algebra with
residue field K. Then the functor A KW OA ! OK induces a bijection on simple
isomorphism classes.
As observed above, OK is just a direct summand of the usual category O over
the Lie algebra gK that contains all objects with weights of the form C . Hence
the simple objects in OK are the irreducible highest weight modules corresponding
to these weights. Denote the one corresponding to the highest weight C by
LK ./. Then there is an up to isomorphism unique simple object LA ./ in OA with
LA ./ A K LK ./.
Our next objective is to study the projective objects in OA .
3.4.3 Projectives in OA
Let A be a deformation algebra. Let J be an open and locally bounded subset of

h? and 2 J . Set J 0 D J D f j 2 J g. Then J 0 is an open
and locally bounded subset of h? containing 0. Denote by B D U.b/ the universal
Lalgebra. Note that it contains S D U.h/ and B is a weight module,

enveloping
i.e. B D 2h? B for the adjoint action of h. Moreover,
L B 0 implies (and is
equivalent to) 0 and it holds that B0 D S. So B0 WD 2h? nJ 0 B is a two-sided
0
ideal in B (as J 0 is open). Define BJ D B=B0 and consider this as a B-module
0
from the left, and as an S-module from the right. Then BJ S A is a B-module
(B acting on the first factor) and an A-module (here A acts on the right factor). Both
actions clearly commute, i.e. we have constructed a B-A-bimodule. Now induce:
J 0
QA ./ WD U.g/ B .BJ S A /:
This is a U.g/-A-bimodule.
Proposition 3.4.9 ([9, Lemma 2.3])
J J
1. QA ./ is an object in OA and it admits a Verma flag with multiplicities
J
.QA ./ W A .// D dimC B =B0
for all 2 J .
J
2. For any M 2 OA there is an isomorphism

HomOA .QA ./; M/
! M ;
J J
functorial in M. In particular, QA ./ is a projective object in OA .
J J
By the above proposition, .QA ./ W A .// D dimC B0 =B0 D 1 and .QA ./ W
A .// 0 implies . From Remark 3.4.7 we deduce that A ./, and hence
J
LA ./, is as a quotient of QA ./.
Theorem 3.4.10 Let J be an open and locally bounded subset of h? . For any
J J
2 J there exists a projective cover PA ./ ! LA ./ in OA . Moreover, LA ./ is
J
the only simple quotient of PA ./.
Proof Let us introduce a convenient notation: write J for f 2 J j
g. We prove the theorem by induction on the number of elements in the finite
set J . If this set contains only the element , then is maximal in J and
J
the universal property in Lemma 3.4.6 implies that PA ./ WD A ./ is projective
J J
in OA . Clearly it admits a surjection f W A ./ ! LA ./. If gW M ! PA ./ is
such that f gW M ! LA ./ is surjective, then im g contains the -weight space of
A ./ (by Nakayamas lemma). As this weight space generates A ./ we obtain
im g D A ./, so g is surjective. Hence A ./ ! LA ./ is a projective cover. As
A ./ is a highest weight module of highest weight , LA ./ is the only simple
quotient.
Now suppose that the statement for all 2 J with jJ j < jJ j is shown.
Recall from the preceding paragraph that A ./ and hence LA ./ are quotients of
56 P. Fiebig
J
QA ./. As the highest weight space of A ./ is free of rank 1 over the local ring
J
A, there must be an indecomposable direct summand of QA ./ that has A ./
as a quotient. Let us denote one of those by P. We claim that any surjective
homomorphism f W P ! LA ./ is in fact a projective cover. First we show that LA ./
is the unique simple quotient of P. So suppose that there is a surjection f W P ! LA ./
and . As P admits a Verma flag (as a direct summand of an object admitting
a Verma flag) with subquotients with highest weights in J , we deduce > .
J
In particular, by induction there already is a projective cover f 0 W PA ./ ! LA ./ in
J J
OA . The projectivity of P implies that there is a homomorphism hW P ! PA ./
such that the diagram
h
P PA
f f
LA
J
commutes. The projective cover property of PA ./ then implies that h is surjective,
and hence h must split. As P is indecomposable, we obtain P PJ ./ which is
J
impossible, as it is already shown that PA ./ has no quotient isomorphic to LA ./.
So LA ./ is the only simple quotient of P.
It remains to show that f W P ! LA ./ is a projective cover. So let gW M ! P be
J
any homomorphism in OA such that f g is surjective. If g is not surjective, then
P=im g must have a simple quotient (since it has a finite filtration by highest weight
modules). But this simple quotient cannot be isomorphic to LA ./ (as .P W A .// D
1 and .P W A .// 0 implies ). But this is a contradiction to what is
shown in the preceding paragraph. Hence g must be surjective, so f W P ! LA ./ is
a projective cover.
Remarks 1
1. The proof of the Theorem above shows that any indecomposable projective
J J
object in OA is a direct summand of QA ./ for some .
J
2. In particular, if P is an indecomposable projective object in OA , then
J
HomOA .P; M/ is a direct summand of HomOA .QA ./; M/ for some 2 J .
Now the latter space can be identified with M . So if M is free as an A-module
of finite rank, then HomOA .P; M/ is free of finite rank as an A-module as well.
Proposition 3.4.11 Let A be a local deformation algebra with residue field K. Let
J h? be open and locally bounded and let be an element in J .
J J J
1. There is an isomorphism PA ./ A K PK ./ in OK .
J
2. If P is projective in OA and A ! A0 is a homomorphism of deformation
J
algebras, then for any M 2 OA the natural homomorphism
HomO J .P; M/ A A0 HomO J .P A A0 ; M A A0 /

A K
is an isomorphism.
J
Proof We prove part (2) first. Note that for P D QA ./ the statement follows from
J
the fact that QA ./ represents the functor M 7! M , and that there is a canonical
identification .MA A0 / D .M /A A0 . As any indecomposable projective object in
J J
OA is a direct summand of some QA ./, this also implies that the functor A A0
preserves projectivity. In order to prove part (1) we check that LK ./ is the only
J
irreducible quotient of PA ./ A K. Indeed, as there is a surjective homomorphism
J J J
PA ./ ! PA ./ A K (which can be considered in the category OA !), every
J J
irreducible quotient of PA ./ A A0 is an irreducible quotient of PA ./ as well.
3.4.4 The BGG-Reciprocity
The following result is a cornerstone of our approach towards the character problem.
Let A be a local deformation algebra with residue field K.
Theorem 3.4.12 (BGG-Reciprocity) Let J be an open and locally bounded
J
subset of h? . Then for any 2 J the object PA ./ admits a Verma flag and
for the multiplicities holds
(
J 0; if 62 J ;
.PA ./ W A .// D
K ./ W LK ./; if 2 J :
In order to prove the above result, we need some preparation. First, we introduce
the dual Verma modules rA ./. There is a duality functor ./_ W OA ! OA that
is defined as follows. For any A-module M consider the dual A-module M D
HomA .M; A/. If M is a g-A-bimodule, then so is M , where the g-action is given
by .x:f /.m/ D f .x:m/ for m 2 M, f 2 M and x 2 g. Note that L the minus sign is
needed to obtain a left action of g on M . It turns out that M ~ WD 2h? .M / is a
g-A-submodule of M , and it is a weight module with weight spaces
.M ~ / D .M / :
Now this is not an object in OA in general, as b will not act locally finitely in all
cases. But if we finally twist the g-action using the Chevalley involution !W g ! g
58 P. Fiebig
that interchanges the root spaces g and g and acts as multiplication by 1 on h,

we obtain the dual object M _ in OA .
Remark 3.4.13 Note that the duality ./_ is involutive only on the subcategory of
OA that contains the objects M which have the property that each weight space M
is reflexive, the latter considered as an A-module (i.e., the natural homomorphism
M ! ..M / / is an isomorphism).
Set rA ./ WD A ./_ . As every weight space of A ./ is a free A-module of
finite rank, it follows that A ./ rA ./_ .
J
Lemma 3.4.14 Suppose that M is an object in OA that admits a Verma flag. Then
Ext1 J .M; rA .// D 0 for all 2 J .
OA
Proof We prove the statement by induction on the length of a Verma flag of M.

Suppose that M A ./ and let
f
0 ! rA ./ ! N ! A ./ ! 0 ()
be a short exact sequence. If 6> , then is a maximal weight of N and the

universal property of A ./ yields a non-zero homomorphism gW A ./ ! N. Then
f g is an endomorphism of A ./ that is an automorphism on the highest weight
space. So f g is an automorphism. This means that the short exact sequence ()
splits. On the other hand, if 6> , then we can consider the dual sequence
0 ! A ./_ ! N _ ! rA ./_ ! 0: ./_
This is an exact sequence as well, as all occurring objects are free as A-modules. As
rA ./_ A ./ we can argue as before and see that ./_ , hence also () splits.
Hence Ext1 J .A ./; rA .// D 0 in any case.
OA
If the length of a Verma flag of M is > 1, then we can choose M 0 M such
that both M 0 and M=M 0 are non-zero and admit a Verma flag (necessarily of smaller
length). The short exact sequence
0 ! M 0 ! M ! M=M 0 ! 0
yields a long exact sequence of Exti J .; rA .//-groups. The i D 1 part of this
OA
sequence reads
Ext1 J .M=M 0 ; rA .// ! Ext1 J .M; rA .// ! Ext1 J .M 0 ; rA .//:

OA OA OA
As the groups on the left and on the right are trivial by induction assumption, we
deduce Ext1 J .M; rA .// D 0.
OA
J
Lemma 3.4.15 Suppose that M is an object in OA that admits a Verma flag. Then
HomO J .M; rA .// is a free A-module for all 2 J and
A
.M W A .// D rkA HomO J .M; rA .//:

A
Proof Again we prove this by induction on the length of a Verma flag of M. If

M A ./, then HomO J .M; rA .// D 0 unless D (by the universal property
A
of A ./ and its dual statement). For D we have that HomO J .A ./; rA .//
A
is isomorphic to rA ./ , i.e. it is free of rank 1. Hence the claim for M D
A ./ is settled. Now, if the length of M is > 1, then consider the short exact
sequence 0 ! M 0 ! M ! M=M 0 ! 0 as in the proof of Lemma 3.4.14. As
Ext1 J .M=M 0 ; rA .// D 0 we obtain a short exact sequence
OA
0 ! HomO J .M=M 0 ; rA .// ! HomO J .M; rA .//

A A
! HomO J .M 0 ; rA .// ! 0:
A
By the induction hypothesis the end terms are free A-modules, hence so is the middle
term. Observing the additivity of the rank with respect to short exact sequences and
the fact that
.M W A .// D .M 0 W A .// C .M=M 0 W A .//
and using the induction hypothesis on M 0 and M=M 0 yield the claim.
Now we can prove the BGG-reciprocity result.
J J J
Proof As PA ./ is isomorphic to a direct summand of QA ./ and as QA ./
J
admits a Verma flag, so does PA ./. Then
J J
.PA ./ W A .// D rkA HomO J .PA ./; rA .//
A
D rK ./ W LK ./
D K ./ W LK ./:
Here the first equality results from Lemma 3.4.15, the second one from using the
J
projective cover property of PA ./, and the last identity is a consequence of the
fact that the characters of K ./ and rK ./ coincide.
3.4.5 Another Result of BernsteinGelfandGelfand
In this section suppose that A D K is a field.

60 P. Fiebig
Definition 3.4.16 Denote by "K the partial order on the set h? that is generated by
"K if there is 2 RC such that h C C ; _ i 2 Z, D s : and < .
Then the following holds (cf. [13, Sect. 5.1]):
Theorem 3.4.17 For ; 2 h? we have K ./ W LK ./ 0 if and only if
"K .
The if part was already proven by Verma.
Definition 3.4.18 Let
A be the equivalence on h? that is generated by
A if
K ./ W LK ./ 0.
By definition,
A only depends on the residue field K (more precisely, on the
residue field K as an S-algebra). From Theorem 3.4.17 it follows that
A is the
equivalence relation generated by the partial order "K . Using this, we can describe
the equivalence classes in more detail.
Definition 3.4.19 For 2 h? define
RA; WD f 2 R j h C C ; _ i 2 Zg;
TA; WD fs j 2 RA; g
WA; WD ht j t 2 TA; i:
Then the following holds:

Lemma 3.4.20
1. If
A , then RA; D RA; and hence TA; D TA; and WA; D WA; .
2. For an equivalence class 2 h? =
A we have D WA; : for all 2 .
In particular, A ./ W LA ./ 0 implies 2 W and 6 .
3.4.6 The Block Decomposition of OA
Let A be a local deformation algebra with residue field K. Using the BGG-
J
reciprocity we could replace the condition in the definition of
A by .PA ./ W
A .// 0 for some open and locally bounded subset J that contains and .
J
In particular, it also follows that
A whenever PA ./ W LA ./ 0.
Let 2 h? =
A be an equivalence class and M an object in OA . Denote by M
the submodule of M that is generated by the images im f for all homomorphisms
J
f W PA ./ ! M with 2 and all open and locally bounded subsets J of h? with
2 J . We also denote by OA; the full subcategory of OA that contains all objects
N with N W LA ./ 0 only if 2 .
Theorem 3.4.21
is contained in O .
1. The object ML
2. We have M D M .
Remark 3.4.22 We can also formulate the above result in the following way: The
functor
M
OA; ! OA ;
2h? =A
M
fM g 7! M

is an equivalence of categories.
J
Proof Statement (1) follows from the fact that for 2 we have that PA ./ W
LA ./ 0 for some J implies 2 . Clearly M is the sum of its submodules
M . But this must be a direct sum, as there are no non-zero homomorphisms
J
PA ./ ! N for 2 and N 2 O0 if 0 .
3.4.7 On the Structure of Some Projectives
The following result is one of the reasons why we deform the category O. Recall
the localization e
S of the symmetric algebra S (cf. Remark 3.4.2).
Proposition 3.4.23 Let p e
S be a prime ideal and A D e
Sp .
1. If . _ / 62 p for all 2 R, then each equivalence class with respect to
A is
a singleton, i.e. D fg for some 2 h? .
2. If p D . _ /e S for some 2 R, then an equivalence class with respect to
A
is either a singleton, or contains exactly two elements, i.e. D f; g for some
; 2 h? . In the latter case, D s :.
Proof For A D e Sp with residue field K one checks that the condition h C C
; _ i 2 Z K is equivalent to . _ / D 0, i.e. . _ / 2 p and h C ; _ i 2 Z
C.
In a next step we would like to understand the equivalence relation
A . We start
with a definition.
Definition 3.4.24 Let 2 h? =
A be an equivalence class.
1. is called generic, if is a singleton, i.e. D fg for some 2 h? .
2. is called subgeneric, if contains exactly two elements.
62 P. Fiebig
Remark 3.4.25 Suppose that D f; g is subgeneric. One can then show (for
example, using the Jantzen filtration, cf. Sect. 5.3 in [13]) that there is a unique
2 R such that h C C ; _ i 2 Z A and D s :.
It is now easy to describe the structure of blocks corresponding to a generic
equivalence class.
Proposition 3.4.26 Suppose that D fg is generic. Then PA ./ A ./ and
K ./ D LK ./. In particular, EndOA .PA .// D A idPA ./ .
Proof As K ./ W LK ./ D 1, the second statement follows readily from the
definition of a generic equivalence class. The first statement then is a consequence
of BGG-reciprocity.
Next we consider the structure of subgeneric blocks.
Proposition 3.4.27 Suppose that D f; g is subgeneric, and suppose that D
s : > . Then PA ./ A ./ and K ./ D LK ./. There are short exact
sequences
0 ! A ./ ! PA ./ ! A ./ ! 0
and
0 ! LK ./ ! K ./ ! LK ./ ! 0:
Proof Note that K ./ W LK ./ D 0 as > . We can argue as in the generic case
and deduce K ./ D LK ./. So K ./ is simple. Clearly K ./ W LK ./ 1.
But the maximal submodule of K ./ must be generated by primitive vectors of
weight . As dimK Hom.K ./; K .// 6 1, this yields K ./ W LK ./ 6 1,
hence K ./ W LK ./ D 1. Now the rest of the statements follow from BGG-
reciprocity.
Remark 3.4.28 In the generic and the subgeneric situations we deduce that the
J J J0
projectives PA ./ stabilize with respect to J , i.e. PA ./ is isomorphic to PA ./
as long as J and J 0 contain the generic or subgeneric equivalence class of .
Hence LA ./ admits a projective cover already in OA . Denote it by PA ./.
3.4.8 Endomorphism Rings in the Subgeneric Situations
We now study the categorical structure of subgeneric blocks in detail. Let D

f; g and suppose that < . First assume that A D K is a field. Then K ./ W
LK ./ D K ./ W LK ./ D 1 and K ./ D LK ./. Moreover, PK ./ K ./
and there is a short exact sequence
0 ! K ./ ! PK ./ ! K ./ ! 0:

For any projective cover P of a simple object L in OK it is easy to show that
dimK HomOK .P; M/ D M W L
for all M 2 OK . Then
dimK HomOK .PK ./; PK .// D dimK HomOK .PK ./; PK .//
D dimK EndOK .PK .// D 1
and
dimK EndOK .PK .// D 2:
Now choose generators
jW PK ./ ! PK ./;
iW PK ./ ! PK ./
of the respective Hom-spaces. Then j is injective, as PK ./ is a Verma module

and PK ./ is free over U.n /. Moreover, i j D 0, as otherwise it would be an
automorphism of PK ./, hence i would split. Then j iW PK ./ ! PK ./ is non-
zero, but nilpotent. Hence fidPK ./ ; j ig is a basis of EndOK .PK .//. It is now not
difficult to check that EndOK .PK ./ PK .// is isomorphic to the path algebra of
the quiver
with relation i j D 0.
Now let A be an arbitrary deformation algebra with residue field K.
Proposition 3.4.29 Let D f; g as above. Then EndOA .PA ./ PA .// is
isomorphic to the quiver depicted above with relation i j D . _ /.
Proof Recall that there is a positive root such that 2 Z. From Remark 1
we know that the Hom-spaces between projectives in OA are free A-modules of
finite rank, and the base change result in Proposition 3.4.11 shows that we can lift
the homomorphisms i and j from above to homomorphisms QjW PA ./ ! PA ./ and
QiW PA ./ ! PA ./ in such a way that they are bases in the respective Hom-spaces.
Using the Jantzen sum formula one can show that Qj Qi equals . _ / idPA ./ up
to an invertible scalar (cf. [6, Proposition 3.4]). Hence EndOA .PA ./ PA .//
is isomorphic to the path algebra of the quiver depicted above, but now with the
relation i j D . _ /.
64 P. Fiebig
3.5 Soergels Theory
In his fundamental paper [19], Soergel developed a theory that eventually produced
an alternative proof of the KazhdanLusztig Conjecture 3.2.9. His proof is quite
different from both original proofs of this conjecture, the one given by Beilinson and
Bernstein in [2] and the other one given by Brylinski and Kashiwara in [4]. Indeed,
Soergels proofs amounts to a detailed study of the combinatorics of translation
functors. A short review of the main ideas will be included in the sequel.
3.5.1 Endomorphisms of Multiplicity Free Projectives
We start with a result on the endomorphism rings of certain projective objects. Let
A be deformation algebra with residue field K and let be a
A -equivalence class
in h? . Let K be a W -set, i.e. a set with a homomorphism W ! Aut.K/. Define
the following commutative algebra
( )
M
ZA .K/ WD .zx / 2 A j zx ztx 2 .t_ /A 8x 2 K; t 2 T :
x2K
Let M be an object in OA; that admits a Verma flag. It is called multiplicity free,
if .M W A .// 6 1 for all 2 h? . Now choose an enumeration D f1 ; : : : ; n g
of the elements in in such a way that i > j implies i < j. Remark 3.4.7 implies
that there is a Verma flag 0 D M0 M1 Mn D M in such a way that
Mi =Mi1 is isomorphic to A .i /. Define Mi WD Mi =Mi1 .
For the following result assume that A D e S. Recall that an eS-module M is
called Treflexive if M is torsion free as an e S-module and if the natural inclusion
M p2P Mp , where P is the set of prime ideals of height one in e S and the
e
inclusion is taken inside M.0/ D M eS Quot .S/, is a bijection. Now suppose that M
and N are objects in OeS that are reflexive as S-modules. Then
\
HomOeS .M; N/ D HomOeSp .Mp ; Np /;
p2P
where the intersection is taken inside HomOeS .M.0/ ; N.0/ /. In particular,

.0/
HomOeS .M; N/ is a reflexive eS-module. If P is projective in OeS; , then P admits
a Verma flag. So it is free as an e
S-module, in particular reflexive. The following
result generalizes the Endomorphismensatz of [19]:
Proposition 3.5.1 Suppose that A D e S and that P is projective in OeS; and

multiplicity free. Then the natural homomorphism
M M
EndOeS .P/ ! EndOeS .P / D e
S
2supp .P/ 2supp .P/
is injective with image ZeS .supp P/. In particular, EndOeS .P/ is a commutative
algebra.
S
e
Proof As P is free over SS (in particular reflexive) it follows that P D p2P Pp
P.0/ and EndOeS .P/ D p2P EndOeSp .Pp / EndOeS .P.0/ /. By Proposition 3.4.23,
.0/
Pp splits into a direct sum of projectives in generic and subgeneric situations, and
so does EndOeSp .Pp /. But in these cases we determined the endomorphism rings
already explicitly in Propositions 3.4.26 and 3.4.29. The result follows.
3.5.2 Big Projectives
A main ingredient for Soergels approach is the study of the big projectives in the
category O. They correspond to weights that are anti-dominant. They are called
big because their support is maximal. The following is an important result and one
of the cornerstones of the theory.
Lemma 3.5.2 Suppose that 2 is anti-dominant. Then .PA ./ W A .// D 1
for all 2 . In particular, the object PA ./ is multiplicity free.
Proof Let s 2 S . Then it is well known that K .s:/ W LK ./ D 1. Using BGG-
reciprocity and the combinatorics of translation functors one shows that Ts PA ./
must contain at least two direct summands isomorphic to PA ./. It follows that
Ts PA ./ D PA ./ PA ./, which can only be the case if .PA ./ W A .// D 1 for
all 2 .
3.5.3 The Strukturfunktor and the Struktursatz
Here is Soergels definition:

Definition 3.5.3 Let 2 be antidominant. The functor
V D VA; WD HomOA; .PA ./; /W OA; ! mod-EndOA; .PA .//
is called the Strukturfunktor associated to A and .

As seen in Lemma 3.5.2, the object PA ./ is multiplicity free and supp PA ./ D
. In particular, EndOA; .PA .// D ZA ./ is commutative. So there is no
66 P. Fiebig
distinction between left and right ZA ./-modules, and the functor V is in fact a
functor from OA; to ZA ./-mod.
Assume now that A D e S is the localization of S at the maximal ideal generated
by h. We show that VeS; is fully faithful on projective objects. Note that V.P/ D
HomOeS .PeS ./; P/ is reflexive as a e
S-module.
Theorem 3.5.4 (Soergels Struktursatz) The functor V is fully faithful on
projectives, i.e. if P; P0 2 OeS; are projective, then HomOeS; .P; P0 / D
HomZeS .V.P/; V.P0 //.
Proof By what we observed above,
\
HomOeS; .P; P0 / D HomOeSp ; .Pp ; P0p /:
p2P
Likewise one shows that

\
HomZeS .VP; VP0 / D HomZeSp .VPp ; VP0p /:
p2P
Everything is compatible with localizations, hence we only have to check that the
natural homomorphism HomOeSp ; .Pp ; P0p / ! HomZeSp .VPp ; VP0p / is a bijection
for all prime ideals of height one. But these cases split into a direct sum of
subgeneric cases, and in the subgeneric case it is an exercise to check the statement
using the categorical structure in Sect. 3.4.8.
3.5.4 The Combinatorics of Translation Functors
As V is a fully faithful functor, the problem of decomposing objects like

Ts1 Tsn eS .0 / in OeS; is equivalent to decomposing the object V.Ts1 Tsn eS
.0 // in Z -mod. Fortunately, we can describe this Z -module quite explicitly.
Proposition 3.5.5 ([9, Theorem 9]) There is a natural isomorphism V Ts
Z Z s V of functors from OeS; to Z -mod.
Before the last problem can be stated we have to understand VeS .0 / as a Z -
module. It is free of rank 1 as an e S-module, and from the identification Z
EndOeS; .PeS .// it follows that an element .z / acts on VeS .0 / as multiplication
with z0 . Denote this Z -module by e S0 .
As the functor V is fully faithful on projectives, Proposition 3.5.5 implies that
the Decomposition Problem I is equivalent to the following:
Decomposition Problem II Decompose Z Z s Z Z t Z u e S0 in the category
of Z -modules for all sequences s; t; : : : ; u of simple reflections.
3.6 Soergel Bimodules, Parity Sheaves, and Moment Graph

Sheaves
Having started from the KazhdanLusztig Conjecture 3.2.9, through a successive

series of translations we moved first to the equivalent Decomposition Problem I and
from there, using Soergels theory, to another equivalent problem, Decomposition
Problem II. The latter problem does not refer to modules over a Lie algebra any
more, it is rather stated in the representation theory of a commutative, associative
algebra Z . The following gives an overview on three approaches to deal with it.
3.6.1 Moment Graph Sheaves
In [8] we studied sheaves on moment graphs that were introduced in [3]. A moment
graph G is a graph with a labeling of its edges by element in a lattice X. For any
field k one can define a commutative algebra Zk .G / over the symmetric algebra Sk
of the k-vector space X Z k. One then has the notion of a k-sheaf on this graph. The
space of global sections of every moment graph sheaf gives rise to a module over
Zk .G /.
To any equivalence class in h? with respect to
one can associate a moment
graph G in such a way that ZC .G / is the algebra ZS ./ that we considered
above. Hence every sheaf on G gives rise to a ZS ./-module. It is a very nice
feature of the theory that one can describe certain sheaves B./ on G for any
2 algorithmically that have the property that the global sections yields the
object V.PeS .// (up to extending scalars from S to e S). Hence one can construct
the indecomposable direct summands that occur in the Decomposition Problem II
directly! This gives a way to study the structure of the projectives locally (i.e., on
the vertices of the graph G ).
3.6.2 Soergel Bimodules
For an integral and regular equivalence class one can define a homomorphism
S S ! Z ./ (cf. [9]). Restriction along this homomorphism gives a functor
from ZS ./-modules to S-bimodules. Using this functor one can show that the
Decomposition Problem II is equivalent to the following:
Decomposition Problem III Decompose S Ss S St Su S in the category of
S-bimodules for all sequences s; t; : : : ; u of simple reflections.
The indecomposable direct summands that occur are called Soergel bimodules.
They were introduced in [20]. Soergel bimodules recently found applications in
faraway subjects such as knot theory!
68 P. Fiebig
3.6.3 Parity Sheaves
The last and probably most important connection of the above problems is to
geometry. Again suppose that is integral and regular. Consider the flag variety
G=B associated to the Lie algebra1 g, and its T-equivariant derived category
DbT;c .X; C/ of sheaves of C-vector spaces that are constructible along Schubert
cells. In [15], Juteau et al. described the remarkable additive subcategory of parity
sheaves that are characterized by certain cohomological vanishing properties. For
any w 2 W there is an up to shift unique parity sheaf E .w/ whose support is the
Schubert variety associated with w inside G=B.
Now there is a hypercohomology functor from DbT;c .G=B; C/ into the category
of modules over the equivariant cohomology HT .G=B/ of G=B. Again it turns out
that HT .G=B/ coincides with ZS ./ (canonically if we replace g by g_ )! Hence the
hypercohomology of parity sheaves yields ZS ./-modules, and again one can show
that
HT .E .w// S e
S V.PeS .w://
for all w 2 W . This yields a geometric way to understand the projective objects in
OeS; .
3.6.4 Another Proof of the KazhdanLusztig Conjecture
Apart from the original proofs of BeilinsonBernstein and BrylinskiKashiwara,

Soergels approach that we reviewed above also yields a proof of Conjecture 3.2.9.
It uses the geometric description of the indecomposable special Z -modules above,
and it amounts to showing (and this is extremely difficult) that the parity sheaves
(with coefficients in C) are in fact the intersection cohomology complexes on the
Schubert varieties in the flag variety. They have some remarkable properties, in
particular, the ranks of their stalk are given by KazhdanLusztig polynomials (cf.
[17]), and this fact implies Conjecture 3.2.9 and hence completes our journey.
There is a new proof due to Ben Elias and Geordie Williamson [5] that directly
solves the Decomposition Problem III. It is geometrically inspired, but does not
use any geometric result at all. It is quite difficult as well, but it has the advantage
that it also works in the more general situation of Coxeter systems, where the
Decomposition Problem III makes perfect sense, although a reformulation in terms
of geometry and parity sheaves is not available in general.
1
To be precise, one should replace g here by its Langlands dual g_ .
3.7 An Epilogue: The Modular Case
One can approach the problem of determining the irreducible rational characters
of a reductive algebraic group over a field of positive characteristic in a similar
way. However, the analogous results are much more difficult to obtain. For example,
the analogue of Soergels theory, i.e. the translation of the representation theoretic
problem into a combinatorial framework, is the main content of the book [1]. There
the authors define a combinatorial category over an arbitrary field k, that, in the
characteristic zero case, yields JordanHlder multiplicities for quantum groups at a
root of unity, and in positive characteristic, JordanHlder multiplicities for modular
Lie algebras. In both cases, the multiplicities are encoded in a decomposition
problem that very much resembles the decomposition problems that we encountered
in this overview article. One can approach this problem again by geometric or
moment graph methods (cf. [10]).
A toy example of the modular case is W. Soergels modular category O
introduced in [18]. The combinatorial counterpart of this is the category of Soergel
bimodules defined over a field of positive characteristic. It is in this category that
G. Williamson recently found counterexamples to Lusztigs modular character for-
mular for characteristics above the Coxeter number (cf. [21]). This is a very exciting,
intriguing and challenging result. One might hope that a better understanding of
the interplay between representation theoretic, combinatorial, and geometric ideas
will lead us to a better understanding of what happens in the case of small
characteristics.
Acknowledgements The author was partially supported by the DFG grant SP1388.
References
1. H.H. Andersen, J.C. Jantzen, W. Soergel, Representations of quantum groups at a p-th root of
unity and of semisimple groups in characteristic p: independence of p. Astrisque 220, 321
(1994)
2. A. Beilinson, J. Bernstein, Localisation de g-modules. C. R. Acad. Sci. Paris Sr. I 292, 1518
(1981)
3. T. Braden, R. MacPherson, From moment graphs to intersection cohomology. Math. Ann.
321(3), 533551 (2001)
4. J.-L. Brylinski, M. Kashiwara, Kazhdan-Lusztig conjecture and holonomic systems. Invent.
Math. 64, 387410 (1981)
5. B. Elias, G. Williamson, The Hodge theory of Soergel bimodules. Ann. Math. (2) 180(3),
10891136 (2014)
6. P. Fiebig, Centers and translation functors for category O over Kac-Moody algebras. Math. Z.
243(4), 689717 (2003)
7. P. Fiebig, The combinatorics of category O over symmetrizable Kac-Moody algebras. Trans-
form. Groups 11(1), 2949 (2006)
8. P. Fiebig, Sheaves on moment graphs and a localization of Verma flags. Adv. Math. 217, 683
712 (2008)
70 P. Fiebig
9. P. Fiebig, The combinatorics of Coxeter categories. Trans. Am. Math. Soc. 360, 42114233
(2008)
10. P. Fiebig, Sheaves on affine Schubert varieties, modular representations and Lusztigs conjec-
ture. J. Am. Math. Soc. 24, 133181 (2011)
11. P. Fiebig, Moment graphs in representation theory and geometry (2013, preprint).
arXiv:1308.2873
12. J.E. Humphreys, Reflection Groups and Coxeter Groups. Cambridge Studies in Advanced
Mathematics, vol. 29 (Cambridge University Press, Cambridge, 1990)
13. J.E. Humphreys, Representations of Semisimple Lie Algebras in the BGG Category O .
Graduate Studies in Mathematics, vol. 94 (American Mathematical Society, Providence, RI,
2008)
14. J.C. Jantzen, Moduln mit einem hchsten Gewicht. Lecture Notes in Mathematics, vol. 750
(Springer, New York, 1979)
15. D. Juteau, C. Mautner, G. Williamson, Parity sheaves. J. Am. Math. Soc. 27(4), 11691212
(2014)
16. D. Kazhdan, G. Lusztig, Representations of Coxeter groups and Hecke algebras. Invent. Math.
53(2), 165184 (1979)
17. D. Kazhdan, G. Lusztig, Schubert varieties and Poincar duality, in Geometry of the Laplace
operator (Proceedings of Symposia in Pure Mathematics) (University of Hawaii, Honolulu,
1979), pp. 185203
18. W. Soergel, On the relation between intersection cohomology and representation theory in
positive characteristic. J. Pure Appl. Algebra 152(13), 311335 (2000)
19. W. Soergel, Kategorie O , perverse Garben und Moduln ber den Koinvarianten zur Weyl-
gruppe. J. Am. Math. Soc. 3(2), 421445 (1990)
20. W. Soergel, Kazhdan-Lusztig-Polynome und unzerlegbare Bimoduln ber Polynomringen. J.
Inst. Math. Jussieu 6(3), 501525 (2007)
21. G. Williamson; with a joint appendix with Alex Kontorovich and Peter J. McNamara. Schubert
calculus and torsion explosion. J. Am. Math. Soc. 30, 10231046 (2017)
Chapter 4
The BCH-Formula and Order Conditions
for Splitting Methods
Winfried Auzinger, Wolfgang Herfort, Othmar Koch,

and Mechthild Thalhammer
Abstract As an application of the BCH-formula, order conditions for split-

ting schemes are derived. The same conditions can be obtained by using non-
commutative power series techniques and inspecting the coefficients of Lyndon
Shirshov words.
4.1 Introduction
The main purpose of this note is to present a not so well-known application of the
BakerCampbellHausdorff formula (BCH-formula): Computing order conditions
for exponential splitting schemes. There is vast literature, for an overview we
particularly refer to [14] and [12, Chap. III] and see [24, 57, 9, 13, 1518].
The topic of splitting is a comparatively young field and it is our intention to
present only facetswith Lie-theoretic background. We shall first recall a few
facts from Lie theory and power series as far as needed. An abstract definition
of splitting is given and the computation of order condition is demonstrated with
examples. The last section is devoted to an alternative approach for finding the order
conditions by inspecting the coefficients of leading LyndonShirshov words in an
exponential function of a sum of Lie elements, as currently used by the authors for
computationally generating order conditions for exponential splitting schemes.
W. Auzinger W. Herfort ()

Technische Universitt Wien, Wiedner Hauptstrae 8-10/E101, Vienna, Austria
e-mail: Winfried.Auzinger@tuwien.ac.at; Wolfgang.Herfort@tuwien.ac.at
O. Koch
Fakultt fr Mathematik, Universitt Wien, Oskar Morgensternplatz 1, Vienna, Austria
e-mail: othmar@othmar-koch.org
M. Thalhammer
Institut fr Mathematik, Universitt Innsbruck, Technikerstrasse 13, Innsbruck, Austria
e-mail: Mechthild.Thalhammer@uibk.ac.at

72 W. Auzinger et al.
4.2 Formal Power Series
Let k be a field of characteristic zero. Then khhSii denotes the algebra of formal
power series with coefficients in k and S a set of non-commuting variables. The
natural grading of khhSii is given as follows: elements in k have degree zero, those
in Sn WD fs1 : : : sn j sj 2 Sg have degree n, where n 1. A homogeneous element in
khhSii is a k-linear combination of elements of the same degree. Every element f in
khhSii allows a unique decomposition into homogeneous components
1
X
f D fj
jD0
where for each j the element fj is homogeneous of degree j.

In our context the power series ring Rhhtii in a single variable t and with
coefficients in a (notPnecessarily commutative) ring R will turn out to be useful.
Whenever f D j cj tj 2 khhtii and g 2 khhSii and g does not contain constant
terms then one can define the composition
X
f g WD cj g j
j
as, for given degree say n, for j n C 1, the power series g j does not contribute
homogeneous elements of degree n.
P
Example 4.2.1 The univariate formal power series f WD 1 1 j
jD0 j t will be denoted
P
by et . Hence the composition f g allows to consider eg D 1 1 j
jD0 j g .
The following simple fact will be helpful:
P
Lemma 4.2.1 Let h D 1 jD1 hj be an element in khhSii, with each hj homogeneous
P 1
of degree j. Then eh D jD0 ej with homogeneous terms ej , and the following
statements are equivalent:
(i) hj D 0 for j D 1; : : : ; p;
(ii) ej D 0 for j D 1; : : : ; p.
Proof Suppose that (i) holds. Then
1 2
eh D 1 C hpC1 C C hpC1 C C
2
shows that there cannot exist homogeneous terms ej with 1 6 j 6 p.
4 The BCH-Formula and Order Conditions for Splitting Methods 73
The converse is proved by induction. Suppose for p that e1 D D ep1 D 0

implies h1 D D hp1 D 0. Suppose next that also ep D 0. Then
1 2
eh D 1 C hp C C hp C C D e0 C epC1 C
2
From this one concludes that hp D 0 must hold. t
u
P1 P1
Corollary 4.2.2 Let h D jD1 hj and k D jD1 kj be elements in khhSii. Set e D
h
P1 P1
jD0 ej and e D
k
jD0 fj . Then the following statements are equivalent:
(a) hj D kj for j D 1; : : : ; p;
(b) ej D fj for j D 1; : : : ; p.
Proof Certainly (a) implies (b), as for forming the homogeneous terms in eh only
the terms up to order p contribute.
For proving the converse one again uses induction. Having established that hj D
kj for 1 6 j 6 p 1, one observes that ep D hp C
.h1 ; : : : ; hp1 / D kp C
.k1 ; : : : ; kp1 / D fp . Here

is a certain multivariate polynomial whose form to
know is not needed. Then, as ep D fp conclude that hp D kp . t
u
4.3 Reformulation Using Formal Differentiation
Given the algebra R WD khhSii, one may use it as the set of coefficients and form
the new algebra
P khhSiihhtii. There is a canonical
P1function
W khhSii ! khhSiihhtii that
sends f WD 1 jD0 fj to the element
. f / WD
j
jD0 fj t . In khhSiihhtii we define formal
differentiation by means of
0 1
X1 1
X
@ rj t jA
WD jrj1 tj1 :
jD0 jD1
Formal derivatives f .k/ of higher order, for elements f 2 khhSiihhtii, are defined
inductively.
P
Notation 4.3.1 If an element f D 1 jD0 rj t 2 khhSiihhtii has r0 D rp D 0, we
j
shall denote this by f D O.tpC1 / or even by f D O. p C 1/.

One proves without difficulty:
P
Lemma 4.3.2 For f D 1 jD0 fj the following statements are equivalent:
(i) f0 D D fp D 0;
(ii)
. f / D O.tpC1 /.
Next we prove a key lemma:

Lemma 4.3.3 For X 2 khhSiihhtii and an element C in khhSii of degree 1, the
following statements are equivalent:
(i) X eCt D O.tpC1 /;
(ii) X.0/ D 1 and D WD XP CX enjoys D D O.tp /.
Proof Certainly (i) implies (ii), as can be seen by differentiation. Conversely, if (ii)
holds, one only needs to check that X eCt vanishes when setting t D 0. But this is
a consequence of the assumption that X.0/ D 1. t
u
Here is the main observation about the different method to be used in Sect. 4.5
on splitting schemes. Namely it will imply that for deriving order conditions it is
equivalent to either consider them as the coefficients of a power series or to pass to
the logarithm and use thereby the BCH-formula and look at the coefficients of the
basic commutators.
Proposition 4.3.4 Suppose that h 2PkhhSii has the form h D eZ with Z0 D 0. Then
the following statements about h D 1 C
jD0 hj and e for an element C homogeneous
of degree 1 are equivalent:
p
(A) h0 1 D h1 C D D hp Cp D 0;
(B)
.h/ C
.h/ D O.tp /;
(C) Z C D O.tpC1 /.
P1
Proof As
.h/ D j
jD0 hj t , the equivalence of (A) and (B) follows from
Lemma 4.3.3. The equivalence of (B) and (C) is an immediate consequence of
Corollary 4.2.2. t
u
4.4 The BakerCampbellHausdorff Formula
The BakerCampbellHausdorff formula (BCH-formula, see, for instance, [12])

allows, for given X and Y in khSi without constant terms, to find Z 2 khhSii with
eX eY D eZ . In fact, Z turns out to be a formal infinite sum of X, Y and homogeneous
elements from the Lie algebra L.S/, generated by the set S and the bracket operation
l1 ; l2 D l1 l2 l2 l1 for li 2 L.S/.
Example 4.4.1 The first terms of Z are
1 1
Z D X C Y C X; Y C X; X; Y C Y; Y; X C
2 12
As noted, Z with the exception of the terms of first order is an infinite sum of Lie
elements, i.e., homogeneous elements of the Lie algebra generated by X and Y. One
denotes Z by log eX eY . Inductively one can derive an analog for log.eX1 : : : eXn /, for
Xi 2 S.
4.5 Splitting Schemes
The following abstract definition of a splitting scheme will serve our purpose:
Definition 4.5.1 Given the elements A and B of S and suppose there are, for j D
1; : : : ; s, Aj 2 spanfAg and Bj 2 spanfBg, i.e., Aj D aj A, Bj D bj B with scalar
coefficients aj ; bj ; j D 1; : : : ; s. Then these data determine a splitting scheme of
order at least p, provided
eA1 t eB1 t : : : eAs t eBs t e.ACB/t D O.tpC1 /
Here is an equivalent formulation of this condition. The proof, in the light of the
BCH-formula, is immediate from the definition:
Proposition 4.5.2 The following statements for given A and B in S and elements
Ai 2 spanfAg, Bi 2 spanfBg, where i D 1; : : : ; s, are equivalent:
(A) The data yield a splitting scheme of order at least p;
(B) log.eA1 eB1 eAs eBs / .A C B/ has homogeneous terms equal to zero for j D
1; : : : ; p.
Remark 4.5.3 Splitting techniques can also be successfully applied to nonlinear
evolution equations. The order conditions studied here are also valid for this general
case. This follows from an ingenious idea by Grbner [11], namely formally
to express the flow of a nonlinear evolution equation as the exponential of the
corresponding Lie derivative; see [12, Sect. III.5].
Let us, as a preparation for Sect. 4.6, compute the logarithm in (A) for s D 2 and
s D 3 up to terms of order p 6 3.
The BCH-formula easily yields
1 1
log.eAj eBj / D Aj CBj C Aj ; Bj C Aj ; Aj ; Bj C Bj ; Bj ; Aj CO.4/; (4.1)
2 12
where O.4/ stands for all terms in L.S/ of degree at least 4.

Example 4.5.1 Let X D X1 C X2 C X3 and Y D Y1 C Y2 C Y3 be a decomposition
into homogeneous elements with all nonlinear terms in L.S/. Then log.eX eY / D H
has first homogeneous terms
H1 D X1 C Y1
1
H2 D X2 C Y2 C X1 ; Y1
2
1
H3 D X3 C Y3 C .X1 ; Y2 C X2 ; Y1 /
2
1
C .X1 ; X1 ; Y1 C Y1 ; Y1 ; X1 /
12
Lemma 4.5.4 Let X WD aA C bB C cA; B C dA; B; B C eB; B; A and X 0 WD

a0 A C b0 B C c0 A; B C d0 A; B; B C e0 B; B; A, then H WD log.eX eY / has first
terms
H D H1 A C H2 B C H3 A; B C H4 A; A; B C H5 B; B; A
where
H1 D a C a0
H2 D b C b0
H3 D c C c0 C 12 .ab0 a0 b/
H4 D d C d0 C 12 .ac0 a0 c/ C 121
.ab0 a0 b/.a a0 /
H5 D e C e0 12 .bc0 b0 c/ 12
1
.ab0 a0 b/.b b0 /
Proof Using the preceding example for Y WD X 0 and elementary computation yields
the result. t
u
With the aid of Lemma 4.5.4 one finds:
Corollary 4.5.5 The first homogeneous terms of K WD log.ea1 A eb1 B ea2 A eb2 B ea3 A eb3 B /
are as follows:
K1 D .a1 C a2 C a3 /A C .b1 C b2 C b3 /B
1
K2 D .a1 b1 C a2 b2 C a3 b3 C a1 b2
2
a2 b1 C a1 b3 a3 b1 C a2 b3 a3 b2 / A; B
and K3 D A; A; B C B; B; A with
1 2
D .a b1 C a22 b2 C a33 b3 /
12 1
1
C .a1 a2 b2 a1 a2 b1 C a1 a3 b3 C a2 a3 b3 a1 a3 b1 a2 a3 b2 a1 a3 b2 C a2 a3 b1 /
4
1
C .a21 b2 a1 a2 b1 C a22 b1 a1 a2 b2 C a21 b3 a1 a3 b1 C a1 a2 b3 a1 a3 b2
12
C a1 a2 b3 a2 a3 b1 C a22 b3 a2 a3 b2
C a23 b1 a1 a3 b3 C a23 b2 a2 a3 b3 /
1 2
D .b a1 C b22 a2 C b33 a3 /
12 1
1
C .b1 b2 a2 b1 b2 a1 C b1 b3 a3 C b2 b3 a3 b1 b3 a1 b2 b3 a2 b1 b3 a2 C b2 b3 a1 /
4
1
C .b21 a2 b1 b2 a1 C b22 a1 b1 b2 a2 C b21 a3 b1 b3 a1 C b1 b2 a3 b1 b3 a2
12
C b1 b2 a3 b2 b3 a1 C b22 a3 b2 b3 a2
C b23 b1 b1 b3 a3 C b23 a2 b2 b3 a3 /:
Proof One first computes sj WD log.eaj A ebj B / D aj A C bj B C 12 aj bj A; B C

1
.a2 b A; A; B C aj b2j B; B; A/ C O.4/. Then, using Lemma 4.5.4, compute
12 j j
first H WD log.es1 es2 / and, again using the lemma, find the desired expressions by
computing K WD log.eH es3 /. t
u
4.6 Computing Order Conditions in Examples
4.6.1 Schemes of Order At Least 1
It follows right from the definition that in this case
X
s X
s
Aj D A; Bj D B
jD1 jD1
i.e., for Aj D aj A and Bj D bj B one obtains
X
s X
s
aj D 1; bj D 1:
jD1 jD1
4.6.2 The Order Conditions for s D 2 and p D 3
Elementary computation leads to the following observation:

Lemma 4.6.1 The order conditions for s D 2 and p D 3 are as follows:
1 D a1 C a2
1 D b1 C b2
0 D a1 b1 C a2 b2 C a1 b2 a2 b1
0 D 1 6a1 a2 b1
0 D 1 6b1 b2 a2
Proof The order condition for s D 1 in the first line follows from the previous
subsection. The higher order conditions result from Lemma 4.5.4. t
u
4.6.3 The Order Conditions for s D 3 and p D 3
Making use of Corollary 4.5.5 the conditions on the coefficients aj and bj for
j D 1; 2; 3 in order to let Aj D aj A and Bj D bj determine the necessary and
sufficient conditions for a splitting scheme of order p at least 3 when s D 3.
Lemma 4.6.2 The order conditions for s D p D 3 read as follows:
1 D a1 C a2 C a3
1 D b1 C b2 C b3
1
D a2 b1 C a3 b1 C a3 b2
2
2 D 3.a2 C a3 / 6a2 a3 b2
2 D 3.b1 C b2 / 6b1 b2 a2
Proof In Corollary 4.5.5, one equates the coefficients of A and B to 1, and those of
A; B, A; A; B and B; B; A to zero. Then, using the third equation, terms a3 b2
have been eliminated from the last two equations. t
u
To conclude this section let us remark that expanding eX1 CX2 C as a Taylor series,
one finds from Lemma 4.2.1 and Proposition 4.3.4:
Proposition 4.6.3 The following statements for an exponential function eX for X
a sum of homogeneous Lie elements with the exception of the linear term are
equivalent:
(i) X1 D X2 D D Xp D 0
(ii) As a power series in khhSii the first nonvanishing homogeneous term is
1
X .
. pC1/ pC1
4.7 Alternative Approach via Taylor Expansion

and Computation in the Free Lie Algebra
Generated by A; B
According to the ideas from [1, 3] systems of polynomial equations representing

order conditions for splitting methods are set up in a different way without making
explicit use of the BCH-formula. This is straightforward to implement in computer
algebra. The resulting systems of equations are not identical but equivalent to those
obtained when implementing the BCH-based procedure described above.
This alternative approach described can also easily be adapted and generalized
to cases with various symmetries, pairs of schemes, and more general cases like
splitting involving three operators A; B; C, or more, see [3].
To find conditions for the coefficients aj ; bj such that for Aj D aj A, Bj D bj B a

scheme of order p is obtained,
L.t/ WD eA1 t eB1 t : : : eAs t eBs t e.ACB/t D O.tpC1 /;
we consider the Taylor expansion of L.t/, the local error of the splitting scheme
applied with stepsize t (satisfying L.0/ D 0 by construction),1
Xp
t q dq
L.t/ D q
L.0/ C O.tpC1 /:
qD1
q dt
The method is of order p iff L.t/ D O.tpC1 /; thus, the conditions for order p are
given by
d dp
L.0/ D D p L.0/ D 0: (4.2)
dt dt
Via successive differentiation of L.t/ we obtain the following homogeneous repre-

q
sentation of dtd q L.0/ in terms of power products in the non-commuting variables A
and B: With k D .k1 ; : : : ; ks / 2 Ns0 , one obtains
! s kj !
dq X q Y X kj l kj l
L.0/ D Aj Bj .A C B/q ; q D 0; 1; 2; : : : (4.3)
dtq k jD1 lD0 l
jkjDq
In a computer algebra system, these symbolic expressions can be generated in a

straightforward way.
If conditions (4.2) are satisfied up to a given order p, then the leading term of
tpC1 dpC1
the local error is given by . pC1/ dtpC1
L.0/. Proposition 4.6.3 shows that this leading
local error term is a homogeneous linear combination of Lie elements. With the
terminology
LC.q/ WD the sum (4.3) is a linear combination of Lie elements of degree q
this amounts to LC. p C 1/ being true for a scheme of order p. Exploiting this
statement allows to design a recursive algorithm for generating a set of order
conditions:
(i) By construction, LC.1/ D 0 holds. But, a priori, for q > 1 the expression (4.3)
q
for dtd q L.0/ does not enjoy LC.q/, see Example 4.7.1 below. On the other hand,
from Proposition 4.6.3 we know that
LC.1/ ^ : : : ^ LC.q 1/ ) LC.q/:

1
Here, dq
dtq
L.0/ WD dq
dtq
L.t/tD0 .
By induction over q we see that each solution of (4.2) must satisfy LC.q/ for
q D 1; : : : ; p. (Moreover, for each the resulting solution LC. p C 1/ will hold
true.)
(ii) Due to (i), for the purpose of solving the system (4.2) we may assume
dq X
q
L.0/ D q;k Bq;k ; q D 1; : : : ; p
dt k
where the Bq;k of degree q are elements from a basis of the free Lie algebra
generated by A and B. Now the problem is to identify, on the basis of the
expressions (4.3), coefficients q;k D q;k .aj ; bj / such that the polynomial
system
q;k D 0; q D 1; : : : ; p; k running over all basis elements of degree q

(4.4)
will be equivalent to (4.2). To this end we make use of the one-to-one
correspondence between basis elements of degree q represented by non-
associative, bracketed words (commutators) and associative words of length
q over the alphabet fA; Bg. The implementation described in [1, 3] relies on the
Lyndon basis, also called LyndonShirshov basis, which can be generated by
an algorithm devised in [10]. With this choice,
each basis element Bq;k is uniquely represented by a

Lyndon word of length q associated with the leading term, (4.5)
in lexicographical order, of the expanded version of Bq;k :
Identifying the coefficients of these Lyndon monomials results in the desired

polynomial system (4.4).
To be more precise we note that, in general, a given Lyndon monomial
shows up in different expanded commutators. Therefore, equating coefficients
of Lyndon monomials will not directly result in the system (4.4) but, as a
consequence of (4.5), in an equivalent system which is obtained from (4.4) by
premultiplication with a regular triangular matrix, see [3].
For the underlying theoretical background concerning Lyndon bases in free Lie
algebras, we refer to [8]. For a detailed illustration of our approach for order p D 5,
see Example 2 from [3]. In the following example we reconsider the case s D p D 3.
Example 4.7.1 For s D 3 we have [see (4.3)]
d
L.0/ D a1 C a2 C a3 1 A C b1 C b2 C b3 1 B;
dt
d2
2
L.0/ D .a1 C a2 C a3 /2 1 A2
dt

C 2a1 .b1 C b2 C b3 / C 2a2 .b2 C b3 / C 2a3 b3 1 AB

C 2a2 b1 C 2a3 .b1 C b2 / 1 BA

C .b1 C b2 C b3 /2 1 B2 ;
Assume that a1 C a2 C a3 D 1 and b1 C b2 C b3 D 1 such that dtd L.0/ D 0. Then

2
substituting a1 D 1 a2 a3 and b3 D 1 b1 b2 into dtd 2 L.0/ gives the commutator
expression
d2
2
L.0/ D 2a2 b1 C 2a3 .b1 C b2 / 1 A; B:
dt
Therefore the system
a1 C a2 C a2 D 1
b1 C b2 C b3 D 1
1
a2 b1 C a3 .b1 C b2 / D
2
represents a set of conditions for order p D 2.
Extending this computation to p D 3 by hand is already somewhat laborious.
However, from the above considerations we know that assuming the conditions for
order p D 2 are satisfied, then
d2
L.0/ D AAB A; A; B C ABBA; B; B;
dt2
where
d2
AAB D coefficient of the power product A2 B in the expression (4.3) for L.0/;
dt2
d2
ABB D coefficient of the power product AB2 in the expression (4.3) for L.0/:
dt2
Here the two independent commutators A; A; B and A; B; B are represented by
the associative Lyndon words AAB and ABB. In computer algebra, extraction of
2
the coefficients AAB and ABB from the symbolic expression dtd 2 L.0/ is straightfor-
ward. In this way we end up with the system
a1 C a2 C a3 D 1
b1 C b2 C b3 D 1
1
a2 b1 C a3 .b1 C b2 / D
2
2 1
a2 b21 C a3 b1 C b2 D
3
1
.a2 C a3 /2 b1 C a23 b2 D
3
representing a set of order conditions for order p D 3. The system is equivalent to
the one found in Lemma 4.6.2. We note that there is a one-dimensional zero solution
manifold containing well-known rational solutions, e.g.,
7 3 1
a1 D ; a2 D ; a3 D ;
24 4 24
2 2
b1 D ; b2 D ; b3 D 1:
3 3
Acknowledgements Wolfgang Herfort is indebted to UNIPA for generous support in June 2016.
He also would like to thank the Department of Mathematics at the Brigham Young University
for the great hospitality during the year 2015. Special thanks to JIM LOGAN for his excellent
support during this time with hard- and software.
Othmar Koch acknowledges the support by the Vienna Science and Technology Fund (WWTF)
under the grant MA14-002.
References
1. W. Auzinger, W. Herfort, Local error structures and order conditions in terms of Lie elements
for exponential splitting schemes. Opuscula Math. 34(2), 243255 (2014)
2. W. Auzinger, O. Koch, M. Thalhammer, Defect-based local error estimators for splitting
methods, with application to Schrdinger equations, Part II. Higher order methods for linear
problems. J. Comput. Appl. Math. 255, 384403 (2013)
3. W. Auzinger, H. Hofsttter, D. Ketcheson, O. Koch, Practical splitting methods for the adaptive
integration of nonlinear evolution equations. Part I: construction of optimized schemes and
pairs of schemes. BIT Numer. Math. Published online: 28 July, 2016. http://dx.doi.org/10.1007/
s10543-016-0626-9
4. S. Blanes, F. Casas, On the convergence and optimization of the BakerCampbellHausdorff
formula. Linear Algebra Appl. 378, 135158 (2004)
5. S. Blanes, P.C. Moan, Practical symplectic partitioned Runge-Kutta and Runge-Kutta-Nystrm
methods. J. Comput. Appl. Math. 142, 313330 (2002)
6. S. Blanes, F. Casas, A. Murua, Splitting and composition methods in the numerical integration
of differential equations. Bol. Soc. Esp. Mat. Apl. 45, 89145 (2008)
7. S. Blanes, F. Casas, P. Chartier, A. Murua, Optimized high-order splitting methods for some
classes of parabolic equations. Math. Comput. 82, 15591576 (2013)
8. L. Bokut, L. Sbitneva, I. Shestakov, LyndonShirshov words, Grbner-Shirshov bases, and free
Lie algebras, in Non-associative Algebra and its Applications, Chap. 3 (Chapman & Hall/CRC,
Boca Raton, FL, 2006)
9. P. Chartier, A. Murua, An algebraic theory of order. M2AN Math. Model. Numer. Anal. 43,
607630 (2009)
10. J.-P. Duval, Gneration dune section des classes de conjugaison et arbre des mots de Lyndon
de longueur borne. Theor. Comput. Sci. 60, 255283 (1988)
11. W. Grbner, Die Liereihen und ihre Anwendungen, 2nd edn. (VEB Deutscher Verlag der
Wissenschaften, Berlin, 1967)
12. E. Hairer, C. Lubich, G. Wanner, Geometrical Numerical Integration Structure-Preserving
Algorithms for Ordinary Differential Equations, 2nd edn. (Springer, Berlin, 2006)
13. E. Hansen, A. Ostermann, Exponential splitting for unbounded operators. Math. Comput. 78,
14851496 (2009)
14. R. McLachlan, R. Quispel, Splitting methods. Acta Numer. 11, 341434 (2002)
15. M. Suzuki, General theory of higher-order decomposition of exponential operators and
symplectic integrators. Phys. Lett. A 165, 387395 (1992)
16. M. Thalhammer, High-order exponential operator splitting methods for time-dependent
Schrdinger equations. SIAM J. Numer. Anal. 46, 20222038 (2008)
17. Z. Tsuboi, M. Suzuki, Determining equations for higher-order decompositions of exponential
operators. Int. J. Mod. Phys. B 9, 32413268 (1995)
18. H. Yoshida, Construction of higher order symplectic integrators. Phys. Lett. A 150, 262268
(1990)
Chapter 5
Cohomology Operations Defining Cohomology
Algebra of the Loop Space
Tornike Kadeishvili
Abstract According to the minimality theorem (Kadeishvili, Russ Math Surv

35(3):231238, 1980), on the cohomology H .X/ of a topological space X there
exists a sequence of operations mi W H .X/i ! H .X/; i D 2; 3; : : : which
form a minimal A1 -algebra .H .X/; fmi g/. This structure defines on the bar
construction BH .X/ a correct differential dm so that .BH .X/; dm / gives coho-
mology modules of the loop space H .X/. In this paper we construct algebraic
operations Ep;q W H .X/p H .X/q ! H .X/; p; q D 0; 1; 2; 3; : : : which turn
.H .X/; fmi g; fEp;q g/ into a B1 -algebra. This structure defines on BH .X/ a correct
multiplication, thus determines a cohomology algebra H .X/.
5.1 Introduction
The Adamss bar construction BC .X/ of cochain complex .C .X/; d; ^/ of a
topological space X defines cohomology modules of the loop space H .X/ [1],
but not the cohomology algebra H .X/.
In [2] Baues constructed a sequence of multioperations
E1;k W C .X/ C .X/k ! C .X/; k D 1; 2; 3; : : :
so that they determine on BC .X/ a correct multiplication. That is, this structure
.C .X/; d; ^; fE1;k g/, which now is called homotopy Gerstenhaber algebra [6, 20]
(hGa in short) determines a cohomology algebra H .X/.
Our aim is to transfer these structures to homology level, i.e. from C .X/ to

H .X/.
T. Kadeishvili ()
A. Razmadze Mathematical Institute of Tbilisi State University, Tbilsi, Georgia
e-mail: kade@rmi.ge

86 T. Kadeishvili
Namely, according to the minimality theorem [9], the differential d W C .X/ !

C1
C .X/ and the multiplication (cup product) ^W C .X/ C .X/ ! C .X/
determine a sequence of operations
mi W H .X/i ! H .X/; i D 2; 3; : : :
which form a minimal A1 -algebra .H .X/; fmi g/. This structure defines on the
bar construction BH .X/ a correct differential dm W BH .X/ ! BH .X/ so
that .BH .X/; dm / and BC .X/ have isomorphic homology modules, thus this
object .H .X/; fmi g/ which is called cohomology A1 -algebra of X, determines
cohomology modules of the loop space H .X/. But not the cohomology algebra
H .X/.
In this paper, as the next step, we transport on cohomology level Bauess
operations
E1;k W C .X/ C .X/k ! C .X/; k D 1; 2; 3; : : : ;
so we obtain algebraic operations
Ep;q W H .X/p H .X/q ! H .X/; p; q D 0; 1; 2; 3; : : :
which turn .H .X/; fmi g; fEp;q g/ into a B1 -algebra. These operations define
on BH .X/ a correct multiplication, which determines the cohomology algebra
H .X/.
Now we describe the algebraic framework for the above, where we will work.
For a differential graded algebra .A; d; / (dg algebra in short) the differential
d W A ! A1 and the multiplication W Ap Aq ! ApCq define on the bar
construction
1
X X
BA D An ; dim.a1 an / D dim ak C n
nD0 k
a differential dB W BA ! BA given by (signs are ignored, let us work over Z2 )

X
dB .a1 an / D a1 dak an
k
X
C a1 .ak ; akC1 / an (5.1)
k
which turns .BA; dB ; rB / into a dg coalgebra with comultiplication
rB W BA ! BA BA;
X
n
rB .a1 an / D .a1 ak / .akC1 an /: (5.2)
kD0
5 Cohomology Operations Defining Cohomology Algebra of the Loop Space 87
To catch the multiplicative structure, there exists the notion of homotopy

Gerstenhaber algebra, see [6] and [20] (hGa in short), which we describe below, and
which allows to construct a correct multiplication on the bar construction. This is an
additional structure on a dg algebra .A; d; /, consisting of a sequence of operations
E1;k W A Ak ! A; k D 1; 2; 3; : : :
which determine on BA a multiplication E W BA BA ! BA turning

.BA; dB ; rB ; E / into a dg bialgebra. In [2] Baues has constructed such a structure
in A D C .X/, so that BC .X/ gives the cohomology algebra H .X/.
We are going to transfer this structure to the homology level, i.e. from A to H.A/
aiming BA and BH.A/ to carry the same amount of information.
Note that for a dg algebra .A; d; / its homology H.A/ is also a dga with trivial
differential and induced multiplication W H.A/ H.A/ ! H.A/. But generally
the bar constructions BA of dg algebra .A; d; / and the bar construction BH.A/ of
dg algebra .H.A/; d D 0; / have different homologies. This means that stepping
from .A; d; / to .H.A/; d D 0; / we lose part of information.
This lose compensates the minimality theorem from [9] which states that the
homology H.A/ of a dg algebra .A; d; / (all Hi .X/-s are assumed free) can be
equipped with a sequence of multioperations
mi W H.A/i ! H.A/; i D 1; 2; 3; : : : I m1 D 0; m2 D
turning .H.A/; fmi g/ into a minimal A1 -algebra in the sense of Stasheff [19] which
is weak equivalent to dga .A; d; / in a certain category. These A1 operations fmi g
determine on BH.A/ a new, perturbed differential dm W BH.A/ ! BH.A/ by
X
dm .a1 an / D a1 ak mi .akC1 akCi /akCiC1 an
k;i
so that BA and .BH.A/; dm / have isomorphic homology modules.

The aim of this paper is to construct for a hGa .A; d; ; fE1k g/ on its homology
A1 -algebra .H.A/; fmi g/ a certain additional structure, the so-called B1 algebra
[8], consisting of multioperations
Ep;q W H.A/p H.A/q ! H.A/; p; q D 0; 1; 2; 3; : : : ;
which determines on BH.A/ a correct multiplication so that the bar constructions

BH.A/ and BA will have isomorphic homology algebras.
Remark This can be summarized as follows: If A is a dg algebra, hGa, or a
commutative dg algebra, then H.A/ becomes, respectively, an A1 [9], B1 (present
paper), or C1 [12] algebra (a commutative version of A1 ).
The plan of the paper is as follows. In the section Preliminaries we present the
notions, facts, and constructions needed for the main goal of the paper. Namely,
88 T. Kadeishvili
in Sect. 5.2.1 we present the notions of A1 -algebra and A1 -morphisms, their

interpretation in terms of bar construction. Then in Sect. 5.2.2 we discuss the
notions of twisting cochains and their equivalence, again with interpretation in
terms of bar construction, and present the Berikashvilis theorem about the functor
Dthe set of equivalence classes of twisting cochains. In the next Sect. 5.2.3 we
generalize all these notions and facts about twisting cochains from dg algebras to
A1 -algebras and prove the lifting theorem for these A1 -twisting cochains, which
is our main technical tool. The next Sect. 5.2.4 is dedicated to the notion of B1 -
algebra, its particular case hGa, to the interpretation of these notions in terms of
twisting cochains; moreover, three remarkable and generic examples of homotopy
Gerstenhaber algebras are given. In Sect. 5.3, using this theorem about the lifting
of A1 -twisting cochains, we transport the hGa structure existing on a dg algebra
A to the B1 -algebra structure on its homology H.A/. In the final section we use
the obtained structures in a topological framework, to obtain a B1 -algebra structure
on the cohomology H .X/ which determines the cohomology algebra of the loop
space H .X/.
5.2 Preliminaries
In this section we give some notions and constructions needed for the goal of the
paper.
5.2.1 Category of A1 -Algebras
The notion of A1 -algebra was introduced by James Stasheff in [19]. This is a

generalization of the notion of differential graded algebra, where one of the defining
conditions, namely the associativity of the multiplication, is satisfied not strictly but
up to certain higher homotopies.
The notion of A1 -algebra gave rise to a series of notions of so-called homotopy
algebras, where various axioms are fulfilled not strictly but up to some higher
homotopies, such as: C1 -algebra which is the commutative version of A1 -algebra;
homotopy Gerstenhaber algebra (hGa) where the commutativity is satisfied up to
homotopy; its generalization B1 -algebra; L1 -algebra, where the Jacobi identity is
satisfied up to homotopy; H1 , the homotopy version of dg Hopf algebra where the
connection between multiplication and comultiplication is satisfied up to homotopy,
etc.
5.2.1.1 A1 -Algebras
An A1 -algebra [19] is a graded module M with a given sequence of operations
fmi W M i ! M; i D 1; 2; : : : ; deg mi D i 2g
which satisfies the following conditions
X X
nj
mi .a1 ak mj .akC1 akCj / an / D 0: (5.3)
iCjDnC1 kD0
In particular, for the operation m1 W M ! M we have deg m1 D 1 and m1 m1 D

0, this m1 can be regarded as a differential on M. The operation m2 W M M ! M
has the degree 0 and satisfies
m1 m2 .a1 a2 / C m2 .m1 a1 a2 / C m2 .a1 m1 a2 / D 0;
i.e., m2 can be regarded as a multiplication on M and m1 is a derivation with respect

to it. Thus .M; m1 ; m2 / is a sort of (possibly nonassociative) dg algebra. For the
operation m3 its degree is 1 and the condition (5.3) gives
m2 .m2 .a1 a2 / a3 / C m2 .a1 m2 .a2 a3 //

D m1 m3 .a1 a2 a3 / C m3 .m1 a1 a2 a3 /
Cm3 .a1 m1 a2 a3 / C m3 .a1 a2 m1 a3 /;
this means that the product m2 is homotopy associative and the appropriate chain
homotopy is m3 .
5.2.1.2 Bar Construction of an A1 -Algebra
The meaning of higher operations mi can be clarified using the notion of bar
construction of an A1 -algebra .M; fmi g/. It is, as above
1
X X
BM D M k ; dim.a1 an / D dim ak C n
kD0 k
but it can also be represented as
BM D T c .sM/ D C sM C sM sM C sM sM sM C
90 T. Kadeishvili
here is the ground ring and .sM/k D Mk1 is the standard suspension operator.
BA is the graded coalgebra with comultiplication rB , see (5.2). In fact it is the cofree
graded coalgebra cogenerated by M, that is it has the following universal property:
p
M T c .sM/

- " f
(5.4)
K
for each graded coalgebra .K; rK W K K/ and each homomorphism

W K ! M
of degree 1 there exists coextension, the unique graded coalgebra map f
W K !
T c .sM/ such that pf
D
where p W T c .sM/ ! M is the clear projection. In fact f
is given by
X
f
D .

/rKi (5.5)
i
here rKi W K ! K i is the iteration of comultiplication rK :
rK1 D idK ; rK2 D rK ; rKn D .id rK /rKn1 :
The main meaning of Stashefs defining condition (5.3) of an A1 -algebra

.M; fmi g/ is the following. The sequence of operations fmi g determines on the bar
construction BM a coderivation
X
dm .a1 an / D a1 ak mj .akC1 akCj / an ;
k;j
P
and the Stasheffs condition (5.3) is exactly mdm D 0 where m D i mi W BM !
M; and this, since of cofreeness is equivalent to dm dm D 0. Consequently dm is a
differential and .BM; dm ; r/ is a dg coalgebra, which is called bar construction of
A1 -algebra .M; fmi g/.
5.2.1.3 Morphisms of A1 -Algebras
A morphism of A1 -algebras f W .M; fmi g/ ! .M 0 ; fm0i g/ can be defined as a dg

coalgebra map of bar constructions
f W B.M; fmi g/ ! B.M 0 ; fm0i g/:
Because of the cofreeness of the tensor coalgebra T c .sM/, such f is uniquely

determined by the projection
f p
p0 f W B.M; fmi g/ ! B.M 0 ; fm0i g/ ! M 0 ;
which, in fact, is a collection of homomorphisms
f fi W M i ! M 0 ; i D 1; 2; : : : ; deg fi D i 1g: (5.6)
Such a collection reconstructs a coalgebra map f W T c .sM/ ! T c .M 0 / by
X
i X
f .a1 an / D fk1 .a1 ak1 / fkt .aikt C1 ai /:
tD1 k1 CCkt Di
If we wish f to be also a chain map f W BM ! BM 0 , that is fdm D dm0 f , again

because of the cofreeness it suffices to check p0 fdm D p0 dm0 f and this in terms of
components is equivalent to the condition
X
i1 X
ik
fijC1 .a1 ak mj .akC1 akCj / ai /
kD0 jD1
X
i X
D m0t . fk1 .a1 ak1 / fkt .aikt C1 ai //;
tD1 k1 CCkt Di
(5.7)
see, for example, [9, 12].

So a morphism of A1 -algebras f W B.M; fmi g/ ! B.M 0 ; fm0i g/ can be defined as
a collection of homomorphisms f fi g satisfying the condition (5.7).
These conditions particularly imply f1 m1 D m1 f1 , i.e. f1 W .M; m1 / ! .M 0 ; m01 /
is a chain map. We define a weak equivalence of A1 -algebras as a morphism f fi g
where f1 induces homology isomorphism.
It follows easily from the definitions that an A1 -algebra .M; fmi g/ with m>2 D 0
is just a dg algebra, and an A1 -algebra morphism
f fi g W .M; fm1 ; m2 ; 0; 0; : : :g/ ! .M 0 ; fm01 ; m02 ; 0; 0; : : :g/
with f>1 D 0 is just a multiplicative chain map, thus the category of dg algebras is a
subcategory of A1 -algebras.
Assigning to .M; fmi g/ its bar construction .BM; dm ; r/ we obtain a functor from
the category of A1 -algebras to the category of dg coalgebras.
5.2.1.4 Minimal A1 -Algebras
An A1 -algebra .M; fmi g/ we call minimal if m1 D 0; in this case, .M; m2 / is strictly

associative graded algebra, see (5.3) for n D 3. Suppose
f W .M; fmi g/ ! .M 0 ; fm0i g/

92 T. Kadeishvili
is a weak equivalence of minimal A1 -algebras, then f1 W .M; m1 D 0/ ! .M 0 ; m01 D

0/, which by definition should induce isomorphism of homology, is automatically
an isomorphism. It is not hard to check that, in this case, f is an isomorphism in the
category of A1 -algebras, thus each weak equivalence of minimal A1 -algebras is
an isomorphism. This fact motivates the word minimal in this notion: the Sullivans
minimal model has similar propertya weak equivalence of minimal dg algebras is
an isomorphism.
Let us repeat the minimality theorem from [9] mentioned in Introduction.
Theorem 1 For a dg algebra .A; d; / its homology H.A/ (all Hi .X/-s are assumed
free) can be equipped with a sequence of multioperations
mi W H.A/i ! H.A/; i D 1; 2; 3; : : : I m1 D 0; m2 D
turning .H.A/; fmi g/ into a minimal A1 -algebra in sense of Stasheff [19] for which
m2 D and there exists a weak equivalence of A1 -algebras
f fi g W .H.A/; fmi g/ ! .A; fm1 D d; m2 D ; m3 D 0; m4 D 0; : : :g/:
This A1 -algebra structure is unique up to isomorphism in the category of A1 -

algebras.
In particular, BA and .BH.A/; dm / have isomorphic homology modules.
5.2.2 Twisting Cochains
Let .K; dK ; rK W K ! K K/ be a dg coalgebra and .A; dA ; / be a dg algebra.

Consider the cochain complex Hom.K; A/ which is a dg algebra with differential
d
D dA
C
dK , multiplication
^ D .
/rK , and grading
Homq .K; A/ D f f W K ! ACq g.
A twisting cochain is defined as a homomorphism
W K ! A of degree 1 (that
is,
W K ! A1 ) satisfying the Browns condition [4]
d
D
^
(5.8)
(MaurerCartan equation, master equation in other terms). Let T.K; A/ be the set of
all twisting cochains.
Bellow assume that .A; dA ; / is connected, that is nonnegative A<0 D 0 and
A0 D ; dA jA1 D 0. Then a twisting cochain
W K ! A can be presented as

D
2 C
3 C C
n C ;
n D
jKn W Kn ! An1 ;
and the Browns condition for components

n means
X
n2
dA
n D
n1 dK C
i ^
ni :
iD2
5.2.2.1 Twisting Cochains and the Bar Construction
Because of the universal property (5.4) any homomorphism

W K ! A of degree
1 induces a graded coalgebra map f
W K ! T c .sA/ by
X
f
D .

/rKi :
i
If, in addition,
W K ! A is a twisting cochain, that is satisfies the Browns
condition (5.8), then f
W K ! B.A/ is a chain map, i.e. is a map of dg coalgebras:
the projection of the condition dB f
D f
dK on A gives pdB f
D pf
dK and this is
exactly d
D
^
.
Conversely, any dg coalgebra map f W K ! BA is f
for
D pf W K ! BA ! A.
In fact we have a bijection Mordgcoalg .K; BA/ $ T.K; A/.
5.2.2.2 Equivalence of Twisting Cochains
Two twisting cochains

; W K ! A are equivalent [3] if there exists c W K !
A; deg c D 0, such that
D
C cdK C dA c C ^ c C c ^
; (5.9)
notation

c .
This notion of equivalence allows to perturb twisting cochains. Let

D
2 C
3 C C
n C W K ! A
be a twisting cochain, and lets take an arbitrary cochain c D cn W Kn ! An . Then

there exists a twisting cochain Fcn
D W K ! A such that

cn . Actually the
components of the perturbed twisting cochain
F cn
D D 2 C 3 C C n C
can be solved from (5.9) inductively, and the solution in particular gives that the
perturbation Fcn
does not change the first components, i.e. i D
i for i < n and
n D
n C d A c n .
94 T. Kadeishvili
5.2.2.3 Equivalence Twisting Cochains and Homotopy of Induced Maps
In the category of dg coalgebras there is the following notion of homotopy: two dg

coalgebra maps f ; g W .K; dK ; rK / ! .K 0 ; dK 0 ; rK 0 / are homotopic, if there exists
D W K ! K 0 of degree C1 such that dK 0 D C Ddk D f g, i.e. the chain maps f and g
are chain homotopic, and additionally the homotopy D is a f g-coderivation, that
is rK 0 D D . f D C D g/rK :
If

c , then f
and f are homotopic in this sense: chain homotopy D.c/ W
K ! BA is given by
X
D.c/ D . . j-times/ c

/rKi ; (5.10)
i;j
it satisfies f
f D dB D.c/ C D.c/dK because the projection of this condition on
A gives pf
pf D pdB D.c/ C pD.c/dK and this is exactly the condition (5.9).
Besides, D.c/ is a f
f -coderivation, that is
rB D.c/ D . f D.c/ C D.c/ f

/rK ;
this also follows from the universal property of tensor coalgebra and the extension
rule (5.10).
5.2.2.4 Functor D
Berikasvilis functor D.K; A/ is defined as the factorset D.K; A/ D M.K;A/ .

Assigning to a twisting chain
W K ! A the dg coalgebra map f
W K ! BA
and having in mind that

c implies f

D.c/ f we obtain a bijection
D.K; A/ $ K; BA where K; BA denotes the set of chain homotopy classes in
the category of dg coalgebras.
Any dg algebra map f W A ! A0 induces the map T.K; A/ ! T.K; A0 /: if
W
K ! M is a twisting cochain so is the composition f
W K ! A ! A0 . Moreover, if

c
0 , then f

fc f
0 . Thus we have a map D. f / W D.K; A/ ! D.K; A0 /.
Theorem 2 (Berikashvili [3]) Let .K; dK ; rK / be a dg coalgebra with free Ki s
and .A; dA ; A / be a connected dg algebra. If f W A ! A0 is a weak equivalence of
connected dg algebras (i.e., homology isomorphism), then
D. f / W D.K; A/ ! D.K; A0 /
is a bijection.
In fact this theorem means that K; BA ! K; BA0 is a bijection. The theorem
consists of two parts:
Surjectivity Any twisting cochain
W K ! A0 can be lifted to a twisting cochain
W K ! A so that

f .
0
Injectivity If
2 T.K; A/, then f
f 2 T.K; A0 /.
5.2.2.5 Lifting of Twisting Cochains
Below well need the surjectivity part of this theorem whose proof we sketch here.
Theorem 3 Let .K; dK ; rK / be a dg coalgebra with free Ki s and let f W
.A; dA ; A / ! .A0 ; dA0 ; A0 / be a weak equivalence of connected dg algebras,
then for an arbitrary twisting cochain

D
2 C
3 C C
n C W K ! A 0
there exists a twisting cochain
D 2 C 3 C C n C W K ! A
such that

f .
Proof Start with a twisting cochain
D
2 C
3 C C
n C W K ! A0 : The
condition (5.8) gives dA
2 D 0, and since f W A ! A0 is a homology isomorphism
then there exist 2 W K2 ! A1 and c02 W K2 ! A02 such that dA 2 D 0 and f 2 D
2 C dA0 c02 (we assume all Kn s are free modules). Perturbing

by this c02 we obtain
new twisting cochain Fc02
for which .Fc02
/2 D
2 C dA0 c02 D f 2 . So we can
assume that
2 D f 2 .
Assume now that we already have 2 ; 3 ; : : : ; n1 which satisfy (5.8) in
appropriate dimensions and f k D
k ; k D 2; 3; : : : ; n 1. We need the next
component n W Kn ! An1 such that
X
n2
dA n D n1 dK C i ^ ni
iD2
and c0n W Kn ! A0n such that f n D

n C dA0 c0n . Then perturbing
by c0n we obtain
new
n for which f n D
n , and this will complete the proof.
Let us denote by
X
n2
Un D n1 dK C i ^ ni W Kn ! An2 ;
iD2
X
n2
Un0 D
n1 dK C
i ^
ni W Kn ! A0n2 :
iD2
So we have Un0 D dA0

n and we want n W Kn ! An1 ; c0n W Kn ! A0n such that
Un D dA n and f n D
n C dA0 c0n .
96 T. Kadeishvili
First, it is not hard to check that dA Un D 0 that is Un maps Kn to cycles Z.An2 /

An2 .
Then
!
X
n2 X
n2
f Un D f n1 dK C i ^ ni D f n1 dK C f i ^ f ni
iD2 iD2
X
n2
D
n1 dK C
i ^
ni D Un0 D dA
n ;
iD2
thus, f Un maps Kn to boundaries B.An2 / An2 . Since f W A ! A0 is a homology

isomorphism, there exist n W Kn ! An1 such that dA n D Un :
Now we must take care on the condition f n D
n . It is clear that dA f n D dA
n .
Let us denote by z0n D f n
n ; this is the homomorphism which maps Kn to
cycles Z.A0n1 /. Again, since f W A ! A0 is a homology isomorphism there exist
zn W Kn ! Z.An1 / and c0n W Kn ! A0n such that fzn D z0n dA0 c0n . Let us define
n D n zn . Then dA n D dA n and
f n Df n fzn D .
n C z0n / .z0n dA0 c0n / D
n C dA0 c0n :
Perturbing
by this c0n we obtain Fc0n
for which .Fc0n
/n D
n C dA0 c0n D f n.
5.2.3 A1 -Twisting Cochains
Now we want to replace in the definition of a twisting cochain a dg algebra .A; dA ; /

with an A1 -algebra .M; fmi g/, see [10, 13].
An A1 -twisting cochain we define as a homomorphism
W K ! M of degree
1 satisfying the condition
1
X
mk .

/rKk D
dK : (5.11)
kD1
Let T1 .K; M/ be the set of all A1 -twisting cochains.

We assume that our A1 -algebras .M; fmi g/ are connected, that is M D
fM1 ; M2 ; : : : g. In this case any A1 -twisting cochain
W K ! M has components

D
2 C
3 C C
n C W K ! M;
here
n W Kn ! Mn1 is the restriction of
on Kn . The condition (5.11) restricted
on Kn gives
X
int.n=2/
X
mi .
k1
ki /rKk D
n1 dK : (5.12)
iD1 k1 Cki Dn
5.2.3.1 A1 -Twisting Cochains and the Bar Construction
Because of the universal property (5.4) any homomorphism

W K ! M of degree
1 induces a graded coalgebra map, f
W K ! T c .sM/ (coextension of
) by
X
f
D .

/rKi :
i
If, in addition,
W K ! M is an A1 -twisting cochain, that is satisfies the
condition (5.11), then f
W K ! B.M/ is a chain map, i.e. is a map of dg coalgebras:
the projection of the condition dB f
D f
dK on M gives pdB f
D pf
dK and this is
exactly (5.11).
Conversely, any dg coalgebra map f W K ! BM is f
for
D p f W K ! BM !
M. In fact we have a bijection Mordgcoalg .K; BM/ $ T1 .K; M/.
5.2.3.2 Equivalence A1 -Twisting Cochains
Two A1 -twisting cochains

; W K ! M we call equivalent if there exists c W K !
M; deg c D 0, such that
X

D cdK C mk . . j-times/ c

/r k ; (5.13)
k;j
notation

c .
This notion of equivalence allows us to perturb A1 -twisting cochains. Let

D
2 C
3 C C
n C W K ! M
be an A1 -twisting cochain and lets take an arbitrary cochain cn W Kn ! Mn . Then

there exists a twisting cochain Fcn
D W K ! M such that

cn . Actually the
components of perturbed twisting cochain
F cn
D D 2 C 3 CC n C
can be solved from (5.13) inductively, and the solution in particular gives i D
i
for i < n and n D
n C m1 cn .
98 T. Kadeishvili
5.2.3.3 Equivalence A1 -Twisting Cochains and Homotopy of Induced

Maps
If

c , then f
and f are homotopic in the category of dg coalgebras: chain
homotopy D1 .c/ W K ! BM is given by
X
D1 .c/ D . . j-times/ c

/rKi ; (5.14)
i;j
which automatically is a f
f -coderivation, that is
rB D1 .c/ D . f D1 .c/ C D1 .c/ f

/rK ;
and it satisfies f
f D dB D1 .c/ C D1 .c/dK since the projection of this
condition on M gives pf
pf D pdB D1 .c/ C pD1 .c/dK and this is exactly
the condition (5.13).
5.2.3.4 Functor D1
We denote by D1 .K; M/ the factorset D1 .K; M/ D T1 .K;M/

. Thus we have a
bijection K; BM $ D.K; M/.
Assigning to an A1 -twisting cochain
W K ! M the dg coalgebra map f
W
K ! BM and having in mind that

c implies f

D1 .c/ f we obtain a
bijection D1 .K; A/ $ K; BM.
Suppose f D f fi g W .M; fmi g/ ! .M 0 ; fm0i g/ is a morphism of A1 -algebras and
W K ! M is an A1 -twisting cochain. Then it is possible to show that f .

/ W K !
M 0 given by
X
f .
/ D fi .

/rKi (5.15)
i
is an A1 -twisting cochain too. Actually this follows from bar construction interpre-
tation: the composition of dg coalgebra maps
B. f / f
W K ! BM ! BM 0
is the coextension of the twisting cochain pB. f / f

which is exactly f .
/.
Moreover, if

c , then f .
/
c0 f . / with c0 W K ! M 0 given by
X
c0 D fi . . j-times/ c

/rKi :
i;j
Again this follows from bar construction interpretation: If

c , then f
; f W
K ! BM are homotopic dg coalgebra maps, and the corresponding homotopy
D1 .c/ is given by (5.14). This implies that the compositions B. f / f

; B. f / f W
K ! BM 0 are homotopic by homotopy B. f / D1 .c/ W K ! M 0 and the projection
p B. f / D1 .c/ W K ! BM ! BM 0 ! M 0 is exactly c0 . This gives
f .
/ D pB. f /f

c0 pB. f /f D f . /:
Thus we have a map D1 . f / W D1 .K; M/ ! D1 .K; M 0 /.

The following theorem is an analog of Berikashviliss Theorem 2 for A1 -
algebras and was proved in [10].
Theorem 4 Let .K; dK ; rK / be a dg coalgebra with free Ki s and .M; fmi g/ be a
connected dg algebra. If f D f fi g W .M; fmi g/ ! .M 0 ; fm0i g/ is a weak equivalence
of A1 -algebras, then
D1 . f / W D1 .K; M/ ! D1 .K; M 0 /
is a bijection.
In fact this theorem means that K; BM ! K; BM 0 is a bijection. The theorem
consists of two parts:
Surjectivity Any A1 -twisting cochain
W K ! M 0 can be lifted to an A1 -twisting
cochain W K ! M so that

f .
0
Injectivity If
2 T1 .K; M/, then f
f 2 T1 .K; M 0 /.
5.2.3.5 Lifting of A1 -Twisting Cochains
Below well use the surjectivity part of this theorem whose proof we sketch here.
Theorem 5 Let .K; dK ; rK / be a dg coalgebra with free Ki s and let f W .M; fmi g/ !
.M 0 ; fm0i g/ be a weak equivalence of connected A1 -algebras (that is, M0 D M00 D
0). Then for an arbitrary A1 -twisting cochain

D
2 C
3 C C
n C W K ! M 0
there exists an A1 -twisting cochain
D 2 C 3 CC n C W K ! M
such that

f .
Proof Start with an A1 -twisting cochain

D
2 C
3 C C
n C W K ! M 0 :
The condition (5.11) gives m1

2 D 0, and since f W .M; m1 / ! .M 0 m01 / is homology
isomorphism then there exist 2 W K2 ! M1 and c02 W K2 ! M20 such that m1 2 D 0
100 T. Kadeishvili
and f 2 D
2 C m01 c02 . Perturbing
by this c02 we obtain Fc02
for which .Fc02
/2 D
2 C m01 c02 D f 2 .
Assume now that we already have 2 ; 3 ; : : : ; n1 which satisfy (5.11)
and (5.15) in appropriate dimensions, that is
X
int.k=2/
X
m1 k D k1 dK C mi . k1 ki /rK ;
k
k D 2; 3; : : : ; n 1
iD2 k1 Cki Dk
(5.16)
and
X
int.k=2/
X

k D fi . k1 ki /rK ;
i
k D 1; 2; : : : ; n 1: (5.17)
iD1 k1 CCki Dk
We need the next component n W Kn ! An1 satisfying the condition (5.16) for
kDn
X
int.n=2/
X
m1 n D n1 dK C mi . k1 ki /rK ;
k
(5.18)
iD2 k1 Cki Dn
and c0n W Kn ! A0n such that
X
int.n=2/
X
m1 c0n C
n D fi . k1 ki /rK :
i
(5.19)
iD1 k1 CCki Dn
Then perturbing
by c0n we obtain new
n for which the condition (5.17) will be
satisfied for k D n and this will complete the proof.
Let us put
P P
int.n=2/
Un D n1 dK C k1 Cki Dn mi . k1 ki /rK ;
k
iD2
P P
int.n=2/
Un0 D
n1 dK C k1 Cki Dn mi .
k1
ki /rKk
iD2
P
int.n=2/ P
Vn D fi . k1 ki /rK :
i
iD2 k1 CCki Dn
Then the needed conditions (5.18) and (5.19) look as
Un D m1 n; m01 c0n C
n D f1 n C Vn :
First, it is possible to check that m1 Un D 0 and f1 Un D Un0 C m01 Vn . Having in

mind that Un0 D m01
n the last condition means f1 Un D m01 .
n C Vn /. So Un is a map
to m1 -cycles
Un W Kn ! Z.Mn2 / Mn2
and f1 Un maps Kn to boundaries

0
f1 Un W Kn ! B.Mn2 / Mn2 :
Then since f1 W .M; m1 / ! .M 0 ; m01 / is a homology isomorphism, there exist n W

Kn ! Mn1 such that m1 n D Un : So our n satisfies (5.18). Now let us perturb
this
in order to catch the condition (5.19) too.
Using the above equality f1 Un D m01 .
n C Vn /, we have
m01 f1 n D f1 m1 n D f1 Un D m01 .
n C Vn /
i.e., m01 . f1 n .
n C Vn // D 0. Thus
z0n D . f1 n
0
.
n C Vn // W Kn ! Z.Mn1 /;
and again, since f1 W .M; m1 / ! .M 0 ; m01 / is a homology isomorphism there exist

zn W Kn ! Z.Mn1 / and c0n W Kn ! Mn0 such that z0n D f1 zn m01 c0n . Lets define
n D n zn . Then
f1 n D f1 n f1 zn D .z0n C
n C Vn / .z0n m01 c0n / D
n C Vn :
5.2.4 B1 -Algebras
The notion of B1 -algebra was introduced in [2, 8] as an additional structure on a

dg module .A; d/ which turns the tensor coalgebra T c .sA/ (i.e., the bar construction
BA) into a dg bialgebra. So it requires differential
e
d W BA ! BA;
which should be a coderivation with respect to standard coproduct of BA, and a

(maybe nonassociative) multiplication
W .BA; e
e d/ .BA; e
d/ ! .BA; e
d/
which should be a map of dg coalgebras, with 1 2 BA as a unit element.

102 T. Kadeishvili
It is mentioned above (see also, for example, [9, 12, 18]) that the specification of
such ed is equivalent to the specification on A of a structure of A1 -algebra, namely
of a sequence of operations fmi W i A ! A; i D 1; 2; 3; : : :g subject of Stasheffs
conditions (5.3).
As for the multiplication
e
W B.A; fmi g/ B.A; fmi g/ ! B.A; fmi g/
by definition of a dg bialgebra it must be a map of dg coalgebras. Consequently it is

uniquely determined by an A1 -twisting cochain, say
E; W B.A; fmi g/ B.A; fmi g/ ! .A; fmi g/: (5.20)
In turn such a twisting cochain is represented by a sequence of operations
fEpq W Ap Aq ! A; p; q D 0; 1; 2; 3; : : :g
satisfying certain coherency conditions together with A1 operations fmi g which

follow from the Browns condition (5.11).
So a B1 -algebra is a graded module equipped with two sets of algebraic
multioperations .A; fmi g; fEp;q g/.
The particular case of a B1 -algebra of type m>3 D 0 is called a Hirsch
algebra, and the particular case of a Hirsch algebra with E>1;q D 0 satisfying
certain additional conditions is called homotopy Gerstenhaber algebra, see below.
We address the reader for more explanations of these structures to [13, 15]. In fact
the present description is enough for this paper.
5.2.4.1 Homotopy G-Algebras
A particular case of B1 -algebra is the homotopy G-algebra (hGa in short), the

notion was introduced in [6, 20] by M. Gerstenhaber and A. Voronov. This is
a dg object whose homology carries a Gerstenhaber algebra structure, that is a
commutative multiplication and a Lie bracket well connected to each other. From
another point of view a hGa is a dg algebra with good ^1 product. In fact the
mentioned Lie bracket is induced by the commutator of ^1 .
Definition A homotopy G-algebra is defined as a dg algebra .A; d; / with a given
sequence of operations
E1;k W A .Ak / ! A; k D 0; 1; 2; 3; : : :
(the value of the operation E1;k on a b1 bk 2 A .A A/ we write

as E1;k .aI b1 ; : : : ; bk /) which satisfies the conditions
E1;0 D id; (5.21)

X
dE1;k .aI b1 ; : : : ; bk / C E1;k .daI b1 ; : : : ; bk / C E1;k .aI b1 ; : : : ; dbi ; : : : ; bk /
i
D b1 E1;k1 .aI b2 ; : : : ; bk / C E1;k1 .aI b1 ; : : : ; bk1 / bk

X
C E1;k1 .aI b1 ; : : : ; bi biC1 ; : : : ; bk /; (5.22)
i
E1;k .a1 a2 I b1 ; : : : ; bk /
D a1 E1;k .a2 I b1 ; : : : ; bk / C E1;k .a1 I b1 ; : : : ; bk / a2
X
k1
C E1;p .a1 I b1 ; : : : ; bp / E1;mp .a2 I bpC1 ; : : : ; bk /; (5.23)
pD1
E1;n .E1;m .aI b1 ; : : : ; bm /I c1 ; : : : ; cn /

X
D
06i1 6j1 6:::6im 6jm 6n
E1;n. j1 CCjm /C.i1 CCim /Cm .aI c1 ; : : : ; ci1 ; E1;j1 i1 .b1 I ci1 C1 ; : : : ; cj1 /;
cj1 C1 ; : : : ; ci2 ; E1;j2 i2 .b2 I ci2 C1 ; : : : ; cj2 /; cj2 C1 ; : : : ; cim ;
E1;jm im .bm I cim C1 ; : : : ; cjm /; cjm C1 ; : : : ; cn /: (5.24)
Let us present these conditions in low dimensions.

The condition (5.22) for k D 1 looks as
dE1;1 .aI b/ C E1;1 .daI b/ C E1;1 .aI db/ D a b C b a: (5.25)
So the operation E1;1 is a sort of ^1 product: it is the chain homotopy which

measures the noncommutativity of A. Below we denote E1;1 .aI b/ D a ^1 b.
.a b/ ^1 c C a .b ^1 c/ C .a ^1 c/ b D 0; (5.26)
this means that the operation E1;1 D^1 satisfies the left Hirsch formula.
dE1;2 .aI b; c/ C E1;2 .daI b; c/ C E1;2 .aI db; c/ C E1;2 .aI b; dc/
D a ^1 .b c/ C .a ^1 b/ c C b .a ^1 c/; (5.27)
104 T. Kadeishvili
this means that this ^1 satisfies the right Hirsch formula just up to homotopy and
the appropriate homotopy is the operation E1;2 .
The condition (5.24) for n D m D 2 looks as
.a ^1 b/ ^1 c C a ^1 .b ^1 c/ D E1;2 .aI b; c/ C E1;2 .aI c; b/; (5.28)
this means that the same operation E1;2 measures also the deviation from the
associativity of the operation E1;1 D^1 .
5.2.4.2 hGa as a B.1/-Algebra
Here we show that a hGa structure on A is a particular B.1/-algebra structure: it

induces on B.A/ D .T c .sA/; dB / an associative multiplication but does not change
the differential dB (see [5, 8, 13, 15]).
Let us extend our sequence fE1;k ; k D 0; 1; 2; : : :g to the sequence
fEp;q W .Ap / .Aq / ! A; p; q D 0; 1; : : :g
adding
E0;1 D id; E0;q>1 D 0; E1;0 D id; Ep>1;0 D 0; (5.29)
and Ep>1;q D 0.
This sequence defines a map E W B.A/ B.A/ ! A by E.a1 ; : : : ; am
b1 ; : : : ; bn / D Ep;q .a1 ; : : : ; am I b1 ; : : : ; bn /. The conditions (5.22) and (5.23) mean
exactly dE C E.dB id C id dB / D E ^ E, i.e. E is a twisting cochain. Thus its
coextension is a dg coalgebra map, i.e. a correct multiplication
E W B.A/ B.A/ ! B.A/:
The condition (5.24) can be rewritten as E.E id/ D E.id E / and the
equality of these twisting cochains implies the equality of their coextensions, that is
E .E id D E .id E / so this condition means that the multiplication E is
strictly associative.
And the condition (5.29) implies that D 1 2 B.A/ is the unit for this
multiplication.
Finally we obtained that .B.A/; dB ; ; E / is a dg bialgebra, thus a hGa is a
B.1/-algebra with strictly associative multiplication.
5.2.4.3 Three Examples of hGa-s
There are three remarkable examples of homotopy G-algebras, see [16].

The first one is the cochain complex of 1-reduced simplicial set C .X/. The
operations E1;k here are dual to cooperations defined by Baues in [2], and the starting
operation E1;1 is the classical Steenrods ^1 product.
The second example is the Hochschild cochain complex C .U; U/ of an associa-
tive algebra U. The operations E1;k here were defined in [11] with the purpose of
describing A.1/-algebras in terms of Hochschild cochains although the properties
of those operations which were used as defining ones for the notion of homotopy
G-algebra in [6] did not appear there. These operations were defined also in [7].
Again the starting operation E1;1 is the classical Gerstenhabers circle product which
is sort of ^1 -product in the Hochschild complex. These operations, denoted as
E1;k .aI ; b1 ; : : : ; bk / D afb1 ; : : : ; bk g and called braces, were used in [17] in the
proof of Delignes hypothesis.
The third example is the cobar construction C of a dg-bialgebra C. The cobar
construction C of a DG-coalgebra .C; d W C ! C; W C ! C C/ is,
by definition, a DG-algebra. Suppose now that C is additionally equipped with a
multiplication W C C ! C turning .C; d; ; / into a DG-bialgebra. How
this multiplication reflects on the cobar construction C? There arises natural
hGa structure, the operations E1;k are constructed in [14]. And again the starting
operation E1;1 is classical, it is Adamss ^1 -product defined for C in [1].
5.3 B1 -Algebra Structure in Homology of a hGa
Here we turn to the problem of determining correct multiplication on the bar

construction B.H.A/; fmi g/.
Suppose that .A; d; ; fE1;k g/ is a hGa. Note that by Sect. 5.2.4.2 the sequence of
operations fE1;k g determines a twisting cochain E W BA BA ! A.
By the Minimality Theorem [9] there exists on H.A/ a structure of minimal A1 -
algebra .H.A/; fmi g/ and a weak equivalence of A1 -algebras
f D f fi g W .H.A/; fmi g/ ! .A; fm1 D d; m2 D ; m3 D 0; m4 D 0; : : :g/:

(5.30)
This weak equivalence induces a weak equivalence of dg coalgebras
B. f / W B.H.A/; fmi g/ ! BA:
Composing the tensor product B. f /B. f / with the twisting cochain E W BABA !
A determined by the given hGa structure fE1;k g we obtain a twisting cochain
E .B. f / B. f // W B.H.A/; fmi g/ B.H.A/; fmi g/ ! BA BA ! A: (5.31)

106 T. Kadeishvili
Now, using the lifting Theorem 5, let us lift this twisting cochain along the weak
equivalence of A1 -algebras (5.30). We obtain an A1 -twisting cochain
E; W B.H.A/; fmi g/ B.H.A/; fmi g/ ! .H.A/; fmi g/
which in turn defines multiplication on B.H.A/; fmi g/, a dg coalgebra map
E; W B.H.A/; fmi g/ B.H.A/; fmi g/ ! B.H.A/; fmi g/
i.e., a B1 -algebra structure on .H.A/; fmi g/.

These twisting cochains can be observed from the diagram
E;
B.H.A/; fmi g/ B.H.A/; fmi g/ ! .H.A/; fmi g//
B. f / B. f / # #f
E
BA BA ! A:
This diagram does not commute but the twisting cochains f E; and E .B. f /
B. f // are equivalent. Consequently the following diagram of induced dg coalgebra
maps commutes up to homotopy
E;
B.H.A/; fmi g/ B.H.A/; fmig/ ! B.H.A/; fmi g//
B. f / B. f / # # B. f /
E
BA BA ! BA:
Thus the dg coalgebra weak equivalence
B. f / W B.H.A/; fmi g; fEp;q g/ ! B.A; d; ; fE1;k g/
is multiplicative up to homotopy. Consequently it induces multiplicative isomor-

phism of their homology algebras
H.B.H.A/; fmi g; fEp;q g// H.B.A; d; ; fE1;k g//:
Finally we have the

Theorem 6 Let .A; d; ; fE1;k g/ be a hGa. Then on its homology H.A/ there
exists a structure of B1 -algebra .H.A/; fmi g; fEp;q g/ such that homology algebras
H.B.H.A/; fmig; fEp;q g// and H.B.A; d; ; fE1;k g// are isomorphic.
Remark For a hGa .A; d; ; fE1;k g/ the twisting cochain E W BA BA ! A satisfies
additional conditions (5.24) which guarantee that the induced multiplication on BA
is associative. The twisting cochain E; we have obtained satisfies only Browns
condition, but not that condition for associativity, so the obtained multiplication
E; W B.H.A/; fmi g/ B.H.A/; fmi g/ ! B.H.A/; fmi g/

is a chain map, but nonassociative generally. Thus the bar construction

B.H.A/; fmi g/ is a nonassociative bialgebra. We expect that this nonassociative
multiplication will be a part of certain A1 algebra structure on B.H.A/; fmi g/
which will allow to iterate the process.
5.4 Cohomology Algebra of the Loop Space
Here we turn to the main goal of this paper, cohomology algebra H .X/.
Start with a topological space X and take .A; d; / D .C .X/; d; ^/, the cochain
complex of X. Then, by Adams [1] we have that H.BA/ D H.BC .X// and H .X/
are isomorphic as graded modules.
As for the multiplicative structure of H .X/, in [2] Baues constructed a
sequence of multioperations
E1;k W C .X/ C .X/k ! C .X/; k D 1; 2; 3; : : :
turning .C .X/; d; ^; fE1;k g/ into a hGa, which determines on the bar construction
BA D BC .X/ correct multiplication, so that we have isomorphism of graded
algebras
H.B.C .X/; d; ^; fE1;k g// H .X/:
Now we are going to transfer these structures to cohomology level, i.e. from
C .X/ to H .X/.
Firstly, according to minimality theorem [9], on the homology of dg algebra
.C .X/; d; ^/ appears a structure of minimal A1 -algebra .H .X/; fmi g/ and a weak
equivalence of A1 -algebras
f fi g W .H .X/; fmi g/ ! .C .X/; fm1 D d; m2 D^; m3 D 0; : : : g/;
which induces weak equivalences of their bar constructions and isomorphism of

graded modules
H.B.H .X/; fmi g// H.BC .X// H .X/:
Thus this object .H .X/; fmi g/, which is called cohomology A1 -algebra of X, deter-
mines cohomology modules of the loop space H .X/. But not the cohomology
algebra H .X/.
Now, to obtain correct multiplication on B.H .X/; fmi g/ we use the above
theorem, which for .A; d; / D .C .X/; d; ^/ gives
Theorem 7 Let .C .X/; d; ^; fE1;k g/ be a hGa structure on the cochain complex
of a topological space X equipped with Bauess operations fE1;k g. Then on there
108 T. Kadeishvili
exists on H .X/ structure of B1 -algebra .H .X/; fmi g; fEp;q g/ which determines

cohomology algebra of the loop space, that is there exists isomorphism of graded
algebras
H.B.H .X/; fmi g; fEp;q g// H .X/:
Acknowledgements The author was supported by the European Unions Seventh Framework
Programme (FP7/20072013) under grant agreement no. 317721.
References
1. J. Adams, On the non-existence of elements of Hpof invariant one. Ann. Math. 72, 20104
(1960)
2. H. Baues, The double bar and cobar construction. Compos. Math. 43, 331341 (1981)
3. N. Berikashvili, On the differentials of spectral sequence. Proc. Tbil. Math. Inst. 51, 1105
(1976)
4. E. Brown, Twisted tensor products. Ann. Math. 69, 223246 (1959)
5. M. Franz, Tensor product of homotopy Gerstenhaber algebras (2011). ArXiv:1009.1116v2
[math.AT]
6. M. Gerstenhaber, A. Voronov, Higher-order operations on the Hochschild complex. Funct.
Anal. i Prilojen. 29(1), 16 (1995)
7. E. Getzler, Cartan homotopy formulas and the Gauss-Manin connection in cyclic homology.
Isr. Math. Conf. Proc. 7, 6578 (1993)
8. E. Getzler, J.D. Jones, Operads, homotopy algebra, and iterated integrals for double loop
spaces. Preprint (1994)
9. T. Kadeishvili, On the homology theory of fibrations. Russ. Math. Surv. 35(3), 231238 (1980)
10. T. Kadeishvili, Predifferential of a fiber bundle. Russ. Math. Surv. 41(6), 135147 (1986)
11. T. Kadeishvili, The A.1/-algebra structure and cohomology of Hochschild and Harrison.
Proc. Tbil. Math. Inst. 91, 1927 (1988)
12. T. Kadeishvili, A.1/ -algebra structure in cohomology and rational homotopy type. Proc. Tbil.
Mat. Inst. 107, 194 (1993)
13. T. Kadeishvili, Measuring the noncommutativity of DG-algebras. J. Math. Sci. 119(4), 494
512 (2004)
14. T. Kadeishvili, On the cobar construction of a bialgebra. Homol. Homotopy Appl. 7(2), 109
122 (2005). Preprint at math.AT/0406502
15. T. Kadeishvili, Twisting elements in homotopy G-algebras, in Higher Structures in Geometry
and Physics. Honor of Murray Gerstenhaber and Jim Stasheff. Progress in Mathematics,
vol. 287 (Birkhauser, Boston, 2011), pp. 181200. arXiv:math/0709.3130
16. T. Kadeishvili, Homotopy Gerstenhaber algebras: examples and applications. J. Math. Sci.
195(4), 455459 (2013)
17. J. McLure, J. Smith, A solution of Delignes conjecture. Preprint (1999). math.QA/9910126
18. V. Smirnov, Homology of fiber spaces. Russ. Math. Surv. 35(3), 294298 (1980)
19. J.D. Stasheff, Homotopy associativity of H-spaces. I, II. Trans. Am. Math. Soc. 108, 275312
(1963)
20. A. Voronov, Homotopy Gerstenhaber algebras (1999). QA/9908040
Chapter 6
Half-Automorphisms of CayleyDickson Loops
Maria de Lourdes Merlini-Giuliani, Peter Plaumann, and Liudmila Sabinina
To the memory of Karl Strambach
In the year 1957 W.R. Scott introduced in [28] the concept of a half-homomorphism
between semi-groups. He calls a mapping ' W S1 ! S2 between semi-groups S1 and
S2 a half-homomorphism if the alternative
8x; y 2 S1 W '.xy/ D '.x/'. y/ or '.xy/ D '. y/'.x/ (6.1)
holds and proves the following

Theorem Every half-isomorphism of a cancellation semi-group S1 into a cancella-
tion semi-group S2 is either an isomorphism or an anti-isomorphism.
We call a half-isomorphism proper if it is neither an isomorphism nor an anti-
isomorphism. Then one finds in Scotts paper the following result which in this
article we call Scotts Theorem.
Theorem There is no proper half-homomorphism from a group G1 into a group G2 .
In the article of Scott references are given to Huas Theorem [16] and to Jordan
homomorphisms [17] which give a hint where the origins of the notion of a half-
automorphism are to be searched (see also [26]).
At the end of his short article Scott gives an example of a loop S of order 8
which possesses a proper half-automorphism. Recently a systematic research on the
M.d.L. Merlini-Giuliani
Universidade Federal do ABC, Rua Santa Adlia 166, Santo Andr, SP 09210-170, Brazil
e-mail: maria.giuliani@ufabc.edu.br
P. Plaumann
Universidad Autnoma Benito Jurez de Oaxaca, Oaxaca de Jurez, Mexico
e-mail: peter.plaumann@mi.uni-erlangen.de
L. Sabinina ()
Universidad Autnoma del Estado de Morelos, Av. Universidad 1001, Cuernavaca 62209, Mexico
e-mail: liudmila@uaem.mx

110 M.d.L. Merlini-Giuliani et al.
existence of proper half-automorphisms of Moufang loops has begun (see [9, 10,
12, 13]). In a new manuscript [19] Kinyon et al. have shown that Scotts Theorem
holds for a class of Moufang loops which contains all automorphic Moufang loops.
Petr Vojtechovsky gave us an example of a loop V in which the inversion map
J satisfying xJ x D 1 D xxJ for all x 2 V exists (Vojtechovsky, July 2015, private
communication). The Cayley table of V is
1 2 3 4 5 6 7 8
2 1 4 7 8 5 3 6
3 5 8 1 6 4 2 7
4 3 1 8 2 7 6 5
5 4 2 6 7 8 1 3
6 7 5 3 4 1 8 2
7 8 6 2 1 3 5 4
8 6 7 5 3 2 4 1
It is easy to check that the inversion mapping J is a proper half-automorphism of

the loop V. Note that V is not a diassociative loop since in diassociative loops the
inversion always is an anti-automorphism.
In this note we exclusively study half-automorphisms of diassociative 2-loops.
A special type of half-automorphisms, which we call elementary maps [see (6.5)],
plays a particular role. We will use them as a part of a generating system of the
group of all half-automorphisms of the loops considered, if this is possible.
Sections 6.1 and 6.3 of this article are used to describe topics which are
background knowledge. In Sect. 6.1 we collect fundamentals of the Theory of Loops
which we need. Would this be an article on Group Theory, this part would not be
necessary since these facts belong to common knowledge for the variety of groups.
We try to write there only things which are necessary for our considerations. In
Sect. 6.3 we present the well-known so-called CayleyDickson construction of a
sequence of algebras An of dimension 2n ; n > 0. We restrict ourselves to the
classical case of algebras over the field of real numbers where the multiplication
is given by (6.9). The well-known generalizations of this definition we do not need
here. The material in Sect. 6.3 is included in our text mostly to fix notations in such
a way that for the iterated doubling process one obtains canonical bases Bn for the
algebras An which allow to assume that An < AnC1 for all n.
In Sect. 6.4 we turn our view to the fact that sets Cn D Bn [ .Bn / are
closed under multiplication in An . Hence they form an infinite sequence of loops
Cn which are called CayleyDickson loops. Note that this does not contradict the
fact that CayleyDickson algebras of dimension > 24 have zero divisors. Cayley
Dickson loops in particular the octonion loop C4 and the sedenion loop C5 have
been considered by several authors. We mention here publications by Kivunge and
Smith [21], Kinyon et al. [18, Sect. 7], Cawagas et al. [5] and Kirshtein [20].
Already before, in Sect. 6.2, we had presented the octonion loop C4 of order 16
which is small enough to be treated by direct computation. Theorem 6.2.2, which
6 Half-Automorphisms of CayleyDickson Loops 111
includes the fact that C4 is a Moufang loop, allows us to describe the group of all
half-automorphisms of C4 (Theorem 6.2.5). In the proof of this theorem mappings
Cn ! Cn which we call elementary mappings play a particular role.
For all CayleyDickson loops we have with Theorem 6.4.3 a result analogous
to Theorem 6.2.2. This allowed us to prove the existence of elementary half-
automorphisms for every CayleyDickson loop in Theorem 6.4.5. We could,
however, not determine the full structure of the group of half-automorphisms for
all CayleyDickson loops.
In Sect. 6.4 we also discuss the existence of proper half-automorphisms of code
loops (see [11]). Since the octonion loop is an example of a code loop (see [14, 15])
the elementary half-automorphisms should play a great role in the class of code
loops.
6.1 Basic Facts About Loops
The fundamental definitions and facts from the Theory of Loops are found in the
monographs of Bruck [4] and of Pflugfelder [23]. We give a short description of
facts that we need. We call a set .S; / with a multiplication
.a; b/ 7! a b W S S ! S
a magma. Consider the left translations
La D .x 7! a x/ W S ! S
and the right translations
Rb D .x 7! x b/ W S ! S:
Definition 6.1.1 A loop .L; / is a magma such that

(1) There is an identity element 1 2 L such that 1 x D x D x 1 for all x 2 L.
(2) All mappings La ; Rb are bijections.
Given a loop L put x=y WD R1 1
y .x/ and ynx WD Ly .x/. Then one has a universal
algebra .L; ; =; n; 1/ of type .3; 0; 1/.
Definition 6.1.2 A loop L is a Moufang loop if it satisfies the equivalent identities
..xy/x/z D x. y.xz// (6.2)

..xy/z/y D x. y.zy// (6.3)
.xy/.zx/ D .x. yz//x (6.4)
A loop L is called diassociative if all subloops hx; yi generated by elements x; y 2 L

are groups. Similarly a loop Q is called power-associative if all subloops of the form
hxi; x 2 Q, are groups. In [23] one finds the
Proposition 6.1.3 All Moufang loops are diassociative.
Let L be a diassociative loop. Then hxi is a group. Hence there is a unique element
x0 2 L such that xx0 D 1 D x0 x. Considering the group hx; yi one sees L1 y D Ly0
1 0 0
and Ry D Ry0 . It follows that x=y D xy and ynx D y x. Thus a diassociative loop
can be considered as a universal algebra of type .1; 1; 1/ as one usually does with
groups. In this article we shall do this permanently.
Definition 6.1.4 A loop Q is called a C-loop if the identity
x. y yz/ D .xy y/z
holds in Q.
Information about C-loops one finds in [24] and [18]. Remembering that a loop
is called flexible if the identity x. yx/ D .xy/x holds one finds there the
Proposition 6.1.5 A flexible C-loop is diassociative.
For a loop Q one considers the multiplication groups
LMult.Q/ D hLa j a 2 Qi (left multiplication group of Q),

RMult.Q/ D hRb j b 2 Qi (right multiplication group of Q),
Mult.Q/ D hLMult.Q/ [ RMult.Q/i (multiplication group of Q).
These groups act transitively on Q and give us reason to consider the inner
mapping groups
LInt.Q/ D f 2 LMult.Q/ j .1/ D 1g (left inner mapping group of Q),

RInt.Q/ D f 2 RMult.Q/ j .1/ D 1g (right inner mapping group of Q),
Int.Q/ D f 2 Mult.Q/ j .1/ D 1g (inner mapping group of Q).
Definition 6.1.6 A subloop of a loop Q is normal in Q if and only if it is invariant

under the group Int.Q/.
We use the notation N G Q to indicate that N is a normal subloop of Q.
Definition 6.1.7 Let Q be a loop and let be a loop homomorphism defined on Q.
Then the set
ker D fx 2 Q j .x/ D 1g
is called the kernel of .

Proposition 6.1.8 A subloop of a loop is normal if and only if it is the kernel of a

homomorphism.
Definitions 6.1.6 and 6.1.7 together with Proposition 6.1.8 are the fundament for
the homomorphism theory of loops which resembles the homomorphism theory for
groups (see [1, 3]).
If the loop Q is a group, then the elements of Int.Q/ are just the inner
automorphisms of Q. For Moufang loops in general the inner mappings are pseudo-
automorphisms (see [23]).
Definition 6.1.9 Let Q be a diassociative loop and let ' W Q ! Q be a bijective
mapping. If
.SA/ '. yxy/ D '. y/'.x/'. y/
holds for all x; y 2 Q, then ' is called a semi-automorphism of Q.

All inner mappings of a Moufang loop are semi-automorphisms (see [23]).
It is easy to see that for a diassociative loop all half-automorphisms are semi-
automorphisms as well (see also [9]).
In particular, characteristic subloops, that is subloops which are invariant under
all automorphisms, in general not are normal.
We denote the group of automorphisms of a loop L by Aut.L/.
Definition 6.1.10 A loop L is called left (right) automorphic if all left (right)
inner mappings of L are automorphisms. If Int.L/ Aut.L/, then L is called
automorphic.
In an automorphic loop all characteristic subloops are normal, a fact which makes
many considerations easier.
We now define some important characteristic respectively normal characteristic
subloops of a given loop L. Detailed information is found in [4] and [23].
1. The commutator subloop of a loop L.
For x; y 2 L the commutator is x; y D xynyx
L; Lnormal subloop generated by all commutators.
Note that L; L is the smallest normal subloop N of L such that L=N is a
commutative loop and that xy=yx 2 L; L for all x; y 2 L.
2. The associator subloop of a loop L.
For x; y; z 2 L the associator is .x; y; z/ D x. yz/n.xy/z
.L; L; L/normal subloop generated by all associators.
Note that .L; L; L/ is the smallest normal subloop N of L such that L=N is a group
and that .xy/znx. yz/ 2 .L; L; L/ for all x; y; z 2 L.
3. We call .L/ D .L; L; L/ \ L; L the radical-defect.
4. The nuclei of a loop.
The left nucleus is Nl D fu 2 L j .u; x; y/ D 1 for all x; y 2 Lg.
Analogously one defines the middle nucleus Nm and the right nucleus Nr .
The nucleus of L is N.L/ D Nl \ Nm \ Nr
5. The commutant of a loop L is C.L/ D fc 2 L j .8x 2 L/ c; x D 1g
6. The center of a loop L
zLnormal closure of the subloop C.L/ \ N.L/.
7. The Frattini subloop of a loop L.

.L/normal closure of the intersection of all maximal subloops.
Note 1 For a Moufang loop M one can show that the center and N.M/ D Nl .M/ D
Nm .M/ D Nr .M/ are normal [4]. The same holds for the commutant (see [8]).
At the end of this section we cannot resist the temptation to cite R. Baer who
writes in the article [2]: It is not the object of this address to introduce you to new
theories or to tell of great discoveries.
6.2 Some Small Moufang Loops
1. Chein Loops In 1974 O. Chein found a construction to obtain from a given group
G a Moufang loop M.G; 2/. He used this construction efficiently in the classification
of finite Moufang loops of small order. For a finite group G and a group of order
two hui such that u G define on the set M.G; 2// D G hui a multiplication by
.g; 1/.h; 1/ D .gh; 1/

.g; u/.h; 1/ D .gh1 ; u/
.g; 1/.h; u/ D .hg; u/
.g; u/.h; u/ D .h1 g; 1/
Theorem 6.2.1 (O. Chein [6]) If G is a finite group, then M.G; 2/ is a Moufang
loop and G h1i is a normal subloop of M.G; 2/. Furthermore, M.G; 2/ is a group
if and only if the group G is abelian. q.e.d.
In [10] one finds much information about half-automorphisms of Chein loops.
2. Moufang Loops of Order 16 The LOOPS package of GAP1 contains a library
of loops of small order. There one finds 5 Moufang loops of order 16 which are not
groups.
MoufangLoop(16,1) - the Chein double of the dihedral group D8

MoufangLoop(16,2) - the Chein double of the quaternion group Q8
MoufangLoop(16,3) - the octonion loop o
MoufangLoop(16,4) -
MoufangLoop(16,5) -
For 4 of these Moufang, the loops MoufangLoop(16,1), MoufangLoop

(16,2), MoufangLoop(16,4), and MoufangLoop(16,5), we have no
satisfactory theory of the group of half-automorphisms. In [10, Theorem 7] it is
1
The GAPGroups, Algorithms, and Programming, Version 4.4.10.
shown that a Moufang loop of even order which admits a proper half-automorphism
contains elements of order 4, but this does not help us. For MoufangLoop(16,2)
in [10, Example 8] the existence of a proper half-automorphism is shown.
We can, however, easily determine the full group of half-automorphisms of
MoufangLoop(16,3) in detail.
3. The Octonion Loop o. The octonion loop MoufangLoop(16,3) carries its
name since it is closely connected with the classical alternative division algebra
O of dimension 8 over the field R of real numbers. To obtain this finite loop one
considers the standard basis B D .1 D b1 ; : : : ; b8 / of the vector space O over R.
Put L D fb1 ; : : : ; b8 g. Then L is closed under the octonion multiplication; below
one sees the multiplication table of L in some ordering of B.
Studying Table 6.1 it is easy to get the following
Theorem 6.2.2 In the octonion loop o the following statements are true:
(1) o is a loop of order 16
(2) o is a Moufang loop.
(3) In o there are one element z D 1 of order 2 and 14 elements of order 4.
(4) zx D xz for all x 2 o.
(5) If H is a subloop of o, then z 2 H.
(6) Z is a normal subloop of o which coincides with the nucleus, the center, the
associator subloop, the commutator subloop and the Frattini subloop of o.
(7) o=Z is an elementary abelian group of order 8. q.e.d.
Some details of the following construction hold for arbitrary diassociative loops.
Table 6.1 Cayley table of octonion loop

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
2 2 4 8 6 3 1 5 7 14 9 16 10 11 12 13 15
3 3 5 4 7 6 8 1 2 15 13 9 11 14 16 12 10
4 4 6 7 1 8 2 3 5 12 14 15 9 16 10 11 13
5 5 7 2 8 4 3 6 1 13 11 14 16 12 15 10 9
6 6 1 5 2 7 4 8 3 10 12 13 14 15 9 16 11
7 7 8 1 3 2 5 4 6 11 16 12 15 10 13 9 14
8 8 3 6 5 1 7 2 4 16 15 10 13 9 11 14 12
9 9 10 11 12 16 14 15 13 4 6 7 1 5 2 3 8
10 10 12 16 14 15 9 13 11 2 4 5 6 3 1 8 7
11 11 13 12 15 10 16 9 14 3 8 4 7 6 5 1 2
12 12 14 15 9 13 10 11 16 1 2 3 4 8 6 7 5
13 13 15 10 16 9 11 14 12 8 7 2 5 4 3 6 1
14 14 9 13 10 11 12 16 15 6 1 8 2 7 4 5 3
15 15 16 9 11 14 13 12 10 7 5 1 3 2 8 4 6
16 16 11 14 13 12 15 10 9 5 3 6 8 1 7 2 4
Definition 6.2.3 Let Q be a diassociative loop and let c be an element of Q. Then

the mapping c W Q ! Q defined by

x1 if x 2 fc; c1 g

c .x/ D (6.5)
x if x fc; c1 g
is called an elementary mapping.

Obviously the following proposition holds.
Proposition 6.2.4 For a diassociative loop Q of exponent 4 put
X D fx 2 Q j x2 1g:
Then
(1) If c 2 X, then the mapping c is an involution.
(2) For a b 2 X one has a D b if and only if ab D 1.
(3) For a; b 2 X the group ha ; b i is abelian. q.e.d.
In o there are 14 elements of order 4 which give us precisely 7 different sets
fx; x1 g with 2 elements. The set I of these 2-element sets consists of
f2; 6g; f3; 7g; f8; 5g; f9; 12g; f14; 10g; f15; 11g; f16; 13g: (6.6)
To study one typical example choose the pair f8; 5g 2 I and consider the
transposition D .5; 8/ in the symmetric group S14 .
One computes
.2 7/ D .5/ D 8 D .7/ .2/ .2/ .7/ D 5;
.3 9/ D .15/ D 15 D .3/ .9/ .9/ .3/ D 11.
Thus is a proper half-automorphism of o.
Because of the symmetry of o every transposition of a pair from the list (6.6)
is a proper half-automorphism. One can verify this fact directly using Table 6.1.
This way one obtains seven commuting proper half-automorphisms of o of order 2.
Denote these seven transpositions by 1 ; : : : ; 7 . They are elementary mappings and
generate a group D .o/ which is an elementary abelian 2-group of order 27 .
Thus an inspection of Table 6.1 gives us the
Theorem 6.2.5 Let c be an element of order 4 of the octonion loop o. Then the
elementary map c is a proper half-automorphism of o. q.e.d.
Theorem 6.2.6 Let o be the octonion loop and be the group generated by
all elementary maps. Then every half-automorphism of o is the product of an
automorphism of o and an element 2 .
Proof Suppose that ' W o ! o is a proper half-automorphism. Since o is a Moufang
loop by Gagola III and Merlini Giuliani [9] thus there exists a triple .x; y; z/ such
that ' restricted to hx; yi is an automorphism, ' restricted to hx; zi is an anti-
automorphism, and x; y 1; x; z 1. Since ' is proper, the loop hx; y; zi cannot
be a group by the theorem of Scott. Hence hx; y; zi D o. In what follows we always

will consider these generators x; y; z of the loop o: In particular yz zy: In o there
are 16 elements:
f1; x; y; z; xy; xz; yz; x. yz/g:
Note that for any half-automorphism of o and for any t 2 o one has:
.1/ D 1; .t/ D .t/
since 1 is the unique element of order 2 in o and central. Note also that the
following relations hold:
xy D yx; xz D zx; yz D zy;
x. yz/ D z.xy/ D y.zx/ D . yz/x D .xy/z D .zx/y:
In order to understand the behavior of proper half-automorphism ' on o it is

enough to consider only the set S D fx; y; z; xy; xz; yz; x. yz/g of seven elements and
to analyze the behavior of ' on all possible subgroups generated by two elements of
S. There are 21 different pairs, some of them generate the same subgroups. Indeed:
(i) hx; yi D hx; xyi D hxy; yi

(ii) hx; zi D hx; xzi D hxz; zi
(iii) hy; zi D hy; yzi D h yz; zi
(iv) hx; yzi D hx; x. yz/i D h yz; x. yz/i
(v) hy; xzi D hy; x. yz/i D hxz; x. yz/i
(vi) hz; yxi D hz; x. yz/i D h yx; x. yz/i
(vii) hxy; yzi D hxy; xzi D hxz; yzi
One obtains .vii/ using the fact that u2 D u; v D .u; v; t/ D 1 and the
Moufang identity in o in the following form. For any elements u; v; t 2 S we have:
ut vt D tu vt D .t.uv/t/ D t2 .uv/ D .uv/ (6.7)
Thus we get seven different 2-generated subgroup of o, every one of them is

isomorphic to Q8 : For example, hx; yi D f1; x; y; xyg:
There are two possibilities: ' restricted to h y; zi is an automorphism or '
restricted to h y; zi is an antiautomorphism.
Case 1 ' restricted to h y; zi is an automorphism.
Consider the mapping ' xz D '1 :
We see that '1 is an automorphism restricted to the following three subgroups:
hx; yi hx; zi hz; yi
Let us analyze the behavior of '1 restricted to hxy; yzi:

By (6.7) one has
'1 .xy yz/ D '1 . yx zy/ D '1 . y.xz/y/ D '1 .xz/ D '1 .x/'1 .z/:
Suppose that '1 restricted to hxy; yzi is an anti-automorphism.

Then we get
'1 .xy yz/ D '1 . yz/ '1 .xy/ D .'1 . y/'1 .z// .'1 .x/'1 . y//
D '1 .z/'1 .x/ '1 .x/'1 .z/:
This is a contradiction. Thus '1 is an automorphism restricted to the group hxy; yzi:
On the subgroup hx; yzi the map '1 can behave as an automorphism or as an
anti-automorphism. We will show that in both cases the behavior of '1 restricted to
subgroup hx; yzi determines the behavior of '1 restricted to subgroups h y; xzi and
hz; yxi . In other words there are two possibilities: the first one is the case that '1 is
an automorphism restricted to all three subgroups
hx; yzi; h y; xzi; hz; yxi
and the second one is the case that '1 is an anti-automorphism restricted to these
subgroups.
Suppose that '1 restricted to hx; yzi is an automorphism. Then
'1 .x. yz// D '1 .x/.'1 . y/'1 .z//:
Let us analyze the behavior of '1 restricted to h y; xzi:
'1 . y.xz// D '1 .x. yz// D '1 .x/.'1 . y/'1 .z//:
Assume that '1 is an anti-automorphism restricted to h y; xzi. Then we get
'1 . y.xz// D '1 .xz/'1 . y/ D .'1 .x/'1 .z//'1 . y/ D '1 .x/.'1 . y/'1 .z//:
This is a contradiction. Thus '1 restricted to h y; xzi is an automorphism too.

By the same arguments one can show that if '1 is an automorphism restricted to
hx; yzi then it is automorphism restricted to hz; xyi as well.
In the same way one can see that if '1 is an anti-automorphism restricted to
one of the subgroups of fhx; yzi; h y; xzi; hz; xyig then '1 is an anti-automorphism
restricted to any of two others, too. Hence if '1 restricted to all of subgroups
fhx; yzi; h y; xzi; hz; xyig is an automorphism then '1 2 Aut.o/. If '1 restricted to
all of subgroups fhx; yzi; h y; xzi; hz; xyig is an anti-automorphism then '2 D '1
x. yz/ 2 Aut.o/:
Therefore our Theorem is proved for Case 1.
Case 2 ' restricted to h y; zi is an anti-automorphism.

Now we can introduce the map 'N D ' xz yz which is an automorphism on
hx; yi hx; zi hz; yi Q8 and can repeat all considerations from the Case 1.
For a diassociative loop Q the inversion mapping
.J W x 7! x1 / W Q ! Q
is an anti-automorphism of Q. Above we had defined seven elementary half-

automorphisms f1 ; : : : 7 g which correspond to the set S of the elements of o: Note
that in the octonion loop o one has
7
Y
i D J:
iD1
Obviously, if ' is an anti-automorphism, then ' J is an automorphism.

Thus we have proved that for any half-automorphism ' W o ! o there exists an
element 2 such that D ' is an automorphism. q.e.d.
Note that the elementary maps of the octonion loop o considered as permutations
of the set of elements of o generate the elementary abelian group .o/ of order 27 .
But not all elements of are proper half-automorphisms. Here we do not discuss
the intersection HAut.o/ \ .o/.
For a diassociative loop Q we consider the group
SAut.Q/ D Aut.Q/ hJi
of all automorphisms and all anti-automorphisms of Q which is a subgroup of

HAut.Q/.
Proposition 6.2.7 If Q is a diassociative loop of exponent 4 and .Q/ is a subgroup
of HAut.Q/, then .Q/ is normal in HAut.Q/:
Proof For an element c 2 Q of order 4 and for 2 Aut.Q/ one has
.c; c1 / 1 D ..c/; .c1 //. Hence c 1 D .c/ : The proposition
follows from Lemma 6.2.6 q.e.d.
6.3 The CayleyDickson Extension Process for Algebras
An algebra A with an involution over a field F is a non-associative F-algebra with a

linear mapping a 7! a satisfying the identity
.ab/ D b a : (6.8)
In [27, III.4] for a given algebra A of finite dimension n with an involution it is

described how one can construct F-algebras CD.A/ of finite dimension 2n with an
involution which contains an isomorphic image of A. For our purposes it is enough
to have this construction for the field Q of the rational numbers or the field R of real
numbers.
Consider the vector space CD.A/ D A A with the operations
.x1 ; y1 /.x2 ; y2 / D .x1 x2 y2 y1 ; y2 x1 C y1 x2 /; (6.9)
.x; y/ D .x ; y/: (6.10)
Then an elementary computation shows that CD.A/ is an algebra with an involution.

Using (6.9) one obtains the identities
.u; 0/.v; 0/ D .uv v 0; 0u C v0/ D .uv; 0/; (6.11)
.u; 0/.0; v/ D .u0 C v 0; vu C 00/ D .0; vu/; (6.12)
.0; v/.u; 0/ D .0u v 0; 0 C vu / D .0; vu /; (6.13)
.0; u/.0; v/ D .0 v u; v0 C u0/ D .v u; 0/ (6.14)
and from (6.10) one obtains
.u; 0/ D .u ; 0/; .0; v/ D .0; v/: (6.15)
Consider the linear mappings
D .a 7! .a; 0// W A ! A A;
D .b 7! .0; b// W A ! A A:
Then W A ! CD.A/ is an injective homomorphism of algebras with an involution.

If the algebra A has a neutral element e, then .e/ D .e; 0/ is the neutral element of
CD.A/.
If .b1 ; : : : ; bm / is an F-basis of A, then

.b1 / : : : ; .bm /; .b1 /; : : : ; .bm / (6.16)
is an F-basis of CD.A/.

Given data A; B , where A is an algebra over F with an involution which has a
neutral element e and

B D b1 ; : : : ; bm
is an F-basis of A, we have defined by (6.9) and (6.10) on the vector space AA the
structure of a non-associative algebra CD.A/ with an involution and injective linear
mappings
; W A ! A A
such that
(a) is a monomorphism of algebras with an involution,
(b) e.;/ D ..e/;
0/ is a neutral element of the algebra A A,
(c) B.;/ D .b1 / : : : ; .bm /; .b1 /; : : : ; .bm / in an F-basis of A A.
In our situation for a given F-algebra A with an involution and x 2 A we put qA .x/ D
kxk D xx . Then
qA .x/ D .xx / D .x / x D xx D qA .x/ (6.17)

1
We now specialize F D R and define inductively a sequence Ak kD1 of algebras
with an involution of dimension 2k1 having bases B.k/ and neutral elements e.k/
such that

(i) A1 D R; B.1/ D 1 ,

(ii) AkC1 D CD.Ak /; B.kC1/ D .B.k/ /; .B.k/ / .
Here and denote the mappings A ! CD.A/ defined above. The algebras Ak ; k >
1 are called CayleyDickson algebras. The injective homomorphisms W Ak !
AkC1 allow to assume that
A1 < A2 < < Ak <

S
and to form the infinite-dimensional
T CayleyDickson algebra A1 D k Ak with
its basis B1 D k B.k/ . Put q.a/ D aa for a 2 A1 . Equation (6.17) says that
q.a/ D q.a/ for all a 2 A. One shows by induction the
Lemma 6.3.1 In A1 the following statements hold:
(i) 0 < aa 2 A1 D R for all 0 a 2 A1 .
(ii) b b D 1 D bb for all b 2 B1 . q.e.d.
6.4 Half-Automorphisms of CayleyDickson Loops

and of Code Loops
We now use the CayleyDickson algebra Am over R of dimension m to construct a

finite loop Cm of order 2m . Consider the set
Cm D fb1 ; : : : ; bm g;
where b1 ; : : : ; bm are the elements of the basis B.m/ constructed for Am . Here we
choose b1 D 1, the neutral element of Am . By the definition of the multiplication of
Am the set Cm is closed under this multiplication.
Proposition 6.4.1 The multiplication of Cm is diassociative.
Proof One can check this easily looking at the multiplication rules (6.11)(6.14).
q.e.d.
Note 2 Observe that the CayleyDickson algebras An ; n > 5 are flexible but not
diassociative.
Now observing Sthat by the conventions made before Cm 6 CmC1 for all m > 1 we
define C1 D m Cm .
Corollary 6.4.2 With respect to the multiplication in A1 the set C1 is an infinite
diassociative loop. q.e.d.
Remember that Cm D B.m/ [ .B.m/ /, where B.m/ is the standard basis of the
vector space Am . Given a half-automorphism of the loop Cm the set .B.m/ / is a
basis of Am , too. Hence we have the linear extension Q W Am ! Am . Obviously Q is
a linear map of the algebra Am such that
8x; y 2 Cm W .xy/
Q D .x/
Q . Q y/ or .xy/
Q D .
Q y/.x/
Q (6.18)
is satisfied. However Q is not a half- automorphism of the algebra Am :

Let us collect fundamental properties of the loop C1 , respectively of the loops
Cn in analogy to Theorem 6.2.2.
Theorem 6.4.3 In the CayleyDickson loops Cn ; n > 4 and C1 the following
statements are true:
(1) Cn is a loop of order 2n .
(2) C1 is a diassociative loop.
(3) In Cn there are one element z D 1 of order 2 and 2n 2 elements of order 4.
(4) zx D xz for all x 2 C1 .
(5) If H is a subloop of C1 , then z 2 H.
(6) Z D hzi is a normal subloop of C1 which coincides with the nucleus, the center,
the associator subloop, the commutator subloop, and the Frattini subloop
of C1 .
(7) Cn =Z is an elementary abelian group of order 2n 1.
Proof The proof is a straightforward induction over n beginning with Theorem 6.2.2
for n D 4. q.e.d.
Furthermore, all the loops Cn are Hamiltonian loops, i.e. they have the property
that all subloops are normal. For o this fact has been discussed in detail by D. Norton
[22] in the year 1951.
We now give a proof that Theorem 6.2.5 holds for arbitrary CayleyDickson
loops. Note that in the quaternion group C3 for all c 2 C3 with c2 1 the
elementary mappings c are anti-automorphisms. A. Grishkov suggested to study
the half-automorphisms of the class L of all loops L satisfying the following

conditions:
(L.1) L is a diassociative loop of exponent 4.
(L.2) In L there is a central subloop Z D hzi of order 2.
(L.3) x2 D z for all x 62 Z and x; y D z for all y 62 f1; z; x; xzg:
Lemma 6.4.4 Given a loop L 2 L and an element c 2 L of order 4 the elementary
map c is a half-automorphism.
Proof Let us show that for all x; y 2 L
c .xy/ D xy D c .x/c . y/ or c .xy/ D xy D c . y/c .x/:
To do this it is sufficient to assume that xy yx and to consider the following three

cases.
First case: x fc; c1 g; y fc; c1 g; xy fc; c1 g
Then by definition of c one has c .xy/ D xy D c .x/c . y/
Second case: x fc; c1 g; y 2 fc; c1 g
Then xy fc; c1 g and we get
c .xy/ D xy D x1 ; y1 yx D y2 yx D y1 x D c . y/c .x/:
and by the same arguments we get
c . yx/ D c .x/c . y/:
Third case: x fc; c1 g; y fc; c1 g; xy 2 fc; c1 g

By definition of c we have:
c .xy/ D .xy/1 D .xy/3 D .xy/2 .xy/ D y1 ; x1 .xy/ D yx D c . y/c .x/:
q.e.d.
Theorem 6.4.5 In a not associative CayleyDickson loop all elementary mappings
are proper half-automorphisms.
Proof If C a CayleyDickson loop, it follows from Theorem 6.4.3 that C belongs
to L . Hence all elementary mappings c ; c 2 C are half-automorphisms. If C is
not associative for c 2 C there exist quaternion subgroups Q.C/ and Q./ such that
c 2 Q./ and c Q.C/ . Hence c restricted to Q./ is an anti-automorphism and c
restricted to Q.C/ is an automorphism. Thus the Theorem is proved. q.e.d.
One can easily show that for the quaternion group Q8 D C3 the equality
HAut.Q8 / D SAut.Q8 / D .Q8 /Aut.Q8 / holds. For octonion loop o D C4 by
our Theorem 6.2.6 we have the factorization HAut.o/ D .o/Aut.o/. For r > 5 one
only knows the inclusion HAut.Cr / > .Cr /Aut.Cr /. The reason for this lies in
the fact that our proof of Theorem 6.2.6 uses the known subloop structure of o: All
subloops of o are CayleyDickson loops. The sedenion loop C5 of order 32 contains

a subloop oQ of order 16 which is not a CayleyDickson loop (see [5, 21]).
Question Is the statement HAut.C/ D .C/Aut.C/ true for every CayleyDickson
loop C?
Turning to code loops (see [11]) and using their description by Chein and
Goodaire [7] and Hsu [14, 15] we note that the unique non-associative code loop
belongs to the class L is o.
To decide when for an element c of an arbitrary code loop the elementary
mapping c is a half-automorphism stays an open question.
In [25] Pires and Grishkov study the variety of loops E generated by all finite
code loops and determine the free elements of E. To us it seems a demanding task
to discuss half-automorphisms of loops in E.
On page 114 we discussed the five Moufang loops of order 16 which are not
groups, among them the octonion loop. It is known that all these loops are code
loops. Two of these loops are Chein loops. In [10] necessary conditions are given
that for finite group G of even order the Chein loop M.G; 2/ admits a proper half-
automorphism and it is shown that on M.Q8 ; 2/ the only three elementary mappings
really are half-automorphisms [10, Example 8].
Question For which finite groups G of order 2n there exist elementary mappings of
the Chein loop M.G; 2/ such that they are half-automorphisms?
Acknowledgements The first author thanks to SRE, Mexico. The second author is grateful to
SNI, Mexico. The third author expresses her gratitude to the University of Sao Paulo for hospitality
during the JulyAugust of 2015 and to FAPESP, proceso 2015/07245-4.
References
1. A.A. Albert, Quasigroups II. Trans. Am. Math. Soc. 55, 401419 (1944)
2. R.Baer, The higher commutator subgroups of a group. Bull. Am. Math. Soc. 50, 143160
(1944)
3. R. Baer, The Homomorphism theorems for loops. Am. J. Math. 67(3), 450460 (1945)
4. R. Bruck, A Survey of Binary Systems (Springer, Berlin, 1966)
5. R.E. Cawagas, A.S. Carrascal, L.A. Bautista, J.P. Sta. Maria, J.D. Urrutia, B. Nobles,
The subalgebra structure of the CayleyDickson algebra of dimension 32 (2009). arXiv:
0907.2047v3
6. O. Chein, Moufang loops of small order. I. Trans. Am. Math. Soc. 188, 3151 (1974)
7. O. Chein, E. Goodaire, Moufang loops with a unique nonidentity commutator (associator,
square). J. Algebra 130(2), 369384 (1990)
8. S. Gagola III, A Moufang loops commutant. Math. Proc. Camb. Philos. Soc. 152(2), 193212
(2012)
9. S. Gagola III, M.L. Merlini Giuliani, Half-isomorphisms of Moufang loops of odd order. J.
Algebra Appl. 11(5), 194199 (2012)
10. S. Gagola III, M.L. Merlini Giuliani, On half-automorphisms of certain Moufang loops with
even order. J. Algebra 386, 131141 (2013)
11. R. Griess, Code loops. J. Algebra 100, 224234 (1986)
12. A. Grishkov, M.L. Merlini Giuliani, M. Rasskazova, L. Sabinina, Half-isomorphisms of finite

automorphic Moufang loops. Commun. Algebra 44(10), 42524261 (2016)
13. A. Grishkov, P. Plaumann, M. Rasskazova, L. Sabinina, Half-automorphisms of free automor-
phic Moufang loops. Math. Notes 98(2), 133135 (2015)
14. T. Hsu, Explicit constructions of code loops as centrally twisted products. Math. Proc. Camb.
Philos. Soc. 128(2), 223232 (2000)
15. T. Hsu, Class 2 Moufang loops, small Frattini Moufang loops, and code loops.
arXiv:math/9611214v1 [math.GR] (1996)
16. L.K. Hua, On the automorphisms of a Sfield. Proc. Natl. Acad. Sci. USA 35, 386389 (1949)
17. N. Jacobson, C.E. Rickart, Jordan homomorphisms of rings. Trans. Am. Math. Soc. 69,
479502 (1950)
18. M.K. Kinyon, J.D. Phillips, P. Vojtechovsky, C-loops: extensions and constructions. J. Algebra
Appl. 6(1), 120 (2007)
19. M. Kinyon, I. Stuhl, P. Vojtechovsky, Half-automorphisms of Moufang loops. J. Algebra 450,
152161 (2016)
20. J. Kirshtein, Automorphism groups of Cayley-Dickson loops. J. Gen. Lie Theory Appl. 6
(2012). Article ID G110901
21. B.M. Kivunge, J.D.H. Smith, Subloops of sedenions. Commentat. Math. Univ. Carol. 45(2),
295302 (2004)
22. D.A. Norton, Hamiltonian loops. Proc. Am. Math. Soc. 3, 5665 (1952)
23. H. Pflugfelder, Quasigroups and Loops: Introduction. Sigma Series in Pure Mathematics, vol. 7
(Heldermann Verlag, Berlin, 1990)
24. J.D. Philips, P. Vojtechovsky, C-loops: an introduction. Publ. Math. Debrecen 68, 115137
(2006)
25. R. Pires, A. Grishkov, Code loops: automorphisms and representations (in preparation). arXiv:
1412 2185 (2014)
26. S. Saeki, On square-preserving isometries of convolution algebras. Trans. Am. Math. Soc. 346,
707718 (1994)
27. R.D. Schafer, An Introduction to Nonassociative Algebras. Pure and Applied Mathematics,
vol. 22 (Academic, New York, 1966)
28. W.R. Scott, Half-homomorphisms of groups. Proc. Am. Math. Soc. 8, 11411144 (1957)
Chapter 7
Invariant Control Systems on Lie Groups
Rory Biggs and Claudiu C. Remsing
Abstract This is a survey of our research (conducted over the last few years) on
invariant control systems, the associated optimal control problems, and the asso-
ciated HamiltonPoisson systems. The focus is on equivalence and classification.
State space and detached feedback equivalence of control systems are characterized
in simple algebraic terms; several classes of systems (in three dimensions, on the
Heisenberg groups, and on the six-dimensional orthogonal group) are classified.
Equivalence of cost-extended systems is shown to imply equivalence of the
associated HamiltonPoisson systems. Cost-extended systems of a certain kind are
reinterpreted as invariant sub-Riemannian structures. A classification of quadratic
HamiltonPoisson systems in three dimensions is presented. As an illustrative
example, the stability and integration of a typical system is investigated.
7.1 Introduction
Geometric control theory began in the late 1960s with the study of (nonlinear)
control systems by using concepts and methods from differential geometry (cf.
[10, 64, 96]). A smooth control system may be viewed as a family of vector fields (or
dynamical systems) on a manifold, smoothly parametrized by controls. As soon as
a control function is fixed, one has a nonautonomous vector field (or, in coordinates,
a nonautonomous ordinary differential equation). An integral curve of such a vector
field is called a trajectory of the system. The first basic question one asks of a control
system is whether or not any two points can be connected by a trajectory: this is
known as the controllability problem. Once one has established that two points can
be connected by a trajectory, one may wish to find a trajectory that minimizes some
(practical) cost function: this is known as the optimality problem.
R. Biggs ()
Department of Mathematics and Applied Mathematics, University of Pretoria, 0002 Pretoria,
South Africa
e-mail: rory.biggs@up.ac.za
C.C. Remsing
Department of Mathematics, Rhodes University, 6140 Grahamstown, South Africa
e-mail: c.c.remsing@ru.ac.za

128 R. Biggs and C.C. Remsing
A significant subclass of control systems rich in symmetry are those evolving on

Lie groups and invariant under left translations; for such a system the left translation
of any trajectory is a trajectory. This class of systems was first considered in 1972
by Brockett [43] and by Jurdjevic and Sussmann [68]; it forms a natural geometric
framework for various (variational) problems in mathematical physics, mechanics,
elasticity, and dynamical systems (cf. [10, 40, 64, 66]). In the last few decades
substantial work on applied nonlinear control has drawn attention to invariant
control affine systems evolving on matrix Lie groups of low dimension (see, e.g.,
[67, 89, 92, 93] and the references therein).
This paper serves as a survey of our recent research on left-invariant control affine
systems, the associated optimal control problems, and the associated Hamilton
Poisson systems. Ideas and key results from several papers published over the
last couple of years are reexamined and restructured; some elements are also
reinterpreted.
The first aspect we address is the equivalence of control systems. Simple
algebraic characterizations of state space equivalence and detached feedback equiv-
alence are presented. Using these characterizations, several classes of systems are
classified. Specifically, classifications (under detached feedback equivalence) of the
full-rank systems on the three-dimensional Lie groups, the .2n C 1/-dimensional
Heisenberg groups, and the six-dimensional orthogonal group are treated.
The second aspect we address is the equivalence of invariant optimal control
problems, or rather, their associated cost-extended systems. If two cost-extended
systems are equivalent, then their optimal trajectories (resp. extremal trajectories)
are related by a Lie group isomorphism. The classification problem is briefly dis-
cussed and a few examples are presented in some detail. One associates to each cost-
extended system, via the Pontryagin Maximum Principle, a quadratic Hamilton
Poisson system on the associated LiePoisson space. It is shown that equivalence
of cost-extended systems implies equivalence of the associated HamiltonPoisson
systems. Additionally, the subclass of drift-free systems with homogeneous cost are
reinterpreted as invariant sub-Riemannian structures; some examples are revisited.
The third and last aspect we address is the equivalence, stability, and integration
of quadratic HamiltonPoisson systems. A classification of the homogeneous
systems in three dimensions is presented; a classification of the inhomogeneous
systems on the three-dimensional orthogonal LiePoisson space so .3/ is also
exhibited. As an illustrative example, we investigate the stability and the explicit
integration of a typical inhomogeneous system on the Euclidean LiePoisson space
se .2/ .
Throughout, we make use of the classification of three-dimensional Lie groups
and Lie algebras; relevant details are given in the Appendix.
7.1.1 Invariant Control Affine Systems
An `-input left-invariant control affine system on a (real, finite-dimensional,

connected) Lie group G consists of a family of left-invariant vector fields u on
7 Invariant Control Systems on Lie Groups 129
G, affinely parametrized by controls u 2 R` . Such a system is written, in classical

notation, as
gP D u .g/ D .g; u/ D g .A C u1 B1 C C u` B` /; g 2 G; u 2 R` :
Here A; B1 ; : : : ; B` are elements of the Lie algebra g with B1 ; : : : ; B` linearly

independent. The product gA denotes the left translation T1 Lg A of A 2 g
by the tangent map of Lg W G ! G, h 7! gh; in fact, TG has trivialization
TG G g, gA $ .g; A/. (When G is a matrix Lie group, this product is simply
matrix multiplication.) Note that the dynamics W G R` ! TG are invariant
under left translations, i.e., .g; u/ D g .1; u/. A system is completely
determined by the specification of its state space G and its parametrization map
.1; / W R` ! g. When G is fixed, we specify by simply writing
W A C u 1 B1 C C u ` B` :
The trace of a system is the affine subspace D A C 0 D A C

hB1 ; : : : ; B` i of g. (Here 0 D hB1 ; : : : ; B` i is the subspace of g spanned by
B1 ; : : : ; B` .) A system is called homogeneous if A 2 0 , and inhomogeneous
otherwise; is said to be drift free if A D 0. Also, is said to have full rank if
its trace generates the whole Lie algebra, i.e., Lie. / D g.
Remark 1 We have the following characterization of the full-rank condition for
systems on G when dim G D 3. No homogeneous single-input system has full
rank. An inhomogeneous single-input system has full rank if and only if A, B1 , and
A; B1 are linearly independent. A homogeneous two-input system has full rank
if and only if B1 , B2 , and B1 ; B2 are linearly independent. Any inhomogeneous
two-input or (homogeneous) three-input system has full rank.
The admissible controls are piecewise continuous maps u./ W 0; T ! R` . A
trajectory for an admissible control u./ is an absolutely continuous curve g./ W
0; T ! G such that gP .t/ D g.t/ .1; u.t// for almost every t 2 0; T. We say
that a system is controllable if for any g0 ; g1 2 G, there exists a trajectory
g./ W 0; T ! G such that g.0/ D g0 and g.T/ D g1 . The attainable set (from
the identity 1 2 G ) is the set A of all terminal points g.T/ of all trajectories
g./ W 0; T ! G starting at 1. (A system is controllable if and only if A D G.)
If is controllable, then it has full rank.
For more details about invariant control systems, see, e.g., [10, 64, 68, 94].
7.1.2 Examples of Invariant Optimal Control Problems
We present three illustrative examples of optimal control problems. Numerous such

problems have been investigated over the last few decades; these problems include
the dynamic equations of the rigid body, various versions of Euler and Kirchhoff
elastic rod problem, Dubins problem, the motion of a particle in a magnetic or
YangMills field, the control of quantum systems, and the sub-Riemannian geodesic
problem (see, e.g., the monographs [10, 40, 64] as well as [89]).
The Plate-Ball Problem [63] The plate-ball problem consists of the following
kinematic situation: a ball rolls without slipping between two horizontal plates;
the lower plate is fixed and the ball is rolled through the horizontal movement of
the upper plate. The problem is to transfer the ball from a given initial position
and orientation to a prescribed final position and orientation along a path which
RT
minimizes 0 kv.t/kdt. Here v denotes the velocity of the moving plate and T is
the time of transfer.
This problem can be regarded as an invariant optimal control problem on the
five-dimensional Lie group
82 x 3 9
< e1 0 0 =
R2 SO .3/ D 4 0 ex2 0 5 W x1 ; x2 2 R; R 2 SO .3/
: ;
0 0 R
specified by (the dynamical law)

0 2 3 2 31
1 000 0 0 00 0 0
B 60 000 077 6 0 07 C
B 6 60 10 7C
B 6 7 6 7C
gP D g Bu1 60 000 17 C u2 60 00 0 0 7C ; g 2 R2 SO .3/
B 6 7 6 7C
@ 40 000 05 40 00 0 15A
0 010 0 0 00 1 0
the boundary conditions g.0/ D g0 , g.T/ D g1 , and the cost functional

Z T
u1 .t/2 C u2 .t/2 dt
0
to be minimized.
The Airplane Problem [100] The airplane landing problem consists of finding
a trajectory for an airplane movingat a constant velocityfrom some starting
point and orientation to some final point and orientation; the airplane is constrained
to arrive at the final point and orientation at (some fixed) time T. A simplified
kinematic model of the airplane entails that it always fly forward and the controls
may yaw, pitch, and roll the aircraft; the cost assigned to a particular solution will
be the sum square of the inputs (this cost encodes the requirement to minimize the
amount of maneuvering an aircraft will do).
Such a problem may be formulated as an optimal control problem on the six-
dimensional Euclidean group

10
SE .3/ D R3 SO .3/ D W x 2 R3 ; R 2 SO .3/
xR
specified by (the equations of motion)
gP D g .A C u1 B1 C u2 B2 C u3 B3 / ; g 2 SE .3/
with
2 3 2 3 2 3 2 3
00 00 000 0 0 0 00 00 0 0
61 0 7
0 07 6000 07 7 6 7 6 1 07
AD6 ; B1 D 6 60 0 0 17
; B3 D 6
00 7
40 0 0 05 40 0 0 15 ; B2 D 40 0 0 05 40 1 0 05
00 00 001 0 0 1 00 00 0 0
the boundary conditions g.0/ D g0 , g.T/ D g1 and the cost functional

Z
1 T
J D c1 u1 .t/2 C c2 u2 .t/2 C c3 u3 .t/2 dt; c1 ; c2 ; c3 > 0:
2 0
If the cost weights c1 ; c2 ; c3 are equal, then the resulting system of equations is
related to Lagranges top (see, e.g., [64]); the Kovalevskayas top is obtained for a
different ratio of weights (see, e.g., [65]).
Control of SerretFrenet Systems [64] A differentiable curve in the Euclidean
plane E2 , parametrized by arc-length, can be lifted to the group of motions of E2
by means of a positively oriented orthonormal moving frame v1 ; v2 defined by
P D v1 ; vP1 D v2 ; vP2 D v1 (7.1)
where is the signed curvature of . The moving frame can be expressed by a

rotation matrix R whose columns consist of the coordinates of v1 and v2 relative to
a fixed orthonormal frame e1 ; e2 2 E2 . Omitting any notational distinctions between
vectors and their coordinate vectors, then Rei D vi . The curve along with its
moving frame can be represented as an curve g./ on the group of motions of E2 ,
namely
82 3 9
< 1 00 =
SE .2/ D 4 1 5 W 1 ; 2 2 R; R 2 SO .2/ :
: R ;
2
Interpreting the curvature as a control function, the SerretFrenet differential

system (7.1) can then be written as an (inhomogeneous) invariant control system
02 3 2 31
000 00 0
gP D g @41 0 05 C 40 0 15A ; g 2 SE .2/:
000 01 0
In this way, many classic variational problems in geometry become problems in

optimal control. For example, the problem of finding a curve that will satisfy
the given boundary conditions .0/ D a, P .0/ D aP , .T/ D b, .T/
P P and
D b,
RT 2
will minimize 0 .t/ dt goes back to Euler; its solutions are known as the elastica
(see, e.g., [92, 93]).
7.2 Equivalence of Control Systems
The most natural equivalence relation for control systems is equivalence up to

coordinate changes in the state space. This is called state space equivalence
(see [61]). State space equivalence is well understood; it establishes a one-to-one
correspondence between the trajectories of the equivalent systems. However, this
equivalence relation is very strong. For analytic control systems on manifolds,
Krener characterized local state space equivalence in terms of the existence of a
linear isomorphism preserving iterated Lie brackets of the systems vector fields
([72], see also [10, 95, 96]).
Another fundamental equivalence relation for control systems is that of feedback
equivalence. Two feedback equivalent control systems have the same set of trajecto-
ries (up to a diffeomorphism in the state space) which are parametrized differently
by admissible controls. Feedback equivalence has been extensively studied in the
last few decades (see [90] and the references therein). There are a few basic methods
used in the study of feedback equivalence. These methods are based either on
(studying invariant properties of) associated distributions or on Cartans method
of equivalence [52] or inspired by the Hamiltonian formalism [61]; also, another
fruitful approach is closely related to Poincars technique for linearization of
dynamical systems. Feedback transformations play a crucial role in control theory,
particularly in the important problem of feedback linearization [62]. The study of
feedback equivalence of general control systems can be reduced, by a simple trick,
to the case of control affine systems [61]. For a thorough study of the equivalence
and classification of control affine systems, see [50].
We consider state space equivalence and feedback equivalence in the context
of left-invariant control affine systems ([38], see also [25]). Characterizations of
state space equivalence and (detached) feedback equivalence are obtained in terms
of Lie group isomorphisms. A complete classification of systems on the (simply
connected) three-dimensional Lie groups is discussed; the controllability of these
systems is also briefly treated. Classifications of the systems on the .2n C 1/-
dimensional Heisenberg groups and the six-dimensional orthogonal group are also
covered.
7.2.1 State Space Equivalence
Two systems and 0 are called state space equivalent if there exists a
diffeomorphism
W G ! G0 such that, for each control value u 2 R` , the vector
fields u and u0 are
-related, i.e., Tg
.g; u/ D 0 .
.g/; u/ for g 2 G and
u 2 R` . We have the following simple algebraic characterization of this equivalence.
Theorem 1 ([38], see also [72]) Two full-rank systems and 0 are state space
equivalent if and only if there exists a Lie group isomorphism
W G ! G0 such
that T1
.1; u/ D 0 .1; u/ for all u 2 R` .
Proof (Sketch) Suppose that and 0 are state space equivalent. By composition
with a left translation, we may assume
.1/ D 1. As the elements u .1/,
u 2 R` generate g and the push-forward
u of left-invariant vector fields
u are left invariant, it follows that
is a Lie group isomorphism satisfying
the requisite property (cf. [25]). Conversely, suppose that
W G ! G0 is a Lie
group isomorphism as prescribed. Then Tg
.g; u/ D T1 .
Lg / .1; u/ D
T1 .L
.g/
/ .1; u/ D 0 .
.g/; u/. t
u
Remark 2 If
is defined only between some neighbourhoods of identity of G and
G0 , then and 0 are said to be locally state space equivalence. A characterization
similar to that given in Theorem 1, in terms of Lie algebra automorphisms, holds. In
the case of simply connected Lie groups, local and global equivalence are the same
(as d Aut.G/ D Aut.g/).
State space equivalence is quite a strong equivalence relation. Hence, there are so
many equivalence classes that any general classification appears to be very difficult
if not impossible. However, there is a chance for some reasonable classification in
low dimensions. We give an example to illustrate this point.
Example 1 ([2]) Any two-input inhomogeneous full-rank control affine system on
the Euclidean group SE .2/ is state space equivalent to exactly one of the following
systems
1; W E3 C u1 .E1 C 1 E2 / C u2 .E2 /

2; W E1 C 1 E2 C 2 E3 C u1 .E3 / C u2 E2
3; W E1 C 1 E2 C 2 E3 C u1 .E2 C 3 E3 / C u2 .E3 /:
Here > 0; 0 and 1 ; 2 ; 3 2 R, with different values of these parameters

yielding distinct (non-equivalent) class representatives.
Proof (Sketch) The group of linearized automorphisms of SE .2/ is given by
82 3 9
< x y v =
d Aut.SE .2// D 4&y &x w5 W x; y; v; w 2 R; x2 C y2 0; & D 1 :
: ;
0 0 &
Indeed, it is straightforward to show that every automorphism of the simply

connected Lie group SE f .2/ preserves the kernel of the universal covering q W
f
SE .2/ ! SE .2/ [33]; hence, d Aut.SE
P .2// D Aut.seP.2// (see the Appendix).
P
Let D .SE .2/; /, .1; u/ D 3iD1 ai Ei C u1 3iD1 bi Ei C u2 3iD1 ci Ei , or
in matrix form
2 3
a 1 b 1 c1
W 4 a 2 b 2 c2 5 :
a 3 b 3 c3
It is then straightforward to show that there exists 2 d Aut.SE .2// such that
2 3 2 3
a1 b 1 c1 0 1 0
4
a2 b 2 c2 5 D 4 0 1 5 if b3 D 0; c3 D 0
a3 b 3 c3 0 0
2 3 2 3
a1 b 1 c1 00
4 a2 b 2 c2 5 D 4 1 0 1 5 if b3 0; c3 D 0
a3 b 3 c3 2 0
2 3 2 3
a 1 b 1 c1 0 0
or 4 5
a 2 b 2 c2 D 1 14 05 if c3 0:
a 3 b 3 c3 2 3
Thus is state space equivalent to 1; , 2; , or 3; . It is a simple matter

to verify that these class representatives are non-equivalent. t
u
Note 1 A full classification (under state space equivalence) of systems on SE .2/
appears in [2], whereas a classification of systems on SE .1; 1/ appears in [17]. For
a classification of systems on SO .2; 1/0 , see [39].
7.2.2 Detached Feedback Equivalence
We specialize feedback equivalence in the context of invariant systems by requiring

that the feedback transformations are compatible with the Lie group structure.
Two systems and 0 are called detached feedback equivalent if there exist
diffeomorphisms
W G ! G0 and ' W R` ! R` such that for each control
value u 2 R` the vector fields u and '.u/0
are
-related, i.e., Tg
.g; u/ D
0 `
.
.g/; '.u// for g 2 G and u 2 R . We have the following simple algebraic
characterization of this equivalence in terms of the traces D im .1; / and
0 D im 0 .1; / of and 0 .
Theorem 2 ([38]) Two full-rank systems and 0 are detached feedback

equivalent if and only if there exists a Lie group isomorphism
W G ! G0 such
that T1
D 0 .
Proof (Sketch) Suppose and 0 are detached feedback equivalent. By com-
posing
with an appropriate left translation, we may assume
.1/ D 10 . Hence
T1
.1; u/ D 0 .10 ; '.u// and so T1
D 0 . Moreover, as the elements
u .1/, u 2 R` generate g and the push-forward of the left-invariant vector fields
u are left invariant, it follows that
is a group isomorphism (cf. [25]). On the
other hand, suppose there exists a group isomorphism
W G ! G0 such that
0
T1
D 0 . Then there exists a unique affine isomorphism ' W R` ! R`
such that T1
.1; u/ D 0 .10 ; '.u//. As with state space equivalence, by left-
invariance and the fact that
is a Lie group isomorphism, it then follows that
Tg
.g; u/ D 0 .
.g/; '.u//. t
u
Remark 3 If
is defined only between some neighbourhoods of identity of G
and G0 , then and 0 are said to be locally detached feedback equivalent.
A characterization similar to that given in Theorem 2, in terms of Lie algebra
automorphisms, holds. As for state space equivalence, in the case of simply
connected Lie groups local and global equivalence are the same (as d Aut.G/ D
Aut.g/).
Detached feedback equivalence is notably weaker than state space equivalence.
To illustrate this point, we give a classification, under detached feedback equiva-
lence, of the same class of systems considered in Example 1.
Example 2 ([29]) Any two-input inhomogeneous full-rank control affine system
on SE .2/ is detached feedback equivalent to exactly one of the following systems
1 W E1 C u1 E2 C u2 E3
2; W E3 C u1 E1 C u2 E2 :
Here > 0 parametrizes a family of class representatives, each different value

corresponding to a distinct non-equivalent representative.
Proof (Sketch) As argued in Example 1, we have d Aut.SE .2// D Aut.se .2//. Let
D .SE .2/; / be an inhomogeneous system with trace
3
* 3 3
+
X X X
D a i Ei C b i Ei ; ci E i :
iD1 iD1 iD1

If c3 0 or b3 0, then D a01 E1 C a02 E2 C b01 E1 C b02 E2 ; c01 E1 C c02 E2 C E3 .
Now either b01 0 or b02 0, and so
0 0
b2 b01 v1 a
D 20
b01 b02 v2 a1
has a unique solution (with v2 0). Therefore

2 3
v2 b02 v2 b01 c01
D 4v2 b01 v2 b02 c02 5
0 0 1
is a (Lie algebra) automorphism such that 1 D .E1 C hE2 ; E3 i/ D . Thus

is detached feedback equivalent to 1 . On the other hand, suppose b3 D 0
and c3 D 0. Then D a3 E3 C hE1 ; E2 i. Hence D diag .1; 1; sgn.a3 // is
an automorphism such that D E3 C hE1 ; E2 i with > 0. Thus is
detached feedback equivalent to 2; . As the subspace hE1 ; E2 i is invariant (under
automorphisms), 1 and 2; cannot be equivalent. It is easy to show that 2; a
2;0 are equivalent only if D 0 . t
u
Remark 4 and 0 are feedback equivalent if there exists a diffeomorphism
W
G ! G0 such that (the push-forward)
F D F 0 . Here g 7! F .g/ D g is the
field of admissible velocities. The specialization to detached feedback equivalence
corresponds to the existence of a Lie group isomorphism
such that
F D
F 0 . Thus feedback equivalence is weaker than detached feedback equivalence. For
example, suppose D g, 0 D g0 , and G is diffeomorphic to G0 . Then
and 0 are feedback equivalent. However, and 0 will be detached feedback
equivalent only if G and G0 are, in addition, isomorphic as Lie groups.
7.2.3 Classification in Three Dimensions
We exhibit a classification, under detached feedback equivalence, of the full-rank

systems evolving on three-dimensional simply connected Lie groups. A represen-
tative is identified for each equivalence class. Systems on the Euclidean group and
the orthogonal group are discussed as typical examples. The three-dimensional Lie
groups, their Lie algebras, and automorphism groups are listed in the Appendix. For
a classification of invariant control affine systems on all three-dimensional matrix
Lie groups, see [33].
7.2.3.1 Solvable Groups
The classification procedure is as follows. First the group of automorphisms is deter-

mined (see the Appendix). Equivalence class representatives are then constructed by
considering the action of an automorphism on the trace of a typical system. Lastly,
one verifies that none of the representatives are equivalent.
Theorem 3 ([28, 29], see also [26, 31]) Suppose is a full-rank system evolving
on a simply connected solvable Lie group G. Then G is isomorphic to one of
the groups listed below and is detached feedback equivalent to exactly one of
accompanying (full-rank) systems on that group.
1. On R3 we have the systems
.2;1/ W E1 C u1 E2 C u2 E3 .3;0/ W u1 E1 C u2 E2 C u3 E3 :
2. On H3 we have the systems
.1;1/ W E2 C uE3 .2;0/ W u1 E2 C u2 E3

.2;1/ .2;1/
1 W E1 C u 1 E2 C u 2 E3 2 W E3 C u 1 E1 C u 2 E2
.3;0/ W u1 E1 C u2 E2 C u3 E3 :
3. On G3:2 we have the systems
.1;1/ .1;1/
1 W E2 C uE3 2; W E3 C uE2
.2;1/
.2;0/ W u2 E2 C u2 E3 1 W E1 C u 1 E2 C u 2 E3
.2;1/ .2;1/
2 W E2 C u 1 E3 C u 2 E1 3; W E3 C u1 E1 C u2 E2
.3;0/ W u1 E1 C u2 E2 C u3 E3 :
4. On G3:3 we have the systems
.2;1/ .2;1/
1 W E1 C u 1 E2 C u 2 E3 2; W E3 C u1 E1 C u2 E2
.3;0/ W u1 E1 C u2 E2 C u3 E3 :
5. On SE .1; 1/ we have the systems
.1;1/ .1;1/
1 W E2 C uE3 2; W E3 C uE2
.2;1/
.2;0/ W u1 E2 C u2 E3 1 W E1 C u 1 E2 C u 2 E3
.2;1/ .2;1/
2 W E1 C u1 .E1 C E2 / C u2 E3 3; W E3 C u1 E1 C u2 E2
.3;0/ W u1 E1 C u2 E2 C u3 E3 :
6. On Ga3:4 (resp. on Aff .R/0 R) we have the systems
.1;1/ .1;1/
1 W E2 C uE3 2; W E3 C uE2
.2;1/
.2;0/ W u1 E2 C u2 E3 1 W E1 C u 1 E2 C u 2 E3
.2;1/ .2;1/
2 W E1 C u1 .E1 C E2 / C u2 E3 3 W E1 C u1 .E1 E2 / C u2 E3
.2;1/
4; W E3 C u1 E1 C u2 E2 .3;0/ W u1 E1 C u2 E2 C u3 E3 :
f .2/ we have the systems

7. On SE
.1;1/ .1;1/
1 W E2 C uE3 2; W E3 C uE2
.2;1/
.2;0/ W u1 E2 C u2 E3 1 W E1 C u 1 E2 C u 2 E3
.2;1/
8. On Ga3:5 we have the systems
.1;1/ .1;1/
1 W E2 C uE3 2; W E3 C uE2
.2;1/
.2;0/ W u1 E2 C u2 E3 1 W E1 C u 1 E2 C u 2 E3
.2;1/
Here > 0 and 0 parametrize families of distinct (non-equivalent) class

representatives.
Proof We treat, as typical case, only item (7). The group of linearized automor-
f .2/ is given by
phisms of SE
82 3 9
< x y u =
d Aut.SE/ .2/ D 4 y x v 5 W x; y; u; v 2 R; x2 C y2 0; D 1 :
: ;
0 0
Let be a single-input inhomogeneous system with trace D A C 0

see .2/. Suppose E3 . 0 / f0g: (Here E3 is the corresponding element of the dual
basis.) Then D a1 E1 C a2 E2 C hb1 E1 C b2 E2 C E3 i. Thus
2 3
a1 a1 b1
D 4a1 a2 b2 5
0 0 1
.1;1/ .1;1/
is an automorphism such that 1 D . So is equivalent to 1 .
On the other hand, suppose E3 . 0 / D f0g. Then D a1 E1 C a2 E2 C a3 E3 C
hb1 E1 C b2 E2 i with a3 0. Hence

2 3
b2 sgn.a3 / b1 a3 sgn.a
a1
3/
6 7
D 4b1 sgn.a3 / b2 a sgn.a
a2
5
3 3/
0 0 sgn.a3 /
.1;1/
is an automorphism such that 2; D , where D a3 sgn.a3 /.
Let be a two-input homogeneous system with trace D hB1 ; B2 i. Then
b W B1 C hB2 i is a (full-rank) single-input inhomogeneous system. Therefore,
there exists an automorphism such that .B1 C hB2 i/ equals either E2 C hE3 i
or E3 C hE2 i. Hence, in either case, we get hB1 ; B2 i D hE2 ; E3 i. Thus is
equivalent to .2;0/ .
The two-input inhomogeneous systems were covered in Example 2. If is a
three-input system, then it is equivalent to .3;0/ .
Again, most pairs of systems cannot be equivalent due to different homogeneities
or different number of inputs. As the subspace hE1 ; E2 i is invariant (under the action
.1;1/ .1;1/
of automorphisms), 1 is not equivalent to any system 2; . For A 2 see .2/
.1;1/ .1;1/
and 2 d Aut.SE/ .2/, we have that E3 . E3 / D . Thus 2; and 2;0
are equivalent only if D 0 . For the two-input inhomogeneous systems, similar
arguments hold. t
u
7.2.3.2 Semisimple Groups
The procedure for classification is similar to that of the solvable groups. However,
here we employ an invariant bilinear product ! (the Lorentzian product and the dot
product, respectively); the inhomogeneous systems are (partially) characterized by
the level set fA 2 g W !.A; A/ D g that their trace is tangent to.
Theorem 4 ([30], see also [26, 31]) Suppose is a full-rank system evolving
on a simply connected semisimple Lie group G. Then G is isomorphic to one of
the groups listed below and is detached feedback equivalent to exactly one of
accompanying (full-rank) systems on that group.
f .2; R/ we have the systems
1. On SL
.1;1/ .1;1/
1 W E3 C u.E2 C E3 / 2; W E2 C uE3
.1;1/ .1;1/
3; W E1 C uE2 4; W E3 C uE2
.2;0/ .2;0/
1 W u 1 E1 C u 2 E2 2 W u 1 E2 C u 2 E3
.2;1/ .2;1/
1 W E3 C u1 E1 C u2 .E2 C E3 / 2; W E1 C u1 E2 C u2 E3
.2;1/
2. On SU .2/ we have the systems
.1;1/ W E2 C uE3 .2;0/ W u1 E2 C u2 E3

.2;1/ W E1 C u1 E2 C u2 E3 .3;0/ W u1 E1 C u2 E2 C u3 E3 :
Here > 0 parametrizes families of distinct (non-equivalent) class representatives.

Proof We consider only item (2), i.e., systems on the unitary group SU .2/. (The
proof for item (1), although more involved, is similar.) The group of linearized
automorphisms of SU .2/ is d Aut.SU .2// D SO .3/ D fg 2 R33 W gg> D
1; det g D 1g.
PThe dot product on P su .2/ is given by A B D a1 b1 C a2 b2 C a3 b3 .
(Here A D 3iD1 ai Ei and B D p3iD1 bi Ei .) The level sets S D fA 2 su .2/ W
A A D g are spheres of radius (and are preserved by automorphisms). The
group of automorphisms acts transitively on each sphere S . The critical point
C . /, at which an affine subspace D A C hBi or A C hB1 ; B2 i is tangent to a
sphere S , is given by
A B
C . / D A B
B B

B1 B1 B1 B2 1 A B1
C . / D A B1 B2
B1 B2 B2 B2 A B2
respectively. Critical points behave well under the action of automorphisms, i.e.,
C . / D C . / for any automorphism .
Let be a single-input inhomogeneous system with trace . There exists an
automorphism
p such that D sin E1 C cos E2 C hE3 i, where D
C . / C . /. Hence
2 3
cos sin 0
0
D 4 sin cos 05
0 0 1
.1;1/
is an automorphism such that 0 D .
Let be a two-input homogeneous system with trace D hB1 ; B2 i. Then
b W B1 C hB2 i is a (full-rank) single-input inhomogeneous system. Therefore,
there exists an automorphism such that .B1 C hB2 i/ D E2 C hE3 i. Hence,
hB1 ; B2 i D hE2 ; E3 i. Thus is equivalent to .2;0/ .
Let be a two-input inhomogeneous system with trace . We have C . /
C . / D 2 for some > 0. As C .1; / C .1; / D 2 , there exists an
automorphism such that C . / D C .1; /. Hence and 1; are both
equal to the tangent plane of S2 at C . /, and are therefore identical.
If is a three-input system, then it is equivalent to .3;0/ .
Lastly we note that none of the representatives obtained are equivalent. (Again,
we first distinguish representatives in terms of homogeneity and number of inputs.)
.1;1/ .1;1/ .2;1/ .2;1/
As 2 D C . / C . / (resp. 2 D C . / C . /) is an invariant
.1;1/ .1;1/ .2;1/ .2;1/
quantity, the systems and 0 (resp. and 0 ) are equivalent
only if D 0 . t
u
7.2.3.3 Controllability
The controllability property is invariant under detached feedback equivalence. The

following results, concerning control systems on Lie groups, are well known. If a
system is controllable, then it has full rank. Moreover, if a system is homogeneous or
if its state space is compact, then the full-rank condition implies controllability. Each
of the following conditions is sufficient for a full-rank system to be controllable
[68]:
the direction space 0 generates g, i.e., Lie . 0 / D g;
the identity element 1 is in the interior of the attainable set
A D fg.t1 / W g./ is a trajectory such that g.0/ D 1; t1 > 0gI
there exists C 2 such that t 7! exp.tC/ is periodic.

If the state space is simply connected and completely solvable, then the condition
Lie . 0 / D g is necessary for controllability [94].
In several cases, these results allow one to immediately determine the controlla-
bility nature of a given system; for instance, for a system on a compact Lie group
one need only check whether or not Lie. / D g. In some cases determining the
controllability of a system can be much more involved; for instance, we could only
.1;1/
show that the system 1 on Ga3:5 is controllable by explicitly constructing a
family of trajectories showing that 1 2 int A .
Theorem 5 ([33], see also [34, 42, 94]) For full-rank systems on three-dimensional
matrix Lie groups, the following criteria for controllability hold.
1. On each of the groups Aff .R/0 R, H3 , G3:2 , G3:3 , SE .1; 1/, and Ga3:4 , a
system is controllable if and only if Lie . 0 / D g.
2. On each of the groups SEn .2/, n 2 N, SO .3/, and SU .2/, all systems are
controllable.
3. On each of the groups Aff .R/ T, SL .2; R/, and SO .2; 1/0 , a system is
controllable if and only if it is homogeneous or there exists A 2 such that
t 7! exp.tA/ is periodic.
4. On each of the groups SE f .2/ and Ga , a system is controllable if and only if
3:5
E3 . / f0g. (Here E3 denotes the corresponding element of the dual basis.)
0
7.2.4 Classification Beyond Three Dimensions
Detached feedback equivalence has proved useful and feasible not only in three
dimensions, but also in some higher dimensional cases. In this spirit we classify
controllable systems on the Heisenberg groups H2nC1 and revisit a classification of
homogeneous systems on the six-dimensional orthogonal group SO .4/.
Note 2 Homogeneous systems on four-dimensional Lie groups have been classified
on the oscillator group [a central extension of SE .2/] in [36], and on the four-
dimensional central extension of SE .1; 1/ in [19]. Recently, subspaces of the
four-dimensional Lie algebras were classified up to automorphism in [37]; from this
classification one can easily derive a classification (under detached feedback equiv-
alence) of the homogeneous systems on the simply connected four-dimensional Lie
groups.
7.2.4.1 The Heisenberg Groups H2nC1
The .2n C 1/-dimensional Heisenberg group may be realized as a matrix Lie group
82 3 9

1 x1 x2 xn z >
>

60 1 >

6 0 0 y1 77
>
>
>
>
<6
7 >
=
60 0 1 0 y2 7
H2nC1 D 66 :: ::
7
:: 7 W x ;
i iy ; z 2 R :

6: : :7 >
>

6 7 >
>

>
>
40 1 yn 5 >
>
: ;
0 0 1
H2nC1 is a simply connected nilpotent Lie group with one-dimensional center; its
Lie algebra
82 3 9

0 x1 x2 xn z >
>

60 0 7 >
>

6 0 0 y1 7 >
>

6 7 >
>
< 60 0 0 0 y2 7 Xn =
h2nC1 D 66: ::
7
:: 7 D zZ C .x X C y Y / W x ; y ; z 2 R
6 ::
i i i i i i
: :7 >
>

6 7 iD1 >
>

>
>
40 0 yn 5 >
>
: ;
0 0 0
has non-zero commutators Xi ; Yj D ij Z.

The automorphisms of h2nC1 are exactly those linear isomorphisms that preserve
the centre z of h2nC1 and for which the induced map on h2nC1 =z preserves an
appropriate symplectic structure (cf. [53]). More precisely, let ! be the skew-
symmetric bilinear form on h2nC1 specified by A; B D !.A; B/Z for A; B 2
h2nC1 . A linear isomorphism W hn ! h2nC1 is a Lie algebra automorphism if and

only if Z D cZ and !. A; B/ D c !.A; B/ for some c 0.
We give a matrix representation for the group of automorphisms. Throughout,
we use the ordered basis
.Z; X1 ; Y1 ; X2 ; Y2 ; : : : ; Xn ; Yn /
for h2nC1 . The bilinear form ! takes the form

2 3
0 1 0
61 0 7
6 7
00 6 : 7
!D ; where J D 6 :: 7:
0J 6 7
4 0 15
0 1 0
We note that the linear map

2 3
1 0 0
60 01 07
6 7
6 7
6 10 7
6
& D6 : 7 (7.2)
: :: 7
6 : : 7
6 7
4 0 15
0 0 10
is an automorphism.
Proposition 1 (cf. [24, 91]) The group of automorphisms Aut.h2nC1 / is given by

2 2
r v r v 2n
; & W r > 0; v 2 R ; g 2 Sp .n; R/
0 rg 0 rg
where

Sp .n; R/ D g 2 R2n2n W g> Jg D J
is the n.2n C 1/-dimensional symplectic group over R.

Theorem 6 (cf. [24]) Every controllable system on H2nC1 is detached feedback
equivalent to one of the following three systems
.2n;0/ W u1 X1 C C un Xn C unC1 Y1 C C u2n Yn

.2n;1/ W Z C u1 X1 C C un Xn C unC1 Y1 C C u2n Yn
.2nC1;0/ W u1 X1 C C un Xn C unC1 Y1 C C u2n Yn C u2nC1 Z:
Proof Let be a system on H2nC1 with trace D A C 0 . As H2nC1 is

completely solvable and simply connected, controllability of is equivalent to the
condition Lie. 0 / D h2nC1 [94]. We have Lie. 0 / span. 0 ; Z/. Therefore, if
is controllable, then 0 has codimension zero or one. Moreover, if Lie. 0 / D
h2nC1 and 0 has codimension one, then
0 D hX1 C v1 Z; : : : ; Xn C vn Z; Y1 C vnC1 Z; : : : ; Yn C v2n Zi
for some v1 ; : : : ; v2n 2 R.

Let be a homogeneous 2n-input system on H2nC1 with trace D 0 . As
is controllable, we have Lie. / D h2nC1 . As H2nC1 is simply connected, we
have d Aut.H2nC1 / D Aut.h2nC1 /. Hence, it suffices to show that there exists an
(inner) automorphism 2 Aut.hn / such that 0 D hX1 ; : : : ; Xn ; Y1 ; : : : ; Yn i.
By the above argument
D hX1 C v1 Z; : : : ; Xn C vn Z; Y1 C vnC1 Z; : : : ; Yn C v2n Zi :
Accordingly,

1 v
D ; v D v1 vnC1 v2 vnC2 vn v2n
0 I2n
is an inner automorphism such that D hX1 ; : : : ; Xn ; Y1 ; : : : ; Yn i.

Let be a inhomogeneous 2n-input system on H2nC1 with trace D A C 0 .
As is controllable, we have Lie. 0 / D h2nC1 . By the above argument, there
exists an automorphism such that 0 D hX1 ; : : : ; Xn ; Y1 ; : : : ; Yn i. Thus
D AChX1 ; : : : ; Xn ; Y1 ; : : : ; Yn i D aZ ChX1 ; : : : ; Xn ; Y1 ; : : : ; Yn i for some a 0.
1 p1
Accordingly, 0 D diag. jaj ; jaj ; : : : ; p1jaj / or 0 D & diag. jaj 1 p1
; jaj ; : : : ; p1jaj / is
an automorphism such that D Z C hX1 ; : : : ; Xn ; Y1 ; : : : ; Yn i.
Clearly, any .2n C 1/-input system is equivalent to .2nC1;0/ . t
u
7.2.4.2 The Orthogonal Group SO .4/
The orthogonal group

SO .4/ D g 2 GL .4; R/ W g> g D 1; det g D 1
is a six-dimensional
semisimple compact connected Lie group. Its Lie alge-
bra so .4/ D A 2 R44 W A> C A D 0 is isomorphic to so .3/ so .3/. Let
.E1 ; E2 ; E3 / denote the standard (ordered) basis for so .3/ (see the Appendix). The
map & W so .3/ so .3/ ! so .4/, given by

2 3
02 3 2 31 0 x3 y3 x2 y2 x1 y1
0 x3 x2 0 y3 y2 6 7
B6 7 6 7C 1 6x3 C y3 0 x1 C y1 x2 y2 7
B6 x3 0 x1 7 ; 6 y3 0 y1 7C 7! 6 7
@4 5 4 5A 2 6 7
4x2 C y2 x1 y1 0 x3 C y3 5
x2 x1 0 y2 y1 0
x1 C y1 x2 C y2 x3 y3 0
is a Lie algebra isomorphism. We define an ordered basis .E1 ; : : : ; E6 / for so .4/

by
Ei D & .Ei ; 0/; i D 1; 2; 3 Ej D & .0; Ej3 /; j D 4; 5; 6:
The group of inner automorphisms of so .4/ is given by

1 0
Int .so .4// D W 1; 2 2 SO .3/ :
0 2
The group of automorphisms .4// is generated by Int .so .4// and the
Aut.so
0 I3
swap automorphism D [3]. Moreover, the group of automorphisms
I3 0
decomposes as a semi-direct product Aut.so .4// D Int .so .4// f1; g of the
inner automorphisms (normal) and the two-element subgroup generated by the
involution .
Proposition 2 d Aut.SO .4// D Aut.so.4//.
Proof As Aut.so .4// is generated by Int .so .4// and , and clearly Int .so .4//
d Aut.SO .4//, it suffices to show that 2 d Aut.SO .4//.
The mapping W su .2/ su .2/ ! so .4/ given by
1
1

i
x
2 1
.ix3 C x2 / i
y1 .iy3 C y2 /
1
2 ; 1 2 2
2 .ix3 x2 / 2i x1 2 .iy3 y2 / 2i y1
2 3
0 x3 y3 x2 y2 x1 y1
166x3 C y3 0 x1 C y1 x2 y2 7 7
7! 4
2 x2 C y2 x1 y1 0 x3 C y3 5
x1 C y1 x2 C y2 x3 y3 0
is a Lie algebra isomorphism. Accordingly, as SU .2/SU .2/ is simply connected,

there exists a unique Lie group homomorphism q W SU .2/SU .2/ ! SO .4/ with
T1 q D (see, e.g., [53]). It is not difficult to see that q is in fact an epimorphism
and so q is a universal covering of SO .4/. Consequently, ker q is a subgroup
of the centre Z.SU.2/

SU.2// D f.I2 ; I2 /; .I2 ; I2 /; .I2 ; I2 /; .I2 ; I2 /g. Let
i 0
C D 2 su .2/. We have exp.C/ D I2 , exp. .C; 0// D exp.
0 i
.0; C// D I4 , and exp. .C; C// D I4 . Hence, as q.exp.A// D exp. A/
for A 2 su .2/ su .2/, it follows that ker q D f.I2 ; I2 /; .I2 ; I2 /g. Let
be
the unique automorphism of SU .2/ SU .2/ such that T1
D 1 . We
have that exp.T1
.C; C// D exp..C; C//. Hence
.ker q/ D ker q. It follows that
2 d Aut.SO .4// (see Proposition 5). t
u
Theorem 7 ([5], see also [3]) Any homogeneous system on SO .4/ is detached
feedback equivalent to exactly one of the following systems
.1;0/
W u1 .E1 C E4 /; 0661
.2;0/
1 W u1 E1 C u2 E4
.2;0/
2; W u1 .E1 C 1 E4 / C u2 .E2 C 2 E5 /;
1
.0 D 2 6 1 / _ .1 6 2
6 1 / _ .0 < 2 6 1 < 1/
.3;0/
1; W u1 .E1 C E4 / C u2 E2 C u3 E6 ; 0661
.3;0/
2; W u1 .E1 C 1 E4 / C u2 .E2 C 2 E5 / C u3 .E3 C 3 E6 /;
1
.0 D 3 6 2 6 1 / _ .0 < j3 j 6 2 < 1 ^ 2 6 1 / _ .2 D 1 6 j3 j 6 1 /
.4;0/
1 W u1 E2 C u2 E3 C u3 E5 C u4 E6
.4;0/
2; W u1 .E4 1 E1 / C u2 .E5 2 E2 / C u3 E3 C u4 E6 ;
1
.0 D 2 6 1 / _ .1 6 2
6 1 / _ .0 < 2 6 1 < 1/
.5;0/
W u1 .E4 E1 / C u2 E2 C u3 E3 C u4 E5 C u5 E6
.6;0/ W u1 E1 C u2 E2 C u3 E3 C u4 E4 C u5 E5 C u6 E6 :
Here and parametrize families of class representatives.

Remark 5 As SO .4/ is compact, a system on SO .4/ is controllable if and only
if it has full rank. For the homogeneous systems on SO .4/ we have the following.
No single input system has full rank; every five- or six-input system has full rank.
.2;0/ .2;0/
1 does not have full rank; 2; does not have full rank exactly when 2 D 0
.3;0/ .3;0/
or 1 D 2 D 1. 1; has full rank exactly when > 0; 2; does not have full
.4;0/ .4;0/
rank exactly when 1 D 2 D 3 D 1 or 2 D 0. 1 has full rank; 2; does
not have full rank exactly when 1 D 2 D 0.
Proof (Sketch) Any Plinearized automorphismP 2 d Aut.SO .4//

P preserves the dot
product A B D 6iD1 ai bi . (Here A D 6iD1 ai Ei and B D 6iD1 bi Ei .) Let ?
denote the orthogonal complement of a subspace so .4/. It is then easy to show
that, if ; e are subspaces of so .4/ and 2 Aut.so .4//, then D e if
? e ?
and only if D . The classification of the .6 `/-input systems therefore
follows from the classification of the `-input systems. Hence, we need only classify
the single-input, two-input, and three-input systems. In order to find normal forms
for the four- and five-input systems we need only take the orthogonal complements
of the traces of the two- and one-input systems, respectively. (The classification for
the six-input systems is trivial.) P
We write an `-input homogeneous system W u1 6iD1 bi1 Ei C C
P6
u` iD1 bi` Ei (in matrix form) as
2 1 3
b1 : : : b1`
6 :: 7 :
D 4 :::
M1
W :5
M2
b61 : : : b6`
Here M1 ; M2 2 R3` . The evaluation .1; u/ then becomes

a matrix multipli-
0
M1 M1
cation. Accordingly, two `-input homogeneous systems W and 0 W
M2 M20
are equivalent if and only if there
exist
an automorphism
0 2 d Aut.SO .4// and
M1 M1
K 2 GL .`; R/ such that K D . More precisely, and 0 are
M2 M20
equivalent if and only if there exist R1 ; R2 2 SO .3/ and K 2 GL .`; R/ such that

R1 M1 K D M10 and R2 M2 K D M20

or R1 M2 K D M10 and R2 M1 K D M20 :
The singular value decomposition (see, e.g., [80]) plays a key role in our
argument. For any matrix M 2 Rmn of rank r, there exist orthogonal matrices
U 2 Rmm , V 2 R
nn
and a diagonal matrix D 2 R
rr
D diag. 1 ; : : : ; r /
D0 >
such that M D U V with 1 > > r > 0. Using this result,
0 0
together with the above characterization of equivalence (in matrix form), it is
fairly straightforward to compute normal forms for the one-, two-, and three-input
homogeneous systems on SO .4/. t
u
Note 3 The inhomogeneous systems on SO .4/ were also classified in [5]; the
approach is similar though the computations are more involved.
7.3 Invariant Optimal Control
We consider the class of left-invariant optimal control problems on Lie groups with
fixed terminal time, affine dynamics, and affine quadratic cost. Formally, such a
problem is written as
gP D g .A C u1 B1 C C u` B` / ; g 2 G; u 2 R` (7.3)
g.0/ D g0 ; g.T/ D g1 (7.4)
Z T
J D .u.t/ /> Q .u.t/ / dt ! min: (7.5)
0
Here G is a (real, finite-dimensional, connected) Lie group with Lie algebra g, A,

B1 ; : : : ; B` 2 g (with B1 ; : : : ; B` linearly independent), u D .u1 ; : : : ; u` / 2 R` ,
2 R` , and Q is a positive definite ` ` matrix. To each such problem, we
associate a cost-extended system .; /. Here is the control system (7.3) and
the cost function W R` ! R has the form .u/ D .u /> Q .u /. Each cost-
extended system corresponds to a family of invariant optimal control problems; by
specification of the boundary data .g0 ; g1 ; T/, the associated problem is uniquely
determined.
Optimal control problems of this kind have received considerable attention in
the last few decades. Various physical problems have been modelled in this manner
(cf. Sect. 7.1.2), such as optimal path planning for airplanes, motion planning for
wheeled mobile robots, spacecraft attitude control, and the control of underactuated
underwater vehicles [77, 89, 100]; also, the control of quantum systems and the
dynamic formation of DNA [48, 54]. Many problems (as well as sub-Riemannian
structures) on various low-dimensional matrix Lie groups have been considered by
a number of authors (see, e.g., [22, 23, 41, 63, 67, 69, 81, 83, 88, 92, 93]).
We introduce a form of equivalence for problems of the form (7.3)(7.4)(7.5),
or rather, for the associated cost-extended systems (cf. [27, 35]). Cost equivalence
establishes a one-to-one correspondence between the associated optimal trajecto-
ries, as well as the associated extremal curves. We associate to each cost-extended
system, via the Pontryagin Maximum Principle, a quadratic HamiltonPoisson
systems on the associated LiePoisson space. We show that cost equivalence of cost-
extended systems implies equivalence of the associated Hamiltonian systems. In
addition, we reinterpret drift-free cost-extended systems (with homogeneous cost)
as invariant sub-Riemannian structures.
A number of illustrative examples concerning classification (some quite exten-
sive) are included.
7.3.1 Pontryagin Maximum Principle
The Pontryagin Maximum Principle provides necessary conditions for optimality

which are naturally expressed in the language of the geometry of the cotangent
bundle T G of G (see [10, 51, 64]). The cotangent bundle T G can be trivialized
(from the left) such that T G D G g ; here g is the dual of the Lie algebra g.
To an optimal control problem (7.3)(7.4)(7.5) we associate, for each real number
and each control parameter u 2 R` a Hamiltonian function on T G D G g :
Hu ./ D .u/ C .u .g//

D .u/ C p .u .1//; D .g; p/ 2 T G: (7.6)
#
We denote by H u the corresponding Hamiltonian vector field (with respect to the
symplectic structure on T G).
Lemma 1 ([64], see also [10]) A curve ./ D .g./; p.// is an integral curve of
#
(the time varying Hamiltonian vector field) H u.t/ if and only if
gP .t/ D .g.t/; u.t// pP .t/ D ad u.t/ .1/ p.t/:
Here .ad A p/.B/ D p.A; B/ for A; B 2 g and p 2 g .

In terms of the above Hamiltonians, the Maximum Principle can be stated as
follows.
Theorem 8 (Maximum Principle) Suppose the controlled trajectory .Ng./; uN .//
defined over the interval 0; T is a solution for the optimal control problem (7.11).
Then, there exists a curve ./ W 0; T ! T G with .t/ 2 TgN.t/ G; t 2 0; T,
and a real number 6 0, such that the following conditions hold for almost every
t 2 0; T :
.; .t// 6 .0; 0/ (7.7)

P DH #
.t/ uN .t/ ..t// (7.8)
HuN .t/ ..t// D max Hu ..t// D constant: (7.9)

u
An optimal trajectory, g./ W 0; T ! G is the projection of an integral curve

#
./ of the (time-varying) Hamiltonian vector field H uN .t/ . A trajectory-control pair
../; u.// is said to be an extremal pair if ./ satisfies the conditions (7.7)(7.9).
The projection ./ of an extremal pair is called an extremal. An extremal curve is
called normal if < 0 and abnormal if D 0.
For the class of optimal control problems under consideration, the maximum
condition (7.9) eliminates the parameter u from the family of Hamiltonians .Hu /;
as a result, we obtain a smooth G-invariant function H on T G D G g . This
HamiltonPoisson system on T G can be reduced to a HamiltonPoisson system
on the (minus) LiePoisson space g , with Poisson bracket given by
fF; Gg D p.dF. p/; dG. p//:
Here F; G 2 C1 .g / and dF. p/; dG. p/ are elements of the double dual g
which is canonically identified with the Lie algebra g.
7.3.2 Equivalence of Cost-Extended Systems
Let .; / and . 0 ; 0 / be two cost-extended systems. .; / and . 0 ; 0 / are

said to be cost equivalent if there exist a Lie group isomorphism
W G ! G0 and
an affine isomorphism ' W R` ! R` such that
Tg
.g; u/ D 0 .
.g/; '.u// and 0 ' D r
for g 2 G, u 2 R` and some r > 0. Equivalently, .; / and . 0 ; 0 / are cost

equivalent if and only if there exist a Lie group isomorphism
W G ! G0 and
an affine isomorphism ' W R` ! R` such that T1
.1; u/ D 0 .10 ; '.u// and
0 ' D r for some r > 0. Accordingly:
If .; / and . 0 ; 0 / are cost equivalent, then and 0 are detached
feedback equivalent.
If two full-rank systems and 0 are state space equivalent, then .; / and
. 0 ; / are cost equivalent for any cost .
If two full-rank systems and 0 are detached feedback equivalent with
respect to a feedback transformation ', then .; '/ and . 0 ; / are cost
equivalent for any cost .
Remark 6 The cost-preserving condition 0 ' D r is partially motivated by
the following considerations. Each cost on R` induces a strict partial ordering
u < v .u/ < .v/. It turns out that and 0 induce the same strict
partial ordering on R` if and only if D r0 for some r > 0. The dynamics-
preserving condition Tg
.g; u/ D 0 .
.g/; '.u// is just that of detached
feedback equivalence (on full-rank systems).
Let .g./; u.// be a controlled trajectory, defined over an interval 0; T, of a
cost-extended system .; /. We say that .g./; u.// is a virtually optimal con-
trolled trajectory (shortly VOCT) if it is a solution for the associated optimal control
problem with boundary data .g.0/; g.T/; T/. Similarly, we say that .g./; u.// is an
extremal controlled trajectory (shortly ECT) if it satisfies the necessary conditions
of the Pontryagin Maximum Principle (with 6 0). Clearly, any VOCT is an ECT.
A map
' defining a cost equivalence between two cost-extended systems
establishes a one-to-one correspondence between their respective VOCTs (and
ECTs).
Theorem 9 ([27, 35]) Suppose

' defines a cost equivalence between .; /
and . 0 ; 0 /. Then
1. .g./; u.// is a VOCT if and only if .
g./; ' u.// is a VOCT;
2. .g./; u.// is an ECT if and only if .
g./; ' u.// is an ECT.
Proof
(1) Suppose .g./; u.// is a controlled trajectory of .; / and .
g./; ' u.//
is a VOCT of . 0 ; 0 /. Further suppose .g./; u.// is not a VOCT of .; /.
Then there exists another controlled trajectory .h./; v.// such that h.0/ D
g.0/, h.T/ D g.T/, and
Z T Z T
J .v.// D .v.t// dt < .u.t// dt D J .u.//:
0 0

Hence
h./; ' v./ is a controlled trajectory of . 0 ; 0 / such that (for
some r > 0)
Z T Z T
J 0 .' v.// D .0 '/.v.t// dt D r .v.t// dt
0 0
Z T Z T
<r .u.t// dt D .0 '/.u.t// dt D J 0 .' u.//:
0 0
However, this contradicts the fact that .

g./; ' u.// is a VOCT of . 0 ; 0 /.
Thus if .
g./; ' u.// is a VOCT, then so is .g./; u.//. As
1 ' 1
defines a cost equivalence between . 0 ; 0 / and .; /, the same argument
can be used to show the converse.
(2) The Hamiltonian functions (7.6), associated to .; / and . 0 ; 0 /, are given
by
Hu .g; p/ D p.u .1// C .u/ and Hu0 0 .g0 ; p0 / D p0 .u0 0 .1// C 0 .u/
respectively (for a fixed ). Assume .g./; u.// is a controlled trajectory of

such that .g0 ./; u0 .// D .
g./; ' u.// is an ECT of . 0 ; 0 /. Then there
exists p0 ./ W 0; T ! g such that . 0 ./; u0 .//, 0 .t/ D .g0 .t/; p0 .t// satisfies
#
(7.7)(7.8)(7.9). As P 0 .t/ D H 0u.t/ . 0 .t// (7.8) we have (see Lemma 1)
gP 0 .t/ D 0 .g0 .t/; u0 .t// pP 0 .t/ D ad u0 0 .t/ .1/ p0 .t/:
Let p./ D 1r .T1

/ p0 ./ and .t/ D .g.t/; p.t//. (Here r > 0 is the unique
constant associated to
'.)
We show that ../; u.// satisfies (7.8)(7.9). By assumption we have that
gP .t/ D .g.t/; u.t//. Thus, in order to satisfy (7.8), we are left to show that
pP .t/ D ad u.t/ .1/ p.t/. Indeed, for A 2 g we have

1

pP .t/ A D .T1
/ pP 0 .t/ A
r

D 1r ad u0 0 .t/ .1/ p0 .t/ .T1
A/

D 1r p0 .t/ T1
u.t/ .1/; T1
A

D ad u.t/ .1/ p.t/ A:
In order to show that ../; u.// satisfies (7.9), we first show that Hu ..t// D
1 0 0
r H'.u/ . .t// for u 2 R . Indeed,
k
Hu ..t// D p.t/ u .1/ C .u/

D . 1r .T1
/ p0 .t// u .1/ C .u/
D 1r p0 .t/ '.u/
0
.1/ C 1r 0 .'.u//
0
D 1r H'.u/ . 0 .t//:
Hence, it follows that
Hu.t/ ..t// D 1r Hu0 0 .t/ . 0 .t//

D max 1r Hu0 0 0 .t/ .D constant/
0
u0 2Rk
0
0
D max 1r H'.u/ .t/
u2Rk
D max Hu ..t// D constant

u2Rk
It is left to show that (7.7) holds. If .g0 ./; u0 .// is normal, then < 0 and
so .; .// 6 0. Suppose .g0 ./; u0 .// is abnormal. We have that .T1
/ is
injective and as .; 0 .// 6 0, i.e., p0 ./ 6 0, it follows that p./ D 1r .T1
/
p0 .t/ 6 0, i.e., .; .// 6 0. Again, the converse may be found by a similar
argument (utilizing
1 ' 1 ). t
u
7.3.2.1 Classification
The classification of a given class of cost-extended systems may be approached

as follows. First, we classify the underlying systems under detached feedback
equivalence and obtain normal forms i , i 2 I. Any cost-extended system must
then be equivalent to a cost-extended system .i ; / for some i 2 I and some
cost . Note that .i ; / and .j ; 0 / cannot be equivalent unless i D j. Therefore
the classification problem reduces to finding normal forms of the cost for each
i . Accordingly, we wish to characterize cost equivalence between cost-extended

systems for which the underlying system is identical.
Let .; / and .; 0 / be two cost-extended systems (with identical underlying
systems). Let T be the group of feedback transformations leaving invariant.
More precisely,
T D f' 2 Aff .R` / W 9 2 d Aut.G/; D ; .1; u/ D .1; '.u//g:
(Here Aff .R` / is the group of affine isomorphisms of R` .) The following result is
easy to prove.
Proposition 3 .; / and .; 0 / are cost equivalent if and only if there exists an
element ' 2 T such that 0 D r ' for some r > 0.
Four examples of the classification of cost-extended systems (on the Euclidean,
semi-Euclidean and Heisenberg groups) are exhibited. In the first two cases the
classification procedure (as outlined above) is carried out in some detail.
Example 3 ([27]) On SE .2/, any full-rank two-input drift-free cost-extended
system .; / with homogeneous cost (i.e., .0/ D 0) is cost equivalent to
(
W u 1 E2 C u 2 E3
.2;0/ .2;0/
. ; / W
.u/ D u21 C u22 :
Proof Let .; / be a cost-extended system as prescribed. By Theorem 3 and the

fact that d Aut.SE .2// D Aut.se .2// (see Example 1), we have that is detached
feedback equivalent to the system .2;0/ . Moreover, it turns out that the feedback
transformation ' for this equivalence must be linear. Thus .; / is cost equivalent
to a cost-extended system . .2;0/ ; 0 / for some 0 W u 7! u> Q0 u.
A straightforward calculation shows that the group T .2;0/ of feedback transfor-
mations leaving .2;0/ invariant is given by

x w
T .2;0/ D u 7! u W x 0; w 2 R; D 1 :
0
" #
0 a1 b 1 ab1
Let Q D . Now '1 D 2 T1 and .0 '1 /.u/ D
b a2 0 1
q 0
2 2 a
u> diag.a1 ; a2 ba1 / u. Let a02 D a2 ba1 and let '2 D diag. a21 ; 1/ 2 T1 .
Then .0 .'1 '2 //.u/ D a02 u> u D a02 .2;0/ .u/. Consequently, by Proposition 3,
.; / is cost equivalent to . .2;0/ ; .2;0/ /. t
u
Example 4 ([21, 35]) Any controllable two-input inhomogeneous cost-extended

system on H3 is cost equivalent to exactly one of
(
.2;1/ W E1 C u1 E2 C u2 E3
.2;1/
. ; .2;1/
/ W
.2;1/
.u/ D .u1 /2 C u22 :
Here > 0 parametrizes a family of (non-equivalent) class representatives.

Proof Let .; / be a cost-extended system as prescribed. By Theorem 6, is
detached feedback equivalent to the system .2;1/ . Thus, .; / is cost equivalent
to . .2;1/ ; 0 / for some cost 0 .
As H3 is simply connected, d Aut.H3 / D Aut.h3 /. A simple calculation shows
that T .2;1/ D SL .2; R/. Now
0 W u 7! .u /> Q .u / for some positive
a b .2;1/
definite matrix Q D 1 and 2 R2 . We transform 0 into multiple of
b a2
" #
1 ab1
by successively composing with elements of T .2;1/ . We have '1 D 2
0 1
T .2;1/ and
b2
1 .u/ D .0 '1 /.u/ D .u 0 /> diag.a1 ; a2 a1
/ .u 0 /
2 p
for some 0 2 R2 . Let a02 D a2 ba1 Then '2 D 4
a1 a02 diag p1 ; p1 0 2 T .2;1/
a1 a2
and
2 .u/ D p 1 .1 '2 /.u/ D .u 00 /> .u 00 /

a1 a02
for some 00 2 R2 . There exist > 0 and 2 R such that 001 D cos and

00 cos sin
2 D sin . Hence '3 D 2 T .2;1/ and
sin cos
>

.2;1/
.u/ D .2 ' 3 /.u/ D u u :
0 0
.2;1/
Therefore D p1 0 .'1 '2 '3 /. Consequently, by Proposition 3, .; /
a1 a02
.2;1/ .2;1/
is cost equivalent to . .2;1/ ; /. A simple verification shows that '
.2;1/
r0 for any 0 , r > 0, and ' 2 T .2;1/ . t
u
Example 5 ([18]) On SE .1; 1/, any controllable drift-free cost-extended system
.; / with homogeneous cost (i.e., .0/ D 0) is cost equivalent to exactly one of
the systems
(
.2;0/ W u1 E1 C u2 E3
.2;0/ .2;0/
. ; / W
.2;0/ .u/ D u21 C u22
8
< .3;0/ W u1 E1 C u2 E2 C u3 E3
.3;0/
. .3;0/ ; / W
: .3;0/ .u/ D u2 C u2 C u2 :
1 2 3
Here 0 < 6 1 parametrizes a family of cost-extended systems, with each distinct

value corresponding to a distinct (non-equivalent) class representative.
Example 6 ([24]) On H2nC1 , any drift-free controllable cost-extended system with
homogeneous cost (i.e., .0/ D 0) is cost-equivalent to exactly one of the following
cost-extended systems:
8 Pn
< .2n;0/ W iD1 .ui Xi C unCi Yi /
.2n;0/
. .2n;0/ ; / W
: .2n;0/ .u/ D Pn u2 C u2
iD1 i i nCi
8 P
< .2nC1;0/ W n
iD1 .ui Xi C unCi Yi / C u2nC1 Z
.2nC1;0/ .2nC1;0/
. ; / W
: .2nC1;0/ .u/ D n u2 C u2 C u2 :
P
iD1 i i nCi 2nC1
Here 1 D 1 > 2 > > n > 0 parametrize families of (non-equivalent) class

representatives.
7.3.3 Pontryagin Lift
To any cost-extended system .; / on a Lie group G we associate, via the

Pontryagin Maximum Principle, a HamiltonPoisson system on the associated Lie
Poisson space g (cf. [10, 64, 94]). We show that equivalence of cost-extended
systems implies equivalence of the associated HamiltonPoisson systems.
Note 4 The Pontryagin lift may be realized as a contravariant functor between the
category of cost-extended control systems and the category of HamiltonPoisson
systems ([35], see also [51]).
A quadratic HamiltonPoisson system .g ; HA;Q / is specified by
HA;Q W g ! R; p 7! p.A/ C Q. p/:
Here A 2 g and Q is a quadratic form on g . If A D 0, then the system is

called homogeneous; otherwise, it is called inhomogeneous. (When g is fixed, a
system .g ; HA;Q / is identified with its Hamiltonian HA;Q .) To each function H 2
# #
C1 .g /, we associate a Hamiltonian vector field H on g specified by HF D
1
fF; Hg. A function C 2 C .g / is a Casimir function if fC; Fg D 0 for all F 2
#
C1 .g /, or, equivalently C D 0. A linear map W g ! h is a linear Poisson
morphism if fF; Gg D fF ; G g for all F; G 2 C1 .h /. Linear Poisson
morphisms are exactly the dual maps of Lie algebra homomorphisms.
Let .E1 ; : : : ; En / be a ordered basis for the Lie algebra g and let .E1 ; : : : ; En /
denote the corresponding dual basis for g . We write elements B 2 g as column
vectors and elements p 2 g as row vectors. The equations of motion for the
#
integral curve p./ of the Hamiltonian vector field H corresponding to H 2
1
C .g / then take the form pP i D p.Ei ; dH. p//. Whenever convenient, linear
maps will be identified with their matrices. Writing elements u 2 R` as column
we can express u .1/ D ACu1 B1 C Cu` B` as u .1/ D ACB u,
vectors as well,
where B D B1 B` is a n ` matrix.
Let .; / be a cost-extended system with
u .1/ D A C B u; .u/ D .u /> Q.u /:
By the Pontryagin Maximum Principle we have the following result.

Theorem 10 ([27]) Any normal ECT .g./; u.// of .; / is given by
gP .t/ D .g.t/; u.t//; u.t/ D Q1 B> p.t/> C
where p./ W 0; T ! g is an integral curve for the HamiltonPoisson system on

g specified by
H. p/ D p .A C B / C 1
2
p B Q1 B> p> : (7.10)
Proof The Hamiltonian (7.6) is given by Hu .g; p/ D p ACp B uC .u/> Q .u
/. Now, @H >
@u .g; p/ D p B C 2.u / Q. Hence the maximum maxu Hu .g; p/
u
1
occurs at u> D 2 p B Q1 C > (for a normal ECT we have < 0). Therefore
the maximized Hamiltonian is given by
H.g; p/ D max Hu .g; p/

u2Rk
1 1 > >
D p A C p B . 2 Q B p C / C
42
p B Q1 Q Q1 B> p>
D p .A C B / 1
4
p B Q1 B> p> :
As the maximized Hamiltonian H is defined and smooth, it follows that

P #
if ../; u.// satisfies (7.7)(7.8)(7.9), then .t/ D H..t// and u./ D
1
2 Q1 B> p./ C ; moreover, if ./ D .g./; p.// is an integral curve of
# 1
H, then the pair ../; uQ .//, uQ ./ D 2 Q1 B> p./ C satisfies (7.7)(7.8)
(7.9) (see [10]). Accordingly the normal ECTs are the projection .g./; p.// of the
1
pairs ../; u.// where u./ D 2 Q1 B> p./ C and ./ D .g./; p.// is an
#
integral curve of the Hamiltonian vector field H on T G D G g . However, as
the Hamiltonian H is G-invariant, the integral curves .g./; p.// of H are given
by (see, e.g., [64])
gP .t/ D g.t/ dH. p.t// and pP .t/ D ad dH. p.t// p.t/
where H is viewed as a function on g and dH. p/ 2 g is canonically identified

with an element in g. Accordingly we get gP .t/ D .g.t/; u.t// where u.t/ D
Q B> p.t/> C and p.t/ is an integral curve of the Hamiltonian system
1 1
2

.g ; H/. Finally, as the pair .; .// can be multiplied by any positive number
[10], we take D 12 for convenience. t
u
We say that two quadratic HamiltonPoisson systems .g ; G/ and .h ; H/ are
linearly equivalent if there exists a linear isomorphism W g ! h such that the
# # # #
Hamiltonian vector fields G and H are -related, i.e., Tp G. p/ D H. . p//

for p 2 g .
Proposition 4 The following pairs of HamiltonPoisson systems (on g , specified
by their Hamiltonians) are linearly equivalent:
1. HA;Q and HA;Q , where W g ! g is a linear LiePoisson automorphism;
2. HA;Q and HA;rQ , where r 0;
3. HA;Q and HA;Q C C, where C is a Casimir function.
Proof
#
(1) As is a (linear) Poisson isomorphism, it follows that T HA;Q D
# # #
H A;Q (see, e.g., [79]), i.e., HA;Q D H A;Q .
#
(2) Let p./ be an integral curve of H A;Q , i.e., suppose pP i .t/ D p.t/ Ei ; A C
p.t/> Q. Let pN ./ D 1r p./. Then pPN i .t/ D 1r p.t/ Ei ; A C p.t/> Q D Np.t/
# #
Ei ; A C pN .t/> rQ. Thus pN ./ is an integral curve of H A;rQ . Thus, H A;Q and
#
H A;rQ are compatible with the map 1=r W p 7! 1r p (as 1=r maps the flow of
# #
H A;Q to the flow of H A;rQ ).
# #
(3) We have fHA;Q CC; Fg D fHA;Q ; Fg for F 2 C1 .g /. Thus HA;Q C C D H A;Q ,
i.e., these vector fields are compatible with the identity map. t
u
Theorem 11 If two cost-extended systems are cost equivalent, then their associated
HamiltonPoisson systems, given by (7.10), are linearly equivalent.
Proof Let .; / and . 0 ; 0 / be cost-extended systems with u .1/ D A C B u
and u0 .1/ D A0 C B0 u0 , respectively. The associated HamiltonPoisson systems
(on g and .g0 / , respectively) are given by
H.;/ . p/ D p .A C B / C 12 p B Q1 B> p>

H. 0 ;0 / . p/ D p .A0 C B0 0 / C 12 p B0 Q01 B0> p> :
Suppose
' defines a cost equivalence between .; / and . 0 ; 0 /, where
'.u/ D Ru C '0 and R 2 R`` . We have 0 ' D r for some r > 0. A simple
calculation yields
T1
A D A0 C B0 '0 ; R C '0 D 0 ; T1
B D B0 R; R Q1 R> D r .Q0 /1 :
Thus .H.;/ .T1

/ /. p/ D p .A0 C B0 0 / C 2r p B0 .Q0 /1 B0> p> . Here .T1
/ W
.g0 / ! g is the dual of the linear map T1
. Hence, the vector fields associated
with H. 0 ;0 / and H.;/ .T1
/ , respectively, are related by the dilation 1=r W
.g0 / ! .g0 / , p 7! 1r p (Proposition 4). Moreover, the vector fields associated
with H.;/ .T1
/ and H.;/ , respectively, are related by the linear Poisson
isomorphism .T1
/ (Proposition 4). Consequently 1r .T1
/ defines a linear
equivalence between ..g0 / ; H. 0 ;0 / / and .g ; H.;/ /. t
u
Remark 7 The converse of the above statement is not true in general. In fact, one
can construct cost-extended systems with different number of inputs but equivalent
Hamiltonians (cf. [35]).
Accordingly, from a classification of cost-extended systems, one can generally
get a (partial) classification of the corresponding quadratic HamiltonPoisson
systems. For instance, we have the following result.
Example 7 (cf. Example 6) Any quadratic HamiltonPoisson systems ..h3 / ; H/,
where H is a positive definite quadratic form, is linearly equivalent to the system
on .h3 / with Hamiltonian H 0 . p/ D 12 . p21 C p22 C p23 /.
Proof Let ..h3 / ; H/ be a quadratic HamiltonPoisson system with Hamiltonian
H. p/ D p Q p> being a positive-definite quadratic form. For the cost-extended
system .; / on H3 specified by

0 0 0 W u1 X1 C u2 Y1 C u3 Z D u1 E2 C u2 E3 C u3 E1
; W
.u/ D 12 u> Q1 u
we have that the associated Hamiltonian system (7.10) is ..h3 / ; H/. Furthermore,
by Example 6, we have that .; / is cost equivalent to the system
(
.2;0/ W u1 Y1 C u2 X1 C u3 Z
.2;0/
.2;0/ ; 1 W .2;0/
1 .u/ D u21 C u22 C u23
which has associated Hamiltonian system ..h3 / ; H 0 /. Consequently, by Theo-

rem 11, we have that ..h3 / ; H/ is L-equivalent to ..h3 / ; H 0 /. t
u
Remark 8 It is a simple matter to expand the above result to positive definite
quadratic HamiltonPoisson systems on .h2nC1 / .
7.3.4 Sub-Riemannian Structures
Left-invariant sub-Riemannian (and, in particular, Riemannian) structures on Lie

groups can naturally be associated to drift-free cost-extended systems with homo-
geneous cost. We show that if two cost-extended systems are cost equivalent, then
the associated sub-Riemannian structures are isometric up to rescaling. At least
for some classes (namely Riemannian structures on nilpotent groups and sub-
Riemannian Carnot groups), the converse holds as well.
A left-invariant sub-Riemannian manifold is a triplet .G; D; g/, where G is a
(real, finite dimensional) connected Lie group, D is a nonintegrable left-invariant
distribution on G, and g is a left-invariant Riemannian metric on D. More
precisely, D.1/ is a linear subspace of the Lie algebra g of G and D.g/ D
gD.1/; the metric g1 is a positive definite symmetric bilinear from on g and
gg .gA; gB/ D g1 .A; B/ for A; B 2 g, g 2 G. When D D TG (i.e., D.1/ D g)
then one has a left-invariant Riemannian structure. An absolutely continuous curve
g./ W 0; T ! G is called a horizontal curve if gP .t/ 2 D.g.t// for almost all
t 2 0; T. We shall assume that D satisfies the bracket generating condition, i.e.,
D.1/ has full rank; this condition is necessary and sufficient for any two points in
G to be connected by a horizontal curve. An isometry between two left-invariant
sub-Riemannian manifolds .G; D; g/ and .G0 ; D 0 ; g0 / is a diffeomorphism
W
G ! G0 such that
D D D 0 and g D
g0 , i.e., Tg
D.g/ D D 0 .
.g// and
gg .gA; gB/ D g0
.g/.Tg
gA; Tg
gB/ for g 2 G and A; B 2 g. If the isometry
is a Lie group isomorphism, we shall say it is an L-isometry.

A standard argument shows that the length minimization problem
gP .t/ 2 D.g.t//; g.0/ D g0 ; g.T/ D g1 ;

Z Tp
g.Pg.t/; gP .t// ! min
0
is equivalent to the energy minimization problem, or invariant optimal control

problem:
gP D u .g/; u 2 R` g.0/ D g0 ; g.T/ D g1

Z T (7.11)
.u.t// dt ! min:
0
Here u .1/ D u1 B1 C C u` B` where B1 ; : : : ; B` are some linearly independent

elements of the Lie algebra g such that hB1 ; : : : ; B` i D D.1/; .u.t// D
u.t/> Q u.t/ D g1 .u.t/ .1/; u.t/ .1// for some ` ` positive definite (symmetric)
matrix Q. More precisely, energy minimizers are exactly those length minimizers
which have constant speed. In other words, the VOCTs of the cost-extended system
.; / associated with (7.11) are exactly the (constant speed) minimizing geodesics
of the sub-Riemannian structure .G; D; g/; the normal (resp. abnormal) ECTs of
.; / are the normal (resp. abnormal) geodesics of .G; D; g/.
Accordingly, to a (full-rank) cost-extended system .; / on G of the form
W u 1 B1 C C u ` B` ; .u/ D u> Qu
we associate a sub-Riemannian structure .G; D; g/ specified by
D.1/ D D hB1 ; : : : ; B` i ; g1 .u1 B1 C C u` B` ; u1 B1 C C u` B` / D .u/:
Let .G; D; g/ and .G0 ; D 0 ; g0 / be two sub-Riemannian structures associated to

.; / and . 0 ; 0 /, respectively.
Theorem 12 .; / and . 0 ; 0 / are cost equivalent if and only if .G; D; g/ and
.G0 ; D 0 ; g0 / are L-isometric up to rescaling.
Proof Suppose
' defines a cost equivalence between .; / and . 0 ; 0 /, i.e.,
0
u D '.u/ and 0 ' D r for some r > 0. As T1

u .1/ D '.u/ .1/, it
follows that T1
D.1/ D D 0 .1/. Hence, as
is a Lie group isomorphism, by left
invariance we have
D D D 0 . Furthermore
r .u/ D 0 .'.u//
r g1 .u .1/; u .1// D g01 .'.u/
0 0
.1/; '.u/ .1//
r g1 .u .1/; u .1// D g01 .T1
u .1/; T1
u .1//: (7.12)
Hence, as
is a Lie group isomorphism, by left invariance we have r g D
g0 .
Conversely, suppose
D D D 0 and g D r
g0 . We have T1
D.1/ D D 0 .1/
and so T1
D 0 . Hence there exists a unique linear map ' W R` ! R` such
0
that T1
u .1/ D '.u/ .1/. Thus
' defines a detached feedback equivalence
between and . By (7.12), it follows that 0 ' D r. Thus .; / and
0
. 0 ; 0 / are cost equivalent. t

u
Clearly every L-isometry is an isometry. For some classes of sub-Riemannian
structures, every isometry is the composition of a left translation and an L-
isometry. Hence, for these classes, if .G; D; g/ and .G0 ; D 0 ; g0 / are isometric (up
to rescaling), then .; / and . 0 ; 0 / are cost equivalent.
Theorem 13 (cf. [102], also [74]) Let .G; g/ and .G0 ; g0 / be two invariant
Riemannian structures on simply connected nilpotent Lie groups G and G0 ,
respectively. A diffeomorphism
W G ! G0 is an isometry between .G; g/ and
.G0 ; g0 / if and only if
is the composition
D L
.1/
0 of a left translation
L
.1/ on G0 and a Lie group isomorphism
0 D L
.1/1
W G ! G0 such that
g1 .A; B/ D g010 .T1
0 A; T1
0 B/.
Corollary 1 Two invariant Riemannian structures on simply connected nilpotent

Lie groups are isometric if and only if they are L-isometric.
A k-step Carnot group G is a simply connected nilpotent Lie group whose Lie
algebra g has stratification g D g1 g2 gk with g1 ; gj D g1Cj , j D
1; : : : ; k1 and gj ; gk D f0g, j D 1; : : : ; k. Let D be the left-invariant distribution
on G specified by D.1/ D g1 . Once we fix a left-invariant metric g on D we have
a left-invariant structure on G (note that D is bracket generating). We shall refer to
such a structure as a sub-Riemannian Carnot group. We likewise have the following
characterization of the isometries between such structures.
Theorem 14 (cf. [56, 70], see also [44, 76]) Let .G; D; g/ and .G0 ; D 0 ; g0 / be
two sub-Riemannian Carnot groups. A diffeomorphism
W G ! G0 is an isometry
between .G; D; g/ and .G0 ; D 0 ; g0 / if and only if
is the composition
D L
.1/
0 of a left translation L
.1/ on G0 and a Lie group isomorphism
0 D L
.1/1
W
G ! G0 such that T1
0 D.1/ D D 0 .10 / and g1 .A; B/ D g010 .T1
0 A; T1
0 B/.
Corollary 2 Two sub-Riemannian Carnot groups are isometric if and only if they
are L-isometric.
Remark 9 Among the three-dimensional simply connected Lie groups, only the
Abelian R3 and Heisenberg H3 groups are nilpotent. Among the four-dimensional
simply connected Lie groups, only the Abelian group R4 , the trivial extension
H3 R of H3 , and the Engel group are nilpotent (see, e.g., [82, 86, 87]). Except for
the Abelian groups, all these groups admit sub-Riemannian Carnot group structures.
Analogous to Example 3, we have the following classification of sub-Riemannian
structures on the Euclidean group SE .2/.
Example 8 On SE .2/, any left-invariant sub-Riemannian structure .D; g/ is
N gN / specified by
isometric (up to rescaling) to the structure .D;
8
D.1/ D hE2 ; E3 i
<

10
: g1 D
01
i.e., with orthonormal frame .E2 ; E3 /.

Analogous to Example 6, we have the following classification of structures on
the Heisenberg groups.
Example 9 ([24]) Any left-invariant sub-Riemannian structure .D; g/ on H2nC1
is isometric to exactly one of the structures .D; g / specified by
(
D.1/ D hX1 ; Y1 ; : : : ; Xn ; Yn i
g1 D D diag.1 ; 1 ; 2 ; 2 ; : : : ; n ; n /
i.e., with orthonormal frame

p1 X1 ; p1 Y1 ; p1 X2 ; p1 Y2 ; : : : ; p1 Xn ; p1 Yn :
1 1 2 2 n n
Here 1 D 1 > 2 > > n > 0 parametrize a family of (non-equivalent) class

representatives.
Example 10 ([24]) Any left-invariant Riemannian structure g on H2nC1 is isomet-
ric to exactly one of the structures

1 0
g1 D ; D diag.1 ; 1 ; 2 ; 2 ; : : : ; n ; n /
0
i.e., with orthonormal frame
.Z; p1 X1 ; p1 Y1 ; p1 X2 ; p1 Y2 ; : : : ; p1 Xn ; p1 Yn /:

1 1 2 2 n n
Here 1 > 2 > > n > 0 parametrize a family of (non-equivalent) class

representatives.
7.4 Quadratic HamiltonPoisson Systems
The dual space of a Lie algebra admits a natural Poisson structure, namely the
LiePoisson structure; these structures are in a one-to-one correspondence with
linear Poisson structures [73]. LiePoisson structures arise naturally in a variety of
fields of mathematical physics and engineering such as classical dynamical systems,
robotics, fluid dynamics, and superconductivity, to name but a few. Many interesting
dynamical systems are naturally expressed as quadratic HamiltonPoisson systems
on LiePoisson spaces; prevalent examples are Eulers classic equations for the rigid
body, its extensions and its generalizations [59, 60, 64, 79, 97].
As described in Sect. 7.3 (Theorem 10), quadratic HamiltonPoisson systems
arise in the study of invariant optimal control problems. In this vein, a number
of quadratic HamiltonPoisson systems on lower-dimensional LiePoisson spaces
have been considered (see, e.g., [4, 7, 23, 47, 81, 93]). Several classes of quadratic
(and especially homogeneous) systems have also recently been studied in their own
right (see, e.g., [1, 6, 8, 1215, 20, 32, 97, 98]). Equivalence of quadratic Hamilton
Poisson systems on LiePoisson spaces has been investigated by a few authors
[6, 8, 20, 27, 35, 49, 97]; in these papers, two systems are considered equivalent
if their Hamiltonian vector fields are compatible with either an affine isomorphism,
a linear isomorphism, or a LiePoisson isomorphism.
In this section, we shall give a brief overview of our classification of the

(positive semidefinite) homogeneous systems in three dimensions; a classification
of the inhomogeneous systems on so .3/ is also exhibited. Finally, for a typical
inhomogeneous system (on the Euclidean space se .2/ ), we determine the stability
nature of its equilibria and find explicit expressions for some of its integral curves.
7.4.1 Classification in Three Dimensions

7.4.1.1 Homogeneous Systems
We consider those three-dimensional LiePoisson spaces that admit global Casimir

functions. Up to linear Poisson isomorphisms, this class is comprised of
.g2:1 g1 / ; .g3:1 / ; .g03:4 / ; .g03:5 / ; .g3:6 / ; and .g3:7 / :
We present a classification of the positive semidefinite homogeneous quadratic

HamiltonPoisson systems on these spaces, i.e., Hamiltonians of the form H. p/ D
p> Q p with matrix Q positive semidefinite. (The quadratic HamiltonPoisson
system associated to any cost-extended system is positive semidefinite, see Theo-
rem 10.)
We omit any mention of the Abelian LiePoisson space .3g1 / as any system on
.3g1 / has trivial dynamics. We find it convenient here to use a basis for g2:1 g1
different from the one listed in Table 7.1 in the Appendix. More precisely, we use
the basis E10 D 12 .E1 E2 /, E20 D 12 E3 , E30 D 12 .E1 C E2 /; the only non-zero
commutator is then E10 ; E20 D E10 .
The classification is presented in two stages. First, a classification of the
HamiltonPoisson systems on each LiePoisson space g is given.
Theorem 15 ([32], see also [6, 8, 20, 27, 35]) Let .g ; HQ / be a homogeneous
positive semidefinite quadratic HamiltonPoisson system.
1. If g .g2:1 g1 / , then .g ; HQ / is equivalent to exactly one of the following
systems on .g2:1 g1 /
H1 . p/ D p21 ; H2 . p/ D p22 ; H3 . p/ D p21 C p22 ;

H4 . p/ D . p1 C p3 /2 ; H5 . p/ D p22 C . p1 C p3 /2 :
2. If g .g3:1 / , then .g ; HQ / is equivalent to exactly one of the following
systems on .g3:1 /
H1 . p/ D p23 ; H2 . p/ D p22 C p23 :

systems on .g03:4 /
H1 . p/ D p21 ; H2 . p/ D p23 ; H3 . p/ D p21 C p23 ;

H4 . p/ D . p1 C p2 /2 ; H5 . p/ D . p1 C p2 /2 C p23 :
systems on .g03:5 /
H1 . p/ D p22 ; H2 . p/ D p23 ; H3 . p/ D p22 C p23 :
systems on .g3:6 /
H1 . p/ D p21 ; H2 . p/ D p23 ; H3 . p/ D p21 C p23 ;

H4 . p/ D . p2 C p3 /2 ; H5 . p/ D p22 C . p1 C p3 /2 :
systems on .g3:7 /
H1 . p/ D p21 ; H2 . p/ D p21 C 12 p22 :
Proof (Sketch) The classification on each LiePoisson space g may be obtained
as follows. First one calculates the group of linear Poisson automorphisms; these
are exactly the dual maps of the Lie algebra automorphisms. Using Proposition 4,
one then reduces the Hamiltonian H. p/ D p> Qp as much as possible, yielding a
collection fHi W i 2 Ig of class representatives. (In many cases, this produces
the given normal forms.) Thereafter additional normalization is performed by
#
determining whether or not there exists a linear isomorphism such that H i D
#
H j for i; j 2 I. This amounts to solving a system of equations in nine variables; a
computational software program such as Mathematica can be used to facilitate these
computations. A classification is obtained once the indexing set I has been refined
to the extent that there is no solution for any i; j 2 I, i j. t
u
Systems on three-dimensional LiePoisson spaces can be partitioned into three
classes based on the geometry of the integral curves. A system .g ; H/ is said to
#
be ruled, if for each integral curve of H there exists a line containing its trace.
Likewise, .g ; H/ is called planar if it is not ruled and for each integral curve of
#
H there exists a plane containing its trace. Otherwise, .g ; H/ is called non-planar.
The ruled, planar, and non-planar properties are each invariant under equivalence,
i.e., if two systems are equivalent, then they must belong to the same class.
By determining whether or not any of the normal forms (on distinct LiePoisson
space) are equivalent to one another, we then get the following classification of
system in the context of all three-dimensional LiePoisson spaces admitting global
Casimirs.
Theorem 16 ([32]) Let .g ; HQ / be a homogeneous positive semidefinite

quadratic HamiltonPoisson system.
1. If .g ; H/ is linear, then it is equivalent to exactly one of the systems
..g2:1 g1 / ; p22 /; ..g03:4 / ; . p1 C p2 /2 /; ..g03:5 / ; p22 /:
2. If .g ; H/ is planar, then it is equivalent to exactly one of the systems
..g2:1 g1 / ; p21 C p22 /; ..g2:1 g1 / ; p22 C . p1 C p3 /2 /;

..g03:4 / ; p23 /; ..g03:5 / ; p23 /; ..g3:6 / ; . p2 C p3 /2 /:
3. If .g ; H/ is non-planar, then it is equivalent to exactly one of the systems
..g03:4 / ; . p1 C p2 /2 C p23 /; ..g03:5 / ; p22 C p23 /:
Remark 10 Any system on .g3:1 / or .g3:7 / is equivalent to one on .g03:5 / .
Any system on .g2:1 g1 / or .g3:1 / is a planar (or linear) one. Every system
on .g3:1 / , .g03:4 / , .g03:5 / , or .g3:7 / may be realized on more than one Lie
Poisson space (for .g3:6 / , the only exceptions are those systems equivalent to
..g3:6 / ; . p2 C p3 /2 /).
7.4.1.2 Inhomogeneous Systems
In the context of inhomogeneous systems we find it desirable to weaken the

equivalence relation under consideration. We say that two quadratic Hamilton
Poisson systems .g ; G/ and .h ; H/ are affinely equivalent if there exists an
#
affine isomorphism W g ! h such that the Hamiltonian vector fields G
# # #
and H are -related, i.e., Tp G. p/ D H. . p// for p 2 g . Clearly, if two
systems are linearly equivalent, then they are affinely equivalent. It turns out that for
homogeneous systems the converse holds as well.
We exhibit a classification of the inhomogeneous systems on the orthogonal Lie
Poisson space .g3:7 / D so .3/ . The classification procedure is very similar to that
used for homogeneous systems.
Theorem 17 ([8]) Any inhomogeneous quadratic system ..g3:7 / ; H/ is affinely
equivalent to exactly one of the systems:
0
H1; . p/ D p1 ; >0
H01 . p/ D 1 2
2 p1
H11 . p/ D p2 C 12 p21
1
H2; . p/ D p1 C p2 C 12 p21 ; >0
2
H1; . p/ D p1 C p21 C 12 p22 ; >0
2
H2; . p/ D p2 C p21 C 12 p22 ; >0
2
H3; . p/ D 1 p1 C 2 p2 C p21 C 12 p22 ; 1 ; 2 > 0
2
H4; . p/ D 1 p1 C 3 p3 C p21 C 12 p22 ; 1 > 3 > 0
2
H5; . p/ D 1 p1 C 2 p2 C 3 p3 C p21 C 12 p22 ;
2 > 0; 1 > j3 j > 0 or 2 > 0; 1 D 3 > 0:
Here ; 1 ; 2 ; 3 parametrize families of class representatives, each different value

corresponding to a distinct (non-equivalent) representative.
7.4.2 On Integration and Stability
Among the three-dimensional LiePoisson spaces admitting global Casimirs, most

homogeneous quadratic HamiltonPoisson systems are integrable by elementary
functions. More precisely, among the normal forms presented in Theorem 16, all
systems are integrable by elementary functions except for ..g03:5 / ; p22 C p23 / which
is integrable by Jacobi elliptic functions. A number of inhomogeneous systems are
also integrable by Jacobi elliptic functions (cf. [7]).
It is generally not difficult to determine the (Lyapunov) stability nature of
the equilibria of the quadratic systems on three-dimensional LiePoisson spaces.
Instability usually follows from spectral instability; stability can usually be proved
by using the (extended) energy Casimir method [85].
As an illustrative and typical
example,
we examine the family of inhomogeneous
systems H . p/ D p1 C 12 p22 C p23 on se .2/ . The equations of motion are
8

pP 1 D p2 p3
<
pP 2 D p1 p3

:
pP 3 D .p1 1/ p2 :
The equilibrium states are

e1 D .; 0; 0/; e2 D . 1 ; ; 0/ and e3 D .0; 0; /
where ; 2 R; 0.
Theorem 18 (cf. [7]) The equilibrium states have the following behaviour:

1. The states e1 , 0 < < 1 are spectrally unstable.

2. The state e1 , D 1 is unstable.

3. The states e1 , 2 .1; 0 [ . 1 ; 1/ are stable.
4. The states e2 are spectrally unstable
5. The states e3 are stable.
Proof
(1) The linearization
2 3
0 0 0
#
D H .e1 / D 40 0 5
0 1 C 0
# p
of H at e1 has eigenvalues 1 D 0, 2;3 D .1 /. Hence for

0 < < 1 the state e1 is spectrally unstable. Item (4) follows similarly.
# 1=
(2) The linearization of H at e1 has eigenvalues all zero; hence, we resort to
a direct approach to prove instability. We have that
p
t2 2t 2
p.t/ D ; p ;
.t2 C / .t2 C / t2 C
# 1=
is an integral
curve of H such that limt!1 p.t/ pD e1 and p.0/ D
1= 1C
1 ; 0; p2 . Let U D fp W kp e1 k < g. (Here kpk D
q
1=
p21 C p22 C p23 .) Hence, for any neighbourhood V U of e1 , there exists
1=
t0 < 0 such that p.t0 / 2 V but p.0/ V. Consequently, the state e1 is
unstable.
(3) Consider the energy function G D 1 HC2 C. Let 1 D 22 and 2 D .
We have
2 3
2 0 0

dG.e1 / D 0 and d 2 G.e1 / D 4 0 2.1 C / 0 5 :
0 0 22

The restriction of quadratic form d2 G.e1 / to

ker dH.e1 / \ ker dC.e1 / D f.0; p2 ; p3 / W p2 ; p3 2 Rg
is positive definite for < 0 and > 1 . Therefore, by the extended energy

Casimir method, it follows that the states e1 , 2 .1; 0/ [ . 1 ; 1/ are
stable. On the other hand, we notice that the intersection C1 .0/ \ H1 .0/
2 2
E3 0 E3 0
2 2
2 2
E2 0 2 E2 0 2
0 E 0 E
2 1
2 1
2 2
Fig. 7.1 Intersection of C1 .c0 / and H1 .h0 / for some typical values
is a singleton, namely the origin e01 . Thus, by the continuous energy Casimir
method, the origin is stable. Item (5) can likewise be proved by use of the
extended energy Casimir method. t
u
#
There are a number of qualitatively different types of integral curves of H ,
each type having a different explicit expression. These qualitative different types
can be characterized in terms of the energy values .c0 ; h0 / for the Casimir C
and Hamiltonian H : we have a change in qualitative behaviour when the surfaces
C1 .c0 / and H1 .h0 / are tangent to one another. We consider here the case when
2c
c0 > 12 and h0 > 1C 1 1
2 . In Fig. 7.1 the cylinder C .c0 /, the paraboloid H .h0 /,
0
and their intersection are graphed some typical values of c0 , h0 , and satisfying
2c
c0 > 12 and h0 > 1C 0
2 ; the equilibria are also graphed (the stable equilibria in
blue and the unstable equilibria in red).
Given a modulus k 2 0; 1, the basic Jacobi elliptic functions sn.; k/, cn.; k/
and dn.; k/ can be defined as (see, e.g., [11, 75, 101])
sn.x; k/ D sin am.x; k/

cn.x; k/ D cos am.x; k/
q
dn.x; k/ D 1 k2 sin2 am.x; k/
R'
where am.; k/ D F.; k/1 is the amplitude and F.'; k/ D 0 1k2dtsin2 t (For
the degenerate case k D 0 and k D 1, we recover the circular functions and the
hyperbolic functions, respectively.) The number K is defined as K D F. 2 ; k/. The
functions sn.; k/ and cn.; k/ are 4K periodic, whereas dn.; k/ is 2K periodic.
#
In terms of these functions, the integral curves of H can be expressed as follows.
#
Theorem 19 ([7]) Let p./ W ."; "/ ! se .2/ be an integral curve of H and
2
let h0 D H. p.0//, c0 D C. p.0//. If c0 > 12 and h0 > 1C c0
2 , then there exists
t0 2 R and 2 f1; 1g such that p.t/ D pN .t C t0 / for t 2 ."; "/, where
8 p p
p h0 h0 C cn . t; k/

pN 1 .t/ D c0 p p

h0 C h0 cn . t; k/

< p sn . t; k/
pN 2 .t/ D 2c0 p p

h C h0 cn . t; k/

0

dn . t; k/
: pN 3 .t/ D 2 p p
h0 C h0 cn . t; k/
q p p
Here D h20 c0 , D 2 and k D p1 .h0 / .h0 C 1/.
2
2 2 2
Proof (Sketch) We square the equation pP 1 D p2 p3 both sides to obtain
2pP 1 D2 p2 p3 .
2 2 1
By use of the constants of motion c0 D p1 C p2 and h0 D p1 C 2 p2 C p3 , we
obtain the separable differential equation
dp1 p
D .c0 p1 2 / .h0 2p1 .c0 p1 2 / /:
dt
The integral
Z
dp1
p
.c0 p1 / .h0 2p1 .c0 p1 2 / /
2
is transformed into a standard form and an elliptic integral formula is applied (see,
e.g., [11, 75, 101]) to obtain
p p
p h0 h0 C cn . t; k/
pN 1 .t/ D c0 p p :
h0 C h0 cn . t; k/
The above step is quite involved computationally. Moreover, the transformation into
2c
standard form is dependent on the inequalities c0 > 12 and h0 > 1C 2
0
; in general,
more inequalities may arise in the process of integration (than just those obtained by
distinguishing qualitatively different cases) and other elliptic integral formulas may
be applicable. We note that in this computation some translations in the independent
variable t have been discarded.
Once a prospective expression has been found for pN 1 .t/, the coordinates pN 2 .t/
2 2
pN23 .t/ are
and recovered by means of the identities c0 D p1 C p2 and h0 D p1 C
1
2
p2 C p3 , up to some possible changes in sign. By explicitly calculating pPN .t/
2
# #
H .Np.t//, we identify for which choices of sign pN .t/ is an integral curve of H .
(This step then also ensures that no mistakes were made in computing pN .t/.)
Finally, it remains to be shown that p./ is identical to pN ./ up to a translation

p p
in the independent variable. We have that pN 1 .0/ D c0 and pN 1 . 2K / D c0 .
p p
As p21 C p22 D c0 , we have that c0 6 p1 .0/ 6 c0 . Thus there exists t1 2
0; 2K

such that pN 1 .t1 / D p1 .0/. By choosing 2 f1; 1g appropriately and
using the constant of motions, we get that p.0/ D pN .t0 / for t0 D t1 or t0 D t1 .
Consequently the integral curves t 7! p.t/ and t 7! pN .t C t0 / solve the same
Cauchy problem and therefore are identical. t
u
Note 5 Inhomogeneous systems on se .2/ were studied in [7] (see also [4])
whereas homogeneous systems were treated in [6]. An extensive investigation
(covering equivalence, stability, and some integration) of systems on so .3/ and
se .1; 1/ is carried out in [1, 6] and [16, 20], respectively.
7.5 Conclusion
A central theme of this survey is the equivalence and classification of invariant

control systems, cost-extended systems, and (the associated) HamiltonPoisson
systems. In Sect. 7.2.3, we presented a complete classification of the invariant
control affine systems in three dimensions. There is no complete classification of the
cost-extended systems in three dimensions. However, there are classifications of the
invariant sub-Riemannian structures up to isometry [9] and invariant Riemannian
structures up to L-isometry [55]. It turns out that two sub-Riemannian (but
not Riemannian) structures on the same three-dimensional simply connected Lie
group are isometric up to rescaling if and only if the associated cost-extended
systems are equivalent (cf. [9, 99]). The extent to which this holds true in higher
dimensions is an interesting open problem. Classifications of cost-extended systems
in four dimensions (and beyond) are also topics for future research. Another
avenue of investigation would be to interpret the inhomogeneous cost-extended
systems as invariant affine distributions together with quadratic forms and consider
diffeomorphisms relating such structures (cf. [45, 46]).
In order to find the extremal trajectories for a cost-extended system, one needs
to integrate the associated HamiltonPoisson system (see Theorem 10). In the
last decade or so several authors have considered quadratic HamiltonPoisson
systems on low-dimensional LiePoisson spaces (see, e.g., [4, 7, 13, 23, 49, 97]).
In Sect. 7.4.1, we presented a classification of a subclass of such systems in three
dimensions. As several systems may be equivalently realized on more than one (non-
isomorphic) LiePoisson spaces, it is desirable to have a complete classification.
A systematic treatment (involving classification, stability, and integration) of the
homogeneous positive semidefinite quadratic systems is currently in preparation
(for some partial results, see [6, 8, 20, 32]). So far, inhomogeneous systems have
only been studied, in this manner, on some specific spaces [1, 4, 7, 16].
Acknowledgements The research leading to these results has received funding from the European
Unions Seventh Framework Programme (FP7/20072013) under grant agreement no. 317721.
Also, Rory Biggs would like to acknowledge the financial assistance of the Claude Leon
Foundation.
Appendix: Three-Dimensional Lie Algebras and Groups
There are eleven types of three-dimensional real Lie algebras; in fact, nine algebras
and two parametrized infinite families of algebras (see, e.g., [71, 78, 82]). In terms
of an (appropriate) ordered basis .E1 ; E2 ; E3 /, the commutation operation is given
by
E2 ; E3 D n1 E1 aE2
E3 ; E1 D aE1 C n2 E2
E1 ; E2 D n3 E3 :
The structure parameters a; n1 ; n2 ; n3 for each type are given in Table 7.1.
A classification of the three-dimensional (real connected) Lie groups can be
found in [84]. Let G be a three-dimensional (real connected) Lie group with Lie
algebra g.
1. If g is Abelian, i.e., g 3g1 , then G is isomorphic to R3 , R2 T, R T, or
T3 .
2. If g g2:1 g1 , then G is isomorphic to Aff .R/0 R or Aff .R/0 T.
3. If g g3:1 , then G is isomorphic to the Heisenberg group H3 or the Lie group
H3 D H3 = Z.H3 .Z//, where Z.H3 .Z// is the group of integer points in the centre
Z.H3 / R of H3 .
4. If g g3:2 , g3:3 , g03:4 , ga3:4 , or ga3:5 , then G is isomorphic to the simply
connected Lie group G3:2 , G3:3 , G03:4 D SE .1; 1/, Ga3:4 , or Ga3:5 , respectively.
(The centres of these groups are trivial.)
5. If g g03:5 , then G is isomorphic to the Euclidean group SE .2/, the n-fold
covering SEn .2/ of SE1 .2/ D SE .2/, or the universal covering group SE f .2/.
6. If g g3:6 , then G is isomorphic to the pseudo-orthogonal group SO .2; 1/0 ,
the n-fold covering An of SO .2; 1/0 , or the universal covering group e A. Here
A2 SL .2; R/.
7. If g g3:7 , then G is isomorphic to either the unitary group SU .2/ or the
orthogonal group SO .3/.
Among these Lie groups, only H3 , An , n > 3, and e
A are not matrix Lie groups.
172
Table 7.1 Three-dimensional Lie algebras

a n1 n2 n3 Unimodular Nilpotent Compl. solv. Exponential Solvable Simple Connected groups
3g1 0 0 0 0 R3 ; R2 T; R T2 ; T3
g2:1 g1 1 1 1 0 Aff .R/0 R; Aff .R/0 T
g3:1 0 1 0 0 H3 , H
3
g3:2 1 1 0 0 G3:2
g3:3 1 0 0 0 G3:3
g03:4 0 1 1 0 SE .1; 1/
a>0
ga3:4 a1 1 1 0 Ga3:4
g03:5 0 1 1 0 SE
f .2/, SEn .2/, SE .2/
ga3:5 a>0 1 1 0 Ga3:5
g3:6 0 1 1 1 e
A, An , SL .2; R/, SO .2; 1/0
g3:7 0 1 1 1 SU .2/, SO .3/
R. Biggs and C.C. Remsing
Matrix Representations for Solvable Groups
We have the following parametrizations of the solvable three-dimensional matrix

Lie groups and their Lie algebras. (We omit the Abelian groups.)
2 3 2 3
1 0 0 0 0 0
Aff .R/0 R W 4x y e2z 0 5 g2:1 g1 W 4x y 2z 0 5
0 0 exCy 0 0 xCy
2 3 2 3
1 0 0 0 0 0
Aff .R/0 T W 4x y e2z 0 5 g2:1 g1 W 4x y 2z 0 5
0 0 ei.xCy/ 0 0 i.x C y/
2 3 2 3
1yx 0yx
H3 W 4 0 1 z 5 h3 W 40 0 z 5
001 000
2 3 2 3
1 0 0 0 0 0
G3:2 W 4y ez 0 5 g3:2 W 4y z 05
x zez ez x z z
2 3 2 3
1 0 0 000
G3:3 W 4y ez 0 5 g3:3 W 4y z 05
x 0 ez x0z
2 3 2 3
1 0 0 0 0 0
SE .1; 1/ W 4x cosh z sinh z5 se .1; 1/ W 4x 0 z5
y sinh z cosh z y z 0
2 3 2 3
1 0 0 0 0 0
Ga3:4 W 4x eaz cosh z eaz sinh z5 ga3:4 W 4x az z5
a>0
a1 y eaz sinh z eaz cosh z a>0
a1 y z az
2 3 2 3
1 0 0 0 00 0 0
6 7 6x 0 z 07
f .2/ W 6x cos z sin z 0 7
SE see .2/ W 6 7
4y sin z cos z 0 5 4 y z 0 05
0 0 0 ez 00 0 z
2 3 2 3
1 0 0 0 00 0 0
6x cos z sin z 0 7 6x 0
6 7 z 0 7
SEn .2/ W 6y sin z cos z 0 7 sen .2/ W 6
4y z
7
4 5 0 05
iz
0 0 0 en 00 0 izn
2 3 2 3
1 0 0 00 0
SE.2/ W 4x cos z sin z5 se.2/ W 4x 0 z5
y sin z cos z yz 0
2 3 2 3
1 0 0 0 0 0
Ga3:5 W 4x eaz cos z eaz sin z5 ga3:5 W 4x az z5
y eaz sin z eaz cos z y z az
An appropriate ordered basis for the Lie algebra in each case is given by setting
.x; y; z/ D .1; 0; 0/ for E1 , .x; y; z/ D .0; 1; 0/ for E2 , and .x; y; z/ D .0; 0; 1/
for E3 .
Matrix Representations for Semisimple Groups

Matrix Lie Groups with Algebra g3:6
There are only two connected matrix Lie groups with Lie algebra g3:6 , namely
the pseudo-orthogonal group SO .2; 1/0 and the special linear group SL .2; R/;
SL .2; R/ is a double cover of SO .2; 1/0 .
The pseudo-orthogonal group
SO .2; 1/ D fg 2 R33 W g> Jg D J; det g D 1g
has two connected components. Here J D diag.1; 1; 1/. The identity component

of SO .2; 1/ is SO .2; 1/0 D fg 2 SO .2; 1/ W g33 > 0g where g D gij (for
g 2 SO .2; 1/). Its Lie algebra is given by
so .2; 1/ D fA 2 R33 W A> J C JA D 0g

82 3 9
< 0 zy =
D 4z 0 x5 W x; y; z 2 R :
: ;
y x0
On the other hand, the special linear group is given by
SL .2; R/ D fg 2 R22 W det g D 1g:
Its Lie algebra is given by

x yz

sl .2; R/ D 2 2 W x; y; z 2 R :
yCz
2
2x
Matrix Lie Groups with Algebra g3:7
There are exactly two connected Lie groups with Lie algebra g3:7 ; both are matrix
Lie groups. The special unitary group and its Lie algebra are given by

SU .2/ D g 2 C22 W gg D 1; det g D 1

i 1

su .2/ D 2x 2 .iz C y/ W x; y; z 2 R
1
2 .iz y/ 2i x
SU .2/ is a double cover of the orthogonal group SO .3/. The orthogonal group
SO .3/ and its Lie algebra are given by

SO .3/ D g 2 R33 W gg> D 1; det g D 1
82 3 9
< 0 z y =
so .3/ D 4 z 0 x5 W x; y; z 2 R :
: ;
y x 0
Note 6 Again, an appropriate ordered basis for the Lie algebra in each case is given
by setting .x; y; z/ D .1; 0; 0/ for E1 , .x; y; z/ D .0; 1; 0/ for E2 , and .x; y; z/ D
.0; 0; 1/ for E3 .
Automorphism Groups
A standard computation yields the automorphism group for each three-dimensional

Lie algebra (see, e.g., [57]). With respect to the given ordered basis .E1 ; E2 ; E3 / ,
the automorphism group of each solvable Lie algebra has parametrization:
2 3 2 3
yw vz x u xyu
Aut.g3:1 / W 4 0 y v5 Aut.g2:1 g1 / W 4y x v 5
0 zw 001
2 3 2 3
uxy xy z
Aut.g3:2 / W 40 u z 5 Aut.g3:3 / W 4u v w5
001 00 1
2 3 2 3 2 3
xyu x y u xyu
0 4 5 4
Aut.g3:4 / W y x v ; y x v5 Aut.ga3:4 / W 4y x v 5
001 0 0 1 001
2 3 2 3 2 3
x yu x y u x yu
Aut.g03:5 // W 4y x v 5 ; 4y x v5 Aut.ga3:5 / W 4y x v 5
0 01 0 0 1 0 01
For the semisimple Lie algebras, we have
Aut.g3:6 / D SO .2; 1/ and Aut.g3:7 / D SO .3/:
In several cases, for a connected Lie group G, we wish to find the subgroup of
linearized group automorphisms d Aut.G/ D fT1
W
2 Aut.G/g 6 Aut.g/. If G
is simply connected, then d Aut.G/ D Aut.g/. If G is not simply connected, then
the subgroup d Aut.G/ may be determined by use of the following result.
Proposition 5 (cf. [58]) Let G be a connected Lie group with Lie algebra g and
e ! G be a universal covering. If
let q W G 2 Aut.g/, then 2 d Aut.G/ if and
e is the unique automorphism such that
only if
.ker q/ D ker q, where
2 Aut.G/
T1
D .T1 q/1 T1 q.
Casimir Functions for LiePoisson Spaces
An exhaustive list of Casimir functions (not necessarily globally defined), for

low-dimensional Lie algebras, was obtained by Patera et al. [86]. For each three-
dimensional LiePoisson space g (associated to the three-dimensional Lie algebra
g) we exhibit its Casimir function:
g2:1 g1 W C. p/ D p3 g3:1 W C. p/ D p1
p2 p2
g3:2 W C. p/ D p1 e p1 g3:3 W C. p/ D
p1
1
p C 12 p2
2 1
g03:4 W C. p/ D p21 p22 g3:4 W C. p/ D 1
1>0 . 12 p1 12 p2 / C1
i
p1 ip2
g03:5 W C. p/ D p21 C p22 g3:5 W C. p/ D . p21 C p22 /
>0 p1 C ip2
g3:6 W C. p/ D p21 C p22 p23 g3:7 W C. p/ D p21 C p22 C p23 :
On the trivial LiePoisson space .3g1 / , every function is a Casimir function. Here
p 2 g is written in coordinates as p D p1 E1 C p2 E2 C p3 E3 where .E1 ; E2 ; E3 /
is the dual of the ordered basis .E1 ; E2 ; E3 /. Note that only 3g1 , g2:1 g1 , g3:1 ,
g03:4 , g03:5 , g3:6 , and g3:7 admit globally defined Casimir functions.
References
1. R.M. Adams, R. Biggs, W. Holderbaum, C.C. Remsing, Stability and integration of Hamilton-
Poisson systems on so.3/ . Rend. Mat. Appl. 37(12), 142 (2016)
2. R.M. Adams, R. Biggs, C.C. Remsing, Equivalence of control systems on the Euclidean
group SE.2/. Control Cybern. 41(3), 513524 (2012)
3. R.M. Adams, R. Biggs, C.C. Remsing, On the equivalence of control systems on the
orthogonal group SO.4/, in Recent Researches in Automatic Control, Systems Science and
Communications, Porto, 2012, ed. by H.R. Karimi (WSEAS Press, 2012), pp. 5459. ISBN:
978-1-61804-103-6
4. R.M. Adams, R. Biggs, C.C. Remsing, Single-input control systems on the Euclidean group
SE .2/. Eur. J. Pure Appl. Math. 5(1), 115 (2012)
5. R.M. Adams, R. Biggs, C.C. Remsing, Control systems on the orthogonal group SO .4/.
Commun. Math. 21(2), 107128 (2013)
6. R.M. Adams, R. Biggs, C.C. Remsing, On some quadratic Hamilton-Poisson systems. Appl.
Sci. 15, 112 (2013)
7. R.M. Adams, R. Biggs, C.C. Remsing, Two-input control systems on the Euclidean group
SE .2/. ESAIM Control Optim. Calc. Var. 19(4), 947975 (2013)
8. R.M. Adams, R. Biggs, C.C. Remsing, Quadratic Hamilton-Poisson systems on so .3/:
classifications and integration, in Proceedings of the 15th International Conference on
Geometry, Integrability and Quantization, Varna, 2013, ed. by I.M. Mladenov, A. Ludu, A.
Yoshioka (Bulgarian Academy of Sciences, Sofia, 2014), pp. 5566
9. A. Agrachev, D. Barilari, Sub-Riemannian structures on 3D Lie groups. J. Dyn. Control Syst.
18(1), 2144 (2012)
10. A.A. Agrachev, Y.L. Sachkov, Control Theory from the Geometric Viewpoint (Springer,
Berlin, 2004)
11. J.V. Armitage, W.F. Eberlein, Elliptic Functions (Cambridge University Press, Cambridge,
2006)
12. A. Aron, C. Pop, Quadratic and homogeneous Hamilton-Poisson systems on the 13th Lie
algebra from Bianchis classification, in Proceedings of the International Conference of
Differential Geometry and Dynamical Systems, Bucharest, 2009, vol. 17 (Geometry Balkan
Press, Bucharest 2010), pp. 1220
13. A. Aron, C. Daniasa, M. Puta, Quadratic and homogeneous Hamilton-Poisson system on
.so.3//. Int. J. Geom. Methods Mod. Phys. 4(7), 11731186 (2007)
14. A. Aron, C. Pop, M. Puta, Some remarks on .sl.2; R// and Kahans integrator. An. Stiint.
Univ. Al. I. Cuza Iasi. Mat. (N.S.) 53(Suppl. 1), 4960 (2007)
15. A. Aron, M. Craioveanu, C. Pop, M. Puta, Quadratic and homogeneous Hamilton-Poisson
systems on A 3;6;1 . Balkan J. Geom. Appl. 15(1), 17 (2010)
16. D.I. Barrett, R. Biggs, C.C. Remsing, Quadratic Hamilton-Poisson systems on se .1; 1/ : the
inhomogeneous case. Preprint
17. D.I. Barrett, R. Biggs, C.C. Remsing, Affine subspaces of the Lie algebra se .1; 1/. Eur. J.
Pure Appl. Math. 7(2), 140155 (2014)
18. D.I. Barrett, R. Biggs, C.C. Remsing, Optimal control of drift-free invariant control systems
on the group of motions of the Minkowski plane, in Proceedings of the 13th European
Control Conference, Strasbourg, 2014 (European Control Association, 2014), pp. 24662471.
doi:10.1109/ECC.2014.6862313
19. D.I. Barrett, R. Biggs, C.C. Remsing, Affine distributions on a four-dimensional extension of
the semi-Euclidean group. Note Mat. 35(2), 8197 (2015)
20. D.I. Barrett, R. Biggs, C.C. Remsing, Quadratic Hamilton-Poisson systems on se .1; 1/ : the
homogeneous case. Int. J. Geom. Methods Mod. Phys. 12, 1550011 (17 pp.) (2015)
21. C.E. Bartlett, R. Biggs, C.C. Remsing, Control systems on the Heisenberg group: equivalence
and classification. Publ. Math. Debr. 88(12), 217234 (2016)
22. J. Biggs, W. Holderbaum, Planning rigid body motions using elastic curves. Math. Control
Signals Syst. 20(4), 351367 (2008)
23. J. Biggs, W. Holderbaum, Integrable quadratic Hamiltonians on the Euclidean group of
motions. J. Dyn. Control Syst. 16(3), 301317 (2010)
24. R. Biggs, P.T. Nagy, A classification of sub-Riemannian structures on the Heisenberg groups.
Acta Polytech. Hungar. 10(7), 4152 (2013)
25. R. Biggs, C.C. Remsing, A category of control systems. An. Stiint. Univ. Ovidius Constanta
Ser. Mat. 20(1), 355367 (2012)
26. R. Biggs, C.C. Remsing, A note on the affine subspaces of three-dimensional Lie algebras.
Bul. Acad. Stiinte Repub. Mold. Mat. 2012(3), 4552 (2012)
27. R. Biggs, C.C. Remsing, On the equivalence of cost-extended control systems on Lie groups,
in Recent Researches in Automatic Control, Systems Science and Communications, Porto,
2012, ed. by H.R. Karimi (WSEAS Press, 2012), pp. 6065. ISBN: 978-1-61804-103-6
28. R. Biggs, C.C. Remsing, Control affine systems on solvable three-dimensional Lie groups, I.
Arch. Math. (Brno) 49(3), 187197 (2013)
29. R. Biggs, C.C. Remsing, Control affine systems on solvable three-dimensional Lie groups,
II. Note Mat. 33(2), 1931 (2013)
30. R. Biggs, C.C. Remsing, Control affine systems on semisimple three-dimensional Lie groups.
An. Stiint. Univ. Al. I. Cuza Iasi. Mat. (N.S.) 59(2), 399414 (2013)
31. R. Biggs, C.C. Remsing, Feedback classification of invariant control systems on three-
dimensional Lie groups, in Proceedings of the 9th IFAC Symposium on Nonlinear Control
Systems, Toulouse (2013), pp. 506511
32. R. Biggs, C.C. Remsing, A classification of quadratic Hamilton-Poisson systems in three
dimensions, in Proceedings of the 15th International Conference on Geometry, Integrability
and Quantization, Varna, 2013, ed. by I.M. Mladenov, A. Ludu, A. Yoshioka (Bulgarian
Academy of Sciences, Sofia, 2014), pp. 6778
33. R. Biggs, C.C. Remsing, Control systems on three-dimensional Lie groups: equivalence and
controllability. J. Dyn. Control Syst. 20(3), 307339 (2014)
34. R. Biggs, C.C. Remsing, Control systems on three-dimensional Lie groups, in Proceedings
of the 13th European Control Conference, Strasbourg, 2014 (European Control Association,
2014), pp. 24422447. doi:10.1109/ECC.2014.6862312
35. R. Biggs, C.C. Remsing, Cost-extended control systems on Lie groups. Mediterr. J. Math.
11(1), 193215 (2014)
36. R. Biggs, C.C. Remsing, Some remarks on the oscillator group. Differ. Geom. Appl.
35(Suppl.), 199209 (2014)
37. R. Biggs, C.C. Remsing, Subspaces of the real four-dimensional Lie algebras: a classification
of subalgebras, ideals, and full-rank subspaces. Extracta Math. 31(1), 4193 (2015)
38. R. Biggs, C.C. Remsing, On the equivalence of control systems on Lie groups. Commun.
Math. 23(2), 119129 (2015)
39. R. Biggs, C.C. Remsing, Equivalence of control systems on the pseudo-orthogonal group
SO .2; 1/. An. Stiint. Univ. Ovidius Constanta Ser. Mat. 24(2), 4565 (2016)
40. A.M. Bloch, Nonholonomic Mechanics and Control (Springer, New York, 2003)
41. A.M. Bloch, P.E. Crouch, N. Nordkvist, A.K. Sanyal, Embedded geodesic problems and
optimal control for matrix Lie groups. J. Geom. Mech. 3(2), 197223 (2011)
42. B. Bonnard, V. Jurdjevic, I. Kupka, G. Sallet, Transitivity of families of invariant vector fields
on the semidirect products of Lie groups. Trans. Am. Math. Soc. 271(2), 525535 (1982)
43. R.W. Brockett, System theory on group manifolds and coset spaces. SIAM J. Control 10(2),
265284 (1972)
44. L. Capogna, E. Le Donne, Smoothness of subRiemannian isometries. Am. J. Math. 138(5),
14391454 (2016)
45. J.N. Clelland, C.G. Moseley, G.R. Wilkens, Geometry of control-affine systems. SIGMA
Symmetry Integrability Geom. Methods Appl. 5(Paper 095), 28 (2009)
46. J.N. Clelland, C.G. Moseley, G.R. Wilkens, Geometry of optimal control for control-affine
systems. SIGMA Symmetry Integrability Geom. Methods Appl. 9(Paper 034), 31 (2013)
47. M. Craioveanu, C. Pop, A. Aron, C. Petrisor, An optimal control problem on the

special Euclidean group SE.3; R/, in International Conference of Differential Geometry
and Dynamical Systems, Bucharest, 2009, vol. 17 (Geometry Balkan Press, Sofia, 2010),
pp. 6878
48. D. DAlessandro, M. Dahleh, Optimal control of two-level quantum systems. IEEE Trans.
Autom. Control 46(6), 866876 (2001)
49. C. Daniasa, A. Grban, R.M. Tudoran, New aspects on the geometry and dynamics of
quadratic Hamiltonian systems on .so.3// . Int. J. Geom. Methods Mod. Phys. 8(8), 1695
1721 (2011)
50. V.I. Elkin, Affine control systems: their equivalence, classification, quotient systems, and
subsystems. J. Math. Sci. 88(5), 675721 (1998)
51. R.V. Gamkrelidze, Discovery of the maximum principle. J. Dyn. Control Syst. 5(4), 437451
(1999)
52. R.B. Gardner, W.F. Shadwick, Feedback equivalence of control systems. Syst. Control Lett.
8(5), 463465 (1987)
53. V.V. Gorbatsevich, A.L. Onishchik, E.B. Vinberg, Foundations of Lie Theory and Lie
Transformation Groups (Springer, Berlin, 1997)
54. S. Goyal, N.C. Perkins, C.L. Lee, Nonlinear dynamics and loop formation in Kirchhoff rods
with implications to the mechanics of DNA and cables. J. Comput. Phys. 209(1), 371389
(2005)
55. K.Y. Ha, J.B. Lee, Left invariant metrics and curvatures on simply connected three-
dimensional Lie groups. Math. Nachr. 282(6), 868898 (2009)
56. U. Hamenstdt, Some regularity theorems for Carnot-Carathodory metrics. IEEE Trans.
Autom. Control 32(3), 819850 (1990)
57. A. Harvey, Automorphisms of the Bianchi model Lie groups. J. Math. Phys. 20(2), 251253
(1979)
58. J. Hilgert, K.-H. Neeb, Structure and Geometry of Lie Groups (Springer, New York, 2012)
59. D.D. Holm, Geometric Mechanics. Part I: Dynamics and Symmetry (Imperial College Press,
London, 2008)
60. D.D. Holm, Geometric Mechanics. Part II: Rotating, Translating and Rolling, 2nd edn.
(Imperial College Press, London, 2011)
61. B. Jakubczyk, Equivalence and invariants of nonlinear control systems, in Nonlinear
Controllability and Optimal Control, ed. by H.J. Sussmann (Dekker, New York, 1990), pp.
177218
62. B. Jakubczyk, W. Respondek, On linearization of control systems. Bull. Acad. Polon. Sci.
Sr. Sci. Math. 28(910), 517522 (1980)
63. V. Jurdjevic, The geometry of the plate-ball problem. Arch. Ration. Mech. Anal. 124(4),
305328 (1993)
64. V. Jurdjevic, Geometric Control Theory (Cambridge University Press, Cambridge, 1997)
65. V. Jurdjevic, Integrable Hamiltonian systems on Lie groups: Kowalewski type. Ann. Math.
150(2), 605644 (1999)
66. V. Jurdjevic, Optimal control on Lie groups and integrable Hamiltonian systems. Regul.
Chaotic Dyn. 16(5), 514535 (2011)
67. V. Jurdjevic, F. Monroy-Prez, Variational problems on Lie groups and their homogeneous
spaces: elastic curves, tops, and constrained geodesic problems, in Contemporary Trends
in Nonlinear Geometric Control Theory and its Applications, Mxico City, 2000 (World
Scientific, River Edge, NJ, 2002), pp. 351
68. V. Jurdjevic, H.J. Sussmann, Control systems on Lie groups. J. Differ. Equ. 12, 313329
(1972)
69. V. Jurdjevic, J. Zimmerman, Rolling sphere problems on spaces of constant curvature. Math.
Proc. Camb. Philos. Soc. 144(3), 729747 (2008)
70. I. Kishimoto, Geodesics and isometries of Carnot groups. J. Math. Kyoto Univ. 43(3), 509
522 (2003)
71. A. Krasinski, C.G. Behr, E. Schcking, F.B. Estabrook, H.D. Wahlquist, G.F.R. Ellis, R.
Jantzen, W. Kundt, The Bianchi classification in the Schcking-Behr approach. Gen. Relativ.
Gravit. 35(3), 475489 (2003)
72. A.J. Krener, On the equivalence of control systems and linearization of nonlinear systems.
SIAM J. Control 11, 670676 (1973)
73. C. Laurent-Gengoux, A. Pichereau, P. Vanhaecke, Poisson Structures (Springer, Heidelberg,
2013)
74. J. Lauret, Modified H-type groups and symmetric-like Riemannian spaces. Differ. Geom.
Appl. 10(2), 121143 (1999)
75. D.F. Lawden, Elliptic Functions and Applications (Springer, New York, 1989)
76. E. Le Donne, A. Ottazzi, Isometries of Carnot groups and subFinsler homogeneous
manifolds. J. Geom. Anal. 26(1), 330345 (2016)
77. N.E. Leonard, P.S. Krishnaprasad, Motion control of drift-free, left-invariant systems on Lie
groups. IEEE Trans. Automat. Control 40, 15391554 (1995)
78. M.A.H. MacCallum, On the classification of the real four-dimensional Lie algebras, in On
Einsteins Path, New York, 1996 (Springer, New York, 1999), pp. 299317
79. J.E. Marsden, T.S. Ratiu, Introduction to Mechanics and Symmetry, 2nd edn. (Springer,
New York, 1999)
80. C. Meyer, Matrix Analysis and Applied Linear Algebra (Society for Industrial and Applied
Mathematics, Philadelphia, PA, 2000)
81. F. Monroy-Prez, A. Anzaldo-Meneses, Optimal control on the Heisenberg group. J. Dyn.
Control Syst. 5(4), 473499 (1999)
82. G.M. Mubarakzyanov, On solvable Lie algebras. Izv. Vys. Ucehn. Zaved. Matematika 1(32),
114123 (1963)
83. P.T. Nagy, M. Puta, Drift Lee-free left invariant control systems on SL.2; R/ with fewer
controls than state variables. Bull. Math. Soc. Sci. Math. Roumanie (N.S.) 44(92)(1), 311
(2001)
84. A.L. Onishchik, E.B. Vinberg, Lie Groups and Lie Algebras, III (Springer, Berlin, 1994)
85. J.-P. Ortega, V. Planas-Bielsa, T.S. Ratiu, Asymptotic and Lyapunov stability of constrained
and Poisson equilibria. J. Differ. Equ. 214(1), 92127 (2005)
86. J. Patera, R.T. Sharp, P. Winternitz, H. Zassenhaus, Invariants of real low dimension Lie
algebras. J. Math. Phys. 17(6), 986994 (1976)
87. R.O. Popovych, V.M. Boyko, M.O. Nesterenko, M.W. Lutfullin, Realizations of real low-
dimensional Lie algebras. J. Phys. A 36(26), 73377360 (2003)
88. M. Puta, Stability and control in spacecraft dynamics. J. Lie Theory 7(2), 269278 (1997)
89. M. Puta, Optimal control problems on matrix Lie groups, in New Developments in Differential
Geometry, ed. by J. Szenthe (Kluwer, Dordrecht, 1999), pp. 361373
90. W. Respondek, I.A. Tall, Feedback equivalence of nonlinear control systems: a survey on
formal approach, in Chaos in Automatic Control (Control Engineering) (CRC, Taylor &
Francis, Boca Raton, 2006), pp. 137262
91. L. Saal, The automorphism group of a Lie algebra of Heisenberg type. Rend. Sem. Mat. Univ.
Politec. Torino 54(2), 101113 (1996)
92. Y.L. Sachkov, Conjugate points in the Euler elastic problem. J. Dyn. Control Syst. 14(3),
409439 (2008)
93. Y.L. Sachkov, Maxwell strata in the Euler elastic problem. J. Dyn. Control Syst. 14(2),
169234 (2008)
94. Y.L. Sachkov, Control theory on Lie groups. J. Math. Sci. 156(3), 381439 (2009)
95. H.J. Sussmann, An extension of a theorem of Nagano on transitive Lie algebras. Proc. Am.
Math. Soc. 45(3), 349356 (1974)
96. H.J. Sussmann, Lie brackets, real analyticity and geometric control, in Differential Geometric
Control Theory, ed. by Brockett, R.W., Millman, R.S., Sussmann, H.J. (Birkhuser, Boston,
MA, 1983), pp. 1116
97. R.M. Tudoran, The free rigid body dynamics: generalized versus classic. J. Math. Phys.
54(7), 072704, 10 (2013)
98. R.M. Tudoran, R.A. Tudoran, On a large class of three-dimensional Hamiltonian systems. J.
Math. Phys. 50(1), 012703, 9 (2009)
99. A.M. Vershik, V.Ya. Gershkovich, Nonholonomic dynamical systems, geometry of distribu-
tions and variational problems, in Dynamical Systems VII, ed. by V.I. Arnold, S.P. Novikov
(Springer, Berlin, 1994), pp. 181
100. G.C. Walsh, R. Montgomery, S.S. Sastry, Optimal path planning on matrix Lie groups, in
Proceedings of the 33rd Conference on Decision and Control, Lake Buena Vista (1994), pp.
12581263
101. E.T. Whittaker, G.N. Watson, A Course of Modern Analysis (Cambridge University Press,
Cambridge, 1927)
102. E.N. Wilson, Isometry groups on homogeneous nilmanifolds. Geom. Dedicata 12(3), 337
346 (1982)
Chapter 8
An Optimal Control Problem for a Nonlocal
Problem on the Plane
D. Devadze, Ts. Sarajishvili, and M. Abashidze
Abstract The present paper is dedicated to the investigation of optimal control

problems whose behavior is described by quasilinear differential equations of
first order on the plane with nonlocal BitsadzeSamarski boundary conditions.
A theorem of the existence and uniqueness of a generalized solution in the space
C .G/ is proved for quasilinear differential equations; necessary conditions of
optimality are obtained in terms of the principle of maximum; the Bitsadze
Samarski boundary value problem is proved for a linear differential equation of
p
first order; the existence of a solution in the space C .G/ is proved and an a
priori estimate is derived. A theorem on the necessary and sufficient condition of
optimality is proved for a linear optimal control problem.
Nonlocal boundary value problems are quite an interesting generalization of

classical problems and at the same time they are naturally obtained when construct-
ing mathematical models in physics, engineering, sociology, ecology, and so on
[15].
The investigation of nonlocal problems for differential equations originated in
the last century. Here we should in the first place refer to the works of T. Carleman,
R. Beals, F. Browder, and other works. The problems posed in [68] are the
problems with nonlocal conditions, which are considered only on the boundary of
the definition domain of a differential operator. In 1963, J. Cannon posed a nonlocal
problem in his work [9] and thus gave an impetus to the development of a new trend
in the investigation of nonlocal boundary value problems [1013].
In 1969, the work of Bitsadze and Samarski [14] was published, which was
dedicated to the investigation of a nonlocal problem of a new type. That problem
arose in connection with the mathematical modeling of plasma processes . Intensive
studies of BitsadzeSamarsk nonlocal problems and their various generalizations
began in the 1980s of the last century (see the papers of D.G. Gordeziani,
A.L. Skubachevski, V.P. Paneyakh, V.A. Ilyin, I. Moiseyev, G.V. Meladze,
M.P. Sapagovas, D.V. Kapanadze, V.P. Mikhailov, A.K. Gushchin, G. Avalishvili,
D. Devadze () Ts. Sarajishvili M. Abashidze

Batumi Shota Rustaveli State University, 35 Ninoshvili St., Batumi 6010, Georgia
e-mail: david.devadze@gmail.com; tsis55@yahoo.com; m.abashidze1@gmail.com

184 D. Devadze et al.
L. Gurevich [1526], and other works). The papers of D.G. Sapagovas,

G.K. Berikelashvili are certainly interesting from the standpoint of application
and numerical methods (see, e.g., [11, 17]). The algorithm of reducing nonlocal
problems of the BitsadzeSamarski type to an iteration sequence of Dirichlet
problems is investigated in Gordezianis papers [15, 16].
Problems of the existence and uniqueness of a generalized analytic function
for the RiemannHilbert problem are investigated in Vekuas monograph [27].
Problems of the existence and uniqueness of a generalized solution for quasilinear
equations of first order on the plane with RiemannHilbert boundary conditions
are considered in [28]. In [29, 30] the BitsadzeSamarski nonlocal boundary value
problem is considered for quasilinear differential equations.
For many optimization problems in elasticity theory, mechanics, diffusion
processes, the kinetics of chemical reactions, and so on, the state of a system is
described by partial differential equations [3134]. Therefore control problems for
systems with distributed parameters attract a great deal of attention [33, 35, 36].
The present paper is devoted to optimal control problems whose behavior is
described by quasilinear differential equation of first order on the plane with
nonlocal BitsadzeSamarski boundary conditions. When dealing with optimization
problems for systems with distributed parameters it is important to investigate the
questions of existence of a generalized solution for discontinuous right-hand parts of
an equation [27]. In the proposed paper, the necessary conditions of optimality are
obtained using the approach elaborated in [3739] for controlled systems of general
form.
In the paper, the theorem on the existence and uniqueness theorem of a gener-
alized solution in the space C .G/ is proved for quasilinear differential equations
of first order with nonlocal boundary conditions; the BitsadzeSamarski boundary
value problem is considered for a linear differential equation of first order. The

existence of a solution in the space C .G/ is proved and an a priori estimate
is obtained. An optimal control problem is stated for a nonlocal boundary value
problem with an integral quality criterion. Necessary conditions of optimality are
obtained in terms of the maximum principle when the control domain is an arbitrary
bounded set. An adjoint equation is constructed in the differential form. The optimal
control problem for linear differential equations of first order is considered on the
plane with nonlocal boundary conditions. The theorem on a necessary and sufficient
condition of optimality is proved.
8.1 Existence of a Generalized Solution
Let G be a bounded domain of the complex plane E with boundary which

is a closed simple Lyapunov curve (i.e. the angle formed between the constant
direction and the tangent to this curve is Hlder-continuous). Denote by the part
of the boundary , which is an open Lyapunov curve with the parametric equation
z D z.s/, 0 6 s 6 . Let 0 be the diffeomorphic image z0 D I.z/ of , which lies
8 An Optimal Control Problem for a Nonlocal Problem on the Plane 185
in the domain G, with the parametric equation z0 D z0 .s/, 0 6 s 6 . Assume that

0 intersects with , but not tangentially to it, z D x C iy 2 G, @Z D 12 @x
@ @
C i @y is
the generalized Sobolev derivative [27, 40, 41].
Let us consider in the domain G the BitsadzeSamarski boundary value problem
[14] for quasilinear differential equations of first order
@Z w D f .z; w; w/; z 2 G; (8.1)

Rew.z/ D '1 .z/; z 2 n ; (8.2)

Re w.z.s// D Re w.z0 .s// ; z.s/ 2 ; z0 .s/ 2 0 ; 0 6 s 6 ; (8.3)
Imw.z / D c; z 2 n ; 0 < D const. (8.4)
It is assumed that the following conditions are fulfilled:

(A1) The function f .z; w; w/ is defined for z 2 G, jwj < R, f .z; 0; 0/ 2 Lp .G/,
p > 2 and

f .z; w; w/ f .z; w0 ; w0 / 6 L jw w0 j C jw w0 j :
(A2) '1 .z/ 2 C . n /, '2 .z/ 2 C . /, > 1=2, and the values of the function
'1 .z/, '2 .z/ at the ends of the contours and 0 coincide.
In order to prove the existence of a solution of the problem (8.1)(8.4), we
consider the following iteration process
@Z wn D f .z; wn ; wn /; z 2 G; (8.5)
Rewn .z/ D '1 .z/; z 2 n ; (8.6)

Re wn .z.s// D Re wn1 .z0 .s// ; z.s/ 2 ; z0 .s/ 2 0 ; (8.7)

Imwn .z / D c; z 2 n ; 0 6 s 6 ; n D 1; 2; : : : ; (8.8)
where w0 .z/ is any function from C . /, > 1=2, which continuously adjoins the
values of the function '1 .z/ at the ends of the contour .
For each n 2 N the problem (8.5)(8.8) is the RiemannHilbert problem and its
regular generalized solution belongs to the space C .G/ [27, 28].
Let us consider the function vn D wnC1 wn . Then from (8.5)(8.8) it follows
that the function vn is the solution of the following problem
@Z vn D f .z; wnC1 ; wnC1 / f .z; wn ; wn /

F.z; wn ; wn ; wnC1 ; wnC1 /; z 2 G; (8.9)
Revn .z/ D 0; z 2 n ; (8.10)

Re vn .z.s// D Re vn1 .z0 .s// ; z.s/ 2 ; z0 .s/ 2 0 ; 0 6 s 6 ; (8.11)

Imvn .z / D c; z 2 n ; n D 1; 2; : : : : (8.12)
The solution of the problem (8.9)(8.12) can be reduced to the following

nonlinear integral equation [28]:
v .z/ D n .z/
C
n .z/

1 F.; wn ./; wn ./; wnC1 ./; wnC1 .//
d d; (8.13)
z
G
where D C i, n .z/ is a holomorphic function satisfying the conditions (8.10)

(8.12), and
n .z/ is a holomorphic function such that the difference

1 F.; wn ; wn ; wnC1 ; wnC1 /

n .z/ d d
z
G
satisfies the homogeneous boundary conditions; an a priori estimate has the form
k
n kC .G/ 6 C1 kFkLp .G/ ; C1 D const > 0:
Denote by TG the integral operator in the right-hand side of Eq. (8.13). Note that
the operator TG maps the space Lp .G/ into C .G/, D . p 2/=p < [27].
Consider the following conditions:
(A3) There exists a number R1 > 0, R1 6 R such that the following inequalities are
fulfilled:

k n kC .G/ C C1 C kTG kLp .G/;C .G/ 2LjGj1=p R1 6 R1 :
(A4) 2jGj1=p L.C1 C kTG kLp .G/;C .G/ / < 1.

Assume that the conditions (A1)(A4) are fulfilled, then there exists a unique
solution of the RiemannHilbert problem (8.9)(8.12) in the ball kvn kC .G/ 6 R1
[28].
Let us estimate the function vn .z/ from the equality (8.13) in the metric of the
space C.G/:

kvn kC.G/ 6 k n kC.G/ C k
n kC.G/ C TG F : (8.14)
C.G/
Using the previous estimates, from the inequality (8.14) we obtain

kvn kC.G/ 6 k n kC.G/ C1 C kTG kLp .G/;C.G/ kFkLp .G/ : (8.15)
By virtue of the condition (A1) we have

F.z; wn ; wn ; wnC1 ; wnC1 /

D f .z; wnC1 ; wnC1 / f .z; wn ; wn / 6 2LjwnC1 wn j D 2Ljvn j:
Due to the condition (A1), from the last inequality it follows that the function
F.z; wn ; wn ; wnC1 ; wnC1 / belongs to the space Lp .G/. Then
kFkLp .G 6 2Lkvn kLp .G/ 6 2LjGj1=p kvn kC.G/ :
Thus, by the inequality (8.15) we can write that

kvn kC.G/ 6 k n kC.G/ C 2LjGj1=p C1 C kTG kLp .G/;C.G/ kvn kC.G/ ;
i.e., in view of the condition (A4), we finally obtain
k n kC.G/
kvn kC.G/ 6 : (8.16)
1 2LjGj1=p .C1 C kTG kLp .G/;C .G/ /
Note that the function n .z/ is the solution of the following problem
@Z n .z/ D 0; z 2 G;
Re n .z/ D 0; z 2 n ;
Re n .z/ D Re n1 .z0 /; z 2 ; z0 2 0 ;

Im n .z / D 0; n D 1; 2; : : : ;
0 .z/ D w1 .z/ w0 .z/:
Since Re n .z/ is a harmonic function, all the conditions of Schwartz lemma

[42] are fulfilled for it and there exists 0 < q < 1 not depending on n , such that
the following inequality [16] holds for it:
k n kC.G/ 6 Mqn ;
where M > 0 is a constant depending only on '1 .z/.

Taking this estimate from (8.16) we can write
M
kvn kC.G/ 6 qn : (8.17)
1 2LjGj1=p .C1 C kTG kLp .G/;C.G/ /
P
1
Then from (8.17) we can conclude that the series vk converges uniformly to
kD1
zero in the domain G. Hence it follows that the sequence fwn .z/g is fundamental in
C.G/ and has the limit w.z/ 2 C.G/.
Let us consider the integral representation for the function wn .z/:

0 1 f .; wn ; wn /
wn .z/ D n .z/ C
n0 .z/ d d; (8.18)
z
G
where n0 .z/ is a holomorphic function that satisfies the conditions (8.6)(8.8), and
n0 .z/ is a holomorphic function such that the difference

1 f .; wn ; wn /

n0 .z/ d d
z
G
satisfies the homogeneous boundary conditions.

The representation (8.18) implies that w.z/ is the solution of the problem (8.1)
(8.4) and w.z/ 2 C .G/. From the uniqueness of the holomorphic solution and the
integral representation (8.18) we can conclude that this solution is unique in the
class C .G/.
We have thereby proved.
Theorem 1 Let the conditions (A1)(A4) be fulfilled, then the solution of the
problem (8.1)(8.4) exists in the space C .G/ and is unique.
8.2 Linear Problem
Let us consider in the domain G the BitsadzeSamarski boundary value problem for
a linear differential equation of first order
@Z w D A.z/w C B.z/w C d.z/; z 2 G;

Rew.z/ D 0; z 2 n ;
(8.19)
Rew.z.s// D Re w.z0 .s// ; z.s/ 2 ; z0 .s/ 2 0 ;
Imw.z / D 0; z 2 n ; 0 6 s 6 :
Assume that A.z/; B.z/; d.z/ 2 Lp .G/, p > 2, jAj; jBj 6 N.

p
Denote by C .G/ the set of functions w.z/ 2 C .G/ such that
Rew.z/ D 0; z 2 n ;

Rew.z.s// D Re w.z0 .s// ; z.s/ 2 ; z0 .s/ 2 0 ; (8.20)
Imw.z / D 0; z 2 n ; 0 6 s 6
and having the finite norm
kwkCp .G/ D kwkC .G/ C k@Z wkLp .G/ < 1: (8.21)

p
It is easy to verify that the set C .G/ is a linear normed space over the real
field with norm defined by means of the equality (8.21). If p > q > 2, then
p q
C .G/ C .G/ and kwkCq .G/ 6 kwkCp .G/ , where is a positive constant and

p
w is any element from C .G/.
Lemma 8.2.1 For any function d.z/ 2 Lp .G/, p > 2, the solution w.z/ of the
p
problem (8.19) exists, belongs to the space C .G/ and the following a priori
estimate holds for it
kwkCp .G/ 6 kdkLp .G/ ; (8.22)

where is a positive constant depending only on p, N and jGj D mes G.

Proof The existence and uniqueness of the problem (8.19) immediately follow from
Theorem 1. It remains for us to prove the validity of the a priori estimate (8.22).
Let us reduce the problem (8.19) to an integral equation. For this we apply the
operator

1 f .t/
TG z; f D d d; t D C i
tz
G
and an operator SG z; f from Lp .G/ to the set of analytic functions which satisfy the
conditions:

Re TG z; f C SG z; f D 0; z 2 n ;

Re TG z; f C SG z; f
(8.23)
D Re TG z0 ; f C SG z0 ; f ; z0 2 0 ; z 2 ;

Im TG z ; f C SG z ; f D 0;
where z 2 n is a fixed point.

Using these conditions, the operator SG z; f is defined uniquely. We define the

operators
P. f / D TG z; f C SG z; f ;
(8.24)
PAB . f / D P.Af / C P.Bf /;
where the functions A.z/ and B.z/ are from the right-hand part of Eq. (8.19).
If we now take into account that @Z P. f / D f .z/, then it can be easily verified
that the solution of the problem (8.19) satisfies the integral equation
w.z/ D PAB .w/ C P.d/: (8.25)
It is not difficult to show that the problems (8.19) and (8.25) are equivalent. Using
the properties of the operators TG z; f and P. f / [27], it can be shown that these
operators are completely continuous over the field of real numbers. It is obvious
that the operator PAB . f / is also completely continuous.
Since for d.z/ D 0 Eq. (8.19) has only a trivial solution, the equation w.z/ D
PAB .w/ will also have only a trivial solution. Hence, due to the complete continuity
of the operator PAB . f /, the existence and boundedness of the operator .I PAB /1
follow, where is an identical operator.
Let us introduce the notations
kI PA k1
C .G/;L
D M; kPkLp .G/;C .G/ D Mp ;
p .G/
where M and Mp are positive constants. From Eq. (8.25) we immediately obtain
kw.z/kC .G/ 6 MMp kdkLp .G/ : (8.26)
From Eq. (8.19) we can immediately obtain
k@Z wkLp .G/ 6 2NkwkC .G/ C kdkLp .G/ : (8.27)
The inequalities (8.26), (8.27) imply the estimate
kwkCp .G/ D kwkC .G/ C k@Z wkLp .G/ 6 kdkLp .G/ ;

where D MMp .2N C 1/ C 1. The lemma is proved.

8.3 Statement of an Optimal Control Problem
Let U be some bounded subset from E. Each function !.z/ W G ! U will be

called a control. The set U is called the control domain. We call the function !.z/
an admissible control if !.z/ 2 Lp .G/, p > 2. The set of all admissible controls is
denoted by .
For each fixed ! 2 in the domain G we consider the following Bitsadze
Samarski boundary value problem for a system of quasilinear differential equations
of first order
@Z w D f .z; w; w; !/; z 2 G;
Rew.z/ D '1 .z/; z 2 n ;
(8.28)
Rew.z.s// D Re w.z0 .s// ; z.s/ 2 ; z0 .s/ 2 0 ;
Imw.z / D c; z 2 n ; c D const; 0 < D const;
and the functional

I.u/ D F.x; y; w1 ; w2 ; !1 ; !2 / dx dy; (8.29)
G
where z D x C iy, w D w1 C iw2 , ! D !1 C i!2 .

Let us pose the following optimal control problem: Find a function !0 .z/ 2 ,
for which the solution of the BitsadzeSamarski boundary value problem (8.28)
gives the functional (8.29) a minimal value. We call the function !0 .z/ 2 an opti-
mal control, and the corresponding generalized solution w0 .z/ of the problem (8.28)
an optimal solution.
Assume that the conditions (A1)(A4) are fulfilled. Then for each fixed ! 2
the solution of the problem (8.1) exists in the space C ./ and is unique.
Furthermore, to obtain optimality conditions, it is additionally assumed that:
(A5) The function f .z; w; q; !/ has continuous partial derivatives which are also
continuous with respect to the same arguments as f . The function F is
continuous with respect to w1 , w2 , !1 , !2 , continuously differentiable with
respect to w1 , w2 and belongs to the space Lp .G/, p > 2.
(A6) For any .w; q/ 2 SwqR1
D f.w; q/ W jwj; jqj < Rg, the following estimates hold
in the domain G W j fw0 j; j fq0 j 6 N1 .R/ < C1, j f j 6 N2 .R/ < C1.
The following theorem is true:

Theorem 8.3.2 Let the conditions (A1)(A6) be fulfilled, w0 .z/ be an optimal
control, w0 .z/ be the corresponding solution if the problem (8.28), .x/ be a weight
function, .z/ be the solution of the adjoint problem
@f .!0 / @f .!0 /
@Z ..x/ .z// C .x/ .z/ C .x/ .z/
@w @w
D 2@w F.!0 /; z 2 G; (8.30)

Re .x/ .z/ D 0; z 2 ;

Re .x/ .zC
0 / .x/ .z0 / D Re .x/ .z/ ; z0 2 0 ; z 2
then the relation

h i
Re f .z; w0 ; w0 ; !0 / 0 .z/ C F.x; y; w01 ; w02 ; !01 ; !02 /
h i
D inf Re f .z; w0 ; w0 ; !/ 0 .z/ C F.x; y; w01 ; w02 ; !1 ; !2 /
!2U
is fulfilled everywhere on G.
Proof Let us construct a variant of the impulse control !0 .z/ [43]. Choose any finite
set of pairwise different Lebesgue points fzi g from the domain G. For each finite set
j
of non-negative numbers D fi g there exists a positive number "0 such that for
0 6 " 6 "0 the rectangles

ij" z W z D x C iy 2 G;
X
j
X
j1
xi " ik < x 6 xi " ik ; yi j" < y 6 yi . j 1/"
kD1 kD1
do not intersect pairwise and all of them belong to the domain G. Let u fui;j g be
the finite set of points from U. Consider an arbitrary set
j
fzi g; f!i;j g; fi g: (8.31)
Define the variant !" .z/ as follows

8

<!ij ; t 2 ij" ;
[
!" .z/ D !0 .t/; t 2 a; b n ij" : (8.32)
:
i;j
Assume that the set of parameters (8.31) is fixed. For fixed !0 and !" , 0 6 " 6 "0 ,
find an increment of the functional I.!/. Applying Lagranges theorem, we find

" I.!" / I.!0 / D Re " .z/! f .z; w0 ; !0 ; !" / dx dy
G

C ! F.x; y; w0 ; !0 ; !" / dx dy; (8.33)
G
where
u F.x; y; w0 ; !0 ; !" / D F.x; y; w01 ; w02 ; !"1 ; !"2 / F.x; y; w01 ; w02 ; !01 ; !02 /;
and the pulse variant !" .z/ is defined by means if the equality (8.28) for an arbitrary
set of parameters
j j
fzi g; ! D f!i;j g; D fi g; zi 2 G; !ij 2 U; i > 0: (8.34)
For an increment of the functional " I, let us construct some integral represen-
p
tation. For this, in the space C .G/, for any ", 0 6 " 6 "0 , we define the following
linear operator
Z1 Z1
@f ." / @f ." /
S" .v/ D @Z v dt v dt v; (8.35)
@w @w
0 0
where
" D .z; w0 C tw; w0 C tw; u" /; 0 6 t 6 1:

p
The operator S" maps the space C .G/ onto Lp .G/. Indeed, on the one hand, the
p
inclusion S" .v.z// 2 Lp .G/ is valid for any function v.z/ 2 C .G/ and, on the other
hand, by virtue of Lemma 8.2.1 we can conclude that the equation
S"0 .v/ D r.z/ (8.36)
has a unique solution for any function r.z/ 2 Lp .G/. In this case the estimate

kvkCp .G/ 6 krkLp .G/ ; D p; N1 ; jGj > 0 (8.37)

holds, from which it follows that for the operator S" there exists an inverse operator
S"1 W Lp .G/ ! C .G/ and kS"1 k 6 .
p
While considering the increment " I, we observe that the second summand in
the right-hand of equalities (8.33) defines uniquely some linear functional T" from
the adjoint space .C .G// :
p

hT" ; wi D 2 Re @w F./w dx dy; (8.38)
G
p
where hT" ; wi denotes a value of the functional T" on an element w 2 C .G/,
D .x; y; w01 C w1 ; w02 C w2 ; u01 C u1 ; u02 C u2 /, 0 6 6 1.
Consider the adjoint operator .S"1 / to the operator S"1 . This operator maps the
space .C .G// onto Lp .G/ since, by definition, hT" ; S"1 f i D h.S"1 / ; f i and
p

hT" ; wi D T" ; S"1 .! f / D .S"1 / T" ; ! f : (8.39)
By Riesz theorem [44] we can conclude that there exists a function " 2 Lp .G/,
1=p C 1=q D 1 such that for ! f 2 Lp .G/, p > 2, there hold the following
representations

.S"1 / T" ; ! f D Re " .z/! f dx dy: (8.40)
G
Taking into account the equalities (8.38)(8.40), the increment of the functional
" I (8.23) can be written in the form

" I D Re " .z/! f dx dy C 2 Re @! F./! dx dy; (8.41)
G G
where the function " 2 Lp .G/ is uniquely defined by the relation
" D .S"1 / T" : (8.42)
It can be easily shown that lim k " 0 kLq .G/ ! 0 [29]. Thus the function 0
"!0
is defined by means of the equality
0 D .S01 / T0 : (8.43)
Consider the first variation I of the functional (8.29):

1 1
I D lim 2
" I D lim 2 Re ! f .z; w0 ; !0 ; !" / " .z/ dx dy
"!0 " "!0 "
G

C ! F.x; y; w0 ; !0 ; !" / dx dy
G
X

D lim Re ! f .z; w0 ; !0 ; !" / " .z/ dx dy
"!0
16i6s
16j6q ij" .zi /

C ! F.x; y; w0 ; !0 ; !" / dx dy :
ij" .zi /
So, on any variant !" .z/, for the finite set (8.41) the variation I takes the form
X
h i
j
I D i Re ! f .zi ; w0 .zi /; !0 .zi /; !i;j / 0 .zi /C! F.zi ; w0 .zi /; !0 .zi /; !i;j / :
i;j
Therefore the relation

h i
Re f .z; w0 ; w0 ; !0 / 0 .z/ C F.x; y; w01 ; w02 ; !01 ; !02 /
h i
D inf Re f .z; w0 ; w0 ; !0 / 0 .z/ C F.x; y; w01 ; w02 ; !1 ; !2 /
!2U
holds almost everywhere on G [29, 43].
8.4 Construction of an Adjoint Equation in the Differential

Form
From the relation (8.41) we obtain the equality
h ; S0 wi D hT0 ; wi; 2 Lp .G/ (8.44)
p
for any w 2 C .G/.
By Riesz theorem [44], we can conclude that there exists a unique element from
the space Lq .G/, which is identified with an element 2 Lp .G/, 1=p C 1=q D 1
for which the equality (8.44) has the form

Re .z/S0 w dx dy D hT0 ; wi: (8.45)
G
Taking into account the form of the operator S0 and also the form of the functional
T0 , the equality (8.45) can be rewritten as follows
h @f .!0 / @f .!0 / i
Re .z/ @Z w w w dx dy
@w @w
G

D2 Re @w F.!0 /w dx dy: (8.46)
G
By Greens formula [27],

Z
1
Re .z/@Z w dx dy D w dz w@Z dx dy;
2i
G G
p
from the relation (8.46), for any function w 2 C .G/ we obtain the equality
Z
1
Re w dz
2i

h i
@f .!0 / @f .!0 /
Re @Z C C C 2@w F.!0 / w dx dy D 0:
@w @w
G
(8.47)
Let the domain G be the rectangle fz D x C iy W 0 6 x 6 1; 0 6 y 6 1g, be

the boundary of the domain G, and 0 D fz0 D x0 C iy W 0 6 y 6 1g, 1 D fz D
1 C iy W 0 6 y 6 1g, 2 D fz D x C i W 0 6 x 6 1g, 3 D fz D iy W 0 6 y 6 1g,
4 D fz D x W 0 6 x 6 1g, 0 < < 1, .x/ be the real weight function of a real
argument x defined as follows
(
1 .x/; 0 6 x 6 x0 ;
.x/ D
2 .x/; x0 < x 6 1;
1 .x/ 2 C 0; x0 ; 2 .x/ 2 C x0 ; 1; lim 1 .x/ lim 2 .x/:
x!x
0 x!xC
0
Then from the equality (8.47) we obtain for .z/ the following problem
@f .!0 / @f .!0 /
@Z C C D 2@w F.!0 /; z 2 G;
@w @w
Re D 0; z 2 2 [ 4 ; (8.48)
Rei D 0; z 2 3 [ 1 ;

Re .zC
0 / .z0 / D Re .z/; z0 2 0 ; z 2 1 :
8.5 Necessary and Sufficient Conditions of Optimality
Let the domain G be the rectangle fz D x C iy W 0 6 x 6 1; 0 6 y 6 1g, be

the boundary of the domain G, and 0 D fz0 D x0 C iy W 0 6 y 6 1g, 1 D fz D
1 C iy W 0 6 y 6 1g, 2 D fz D x C i W 0 6 x 6 1g, 3 D fz D iy W 0 6 y 6 1g,
4 D fz D x W 0 6 x 6 1g, z 2 n 1 , c.z/; d.z/; B.z/; f .z/ 2 Lp .G/, p > 2,
g.z/ 2 C . n 1 /, 0 < < 1. For each fixed ! 2 in the domain G we consider
the following BitsadzeSamarski boundary value problem for the system of linear
differential equations of first order
@Z w C B.z/w D f .z/!; z 2 G;
Rew.z1 / D g.z/; z 2 n 1 ;
(8.49)
Rew.z1 / D Rew.z0 /; z0 2 0 ; z1 2 1 ;
Imw.z / D const; 0 < D const.
Let us consider the functional

I.!/ D Re c.z/w.z/ C d.z/!.z/ dx dy (8.50)
G
and state the following optimal control problem: Find a function !0 .z/ 2 , for
which the solution of the BitsadzeSamarski boundary value problem (8.49) gives
the functional (8.50) a minimal value.
Theorem 8.5.1 Let .z/ be a solution of the adjoint problem
@Z ..x/ .z// B.z/.x/ .z/ D c.z/;

Re D 0; z 2 2 [ 4 ;
(8.51)
Rei D 0; z 2 3 [ 1 ;

Re .zC
0 / .z0 / D Re .z/; z0 2 0 ; z 2 1 ;
then for the optimality of !0 .z/, w0 .z/ it is necessary and sufficient that the following
equality be fulfilled almost everywhere on G
h i h i
Re d.z/ .x/ .z/f .z/ !0 .z/ D inf Re d.z/ .x/ .z/f .z/ !.z/ :
!2U
Proof Let !0 .z/ 2 be an optimal control, !" .z/ 2 be an arbitrary admissible

control, w0 .z/, w" .z/ be the solutions of the problem (8.49), which correspond to
the controls !0 .z/, !" .z/. Let us introduce the notations w.z/ D w" .z/ w0 .z/,
!.z/ D !" .z/ !0 .z/. Then we obtain the following problem
@Z w C B.z/w D f .z/!; z 2 G;
Rew.z/ D g.z/; z 2 n 1 ;
(8.52)
Rew.z1 / D Rew.z0 /; z0 2 0 ; z1 2 1 ;
Imw.z / D const; 0 < D const.
Let .z/ D 1 .x; y/ C i 2 .x; y/ 0 be an arbitrary integrable function. After

multiplying (8.52) by .x/ .z/, we obtain the equality

Re .x/ .z/ @Z w C B.z/w dx dy D Re .x/ .z/f .z/! dx dy:
G G
(8.53)
For fixed !0 .z/ and !" .z/ let us find an increment of the functional (8.54):

I D I.!" / I.!0 / D Re c.z/w.z/ C d.z/!.z/ dx dy:
G
In view of (8.53), we obtain the equality

h
i
I D Re .x/ .z/@Z w C c.z/ C .x/B.z/ .z/ w dx dy
G

C Re !.z/ d.z/ .x/ .z/f .z/ dx dy: (8.54)
G
Let w D u C iv and consider the first summand in the right-hand part of the
equality (8.54):

Re .x/ .z/@Z w dx dy
G
Z 1 Z1 @u
@v @u @v
D .x/ 1 .x; y/ .x/ 2 .x; y/ C dx dy:
@x @y @y @x
0 0
(8.55)
Furthermore, taking into account the boundary conditions from (8.52), we have
Z1 Z1 h #
@u @v
.x/ 1 .x; y/ .x/ 2 .x; y/ dx dy
@x @x
0 0
Z1 Zx0
@u @v
D 1 .x/ 1 .x; y/ 1 .x/ 2 .x; y/ dx
@x @x
0 0
Z1
@u @v
C 2 .x/ 1 .x; y/ 2 .x/ 2 .x; y/ dx dy
@x @x
x0
Z1
D 1 .x
0/

1 .x0 ; y/u.x0 ; y/ 1 .0/ 1 .0; y/u.0; y/
0
1 .x
0/

2 .x0 ; y/v.x0 ; y/ 1 .0/ 2 .0; y/v.0; 1/
Zx0 h@ i
@
.1 1 /u .1 2 /v dx
@x @x
0
C 2 .1/ 1 .1; y/u.1; y/ 2 .xC

0 /
C
1 .x0 ; y/u.x0 ; y/
2 .1/ 2 .1; y/v.1; y/ C 2 .xC

0 /
C
2 .x0 ; y/v.x0 ; y/
Z1 h i
@ @
.2 1 /u .2 2 /v dx dy
@x @x
x0
Z1
h i
D 1 .x
0/

1 .x0 ; y/ 2 .xC
0 /
C
1 .x0 ; y/ C 2 .1/ 1 .1; y/ u.x0 ; y/
0
h i
C 2 .xC
0 /
C
2 .x0 ; y/ 1 .x
0/

2 .x0 ; y/ v.x0 ; y/
1 .0/ 2 .0; y/v.0; y/ 2 .1/ 2 .1; y/v.1; y/
Zx0 h
@ @ i Z1 h
@ @ i
. 1 /u . 2 /v dx . 1 /u . 2 /v dx dy:
@x @x @x @x
0 x0
Analogously, we obtain the equality
Z1 Z1 h
@u @v i
.x/ 2 .x; y/ C .x/ 1 .x; y/ dy dx
@y @y
0 0
Zx0
D .x/ 1 .x; 0/v.x; 0/ C .x/ 1 .x; 1/v.x; 1/

0
Z1 h i
@ @
. 2 /u C . 1 /v dy dx
@y @y
0
Z1
C .x/ 1 .x; 0/v.x; 0/ C .x/ 1 .x; 1/v.x; 1/

x0
Z1 h i
@ @
. 2 /u C . 1 /v dy dx:
@y @y
0
Let us assume that the following conditions are fulfilled:
1 .x
0/

1 .x0 ; y/ 2 .xC
0 /
C
1 .x0 ; y/ C 2 .1/ 1 .1; y/ D 0;
2 .xC
0 /
C
2 .x0 ; y/ 1 .x
0/

2 .x0 ; y/ D 0;
(8.56)
1 .0/ 2 .0; y/ D 0; 2 .1/ 2 .1; y/ D 0;
.x/ 1 .x; 0/ D 0; .x/ 1 .x; 1/ D 0:
By the above reasoning, from the equality (8.55) we obtain the relation

Re .x/ .z/@Z w dx dy D Re @Z ..x/ .z//w dx dy
G G1

Re @Z ..x/ .z//w dx dy; (8.57)
G2
where

G1 D z D x C iy W 0 6 x 6 x0 ; 0 6 y 6 1 ;

G2 D z D x C iy W x0 6 x 6 1; 0 6 y 6 1 :
Using (8.57), for the increment of the functional I from (8.54) we obtain the
equality
h i
I D Re @Z ..x/ .z// C c.z/ C .x/B.z/ .z/ w dx dy
G1

C Re !.z/ d.z/ .x/ .z/f .z/ dx dy
G
h i
C Re @Z ..x/ .z// C c.z/ C .x/B.z/ .z/ w dx dy: (8.58)
G2
Let us consider the equation
@Z .x/ .z// B.z/..x/ .z// D c.z/ (8.59)
and the function

(
.1/
i .x; y/; .x; y/ 2 G1 ;
i .x; y/ D .2/
i D 1; 2:
i .x; y/; .x; y/ 2 G2 ;
.1/ .1/ .2/

where 1 . 1 C i 2 / is a solution of Eq. (8.59) in the domain G1 , and 2 . 1 C
.2/
i 2 / is a solution in the domain G2 . It is assumed that the function .x/ .z/
satisfies also the conditions (8.56). In that case, the increment of the functional
I (8.58) takes the form

I D I.!" / I.!0 / D Re !.z/ d.z/ .x/ .z/f .z/ dx dy: (8.60)
G
From the representation (8.60) it obviously follows that for the optimality of
!0 .z/, w0 .z/ it is necessary and sufficient that the following equality be fulfilled
almost everywhere on G
h i h i
Re d.z/ .x/ .z/f .z/ !0 .z/ D inf Re d.z/ .x/ .z/f .z/ !.z/ :
!2U
The theorem is proved.

Acknowledgements The authors were supported by the European Unions Seventh Framework
Programme (FP7/20072013) under grant agreement no. 317721.
References
1. V.V. Shelukhin, A non-local in time model for radionuclides propagation in Stokes fluid.
Dinamika Sploshn. Sredy 107, 180193 (1993)
2. C.V. Pao, Reaction diffusion equations with nonlocal boundary and nonlocal initial conditions.
J. Math. Anal. Appl. 195(3), 702718 (1995)
3. E. Obolashvili, Nonlocal problems for some partial differential equations. Appl. Anal. 45(14),
269280 (1992)
4. T.V. Aloyev, E.N. Aslanova, Nonlocal problems of conductive radial heat exchange, in
Abstracts of International Conference Non-local Boundary Problems and Related Mathe-
matical Biology, Informatic and Physic Problems, Nalchik, 1996
5. J.I. Diaz, J.-M. Rakotoson, On a nonlocal stationary free-boundary problem arising in the
confinement of a plasma in a stellarator geometry. Arch. Rational Mech. Anal. 134(1),
5395 (1996)
6. T. Carleman, Sur la thorie des equations integrals et ses applications. in Verh. Internat. Math.
Kongr., Zurich (Orell Fussli, Zurich, 1932/1933), pp. 138151
7. R. Beals, Nonlocal elliptic boundary value problems. Bull. Am. Math. Soc. 70, 693696 (1964)
8. F.E. Browder, Non-local elliptic boundary value problems. Am. J. Math. 86, 735750 (1964)
9. J.R. Cannon, The solution of the heat equation subject to the specification of energy. Q. Appl.
Math. 21, 155160 (1963)
10. N.I. Ionkin, The solution of a certain boundary value problem of the theory of heat conduction
with a nonclassical boundary condition. Differ. Uravn. 13(2), 294304, 381 (1977, in Russian)
11. G. Berikelashvili, To a nonlocal generalization of the Dirichlet problem. J. Inequal. Appl. 2006,
Art. ID 93858, 6 pp. (2006)
12. F. Shakeri, M. Dehghan, The method of lines for solution of the one-dimensional wave equation
subject to an integral conservation condition. Comput. Math. Appl. 56(9), 21752188 (2008)
13. A. Ashyralyev, O. Gercek, Nonlocal boundary value problems for elliptic-parabolic differential
and difference equations. Discret. Dyn. Nat. Soc. 2008, Art. ID 904824, 16 pp. (2008)
14. A.V. Bitsadze, A.A. Samarski, On some simple generalizations of linear elliptic boundary
problems. Dokl. Akad. Nauk SSSR 185, 739740 (1969, in Russian); English transl.: Sov.
Math. Dokl. 10, 398400 (1969)
15. D.G. Gordeziani, T.Z. Dzhioev, The solvability of a certain boundary value problem for a
nonlinear equation of elliptic type. Sakharth. SSR Mecn. Akad. Moambe 68, 289292 (1972,
in Russian)
16. D.G. Gordeziani, Methods for Solving a Class of Nonlocal Boundary Value Problems. With
Georgian and English summaries (Tbilis. Gos. Univ., Inst. Prikl. Mat., Tbilisi, 1981, in
Russian), 32 pp.
17. M.P. Sapagovas, R.Yu. Chegis, Boundary-value problems with nonlocal conditions. Differ.
Uravn. 23(7), 12681274 (1987, in Russian); translation in Differ. Equ. 23(7), 858863 (1987)
18. B.P. Paneyakh, Some nonlocal boundary value problems for linear differential operators. Mat.
Zametki 35(3), 425434 (1984, in Russian)
19. A.L. Skubachevski, On a spectrum of some nonlocal boundary value problems. Mat. Sb.
117(7), 548562 (1982, in Russian)
20. D. Devadze, V. Beridze, An optimal control problem for Helmholtz equations with Bitsadze
Samarski boundary conditions. Proc. A. Razmadze Math. Inst. 161, 4753 (2013)
21. D. Devadze, M. Dumbadze, An optimal control problem for a non-local boundary value
problem. Bull. Georgian Natl. Acad. Sci. (N.S.) 7(2), 7174 (2013)
22. D.V. Kapanadze, On a nonlocal BitsadzeSamarski boundary value problem. Differ. Uravn.
23(3), 543545, 552 (1987, in Russian)
23. V.A. Ilin, E.I. Moiseev, A two-dimensional nonlocal boundary value problem for the Poisson
operator in the differential and the difference interpretation. Mat. Model. 2(8), 139156 (1990,
in Russian)
24. D. Gordeziani, N. Gordeziani, G. Avalishvili, Non-local boundary value problems for some
partial differential equations. Bull. Georgian Acad. Sci. 157(3), 365368 (1998)
25. A.K. Gushchin, V.P. Mikhailov, On the solvability of nonlocal problems for a second-order
elliptic equation. Mat. Sb. 185(1), 121160 (1994, in Russian); translation in Russian Acad.
Sci. Sb. Math. 81(1), 101136 (1995)
26. P.L. Gurevich, Asymptotics of solutions for nonlocal elliptic problems in plane bounded
domains. Funct. Differ. Equ. 10(12), 175214 (2003). Functional differential equations and
applications (Beer-Sheva, 2002), No. 4, 773775
27. I.N. Vekua, Generalized Analytic Functions, 2nd edn. Edited and with a preface by
O.A. Oleinik and B.V. Shabat (Nauka, Moscow,1988, in Russian)
28. G.F. Mandzhavidze, V. Tuchke, Some boundary value problems for first-order nonlinear
differential systems on the plane, in Boundary Value Problems of the Theory of Generalized
Analytic Functions and Their Applications (Tbilis. Gos. Univ., Tbilisi, 1983, in Russian),
pp. 79124
29. G.V. Meladze, T.S. Tsutsunava, D.Sh. Devadze, in The Optimal Control Problem for Quasilin-
ear Differential Equations of First Order on the Plane with Nonlocal Boundary Conditions
(Tbilisi State University Press, Tbilisi, 1987). Deposited at Georgian Res. Inst. Sci. Eng.
Inform., 25.12.87, No. 372, 87, 61 pp.
30. D.Sh. Devadze, V.Sh. Beridze, Optimality conditions for quasilinear differential equations with
nonlocal boundary conditions. Uspekhi Mat. Nauk 68(4(412)), 179180 (2013, in Russian);
translation in Russian Math. Surv. 68(4), 773775 (2013)
31. Z.-L. Lions, The optimal control of distributed systems. Uspehi Mat. Nauk 28(4(172)), 1546
(1973, in Russian); Translated from the French: Enseignement Math. (2) 19, 125166 (1973)
32. V.A. Troitski, L.V. Petukhov, Optimization of the Form of Elastic Bodies (Nauka, Moscow,
1982, in Russian)
33. A.G. Butkovski, Theory of Optimal Control of Systems with Distributed Parameters (Nauka,
Moscow, 1977)
34. R. Glovinski, Zh.L. Lions, R. Tremoler, Numerical Analysis of Variational Inequalities.
Translated from the French by A.S. Kravchuk (Mir, Moscow, 1979, in Russian)
35. A.G. Butkovskiy, Metody upravleniya sistemami s raspredelennymi parametrami (Distributed-
Parameter Systems: Methods of Control) (Nauka, Moscow, 1975)
36. D. Devadze, V. Beridze, An algorithm of the solution of an optimal control problem for a
nonlocal problem. Bull. Georgian Natl. Acad. Sci. (N.S.) 7(1), 4448 (2013)
37. L.S. Pontryagin, V.G. Boltyanski, R.V. Gamkrelidze, E.F. Mishchenko, The Mathematical
Theory of Optimal Processes, 4th edn. (Nauka, Moscow, 1983, in Russian)
38. Z.-L. Lions, in Optimal Control of Systems That Are Governed by Partial Differential
Equations, ed. by R.V. Gamkrelidze. Translated from the French by N.H. Rozov (Izdat. Mir,
Moscow, 1972, in Russian)
39. V.I. Plotnikov, Necessary and sufficient conditions for optimality and conditions for uniqueness
of the optimizing functions for control systems of general form. Izv. Akad. Nauk SSSR Ser.
Mat. 36, 652679 (1972, in Russian)
40. O.A. Ladyzhenskaya, N.N. Uraltseva, Linear and Quasilinear Equations of Elliptic Type, 2nd
rev. edn. (Izdat. Nauka, Moscow, 1973, in Russian)
41. V.S. Vladimirov, The Equations of Mathematical Physics, 4th edn. (Nauka, Moscow, 1981, in
Russian)
42. R. Courant, D. Hilbert, Methods of Mathematical Physics, vol. I (Interscience, New York,
1953)
43. V.I. Plotnikov, V.I. Sumin, Optimization of objects with distributed parameters that can be
described by GoursatDarboux systems. Z. Vycisl. Mat. i Mat. Fiz. 12, 6177 (1972, in
Russian)
44. N. Danford, Dz. Shvarc, Linear Operators. Part I: General Theory (Izdat. Inostran. Lit.,
Moscow, 1962, in Russian)
Chapter 9
On the Geometry of the Domain of the Solution
of Nonlinear Cauchy Problem
. Figula and M.Z. Menteshashvili
Abstract We consider the Cauchy problem for a second order quasi-linear partial
differential equation with an admissible parabolic degeneration such that the given
functions described the initial conditions are defined on a closed interval. We study
also a variant of the inverse problem of the Cauchy problem and prove that the
considered inverse problem has a solution under certain regularity condition. We
illustrate the Cauchy and the inverse problems in some interesting examples such
that the families of the characteristic curves have either common envelopes or
singular points. In these cases the definition domain of the solution of the differential
equation contains a gap.
9.1 Introduction
Let us consider a Cauchy problem for second order non-strictly quasi-linear

hyperbolic partial differential equations: find a solution u.x; y/ of the equation by
the initial conditions ujyD0 D .x/, uy jyD0 D .x/, where .x/ 2 C1 .R/, .x/ 2
C2 .R/ are given functions such that .x/ is once-, and .x/ is twice-continuously
differentiable. To solve this problem one can use the method of characteristics (cf.
[18, pp. 164166]). The characteristic roots of the partial differential equation are
the solutions of the corresponding characteristic equation. The class of hyperbolic
equations is defined through the characteristic roots by the inequality 1 2 .
. Figula ()
Institute of Mathematics, University of Debrecen, P.O. Box 400, 4002 Debrecen, Hungary
e-mail: figula@science.unideb.hu
M.Z. Menteshashvili
Muskhelishvili Institute of Computational Mathematics of the Georgian Technical University, 4,
Grigol Peradze Str., 0131 Tbilisi, Georgia
Sokhumi State University, 0186 Tbilisi, Georgia
e-mail: m.menteshashvili@gtu.ge

206 . Figula and M.Z. Menteshashvili
Characteristic directions are defined at every point by the relations
dy dy
D 1 .x; y; u; ux ; uy /; D 2 .x; y; u; ux ; uy /: (9.1)
dx dx
Solving the differential equations (9.1) we obtain the characteristic invariants (cf.
[17]).
Among the papers based on the application of the method of characteristics to
nonlinear hyperbolic problems we refer to [2, 6, 7, 913, 15, 16], where the structure
of the definition domains of the solutions and that of influence domains of initial and
characteristic perturbations are investigated in singular cases.
The aim of our investigation is to discuss some questions stating an initial
problem on the data support a; b 2 R for the following second order non-strictly
hyperbolic equation
.u2y 1/uxx 2ux uy uxy C u2x uyy D 0 (9.2)
(cf. [5, p. 442]). We consider also a variant of the inverse problem of the Cauchy
problem for Eq. (9.2) on some open and closed data supports. Equation (9.2) is
interesting for physical applications since the two-dimensional flow of an inviscid
incompressible fluid in gravitational field can be described by it (cf. [19, p. 535]).
The characteristic roots of (9.2), i.e. the solution of the equation .u2y 1/2 C
2ux uy C u2x D 0, are
ux ux
; :
uy C 1 uy 1
Therefore Eq. (9.2) is hyperbolic everywhere, except for the points at which the
derivative ux of the sought solution u.x; y/ has zero values. At these points, Eq. (9.2)
parabolically degenerates. Hence Eq. (9.2) has mixed type (cf. [4]). It is one of
the most important problems of mathematical physics to study the properties of
solutions of equations of mixed type (cf. [3, 20, 22, 23]). The first fundamental
results in this direction were obtained by the Italian mathematicians Francesco
Tricomi in the twenties of the nineteenth century.
The combinations ux uy , ux C uy of the first derivatives of the solution which are
constant for the string vibration equation uxx uyy D 0 differ from those for non-
linear equation (9.2). In the latter case these combinations depend on an unknown
solution and its first derivatives. Hence it is rather difficult to use the characteristic
invariants for the solution of problems in the case of quasi-linear equations. In
aerodynamics, these combinations are called Riemann invariants (cf. [4, p. 23]).
Solving the equations dy dx
D uyuC1
x dy
, dx D uyu1x
we obtain that the characteristic
invariants for Eq. (9.2) are u C y D const and u y D const.
According to Hadamard (cf. [14]) the Cauchy problem can appear to be well-
posed in some cases involving the closed initial support. Such problems were
9 On the Geometry of the Domain of the Solution of Nonlinear Cauchy Problem 207
studied by Aleksandryan [1], Sobolev [21], Vakhania [24], Wolfersdorf [25], and
others for linear equations.
The Cauchy problem of Eq. (9.2) such that the given functions , are defined
in the circumference of the unit circle, i.e. the initial problem on the closed data
support W D 1; x D cos #; y D sin #, 0 6 # 6 2, was investigated in [16].
In this paper we study the Cauchy problem of Eq. (9.2) on the data support a; b
(see Problem 1 in Sect. 9.2). In Theorem 1 we give the solution of the Cauchy
problem of (9.2) in integral form. To establish this result we use the representation
formula of the general integral for (9.2). After this we find a domain where the
solution can be completely defined. For this we have to write equations for the
characteristic curves of both families. Then we consider the set of intersection points
for both families of characteristic curves. This set creates the domain within which
the initial support is located. In Sect. 9.3 we formulate a variant of the inverse
problem of the Cauchy problem for Eq. (9.2) (see Problem 2). We prove that the
considered inverse problem has a solution under certain regularity condition (cf.
Theorem 2). We illustrate the Cauchy and the inverse problems for Eq. (9.2) in
Examples 13. In Examples 2, 3 we deal with the interesting cases that the families
of the characteristic curves have either common envelopes or singular points.
9.2 Cauchy Problem on Eq. (9.2)
Problem 1 Let .x/, respectively .x/ be real functions, which are twice, respec-
tively once continuously differentiable. Find a function u.x; y/, which satisfies
Eq. (9.2) and the conditions
u.x; 0/ D .x/; uy .x; 0/ D .x/; a 6 x 6 b: (9.3)
It is also required to find a domain where the solution can be completely defined.
Theorem 1 If 0 .x/ 0 for all x 2 a; b, then the integral of problem (9.2), (9.3)
can be written into the form
Z Z
1 T.uCy/
1 T.uy/
xDaC .1 .t//dt C .1 C .t//dt; (9.4)
2 a 2 a
where x D T.z/ denotes the inverse function of z D .x/.

Proof We start the investigation of the problem by considering the form of the
general integral of Eq. (9.2). Using a straightforward calculation it can be checked
that the solution u.x; y/ of Eq. (9.2) satisfies the following functional equation
f .u C y/ C g.u y/ D x; (9.5)
where f , g are arbitrary functions (see [8]). In order to provide the required
smoothness of the solution u.x; y/, here it is assumed that the arbitrary functions
f , g belong to the class C2 .R/. Taking the derivations of (9.5) with respect to the
variables x and y and putting into these y D 0 we obtain
f 0 .u.x; 0/ C 0/ux .x; 0/ C g0 .u.x; 0/ 0/ux .x; 0/ D 1 (9.6)
f 0 .u.x; 0/ C 0/.uy.x; 0/ C 1/ C g0 .u.x; 0/ 0/.uy .x; 0/ 1/ D 0: (9.7)
Hence we come to a system of two linear algebraic equations (9.6), (9.7) with
respect to the variables f 0 .u.x; 0//, g0 .u.x; 0//. Solving this system of linear
equations one gets
1 .x/
f 0 .u.x; 0// D WD F.x/ (9.8)
2 0 .x/
1 C .x/
g0 .u.x; 0// D WD G.x/: (9.9)
2 0 .x/
If the condition 0 .x/ 0 is fulfilled for all x 2 a; b, then on the closed interval
a; b the equation .x/ D z is uniquely solvable in the class of real solutions. We
denote this solution by x D T.z/. Integrating relations (9.8), (9.9) between .a/ and
an arbitrary value z 2 .a/; .b/ we obtain
Z z Z x
0
f .z/ f ..a// D f .v/dv D F.t/ 0 .t/dt;
.a/ a
Z z Z x
g.z/ g..a// D g0 .v/dv D G.t/ 0 .t/dt;
.a/ a
or equivalently
Z T.z/
f .z/ D f ..a// C F.t/ 0 .t/dt; (9.10)
a
Z T.z/
g.z/ D g..a// C G.t/ 0 .t/dt: (9.11)
a
Their sum already yields an implicit solution of problem (9.2), (9.3) which contains
the undefined free functions for x D a. They can be identified by normalization. In
particular, if for the functions f , g defined by (9.10), (9.11), relation (9.5) is fulfilled
at the point .a; 0/, then we have
Z Tu.a;0/C0 Z Tu.a;0/0
0
f ..a//Cg..a//C F.t/ .t/dtC G.t/ 0 .t/dt D a: (9.12)
a a
Note that the identity T..x// D x is valid for all values of x 2 a; b, including
x D a, and therefore the upper bounds of both integrals in (9.12) are a. Hence one
has f ..a// C g..a// D a: Finally, the implicit solution of problem (9.2), (9.3) can
be written as
Z T.uCy/ Z T.uy/
xaD F.t/ 0 .t/dt C G.t/ 0 .t/dt; (9.13)
a a
where the functions F.t/ and G.t/ are defined by (9.8), (9.9). This representation
is obtained if condition 0 .x/ 0 is fulfilled for all x 2 a; b and the expressions
under the integral sign are integrable. t
u
In order to establish the definition domain of the solution u.x; y/ of (9.2) given
by (9.13), it is necessary to investigate the structure of the characteristic curves.
Keeping in mind representation (9.13) and that the relation u C y D const is fulfilled
along the characteristic curves of the first family, we can obtain an equation for
each characteristic of this family that passes through an arbitrary point x of the data
support a; b. At this arbitrarily chosen point, with which we associate the argument
x0 ; we have
u.x0 ; 0/ D .x0 /:
Hence the relation
u.x; y/ C y D .x0 / (9.14)
is fulfilled along the characteristic curve satisfying the relation u C y D const and
passing through the point .x0 ; 0/. Since the right-hand side of this equality is a
completely defined expression, we conclude that
u.x; y/ y D .x0 / 2y: (9.15)
Substituting (9.14) and (9.15) into (9.13), we obtain the equation for a characteristic
curve of the first family in the form
Z T. .x0 // Z T. .x0 /2y/
xaD F.t/ 0 .t/dt C G.t/ 0 .t/dt: (9.16)
a a
Keeping in mind that the identity T..x0 // D x0 is valid, the upper bound of the first
integral in (9.16) will be x0 . If we introduce the notation x0 D c, then this expression
becomes the parameter which takes values from the interval a; b. Therefore, all
characteristic curves of the family u C y D const which pass through the points of
the data support have the representation
Z c Z T. .c/2y/
xDaC F1 .t/dt C G1 .t/dt; (9.17)
a a
where
1 .t/
F1 .t/ D F.t/ 0 .t/ D
2
1 C .t/
G1 .t/ D G.t/ 0 .t/ D :
2
Analogously, for all characteristic curves of the family u y D const we obtain
Z T. .c/C2y/ Z c
xDaC F1 .t/dt C G1 .t/dt: (9.18)
a a
Here the parameter c takes its values from the interval a; b. For both families,
in (9.17) and (9.18) there is a one-to-one correspondence between the equation for a
characteristic curve and the parameter value. This fact is stipulated by the condition
0 .x/ 0 for all x 2 a; b.
9.3 Inverse Problem
Now we consider certain inverse problems of the Cauchy problem for Eq. (9.2).
Below we will treat a variant of inverse problem which requires the construction of
a given equation under the condition that the characteristic curves of both families
are known a priori. The cases that the families of characteristic curves have either
common envelopes or singular points are particularly interesting to investigate. In
these situations, gaps may appear in the definition domain of a solution of the
problem.
As has been mentioned above, Eq. (9.2) is of hyperbolic type and the set of
parabolic degeneration is not defined a priori because it depends on the behavior
of an arbitrary solution u.x; y/ and its derivative with respect to the variable x.
The characteristic invariants (cf. [4, p. 23]) corresponding to Eq. (9.2) are given
as follows
ux
u C y D const; for 1 D ; (9.19)
uy C 1
ux
u y D const; for 2 D : (9.20)
uy 1
From these equations we conclude that the families of characteristic curves of the
equation are not defined a priori because they depend on the sought solution and on
its first order derivatives.
In the case of Eq. (9.2), we can admit in the role of characteristics the families
of characteristic curves along which relations (9.19) and (9.20) are fulfilled.
Characteristic families can therefore be given a priori in an arbitrary manner. The

statement of the inverse problem considered here rests on this fact.
Let us consider two one-parameter families of plane curves which are given
explicitly by the equations
y D '1 .x; c/; (9.21)
y D '2 .x; c/; (9.22)
where '1 , '2 are given, twice differentiable functions with respect to the variable x.
Let '1 , '2 be defined for any value of the real parameter c. Assume that any curve
of these families necessarily intersects once the straight line y D 0. We denote
by D1 the domain of the plane .x; y/ which is completely covered by the family
of characteristic curves given by Eq. (9.21) if the parameter c runs continuously
through all real values. Analogously, we denote by D2 the domain of the plane .x; y/
which is completely covered by the characteristic lines given by Eq. (9.22). Also,
we introduce the notation D D D1 \ D2 , I D D \ fy D 0g:
Problem 2 (Inverse Problem) Find the initial conditions of a regular solution
u.x; y/ of Eq. (9.2) and its derivative with respect to the normal direction on the
interval I of the straight line y D 0 if the families of plane curves given by Eq. (9.21)
are the characteristics corresponding to invariants (9.19), while (9.22) is the family
of characteristics corresponding to invariants (9.20).
Theorem 2 If the condition '10 .x; c0 / '20 .x; c00 / is fulfilled for all c0 , c00 2 R, then
there exists a solution of Problem 2.
Proof To investigate the posed problem, we need a structural analysis of the
characteristic curves of Eq. (9.2). The characteristic roots corresponding to Eq. (9.2)
define at every point two characteristic directions
dy ux
D 1 .x; y/; (9.23)
dx uy C 1
dy ux
D 2 .x; y/: (9.24)
dx uy 1
It follows from the posed problem that the family of characteristic curves defined by
Eq. (9.21) corresponds to the root 1 given by (9.23), while that given by Eq. (9.22)
corresponds to the root 2 given by (9.24). Therefore we can calculate the values of
first derivatives ux , uy of the unknown solution u.x; y/ along any characteristic curve.
Note that the parameters contained in families (9.21) and (9.22) can be defined by
the abscissa of the intersection point of the concrete curve and the axis y D 0.
Indeed, let the curve corresponding to the parameter c ; intersecting the straight
line y D 0 at a point x0 be
'1 .x0 ; c / D 0:
Solving the equation for the parameter c .x0 / D c; family (9.21) takes the form
y D '1 .x; x0 /;
which is equal to zero for x D x0 . If family (9.21) corresponds to the characteristic

root 1 , then the relation
d'1 .x; x0 / ux .x; '1 .x; x0 //

D : (9.25)
dx uy .x; '1 .x; x0 // C 1
is fulfilled. Analogously, along the second family of the characteristic curves the
following equality is fulfilled
d'2 .x; x0 / ux .x; '2 .x; x0 //

D : (9.26)
dx uy .x; '2 .x; x0 // 1
Equalities (9.25), (9.26) have the following at the points of the interval I:
d'1 .x; x0 / 0 .x0 /

jxDx0 D ; x0 2 I; (9.27)
dx .x0 / C 1
d'2 .x; x0 / 0 .x0 /

jxDx0 D ; x0 2 I: (9.28)
dx .x0 / 1
Since the point .x0 ; 0/ is arbitrarily chosen from the interval I, x0 in

Eqs. (9.27), (9.28) can be replaced with x. Hence we can easily define the unknown
functions 0 and . Integrating the function 0 , we finally obtain the solutions .x/,
.x/ of the problem. Naturally, the function is defined up to an integration constant
which will be defined uniquely if we give the value of the function u.x; y/ at one
arbitrary point of the interval I. The theorem is thereby proved. t
u
From the above general argumentation we can draw a conclusion. If fami-
lies (9.21) and (9.22) do not have common characteristic directions at anyone of
the points, i.e. if Eq. (9.2) is of strictly hyperbolic type, then the solution of the
inverse problem presents no difficulty (see Example 1). However, if the equation
parabolically degenerates on certain set of points, then the situation drastically
changes. We illustrate this in Examples 2, 3.
Example 1 We solve the Cauchy and inverse problems in the following concrete
example. Cauchy problem: Let u.x; 0/ D .x/ be the function x and uy .x; 0/ D .x/
be the function 1 ex . According to (9.4) the solution of the Cauchy problem in
implicit form is given by equation
1 uCy 1
e C u y euy x D 0:
2 2
The characteristic curves (9.17) of the family u C y D const, as parametric curve

f .x; y; c/ with respect to the parameter c are given by
1 c 1
0D e C c 2y ec2y x D f1 .x; y; c/: (9.29)
2 2
The characteristic curves (9.18) of the family u y D const, as parametric curve
f .x; y; c/ with respect to the parameter c are given by
1 cC2y 1
0D e C c ec x D f2 .x; y; c/: (9.30)
2 2
Differentiating the function f1 .x; y; c/ in (9.29) with respect to the parameter c we

get
1 c
0D e .1 e2y / C 1 D f1;c .x; y; c/: (9.31)
2
Expressing from (9.31) the parameter c we have

2e2y
c D ln : (9.32)
1 e2y
Putting expression (9.32) into Eq. (9.29) we obtain
2
e1Cx D : (9.33)
1 e2y
@f .x;y;c/ @f .x;y;c/
Since the derivative 1;c@y D ec2y is non-zero and the derivative 1;c@c D
1 2y c
2
.1 e /e is non-zero for all y 0 we obtain that the envelope of the character-
istic curves (9.29) of the first family is given by (9.33). An analogous computation
shows that this envelope is also the envelope of characteristic curves (9.30) of the
second family. In the plane .x; y/ for the domain which is below of envelope (9.33)
there does not exist solution of the Cauchy problem (see Fig. 9.1 for parameter
c D 1).
Inverse problem: Assume that the families of characteristic curves are given by:
1 c 1 c2y
x D '1 .y; c/ e e C c 2y; for 1 ; (9.34)
2 2
1 1
x D '2 .y; c/ ec C ecC2y C c; for 2 : (9.35)
2 2
From (9.27), (9.28) we get
ux .x0 ; 0/ 1
D x ; (9.36)
uy .x0 ; 0/ C 1 e 0 2
c = 1.
y 1
4 2 0 2 4
x
1
characteristic curve (9.29) characteristic curve (9.30)

envelop of characteristics
Fig. 9.1 The characteristic curves (9.29), (9.30) and their envelopes
for family (9.34) and
ux .x0 ; 0/ 1
D x ; (9.37)
uy .x0 ; 0/ 1 e0
for family (9.35).

The right-hand sides of the equalities (9.36), (9.37) depend only on x0 . From the
relations (9.36) and (9.37) for the point .x0 ; 0/ we find the values of the functions 0
and :
0 .x/ D 1; (9.38)
.x/ D 1 ex :
From equality (9.38) we obtain by integration the value of the solution u.x; y/ on
the axis y D 0:
.x/ D x C c; c D const. (9.39)
The constant c in (9.39) can be obtained if we know the value of .x/ in one arbitrary
point.
Example 2 Now we consider the following inverse problem. Assume that the
families of characteristic curves have common envelopes and are given in the form
r
a
x D '1 .y; c/ c a 1; b 6 y 6 a b; (9.40)
yCb
r
a
x D '2 .y; c/ c C a 1; b 6 y 6 a b; (9.41)
yCb
a > b > 0; c D const.
Each curve of family (9.40) defined on an interval x 2 .1; c monotonically

increases, has the asymptote y D b. For x D c each curve of family (9.40) is
tangent to the straight line y D a b and therefore is tangent to one of the curves of
family (9.40). One of the curves of family (9.40) passes at every point of the straight
line y D ab and smoothly continues from the same point to the completely defined
curve of family (9.41). Therefore both characteristic directions at the points of this
straight line coincide. Hence the straight line y D a b is the line of parabolic
degeneration for Eq. (9.2). The straight line y D b is also the line of parabolic
degeneration because all characteristic curves of both families are tangent to the
straight line at infinity (see Fig. 9.2).
By direct calculations we establish that
r
ux .x; y/ 2a3 .x c1 / ab
D ; c1 D x0 C a ; (9.42)
uy .x; y/ C 1 .x c1 /2 C a2 2 b
Fig. 9.2 The characteristic curves (9.40), (9.41) and their envelopes
holds for family (9.40) and that

r
ux .x; y/ 2a3 .x c2 / ab
D ; c2 D x0 a ; (9.43)
uy .x; y/ 1 .x c2 /2 C a2 2 b
is satisfied for family (9.41). Note that the right-hand sides of the equali-
ties (9.42), (9.43) are represented by expressions depending solely on the variable
x. Relations (9.42) and (9.43) for the point .x0 ; 0/ yield the equalities
0 .x0 / 2a3 .x0 c1 /

D ;
.x0 / C 1 .x0 c1 /2 C a2 2
for the first family of the characteristic curves, and
0 .x0 / 2a3 .x0 c2 /

D ;
.x0 / 1 .x0 c2 /2 C a2 2
for the second family. Let us assume that the point x0 2 .1; C1/ is arbitrary,
then for any .x; 0/ we obtain
r
0 .x/ 2b2 ab
D 2 ; (9.44)
.x/ C 1 a b
r
0 .x/ 2b2 ab
D 2 : (9.45)
.x/ 1 a b
Hence we find the values of the functions 0 and :

r
0 2b2 a b
.x/ D 2 ; (9.46)
a b
.x/ D 0:
From equality (9.46) we find by integration the values of the solution u.x; y/ on the
axis y D 0:
r
2b2 ab
.x/ D 2 x C c; c D const.
a b
As relations (9.44), (9.45) show the characteristic roots uyuC1x

; uyu1
x
of both
families take a constant value. Despite this, we cannot assert that the straight line
y D 0 is the line of parabolic degeneration. To make such an assertion it is necessary
that the invariants u C y and u y are constant along y D 0. The constancy of the
invariants is due to the fact that families (9.40), (9.41) are sets of curves obtained
by a parallel translation along the axis of abscissas. Therefore every curve of both
families has one and the same slope with respect to the axis of abscissas. This
indicates that the values of the derivative dy
dx along this axis preserve constancy.
Example 3 The situation is more difficult if the families of the characteristic curves
have, besides the lines of parabolic degeneration, also common singular points. As
an example let us consider the case that families (9.21), (9.22) have the common
node. Each contour of the one-parameter family of curves
( 2 ) 1
.1 r cos #/ r sin #
F.r; #; #0 / D 2 1C 2 C #0 D 0;
r 2r cos # C 1 r 2r cos # C 1
(9.47)
0 6 # 6 2 is closed. The whole family lies in the half plane to the left of the
straight line x D 1. The circumference
.2x 1/2 C 4y2 D 1 (9.48)
is a common envelope of a family of curves. The point .1; 0/ is the node of

the considered family. The unit circumference completely lies in the definition
domain of family (9.47). Every curve of this family intersects twice the unit
circumference. Every contour of family (9.47) can be represented as the union of
two arcs. The first arc is considered from the point .1; 0/ to the point of tangency
to circumference (9.48) if the movement occurs in the positive direction. The
remaining part is considered to be the arc of the second family. If the polar angle
of the point of tangency of the concrete curve to circumference (9.48) is denoted by
# , then family (9.47) can be divided into two parts:
F.r; #; #0 / D 0; 0 6 # 6 # ; (9.49)
F.r; #; #0 / D 0; # 6 # 6 2: (9.50)
Problem 3 (Inverse Problem) Find the values of a regular solution of Eq. (9.2)
and its derivative with respect to the normal direction on the circumference x2 Cy2 D
1, i.e. find the functions
lim u.r; #/ D .#/; lim ur .r; #/ D .#/; (9.51)

r!1 r!1
if the family of plane curves (9.49) represents the characteristic curves corre-
sponding to invariants (9.19), while family (9.50) corresponds to the characteristic
invariants (9.20).
Let .1; '0 /; .'0 2k; k 2 Z/ be a point of the unit circumference. Then the
following curves of families (9.21) and (9.22) pass, respectively, through this point:
( 2 ) 1
.1 r cos #/ r sin # sin '0
1C 2 C C1 D 0;
r2 2r cos # C 1 r 2r cos # C 1 2.1 cos '0 /
(9.52)
where 0 6 # 6 #1 ,
( 2 ) 1
.1 r cos #/ r sin # sin '0
1C 2 1 D 0;
r2 2r cos # C 1 r 2r cos # C 1 2.1 cos '0 /
(9.53)
where #2 6 # 6 2.
Here #1 ; #2 are the polar angles of points of tangency of these curves with the
circumference (9.48), respectively.
The relation
ux .r; #; '0 / f1 .#; '0 /

D (9.54)
uy .r; #; '0 / C 1 g1 .#; '0 /
is fulfilled for the characteristic curve (9.52), and the relation
ux .r; #; '0 / f2 .#; '0 /

D (9.55)
uy .r; #; '0 / 1 g2 .#; '0 /
is fulfilled along curve (9.53), where
4hi r sin #.1 cos #/

fi .#; '0 / D 2 C .r2 cos 2# 2r cos # C 1/;
1 C h2i
2hi .r2 cos 2# 2r cos # C 1/

gi .#; '0 / D 2 2r sin #.1 cos #/;
1 C h2i
r sin # sin '0

hi D .1/i ; i D 1; 2:
r2 2r cos # C 1 2.1 cos '0 /
From (9.54) and (9.55) we conclude that the following equalities hold at the point
.1; '0 /:
ux .1; '0 / f0
D 10 ; (9.56)
uy .1; '0 / C 1 g1
ux .1; '0 / f0
D 20 ; (9.57)
uy .1; '0 / 1 g2
where
1
fi0 D .1/i sin '0 C cos '0 ;
2
1
g0i D .1/iC1 cos '0 C sin '0 :
2
Using relations (9.56) and (9.57) we obtain for the values of the derivatives ux and
uy at the point .1; '0 /
2f10 f20
ux .1; '0 / D ;
f20 g01 f10 g02
f10 g02 C f20 g01

uy .1; '0 / D :
f20 g01 f10 g02
Assuming that '0 takes all values on the circumference x2 C y2 D 1; 0 6 '0 6 2,
we have
1 2
ux .1; #/ D 2 cos2 # sin #;
2
5
uy .1; #/ D sin 2#:
4
Since the equalities
u# .1; #/ D ux sin # C uy cos #;
ur .1; #/ D ux cos # C uy sin #
are fulfilled, we obtain
1
u# .1; #/ D sin #; (9.58)
2
ur .1; #/ D 2 cos #: (9.59)
Finally, integrating the value of (9.58) on the circumference x2 C y2 D 1; 0 6 '0 6

2, we define also the function
1
u.1; #/ D cos # C c: (9.60)
2
If we consider the initial problem: find a function u.x; y/, which satisfies Eq. (9.2)
with initial data defined by functions (9.59), (9.60) on the unit circumference :
r D 1, x D r cos #, y D r sin #, then the definition domain of a solution
of this problem is constructed by the set of characteristic curves represented by
formulas (9.49), (9.50). By the structure of these characteristic curves we conclude
that for a > b the definition domain of a solution of the initial problem is the domain
of the plane .x; y/ which lies to the left of the straight line x D 1, with a gap given
the circumference (9.48).
Acknowledgements The authors were supported by the European Unions Seventh Framework
Programme (FP7/20072013) under grant agreements no. 317721, no. 318202.
References
1. R.A. Aleksandryan, The boundary problem of Dirichlet type for degenerative hyperbolic
equations. Reports of Symposium in Continuum Mechanics and related problems of Analysis.
Tbilisi (1971, in Russian)
2. G. Baghaturia, Nonlinear versions of hyperbolic problems for one quasi-linear equation of
mixed type. J. Math. Sci. 208, 621634 (2015)
3. L. Bers, Mathematical Aspects of Subsonic and Transonic Gas Dynamics. Surveys in Applied
Mathematics, vol. 3 (Wiley, New York; Chapman and Hall, London, 1958)
4. A.V. Bitsadze, Equations of Mixed Type (Pergamon Press, Oxford, 1964)
5. A.V. Bitsadze, Some Classes of Partial Differential Equations (Gordon and Breach Science,
New York, 1988)
6. R. Bitsadze, On one version of the initial-characteristic Darboux problem for one equation of
nonlinear oscillation. Reports of an Enlarged Session of the Seminar of I. Vekua Inst. Appl.
Math. 8, 46 (1993)
7. R.G. Bitsadze, M. Menteshashvili, On one nonlinear analogue of the Darboux problem. Proc.
A. Razmadze Math. Inst. 169, 921 (2015)
8. J.K. Gvazava, On Some Classes of Quasi-Linear Equations of Mixed Type (Metsniereba,
Tbilisi, 1981, in Russian)
9. J.K. Gvazava, Second-order nonlinear equations with complete characteristic systems and
characteristic problems for them. Trudy Tbiliss. Mat. Inst. Razmadze Akad. Nauk Gruzin. SSR
87, 4553 (1987, in Russian)
10. J.K. Gvazava, Nonlocal and initial problems for quasi-linear, non-strictly hyperbolic equations
with general solutions represented by superposition of arbitrary functions. Georgian Math. J.
10, 687707 (2003)
11. J.K. Gvazava, The mean value property for nonstrictly hyperbolic second order quasilinear
equations and the nonlocal problems. Proc. A. Razmadze Math. Inst. 135, 7996 (2004)
12. J.K. Gvazava, On one nonlinear version of the characteristic problem with a free support of
data. Proc. A. Razmadze Math. Inst. 140, 91107 (2006)
13. J. Gvazava, M. Menteshashvili, G. Baghaturia, Cauchy problem for a quasi-linear hyperbolic
equation with closed support of data. J. Math. Sci. 193, 364368 (2013)
14. J. Hadamard, Leons sur la propagation ds ondes et les quations de lhydrodinamique
(Hermann, Paris, 1903)
15. M. Klebanskaya, Some nonlinear versions of Darboux and Goursat problems for a hyperbolic
equation with parabolic degeneracy, in International Symposium on Differential Equations and
Mathematical Physics dedicated to the 90th Birthday Anniversary of Academician I. Vekua,
Tbilisi, 2125 June 1997
16. M.Z. Menteshashvili, On the Cauchy problem on the unit circumference for a degenerating
quasi-linear equation. Soobsh. Akad. Nauk Gruzii 148, 190193 (1993, in Russian)
17. G. Monge, Second mmoire sur le calcul integral de quelques quations aux differences
partilelles. Mem. R. Accad. Sci. Turin, annes 5, 79122 (17701773)
18. P.J. Olver, Applications of Lie Groups to Differential Equations. Graduate Texts in Mathemat-
ics, vol. 107 (Springer, Berlin, 1993)
19. I.Yu. Popov, Stratified flow in an electric field, the Schrdinger Equation, and the operator
extension theory model. Theor. Math. Phys. 103, 535542 (1995)
20. L.I. Sedov, Problems in the Plane in Hydrodynamics and Aerodynamics (Gastekhizdat,
Moscow-Leningrad, 1950)
21. S.L. Sobolev, On mixed problems for partial differential equations with two independent
variables. Dokl. Akad. Nauk SSSR, 122, 555558 (1958, in Russian)
22. F. Tricomi, Beispiel einer Strmung mit Durchgang durch die Schallgeschwindigkeit. Monat-
shefte Math. 58, 160171 (1954)
23. F. Tricomi, Lectures on Partial Equations, Moscow, 1957
24. N. Vakhania, On a boundary problem with the prescription on the whole boundary for the
hyperbolic system equivalent to the vibrating string equation. Dokl. Akad. Nauk SSSR, 116,
906909 (1957, in Russian)
25. L. Wolfersdorf, Zum Problem der Richtungsableitung fr die Tricomi-Gleichung. Math. Nachr.
25, 111127 (1963)
Chapter 10
Reduction of Some Semi-discrete Schemes
for an Evolutionary Equation to Two-Layer
Schemes and Estimates for the Approximate
Solution Error
Jemal Rogava, David Gulua, and Romeo Galdava
Abstract In the paper, using the perturbation algorithm, purely implicit three-
layer and four-layer semi-discrete schemes for an abstract evolutionary equation are
reduced to two-layer schemes. The solutions of these two-layer schemes are used
to construct an approximate solution of the initial problem. By using the associated
polynomials the estimates for the approximate solution error are proved.
10.1 Introduction
Various initial boundary value problems for evolutionary equations with partial
derivatives can be reduced to the Cauchy problem for an abstract parabolic equation.
One of the methods used to solve these problems is the semi-discretization method
(this is the method based on the discretization of a derivative with respect to a time
variable). The semi-discretization method for a parabolic equation is also known as
the Rothe method [1]. The advantage of this method is that it enables us to solve the
obtained system by the finite-difference method with subsequent discretization of
derivatives with respect to spatial derivatives or to apply other methods (including
analytical ones) which are easy to realize. Among these methods we want to mention
the projective-network, variational and finite-element methods (see, e.g., Marchuk
and Agoshkov [2], Mikhlin [3], Streng and Fiks [4]).
J. Rogava
I. Vekua Institute of Applied Mathematics, Iv. Javakhishvili Tbilisi State University, 2 University
St., Tbilisi 0186, Georgia
e-mail: jemal.rogava@tsu.ge
D. Gulua ()
Georgian Technical University, 77 M. Kostava St., Tbilisi 0175, Georgia
e-mail: d_gulua@gtu.ge
R. Galdava
Sokhumi State University, 9, Anna Politkovskaia St., Tbilisi 0186, Georgia
e-mail: romeogaldava@gmail.com

224 J. Rogava et al.
Questions connected with the construction and investigation of approximate

solution algorithms of evolutionary problems are considered, for example in the
well-known books by Godunov and Ryabenki [5], Marchuk [6], Richtmayer and
Morton [7], Samarski [8], Ianenko [9]. We also refer to the works by Crouzeix
[10], Crouzeix and Raviart [11], and Le Rouxe [12].
The questions connected with construction and research of algorithms of the
approximate solution of evolutionary problems, including research and realization
of multi-layer schemes for these problems are important. The main difficulty
which arises at realization of multi-layer schemes (especially for multidimensional
problems) consists in use of large random access memory, which increases in
proportion with growth of number of layers. One of the opportunities of overcoming
this problem is decomposition of multi-layer schemes. In [13] the perturbation algo-
rithm was considered for decomposition of three-layer scheme. The perturbation
algorithm is widely used when solving problems of mathematical physics (see, e.g.,
Marchuk et al. [14]).
We would note [15, 16] where a purely implicit three-layer semi-discrete scheme
for evolutionary equation is reduced to two-layer schemes and explicit estimates
for the approximate solution at rather general assumptions about data the tasks are
proved in Banach space. Furthermore, in these works, by reducing with the aid of the
perturbation algorithm the four-layer scheme to two-layer schemes we demonstrate
the generality of the algorithm when it is applied to difference schemes.
In the present paper, purely implicit multi-layer semi-discrete schemes are
considered for an approximate solution of the Cauchy problem for an evolutionary
equation with the self-adjoint positively defined operator in Hilbert space. Using
the perturbation algorithm, the considered scheme is reduced to two-layer schemes.
An approximate solution of the original problem is constructed by means of
the solutions of these schemes. Note that the first two-layer scheme gives an
approximate solution to an accuracy of first order, whereas the solution of each
subsequent scheme is the refinement of the preceding solution by one order.
In the present work, for an estimate of the error of the approximate solution,
we applied the approach offered in [17], where the stability of linear many-step
methods is investigated by means of the properties of the class of polynomials
of many variables (which are called associated polynomials). They are a natural
generalization of classical Chebyshev polynomials of second kind.
The present article consists of two paragraphs. In the first paragraph, creation
of algorithm and technique of proving estimate of approximate solution error is
applied for three-layer scheme. In the second paragraph, same technique is applied
for four-layer scheme. In our opinion, such presentation of the material makes it
possible to demonstrate the generality of the proposed method for decomposition
and investigation of multi-layer schemes.
In the first paragraph we also consider the results of the numerical experiments,
which visually confirm the theoretical estimates received in this paragraph.
10 Reduction of Some Semi-discrete Schemes 225
It should be noted that the algorithm proposed in this paper is close to the
methods considered in the works by Marchuk and Shadurov [18] and Pereyra
[19, 20]. The paper is a generalization of the results obtained in [13].
10.2 The Case of Three-Layer Scheme
10.2.1 Reducing a Purely Implicit Three-Layer Scheme

for an Evolutionary Problem to Two-Layer Schemes
Let us consider the following evolutionary problem in the Hilbert space H:
du.t/
C Au .t/ D 0; t 2 0; T ; (10.1)
dt
u .0/ D u0 ; (10.2)
where A is the self-adjoint positively defined operator in H with domain of definition

D.A/; u0 is a given vector from H; u.t/ is the sought function.
On 0; T we introduce a grid tk D k, k D 0; 1; : : : ; n, with the step D T=n.
Using the approximation of the first derivative
3
du u.tk / 2u.tk1 / C 12 u.tk2 /
D 2 C 2 Rk .; u/; Rk .; u/ 2 H;
dt tDtk
Eq. (10.1) can be represented at the point t D tk as

3
2 u.tk / 2u.tk1 / C 12 u.tk2 /
C Au .tk / D 2 Rk .; u/; k D 2; : : : ; n;

(10.3)
We write system (10.3) in the form

u.tk / u.tk1 / u.tk / 2u.tk1 / C u.tk2 /
C Au .tk / C D 2 Rk .; u/:
2 2
It is obvious that expression in brackets in case of 2 is u00 .tk1 / C 2 R1;k1 ,

R1;k 2 H:
By analogy with the obtained system let us consider in the Hilbert space H the
one-parametric family of equations

uk uk1 " uk 2uk1 C uk2
C Auk C D " 2 Rk ; Rk 2 H: (10.4)
2 2
Assume that uk is analytic with respect to ",

1
X . j/
uk D " j uk : (10.5)
jD0
Substituting (10.5) into (10.4) and equating the members of identical powers ", we
obtain the following system of equations
.0/ .0/
uk uk1 .0/ .0/
C Auk D 0; u0 D u0 ; k D 1; : : : ; n; (10.6)

.1/ .1/ .0/
uk uk1 .1/ 1 2 uk2
C Auk D ; k D 2; : : : ; n; (10.7)
2 2
.2/ .2/ .1/
uk uk1 .2/ 1 2 uk2
C Auk D C Rk ;
2 2
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::
where uk1 D uk uk1 .

We introduce the notation
.0/ .1/
vk D uk C uk ; k D 2; : : : ; n: (10.8)
Let the vector vk be an approximate value of the exact solution of problem (10.1)
(10.2) for t D tk , u.tk / vk .
.1/
Note that in scheme (10.7) the starting vector u1 is defined from the equality
.0/ .1/ .0/
v1 D u1 C u1 , where u1 is found by scheme (10.6), and v1 is an approximate
value of u.t1 / with accuracy of O. 2 /. According to this assumption, we can write
kv1 u.t1 /k D O. 2 /:
Then, taking into account that for the solution of the problem (10.1), (10.2) the
formula (see, e.g., Kato [21]) u.t/ D exp.tA/u0 is true, we can write
v1 D .I A/ u0 : (10.9)
.0/ .1/
From representation v1 D u1 C u1 , taking into account a formula (10.9)
.0/ .1/
and equality u1 D Su0 ; where S D .I C A/1 ; for a start vector u1 of the
scheme (10.7), we can take
.0/
.1/ v1 u1 1 1
u1 D D .I A S/ u0 D ..I S/ A/ u0

1
D .AS A/ u0 D A.S I/u0 D A2 Su0 : (10.10)

10.2.2 Estimate of Residual for Solution Obtained

by Perturbation Algorithm
Let estimate residual of purely implicit three-layer scheme for solutions of

scheme (10.6)(10.8).
Multiply Eq. (10.7) by and add result with (10.6). We get that vk is a solution
of system of equations:
.0/
vk vk1 2 uk2
C Avk D ; k D 2; : : : ; n: (10.11)
2 2
Rewrite this system in the form
vk vk1 2 vk2
C Avk C De
Rk ./; (10.12)
2 2
where k > 3,
.0/
e 2 vk2 2 uk2
Rk ./ D :
2 2 2 2
It is obvious that (10.12) can be represented in the following form:
3
2 vk 2vk1 C 12 vk2
C Avk D e
Rk ./; k D 3; : : : ; n: (10.13)

Therefore eRk ./ is a residual of purely implicit three-layer scheme for solutions of

scheme (10.6)(10.8) (see (10.3)).
It is obviously the representation
.0/ .1/
e 2 vk2 2 uk2 2 2 uk2
Rk ./ D 2
2
D : (10.14)
2 2 2 2
Note that representation (10.14) is true for k > 2.
We estimate the difference relation (10.14).
.0/
From (10.6) it is obviously uk D Sk u0 ; where S D .I C A/1 . Then
.0/ .0/ .0/ .0/

2 uk2 D uk 2uk1 C uk2
D .Sk 2Sk1 C Sk2 /u0 D 2 A2 Sk u0 : (10.15)
Taking into account (10.15) from (10.7) we have:

.0/
!
.1/ .1/ 2 uk2
uk D S uk1
2 2
.1/ .1/
D Suk1 A2 SkC1 u0 D Sk1 u1 .k 1/A2 SkC1 u0 : (10.16)
2 2
From here it follows:
.1/ .1/ .1/ .1/

3 .1/
D 2 A3 SkC1 .k 3/A4 SkC1 u0 C 2 A2 Sk1 u1 : (10.17)
2
.1/
From here, taking into account that u1 uniformly bounded by (see (10.10)),
for enough smooth initial data we receive the estimation

2 u.1/
k2
6 c; c D const. > 0: (10.18)
2
Therefore, taking into account (10.18), from (10.14) for residual e

Rk ./, we have
the estimation

e
Rk ./ 6 c 2 ; c D const. > 0: (10.19)
10.2.3 A Priori Estimate for the Approximate Solution Error
From (10.3) and (10.13), for error zk D u.tk / vk we have:

3
z
2 k
2zk1 C 12 zk2
C Azk D rk ./; k D 3; : : : ; n; (10.20)

where rk ./ D 2 Rk .; u/ C e
Rk ./ :
Remark 10.2.1 Taking into account (10.19) we conclude that if the solution of
problem (10.1)(10.2) is enough smooth function, then krk ./k D O. 2 /.
The following theorem is valid.
Theorem 10.2.2 Let A be a self-adjoint positively defined operator in H. Then
1 3 X k
kzkC1 k 6 kz1 k C kz2 k C kriC1 ./k; (10.21)
2 2 iD2
where k D 2; : : : ; n 1:
Proof From (10.20), we have
zkC1 D L1 zk L2 zk1 C gkC1 ; k D 2; : : : ; n 1:
where
4 1 2
L1 D L; L2 D L; gkC1 D LrkC1 ./;
3 3 3
1
2
L D I C A :
3
Then (see [17], ch.1. 3):
X
k
zkC1 D Uk1 z2 L2 Uk2 z1 C Uki giC1 ; (10.22)
iD2
where Uk .L1 ; L2 / are the operator polynomials, which are defined by the following
recurrence relation:
Uk .L1 ; L2 / D L1 Uk1 C L2 Uk2 ;

U0 D I; U1 D 0:
If in equality (10.22) we take the norms, we will have
2 X k
kzkC1 k 6 kUk1 kkz2 k C kL2 kkUk2 kkz1 k C kLk kUki kkriC1 ./k:
3 iD2
(10.23)
Be using the well-known fact that the norm of the operator polynomial is equal to
C norm of corresponding scalar polynomial on the spectrum of this operator when
the argument is self-adjoint bounded operator (see, e.g., [22], ch. IX, 5), we have

1

kUk .L1 ; L2 /k D max Uk x; x : (10.24)
x2Sp.L1 / 4
It is obvious that Sp.L1 / 0; 43 : Then we have

1 3
max Uk x; x 6 :
x20; 43 4 2
Taking into account this estimate from (10.24) it follows
3
kUk .L1 ; L2 /k 6 :
2
Substituting this estimate in (10.23), we obtain the sought estimate (10.21). t
u
Note that since the representation (10.14) is true only if k > 2, proof of the
estimate (10.21) does not apply in the case of k D 1, i.e. for a vector v2 . So it is
necessary to separately estimate the error u.t2 / v2 .
From (10.11) follows
v2 D Sv1 C S'1 ; (10.25)
.0/
2 u
where '1 D 2 20 : Substituting (10.9) and (10.15) in (10.25) and taking into
account the expansion S D I A C 2 A2 S; we have
v2 D .I A C 2 A2 S/ .I A/ u0 C e
r1 ./u0 ; r1 ./u0 k D O. 2 /:
ke
It is obvious that from here follows
v2 D .I 2A/ u0 C e
r2 ./u0 ; r./u0 k D O. 2 /:
ke
Therefore for error u.t2 / v2 we have
ku.t2 / v2 k D k exp.2A/u0 v2 k D O. 2 /: (10.26)
Remark 10.2.3 Taking into account (10.9), (10.26) and Remark 10.2.1 from
inequality (10.21), it follows
ku.tk / vk k D O. 2 /; k D 1; : : : ; n:
It is obvious we imply that the solution of problem (10.1), (10.2) is a smooth enough
function.
10.2.4 Results of Numerical Experiments
Now we will consider results of numerical experiments of solving some model

problems by the algorithm studied in this chapter.
For the beginning we consider the following simple problem
du.t/
C u .t/ D f .t/; t 2 0; 1 ;
dt
u .0/ D u0 :
We will look for approximate solution vk (k D 1; : : : ; 10) of this problem by the

algorithm (10.6)(10.8). Errors
D max ju.tk / vk j
k
for some test solutions u.t/ are given below. Let s be the order of the considered
.0/
correction (s D 0 corresponds to the solution uk and s D 1 corresponds to the
.0/ .1/
specified solution vk D uk C uk ).
1. u.t/ D 5t2 4t
D 2;
sD0W D 0:1045
sD1W D 0:0051:
D 0:001;
sD0W D 0:2373
sD1W D 0:0062:
2. u.t/ D 3 sin.2t/
D 2;
sD0W D 0:1075
sD1W D 0:0032:
D 0:001;
sD0W D 0:2136
sD1W D 0:0017:
3. u.t/ D e3t
D 2;
sD0W D 0:8902
sD1W D 0:0473:
D 0:001;
sD0W D 1:4543
sD1W D 0:0292:
4. u.t/ D 10sin.t/
D 2;
sD0W D 0:0629
sD1W D 0:0021:
D 0:001;
sD0W D 0:1162
sD1W D 0:0014:
5. u.t/ D sin.10t/
D 2;
sD0W D 0:3379
sD1W D 0:0389:
D 0:001;
sD0W D 0:4597
sD1W D 0:0507:
Now consider the following problem
@u.x; t/ @2 u .x; t/
D C f .x; t/; .x; t/ 2a; b0; T;
@t @x2
u.x; 0/ D '.x/;
u.a; t/ D 0 .t/; u.b; t/ D 1 .t/:
We will also look for approximate solution vi;k of this problem by the algo-
rithm (10.6)(10.8). Errors
D max ju.xi ; tk / vi;k j

i;k
for some test solutions u.x; t/ (s -order of the considered correction, n-number of
steps on x; m -number of steps on t) are given below.
1. u.x; t/ D 5x2 t x2 t2 ; D 0:1; T D 2:1; a D 1; b D 5; m D 10; n D 20
sD0W D 0:0636
sD1W D 0:00043:
2. u.x; t/ D 3 sin.2xt2 /; D 2; T D 1:2; a D 1:2; b D 2:4; m D 10; n D 30
sD0W D 0:2851
sD1W D 0:0475:
3. u.x; t/ D ext ; D 2; T D 1:2; a D 1:2; b D 2:4; m D 10; n D 30
sD0W D 0:1331
sD1W D 0:0102:
As one would expect, the numerical experiments showed that the second step of
.0/ .1/
correction (s D 1) reduces an error by one order, that is ku.tk / .uk C uk /k D
2 .0/
O. /; while ku.tk / uk k D O./:
.2/
The numerical experiments also showed that the further correction uk
(see (10.6)(10.8)) cannot practically reduce an approximate solution error. The
.2/
reason is that the equation for looking for uk contains unknown Rk and finding
of this unknown Rk is equivalent to definition of the solution of an initial problem.
This is the reason why the algorithm is limited to two Eqs. (10.6)(10.7).
10.3 The Case of Four-Layer Scheme
10.3.1 Reducing a Purely Implicit Four-Layer Scheme

for an Evolutionary Problem to Two-Layer Schemes
On 0; T we introduce a grid tk D k, k D 0; 1; : : : ; n, with the step D T=n. Using

the approximation of the first derivative
11
du 6
u.tk / 3u.tk1 / C 32 u.tk2 / 13 u.tk3 /
D C 3 Rk .; u/; Rk .; u/ 2 H:
dt tDtk
Equation (10.1) can be represented at the point t D tk as

11
6
u.tk / 3u.tk1 / C 32 u.tk2 / 13 u.tk3 /
C Au .tk / D 3 Rk .; u/; k D 3; : : : ; n:

(10.27)
We write system (10.27) in the form
u.tk / u.tk1 /
C Au .tk /

u.tk / 2u.tk1 / C u.tk2 /
C
2 2

2 u.tk / 3u.tk1 / C 3u.tk2 / u.tk3 /
C D 3 Rk .; u/: (10.28)
3 3
It is obvious that expression in brackets in case of 2 is u00 .tk1 / C 2 R1;k1 and

2
expression in brackets in case of 3 is u000 .tk / C R2;k , R1;k ; R2;k 2 H.
By analogy with the obtained system let us consider in the Hilbert space H the
one-parametric family of equations

uk uk1 " uk 2uk1 C uk2
C Auk C
2 2

"2 uk 3uk1 C 3uk2 uk3
C D " 3 Rk ; Rk 2 H: (10.29)
3 3
Assume that uk is analytic with respect to ",

1
X . j/
uk D " j uk : (10.30)
jD0
Substituting (10.30) into (10.29) and equating the members of identical powers ",
we obtain the following system of equations
.0/ .0/
uk uk1 .0/ .0/
C Auk D 0; u0 D u0 ; k D 1; : : : ; n; (10.31)

.1/ .1/ .0/
uk uk1 .1/ 1 2 uk2
C Auk D ; k D 2; : : : ; n; (10.32)
2 2
.2/ .2/ .1/ .0/
uk uk1 .2/ 1 2 uk2 1 3 uk3
C Auk D ; k D 3; : : : ; n;
2 2 3 3
(10.33)
.3/ .3/ .2/ .1/
uk uk1 .3/ 1 2 uk2 1 3 uk3
C Auk D C Rk ;
2 2 3 3
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::
where uk1 D uk uk1 .

We introduce the notation

.0/ .1/ .2/
vk D uk C uk C 2 uk ; k D 3; : : : ; n: (10.34)
Let the vector vk be an approximate value of the exact solution of problem (10.1)
(10.2) for t D tk , u.tk / vk .
.1/
Note that in scheme (10.32) the starting vector u1 is defined from the equality
.0/ .1/ .0/
v1 D u1 C u1 , where u1 is found by scheme (10.31), and v1 is an approximate
.2/
value of u.t1 / with accuracy of O. 3 /. Similarly, the starting vector u2 is defined
.0/ .1/ .2/ .0/ .1/
from the equality v2 D u2 C u2 C 2 u2 , where u2 and u2 are found by
scheme (10.31), (10.32), respectively, and v2 is an approximate value of u.t2 / with
accuracy of O. 3 /.
According to this assumption, we can write
kv1 u.t1 /k D O. 3 /; kv2 u.t2 /k D O. 3 /:
Then, taking into account that for the solution of the problem (10.1), (10.2) the
formula (see, e.g., Kato [21]) u.t/ D exp.tA/u0 is true, we can write

2 A2
v1 D I A C u0 ; (10.35)
2

v2 D I 2A C 2 2 A2 u0 : (10.36)
.0/ .1/
From representation v1 D u1 C u1 , taking into account a formula (10.35) and
.0/ .1/
equality u1 D Su0 ; for the start vector u1 we have
.0/
.1/ v1 u1 1 2 A2 1 2
u1 D D I A C S u0 D .I S/ A C A2 u0
2 2
2 2
1 1
D AS A C A2 u0 D A.S I/ C A2 u0
2 2
2
1 2
D 2 A2 S C A2 u0 D A A2 S u0 : (10.37)
2 2
.0/ .1/ .2/

From equality v2 D u2 C u2 C 2 u2 , taking into account the formulas (10.16)
and (10.37), it follows
.2/ 1 .0/ .1/

1 .0/

.1/ 2 3
u2 D v2 u 2 u 2 D v2 u 2 Su 1 A S u0
2 2 2

1 .0/ 1 2 1
D 2 v2 u2 2 A S A2 S 2 A2 S 3 u 0 : (10.38)
2 2
.0/ .0/
Transform the difference v2 u2 . Taking into account (10.36) and equation u2 D
S2 u0 , we have
.0/
v2 u2 D .I S2 / 2A C 2 2 A2 u0 D AS.I C S/ 2A C 2 2 A2 u0

D A.I C S/ 2AS1 C 2 2 A2 S1 Su0

D A.I C S/ 2A I 2 A2 Su0

D A.S I/ C 2 3 A3 Su0

D 2 3 A3 2 A2 S Su0 : (10.39)
.2/
Substituting (10.39) into (10.38), for the start vector u2 ; we have

.2/ 1 1 1
u2 D A2 2A I C S2 Su0 D A2 .4A C .S I/.S C I// Su0
2 2 2
3
D A .4I S.S C I// Su0 : (10.40)
2
10.3.2 Estimate of Residual for Solution Obtained

by Perturbation Algorithm
Let us estimate residual of purely implicit four-layer scheme for solutions of

schemes (10.31)(10.34).
Multiply Eqs. (10.32) and (10.33) by and 2 , respectively, and add result
to (10.31). We get that vk is a solution of system of equations:
.0/ .1/ .0/

vk vk1 2 uk2 2 2 uk2 2 3 uk3
C Avk D ; k D 3; : : : ; n:
2 2 2 2 3 3
(10.41)
Rewrite this system in the form
vk vk1 2 vk2 2 3 vk3

C Avk C C De
Rk ./; (10.42)
2 2 3 3
where k > 5,
.0/ .1/ .0/
e 2 vk2 2 3 vk3 2 uk2 2 2 uk2 2 3 uk3
Rk ./ D 2
C :
2 3 3 2 2 2 2 3 3
It is obvious that (10.42) can be represented in the following form:

11
6 vk 3vk1 C 32 vk2 13 vk3
C Avk D e
Rk ./; k D 5; : : : ; n: (10.43)

Therefore e Rk ./ is a residual of purely implicit four-layer scheme for solutions of

scheme (10.31)(10.34) (see (10.27)).
It is obviously the representation
.0/ .1/ .2/

!
e 2 uk2 2 2 uk2 3 2 uk2
Rk ./ D C C
2 2 2 2 2 2
.0/ .1/ .2/
!
2 3 uk3 3 3 uk3 4 3 uk3
C 3
C 3
C
3 3 3 3
.0/ .1/ .0/
2 uk2 2 2 uk2 2 3 uk3

2 2 2 2 3 3
.2/ .1/ .2/
!
3 1 2 uk2 1 3 uk3 3 uk3
D 2
C 3
C : (10.44)
2 3 3 3
Note that representation (10.44) is true for k > 4.

We estimate the difference relations included in eRk ./.
Taking into account (10.16), we have
.1/ .1/ .1/ .1/


3 .1/
D 2 A3 SkC1 .k 3/A4 SkC1 u0 C 2 A2 Sk1 u1 : (10.45)
2
.1/ .1/ .1/

As 3 uk3 D 2 uk2 2 uk3 , taking into account (10.45) we have

.1/ k 3 5 kC1 1 4 k 3 k1 .1/
3 uk3 D 3 4 kC1
A S A S C A S u0 C A S u1 :
2 2
.1/
From here, taking into account that u1 is uniformly bounded by (see (10.37)), for
smooth enough initial data we obtain the estimation

3 u.1/
k3
6 c; c D const. > 0: (10.46)
3
.2/
2 uk2
Estimate the difference relation 2
: From (10.33) we have
.2/ .2/
X
k2
uk D Sk2 u2 C Si wkiC1 ; (10.47)
iD1
where
.1/ .0/
1 2 uk2 1 3 uk3
wk D : (10.48)
2 2 3 3
Taking into account (10.47) we have
.2/ .2/ .2/ .2/ .2/

2 uk2 D uk 2uk1 C uk2 D Sk2 2Sk3 C Sk4 u2

C S.wk wk1 / C .S2 S/wk1 C .S3 2S2 C S/wk2 C

C Sk2 2Sk3 C Sk4 w3
.2/
D 2 A2 Sk2 u2 C S.wk wk1 / 2 AS2 wk1

C 3 A2 S3 wk2 C C A2 Sk2 w3 (10.49)
Taking into account (10.15) we have
.0/ .0/ .0/

3 uk3 D 2 uk2 2 uk3 D 2 A2 Sk u0 2 A2 Sk1 u0 D 3 A3 Sk u0 :
(10.50)
From (10.48), taking into account (10.15), it follows
1 2 .1/ 2 .1/
1 3 .0/ 3 .0/

wk wk1 D u u u u
2 2 k2 k3
3 3 k3 k4
.1/ .1/
3 uk3 1 .0/ 3 uk3 1
D 3
3 4 uk4 D 3 2 A2 2 .Sk2 u0 /
2 3 2 3 3
.1/ .1/
3 uk3 1 .0/ 3 uk3 4 k
D 3
3 2 A2 2 uk2 D A S u0 :
2 3 2 3 3
(10.51)
from (10.49), taking into account (10.51), we have
.2/ .1/
2 uk2 .2/ 1 3 uk3 1 4 kC1
2
D A2 Sk2 u2 S A S u0 AS2 wk1
2 3 3
2 3

CA S wk2 C Swk3 C C Sk5 w3 : (10.52)
Note that from (10.48), taking into account (10.45) and (10.50), follows bounded-
ness wk ,
kwk k 6 c; c D const. > 0: (10.53)
.2/
From (10.52), taking into account that, u2 is uniformly bounded by
(see (10.40)), and taking into account (10.46) and (10.53), for smooth enough
initial data we obtain the estimation

2 u.2/
k2
6 c; c D const. > 0: (10.54)
2
Furthermore, since
.2/ .2/ .2/
3 uk3 D 2 uk2 2 uk3 ;
then taking into account (10.54) we have

3 u.2/
k3
6 c; c D const. > 0: (10.55)
3
Finally taking into account (10.46), (10.54), and (10.55), from (10.44) for
residual e
Rk ./, follows the estimate

e
Rk ./ 6 c 3 ; c D const. > 0: (10.56)
10.3.3 A Priori Estimate for the Approximate Solution Error
From (10.27) and (10.43), for error zk D u.tk / vk we have:

11
6 zk 3zk1 C 32 zk2 13 zk3
C Azk D rk ./; k D 5; : : : ; n; (10.57)

where rk ./ D 3 Rk .; u/ C e Rk ./ :
Remark 10.3.1 Taking into account (10.56) we conclude that if the solution of
problem (10.1)(10.2) is smooth enough function, then krk ./k D O. 3 /.
The following theorem is valid.
Theorem 10.3.2 Let A be a self-adjoint positively defined operator in H. Then

!
X
k
kzkC2 k 6 c kz2 k C kz3 k C kz4 k C kriC2 ./k ; c D const: > 0:
iD3
(10.58)
where k D 3; : : : ; n 2:
Proof From (10.57), we have
zkC1 D L1 zk C L2 zk1 C L3 zk2 C gkC1 : (10.59)
where
18 9 2 6
L1 D L; L2 D L; L3 D L; gkC1 D LrkC1 ./;
11 11 11 11
1
6
L D I C A :
11
From here, by induction, we have (see [17, p. 59])
X
k
zkC2 D Uk2 z4 C .L2 Uk3 C L3 Uk4 /z3 C L3 Uk3 z2 C Uki giC2 ; (10.60)
iD3
where Uk .L1 ; L2 ; L3 / are the operator polynomials, which are defined by the
following recurrence relation:
Uk .L1 ; L2 ; L3 / D L1 Uk1 C L2 Uk2 C L3 Uk3 ;

U0 D I; U1 D U2 D 0:
Consider the characteristic equation associated with the difference equa-

tion (10.59)
18 2 9 2
Q1 ./ D 3 x C x x D 0; (10.61)
11 11 11
where x 2 Sp.L/ 0; 1.

If the real root of Eq. (10.61) is in the unit circle and complex roots are inside the
unit circle then operator polynomials Uk .L1 ; L2 ; L3 / are bounded in the aggregate
(see [17], I, 3). Note that proving the above-mentioned result uses the well-known
fact that the norm of the operator polynomial is equal to C norm of corresponding
scalar polynomial on the spectrum of this operator when the argument is self-adjoint
bounded operator (see, e.g., [22], ch. IX, 5).
We will show that there exists > 1 such that for any x 2 0; 1/, the real root of
2
a characteristic equation (10.61) belongs to an interval 11 x; 1:
11
There exists 2 .1; 9 / such that

2
Q1 x < 0
11
for any x 2 .0; 1.

As
Q1 .1/ D 1 x > 0; x 2 0; 1/;
one of the roots of the characteristic equation (10.61) belongs to an interval

2
11 x; 1. Furthermore, we will show that remaining roots of this equation are
complex.
Assume that 1 is a real root of the characteristic equation (10.61) and 2 and 3
are complex roots, 3 D 2 . By theorem of Vieta
2
j1 2 2 j D x;
11
from here
2x 1
j2 j2 D < < 1:
111
Obviously, when x D 1, a real root of the characteristic equation (10.61) is
1 D 1, and complex roots are 2 and 3 then
2
j2 j2 D j3 j2 D :
11
We will show that the roots 2 and 3 of the characteristic equation (10.61) are
complex.
It is known that if discriminant
2
q p3
D D 108 C ;
4 27
where
2a3 ab a2
qD C c; p D b ;
27 3 3
of cubic equation
3 C a2 C b C c D 0;
is negative, then the cubic equation has complex roots.

In our case we have:

9 12
pD x 1 x ;
11 11
24 33 3 2 33 2 2
qD x C x x:
113 112 11
Obviously, for any x 2 0; 1 inequalities:
9 9 9
p.x/ D x.1 x/ 2 x2 2 x2 ;
11 11 11
5
q.x/ 6 x:
11 24
are true.
From here:
93 6
p3 .x/ x ;
116
25
q2 .x/ x2 ; x 2 0; 1:
112 28
According to these inequalities we have

q2 p3 25 93
D D 108 C 6 108 2 10
x2 x6 < 0:
4 27 11 2 27 116
So one of the roots of the characteristic equation (10.61) is real and two other
roots are complex. Thus the real root is on the unit circle and complex roots are
inside the unit circle. From this follows that operators Uk .L1 ; L2 ; L3 / are bounded
in the aggregate (see [17], I, 3). If we consider this fact and we will pass to norms
in (10.60) then we will have the estimate (10.3.2). t
u
10.3.4 Error Estimation for Vectors v3; v4 and Consequences

of Theorem 10.3.2
Note that since the representation (10.44) is true only if k > 4, proof of the
estimate (10.58) does not apply in the case of k D 1; 2, i.e. for a vector v3 ; v4 .
So it is necessary to separately estimate the error for vectors v3 ; v4 .
At first we estimate the error u.t3 / v3 .
From (10.41) follows
v3 D Sv2 C S'1 ; (10.62)
where
.0/ .1/ .0/
2 u 1 2 2 u 1 2 3 u 0
'1 D 2
2
: (10.63)
2 2 3 3
Taking into account (10.15), (10.45), (10.50), and (10.37) we have:
.0/
2 u 1 D 2 A2 S 3 u 0 ;

.1/ .1/
2 u 1 D 2 A3 S4 u0 C 2 A2 S2 u1 D 2 A3 S4 C A4 S2 A4 S3 u0 ;
2
.0/
3 u0 D 3 A3 S3 u0 :
Substituting these values in (10.63) we have

2 2 3 1 4 2
'1 D A2 S 3 C A3 S 3 A3 S 4 A S A4 S 3 u0 : (10.64)
2 3 2 2 2
Substituting (10.36) in (10.64) and (10.62) we have

2
v3 D S I 2A C 2 2 A2 A2 S3 u0 C e
r1 ./u0 ; r 1 ./u0 k D O. 3 /:
ke
2
As S D I AS, that
2 2 3 2
A S u 0 D A2 u 0 C e
r 2 ./u0 ; r 2 ./u0 k D O. 3 /:
ke
2 2
Therefore for v3 we have

3 2 2
v3 D S I 2A C A u0 C e
r3 ./u0 ; r3 ./u0 k D O. 3 /:
ke (10.65)
2
Taking into account the expansion
S D I A C 2 A2 3 A3 S;
from (10.65) follows

2 2
3 2 2
v3 D I A C A I 2A C A u0 C e
r4 ./u0 ; r4 ./u0 k D O. 3 /:
ke
2
Obviously, from here it follows

9 2 2
v3 D I 3A C A u0 C e
r5 ./u0 ; r 5 ./u0 k D O. 3 /:
ke
2
ku.t3 / v3 k D k exp.3A/u0 v3 k D O. 3 /: (10.66)
We will pass to an error assessment u.t4 / v4 :

From (10.41) it follows
v4 D Sv3 C S'2 ; (10.67)
where
.0/ .1/ .0/
2 u 2 2 2 u 2 2 3 u 1
'2 D 2
2
: (10.68)
2 2 3 3
Taking into account (10.15), (10.45), (10.50), and (10.37) we have:
.0/
2 u 2 D 2 A2 S 4 u 0 ;

.1/ 3 4 5 3 2 3 1 2
2 u 2 2 3 5
D A S A S u0 C A S 2
A A S u0 ;
4 2
.0/
3 u1 D 3 A3 S4 u0 :
Substituting these values in (10.68) we have

'2 D A2 S 4 u 0 C e
r6 ./u0 ; r 6 ./u0 k D O. 2 /:
ke
2
Substituting values of v3 and '2 in (10.67) we have

9 2 2 2 2 4
v4 D S I 3A C A A S u0 C e
r7 ./u0 ; r 7 ./u0 k D O. 3 /:
ke
2 2
Furthermore, arguing as in the previous case, we obtain

v4 D I 4A C 8 2 A2 u0 C e
r8 ./u0 ; r8 ./u0 k D O. 3 /:
ke
ku.t4 / v4 k D k exp.4A/ v4 k D O. 3 /: (10.69)

Theorem 10.3.3 Let A be a self-adjoint positively defined operator in H and let the
solution of problem (10.1)(10.2) be a smooth enough function. If ku.tk / vk k D
O. 3 /; k D 1; 2, then the estimate is true
ku.tk / vk k D O. 3 /; k D 3; : : : ; n: (10.70)
At this point, for k D 3; 4 we already proved the estimate (10.70) provided

that the solution of problem (10.1)(10.2) is smooth enough function and ku.tk /
vk k D O. 3 /; k D 1; 2: The estimate (10.70) for k > 4 follows from a priori
estimate (10.58).
Acknowledgements J. Rogava and D. Gulua were supported by the European Unions Seventh
Framework Programme (FP7/20072013) under grant agreement no. 317721.
References
1. E. Rothe, ber die Wrmeleitungsgleichung mit nichtkonstanten Koeffizienten im rumlichen

Falle. Math. Ann. 104(1), 340354 (1931)
2. G.I. Marchuk, V.I. Agoshkov, Introduction to Projection-Grid Methods (Nauka, Moscow,
1981)
3. S.G. Mikhlin, Numerical Realization of Variational Methods. With an appendix by T.N.
Smirnova (Izdat. Nauka, Moscow, 1966)
4. G. Streng, J. Fiks, in Theory of the Finite Element Method, ed. by G.I. Marchuk. Translated
from the English by V.I. Agoshkov, V.A. Vasilenko, and V.V. Shadurov (Izdat. Mir, Moscow,
1977)
5. S.K. Godunov, V.S. Ryabenki, Difference Schemes (Nauka, Moscow, 1973)
6. G.I. Marchuk, Methods of Computational Mathematics (Nauka, Moscow, 1977)
7. R. Richtmayer, K. Morton, Difference Methods of Solution of Initial-Value Problems (Mir,
Moscow, 1972)
8. A.A. Samarski, Theory of Difference Schemes (Nauka, Moscow, 1977)
9. N.N. Ianenko, A Method of Fractional Steps of Solution of Multi-Dimensional Problems of
Mathematical Physics (Nauka, Novosibirsk, 1967)
10. M. Crouzeix, Une methode multipas implicite-explicite pour lapproximation des quations
dvolution paraboliques. Numer. Math. 35(3), 257276 (1980)
11. M. Crouzeix, P.-A. Raviart, Approximation des quations dvolution linaires par des
mthodes pas multiples. C. R. Acad. Sci. Paris 283, A367A370 (1976)
12. M.-N. Le Roux, Semidiscretization in time for parabolic problems. Math. Comput. 33(147),
919931 (1979)
13. V.I. Agoshkov, D.V. Gulua, A perturbation algorithm for realization of finite-dimensional
realization problems. Comput. Math. Dept. USSR Acad. Sci., Moscow, 35 pp. (1990). Preprint
253
14. G.I. Marchuk, V.I. Agoshkov, V.P. Shutyaev, Adjoint Equations and Perturbation Methods in
Nonlinear Problems of Mathematical Physics (VO Nauka, Moscow, 1993)
15. J.L. Rogava, D.V. Gulua, Perturbation algorithm for implementing a finite-difference approxi-
mation to an abstract evolutionary problem and explicit error estimation of its solution. Dokl.
Math. 89(3), 335337 (2014)
16. J.L. Rogava, D.V. Gulua, Reduction of a three-laver semi-discrete scheme for an abstract
parabolic equation to two-layer schemes. Explicit estimates for the approximate solution error.
J. Math. Sci. 206(4), 424444 (2015)
17. J.L. Rogava, Semidiscrete Schemes for Operator Differential Equations (Technical University
Press, Tbilisi, 1995)
18. G.I. Marchuk, V.V. Shadurov, Difference Methods and Their Extrapolations (Springer, Berlin,
1983)
19. V. Pereyra, On improving an approximate solution of a functional equation by deferred
corrections. Numer. Math. 8(3), 376391 (1966)
20. V. Pereyra, Accelerating the convergence of discretization algorithms. SIAM J. Numer. Anal.
4, 508533 (1967)
21. T. Kato, in Perturbation Theory for Linear Operators, ed. by V.P. Maslov. Translated from the
English by G.A. Voropaeva, A.M. Stepin and I.A. Sismarev (Izdat. Mir, Moscow, 1972)
22. L.V. Kantorovich, G.P. Akilov, Functional Analysis (VO Nauka, Moscow, 1977)
Chapter 11
Hilberts Fourth Problem and Projectively Flat
Finsler Metrics
Xinyue Cheng, Xiaoyu Ma, Yuling Shen, and Shuhua Liu
Abstract This survey article mainly introduces some important research progresses
on the smooth solutions of Hilberts fourth problem in the regular case, which we
call projectively flat Finsler metrics. We characterize and classify projectively flat
Finsler metrics of constant flag curvature. We also discuss and classify projectively
flat Finsler metrics with isotropic S-curvature. In particular, we study and charac-
terize projectively flat Randers metrics, square metrics, .; /-metrics, and their
curvature properties.
11.1 Hilberts Fourth Problem
In 1900, David Hilbert gave his famous speech about mathematical problems [18].
Among Hilberts 23 problems, the fourth problem is Problem of the straight line
as the shortest distance between two points. Concretely, Hilberts fourth problem
asks to construct and study all metrics on an open subset in Rn such that straight
line segments are the shortest paths joining any two points. Busemann called these
metrics projective metrics and proposed an integral-geometric construction which
is inspiringly simple (see [8]). Further, Pogorelov proved that every projective
metric that is sufficiently smooth is of the metric form defined by Busemanns
construction [21]. However, lvarez Paiva pointed out in [1] that a large number of
projective metrics cannot be constructed by taking positive measures on the space
of hyperplanes and it is not clear how to construct all quasi-positive measures and
Busemanns construction, by itself, does not shed much light on the properties of
projective metric spaces. On the other hand, all of the researches in [8] and [1, 21]
are just for the absolutely homogeneous metrics, which means that the distance
functions determined by the metrics are symmetric. Therefore, it is necessary to
complement the results of Busemann and Pogorelov with other characterizations of
X. Cheng () X. Ma Y. Shen S. Liu

School of Mathematics and Statistics, Chongqing University of Technology, Chongqing 400054,
Peoples Republic of China
e-mail: chengxy@cqut.edu.cn; 1066719732@qq.com; sylyiyi@163.com; 12612954@qq.com

248 X. Cheng et al.
projective metric spaces that either yield new explicit examples or shed more light
on the general properties of these spaces.
Because any intrinsic quasimetric induces a Finsler metric and projective metrics
that are sufficiently smooth are Finsler metrics, Hilberts fourth problem in the
regular case is to study the Finsler metrics with straight lines as their geodesics. We
call the smooth solutions of Hilberts fourth problem in the regular case projectively
flat Finsler metrics.
It is easy to see that a Finsler metric F D F.x; y/ on an open subset U Rn is
projectively flat if and only if the spray coefficients are in the following form
Gi D Pyi ;
where P D P.x; y/ satisfies P.x; y/ D P.x; y/; 8 > 0 and Gi are defined by
1 n o
Gi D gil F 2 xm yl ym F 2 xl : (11.1)
4
We call Gi the geodesic coefficients of F. In 1903, G. Hamel found a system of
partial differential equations that characterize projectively flat metrics F D F.x; y/
on an open subset U Rn , that is,
F x m y i ym D F x i ; (11.2)
see [17]. A natural problem is to find projectively flat metrics by solving

p (11.2).
According to the Beltrami Theorem, a Riemannian metric F D gij .x/yi y j is
projectively flat if and only if it is of constant sectional curvature. Thus, Hilberts
fourth problem has been solved in Riemannian geometry. However, this problem
is far from being solved for Finsler metrics.
11.2 Preliminaries
A Finsler metric on a C1 manifold M is a function F W TM ! 0; 1/ with the

(1) Smoothness: F.x; y/ is C1 on TMnf0g;
(2) Homogeneity: F.x; y/D F.x;y/; 8 > 0;
(3) Regularity/Convexity: gij .x; y/ is positive definite, where
1 2
gij .x; y/ WD F yi y j .x; y/:
2
For a Finsler manifold .M; F/ and for each y 2 Tx M n f0g, we can define an inner
product gy W Tx M Tx M ! R W
gy .u; v/ D gij .x; y/ui v j ;

11 Hilberts Fourth Problem and Projectively Flat Finsler Metrics 249
where u D ui @x@ i jx ; v D v j @x@ j jx . By the homogeneity of F, we have

q
F.x; y/ D gij .x; y/yi y j :
Remark 11.2.1 The following are some special Finsler metrics.

p
(1) Minkowski metric: F.x; y/ D gij . y/yi y j . In this case, F.x; y/ is independent
of the position x 2 M. p
(2) Riemann metric: .x; y/ D aij .x/yi y j , where aij are independent of the
direction y 2 Tx M. p
(3) Randers metric [22]: F D C , where D aij .x/yi y j denotes a Riemannian
p
metric and D bi .x/yi denotes a 1-form with kk .x/ WD aij .x/bi .x/bj .x/ <
1; 8x 2 M. Randers metrics were first introduced by physicist G. Randers
in 1941 from the standpoint of general relativity. Randers metrics can also be
naturally characterized as the solution of the Zermelo navigation problem. It is
easy to show that a Finsler metric F is a Randers metric if and only if it is the
solution of Zermelo navigation problem on a Riemann space .M; h/ under the
influence of a force field W with jWjh < 1, where jWjh denotes the length of W
with respect to Riemannian metric h. We call the couple .h; W/ the navigation
data of Randers metric F. This fact shows that it is unavoidable to meet non-
Riemannian Finsler metrics in studying natural sciences. For more details about
Randers metrics, see [13].
(4) .; /-metrics: F D
.s/; s D =; where satisfies kx k < b0 (or
kx k 6 b0 ), 8x 2 M and
.s/ is a C1 function on .bo ; bo / satisfying

.s/ > 0;
.s/ s
0 .s/ C .b2 s2 /
00 .s/ > 0; .jsj 6 b < bo /:
Such metrics are called regular (or almost regular) .; /-metrics [20]. An
.; /-space can be considered as a perturbed Riemannian space by an
external force.
(; /-Metrics form a very important class of Finsler metrics which contains
many important and interesting metrics. When
D 1, we get the Riemannian metric
F D . If
D 1 p C s, the .; /-metric F D C is just
p a Randers metric. More
generally, if
D 1 C ks2 Cs, the .; /-metrics F D 2 C k 2 C are called
1
.; /-metrics of Randers type. When
D 1s , we obtain the Matsumoto metric
2 2

F D . The square metric in the form F D .C/

is defined by
D .1 C s/2 .
The famous Berwalds metric defined by L. Berwald in 1929 is an important class
of square metrics which is positively complete and projectively flat Finsler metric
with K D 0 (see Sect. 11.3 or [4, 15]).
The notion of projectively flatness is closely connected with the curvature
properties of Finsler metrics. Let D .t/ .a 6 t 6 b/ be a geodesic on
250 X. Cheng et al.
a Finsler manifold .M; F/. Let H.t; s/ be a variation of such that each curve
s .t/ WD H.t; s/ .a 6 t 6 b/ is a geodesic. Let
@H
J.t/ WD .t; 0/:
@s
Then the vector field J.t/ is a Jacobi field along satisfying the Jacobi equation:

D P D P J.t/ C R P J.t/ D 0:
Here, R denotes the Riemann curvature of F. Locally, for any x 2 M and y 2

Tx Mnf0g, the Riemann curvature Ry D Ri k @x@ i dxk is defined by
@Gi @2 Gi m 2 i
m @ G @Gi @Gm
Ri k D 2 y C 2G : (11.3)
@xk @xm @yk @ym @yk @ym @yk
The flag curvature of .M; F/ is the function K D K.x; y; / of a two-dimensional

plane called flag Tx M and a flagpole y 2 nf0g defined by
gy .Ry .u/; u/
K.x; y; / WD 2 ;
gy . y; y/gy .u; u/ gy .u; y/
where D spanfy; ug. The flag curvature is the natural generalization of the
sectional curvature in Riemannian geometry. When F is a Riemannian metric,
K.x; y; / D K.x; / is independent of the flagpole y and is just the sectional
curvature. A Finsler metric F is said to be of scalar flag curvature if the flag
curvature is independent of the flag , K D K.x; y/. F is said to be of weakly
isotropic flag curvature if
3
KD C .x/;
F
where D .x/ is a scalar function and is a 1-form on M. When K D constant,

F is said to be of constant flag curvature. The flag curvature governs the Jacobi
equation and the second variation of length.
A fundamental fact is that projectively flat Finsler metrics with Gi D P.x; y/yi
must be of scalar flag curvature,
P 2 P x k yk
KD :
F2
Another fundamental fact is that Beltrami Theorem is no longer true in Finsler
geometry. Let us look at an example.
Example 11.2.1 ([24] Shens Fish Tank) Let R3 be an open domain given by

WD .x1 ; x2 ; x3 /j.x1 /2 C .x2 /2 < 1 :
For each x D .x1 ; x2 ; x3 / 2 and y D fy1 ; y2 ; y3 g 2 Tp , define a Randers metric

on as
F D C ;
where
p
x2 y1 C x1 y2 2 C . y1 /2 C . y2 /2 C . y3 /2 1 .x1 /2 .x2 /2
WD ; (11.4)
1 .x1 /2 .x2 /2
x2 y1 C x1 y2
WD : (11.5)
1 .x1 /2 .x2 /2
It has been proved that K D 0. However, by BcsMatsumoto theorem in [2], F is

not projectively flat Finsler metric because is not closed.
In Finsler geometry, there are some important quantities which all vanish for
Riemannian metrics. Hence they are said to be non-Riemannian. Let F be a Finsler
metric on an n-dimensional manifold M. Let fbi g be a basis for Tx M and f! i g be the
basis for Tx M dual to fbi g. Define the BusemannHausdorff volume form by
dVBH WD BH .x/! 1 ^ ^ ! n ;
where

Vol Bn .1/
BH .x/ WD n o:
Vol . yi / 2 Rn jF.x; y/ < 1
Here Volfg denotes the Euclidean volume function on subsets in Rn and Bn .1/
denotes thepunit ball in Rn .
If F D gij .x/yi y j is a Riemannian metric, then
q
BH .x/ D det.gij .x//:
p
However, in general, for a Finsler metric F, BH .x/ det.gij .x; y//. Define
"p #
det.gij .x; y//
.x; y/ WD ln :
BH .x/
252 X. Cheng et al.
D .x; y/ is well defined, which is called the distortion of F. The distortion

characterizes the geometry of tangent space .Tx M; Fx /. It is well known that a
Finsler metric F is Riemannian if and only if .x; y/ D 0.
It is natural to study the rate of change of the distortion along geodesics. For a
vector y 2 Tx M n f0g, let D .x/ be the geodesic with .0/ D x and P .0/ D y.
Put
d
S.x; y/ WD . .t/; P .t// jtD0 :
dt
Equivalently,
S.x; y/ WD Im .x; y/ym ;
where ; denotes the horizontal covariant derivative with respect to F [15]. S D

S.x; y/ is called the S-curvature of Finsler metric F [23].
The S-curvature S.x; y/ measures the rate of change of .Tx M; Fx / in the direction
y 2 Tx M. As we know, for any piecewise C1 curve c D c.t/ from p to q in any
Berwald manifold .M; F/, the parallel translation Pc is a linear isometry between
.Tp M; Fp / and .Tq M; Fq /. Hence, it is easy to see that, for any Berwald metrics,
S D 0. In particular, S D 0 for Riemannian metrics [15, 23]. Hence, S-curvature is
a non-Riemannian quantity.
We say that F is of isotropic S-curvature if there exists a scalar function c.x/ on
M such that S.x; y/ D .n C 1/c.x/F.x; y/, equivalently,
Im .x; y/ym
D .n C 1/c.x/: (11.6)
F.x; y/
Equation (11.6) means that the rate of change of the tangent space .Tx M; Fx / along
the direction y 2 Tx M at each x 2 M is independent of the direction y but just
dependent on the point x. If c.x/ Dconstant, we say that F has constant S-curvature.
11.3 Projectively Flat Finsler Metrics of Constant

Flag Curvature
Firstly, let us recall three important kinds of projectively flat Finsler metrics.
1. Funk metric [16]. Let D .x; y/ be a Finsler metric on a strongly convex
domain Rn satisfying
y
xC 2 @:
.x; y/
is called a Funk metric. Funk metric has the following important property:
xk D yk : (11.7)

From this, we know that is a projectively flat Finsler metric with constant flag
curvature K D 14 . Further, is of constant S-curvature, S D nC1 2 . When
D Bn .1/, .x; y/ D N C N is actually a Randers metric given by
p
.1 jxj2 /jyj2 C < x; y >2 < x; y >
.x; y/ D C : (11.8)
1 jxj2 1 jxj2
More generally, a Finsler metric D .x; y/ satisfying (11.7) on an open subset
Rn is called a Funk metric.
2. Randers metric [2]. In 1997, S. Bcs and M. Matsumoto proved that a Randers
metric F D C is projectively flat if and only if is of constant sectional
curvature (that is, is projectively flat) and is closed.
3. Berwalds metric [4]. Let Bn .1/ Rn be the standard unit ball. Define
. C /2
FD ; y 2 Tx Bn .1/; (11.9)

N D N and
where D ;
p
.1 jxj2 /jyj2 C < x; y >2 N < x; y >
N D ; D ;
1 jxj2 1 jxj2
1
D :
1 jxj2
The Finsler metric in (11.9) is constructed by Berwald [4]. It is easy to verify that
F D F.x; y/ satisfies the following equations
F x k y l yk D F x l
and Fxk yk D 2F, where D .x; y/ is the Funk metric on Bn .1/ defined
by (11.8). Therefore, we can conclude that F is projectively flat and has zero flag
curvature, K D 0 (see [15]). The Finsler metrics in the form (11.9) are called
square metrics.
11.3.1 Projectively Flat Randers Metrics of Constant

Flag Curvature
As we mentioned before, a Randers metric F D C is projectively flat if and

only if is of constant sectional curvature and is closed. In this case, is locally
isometric to the following metric defined on a ball Bn .r / Rn ,
p
jyj2 C .jxj2 jyj2 hx; yi2 /
.x; y/ D :
1 C jxj2
Moreover, if F is of constant flag curvature, can be completely determined.
254 X. Cheng et al.
Theorem 11.3.1 ([25]) Let F D C be an n-dimensional Randers metric of

constant Ricci curvature Ric D .n 1/ F 2 with 6 0. Suppose that F is locally
projectively flat. Then 6 0: Further, if D 0, F is locally Minkowskian. If
D 1=4, F can be expressed in the following form
p
jyj2 .jxj2 jyj2 hx; yi2 / hx; yi ha; yi
FD 2
2
; y 2 Tx Rn ; (11.10)
1 jxj 1 jxj 1 C ha; xi
where a 2 Rn is a constant vector with jaj < 1: The Randers metric in (11.10) has
the following properties:
(a) K D 1=4I
(b) S D 12 .n C 1/FI
(c) all geodesics of F are straight lines.
The Finsler metrics defined by (11.10) can be expressed as follows
< a; y >
a WD .x; y/ ; (11.11)
1C < a; x >
where D .x; y/ is Funk metric on Bn .1/ given by (11.8). Theorem 11.3.1 is

the first classification theorem in Finsler geometry without the restriction that the
manifold is closed.
11.3.2 Projectively Flat Square Metrics of Constant

Flag Curvature
Berwalds metric in (11.9) is the first projectively flat square metric of constant flag
curvature. In 2008, by Hamel equation (11.2), Z. Shen and G.C. Yildirim first proved
the following
Theorem 11.3.2 ([28]) Let F D . C /2 = be a Finsler metric on a manifold M.
F is projectively flat if and only if
(i) bij j D f.1 C 2b2 /aij 3bi bj g,
(ii) the spray coefficients Gi of are in the form: Gi D yi 2 bi ,
where b WD kx k , bij j denote the covariant derivatives of with respect to ,
D .x/ is a scalar function and D ai .x/yi is a 1-form on M.
There are plenty of non-trivial Finsler metrics satisfying the conditions (i) and
(ii) in Theorem 11.3.2. By Theorem 11.3.2, we can completely determine the local
structure of a projectively flat Finsler metric F in the form F D . C /2 = which
is of constant flag curvature.
Theorem 11.3.3 ([28]) Let F D . C /2 = be a square Finsler metric on a

manifold M. Then F is locally projectively flat with constant flag curvature if and
only if one of the following conditions holds
(a) is flat and is parallel with respect to . In this case, F is locally
Minkowskian;
(b) Up to a scaling, F is isometric to the following metric Fa for any constant vector
a 2 Rn with jaj < 1:
.a N C a Na /2
Fa WD ; (11.12)
a N
where
p
.1 jxj2 /jyj2 C < x; y >2 < x; y > < a; y >
N D ; Na D C
1 jxj2 1 jxj2 1C < a; x >
and
.1C < a; x >/2
a WD :
1 jxj2
In both cases (a) and (b), the flag curvature of F must be zero, K D 0.
Recently, Zhou has proved that, if a square metric F has constant flag curvature
K, then F must be locally projectively flat Finsler metric [30]. In this case, according
to Theorem 11.3.3, K D 0.
11.3.3 Projectively Flat .; /-Metrics of Constant

Flag Curvature
In Finsler geometry, it is very difficult in general to compute the curvatures of a

Finsler metric. Some Finsler metrics are defined by some elementary functions,
but their expressions of curvatures are extremely complicated so that one cannot
easily determine their values. .; /-Metrics are relatively simple and computable
Finsler metrics with interesting curvature properties. In recent years, various
curvatures in Finsler geometry have been studied and their geometric meanings are
better understood. This is partially due to the study of .; /-metrics.
In [27], Shen has characterized all projectively flat .; /-metrics.
Theorem 11.3.4 ([27]) Let F D
.=/ be an .; /-metric onp an open subset
U in the n-dimensional Euclidean space Rn (n > 3), where D aij .x/yi y j and
D bi .x/yi 6D 0. Suppose that the following conditions hold:
(a) is notpparallel with respect to ,
6 a2 1 C a1 s2 C a3 s for some constants a1 ; a2 , and a3 with a2 > 0, that is,
(b)
D
F is not .; /-metric of Randers type.
256 X. Cheng et al.
Then F is projectively flat on U if and only if

(i)
D
.s/ satisfies
n o n o
1 C .k1 C k2 s2 /s2 C k3 s2
00 D .k1 C k2 s2 /
s
0 :
(ii) and satisfy

n o
bij j D 2 .1 C k1 b2 /aij C .k2 b2 C k3 /bi bj ;

Gi D yi k1 2 C k2 2 bi ;
where D .x/ is a scalar function on U and k1 ; k2 and k3 are constants with

.k2 ; k3 / 6D .0; 0/:
Using Theorem 11.3.4, one can classify locally projectively flat .; /-metrics
FDp
.=/ of constant flagpcurvature. Roughly
p speaking, if F is not trivial, then
D 1 C ks2 C s or
D . 1 C ks2 C s/2 = 1 C ks2 .
Theorem 11.3.5 ([19]) Let F D
.=/
p be an .; /-metric on an open subset
U Rn (n > 3), where D aij .x/yi y j and D bi .x/yi 0. Then F is
projectively flat Finsler metric with constant flag curvature K if and only if one
of the following holds:
(i) is projectively
p flat and is parallel with respect to .
(ii)
D 1 C ks2 C s, where k; .6D 0/ are constants. In this case, p F D N C N
is of Randers type with constant flag curvature K < 0, where N D 2 C k 2
and N pD . p
(iii)
D . 1 C ks2 C s/2 = 1 C ks2 , where k; .6D 0/ are constants. In this case,
F Dp N 2 =N is a square metric with constant flag curvature K D 0, where
.N C /
N D C k 2 and N D
2
By Theorem 11.3.1, we can determine the structure of the metrics in Theo-

rem 11.3.5(ii). By Theorem 11.3.3, we can determine the structure of the metrics
in Theorem 11.3.5(iii).
Recently, Yu studied and characterized all projectively flat .; /-metrics. He
classified projectively flat .; /-metrics in dimension n > 3 by a special class of
deformations. His results show that the projective flatness of an .; /-metric always
arises from the projective flatness of some Riemannian metrics when dimension
n > 3. See [29].
11.3.4 Projectively Flat Finsler Metrics of Constant

Flag Curvature
On every strongly convex domain U in Rn , Hilbert constructed a complete

reversible projectively flat metric H D H.x; y/ with negative constant flag curvature
K D 1. Then Funk constructed a positively complete projectively flat metric
D .x; y/ with K D 1=4 on U so that its symmetrization is just the Hilbert
metric, H.x; y/ D 12 ..x; y/ C .x; y// [16]. When U D Bn .1/ is the unit ball in
Rn , the Funk metric is given by (11.8). On the other hand, Berwald [4] constructed
a metric given by (11.9) on the unit ball Bn .1/, which is projectively flat metric with
zero flag curvature.
In [57], Bryant studied and characterized locally projectively flat Finsler
metrics with constant flag curvature K D 1 on S2 and Sn . It is clear that Bryants
metrics cannot be expressed in terms of a Riemannian metric and a 1-form as
Randers metrics and Berwalds metrics.
Further, Shen studied and characterized projectively flat Finsler metrics of
constant flag curvature in [26]. Shen proved that such metrics can be described
using algebraic equations or using Taylor expansions. Shen gave a formula for
x-analytic projectively flat Finsler metrics with constant flag curvature using a
power series with coefficients expressed in terms of F.0; y/ and Fxk .0; y/yk . Shen
also gave a formula for general projectively flat Finsler metrics with constant flag
curvature using some algebraic equations depending on F.0; y/ and Fxk .0; y/yk . By
these formulas, he obtained several interesting projectively flat Finsler metrics of
constant flag curvature which can be used as models in certain problems.
11.4 Projectively Flat Finsler Metrics with Isotropic

S-Curvature
11.4.1 Projectively Flat Randers Metrics with Isotropic

S-Curvature
By Theorem 11.3.1, Z. Shen has classified projectively flat Randers metrics of

constant flag curvature. Later on, Bao and Robles proved the following result: If
a Randers metric F is an Einstein metric with Ric D .n 1/ .x/F 2 , then F is
of constant S-curvature [3]. This naturally leads to the study of projectively flat
Randers metrics with isotropic S-curvature.
Let F D C be a locally projectively flat Randers metric. Then is locally
projectively flat and is closed. According to the Beltrami theorem in Riemann
geometry, is locally projectively flat if and only if it is of constant sectional
curvature. Thus we may assume that is of constant sectional curvature . It is
258 X. Cheng et al.
locally isometric to the following standard metric on the unit ball Bn Rn or the
whole Rn for D 1; 0; C1:
p
jyj2 .jxj2 jyj2 hx; yi2 /
1 .x; y/ D ; y 2 Tx Bn Rn ; (11.13)
1 jxj2
0 .x; y/ D jyj; y 2 Tx Rn Rn ; (11.14)
p
jyj2 C .jxj2 jyj2 hx; yi2 /
C1 .x; y/ D ; y 2 Tx Rn Rn : (11.15)
1 C jxj2
Then we can determine if F is of isotropic S-curvature and get the following

classification theorem.
Theorem 11.4.1 ([14]) Let F D C be a locally projectively flat Randers metric
on an n-dimensional manifold M and denote the constant sectional curvature of
. Suppose that the S-curvature is isotropic, S D .n C 1/c.x/F. Then F can be
classified as follows.
(A) If C 4c.x/2 0, then c.x/ D constant and K D c2 6 0.
(A1) if c D 0, then F is locally Minkowskian with flag curvature K D 0;
(A2) if c 6D 0, then after a normalization, F is locally isometric to the following
Randers metric on the unit ball Bn Rn ,
p
F.x; y/ D 2
; (11.16)
1 jxj 1 C ha; xi
where a 2 Rn with jaj < 1, and the flag curvature of F is negative

constant, K D 14 .
(B) If C 4c.x/2 6D 0, then F is given by
2cxk .x/yk
F.x; y/ D .x; y/ (11.17)
C 4c.x/2
and the flag curvature of F is given by
3cxk .x/yk
KD C 3c.x/2 C : (11.18)
F.x; y/
(B1) when D 1, D 1 can be expressed in the form (11.13) on Bn .

In this case,
C ha; xi
c.x/ D p ; (11.19)
2 . C ha; xi/2 .1 jxj2 /
where 2 R and a 2 Rn with jaj2 < 2 1.

(B2) when D 0, D 0 can be expressed in the form (11.14) on Rn . In

this case,
1
c.x/ D p ; (11.20)
2 C ha; xi C jxj2
where > 0 and a 2 Rn with jaj2 < .

(B3) when D 1, D C1 can be expressed in the form (11.15) on Rn . In
this case,
C ha; xi
c.x/ D p ; (11.21)
2 1 C jxj2 . C ha; xi/2
where 2 R and a 2 Rn with jj2 C jaj2 < 1.

Theorem 11.4.1 (A) follows from the classification theorem in [25] after we
prove that the flag curvature is constant in this case (also see Theorem 11.3.1). From
Theorem 11.4.1 we obtain some interesting projectively flat Randers metrics with
isotropic S-curvature.
Example 11.4.1 Let
p p
.1 jxj2 /jyj2 C hx; yi2 .1 jxj2 / C 2 C hx; yi
F .x; y/ D p ; y 2 Tx Bn ;
2 2
.1 jxj / .1 jxj / C 2
(11.22)
where 2 R is an arbitrary constant. The geodesics of F are straight lines in Bn .

One can easily verify that F is complete in the sense that every unit speed geodesic
of F is defined on .1; 1/. Moreover F has strictly negative flag curvature
K 6 14 .
Example 11.4.2 Let
p
jyj 1 C jxj2 C hx; yi
F0 .x; y/ D p ; y 2 Tx Rn : (11.23)
1 C jxj2
The geodesics of F0 are straight lines in Rn . One can easily verify that F0 is
positively complete in the sense that every unit speed geodesic of F0 is defined
on .a; 1/. Moreover F0 has positive flag curvature K > 0.
Theorem 11.4.1 is a local classification theorem. If we assume that the manifold
is closed (compact without boundary), then the scalar function c.x/ takes much more
special values [14]. In particular, we have the following
Theorem 11.4.2 ([14]) Let Sn D .M; / be the standard unit sphere and F D C
be a projectively flat Randers metric on Sn . Suppose that S-curvature is isotropic,
S D .n C 1/c.x/F. Then
f .x/
c.x/ D p
2 1 f .x/2
260 X. Cheng et al.
and
f k .x/yk
F.x; y/ D .x; y/ p x ;
1 f .x/2
where f .x/ is an eigenfunction of Sn corresponding to the first eigenvalue.

Moreover,
p
(a) WD jrf j2 .x/ C f .x/2 < 1 is a constant and we have the following estimates
for flag curvature
2 2C
6K6 :
2.1 C / 2.1 /
(b) The geodesics of F are the great circles on Sn with F-length 2.
11.4.2 Projectively Flat .; /-Metrics with Isotropic

S-Curvature
Note that projectively flat Finsler metrics must be of scalar flag curvature. We firstly
recall the characterizations on .; /-metrics of scalar flag curvature with isotropic
S-curvature. Based on the main result in [12] on .; /-metrics with isotropic S-
curvature given by the first author and Z. Shen, the first author has proved the
following theorem.
Theorem 11.4.3 ([10]) A regular .; /-metric F D
.=/ of non-Randers
type on an n-dimensional manifold M is of isotropic S-curvature, S D .n C 1/cF, if
and only if satisfies
rij D 0; sj D 0: (11.24)
In this case, S D 0, regardless of the choice of a particular

D
.s/. Here,
1 1
rij WD .bij j C bjji /; sij WD .bij j bjji /;
2 2
si j WD aih shj ; sj WD bi si j D bm smj
and j denotes the covariant derivative with respect to the Levi-Civita connection
of .
Further, the first author obtained the following classification theorem for regular
.; /-metrics of non-Randers type of scalar flag curvature whose S-curvatures are
isotropic.
Theorem 11.4.4 ([9, 10]) Let F D

.=/ be a regular .; /-metric of non-
Randers type on an n-dimensional manifold M .n > 3/. Then F is of scalar flag
curvature and of isotropic S-curvature if and only if the flag curvature K D 0 and
F is a Berwald metric. In this case, F is a locally Minkowski metric.
If we regard locally Minkowski metrics as so-called trivial Finsler metrics, then,
by Theorem 11.4.4, we have the following theorem.
Theorem 11.4.5 ([10]) The non-trivial regular .; /-metrics of scalar flag cur-
vature with isotropic S-curvature on an n-dimensional manifold M .n > 3/ must be
Randers metrics. In this case, the metrics are of weakly isotropic flag curvature and
are completely determined by navigation data .h; W/ (for more details, see [13]).
From Theorems 11.4.4 and 11.4.5, we can obtain the following theorem on
projectively flat .; /-metrics with isotropic S-curvature.
Theorem 11.4.6 Let F D
.=/ be a regular .; /-metric on an n-dimensional
manifold M .n > 3/. Then F is projectively flat metric with isotropic S-curvature
if and only if F is a locally Minkowski metric of non-Randers type, or F is a
projectively flat Randers metric with isotropic S-curvature.
11.4.3 Projectively Flat Finsler Metrics with Isotropic

S-Curvature
We have classified projectively flat Randers metrics and .; /-metrics with

isotropic S-curvature. It is a natural problem to study and characterize projectively
flat Finsler metrics with isotropic S-curvature. In [14], we have shown the following
theorem.
Theorem 11.4.7 ([14]) Let .M; F/ be an n-dimensional Finsler manifold. Assume
that F is of scalar curvature K D K.x; y/. If the S-curvature of F is isotropic, that
is, S D .n C 1/cF, where c D c.x/ is a scalar function, then the flag curvature K is
given by
cx m ym
KD3 C ; (11.25)
F
where D .x/ is a scalar function on M. In particular, c D constant if and only if

K D K.x/ is a scalar function on M.
Shen and Yildirim have proved that the converse of Theorem 11.4.7 is also true
for Randers metrics (see [28]). Because projectively flat Finsler metrics must be
of scalar flag curvature, Theorem 11.4.7 plays a very important role in classifying
projectively flat Randers metrics with isotropic S-curvature. Further, we can study
and characterize locally projectively flat Finsler metrics with isotropic S-curvature
by using Theorem 11.4.7.
262 X. Cheng et al.
There are many projectively flat Randers metrics with isotropic S-curvature [14].
Besides these Randers metrics, there is another special kind of projectively flat
Finsler metrics with isotropic S-curvatureFunk metrics. More generally, we have
the following classification theorem.
Theorem 11.4.8 ([11]) Let F D F.x; y/ be a locally projectively flat Finsler metric
on an open subset Rn . Suppose that F has isotropic S-curvature, S D .n C
1/c.x/F. Then the flag curvature is in the form
cx m ym
KD3 C ;
F
where D .x/ is a scalar function on .

m
(a) If K c2 C cxmFy on , then F D C is a projectively flat Randers metric
with isotropic S-curvature
m
S D .n C 1/cF
(b) If K c2 C cxmFy on , then c D constant and F is either locally
Minkowskian (c D 0) or, up to a scaling, locally isometric to the metric
< a; y > 1
a WD .x; y/ C ; cD
1C < a; x > 2
or its reverse
< a; y > 1
N a WD .x; y/ ; cD ;
1C < a; x > 2
where a 2 Rn is a constant vector and .x; y/ is Funk metric on .

In Theorem 11.4.8(a), the local structure of F has been completely determined in
[14] (also see Theorem 11.4.1). Theorem 11.4.8 tells us that non-trivial projectively
flat Finsler metrics with isotropic S-curvature are either projectively flat Randers
metrics with isotropic S-curvature, or completely determined by Funk metrics.
Acknowledgements X. Cheng was supported by the National Natural Science Foundation of

China (11371386) and the European Unions Seventh Framework Programme (FP7/20072013)
under grant agreement no. 317721.
References
1. J.C. lvarez Paiva, Symplectic geometry and Hilberts fourth problem. J. Differ. Geom. 69(2),
353378 (2005)
2. S. Bcs, M. Matsumoto, On Finsler spaces of Douglas type. A generalization of the notion of
Berwald space. Publ. Math. Debr. 51, 385406 (1997)
3. D. Bao, C. Robles, On Randers metrics of constant curvature. Rep. Math. Phys. 51, 942 (2003)
4. L. Berwald, ber die n-dimansionalen Geometrien konstanter Krmmung, in denen die
Geraden die krzesten sind. Math. Z. 30, 499469 (1929)
5. R. Bryant, Finsler structures on the 2-sphere satisfying K D 1, in Finsler Geometry,

Contemporary Mathematics, vol. 196 (American Mathematical Society, Providence, 1996),
pp. 2742
6. R. Bryant, Projectively flat Finsler 2-spheres of constant curvature. Sel. Math. 3, 161204
(1997)
7. R. Bryant, Some remarks on Finsler manifolds with constant flag curvature. Houst. J. Math.
28(2), 221262 (2002)
8. H. Busemann, Problem IV: Desarguesian spaces in mathematical developments arising from
Hilbert problems, in Proceedings of Symposia in Pure Mathematics (American Mathematical
Society, Providence, 1976)
9. X. Cheng, On .; /-metrics of scalar flag curvature with constant S-curvature. Acta Math.
Sin. 26(9), 17011708 (2010)
10. X. Cheng, The .; /-metrics of scalar flag curvature. Differ. Geom. Appl. 35, 361369 (2014)
11. X. Cheng, Z. Shen, Projectively flat Finsler metrics with almost isotropic S-curvature. Acta
Math. Sci. 26B(2), 307313 (2006)
12. X. Cheng, Z. Shen, A class of Finsler metrics with isotropic S-curvature. Isr. J. Math. 169(1),
317340 (2009)
13. X. Cheng, Z. Shen, Finsler GeometryAn Approach via Randers Spaces (Springer/Science
Press, Berlin, 2012)
14. X. Cheng, X. Mo, Z. Shen, On the flag curvature of Finsler metrics of scalar curvature. J. Lond.
Math. Soc. 68(2), 762780 (2003)
15. S.S. Chern, Z. Shen, Riemann-Finsler Geometry. Nankai Tracts in Mathematics, vol. 6 (World
Scientific, Singapore, 2005)
16. P. Funk, ber Geometrien bei denen die Geraden die krzesten sind. Math. Ann. 101, 226237
(1929)
17. G. Hamel, ber die Geometrien in denen die Geraden die krzesten sind. Math. Ann. 57, 231
264 (1903)
18. D. Hilbert, Mathematical problems. Bull. Am. Math. Soc. 8, 437479 (1902); Bull. Am. Math.
Soc. 37(4), 407436 (2000)
19. B. Li, Z. Shen, On a class of projectively flat Finsler metrics with constant flag curvature. Int.
J. Math. 18(7), 112 (2007)
20. M. Matsumoto, On C-reducible Finsler spaces. Tensor N. S. 24, 2937 (1972)
21. A.V. Pogorelov, Hilberts Fourth Problem. Scripta Series in Mathematics (Winston and Sons,
Washington, 1979)
22. G. Randers, On an asymmetrical metric in the four-space of general relativity. Phys. Rev. 59,
195199 (1941)
23. Z. Shen, Volume comparison and its applications in Riemann-Finsler geometry. Adv. Math.
128(2), 306328 (1997)
24. Z. Shen, Finsler metrics with K=0 and S=0. Can. J. Math. 55(1), 112132 (2003)
25. Z. Shen, Projectively flat Randers metrics with constant flag curvature. Math. Ann. 325, 1930
(2003)
26. Z. Shen, Projectively flat Finsler metrics of constant flag curvature. Trans. Am. Math. Soc.
355(4), 17131728 (2003)
27. Z. Shen, On projectively flat .; /-metrics. Can. Math. Bull. 52(1), 132144 (2009)
28. Z. Shen, G.C. Yildirim, On a class of projectively flat metrics with constant flag curvature.
Can. J. Math. 60(2), 443456 (2008)
29. C. Yu, Deformations and Hilberts fourth problem. Math. Ann. 365, 13791408 (2016)
30. L. Zhou, A local classification of a class of .; / metrics with constant flag curvature. Differ.
Geom. Appl. 28, 170193 (2010)
Chapter 12
Holonomy Theory of Finsler Manifolds
Zoltn Muzsnay and Pter T. Nagy
Abstract The holonomy group of a Riemannian or Finslerian manifold can be

introduced in a very natural way: it is the group generated by parallel translations
along loops with respect to the canonical connection. The Riemannian holonomy
groups have been extensively studied and by now their complete classification is
known. On the Finslerian holonomy, however, only few results are known and, as
our results show, it can be essentially different from the Riemannian one.
In recent papers we have developed a method for the investigation of holon-
omy properties of non-Riemannian Finsler manifolds by constructing tangent Lie
algebras to the holonomy group: the curvature algebra, the infinitesimal holonomy
algebra, and the holonomy algebra. In this book chapter we present this method and
give a unified treatment of our results. In particular we show that the dimension of
these tangent algebras is usually greater than the possible dimensions of Riemannian
holonomy groups and in many cases is infinite. We prove that the holonomy group of
a locally projectively flat Finsler manifold of constant curvature is finite dimensional
if and only if it is a Riemannian manifold or a flat Finsler manifold. We also show
that the topological closure of the holonomy group of a certain class of simply
connected, projectively flat Finsler 2-manifolds of constant curvature (spherically
symmetric Finsler 2-manifolds) is not a finite dimensional Lie group, and we prove
that its topological closure is the connected component of the full diffeomorphism
group of the circle.
12.1 Introduction
The notion of the holonomy group of a Riemannian or Finslerian manifold can be

introduced in a very natural way: it is the group generated by parallel translations
along loops. In contrast to the Finslerian case, the Riemannian holonomy groups
Z. Muzsnay ()
Institute of Mathematics, University of Debrecen, P.O.B. 400, H-4010 Debrecen, Hungary
e-mail: muzsnay@science.unideb.hu
P.T. Nagy
Institute of Applied Mathematics, buda University, Bcsi t 96/b, H-1034 Budapest, Hungary
e-mail: nagy.peter@nik.uni-obuda.hu

266 Z. Muzsnay and P.T. Nagy
have been extensively studied. One of the earliest fundamental results is the theorem
of Borel and Lichnerowicz [3] from 1952, claiming that the holonomy group of a
simply connected Riemannian manifold is a closed Lie subgroup of the orthogonal
group O.n/. By now, the complete classification of Riemannian holonomy groups
is known. Similarly to the Riemannian case the holonomy group of a Finsler
manifold is the subgroup of the diffeomorphism group of an indicatrix, generated by
canonical homogeneous (nonlinear) parallel translations along closed loops. Before
our investigation (c.f. [1823]), the holonomy groups of non-Riemannian Finsler
manifolds have been described only in special cases: for Berwald manifolds there
exist Riemannian metrics with the same holonomy group (cf. Szab, [31]), for
positive definite Landsberg manifolds the holonomy groups are compact Lie groups
consisting of isometries of the indicatrix with respect to an induced Riemannian
metric (cf. Kozma [13, 14]). A thorough study of holonomy groups of homogeneous
(nonlinear) connections was initiated by Barthel in his basic work [2] in 1963; he
gave a construction for a holonomy algebra of vector fields on the tangent space. A
general setting for the study of infinite dimensional holonomy groups and holonomy
algebras of nonlinear connections was initiated by Michor in [17]. However the
introduced holonomy algebras could not be used to estimate the dimension of
the holonomy group since their tangential properties to the holonomy group were
not clarified.
Now we give a unified treatment of our results on Finslerian holonomy theory.
We construct and investigate tangent Lie algebras to the holonomy group and show
that the dimension of these tangent algebras in many cases is infinite, particularly it
is greater than the possible dimensions of Riemannian holonomy groups.
In the second section we collect the necessary definitions and constructions of
spray and Finsler geometry. The third section is devoted to the investigation of
tangential properties of subalgebras of the Lie algebra of vector fields on a manifold
to the infinite dimensional diffeomorphism group of this manifold. Particularly we
consider the case if the manifold is compact, in this case the diffeomorphism group
is an infinite dimensional Lie group modeled on the Lie algebra of vector fields on
the manifold.
In Sect. 12.4 we introduce the notion of curvature algebra of a Finsler manifold
consisting of tangent vector fields on the indicatrix, which is a generalization of the
matrix group generated by curvature operators of a Riemannian manifold. We show
that the vector fields belonging to the curvature algebra are tangent to 1-parameter
families of diffeomorphisms contained in the holonomy group. We prove that for a
positive definite non-Riemannian Finsler manifold of non-zero constant curvature
with dimension n > 2 the dimension of the curvature algebra is strictly greater
than the dimension of the orthogonal group acting on the tangent space and hence
it cannot be a compact Lie group. In addition, we provide an example of a left
invariant singular (non y-global) Finsler metric of Berwald-Mor-type on the three-
dimensional Heisenberg group which has infinite dimensional curvature algebra and
hence its holonomy is not a (finite dimensional) Lie group. These results give a
positive answer to the following problem formulated by Chern and Shen in [6]
12 Holonomy Theory of Finsler Manifolds 267
(p. 85): Is there a Finsler manifold whose holonomy group is not the holonomy
group of any Riemannian manifold?
Section 12.5 contains constructions of further tangent Lie algebras to the
holonomy group consisting of tangent vector fields on the indicatrix, namely the
infinitesimal holonomy algebra and the holonomy algebra of a Finsler manifold.
Our goal is to make an attempt to find the right notion of the holonomy algebra of
Finsler spaces. The holonomy algebra should be the largest Lie algebra such that
all its elements are tangent to the holonomy group. In our attempt we are building
successively Lie algebras having the tangent properties. We define the infinitesimal
holonomy algebra by the smallest Lie algebra of vector fields on an indicatrix,
containing the curvature vector fields and their horizontal covariant derivatives with
respect to the Berwald connection and prove the tangential property of this Lie
algebra to the holonomy group. At the end we introduce the notion of the holonomy
algebra of a Finsler manifold by all conjugates of infinitesimal holonomy algebras
by parallel translations with respect to the Berwald connection. We prove that this
holonomy algebra is tangent to the holonomy group. A different treatment of the
holonomy algebra is given by Crampin and Saunders in [7]. The question of whether
the holonomy algebra introduced in this way is the largest Lie algebra, which is
tangent to the holonomy group, is still open.
In Sect. 12.6 we construct for interesting classes of locally projectively flat
Finsler surfaces and manifolds of non-zero constant curvature infinite dimensional
subalgebras in the tangent infinitesimal holonomy algebras. From the viewpoint
of non-Euclidean geometry the most important Riemann-Finsler manifolds are the
projectively flat spaces of constant flag curvature. We will turn our attention to non-
Riemannian projectively flat Finsler manifolds of non-zero constant flag curvature.
We consider the following classes of locally projectively flat non-Riemannian
Finsler manifolds of non-zero constant flag curvature:
1. Randers manifolds,
2. manifolds having a two-dimensional subspace in the tangent space at some point,
on which the Finsler norm is an Euclidean norm,
3. manifolds having a two-dimensional subspace in the tangent space at some point,
on which the Finsler norm and the projective factor are linearly dependent.
The first class consists of positively complete Finsler manifolds of negative curva-
ture, the second class contains a large family of (not necessarily complete) Finsler
manifolds of negative curvature, and the third class contains a large family of not
necessarily complete Finsler manifolds of positive curvature. The metrics belonging
to these classes can be considered as (local) generalizations of a 1-parameter family
of complete Finsler manifolds of positive curvature defined on S2 by Bryant in [4, 5]
and on Sn by Shen in [30], Example 7.1. It follows from these constructions that the
holonomy group of Finsler manifolds belonging to the above classes and satisfying
some additional technical assumption is infinite dimensional.
In Sect. 12.7 we study the holonomy group of an arbitrary locally projectively flat
Finsler manifolds of constant curvature. Our aim is to characterize all locally pro-
jectively flat Finsler manifolds with finite dimensional holonomy group. To obtain
such a characterization, we investigate the dimension of the infinitesimal holonomy

algebra. We obtain that if .M; F / is a non-Riemannian locally projectively flat
Finsler manifolds of non-zero constant curvature, then its infinitesimal holonomy
algebra is infinite dimensional. Using this general result and the tangent property of
the infinitesimal holonomy algebra we obtain the characterization: The holonomy
group of a locally projectively flat Finsler manifold of constant curvature is finite
dimensional if and only if it is a Riemannian manifold or a flat Finsler manifold.
Section 12.8 is devoted to show that the topological closure of the holonomy
group of a certain class of simply connected, projectively flat Finsler 2-manifolds
of constant curvature is not a finite dimensional Lie group, and we prove that its
topological closure is the connected component of the full diffeomorphism group
of the circle. Before our investigation, perhaps because of technical difficulties, not
a single infinite dimensional Finsler holonomy group has been described. Now, we
provide the first such a description. This class of Finsler 2-manifolds contains the
positively complete standard Funk plane of constant negative curvature (positively
complete standard Funk plane), and the complete irreversible Bryant-Shen-spheres
of constant positive curvature [5, 30]. We obtain that for every simply connected
Finsler 2-manifold the topological closure of the holonomy group is a subgroup of
Diff1 1
C .S /. That means that in the examples mentioned above, the closed holonomy
group is maximal. In the proof we use our constructive method developed in
Sect. 12.6 for the study of infinite dimensional Lie subalgebras of the infinitesimal
holonomy algebra.
12.2 Preliminaries
12.2.1 Spray Manifolds
A spray on a manifold M is a smooth vector field S on the slit tangent bundle

O WD TM nf0g expressed in a standard coordinate system .xi ; yi / on TM as
TM
@ @
S D yi 2Gi .x; y/ i ; (12.1)
@xi @y
where the functions Gi .x; y/ of local coordinates .xi ; yi / on TM satisfy
Gi .x; y/ D 2 Gi .x; y/; > 0: (12.2)
A manifold M with a spray S is called a spray manifold .M; S /, cf. [27], Chapter 5.
A curve c.t/ is called geodesic of the spray manifold .M; S / if its coordinate
functions ci .t/ satisfy the system of 2nd order ordinary differential equations
cR i .t/ C 2Gi .c.t/; cP .t// D 0; (12.3)
where the functions Gi .x; y/ are the geodesic coefficients the spray manifold
.M; S /.
12.2.1.1 Horizontal Distribution, Covariant Derivative and Curvature
Let .TM; ; M/ and .TTM; ; TM/ be the first and the second tangent bundle of
the manifold M, respectively, and let V TM TTM be the (integrable) vertical
distribution on TM given by V TM WD Ker . The horizontal distribution H TM
TTM associated to the spray manifold .M; S / is the image of the horizontal lift
which is the vector space isomorphism ly W Tx M ! Hy TM for x 2 M and y 2 Tx M
defined by

@ @ @ @Gi
ly D Gki .x; y/ k ; where Gij D (12.4)
@xi @xi @y @y j
in the coordinate system .xi ; yi / of TM. The horizontal distribution is complementary

to the vertical distribution, hence we have the decomposition Ty TM D Hy TM
Vy TM. The projectors corresponding to this decomposition will be denoted by h W
TTM ! H TM and v W TTM ! V TM.
The vertical distribution over the slit tangent bundle TM O D TM n f0g will be
denoted by .VO TM; ; TM/
O and the pull-back bundle of .TM; O ; M/ corresponding
to the map W TM ! M by . TM; N ; TM/. O Clearly, the mapping

@ @
x; y; i i 7! x; y; i i W VO TM ! TM (12.5)
@y @x
is a canonical bundle isomorphism. In the following we will use the isomor-

phism (12.5) for the identification of these bundles.
Let X1 .M/ be the vector space of smooth vector fields on the manifold M
and XO 1 .TM/ the vector space of smooth sections of the bundle .VO TM; ; TM/.
O
O 1
The horizontal covariant derivative of a section 2 X .TM/ by a vector field
X 2 X1 .M/ is given by
rX WD l.X/; :
If .x; y/ D i .x; y/ @y@ i and X.x/ D X i .x/ @x@ i , then rX can be expressed as

@ i .x; y/ @ i .x; y/ @
rX D G k
.x; y/ C G i
.x; y/ k
.x; y/ Xj i ; (12.6)
@x j j
@y k jk
@y
@Gi
where Gijk WD @ykj .
Defining the horizontal covariant derivative

@
@
.x; y/
rX
D l.X/
D G k
.x; y/ Xj
@xj j
@yk
O ! R, the horizontal covariant derivation (12.6) can be

of a smooth function
W TM
O
extended to sections of the tensor bundle over . TM; ; TM/, using the canonical
bundle isomorphism (12.5).
The curvature tensor field

K.x;y/ .X; Y/ WD v X h ; Y h ; X; Y 2 Tx M: (12.7)
O
on the pull-back bundle . TM; ; TM/ of the spray manifold .M; S / in a local
coordinate system .x ; y / of TM is given by
i i
@
K.x;y/ D Kjki .x; y/dxj dxk ;
@xi
where
@Gij .x; y/ @Gik .x; y/

Kjki .x; y/ D C Gm
j .x; y/Gkm .x; y/ Gk .x; y/Gjm .x; y/:
i m i
@xk @xj
(12.8)
The curvature tensor field characterizes the integrability of the horizontal distribu-
tion. Namely, if the horizontal distribution H TM is integrable, then the curvature
is identically zero.
12.2.1.2 Parallel Translation
For a spray manifold .M; S / the parallel vector fields X.t/ D X i .t/ @x@ i along a
curvec.t/ are defined by the solutions of the differential equation

dX i .t/ @
DcP X.t/ WD C Gj .c.t/; X.t//Pc .t/
i j
D 0: (12.9)
dt @xi
Using the relations (12.2) and Euler theorem on homogeneous functions we see
that the functions Gij .x; y/ are positive homogeneous of first order with respect to
the variable y, and hence DcP .X.t// D DcP X.t/ for any 0. The differential
equation (12.9) can be expressed by the horizontal covariant derivative (12.6) using
the bundle isomorphism (12.5) as follows: a vector field X.t/ D X i .t/ @x@ i along a
curve c.t/ is parallel if it satisfies the equation

dX i .t/ @
rcP X.t/ D C Gj .c.t/; X.t//Pc .t/
i j
D 0: (12.10)
dt @xi
Clearly, for any X0 2 Tc.0/ M there is a unique parallel vector field X.t/ along the
curve c such that X0 D X.0/. Moreover, if X.t/ is a parallel vector field along c, then
X.t/ is also parallel along c for any 0. Then the homogeneous (nonlinear)
parallel translation
c W Tc.0/ M ! Tc.1/ M along a curve c.t/
of the spray manifold .M; S / is defined by the positive homogeneous map c W

X0 7! X1 given by the value X1 D X.1/ at t D 1 of the parallel vector field with
initial value X.0/ D X0 (Fig. 12.1).
Since the parallel translation of a spray manifold .M; S / is determined by its
horizontal distribution H TM TTM, a spray manifold can be considered as a
particular case of a fibered manifold equipped with an Ehresmann connection (cf.
[8]). An Ehresmann connection of a fibered manifold is given by a horizontal
distribution, which is complement to the vertical distribution consisting of the
tangent spaces of the fibers. For a spray manifold the fibered manifold is the
tangent bundle of M and the horizontal distribution determined by the horizontal
lift ly W Tx M ! Hy TM expressed by Eq. (12.4).
The parallel translation can be introduced in a very nice geometrical way with
the help of the notion of horizontal distribution. Namely, we call a curve in TM
horizontal if the tangent vectors of this curve are contained in the horizontal
distribution H TM TTM. Let now c.t/ be a curve in the manifold M joining the
points p and q. The horizontal lift ch .t/ D .c.t/; X i .t/ @x@ i / of c.t/ is the curve ch .t/ in
TM defined by the properties that ch .t/ projects on c.t/ and ch .t/ is horizontal that
is cP h .t/ 2 Hc.t/ . This means according to Eq. (12.4) that

@ d @ @ @
cP .t/ i C X i .t/ i D
i
Gki .x; y/ k cP i .t/;
@x dt @y @xi @y
i.e., the tangent vector of the lifted curve ch .t/ is the horizontal lift of the tangent
vector cP i .t/ @x@ i of c.t/. It follows that a vector field X.t/ along a curve c.t/ is parallel
if and only if it is a solution of the differential equation

d @
c.t/; X .t/ i D lX.t/ .Pc.t//;
i
(12.11)
dt @x
or equivalently X.t/ satisfies the differential equation (12.9). Hence the parallel
translation along a curve c.t/ joining the points p and q is the map c W Tp M ! Tq M
Fig. 12.1 Parallel translation

X1
X0
c(t)
Fig. 12.2 Geometric

construction of the parallel
translation
TM ch
v w
v c w
determined by the intersection points of the horizontal lifts of the curve c.t/ with
the tangent spaces Tp and Tq . The construction can be illustrated by Fig. 12.2.
12.2.1.3 Totally Geodesic and Auto-Parallel Submanifolds
A submanifold M N in a spray manifold .M; S / is called totally geodesic if any

geodesic of .M; S / which is tangent to MN at some point is contained in M.
N
A totally geodesic submanifold M N of .M; S / is called auto-parallel if the
homogeneous (nonlinear) parallel translations c W Tc.0/ M ! Tc.1/ M along curves
in the submanifold M N leave invariant the tangent bundle T M N and for every 2
XO 1 .T M/
N the horizontal Berwald covariant derivative rX belongs to XO 1 .T M/.
N
Lemma 1 Let M N be a totally geodesic submanifold in a spray manifold .M; S /.

The following assertions hold:
(a) the spray S induces a spray SN on the submanifold M,
N
(b) MN is an auto-parallel submanifold.
Proof Assume that the manifolds M N and M are k, respectively, n D k C p

1
dimensional. Let .x ; : : : ; x ; x ; : : : ; xn / be an adapted coordinate system, i. e. the
k kC1
submanifold M N is locally given by the equations xkC1 D D xn D 0. We denote

the indices running on the values f1; : : : ; kg or fk C 1; : : : ; ng by ; ; or ; ,
respectively. The differential equation (12.3) of geodesics yields that the geodesic
coefficients G .x; y/ satisfy
G .x1 ; : : : ; xk ; 0; : : : ; 0I y1 ; : : : ; yk ; 0; : : : ; 0/ D 0
identically, hence their derivatives with respect to y1 ; : : : ; yk are also vanishing. It

follows that G D 0 and G D 0 at any .x1 ; : : : ; xk ; 0; : : : ; 0I y1 ; : : : ; yk ; 0; : : : ; 0/.
Hence the induced spray SN on M N is defined by the geodesic coefficients
N .x1 ; : : : ; xk I y1 ; : : : ; yk / D G .x1 ; : : : ; xk ; 0; : : : ; 0I y1 ; : : : ; yk ; 0; : : : ; 0/:

G
(12.12)
The homogeneous (nonlinear) parallel translation c W Tc.0/ M ! Tc.1/ M along

curves in the submanifold M N and the horizontal covariant derivative on M N with
respect to the spray S coincide with the translation and the horizontal covariant
N with respect to the spray SN. Hence the assertions are true.
derivative on M
12.2.1.4 Holonomy
The holonomy group of an Ehresmann connection, or particularly of a spray

manifold is the group generated by parallel translations along loops with respect
to the associated connection, cf. [12, 33], pp. 8286. The holonomy properties of
a spray manifold depend essentially on its curvature properties. This can be easily
understood by considering the geometric construction of the parallel translation.
Let .M; S / be a spray manifold, and let us assume that M is connected.
We choose a fixed base point p 2 M. For each closed piecewise smooth curve
c W 0; 1 ! M through p the parallel translation c W Tp M ! Tp M along
the curve c W 0; 1 ! M is a diffeomorphism of the tangent space Tp M. All
these diffeomorphisms form together the holonomy group at the point p, a subgroup
of the diffeomorphism group of Tp M. Clearly, the holonomy group depends on the
base point p only up to conjugation, and therefore the holonomy groups at different
points of M are isomorphic.
Case K 0. Let us consider a point p 2 M, a tangent vector v 2 Tp M at p, and
an arbitrary closed curve c on M starting from p.
If the curvature identically vanishes, then the horizontal distribution is integrable.
Therefore TTM has a horizontal foliation, that is for every v 2 Tp M there is a
unique n-dimensional submanifold Hv in TM such that Hv \ Tp M D fvg and
the tangent spaces are horizontal. In Fig. 12.3 the level surface corresponding to
v 2 Tp M is represented by dashed lines in tangent manifold TM. The horizontal lift
Fig. 12.3 Trivial holonomy:

R 0
TM
v M
p
Fig. 12.4 Nontrivial

holonomy: R 6 0
TM
(v)
v
v = (v)
M
ch D l c of the curve c starts at v and (because the integrability condition) stays

on the horizontal submanifold Hv . Its endpoint is an element of Hv and also an
element of Tp M that is v. Consequently we can obtain that ch .0/ D ch .1/ D v, and
the holonomy is trivial.
Case K 6 0. Although Fig. 12.4 describing this case looks similar to that of
Fig. 12.3, the situation is quite different. The horizontal distribution is non-integ-
rable and therefore there is no horizontal foliation. Let us consider a point p 2 M
and a vector v 2 Tp M. The smallest integrable distribution Nv at v containing
the horizontal distribution is at least .n C1/-dimensional. The reachable sets in
Tp M is at least one-dimensional. In particular fvg Nv \ Tp M and there are
other elements Tp M reachable from v. That is, there are elements w 2 Tp M and
a horizontal curve ch 2 Nv such that ch .0/ D v, ch .1/ D w. Considering the
projection of ch we can obtain a curve c WD ch such that c.0/ D c.1/ D p and
c .v/ v. Consequently we obtained that the holonomy is nontrivial (cf. [27],
Remark 8.1.3).
12.2.2 Finsler Manifolds

12.2.2.1 Finsler Metric, Its Associated Spray and Parallel Translation
A Minkowski functional on a vector space V is a continuous function F , positively

homogeneous of degree two, i.e. F .y/ D 2 F . y/ if > 0, smooth on VO WD
V n f0g, and for any y 2 VO the symmetric bilinear form gy W V V ! R defined by
1 @2 F . y C su C tv/
gy W .u; v/ 7! gij . y/ui v j D
2 @s @t tDsD0
O then F is said positive

is non-degenerate. If gy is positive definite for any y 2 V,
definite and .V; F / is called positive definite Minkowski space. A Minkowski
functional F is called semi-Euclidean if there exists a symmetric bilinear form h ; i
on V such that gy .u; v/ D hu; vi for any y 2 VO and u; v 2 V. A semi-Euclidean
positive definite Minkowski functional is called Euclidean.
A Finsler manifold is a pair .M; F / of an n-manifold M and a function
F W TM ! R (called Finsler metric, cf. [27]) defined on the tangent bundle of M,
which is smooth on TMO WD TM nf0g and its restriction Fx D F jT M is a Minkowski
x
functional on Tx M for all x 2 M. If the Minkowski functional Fx is positive definite
on Tx M for all x 2 M, then .M; F / is called positive definite Finsler manifold.
A point x 2 M is called (semi-)Riemannian if the Minkowski functional Fx is
(semi-)Euclidean.
We remark that in many applications the metric F is defined and smooth only on
an open cone C M TMnf0g, where C M D [x2M Cx M is a fiber bundle over M such
that each Cx M is an open cone in Tx M nf0g. In such case .M; F / is called singular
(or non y-global) Finsler space (cf. [27]).
The symmetric bilinear form
1 @2 Fx2 . y C su C tv/
gx;y W .u; v/ 7! gij .x; y/ui v j D ; u; v 2 Tx M
2 @s @t tDsD0
is called the metric tensor of the Finsler manifold .M; F /. The Finsler function is
called absolutely homogeneous at x 2 M, if Fx .y/ D jjFx . y/ for all 2 R. If
F is absolutely homogeneous at every x 2 M, then the Finsler manifold .M; F / is
reversible.
The simplest non-Riemannian Finsler metrics are the Randers metrics firstly
studied by Randers in [26]. A Finsler manifold .M; F / is called Randers manifold
p metric F can be expressed in the form F.x; y/ D ax . y/Cbx . y/, where
if the Finsler
ax . y/ D aij .x/yi y j is a Riemannian metric and bx . y/ D bi .x/yi is a nowhere zero
1-form.
The canonical spray of a Finsler manifold .M; F / is locally given by S D
yi @x@ i 2Gi .x; y/ @y@ i , where

1 il @gjl @gjk
Gi .x; y/ WD g .x; y/ 2 k .x; y/ l .x; y/ y j yk : (12.13)
4 @x @x
are the geodesic coefficients. The geodesics of a Finsler manifold .M; F / are the
geodesics of the canonical spray of .M; F / determined by the differential equa-
tions (12.3). The horizontal covariant derivative with respect to the spray associated
to the Finsler manifold .M; F / is called the horizontal Berwald covariant deriva-
tive, cf. Eq. (12.6). The corresponding homogeneous (nonlinear) parallel translation
c W Tc.0/ M ! Tc.1/ M along a curve c.t/, given by Eqs. (12.9) and (12.10), is called
the canonical homogeneous (nonlinear) parallel translation of the Finsler manifold
.M; F /. Since the geodesic coefficients Gi .x; y/ are differentiable functions on the
O D TM n f0g the canonical homogeneous (nonlinear) parallel

slit tangent bundle TM
translation c W Tc.0/ M ! Tc.1/ M induces differentiable maps between the slit
tangent spaces Tc.0/ M n f0g and Tc.1/ M n f0g.
12.2.2.2 Curvature
The Riemannian curvature

tensor field of the Finsler manifold .M; F / is the cur-
vature tensor field v X h ; Y h of the canonical spray manifold of .M; F / is defined
on the pull-back bundle . TM; ; O
N TM/, (cf. Eq. (12.7)). If H TM is integrable,
then the Riemannian curvature is identically zero. According to Eq. (12.8) the
expression of the Riemannian curvature tensor R.x;y/ D Rijk .x; y/dxj dxk @x@ i is
@Gij .x; y/ @Gik .x; y/

Rijk .x; y/ D C Gm
j .x; y/Gkm .x; y/ Gk .x; y/Gjm .x; y/
i m i
@xk @xj
in a local coordinate system. The manifold .M; F / has constant flag curvature
2 R, if for any x 2 M the local expression of the Riemannian curvature is

Rijk .x; y/ D ki gjm .x; y/ym ji gkm .x; y/ym : (12.14)
In this case the flag curvature of the Finsler manifold (cf. [6], Sect. 2.1 pp. 4346)
does not depend either on the point or on the 2-flag.
The Berwald curvature tensor field B.x;y/ D Bijkl .x; y/dxj dxk dxl @x@ i is
@Gijk .x; y/ @3 Gi .x; y/

Bijkl .x; y/ D D : (12.15)
@yl @y j @yk @yl
The mean Berwald curvature tensor field E.x;y/ D Ejk .x; y/dxj dxk is the trace
@3 Gl .x; y/
Ejk .x; y/ D Bljkl .x; y/ D : (12.16)
@y j @yk @yl
@
The Landsberg curvature tensor field L.x;y/ D Lijkl .x; y/dxj dxk dxl @xi
is

L.x;y/ .u; v; w/ D g.x;y/ rw B.x;y/ .u; v; w/; y ; u; v; w 2 Tx M:
According to Lemma 6.2.2, Eq. (6.30), p. 85 in [27], one has for u; v; w 2 Tx M
rw g.x;y/ .u; v/ D 2L.x;y/ .u; v; w/:

Lemma 2 The horizontal Berwald covariant derivative of the tensor field
@
Q.x;y/ D ji gkm .x; y/ym ki gjm .x; y/ym dxj dxk dxl i
@x
vanishes.
Proof For any vector field W 2 X1 .M/ we have rW y D 0 and rW IdTM D 0.
Moreover, since L.x;y/ . y; v; w/ D 0 (cf. Eq. (6.28), p. 85 in [27]) we get the assertion.
12.2.2.3 Projectively Flat Finsler Manifold
A Finsler manifold .D; F / on an open subset D Rn is said to be projectively

flat, if all geodesics of .D; F / are contained in straight lines of the affine space
associated to Rn . A Finsler manifold .M; F / is said to be locally projectively flat,
if for any point in p 2 M there exists a local coordinate map x W U ! Rn of a
neighborhood U M of p such that the Finsler manifold induced by the Finsler
function F on the image x.U/ D D is projectively flat. The space Rn containing D
is called to be projectively related to .M; F /.
Let .M; F / be a locally projectively flat Finsler manifold and .x1 ; : : : ; xn / W U !
D a local coordinate map corresponding to canonical coordinates of the space Rn
which is projectively related to .M; F /. Then the geodesic coefficients (12.13) are
of the form
@P i @2 P i @P i @P i
Gi .x; y/ D P.x; y/yi ; Gik D y C Pki ; Gikl D y C k l C l k ;
@yk @yk @yl @y @y
(12.17)
where P is a 1-homogeneous function in y, called the projective factor of .M; F /,

(cf. [6], p. 63). Clearly, the intersections of 2-planes of Rn with the image D of the
coordinate map .x1 ; : : : ; xn / W U ! D are images of totally geodesic submanifolds
of .M; F /.
Remark 1 The canonical homogeneous parallel translation c W Tc.0/ M ! Tc.1/ M
in a locally projectively flat Finsler manifold .M; F / along curves c.t/ contained
in the domain of the coordinate system .x1 ; : : : ; xn / are linear maps if and only if
the projective factor P.x; y/ is a linear function in y. Hence the nonlinearity in y
of the projective factor implies that the locally projectively flat Finsler manifold is
non-Riemannian.
Locally projectively flat Randers manifolds with constant flag curvature were
classified by Shen in [29]. He proved that any locally projectively flat non-
Riemannian Randers manifold .M; F / with non-zero constant flag curvature has
negative curvature. These metrics can be normalized by a constant factor so that the
curvature is 14 . In this case .M; F / is isometric to the Finsler manifold defined by

the metric function
p
F .x; y/ D C (12.18)
1 jxj2 1 jxj2 1 C ha; xi
on the unit ball Dn Rn , where a 2 Rn is any constant vector with jaj < 1.
According to Lemma 8.2.1 in [6], p.155, the projective factor P.x; y/ can be
computed by the formula
1 @F i
P.x; y/ D y:
2F @xi
An easy calculation yields
p !2
@F jyj2 .jxj2 jyj2 hx; yi2 / hx; yi ha; yi 2
i yi D ;
@x 1 jxj2 1 C ha; xi
hence
p !
1 jyj2 .jxj2 jyj2 hx; yi2 / C hx; yi ha; yi
P.x; y/ D : (12.19)
2 1 jxj2 1 C ha; xi
12.2.3 Finsler Holonomy Represented on the Indicatrix Bundle
The notion of the holonomy group of a Riemannian or Finslerian manifolds is an

adaptation of the corresponding notion of spray manifolds: it is the group generated
by parallel translations along loops with respect to the canonical associated con-
nection. However, the parallel translation leaves invariant the indicatrix bundle, the
holonomy group can be identified by its action on the indicatrix at the initial point.
Hence the holonomy group can be considered as a subgroup of the diffeomorphism
group of the indicatrix (Fig. 12.5a, b).
The holonomy properties of Finsler spaces are essentially different from the
Riemannian one, and it is far from being well understood. The main difficulty comes
from the fact that in the general case the canonical connection of a Finsler manifold
is neither linear nor metrical (that is, the parallel translation is not necessarily
preserving the metric). Only much weaker properties are fulfilled: instead of
the linearity it is only 1-homogeneous, and instead of the metrical property it
is preserving only the norm function. Nonetheless these properties allow us to
consider the parallel translations as maps between the indicatrices and therefore
the holonomy group as a subgroup of the diffeomorphism group of the indicatrix.
a b X1
X0
X0 = X1
Fig. 12.5 (a) Trivial holonomy: the plane. (b) Nontrivial holonomy: sphere
Let .M; F / be an n-dimensional Finsler manifold. The indicatrix Ix M at x 2 M

is a hypersurface of Tx M defined by
Ix M WD fy 2 Tx MI F . y/ D 1g:
If the Finsler manifold .M; F / is positive definite, then the indicatrix Ix M is a

compact hypersurface in the tangent space Tx M, diffeomorphic to the standard .n
1/-dimensional sphere.
In the sequel .IM; ; M/ will denote the indicatrix bundle of .M; F / and i W
IM ,! TM the natural embedding of the indicatrix bundle into the tangent bundle
.TM; ; M/.
The parallel translation c W Tc.0/ M ! Tc.1/ M along a curve c W 0; 1 ! R on
a Finsler manifold .M; F / is defined by the parallel translation of the associated
spray manifold (cf. Sect. 12.2.1.2). It is determined by vector fields X.t/ along c.t/
which are solutions of the differential equation (12.10). Since c W Tc.0/ M ! Tc.1/ M
is a differentiable map between the slit tangent spaces TO c.0/ M and TO c.1/ M preserves
the value of the Finsler function, it induces a map
cI W Ic.0/ M ! Ic.1/ M (12.20)
between the indicatrices. Since the parallel translation is 1-homogeneous, the

parallel translation c is entirely characterized by the map cI : we have C .0/ D 0
and for every non-zero vector v 2 Tc.0/ M we have

I v
c .v/ D jvj c :
jvj
It follows from these observations that the holonomy group Hol.x/ of the spray
manifold associated to a Finsler manifold .M; F / (cf. Sect. 12.2.1.4) is uniquely
determined by its action on the indicatrix in the tangent space Tx M at the point x.
Hence we can formulate
Fig. 12.6 Holonomy

transformation induced on the (v)
indicatrix
v
Definition 1 The holonomy group Hol.x/ of a Finsler space .M; F / at x 2 M

is the subgroup of the group of diffeomorphisms Diff.Ix M/ of the indicatrix Ix M
determined by parallel translation of Ix M along piece-wise differentiable closed
curves initiated at the point x 2 M (Fig. 12.6).
We note that the holonomy group Hol.x/ is a topological subgroup of the regular
infinite dimensional Lie group Diff.Ix M/, but its differentiable structure is not
known in general. (Diff.Ix M/ denotes the group of all C1 -diffeomorphism of Ix M
with the C1 -topology.)
12.3 Diffeomorphism Groups and Their Tangent Algebras
The group Diff.M/ of all smooth diffeomorphisms of a differentiable manifold M

is a regular infinite dimensional Lie group modeled on the vector space Xc .M/
of smooth vector fields on M with compact support. The Lie algebra of the
infinite dimensional Lie group Diff.M/ is the vector space Xc .M/, equipped with
the negative of the usual Lie bracket (c.f. Kriegl and Michor [15], Sect. 43.1,
pp. 454456).
Here we discuss the tangential properties of Lie algebras of vector fields to an
abstract subgroup of the diffeomorphism group of a manifold. The results of this
section will be applied in the following to the investigation of tangent Lie algebras
of the holonomy subgroup of the diffeomorphism group of an indicatrix Ix M and
to the fibered holonomy subgroup of the diffeomorphism group of the indicatrix
bundle I.M/.
Let M be a C1 manifold, let G be a (not necessarily differentiable) subgroup of
the diffeomorphism group Diff1 .M/ and let X1 .M/ be the Lie algebra of smooth
vector fields on M.
Definition 2 A vector field X 2 X1 .M/ is called tangent to the subgroup G of
Diff1 .M/, if there exists a C 1 -differentiable 1-parameter family f
t 2 G gt2.";"/

t
of diffeomorphisms of M such that
0 D Id and @
@t tD0
D X: A Lie subalgebra g of
X1 .M/ is called tangent to G , if all elements of g are tangent vector fields to G .
Unfortunately, it is not true in general that tangent vector fields to a group G generate
a tangent Lie algebra to G . This is why we have to introduce a stronger tangential
property in Definition 4.
Definition 3 A C 1 -differentiable k-parameter family f
.t1 ;:::;tk / 2 Diff1.M/gti 2.";"/
of diffeomorphisms of M is called a commutator-like family if it satisfies the
equations

.t1 ;:::;tk / D Id; whenever tj D 0 for some 1 6 j 6 k:
We remark that any C 1 -differentiable 1-parameter family f

t 2 G gt2.";"/ of
diffeomorphisms of M with
0 D Id is a commutator-like family. Moreover, the
commutator of commutator-like families is commutator-like, and the inverse of a
commutator-like family is commutator-like.
Definition 4 A vector field X 2 X1 .M/ is called strongly tangent to the subgroup G
of Diff1 .M/, if there exists a commutator-like family f
.t1 ;:::;tk / 2 Diff1 .M/gti 2.";"/
of diffeomorphisms satisfying the conditions
(A)
.t1 ;:::;tk / 2 G for all ti 2 ."; "/; 1 6 i 6 k;
@k
1 ;:::;tk /
(B) @t1.t@t D X:
k .0;:::;0/
@k
1 ;:::;tk /
It follows from the commutator-like property that @t1.t@t k .0;:::;0/
is the first non-
necessarily vanishing derivative of the diffeomorphism family f
.t1 ;:::;tk / g at any
point x 2 M, and therefore it determines a vector field. On the other hand, by
parameterizing the commutator like family of diffeomorphism, it can be shown
that if a vector field is strongly tangent to a group G , then it is also tangent to
G . Moreover, we have the following
Theorem 1 Let V be a set of vector fields strongly tangent to the subgroup G of
Diff1 .M/. The Lie subalgebra v of X1 .M/ generated by V is tangent to G .
Proof First, we investigate some properties of vector fields strongly tangent to the
group G .
Lemma 1 Let f .t1 ;:::;th / 2 Diff1 .U/gti 2.";"/ be a C 1 -differentiable h-parameter
commutator-like family of (local) diffeomorphisms on a neighborhood U Rn .
Then

@i1 CCih .t1 ;:::;th /
(i) .x/ D 0; if ip D 0 for some 1 6 p 6 h;
@t1i1 : : : @thih .0;:::;0/

@h . .t1 ;:::;th / /1 @h .t1 ;:::;th /
(ii) .x/ D .x/;
@t1 : : : @th .0;:::;0/ @t1 : : : @th .0;:::;0/

@h .t1 ;:::;th / @ p
h
t;:::; h t/
p
(iii) .x/ D .x/
@t1 : : : @th .0;:::;0/ @t tD0
at any point x 2 U.
Proof Assertions (i) and (ii) can be obtained by direct computation. It follows from
@h h/
(i) that @t1.t:::1 ;:::;t
@th
.x/ is the first non-necessarily vanishing derivative of the
.0;:::;0/
diffeomorphism family f .t1 ;:::;th / g at any point x 2 M. Using
.t1 ;:::;tk / .x/ D x C t1 tk .X.x/ C !.x; t1 ; : : : ; tk // ;
where lim !.x; t1 ; : : : ; tk / D 0 we obtain that

ti !0

@ @ p p
t/ .x/ D C C !.x; : : : ; D X.x/;
p p x t X.x/ k
t; k
t/
@t .
@t tD0
tD0 k k
t;:::;
which proves (iii).

We remark that assertion .iii/ means that any vector field strongly tangent to G is
tangent to G . Now, we generalize a well-known relation between the commutator of
vector fields and the commutator of their induced flows.
Lemma 2 Let f
.s1;:::;sk / g and f .t1 ;:::;tl / g be C 1 -differentiable k-parameter, respec-
tively, l-parameter families of (local) diffeomorphisms defined on a neighborhood
U Rn . Assume that
.s1 ;:::;sk / D Id, respectively, .t1 ;:::;tl / D Id, if some of their
variables is 0. Then the family of (local) diffeomorphisms
.s1 ;:::;sk / ; .t1 ;:::;tl / defined
by the commutator of the group Diff1 .U/ fulfills
.s1 ;:::;sk / ; .t1 ;:::;tl / D Id, if some
of its variables equals 0. Moreover
" #
@kCl
.s1 :::sk / ; .t1 :::tl / @k
.s1 :::sk / @l .t1 :::tl /
.x/ D .0:::0/ ; .x/
@s1 : : : @sk @t1 : : : @tl @s1 : : : @sk @t1 : : : @tl .0:::0/
.0:::0I0:::0/
at any point x 2 U.

Proof The group theoretical commutator
.s1 ;:::;sk / ; .t1 ;:::;tl / of the families of
diffeomorphisms satisfies
.s1 ;:::;sk / ; .t1 ;:::;tl / D Id, if some of its variables equals
0. Hence

@i1 CCik Cj1 CCjl
.s1 ;:::;sk / ; .t1 ;:::;tl /
D 0;
j
@si1 : : : @sik @t 1 : : : @til
1 k 1 l .0;:::;0I0;:::;0/
if ip D 0 or jq D 0 for some index 1 6 p 6 k or 1 6 q 6 l. The families of

diffeomorphisms f
.s1;:::;sl / g, f .t1 ;:::;tl / g, f
.s11;:::;sl / g and f .t1
1 ;:::;tl /
g are the constant
family Id, if some of their variables equals 0. Hence one has

@kCl
.s1 :::sk / ; .t1 :::tl /
.x/ (12.21)
@s1 : : : @sk @t1 : : : @tl .0;:::;0I 0;:::;0/
0 1
@l
.s11 :::sk / 1
@k .t1 :::tl /
.s1 :::sk / .t1 :::tl / .x/
D @ A
@s : : : @s
1 k .0:::0/ @t1 : : : @tl

.0:::0/
0 1

@k @l .t1 :::t /
D @d.
.s :::s / /
.s :::s / .x/
1 1 l
.
.s1 :::sk / .x//A ;
@s1 : : : @sk .0:::0/ 1 k 1 k
@t1 : : : @tl
.0;:::;0/

where d
.s11 ;:::;sk /
.s ;:::;s / .x/ denotes the Jacobi operator of the map
.s11 ;:::;sk / at the
1 k
point
.s1 ;:::;sk / .x/. Using the fact, that f
.s1 ;:::;sk / g is the constant family Id, if some
1
of its variables equals 0, and the relation d.
.0;:::;0/ /
.s1 ;:::;sk / .x/ D Id, we obtain
that (12.21) can be written as
0 1 0 1

@k
.s1
1 :::sk /
@l 1
.t1 :::tl / .x/ @l 1

.t1 :::tl / @k
.s1 :::sk / .x/
d@ A Cd@ A :
@s1 : : : @sk @t1 : : : @tl @t1 : : : @tl @s1 : : : @sk .0;:::;0/
.0:::0/ x .0:::0/ .0:::0/ x
According to assertion (ii) of Lemma 1 the last formula gives

! l !
@k
.s1 :::sk / @l .t1 :::tl / .x/ @ .t1 :::tl / @k
.s1 :::sk / .x/
d d ;
@s1 : : : @sk .0:::0/ @t1 : : : @tl .0:::0/ @t1 : : : @tl .0:::0/ @s1 : : : @sk .0:::0/
x x
which is the Lie bracket of vector fields

" #
@l .t1 ;:::;tl / @k
.s1 ;:::;sk /
; W U ! Rn :
@t1 : : : @tl .0;:::;0/ @s1 : : : @sk .0;:::;0/
The previous lemma gives for 1-parameter families f

t 2 G gt2.";"/ and f t 2
G gt2.";"/ of diffeomorphisms of M with
0 D 0 D Id the relation
" #
@2 @
s @l t

s ; t D ; : (12.22)
@s@t sD0;tD0 @s sD0 @t tD0
Lemma 3 Any Lie subalgebra of X1 .M/ algebraically generated by strongly

tangent vector fields to the group G has a basis consisting of vector fields strongly
tangent to the group G .
Proof Let V be a set of strongly tangent vector fields to the group G and v the
Lie algebra algebraically generated by V . The iterated Lie brackets of vector fields
belonging to V linearly generate the vector space v. It follows from Lemma 2 that
these iterated Lie brackets of vector fields are strongly tangent to the group G . Hence
v is linearly generated by vector fields strongly tangent to G .
Lemma 4 Linear combinations of vector fields tangent to G are tangent to G .
Proof If X and Y are vector fields tangent to G then there exist C 1 -differentiable
1-parameter families of diffeomorphisms f
t 2 G g and f t 2 G g such that
@ @

0 D 0 D Id;
t D X; t D Y:
@t tD0 @t tD0
Hence the 1-parameter families of diffeomorphisms f

t tg and f
ct g satisfy
@ @
XCY D .
t t /; cX D
.c t/ ; for all c 2 Rn ;
@t tD0 @t tD0
which proves the assertion.
Lemmas 14 prove Theorem 1.
12.3.1 Diffeomorphism Group of Compact Manifolds
If K is a compact manifold, then the group Diff1 .K/ of diffeomorphisms is an

infinite dimensional Lie group belonging to the class of Frchet Lie groups. The
Lie algebra of Diff1 .K/ is the Lie algebra X1 .K/ of smooth vector fields on K
endowed with the negative of the usual Lie bracket of vector fields. The Frchet
Lie group Diff1 .K/ is modeled on the locally convex topological Frchet vector
space X1 .K/. A sequence f fj gj2N X1 .K/ converges to f in the topology of
X1 .K/ if and only if the functions fj and all their derivatives converge uniformly
to f , respectively to the corresponding derivatives of f . We note that the difficulty
of the theory of Frchet manifolds comes from the fact that the inverse function
theorem and the existence theorems of differential equations, which are well known
for Banach manifolds, are not true in this category. These problems have led to the
concept of regular Frchet Lie groups (cf. [25] Chapter III, [15] Chapter VIII). The
distinguishing properties of regular Frchet Lie groups can be summarized as (a)
the existence of the smooth exponential map from the Lie algebra of the Frchet
Lie groups to the group itself, (b) the existence of product integrals, which produces
the convergence of some approximation methods for solving differential equations
(cf. Sect. III.5. in [25], pp. 8389). J. Teichmann gave a detailed discussion of these
properties in [32].
Proposition 1 If a Lie subalgebra g of the Lie algebra X1 .K/ of smooth vector
fields on a compact manifold K is tangent to a subgroup G of the diffeomorphism
group Diff1 .K/ of K, then the group generated by the exponential image exp.g/ of
g is contained in the topological closure G of G in Diff1 .K/.

Proof Let us denote by exp.g/ the group generated by the exponential image of g.
For any element X 2 g there exists a C 1 -differentiable 1-parameter family f.t/ 2
G gt2R of diffeomorphisms of the manifold K such that

@.s/
.0/ D Id and D X:
@s sD0
Then, considering .t/ as hair and using the argument of Corollary 5.4. in [25],
p. 85, we get that

t n n t t o
D G; n D 1; 2 : : :
n t2R n n t2R
as a sequence of Diff1 .K/ converges uniformly in all derivatives to exp.tX/. It

follows that we have fexp.tX/I t 2 Rg G for any X 2 g/ and therefore
exp.g/ G : Naturally, if for the generated group exp.g/ ; then the containing

relation is preserved, that is exp.g/ G , which proves the proposition.
12.4 Curvature Algebra
12.4.1 Curvature Vector Fields at a Point
Definition 5 A vector field 2 X1 .IM/ on the indicatrix bundle IM is a

curvature vector field of the Finsler manifold .M; F /, if there exist vector fields
X; Y 2 X1 .M/ on the manifold M such that D r.X; Y/, where for every x 2 M
and y 2 Ix M we have
r.X; Y/.x; y/ WD R.x;y/ .Xx ; Yx /: (12.23)
If x 2 M is fixed and X; Y 2 Tx M, then the vector field y ! r.X; Y/.x; y/ on Ix M is

a curvature vector field at the point x.
The Lie algebra R.M/ of vector fields generated by the curvature vector fields of
.M; F / is called the curvature algebra of the Finsler manifold .M; F /. For a fixed
x 2 M the Lie algebra Rx of vector fields generated by the curvature vector fields at
x is called the curvature algebra at the point x.
In this section we investigate the properties of the curvature vector fields and of
the curvature algebra at a fixed point x 2 M. The curvature vector fields D rx .X; Y/
are tangent to the indicatrix

g.x;y/ y; R.x;y/ .ly .X/; ly .Y/ D 0; for any y; X; Y 2 Tx M
according to [27], Eq. (10.9). In the sequel we investigate the tangential properties
of the curvature algebra to the holonomy group of the canonical connection r of a
Finsler manifold.
Proposition 2 Any curvature vector field at x 2 M is strongly tangent to the
holonomy group Hol.x/.
Proof Indeed, let us consider the curvature vector field WD rx .X; Y/ 2 X.Ix M/
corresponding to the directions X; Y 2 Tx M and let X; O YO 2 X.M/ be commuting
O Y
vector fields, i.e. X; O D 0 such that XO x D X, YO x D Y. By the geometric construction,
the flows f
t g and f s g of XO and YO are commuting, that is
s t t
s . For any
sufficiently small s; t 2 R we can consider the curve Ps;t defined as follows:
8
u .x/; 0 6 u 6 t;

<
ut . t .x//; t 6 u 6 t C s;
Ps;t .u/ D

1
u.tCs/ .
s . t .x///; t C s 6 u 6 2t C s;

: 1 1

u.2tCs/ . t .
s . t .x////; 2t C s 6 u 6 2t C 2s:
Because of the commuting property of the flows f

t g and f s g the curves Ps;t
are closed parallelograms: their initial and final point at u D 0 and u D 2t C 2s
are the same x 2 M. Consequently, the parallel translation s;t W Tx M ! Tx M along
the parallelogram Ps;t is a holonomy element for every small value of t; s 2 R (see
Fig. 12.7).
On the other hand, using the geometric construction of parallel translation
presented in Sect. 12.2.1.2, we know that the flows f
th g and f sh g of the horizontal
O and l.Y/
lifts l.X/ O can be considered as parallel translations along integral curves of
XO and Y,O respectively. They can be considered as fiber preserving diffeomorphisms
of the bundle IM for any t; s 2 R. Then the commutator
s;t D
sh ; t
h
D
s
h
h
t
sh h
t W IM ! IM
is also a fiber preserving diffeomorphism of the bundle IM for any t; s 2 R.

Therefore for x 2 M the restriction

s;t .x/ D s;t Ix M W Ix M ! Ix M
Fig. 12.7 Parallel translation

along a parallelogram M
s,t v
v s
x t
Fig. 12.8 Non-zero

curvature vector field
TM
v
s,t v
v h
s
th
M
s,t v
v s
x t
to the fiber Ix M is a 2-parameter C1 -differentiable family of diffeomorphisms

contained in the holonomy group Hol.x/ such that

@2
s;0 .x/ D Id; 0;t .x/ D Id; and s;t .x/ D rx .X; Y/;
@t@s sD0;tD0
which proves that the curvature vector field D rx .X; Y/ is strongly tangent to the
holonomy group Hol.x/ and hence we obtain the assertion.
In order to enlighten the construction and the geometric meaning of the curvature
vector field D rx .X; Y/ we consider Fig. 12.8 which can be seen as the extension
of Fig. 12.7. Here we can present not only the geometric objects at the level of the
manifold M but at the level of the tangent manifold TM too. We remark that, as it
is usual, the tangent vectors of M are represented as arrows at the level of M, but
they are represented as points at the level of the tangent space TM. For example,
the vectors v and s;t v are represented as arrows at x on M and points above x in TM.
The gray vectors at the level of M represent the elements of the parallel vector field
V along the parallelogram Ps;t with the initial condition Vx D v. The gray dots are
the points in TM corresponding to the elements of V. These dots lie on the curves
of the flows
th , sh because V is a parallel field and these flows correspond to the
parallel translations along integral curves of XO and Y, O respectively. As the picture
shows, the parallel translation along the parallelogram Ps;t of a vector v 2 Tx M
can be obtained by following in TM the flows
th , sh ,
t h h
, s above Ps;t . The
indicatrix, or unite ball, at x 2 M is represented by the oval above x. Since the
parallel translation preserves the norm, if v 2 Ix M, then s;t v 2 Ix M. Therefore
t ! 12 t;t .v/ is a curve in Ix M. Its tangent vector at t D 0 is .v/ which is therefore
a tangent vector of Ix M.
We remark that in the case, when the curvature is identically zero, the horizontal
lifts of commuting vector fields are also commuting vector fields. Therefore one
obtains
sh th th
sh and s;t D
sh ; th W IM ! IM is the identity
transformation. In that case s;t v v is a constant map and therefore its derivative is
Fig. 12.9 Vanishing

curvature vector field
Ip TM
v h
s
th
M
v
s
x t
zero, that is v D 0. Geometrically that means that the horizontal lifts of the closed
parallelograms Ps;t are closed parallelograms. See Fig. 12.9.
Theorem 1 The curvature algebra Rx at a point x 2 M of a Finsler manifold
.M; F / is tangent to the holonomy group Hol.x/.
If .M; F / is a positive
definite
Finsler manifold, then the group generated by the
exponential image exp Rx is a subgroup of the topological closure of the holonomy
group Hol.x/.
Proof Since by Proposition 2 the curvature vector fields are strongly tangent to
Hol.x/ and the curvature algebra Rx is algebraically generated by the curvature
vector fields, the first assertion follows from Theorem 1. The second assertion is a
consequence of Proposition 1.
Proposition 3 The curvature algebra Rx of a Riemannian manifold .M; g/ at any
point x 2 M is isomorphic to the linear Lie algebra over the vector space Tx M
generated by the curvature operators of .M; g/ at x 2 M.
Proof The curvature tensor field of a Riemannian manifold given by Eq. (12.7) is
linear with respect to y 2 Tx M and hence
@
R.x;y/ .; / D .Rx .; //kl yl ;
@yk
where Rx .; //kl is the matrix of the curvature operator Rx .; /W Tx M ! Tx M

with respect to the natural basis @x@1 jx ; : : : ; @x@n jx : Hence any curvature vector field
rx .; /. y/ with ; 2 Tx M has the shape rx .; /. y/ D .Rx .; //kl yl @y@k . It follows
that the flow of rx .; /. y/ on the indicatrix Ix M generated by the vector field
rx .; /. y/ is induced by the action of the linear 1-parameter group exp tRx .; //
on Tx M, which implies the assertion.
Since for Finsler surfaces of non-vanishing curvature the curvature vector fields
form a one-dimensional vector space and hence the generated Lie algebra is also
one-dimensional, we have
Remark 2 The curvature algebra of Finsler surfaces is at most one-dimensional.
12.4.2 Constant Curvature
Now, we consider a Finsler manifold .M; F / of non-zero constant curvature. In this

case for any x 2 M the curvature vector field rx .X; Y/. y/ has the shape (cf. (12.14))
@
r.X; Y/. y/ D c ji gkm . y/ym ki gjm . y/ym X j Y k i ; 0 c 2 R:
@y

Putting yj D gjm . y/ym we can write r.X; Y/. y/ D c ji yk ki yj X j Y k @y@ i : Any
linear combination of curvature vector fields has the form
@
r.A/. y/ D Ajk ji yk ki yj ;
@yi
where A D Ajk @x@ j ^ @

@xk
2 Tx M ^ Tx M is arbitrary bivector at x 2 M.
Lemma 1 Let .M; F / be a Finsler manifold of non-zero constant curvature. The
curvature algebra Rx at any point x 2 M satisfies
n.n 1/
dim Rx ; (12.24)
2
where n D dim M.
Proof Let us consider the curvature vector fields rjk D rx . @y@ j ; @y@k /. y/ at a fixed
point x 2 M. If a linear combination
@ @ @
Ajk rjk D Ajk .ji yk ki yj / D .Aik yk Aji yj / i D 2Aik yk i
@y i @y @y
of curvature vector fields rjk with constant coefficients Ajk D Akj satisfies Ajk rjk D
0 for any y 2 Tx M then one has the linear equation Aik yk D 0 for any fixed index
i. Since the covector fields y1 ; : : : ; yn are linearly independent we obtain Ajk D 0
for all j; k 2 f1; : : : ; ng. It follows that the curvature vector fields rjk are linearly
independent for any j < k and hence dim Rx > n.n1/ 2
.
Corollary 1 Let .M; g/ be a Riemannian manifold of non-zero constant curvature

with n D dim M. The curvature algebra Rx at any point x 2 M is isomorphic to the
orthogonal Lie algebra o.n/.
Proof The holonomy group of a Riemannian manifold is a subgroup of the
orthogonal group O.n/ of the tangent space Tx M and hence the curvature algebra
Rx is a subalgebra of the orthogonal Lie algebra o.n/. Hence the previous assertion
implies the corollary.
Theorem 2 Let .M; F / be a Finsler manifold of non-zero constant curvature with
dim M > 2. If the point x 2 M is not (semi-)Riemannian, then the curvature algebra
Rx at x 2 M satisfies
n.n 1/
dim Rx > where n D dim M: (12.25)
2
Proof We assume dim Rx D n.n1/ 2

. For any constant skew-symmetric matrices
fAjk g and fB jk g the Lie bracket of vector fields Aik yk @y@ i and Bik yk @y@ i has the
shape Cik yk @y@ i , where fCik g is a constant skew-symmetric matrix, too. Using the
homogeneity of ghl we obtain
@yh @ghl
D m yl C ghm D ghm (12.26)
@y m @y
and hence

@ @ @yh @yh @
Amk yk m ; Bih yh i D Amk Bih m Bmk Aih m yk i
@y @y @y @y @y
@ @
D Bih ghm Amk Aih ghm Bmk yk i D Cik yk i :
@y @y
ij j
Particularly, for the skew-symmetric matrices Eab D ai b bi aj , a; b 2 f1; : : : ; ng,
we have

ij @ @ ih @ @
Eab yj i ; Ecd yl k D Ecd
kl mk
ghm Eab Eab
ih mk
ghm Ecd yk i D im ab;cd ym ;
@y @y @y @yi
ij ij ji ij ij ij
where the constants ab;cd satisfy ab;cd D ab;cd D ba;cd D ab;dc D cd;ab :
Putting i D a and computing the trace for these indices we obtain
.n 2/.gbd yc gbc yd / D lb;cd yl ; (12.27)
where lb;cd WD ilib;cd . The right-hand side is a linear form in variables y1 ; : : : ; yn .

According to the identity (12.27) this linear form vanishes for yc D yd D 0, hence
.c/ 1
lb;cd D 0 for l c; d. Denoting bd WD c
n2 b;cd
(no summation for the index c)
we get the identities
.c/ .d/
gbd yc gbc yd D bd yc bc yd .no summation for c and d/:

Putting yd D 0 we obtain gbd yd D0 D bd for any c d. It follows bd is
.c/ .c/
.c/
independent of the index c . d/. Defining bd WD bd with some c . d/ we
obtain from (12.27) the identity
gbd yc gbc yd D bd yc bc yd (12.28)
for any b; c; d 2 f1; : : : ; ng. We have
cd yb cb yd D .gbd yc gbc yd /.gdb yc gdc yb / D .bd yc bc yd /.db yc dc yb /:
which implies the identity
.cd yb cb yd / C .db yc dc yb / C .bc yd bd yc /

D .cd dc / yb C .db bd / yc C .bc cb / yd D 0: (12.29)
Since dim M > 2, we can consider three different indices b; c; d and we obtain from
the identity (12.29) that bc D cb for any b; c 2 f1; : : : ; ng.
By derivation the identity (12.28) we get
@gbd @gbc
yc yd C gbd ca gbc da D bd ca bc da :
@ya @ya
Using (12.26) we obtain

@ya @gbd @gbc
yc yd C gbd gcq gbc gdq
@yq @ya @ya
@gbd @gbc
D yc q yd C gbd gcq gbc gdq D bd gcq bc gdq :
@yq @y
Since

@gbd @gbc
y c y d y D0
b
@yq @yq
we get the identity
yd gcq yc gdq D bd yb gcq bc yb gdq :

Multiplying both sides of this identity by the inverse fgqr g of the matrix fgcq g and
taking the trace with respect to the indices c; r we obtain the identity
.n 1/ yd D .n 1/bd yb :
Hence we obtain that gbd yb D bd yb and hence gbd D bd ; which means that the
point x 2 M is (semi-)Riemannian. From this contradiction follows the assertion.
Corollary 2 The curvature algebra Rx at a point x 2 M of a Finsler manifold
.M; F / of non-zero constant curvature satisfies
n.n 1/
dim Rx D ; where n D dim M; (12.30)
2
if and only if n D 2 or the point x 2 M is (semi-)Riemannian.
Theorem 3 Let .M; F / be a positive definite Finsler manifold of non-zero constant
curvature with n D dim M > 2. The holonomy group of .M; F / is a compact Lie
group if and only if .M; F / is a Riemannian manifold.
Proof We assume that the holonomy group of a Finsler manifold .M; F / of non-
zero constant curvature with dim M > 3 is a compact Lie transformation group
on the indicatrix Ix M. The curvature algebra Rx at a point x 2 M is tangent to
the holonomy group Hol.x/ and hence dim Hol.x/ dim Rx . If there exists a not
(semi-)Riemannian point x 2 M, then dim Rx > n.n1/ 2 . The .n 1/-dimensional
indicatrix Ix M at x can be equipped with a Riemannian metric which is invariant
with respect to the compact Lie transformation group Hol.x/. Since the group of
isometries of an n 1-dimensional Riemannian manifold is of dimension at most
n.n1/
2 (cf. Kobayashi [11], p. 46) we obtain a contradiction, which proves the
assertion.
Since the holonomy group of a Landsberg manifold is a subgroup of the isometry
group of the indicatrix, we obtain that any Landsberg manifold of non-zero constant
curvature with dimension > 2 is Riemannian (c.f. Numata [24]).
We can summarize our results as follows:
Theorem 4 The holonomy group of any non-Riemannian positive definite Finsler
manifold of non-zero constant curvature with dimension > 2 does not occur as the
holonomy group of any Riemannian manifold.
12.4.3 Infinite Dimensional Curvature Algebra
Let us consider the singular (non y-global) Finsler manifold .H3 ; F /, where H3 is
the three-dimensional Heisenberg group and F is a left-invariant Berwald-Mor
metric (c.f. [27], Example 1.1.5, p. 8).

1 x1 x2
The group H3 can be realized as the Lie group of matrices of the form 0 1 x3 ;
0 0 1
1 2 3 3
where x D .x ; x ; x / 2 R and hence the multiplication can be written as
.x1 ; x2 ; x3 / . y1 ; y2 ; y3 / D .x1 C y1 ; x2 C y2 C x1 y3 ; x3 C y3 /:
The vector 0 D .0; 0; 0/ 2 R3 gives the

unit element of H3 . The Lie algebra h3 D
0 a1 a2
T0 H3 consists of matrices of the form 0 0 a3 ; corresponding to the tangent vector
0 0 0
a D a1 @x@1 Ca2 @x@2 Ca3 @x@3 at the unit element 0 2 H3 . A left-invariant Berwald-Mor
Finsler metric F is induced by the (singular) Minkowski functional F0 W h3 ! R:
2
F0 .a/ WD a1 a2 a3 3
of the Lie algebra in the following way: if y D . y1 ; y2 ; y3 / is a tangent vector at

x 2 H3 , then
F .x; y/ WD F0 .x1 y/:
The coordinate expression of the singular (non y-global) Finsler metric F is

2
F .x; y/ D y1 y2 x1 y3 y3 3 :
Since F is left-invariant, the associated geometric structures (connection,

geodesics, curvature) are also left-invariant and the curvature algebras at different
points are isomorphic. Using the notation

@ @
rx .i; j/ D rx ; ; i; j D 1; 2; 3;
@xi @xj
for curvature vector fields, a direct computation yields
2 2 2 3
!
1 5y1 y3 @ y1 y3 3 x1 y3 C y2 @ 4y1 y3 @
rx .1; 2/ D C C ;
4 .x1 y3 y2 /3 @y1 . y2 x1 y3 /3 @y2 . y2 x1 y3 /3 @y3
2 2
1 y1 y3 6 x1 y3 11 y2 @ 4y1 y3 x1 2 x1 y3 3 y2 @
rx .1; 3/ D C
4 .x1 y3 y2 /3 @y1 . y2 x1 y3 /3 @y2
2 !
y1 y3 7 x1 y3 11 y2 @
C ;
. y2 x1 y3 /3 @y3
3 2 2 2
!
1 4y1 y3 @ y1 y3 6 x1 y3 y2 @ 5y1 y3 @
rx .2; 3/ D C C :
4 .x1 y3 y2 /3 @y1 . y2 x1 y3 /3 @y2 . y2 x1 y3 /3 @y3
The curvature vector fields r0 .i; j/; i; j D 1; 2; 3; at the unit element 0 2 H3 generate
k m
y1 y3
the curvature algebra r0 . Let us denote Y k;m WD kCm1 ; k; m 2 N, and consider the
y2
vector fields

1 2 3 1 kC1;m @
2 k;m @ 3 k;mC1 @

A k;m
.a ; a ; a / D a Y C a Y C a Y ; (12.31)
@y1 0 @y2 0 @y3 0
with .a1 ; a2 ; a3 / 2 R3 and k; m 2 N. Then the curvature vector fields r0 .i; j/ at

0 2 H3 can be written in the form
1 1;2 1 1
r0 .1; 2/ D A .5; 1; 4/; r0 .1; 3/ D A1;1 .11; 0; 11/; r0 .2; 3/ D A2;1 .4; 1; 5/:
4 4 4
Proposition 4 The curvature algebra rx at any point x 2 M is a Lie algebra of

infinite dimension.
Proof Since the Finsler metric is left-invariant, the curvature algebras at different
points are isomorphic. Therefore it is enough to prove that the curvature algebra r0
at 0 2 H3 has infinite dimension. We prove the statement by contradiction: let us
suppose that r0 is finite dimensional.
A direct computation shows that for any .a1 ; a2 ; a3 /; .b1 ; b2 ; b3 / 2 R3 one has
k;m 1 2 3 p;q 1 2 3
A .a ; a ; a /; A .b ; b ; b / D AkCp;mCq .c1 ; c2 ; c3 /
with some .c1 ; c2 ; c3 / 2 R3 . It follows that any iterated Lie bracket of curvature
vector fields r0 .i; j/; i; j D 1; 2; 3; has the shape (12.31) and hence there exists a
basis of the curvature algebra r0 of the form fAki ;mi .a1i ; a2i ; a3i /gNiD1 , where N 2 N is
the dimension of r0 . We can assume that f.ki ; mi /gNiD1 forms an increasing sequence,
i.e. .k1 ; m1 / 6 .k2 ; m2 / 6 6 .kN ; mN / holds with respect to the lexicographical
ordering of N N. We can consider the vector fields
4
r0 .1; 3/ D A1;1 .1; 0; 1/; 4r0 .1; 2/ D A1;2 .5; 1; 4/; 4r0 .2; 3/ D A2;1 .4; 1; 5/
11
as the first three members of this sequence. Hence 1 6 kN ; mN and
1;1
A .1; 0; 1/; AkN ;mN .a1N ; a2N ; a3N / D A1CkN ;1CmN .c1 ; c2 ; c3 /
belongs to r0 , too, where c1 D .kN mN 1/a1N C 2a2N a3N , c2 D .kN mN /a2N

and c3 D a1N 2a2N C .kN mN C 1/a3N . Since kN < 1 C kN , mN < 1 C mN we have
c1 D c2 D c3 D 0 and hence the homogeneous linear system
0 D .kN mN 1/a1N C 2a2N a3N ;

0 D .kN mN /a2N ;
0 D a1N 2a2N C .kN mN C 1/a3N
has a solution .a1N ; a2N ; a3N / .0; 0; 0/. It follows that kN D mN .

Similarly, computing the Lie bracket

0 D A1;2 .5; 1; 4/; AkN ;kN .a1N ; a2N ; a3N / D A1CkN ;2CkN .d1 ; d2 ; d3 /
Since kN < 1 C kN < 2 C kN we have d1 D d2 D d3 D 0 giving the homogeneous

linear system
0 D.3kN C 5/a1N 15a2N C 10a3N ;

0 D a1N C .3 3kN /a2N 2a3N ;
0 D 4a1N C 12a2N .3kN C 8/a3N
for .a1N ; a2N ; a3N /. The determinant of this system vanishes only for kN D 0 which is
a contradiction.
Corollary 3 The holonomy group of the y-singular Finsler manifold .H3 ; F / has
an infinite dimensional tangent Lie algebra.
We remark here that it remains an interesting open question: Is there a nonsingu-
lar (y-global) Finsler manifold whose curvature algebra is infinite dimensional?
12.5 Holonomy Algebra
12.5.1 Fibered Holonomy Group
Now, we introduce the notion of the fibered holonomy group of a Finsler manifold
.M; F / as a subgroup of the diffeomorphism group of the total manifold IM of
the bundle .IM; ; M/ and apply our results on tangent vector fields to an abstract
subgroup of the diffeomorphism group to the study of tangent Lie algebras to the
fibered holonomy group.
Definition 6 The fibered holonomy group Holf .M/ of .M; F / consists of fiber
preserving diffeomorphisms 2 Diff1 .IM/ of the indicatrix bundle .IM; ; M/
such that for any p 2 M the restriction p D jIp M 2 Diff1 .Ip M/ belongs to the
holonomy group Hol. p/.
We note that the holonomy group Hol. p/ and the fibered holonomy group Holf .M/
are topological subgroups of the infinite dimensional Lie groups Diff1 .Ip M/ and
Diff1 .IM/, respectively.
The definition of strongly tangent vector fields yields
Remark 3 A vector field 2 X1 .IM/ is strongly tangent
to the fibered
holonomy
group Holf .M/ if and only if there exists a family .t1 ;:::;tk / IM ti 2.";"/ of fiber
preserving diffeomorphisms of the bundle .IM; ; M/ such that for any indicatrix

Ip the induced family .t1 ;:::;tk / Ip M ti 2.";"/ of diffeomorphisms is contained in the

holonomy group Hol. p/ and Ip M is strongly tangent to Hol. p/.

Since .t1 ;:::;tk / . p/ p and ./ D 0 for every p 2 U, we get the
Corollary 4 Strongly tangent vector fields to the fibered holonomy group Holf .M/
are vertical vector fields. If 2 X1 .IM/ is strongly tangent to Holf .M/ then its
restriction p WD to any indicatrix Ip is strongly tangent to the holonomy group
Ip
Hol. p/.
Let now U M be an open neighborhood in M diffeomorphic to Rn and consider
the Finsler manifold .U; F jU / induced on U. Now we prove that the first assertion
of Theorem 1 on the tangential property to the holonomy group of curvature vector
fields at a point can be extended to curvature vector fields defined on the full the
indicatrix bundle of .U; F jU / .
Proposition 5 Any curvature vector field 2 X1 .IU/ on the indicatrix bundle
of the Finsler manifold .U; F jU / is strongly tangent to the fibered holonomy group
Holf .U/.
Proof Since U is diffeomorphic to Rn we can identify the manifold U with the

vector space Rn . Let D r.X; Y/ 2 X1 .IRn / be a curvature vector field with X; Y 2
X1 .Rn /. According to Proposition 2 its restriction Ip Rn to any indicatrix Ip Rn is
strongly tangent to theholonomy groups Hol. p/. We have to prove that there exists
a family .t1 ;:::;tk / IRn ti 2.";"/ of fiber preserving diffeomorphisms of the indicatrix
bundle .IRn ; ; Rn / such that for any p 2 Rn the family of diffeomorphisms induced
on the indicatrix Ip is contained in Hol. p/ and Ip Rn is strongly tangent to Hol. p/.
For any p 2 Rn and 1 < s; t < 1 let .sXp ; tYp / be the parallelogram in
R determined by the vertexes p; p C sXp ; p C sXp C tYp ; p C tYp 2 Rn and let
n
.sXp;tYp / W Ip ! Ip denote the (nonlinear) parallel translation of the indicatrix

Ip along the parallelogram .sXp ; tYp / with respect to the associated homogeneous
(nonlinear) parallel translation of the Finsler manifold .Rn ; F /. Clearly we have
.sXp;tYp / D IdIRn , if s D 0 or t D 0 and

@2 .sXp;tYp /
D p for every p 2 Rn :
@s@t
.s;t/D.0;0/
Since .sXp ; tYp / is a differentiable field of parallelograms in Rn , the maps

.sXp;tYp / , p 2 Rn , 0 < s; t < 1, define a 2-parameter family of fiber preserving
the indicatrix bundle IR . The diffeomorphisms induced by the
n
diffeomorphisms
of
family .sXp;tYp / s;t2.1;1/ on any indicatrix Ip are contained in Hol. p/. Hence the
vector field 2 X1 .Rn / is strongly tangent to the fibered holonomy group Holf .U/
and the assertion is proved.
Corollary 5 If M is diffeomorphic to Rn , then the curvature algebra R.M/ of
.M; F / is tangent to the fibered holonomy group Holf .M/.
12.5.2 Infinitesimal Holonomy Algebra
The following assertion shows that similarly to the Riemannian case, the curvature
algebra can be extended to a larger tangent Lie algebra containing all horizontal
Berwald covariant derivatives of the curvature vector fields.
Proposition 6 If 2 X1 .IM/ is strongly tangent to the fibered holonomy group
Holf .M/ of .M; F /, then its horizontal Berwald covariant derivative rX along any
vector field X 2 X1 .M/ is also strongly tangent to Holf .M/.
Proof Let be the (nonlinear) parallel translation along the flow ' of the vector
field X, i.e. for every p 2 M and t 2 ."p ; "p / the map t . p/W Ip M ! I't . p/ M is the
(nonlinear) parallel translation along the integral curve of X. If f.t1 ;:::;tk / gti 2.";"/
is a C 1 -differentiable k-parameter family f.t1 ;:::;tk / gti 2.";"/ of fiber preserving
diffeomorphisms of the indicatrix bundle .IM; jM ; M/ satisfying the conditions
of Definition 4, then the commutator
.t1 ;:::;tk / ; tkC1 WD .t1

1 ;:::;tk /
.tkC1 /1 .t1 ;:::;tk / tkC1

of the group Diff1 IM fulfills .t1 ;:::;tk / ; tkC1 D Id, if some of its variables equals
0. Moreover

@kC1 .t1 :::tk / ; .tkC1 /
D ; X h (12.32)
@t1 : : : @tkC1
.0:::0/

at any point of M, which shows that the vector field ; X h is strongly tangent to
Holf .M/. Moreover, since the vector field is vertical, we have hX h ; D 0, and
using rX WD X h ; we obtain
; X h D X h ; D vX h ; D rX
which yields the assertion.

Definition 7 Let hol .M/ be the smallest Lie algebra of vector fields on the
indicatrix bundle IM satisfying the properties
(i) any curvature vector field belongs to hol .M/,
(ii) if ; 2 hol .M/, then ; 2 hol .M/,
(iii) if 2 hol .M/ and X 2 X1 .M/, then the horizontal Berwald covariant
derivative rX also belongs to hol .M/.
The Lie algebra hol .M/ X1 .IM/ is called the infinitesimal holonomy algebra
of the Finsler manifold .M; F /.
Remark 4 The infinitesimal holonomy algebra hol .M/ is invariant with respect to
the horizontal Berwald covariant derivation, i.e.
2 hol .M/ and X 2 X1 .M/ ) rX 2 hol .M/: (12.33)
Let U M be an open neighborhood in M diffeomorphic to Rn and let .U; F jU /

be the Finsler manifold .U; F jU / induced on U. The results of this sections yield
the following
Theorem 1 The infinitesimal holonomy algebra hol .U/ of the Finsler manifold
.U; F jU / is tangent to the fibered holonomy group Holf .U/.
Let hol .M/ X1 .IM/ be the infinitesimal holonomy algebra of the Finsler
manifold .M; F / and let p be a given point in M.

Definition 8 The Lie algebra hol . p/ WD p I 2 hol .M/ X1 .Ip M/ of
vector fields on the indicatrix Ip M is called the infinitesimal holonomy algebra at
the point p 2 M.
Clearly, for any p 2 M the curvature algebra Rp at p 2 M is contained in the
infinitesimal holonomy algebra hol . p/ at p 2 M.
The following assertion is a direct consequence of the definition. It shows that
the infinitesimal holonomy algebra at a point p of .M; F / can be calculated in a
neighborhood of p.
Remark 5 Let .U; F jU / be an open submanifold of .M; F / such that U M is
diffeomorphic to Rn and let p 2 U. The infinitesimal holonomy algebras at p of the
Finsler manifolds .M; F / and .U; F jU / coincide.
Now, we can prove the following
Theorem 2 The infinitesimal holonomy algebra hol . p/ at a point p 2 M of the
Finsler manifold .M; F / is tangent to the holonomy group Hol. p/.
If .M; F / is a positive definite Finsler manifold, then the group generated by
the exponential image exp hol . p/ is a subgroup of the topological closure of the
Proof Let U M be an open submanifold of M, diffeomorphic to Rn and

p 2 M. According to the previous remark we have hol . p/ WD p I 2
containing
holf .U/ . Since the fibered holonomy algebra holf .U/ is tangent to the fibered
holonomy group Holf .U/ we obtain the first assertion. The second assertion is a
consequence of Proposition 1.
12.5.3 The Berwald Translate
Let x.t/, 0 6 t 6 a be a smooth curve joining the points q D x.0/ and p D x.a/ in the
Finsler manifold .M; F /. If y.t/ D t y.0/ 2 Ix.t/ M is a parallel vector field along
x.t/, 0 6 t 6 a, where t W Iq M ! Ix.t/ M denotes the homogeneous (nonlinear)

dyi .t/ @
parallel translation, then we have DxP y.t/ WD dt C Gij .x.t/; y.t//Pxj .t/ @xi
D 0:
Considering a vector field on the indicatrix Iq M, the map a W . p; y/ 7!a1
a . y.a// gives a vector field on the indicatrix Ip M. Hence we can formulate
Lemma 1 For any vector field 2 hol .q/ X1 .Iq M/ in the infinitesimal
holonomy algebra at q the vector field a a1 2 X1 .Ip M/ is tangent to the
Proof Let f
t 2 Hol.q/gt2.";"/ be a C 1 -differentiable 1-parameter family of
diffeomorphisms of Iq M belonging to the holonomy group Hol.q/ and satisfying

t
the conditions
0 D Id, @
@t tD0 D : Since the 1-parameter family
a
t a1 2 Diff1 .Ip M/gt2.";"/
of diffeomorphisms consists of elements of the holonomy group Hol. p/ and satisfies

the conditions

1 @ a
t a1
a
0 a D Id; D a a1 ;
@t
tD0
the assertion follows.

Definition 9 A vector field B 2 X1 .Ip M/ on the indicatrix Ip M will be called
the Berwald translate of the vector field 2 X1 .Iq M/ along the curve D x.t/ if
B D a .a /1 :
Remark 6 Let y.t/ D t y.0/ 2 Ix.t/ M be a parallel vector field along D x.t/,
0 6 t 6 a, started at y.0/ 2 Ix.0/ M. Then, the vertical vector field t D .x.t/; y.t//
along .x.t/; y.t// is the Berwald translate t D t 0 t 1 if and only if

@ i .x; y/ @ i .x; y/ @
rxP D Gj .x; y/
k
C Gjk .x; y/ .x; y/ xP j i D 0:
i k
@xj @yk @y
Now, Lemma 1 yields the following

Corollary 6 If 2 hol .q/, then its Berwald translate B 2 X1 .Ip M/ along any
curve D x.t/, 0 6 t 6 a, joining q D x.0/ with p D x.a/ is tangent to the holonomy
group Hol. p/.
This last statement motivates the following
Definition 10 The holonomy algebra holp .M/ of the Finsler manifold .M; F / at
the point p 2 M is defined by the smallest Lie algebra of vector fields on the
indicatrix Ip M, containing the Berwald translates of all infinitesimal holonomy
algebras along arbitrary curves x.t/, 0 6 t 6 a joining any points q D x.0/ with the
point p D x.a/.
Clearly, the holonomy algebras at different points of the Finsler manifold .M; F /
are isomorphic. Lemma 1, Corollary 6, and Proposition 1 yield the following
Theorem 3 The holonomy algebra holp .M/ at a point p 2 M of a Finsler manifold
.M; F / is tangent to the holonomy group Hol. p/.
If .M; F / is a positive definite Finsler manifold, then the group generated by
the exponential image exp holp .M/ is a subgroup of the topological closure of the
12.5.4 Finsler Surfaces with hol .x/ D Rx
The relation between the infinitesimal holonomy algebra and the curvature algebra
is enlightened by the following
Theorem 4 Let .M; F / be a Finsler surface with non-zero constant flag curvature.
The infinitesimal holonomy algebra hol .x/ at a point x 2 M coincides with the
curvature algebra Rx at x if and only if the mean Berwald curvature E.x;y/ of .M; F /
vanishes for any y 2 Ix M.
Proof Let U M be a neighborhood of x 2 M diffeomorphic to R2 . Identifying U
with R2 and considering a coordinate system .x1 ; x2 / in R2 we can write

Rijk .x; y/ D ji gkm .x; y/ym ki gjm .x; y/ym ; with 0:
Since the curvature tensor field is skew-symmetric, R.x;y/ acts on the one-
dimensional wedge product Tx M ^ Tx M. According to Lemma 2 the covariant
derivative of the curvature vector field D R.X; Y/ D 12 R.X YY X/ D R.X ^ Y/
can be written in the form
rZ D rZ .r.X; Y// D R .rZ .X ^ Y// D R.rZ X ^ Y C X ^ rZ Y/;
where X; Y; Z 2 X.U/. If X D X i @x@ i , Y D Y i @x@ i and Z D Z i @x@ i , then we have X ^ Y D

1
1 2
2
X Y X 2 Y 1 @x@1 ^ @x@2 and hence we obtain

1 2 1 2 @ @
rZ D R rk .X Y Y X / 1 ^ 2 Z k (12.34)
@x @x

@.X 1 Y 2 Y 1 X 2 / k @ @
DR Z ^
@xk @x1 @x2

@ @
C.X 1 Y 2 Y 1 X 2 /R rk ^ Zk ;
@x1 @x2
@
where we denote the covariant derivative rZ by rk if Z D @xk
, k D 1; 2. For given
1 @.X 1 Y 2 Y 1 X 2 /
vector fields X; Y; Z 2 X .U/ the expression @xk
k
Z is a function on U.
Hence there exists a function on U such that

@.X j Y h Y j X h / k @ @
R Z j^ h D R.X ^ Y/ D R.X; Y/;
@x k @x @x
and R.X; Y/ is an element of the curvature algebra R.U/ of the submanifold

.U; F jU /.
Now, we investigate the second term of the right-hand side of (12.34).
@ @ @ @ @ @
rk ^ D r k ^ C ^ rk D
@x1 @x2 @x1 @x2 @x1 @x2
@ @ @ @ @ @
D Glk1 l ^ 2 C 1 ^ Gm D G1k1 C G2k2 ^ :
@x @x @x k2
@xm @x1 @x2
Hence
@ @ k 1
.X 1 Y 2 Y 1 X 2 /R rk 1
^ 2
Z D Gk1 CG2k2 Z k R.X; Y/ D G1k1 C G2k2 Z k
@x @x
This expression belongs to the curvature algebra if and only if the function G1k1 CG2k2
does not depend on the variable y, i.e. if and only if

@ G1k1 C G2k2
Ekh D D 0; h; k D 1; 2;
@yh
identically.
Remark 7 Let D R.X; Y/ be a curvature vector field. Assume that the vector fields
X; Y 2 X1 .M/ have constant coordinate functions in a local coordinate system
.x1 ; : : : ; xn / of the Finsler surface .M; F /. Then we have in this coordinate system

rZ D G1k1 C G2k2 Z k :
Shen constructed in [28] families of Randers surfaces depending on the real

parameter , which are of constant flag curvature 1 on the unit sphere S2 R3
and of constant flag curvature 1 on a disk D2 R2 . These Finsler surfaces are not
projectively flat and have vanishing S-curvature (c.f. [28], Theorems 1.1 and 1.2).
Their Finsler function is defined by
p
2 h.v; y/2 C h. y; y/ .1 2 h.v; v// h.v; y/
D ; D ; (12.35)
1 2 h.v; v/ 1 2 h.v; v/
where h.v; y/ is the standard metric of the sphere S2 , respectively h.v; y/ is the
standard Klein metric on the unit disk D2 and v denotes the vector field defined by
.x2 ; x1 ; 0/ at .x1 ; x2 ; x3 / 2 S2 , respectively by .x2 ; x1 / at .x1 ; x2 / 2 D2 .
Theorem 5 For any Randers surface defined by (12.35) the infinitesimal holonomy
algebra hol .x/ at a point x 2 M coincides with the curvature algebra Rx .
Proof According to Theorems 1.1 and 1.2 in [28], the above classes of not
locally projectively flat Randers surfaces with non-zero constant flag curvature have
vanishing S-curvature. Moreover, Proposition 6.1.3 in [27], p. 80 yields that the
mean Berwald curvature vanishes in this case. Hence the assertion follows from
Corollary 4.
12.6 Infinite Dimensional Subalgebras of the Infinitesimal

Holonomy Algebra
12.6.1 Projective Finsler Surfaces of Constant Curvature
A Finsler manifold .M; F / of dimension 2 is called Finsler surface. In this case

the indicatrix is one-dimensional at any point x 2 M, hence the curvature vector
fields at x 2 M are proportional to any given non-vanishing curvature vector field.
It follows that the curvature algebra Rx .M/ has a simple structure: it is at most
one-dimensional and commutative. Even in this case, the infinitesimal holonomy
algebra holx .M/ can be higher dimensional, or potentially infinite dimensional. For
the investigation of such examples we use a classical result of Lie claiming that
the dimension of a finite-dimensional Lie algebra of vector fields on a connected
one-dimensional manifold is less than 4 (cf. [1], Theorem 4.3.4). We obtain the
following
Lemma 1 If the infinitesimal holonomy algebra holx .M/ of a Finsler surface
.M; F / contains four simultaneously non-vanishing R-linearly independent vector
fields, then holx .M/ is infinite dimensional.
Proof If the infinitesimal holonomy algebra is finite-dimensional, then the
dimension of the corresponding Lie group acting locally effectively on the one-
dimensional indicatrix would be at least 4, which is a contradiction.
Let .M; F / be a locally projectively flat Finsler surface of non-zero constant
curvature, let .x1 ; x2 / be a local coordinate system centered at x 2 M, corresponding
to the canonical coordinates of the Euclidean space which is projectively related to
.M; F / and let . y1 ; y2 / be the induced coordinate system in the tangent plane Tx M.
In the sequel we identify the tangent plane Tx M with R2 with help p of the coordi-
nate system . y1 ; y2 /. We will use the euclidean norm jj. y1 ; y2 /jj D . y1 /2 C . y2 /2
of R2 and the corresponding polar coordinate system .er ; t/, too.
Let '. y1 ; y2 / be a positively 1-homogeneous function on R2 and let r.t/ be the

2-periodic smooth function r W R ! R determined by
p
'.er.t/ cos t; er.t/ sin t/ D 1 or '. y1 ; y2 / D er.t/ . y1 /2 C. y2 /2 ; (12.36)
where
y1 y2 y2
cos t D p ; sin t D p ; tan t D ;
. y 1 /2 C . y 2 /2 . y 1 /2 C . y 2 /2 y1
i.e. the level set f'. y1; y2 / 1g of the 1-homogeneous function ' in R2 is given by
the parametrized curve t ! .er.t/ cos t; er.t/ sin t/:
Since the curvature of a smooth curve t ! .er.t/ cos t; er.t/ sin t/ in R2 is
er
D p .Rr rP 2 1/; (12.37)
rP 2 C 1
the vanishing of the expression rR rP 2 1 means the infinitesimal linearity of the

corresponding positively homogeneous function in R2 .
Definition 11 Let '. y1 ; y2 / be a positively 1-homogeneous function on R2 and
let .t/ be the curvature of the curve t ! .er.t/ cos t; er.t/ sin t/ defined by the
Eq. (12.36). We say that '. y1 ; y2 / is strongly convex, if .t/ 0 for all t 2 R.
Conditions (A)(C) in the following theorem imply that the projective factor P at
x0 2 M is a nonlinear function, and hence, according to Remark 1, .M; F / is a
non-Riemannian Finsler manifold.
Theorem 1 Let .M; F / be a projectively flat Finsler surface of non-zero constant
curvature covered by a coordinate system .x1 ; x2 /. Assume that there exists a point
x0 2 M such that one of the following conditions hold
(A) F induces a scalar product on Tx0 M and the projective factor P at x0 is a
strongly convex positively 1-homogeneous function,
(B) F .x0 ; y/ is a strongly convex absolutely 1-homogeneous function on Tx0 M, and
the projective factor P.x0 ; y/ on Tx0 M satisfies P.x0 ; y/ D c F .x0 ; y/ with
0 c 2 R,
(C) there is a projectively related Euclidean coordinate system of .M; F / centered
at x0 and one has
1
F .0; y/ D jyj ha; yi and P.0; y/ D .jyj ha; yi/ : (12.38)
2
Assume that the vector fields U D U i @x@ i , V D V i @x@ i 2 X1 .M/ have constant
coordinate functions and let D R.U; V/ be the corresponding curvature vector
field. Then the subalgebra
of the infinitesimal holonomy algebra holx .M/ generated
by the vector fields x0 , r1 jx0 , r2 jx0 and r1 .r2 / jx0 is infinite dimensional.
Proof Since .M; F / is of constant flag curvature, we can write

Rijk .x; y/ D ji gkm .x; y/ym ki gjm .x; y/ym ; with D const:
According to Lemma 2 the horizontal Berwald covariant derivative rW R of the

tensor field R D Rijk .x; y/dxj ^ dxk @x@ i vanishes and hence rW R D 0.
Since the curvature tensor field is skew-symmetric, R.x;y/ acts on the one-
dimensional wedge product Tx M ^ Tx M. The covariant derivative rW of the
curvature vector field D R.U; V/ D 12 R.U V V U/ D R.U ^ V/ can be
written in the form
rW D R .rW .U ^ V// D R.rW U ^ V C U ^ rW V/:

1
1 2
We have U ^ V D 2 U V U 2 V 1 @x@1 ^ @
@x2
and hence

rW D .U 1 V 2 V 1 U 2 /W k R rk @x@1 ^ @
@x2
; (12.39)
where rk WD r @ . Since
@xk
@ @

rk @x1
^ @x2
D rk @x@1 ^ @
@x2
C @
@x1
^ rk @x@2
1 2
@
D Glk1 @x@ l ^ @
@x2
C @
@x1
^ Gm @
k2 @xm D Gk1 C Gk2 @x1 ^
@
@x2
we obtain

rW D G1k1 C G2k2 W k R.U; V/ D G1k1 C G2k2 W k :
Since the geodesic coefficients are given by (12.17) we have
@P k
rW D Gm
km W D 3
k
W : (12.40)
@yk
Hence

@P k @P k @P k @P l
rZ .rW / D 3rZ W D 3 rZ W C W Z :
@yk @yk @yk @yl
Let W be a vector field with constant coordinate functions. Then, using (12.17)
we get
2
@P k @2 P 2
m @ P @P @2 P
rZ W D Gj m k W Z D
k j
P k j W k Zj;
@yk @xj @yk @y @y @xj @yk @y @y
and hence

@2 P @2 P @P @P
rZ .rW / D 3 P C k W k Z j : (12.41)
@x @y
j k @y @y
k j @y @y j
Let x0 2 M be the point with coordinates .0; 0/ in the local coordinate system
of .M; F / corresponding to the canonical coordinates of the projectively related
Euclidean plane. According to Lemma 8.2.1 in [6], p.155, we have
@2 P @2 P @P @P @P @P l @2 F 2 @P @P
1 2
P 1 2
C 1 2
D 2 1 2
1 2
D 2 1 2 g12 :
@x @y @y @y @y @y @y @y 2 @y @y @y @y

Hence the vector fields x0 , r1 jx0 , r2 jx0 and r1 .r2 / jx0 are linearly independent
if and only if the functions

@P @P @P @P
1; ; ; 2 1 2 g12 (12.42)
@y1 x0 @y2 x0 @y @y x0
are linearly independent, where g12 D gy . @x@1 ; @x@2 / is the component of the metric
tensor of .M; F /.
2
Lemma 2 The functions @P.0;y/
@y1
, @P.0;y/
@y2
and P.0; y/ @ @yP.0;y/
1 @y2 can be expressed in
the polar coordinate system .er ; t/ by
@P.0; y/ @P.0; y/
D .cos t C rP sin t/er; D .sin t rP cos t/er;
@y1 @y2
@2 P.0; y/
P.0; y/ D .Pr2 C 1 rR /e2r sin t cos t;
@y1 @y2
where the dot refers to differentiation with respect to the variable t.

2
@er @ y2 d tan t @t
Proof We obtain from @y1
D er rP @y@t1 and from . yy1 /2 D . / dt @y1
@y1 y1
D
2 2
1 @t @er r 2 r
cos2 t @y1
that @y1
D e rP cos t . yy1 /2 De rP . y1 /2 yC. y2 /2 : Hence
p !
@P .0; y/ @ er . y1 /2 C . y2 /2 r y2 y1
D De rP p Cp :
@y1 @y1 . y1 /2 C . y2 /2 . y1 /2 C . y2 /2
1
@er
Similarly, we have @y2
D er rP cos2 t y11 D er rP . y1 /2 yC. y2 /2 : Hence
p !
@P.0; y/ @ er . y1 /2 C . y2 /2 r y1 y2
D De Pr p Cp :
@y2 @y2 . y1 /2 C . y2 /2 . y1 /2 C . y2 /2
Finally we have
@2 P.0; y/ @.sin t rP cos t/er 1

1 2
D D .Rr rP 2 1/er sin t cos t p :
@y @y @y1 . y / C . y 2 /2
1 2
Replacing ' by the function P.0; y/ in the expression (12.36) we get the assertion.
Lemma 3 Let r W R ! R be a 2-periodic smooth function such that the inequality
rR .t/ rP 2 .t/ 1 0 holds on a dense subset of R. Then the functions
1; .cos t C rP sin t/er ; .sin t rP cos t/er ; .cos t C rP sin t/.sin t rP cos t/e2r
(12.43)
are linearly independent.

Proof The derivative of .cos t C rP sin t/er and of .sin t rP cos t/er are .Rr rP 2
1/er sin t and .Rr rP 2 1/er cos t, respectively, hence the functions (12.43) do not
vanish identically. Let us consider a linear combination
ACB.cos t C rP sin t/er CC.sin t rP cos t/er CD.cos t C rP sin t/.sin t rP cos t/e2r D 0
with constant coefficients A; B; C; D. We differentiate and divide by et .Rr rP 2 1/

and we have
B sin t C cos t D.cos 2t C rP sin 2t/er D 0:
Putting t D 0 and t D we get C D Der.0/ D Der./ . Since er.0/ ; er./ > 0

we get C D D D 0 and hence A D B D C D D D 0.
Now, assume that condition (A) of Theorem 1 is fulfilled. According to Propo-
sition 1 if the functions (12.42) are linearly independent, then the holonomy
group Holx0 .M/ is an infinite dimensional subgroup of Diff1 .Ix0 M/. The function
F .x0 ; y/ induces a scalar product on Tx0 M, consequently the component g12 of the
metric tensor is constant on Tx0 M. Hence Holx0 .M/ is infinite dimensional if the
functions

@P @P @P @P
1; ; ; (12.44)
@y1 x0 @y2 x0 @y1 @y2 x0
are linearly independent. This follows from Lemma 3 and hence the assertion of the
theorem is true.
Assume that condition (B) is satisfied. We denote '. y/ D F .x0 ; y/. Using
the expressions (12.42) we obtain that the vector fields x0 , r1 jx0 , r2 jx0 and
r1 .r2 / jx0 are linearly independent if and only if the functions

@P @' @P @'
1; D c 1; Dc 2
@y1 x0 @y @y2 x0 @y

@P @P @' @' @2 '
2 1 2 g12 D .2c2 / 1 2 ' 1 2
@y @y x0 @y @y @y @y
are linearly independent. According to Lemma 2 this is equivalent to the linear

independence of the functions
1; .cos t C rP sin t/ er ; .sin t rP cos t/ er ;

.2c2 /.cos t C rP sin t/.sin t rP cos t/ e2r .Rr rP 2 1/ e2r sin t cos t:
If r D const, then these functions are 1; cos t er ; sin t er ; 2c2 cos t sin t e2r ,
hence the assertion follows from Lemma 3. In the following we can assume that
r.t/ const. Let t0 2 R such that rP .t0 / D 0 and .t0 / 0. Weprotate the coordinate
system at the angle t0 with respect to the euclidean norm . y1 /2 C . y2 /2 , then
we get in the new polar coordinate system that rP .0/ D 0 and .0/ 0. Consider the
linear combination
ACB.cos t C rP sin t/er C C.sin t rP cos t/ er

CD .2c2 /.cos tCPr sin t/.sin t rP cos t/ e2r .Rr Pr2 1/ e2r sin t cos t D 0
(12.45)
with some constants A; B; C; D. Since the function ' is absolutely homogeneous,

the function r.t/ is -periodic. Putting t C into t, the value of
A C D.2c2 /.cos t C rP sin t/.sin t rP cos t/ e2r .Rr rP2 1/ e2r sin t cos t
does not change, but the value of
B.cos t C rP sin t/ er C C.sin t rP cos t/ er
changes sign. Since Lemma 3 implies that .cos t C rP sin t/ er and .sin t rP cos t/ er
are linearly independent, we have B D C D 0 and (12.45) becomes

1
A e2r C D .2c2 / Pr cos 2t C .1 rP 2 / sin 2t .Rr rP 2 1/ sin 2t D 0:
2 2
(12.46)
Since rP .0/ D 0 at t D 0, we have A D 0. If D 0, then (12.46) gives

1
.2c2 / Pr cos 2t C .1 rP2 / sin 2t .Rr rP 2 1/ sin 2t D 0:
2 2
By derivation and putting t D 0 we obtain

.2c2 / rR .0/ C 1 .Rr.0/ 1/ D 2c2 .1 rR .0// D 0:
Using the relation (12.37) condition (B) gives .0/ D er.0/.1 rR .0// 0, which
is a contradiction. Hence D D 0 and the vector fields x0 , r1 jx0 , r2 jx0 and
r1 .r2 / jx0 are linearly independent. Using Proposition 1 we obtain the assertion.
Suppose now that the condition (C) holds. Hence we have
@F y1 @F y2 @2 F y1 y2
.0; y/ D a1 ; .0; y/ D a2 ; .0; y/ D ;
@y1 jyj @y2 jyj @y1 @y2 jyj3
and
D y E y1 y2
y1 y2
g12 D a1 a2 1 a; : (12.47)
jyj jyj jyj jyj2
Similarly, we obtain from condition (C) that
@P y1 @P y2
.0; y/ D a1 ; .0; y/ D a2 :
@y1 jyj @y2 jyj

Using the expressions (12.42) we get that the vector fields x0 , r1 jx0 , r2 jx0 ,
r1 .r2 / jx0 are linearly independent if and only if the functions

@P y1 @P y2
1; D a1 ; D a2
@y1 .0;y/ jyj @y2 .0;y/ jyj
and

@P @P
2 g D
12
1
@y @y 2
.0;y/
D y E y1 y2 y1 y2
D a; C.1/
jyj jyj2 jyj2
1
y y2
.2C/ a2 Ca1 C.2/a1a2
jyj jyj
are linearly independent. Putting
y1 y2
cos t D ; sin t D
jyj jyj
we obtain that this condition is true, since the trigonometric polynomials
1; cos t; sin t; .1 / cos t sin t .a1 cos t C a2 sin t/ cos t sin t

are linearly independent. Hence the vector fields x0 , r1 jx0 , r2 jx0 , r1 .r2 / jx0
generate an infinite dimensional subalgebra of Holx0 .M/.
12.6.2 Projective Finsler Manifolds of Constant Curvature
Now we will prove that the infinitesimal holonomy algebra of a totally geodesic
submanifold of a Finsler manifold can be embedded into the infinitesimal holonomy
algebra of the entire manifold. This result yields a lower estimate for the dimension
of the holonomy group.
12.6.2.1 Totally Geodesic and Auto-Parallel Submanifolds
Lemma 4 Let M N be a totally geodesic submanifold in a spray manifold .M; S /.

N can be extended to a curvature vector
The curvature vector fields at any point of M
field of M.
Proof Assume that the manifolds M N and M are k, respectively n D k C p
dimensional. Let .x1 ; : : : ; xk ; xkC1 ; : : : ; xn / be an adapted coordinate system, i.e. the
submanifold M N is locally given by the equations xkC1 D D xn D 0. Using the
notation of the proof of Lemma 1 we get from Eq. (12.12) that G D 0 and G D 0
for any .x1 ; : : : ; xk ; 0; : : : ; 0I y1 ; : : : ; yk ; 0; : : : ; 0/ we have
@G @G
C G G G G C G G G G D 0
@x @x
at .x1 ; : : : ; xk ; 0; : : : ; 0I y1 ; : : : ; yk ; 0; : : : ; 0/. Hence the curvature tensors KN and K,

corresponding to the spray SN, respectively to the spray S satisfy
N
K.X; Y/.x; y/ D K.X; Y/.x; y/ if N
x2M N
and y; X; Y 2 Tx M:
It follows that for any given X; Y 2 Tx M N y/ D

N the curvature vector field .
N N N
K.X; Y/.x; y/ at x 2 M defined on Tx M can be extended to the curvature vector
N defined on Tx M.
field . y/ D K.X; Y/.x; y/ at x 2 M
Theorem 2 Let M N be a totally geodesic two-dimensional submanifold of a Finsler
N of M
manifold .M; F / such that the infinitesimal holonomy algebra holx .M/ N is
infinite dimensional. Then the infinitesimal holonomy algebra holx .M/ of M is
infinite dimensional.
Proof According to Lemma 1 any curvature vector field of M N at x 2 M N M

defined on Ix M N can be extended to a curvature vector field on the indicatrix Ix M.
Hence the curvature algebra Rx .M/ N of the submanifold M N can be embedded into the
curvature algebra Rx .M/ of the manifold .M; F /. Assume that N is a vector field
belonging to the infinitesimal holonomy algebra holx .M/ N which can be extended
to the vector field belonging to the infinitesimal holonomy algebra holx .M/. Any
vector field XN 2 X1 .M/ N can be extended to a vector field X 2 X1 .M/, hence the
horizontal Berwald covariant derivative along XN 2 X1 .M/ N of N can be extended to
the Berwald horizontal covariant derivative along X 2 X1 .M/ of the vector field
N of the submanifold M
. It follows that the infinitesimal holonomy algebra holx .M/ N
can be embedded into the infinitesimal holonomy algebra holx .M/ of the Finsler
manifold .M; F /. Consequently, holx .M/ is infinite dimensional and hence the
holonomy group Holx .M/ is an infinite dimensional subgroup of Diff1 .Ix M/.
This result can be applied to locally projectively flat Finsler manifolds, as they have
for each tangent 2-plane a totally geodesic submanifold which is tangent to this
2-plane.
Corollary 7 If a locally projectively flat Finsler manifold has a two-dimensional
totally geodesic submanifold satisfying one of the conditions of Theorem 1, then its
infinitesimal holonomy algebra is infinite dimensional.
According to Eqs. (12.18) and (12.19) the projectively flat Randers manifolds of
non-zero constant curvature satisfy condition (C) of Theorem 1. We can apply
Corollary 7 to these manifolds and we get the following
Theorem 3 The infinitesimal holonomy algebra of any projectively flat Randers
manifolds of non-zero constant flag curvature is infinite dimensional.
Bryant in [4, 5] introduced and studied complete Finsler metrics of positive
curvature on S2 . He proved that there exists exactly a 2-parameter family of Finsler
metrics on S2 with curvature = 1 with great circles as geodesics. Z. Shen generalized
a 1-parameter family of complete Bryant metrics to Sn satisfying
F .0; y/ D jyj cos ; P.0; y/ D jyj sin (12.48)
with jj < 2 in a coordinate neighborhood centered at 0 2 Rn , (cf. Example 7.1. in

[30] and Example 8.2.9 in [6]).
We investigate the holonomy groups of two families of metrics, containing the
1-parameter family of complete Bryant-Shen metrics (12.48). The first family in the
following theorem is defined by condition (A), which is motivated by Theorem 8.2.3
in [6]. There is given the following construction:
If D . y/ is an arbitrary Minkowski norm on Rn and ' D '. y/ is an
arbitrary positively 1-homogeneous function on Rn , then there exists a projectively
flat Finsler metric F of constant flag curvature 1, defined on a neighborhood of
the origin, such that F and its projective factor P satisfy F .0; y/ D . y/ and
P.0; y/ D '. y/.
Condition (B) in the next theorem is confirmed by Example 7 in [30], p. 1726,

where it is proved that for an arbitrary given Minkowski norm ' and j#j < 2 there
exists a projectively flat Finsler function F of constant curvature D 1 defined on a
neighborhood of 0 2 Rn , such that
F .0; y/ D '. y/ cos # and P.0; y/ D '. y/ sin #:
Conditions (A) and (B) in Theorem 1 together with Corollary 7 yield the following
Theorem 4 Let .M; F / be a projectively flat Finsler manifold of non-zero constant
curvature. Assume that there exists a point x0 2 M and a two-dimensional totally
geodesic submanifold M N through x0 such that one of the following conditions
holds
(A) F induces a scalar product on Tx0 M, N and the projective factor P on Tx0 M N is
a strongly convex positively 1-homogeneous function,
(B) F .x0 ; y/ on Tx0 MN is a strongly convex absolutely 1-homogeneous function
on Tx0 M, and the projective factor P.x0 ; y/ on Tx0 M N satisfies P.x0 ; y/ D
cF .x0 ; y/ with 0 c 2 R.
Then the infinitesimal holonomy holx .M/ of M is infinite dimensional.
12.7 Dimension of the Holonomy Group
Let .M; F/ be a positive definite Finsler manifold and x 2 M an arbitrary point in

M. According to Proposition 3 of [18], the infinitesimal holonomy algebra holx .M/
is tangent to the holonomy group Holx .M/. Therefore the group generated by the
exponential image of the infinitesimal holonomy algebra at x 2 M with respect to
the exponential map expx W X1 .Ix M/ ! Diff1 .Ix M/ is a subgroup of the closed
holonomy group Holx .M/ (see Theorem 3.1 of [21]). Consequently, we have the
following estimation on the dimensions:
dim holx .M/ 6 dim Holx .M/: (12.49)
Proposition 7 The infinitesimal holonomy algebra holx .M/ of any locally projec-
tively flat non-Riemannian Finsler surface .M; F / of constant curvature 0 is
infinite dimensional.
Proof We use for the proof Lemma 1 and the notations introduced in the proof
of Theorem 1 in the previous section on projective Finsler surfaces of constant
curvature. Using the assertion on the vector fields (12.42) we obtain
Lemma 1 For any fixed 1 6 j; k 6 2
y ! .x; y/; y ! r1 .x; y/; y ! r2 .x; y/; y ! rj .rk / .x; y/; (12.50)
considered as vector fields on Ix M, are R-linearly independent if and only if the
@P @P @2 P
1; ; ; gjk (12.51)
@y1 @y2 @y j @yk 4
are linearly independent functions on Tx M.
Since we assumed that the Finsler function F is non-Riemannian at the point x,
i.e. F 2 .x; y/ is non-quadratic in y, the function P.x; y/ is nonlinear in y on Tx M
(cf. Eq. (12.17)). Let us choose a direction y0 D . y10 ; y20 / 2 Tx M with y10 0,
y20 0 and having property that P is nonlinear 1-homogeneous function in a conic
neighborhood U of y0 in Tx M. By restricting U if it is necessary we can suppose that
for any y 2 U we have y1 0, y2 0.
To avoid confusion between coordinate indexes and exponents, we rename the
fiber coordinates of vectors belonging to U by .u; v/ D . y1 ; y2 /. Using the values
of P on U we can define a 1-variable function f D f .t/ on an interval I R by
1
f .t/ WD P.x1 ; x2 ; tv; v/: (12.52)
v
Then we can express P and its derivatives with f :
@P @P u
P D v f .u=v/; D f 0 .u=v/; D f .u=v/ f 0 .u=v/;
@y1 @y 2 v
@2 P 1 @2 P u @2 P u2 00
1 1
D f 00 .u=v/; 1 2
D 2 f 00 .u=v/; D f .u=v/:
@y @y v @y @y v @y2 @y2 v 3
(12.53)
Lemma 2 The functions 1; @P
@y1
; @P
@y2
are linearly independent.
Proof A nontrivial relation a C b @P

@y1
C c @P
@y2
D 0 yields the differential equation
0 0
a C bf C c. f tf / D 0: It is clear that both b and c cannot be zero. If c 0, we
get the differential equation
.a C cf /0 1
D :
a C cf t bc
The solution is f .t/ D t .a C b/=c and therefore the corresponding P.u; v/ D u

v.a C b/=c is linear which is a contradiction. If c D 0, then b 0 and f D ab t C K.
The corresponding P.u; v/ D ab u C Kv is again linear which is a contradiction.
Let us assume now that the infinitesimal holonomy algebra is finite dimensional. We
will show that this assumption leads to contradiction which will prove then that the
infinitesimal holonomy algebra is actually infinite dimensional.
Since Ix M is one-dimensional, according to Lemma 1, the four vector fields
in (12.50) are linearly dependent for any j; k 2 f1; 2g. Using Lemma 1 we get that
the functions

1; P1 ; P2 ; Pj Pk gjk (12.54)
4
2
(Pi D @P @yi
, Pjk D @y@ jP@yk
) are linearly dependent for any j; k 2 f1; 2g. From
Lemma 2 we know that the first three functions in (12.54) are linearly independent.
Therefore by the assumption, the fourth function must be a linear combination of
the first three, that is there exist constants ai ; bi ; ci 2 R, i D 1; 2; 3, such that

g11 D P1 P1 C a1 C b1 P1 C c1 P2 ;
4

g12 D P1 P2 C a2 C b2 P1 C c2 P2 ; (12.55)
4

g22 D P2 P2 C a3 C b3 P1 C c3 P2 :
4
Since @1 g21 @2 g11 D 0 and @1 g22 @2 g12 D 0 we have
P2 P11 P1 P12 C b2 P11 C .c2 b1 /P12 c1 P22 D 0;

(12.56)
P1 P22 P2 P12 b3 P11 C .b2 c3 /P12 C c2 P22 D 0:
Using the expressions (12.53) we obtain from (12.56) the equations

u 1 00 u 1 u u2
f f0 f C f 0 2 f 00 C b2 f 00 .c2 b1 / 2 f 00 c1 3 f 00 D 0;
v v v v v v
(12.57)
u 2 u u 1 u u2
f 0 3 f 00 C f f 0 2 f 00 b3 f 00 .b2 c3 / 2 f 00 C c2 3 f 00 D 0:
v v v v v v
Since by the nonlinearity of P on U we have f 00 0, Eq. (12.57) can divide by

f 00 =v and we get
u c1 u 2
f C b2 C .b1 c2 / D 0
v v2
(12.58)
u u c2 u 2
f b3 C .c3 b2 / C D 0:
v v v2
for any t D u=v in an interval I R. The solution of this system of quadratic

equations for the function f is f .t/ D c2 t b2 with c1 D b3 D 0, b1 D 2c2 ,
c3 D 2b2 . But this is a contradiction, since we supposed that by the nonlinearity of
P we have f 00 0 on this interval. Hence the functions 1, P1 , P2 , Pj Pk 4 gjk
can not be linearly dependent for any j; k 2 f1; 2g, from which follows the assertion.
Remark 8 From Proposition 7 we get that if .M; F / is non-Riemannian and 0,
then the holonomy group has an infinite dimensional tangent algebra.
Indeed, according to Theorem 6.3 in [18] the infinitesimal holonomy algebra
holx .M/ is tangent to the holonomy group Holx .M/, from which follows the
assertion.
Now, we can prove our main result:

Theorem 1 The holonomy group of a locally projectively flat simply connected
Finsler manifold .M; F / of constant curvature is finite dimensional if and only if
.M; F / is Riemannian or D 0.
Proof If .M; F / is Riemannian, then its holonomy group is a Lie subgroup of
the orthogonal group and therefore it is a finite dimensional compact Lie group.
If .M; F / has zero curvature, then the horizontal distribution associated to the
canonical connection in the tangent bundle is integrable and hence the holonomy
group is trivial.
If .M; F / is non-Riemannian having non-zero curvature , then for each tangent
2-plane S Tx M the manifold M has a totally geodesic submanifold M e M
e e
such that Tx M D S. This M with the induced metric is a locally projectively flat
Finsler surface of constant curvature . Therefore from Proposition 7 we get that
e is infinite dimensional. Moreover, according to Theorem 4.3 in [22], if a
holx .M/
Finsler manifold .M; F / has a totally geodesic two-dimensional submanifold M e
e
such that the infinitesimal holonomy algebra of M is infinite dimensional, then the
infinitesimal holonomy algebra holx .M/ of the containing manifold is also infinite
dimensional. Using (12.49) we get that Holx .M/ cannot be finite dimensional. Hence
the assertion is true.
We note that there are examples of non-Riemannian type locally projectively flat
Finsler manifolds with D 0 curvature, (cf. [16]).
Remark 9 In the discussion before the previous theorem, the key condition for
the Finsler metric tensor was not the positive definiteness but its non-degenerate
property. Therefore Theorem 1 can be generalized as follows.
A pair .M; F / is called semi-Finsler manifold if in the definition of Finsler
manifolds the positive definiteness of the Finsler metric tensor is replaced by the
non-degenerate property. Then we have
Corollary 8 The holonomy group of a locally projectively flat simply connected
semi-Finsler manifold .M; F / of constant curvature is finite dimensional if and
only if .M; F / is semi-Riemannian or D 0.
12.8 Maximal Holonomy
12.8.1 The Group Diff1 1

C .S / and the Fourier Algebra
Let .M; F / be a positive definite Finsler 2-manifold. In this case the indicatrix is
diffeomorphic to the unit circle S1 , at any point x 2 M. Moreover, if there exists a
non-vanishing curvature vector field at x 2 M then any other curvature vector field
at x 2 M is proportional to it, which means that the curvature algebra is at most
one-dimensional. The infinitesimal holonomy algebra, however, can be an infinite
dimensional subalgebra of X1 .S1 /, therefore the holonomy group can be an infinite

dimensional subgroup of Diff1 1
C .S /, cf. [22].
1
Let S D R mod 2 be the unit circle with the standard counterclockwise
orientation. The group Diff1 1
C .S / of orientation preserving diffeomorphisms of
the S is the connected component of Diff1 .S1 /. The Lie algebra of Diff1
1 1
C .S /
1 1 1
is the Lie algebra X .S /denoted also by Vect.S / in the literaturecan be
written in the form f .t/ dtd , where f is a 2-periodic smooth functions on the real
line R. A sequence f fj dtd gj2N Vect .S1 / converges to f dtd in the Frchet topology
of Vect.S1 / if and only if the functions fj and all their derivatives converge uniformly
to f , respectively to the corresponding derivatives of f . The Lie bracket on Vect.S1 /
is given by

d d df dg d
f ;g D g f :
dt dt dt dt dt
The Fourier algebra F.S1 / on S1 is the Lie subalgebra of Vect.S1 / consisting

of vector fields f dtd such that f .t/ has finite Fourier series, i.e. f .t/ is a Fourier

polynomial. The vector fields dtd ; cos nt dtd ; sin nt dtd n2N provide a basis for F.S1 /.
A direct computation shows that the vector fields
d d d d d
; cos t ; sin t ; cos 2t ; sin 2t (12.59)
dt dt dt dt dt
generate the Lie algebra F.S1 /. The complexification F.S1 /R C of F.S1 / is called
the Witt algebra W.S1 / on S1 having the natural basis ieint dtd n2Z , with the Lie
bracket ieimt dtd ; ieint dtd D i.m n/ei.nm/t dtd :
D E
Lemma 1 The group exp .F.S1 // generated by the topological closure of the
exponential image of the Fourier algebra F.S1 / is the orientation preserving
diffeomorphism group Diff1 1
C .S /.
Proof The Fourier algebra F.S1 / is a dense subalgebra of Vect.S1 / with respect
to the Frchet topology, i.e. F.S1 / D Vect.S1 /. This assertion follows from the fact
that any r-times continuously differentiable function can be approximated uniformly
by the arithmetical means of the partial sums of its Fourier series (cf. [10], 2.12
Theorem). The exponential mapping is continuous (c.f. Lemma 4.1 in [25], p. 79),
hence we have

exp Vect.S1 / D exp F.S1 / exp F.S1 / Diff1 1
C .S / (12.60)
which gives for the generated groups the relations

D E
exp Vect.S1 / exp F.S1 / Diff1 1
C .S /: (12.61)
Moreover, the conjugation map Ad W Diff1 1 1

C .S / Vect.S / satisfies the relation
h exp s h1 D exp sAd.h/
for every h 2 Diff1 1 1

C .S / and 2 Vect.S /. Clearly,
the Lie 1algebra
Vect.S1 / is
invariant under conjugation and hence
the group exp Vect.S / is also invariant
under conjugation. Therefore exp Vect.S1 / is a nontrivial normal subgroup of
Diff1 1 1 1
C .S /. On the other hand, DiffC .S / is a simple group (cf. [9])
which
means that
its only nontrivial normal subgroup is itself. Therefore, we have exp Vect.S1 / D
Diff1 1
C .S /; and using (12.61) we get
D E
exp.F.S1 // D Diff1 1
C .S /:
12.8.2 Holonomy of the Standard Funk Plane

and the Bryant-Shen 2-Spheres
Using the results of the preceding chapter we can prove the following statement,
which provides a useful tool for the investigation of the closed holonomy group of
Finsler 2-manifolds.
Proposition 8 If the infinitesimal holonomy algebra holx .M/ at a point x 2 M of
a simply connected Finsler 2-manifold .M; F / contains the Fourier algebra F.S1 /
on the indicatrix at x, then Holx .M/ is isomorphic to Diff1 1
C .S /.
Proof Since M is simply connected we have
Holx .M/ Diff1 1

C .S /: (12.62)
On the other hand, using Proposition 1, we get

exp F.S1 / Holx .M/ ) exp F.S1 / Holx .M/ ) exp F.S1 / Holx .M/;
and from the last relation, using Lemma 1, we can obtain that
Diff1 1
C .S / Holx .M/: (12.63)
Comparing (12.62) and (12.63) we get the assertion.

Using this proposition we can prove our main result:
Theorem 1 Let .M; F / be a simply connected projectively flat Finsler manifold of
constant curvature 0. Assume that there exists a point x0 2 M such that on Tx0 M
the induced Minkowski norm is an Euclidean norm, that is F .x0 ; y/ D kyk, and the
projective factor at x0 satisfies P.x0 ; y/ D ckyk with c 2 R, c 0. Then the closed

holonomy group Holx0 .M/ at x0 is isomorphic to Diff1 1
C .S /.
Proof Since .M; F / is a locally projectively flat Finsler manifold of non-zero

constant curvature, we can use an .x1 ; x2 / local coordinate system centered at
x0 2 M, corresponding to the canonical coordinates of the Euclidean space which
is projectively related to .M; F /. Let . y1 ; y2 / be the induced coordinate system
in the tangent plane Tx M. In the sequel we identify the tangent plane Tx0 M with
R2 by using p the coordinate system . y1 ; y2 /. We will use the Euclidean norm
k. y ; y /k D . y1 /2 C . y2 /2 of R2 and the corresponding polar coordinate system
1 2
.er ; t/, too.

Let us consider the curvature vector field at x0 D 0 defined by

@ @ @
DR ; D 2i g1m .0; y/ym 1i g2m .0; y/ym
@x1 @x2 xD0 @xi
Since .M; F / is of constant flag curvature, the horizontal Berwald covariant

derivative rW R of the tensor field R vanishes, c.f. Lemma 2. Therefore the covariant
derivative of can be written in the form

@ @
rW D R rk ^ Wk:
@x1 @x2
Since

@ @ @ @
rk 1
^ 2 D G1k1 C G2k2 1
^ 2
@x @x @x @x
@P
we obtain rW D G1k1 C G2k2 W k : Using (12.17) we can express Gm
km D 3 @yk D
k
3c kyk
y
and hence
@P yk
rk D 3 D 3c ;
@yk kyk
where we use the notation rk D r @ . Moreover we have

@xk

@P @2 P @2 P @2 P @2 P
rj D Gm D j k P k j;
@yk @x @y
j k j
@y @y
m k @x @y @y @y
and hence

@2 P @2 P @P @P
rj .rk / D 3 P C3 k :
@xj @yk @yk @y j @y @y j
According to Lemma 8.2.1, Eq. (8.25) in [6], p. 155, we obtain
@2 P @P @P @2 P @2 F 2
D CP j k :
@x @y
j k @y @y
j k @y @y 2 @y j @yk
Using the assumptions on F and on the projective factor P we can get at x0

@F @F @2 F 2
rj .rk / D 3 4c2 j k
@y @y 2 @y j @yk
and hence
j k
2 y y
rj .rk / D 3 4 c ;
jk
kyk2
where jk 2 f0; 1g such that jk D 1 if and only if j D k.

Let us introduce polar coordinates y1 D r cos t, y2 D r sin t in the tangent
space Tx0 M. We can express the curvature vector field, its first and second covariant
derivatives along the indicatrix curve f.cos t; sin t/I 0 6 t < 2g as follows:
d d d d
D ; r1 D 3c cos t ; r2 D 3c sin t ; r1 .r2 / D 12 c2 sin 2t ;
dt dt dt dt
d d
r1 .r1 / D 12 c2 cos2 t ; r2 .r2 / D 12 c2 sin2 t :
dt dt
Since c 0, the vector fields
d d d d d d
; cos t ; sin t ; cos t sin t ; cos2 t ; sin2 t
dt dt dt dt dt dt
are contained in the infinitesimal holonomy algebra holx0 .M/. It follows that the
generator system

d d d d d
; cos t ; sin t ; cos 2t ; sin 2t
dt dt dt dt dt
of the Fourier algebra F.S1 / (c.f. Eq. (12.59)) is contained in the infinitesimal
holonomy algebra holx0 .M/. Hence the assertion follows from Proposition 8.
We remark that the standard Funk plane and the Bryant-Shen 2-spheres are
connected, projectively flat Finsler manifolds of non-zero constant curvature.
Moreover, in each of them, there exists a point x0 2 M and an adapted local
coordinate system centered at x0 with the following properties: the Finsler norm
F .x0 ; y/ and the projective factor P.x0 ; y/ at x0 are given by F .x0 ; y/ D kyk and
by P.x0 ; y/ D ckyk with some constant c 2 R, c 0, where kyk is an Euclidean
norm in the tangent space at x0 . Using Theorem 1 we can obtain
Theorem 2 The closed holonomy groups of the standard Funk plane and of
the Bryant-Shen 2-spheres are maximal, that is diffeomorphic to the orientation
preserving diffeomorphism group of S1 .
References
1. M. Ackermann, R. Hermann, Sophus Lies Transformation Group Paper (Mathematical

Science Press, Brookline, 1975).
2. W. Barthel, Nichtlineare Zusammenhnge und deren Holonomiegruppen. J. Reine Angew.
Math. 212, pp. 120149 (1963)
3. A. Borel, A. Lichnerowicz, Groupes dholonomie de varietes riemanniennes. CR Acad. Sci.
Paris 234, 18351837 (1952)
4. R. Bryant, Finsler structures on the 2-sphere satisfying K = 1, in Finsler Geometry, Contempo-
rary Mathematics, vol. 196 (American Mathematical Society, Providence, 1996), pp. 2742
5. R. Bryant, Projectively flat Finsler 2-spheres of constant curvature. Sel. Math. N. Ser. 3, 161
204 (1997)
6. S.S. Chern, Z. Shen, Riemann-Finsler Geometry. Nankai Tracts in Mathematics, vol. 6 (World
Scientific, Singapore, 2005)
7. M. Crampin, D.J. Saunders, Holonomy of a class of bundles with fibre metrics. Publ. Math.
Debr. 81, 199234 (2012)
8. Ch. Ehresmann, Les connexions infinitsimales dans un espace fibr diffrentiable. Colloque
de Topologie, Bruxelles, (1950), pp. 2955
9. M.R. Herman, Sur le groupe des diffomorphismes du tore. Ann. Inst. Fourier 23, 7586 (1973)
10. Y. Katznelson, An Introduction to Harmonic Analysis (Cambridge University Press, Cam-
bridge, 2004)
11. S. Kobayashi, Transformation Groups in Differential Geometry. Ergebnisse der Mathematik
und ihrer Grenzgebiete, vol. 70 (Springer, Berlin, 1972)
12. I. Kolar, P.W. Michor, J. Slovak, Natural Operations in Differential Geometry (Springer, Berlin,
1993)
13. L. Kozma, On Landsberg spaces and holonomy of Finsler manifolds. Contemp. Math. 196,
177185 (1996)
14. L. Kozma, Holonomy structures in Finsler geometry. Part 5, in Handbook of Finsler Geometry,
ed. by P.L. Antonelli (Kluwer Academic, Dordrecht, 2003), pp. 445490
15. A. Kriegl, P.W. Michor, The Convenient Setting for Global Analysis. Surveys and Monographs,
vol. 53 (AMS, Providence, 1997)
16. B. Li, Z. Shen, On a class of projectively flat Finsler metrics with constant flag curvature. Int.
J. Math. 18, 112 (2007)
17. P.W. Michor, Gauge Theory for Fiber Bundles. Monographs and Textbooks in Physical
Sciences. Lecture Notes, vol. 19 (Bibliopolis, Napoli, 1991)
18. Z. Muzsnay, P.T. Nagy, Tangent Lie algebras to the holonomy group of a Finsler manifold.
Commun. Math. 19, 137147 (2011)
19. Z. Muzsnay, P.T. Nagy, Finsler manifolds with non-Riemannian holonomy. Houst. J. Math. 38,
7792 (2012)
20. Z. Muzsnay, P.T. Nagy, Witt algebra and the curvature of the Heisenberg group. Commun.
Math. 20, 3340 (2012)
21. Z. Muzsnay, P.T. Nagy, Characterization of projective Finsler manifolds of constant curvature
having infinite dimensional holonomy group. Publ. Math. Debr. 84, 1728 (2014)
22. Z. Muzsnay, P.T. Nagy, Projectively flat Finsler manifolds with infinite dimensional holonomy.
Forum Mathematicum 27, 767786 (2015)
23. Z. Muzsnay, P.T. Nagy, Finsler 2-manifolds with maximal holonomy group of infinite
dimension. Differ. Geom. Appl. 39, 19 (2015)
24. S. Numata, On Landsberg spaces of scalar curvature. J. Korean Math. Soc. 12, 97100 (1975)
25. H. Omori, Infinite-Dimensional Lie Groups. Translation of Mathematical Monographs, vol.
158 (American Mathematical Society, Providence, 1997)
26. G. Randers, On an asymmetrical metric in the fourspace of general relativity. Phys. Rev. (2)
59, 195199 (1941)
27. Z. Shen, Differential Geometry of Spray and Finsler Spaces (Kluwer Academic, Dordrecht,
2001)
28. Z. Shen, Two-dimensional Finsler metrics of constant flag curvature. Manuscripta Math. 109,
349366 (2002)
29. Z. Shen, Projectively flat Randers metrics with constant flag curvature. Math. Ann. 325, 1930
(2003)
30. Z. Shen, Projectively flat Finsler metrics with constant flag curvature. Trans. Am. Math. Soc.
355, 17131728 (2003)
31. Z.I. Szab, Positive definite Berwald spaces. Tensor, New Ser. 35, 2539 (1981)
32. J. Teichmann, Regularity of infinite-dimensional Lie groups by metric space methods. Tokyo
J. Math. 24, 2958 (2001)
33. J.A. Wolf, Differentiable fibre spaces and mappings compatible with Riemannian metrics.
Michigan Math. J. 11, 6570 (1964)
Chapter 13
Lepage Manifolds
Olga Rossi
Abstract Lepage manifolds represent a geometric structure arising from differen-

tial equations, and related with them as close as possible. Geometric properties
of Lepage manifolds reflect geometrical, topological and dynamical properties of
differential equations, and vice versa. After symplectic geometry, Lepage manifolds
represent a next step in understanding interrelations between differential equations,
geometry and physics. The aim of this chapter is to introduce the new concept, and
to stimulate deeper studies in this direction as well as applications in differential
geometry, differential equations, dynamical systems, exterior differential systems,
calculus of variations, and mathematical physics.
13.1 Introduction
Lepage manifold is a fibred manifold endowed with a certain differential form

Lepage .n C 1/-form where n is the dimension of the base manifold. Lepage
manifolds represent a geometric structure arising from differential equations, and
related with them as close as possible. It turns out that geometric properties
of Lepage manifolds reflect geometric, topological and dynamical properties of
differential equations, and vice versa. The aim of this chapter is to introduce the
new concept and to stimulate further research in this direction, as well as many
potential applications.
After symplectic geometry, Lepage manifolds represent a next step in under-
standing interrelations between differential equations, geometry and physics. In a
similar way as symplectic manifolds constitute foundations of analytical mechanics
and the theory of dynamical systems, Lepage manifolds become fundamental
for differential equations in general, with strong applications in the calculus of
variations and geometric field theories.
O. Rossi ()
Department of Mathematics, Faculty of Science, The University of Ostrava, 30. dubna 22, 701 03
Ostrava, Czech Republic
e-mail: olga.rossi@osu.cz

322 O. Rossi
The use of differential forms in the calculus of variations and its applications
in physics goes back to . Cartan, Poincar, Whittaker, Carathodory, Lepage,
Dedecker and Weinstein (see, e.g., [9, 11, 16, 72, 79, 92, 94]). A great success
was the development of symplectic geometry as a beautiful and powerful tool
for geometrization of classical Hamiltonian mechanics (see, e.g., Abraham and
Marsden [1], Arnold et al. [4], de Len and Rodrigues [14], Libermann and Marle
[73]), and with many applications in physics, the theory of differential equations,
dynamical systems, control theory, and elsewhere. Nevertheless, it has turned out
that there are many variational systems for which the use of symplectic manifolds
is inappropriate. Indeed, a symplectic structure is naturally inherited only by a
particular class of variational problems, namely those described by variational inte-
grals of one independent variable with first-order regular Lagrangians, independent
upon the parameter of the curves (time). The generalization of the symplectic
structure arising in classical mechanics to multiple integrals is by no means trivial.
It means to find out proper geometric structures for general variational problems,
including degenerate Lagrangians (implicit variational equations) and higher-order
Lagrangian systems, in order to generalize Lagrangian theory, Hamilton and
HamiltonJacobi theory and Noether theory. The next challenge is then to include
non-variational equations in the picture, and to understand the property to be
variational.
Despite great efforts, the attempts to include more general Lagrangians or even
non-variational equations in the symplectic setting have not really been satisfactory.
It has turned out that the difficulties are of geometric origin: instead of tangent and
cotangent bundles, one has to consider jet bundles,1 and symplectic forms have to
be replaced by some other geometric objects. There is a vast literature which cannot
be mentioned here; to illustrate the state of art and the diversity of approaches we
refer at least to the books [7, 13, 27, 32, 34, 41, 52, 59, 66], and selected papers
[8, 10, 15, 18, 20, 21, 2426, 2831, 33, 3840, 42, 45, 48, 51, 60, 63, 71, 75, 76,
81, 82, 85, 91] to mention just a few.
Concerning the variational case, roughly speaking, there are two mainstream
settings: one based on multisymplectic or polysymplectic forms, and the other based
on the Cartan form and its generalizations. Within the latter approach, the central
object for formulation of Lagrangian and Hamiltonian field theories is the Poincar
Cartan form, or more generally, a Lepage n-form where n is the dimension of the
base of the underlying fibred manifold. Following the idea of Lepage [72], examples
of these forms were introduced to the calculus of variations in the seminal works of
Carathodory, Dedecker, Krupka, Betounes, Olver, Gotay, Crampin and Saunders
(see e.g. [5, 9, 12, 17, 30, 45, 46, 55, 77]). The general concept of Lepage n-form,
or, Lepage equivalent of a Lagrangian, is due to Krupka [45, 48] who used it to
build, systematically, a general theory of higher order variational integrals on fibred
manifolds, generalizing the Lagrange, Hamilton, HamiltonJacobi and Noether
1
The concept introduced by Ehresmann in the 1950s [22, 23]; a standard book on this subject is
[86].
13 Lepage Manifolds 323
theories of classical mechanics [52]. Subsequently, instead of a Lepage equivalent

of a Lagrangian, we suggested to consider better a certain .n C 1/-form which in
this context becomes a Lepage equivalent of a system of variational equations (see
[56, 5860, 62, 81, 83]). This step was motivated mainly by the idea to obtain a
global Hamilton theory, independent upon a choice of a particular Lagrangian; in
this new setting, Hamilton equations are the same for all equivalent Lagrangians.
At the same time this approach avoids certain unpleasant discrepancies which
appear in Hamiltonian formulation of Lagrangian theories related with particular
Lagrangians (as reported, e.g., in [17, 19, 35, 59, 60, 68, 69, 87]), and allows one
to treat Lagrangian systems globally in terms of the theory of exterior differential
systems even in case when a global Lagrangian does not exist [59, 60, 63, 81].
It took quite a long time to find a proper generalization of Lepage .n C 1/-forms
to non-variational equations [53, 78]. The generalization is not straightforward,
and appears within the theory of variational sequences (for the introduction to the
variational sequence theory we refer to [49, 50], and [52]). This progress motivated
us to introduce the concept of a Lepage manifold as defined in this article.
Every Lepage manifold naturally carries an exterior differential system (EDS). A
fundamental result on the Inverse Problem of the Calculus of Variations [48] implies
that the corresponding system of (generally higher-order) differential equations
(ODEs if n D 1 and PDEs otherwise) for integral sections of this EDS is
variational, and so represents EulerLagrange equations of possibly locally defined
Lagrangians, if and only if the generating Lepage .n C 1/-form is closed (up
to a contact equivalence). This property makes Lepage manifolds very suitable
structures to study variational equations, and at the same time a good background
to study non-variational differential equations in the context of the calculus
of variations making use of variational structures and techniques. With Lepage
manifolds, many particular results known so far and fragmented in the calculus of
variations and in the geometric theory of differential equations can be put together
in a unified framework, linked together, and extended. Lepage manifolds naturally
arise wherever differential equations of dynamical type are concerned (i.e. related
with dynamical forms, vector fields, sprays, semisprays, Ehresmann connections).
Through this relationship they are naturally present in Riemannian geometry, Finsler
geometry, symplectic geometry, and Lagrangian geometry, as well as in analytical
mechanics, classical field theories, relativity, string theories, and other physical
systems. The present exposition aims to show the potential of this new framework,
and to stimulate further research in this direction, as well as study of applications,
namely in differential geometry, differential equations and mathematical physics.
The present article is structured as follows: In Sect. 13.2 we introduce Lepage
forms (of any degree) in the context of the variational sequence, and we remind
the solution of the Inverse Problem of the Calculus of Variations. We follow the
paper [78] where the reader can find more detailed exposition, as well as other
aspects and applications of the theory of variational sequences. Section 13.3 is
devoted to Lepage manifolds. After defining the concept we show the relationship
with variational equations and with non-variational equations, and we introduce the
important class of regular Lepage manifolds. We also remind a theorem by Krupka
324 O. Rossi
et al. [54], which makes possible to transfer non-variational Lepage manifolds to

the variational ones. In Sect. 13.4 we pay a particular attention to Lepage mani-
folds of order zero which are a natural framework for general Hamiltonian systems.
Many results of this section are taken from the paper [36]. Then we present original
results on Legendre transformation and Poisson structure for the generalized (and
not necessarily variational) Hamiltonian systems, and, remarkably, we introduce a
new concept of a directional Poisson bracket in covariant Hamiltonian field theory.
The last two sections summarize some applications of Lepage manifolds in classical
field theories and in mechanics. Finally we briefly show that (and how) Lepage
manifolds naturally arise in Riemann and Finsler geometry.
We assume that the reader has a basic knowledge of jet bundles and related
geometric structures as introduced, e.g., in [86]. As for the calculus of variations
in jet bundles, the reader can be guided by the books [13, 59], and [52].
13.2 Lepage Forms
13.2.1 Jet Bundles and Contact Forms
All manifolds and mappings are smooth, and summation over repeated indices
applies.
Let W Y ! X be a fibred manifold (surjective submersion) with fibre dimension
m, and r W J r Y ! X its r-jet prolongation; usually we consider r D 1; 2, and for
the sake of simplicity we sometimes denote Y as J 0 Y. We also denote by r;k W
J r Y ! J k Y the natural projections. J r Y is the manifold of r-jets of local sections
of ; a point in J r Y (the equivalence class of sections of at x 2 X, with the same
derivatives up to the order r) is denoted by jrx . In what follows we shall omit the
word local and speak just about sections meaning local sections.
Any section of can be naturally prolonged to a section J r of the fibred
manifold r W J r Y ! X, defined by J r .x/ D jrx . Sections of r of this kind are
called holonomic and have to be distinguished from (general) sections of r which
need not be a form of a prolongation of a section of .
Let .xi ; y / denote local fibred coordinates on Y, and .xi ; y J / the associated
coordinates on J r Y, where J is a multiindex, 0 6 jJj 6 r, J D . j1 ; : : : ; jp / with
1 6 j1 6 j2 6 6 jp 6 n and p 6 r. In formulas where the summation
convention applies we shall also use the functions y J , J D . j1 ; : : : ; jp /, p 6 r,
where 1 6 j1 ; j2 ; : : : ; jr 6 n; these functions are defined naturally with help of the
above coordinate functions using the symmetry of partial derivatives. Next, we set
!0 D dx1 ^ dx2 ^ ^ dxn ; !j D i@=@x j !0 ; !Jj D i@=@x j !J : (13.1)
Jet bundles over fibred manifolds carry a canonical contact structure. A differen-
tial form on J r Y is called contact if it vanishes on r-jet prolongations of all sections
of . It easily follows that every q-form for q > n is contact, so it is convenient

to introduce a refined concept: A contact q-form is called k-contact if for every
r -vertical vector field the contraction i is .k 1/-contact, where k D 1; 2; : : : q.
0-contact forms, i.e. such that i D 0 whenever is a r -vertical vector field, are
more often referred to as horizontal forms.
Horizontal forms of degree n where n is the dimension of the base manifold X
are essential in the calculus of variations, since they serve as Lagrangians. In what
follows, we shall denote the sheaf of Lagrangians of order r over by rn;X .
An important structural Lemma states that every q-form on J r Y, if lifted to J rC1 Y,
has an invariant decomposition

rC1;r D h C p1 C C pq (13.2)
where h is horizontal, and pk are k-contact forms (1 6 k 6 q) [48].

If applied to the exterior derivative of a function f , the splitting (13.2) gives
the horizontal differential dH f D hdf and vertical differential dV f D p1 df . The
n components of the horizontal differential are called total derivatives; they take the
form
@f X
r
@f
di f D i C y : (13.3)
@x @y J Ji
jJjD0
In view of the canonical decomposition above we often say that a form is at

most k-contact if pi D 0 for all i > k, and at least k-contact if pi D 0 for all
i < k. Note that if q > n C 1 then is contact, and it is at least .q n/-contact,
i.e. the contact components p0 ; p1 ; : : : ; pqn1 of are equal to zero. If, moreover,
pqn D 0, we speak about a strongly contact form.
Contact 1-forms on J r Y annihilate a distribution called Cartan distribution and
denoted Cr . The annihilator Cr0 of Cr is generated by the following local contact
forms:
!J D dy J y Ji dxi ; 0 6 jJj 6 r 1 : (13.4)
It is also worth notice that the contact ideal of order r is generated by contact 1-
forms and exterior derivatives of contact 1-forms on J r Y [49].
Finally, we recall that a vector field on J r Y is called a contact symmetry if it
is a symmetry of the contact ideal on J r Y; contact symmetries are characterized by
the condition L Cr0 Cr0 . In particular, for every -projectable vector field on
Y the r-jet prolongation J r is a contact symmetry. Conversely, if on J r Y is a r -
projectable contact symmetry, then D J r for some -projectable vector field
on Y.
326 O. Rossi
13.2.2 The Variational Sequence
Now we are prepared to introduce the variational sequence due to Krupka [49] (see
also [52] and [78] for review).2
Let us denote by RY the constant sheaf over R, and by rq the sheaf of q-forms
on J r Y. Next, we let r0;c D f0g, and rq;c be the sheaf of contact q-forms, if q 6 n,
or the sheaf of strongly contact q-forms, if q > n, on J r Y. We set

qr D rq;c C drq1;c ; (13.5)

where drq1;c is the image of rq1;c by the exterior derivative d, and drq1;c
denotes the sheaf generated by the presheaf drq1;c . We note that qr D 0 for
q > corank Cr .
The subsequence
0 ! 1r ! 2r ! 3r ! (13.6)
of the de Rham sequence
0 ! RY ! r0 ! r1 ! r2 ! r3 ! (13.7)
is an exact sequence of soft sheaves. The quotient sequence
0 ! RY ! r0 ! r1 =1r ! r2 =2r ! r3 =3r ! (13.8)
is called the variational sequence of order r.

Theorem 13.2.1 ([49]) The variational sequence is an acyclic resolution of the
constant sheaf RY .
The above assertion means that the variational sequence is locally exact with the
exception of RY . As a consequence, due to the abstract de Rham theorem, we obtain
the following fundamental result:
Corollary 13.2.2 ([49]) The cohomology groups of the cochain complex of global
sections of the variational sequence are identified with the de Rham cohomology
q
groups HdR Y of the manifold Y.
Since the variational sequence is a quotient sequence, the elements in the
individual columns are elements of the quotient sheaves rq =qr , i.e. they are
equivalence classes of local r-th order differential q-forms. We denote by 2
rq =qr the class of 2 rq . Morphisms in the variational sequence are then by
construction quotients of the exterior derivative d. We denote
Eq W rq =qr ! rqC1 =qC1

r
; (13.9)
2
For basics of sheaf theory needed below we refer to standard books, e.g. [6] or [93].
so that
Eq ./ D d: (13.10)
Now, Eq ./ D d D 0 means that there exists 2 rq1 =q1 r

such that
q
D Eq1 ./ D d. If, moreover, HdR Y D f0g, then the class has a global
representative.
The name variational sequence comes from the direct relationship with the
calculus of variations. Namely, as proved in [49, 51], the sheaf rn =nr is isomorphic
with a subsheaf of the sheaf of Lagrangians rC1 n;X , and the quotient mapping En W
rn =nr ! rnC1 =nC1
r
is the EulerLagrange mapping.
The kernel of En then consists of null-Lagrangians (variationally trivial
Lagrangians), and its image, which, by exactness is the kernel of the next morphism,
EnC1 W rnC1 =nC1

r
! rnC2 =nC2
r
; (13.11)
represents variational equations. Therefore EnC1 is called Helmholtz morphism.

The name refers to the famous Helmholtz conditions, necessary and sufficient
conditions for differential equations to be variational (as they stand) [3, 37, 47].
Remarkably, the Helmholtz morphism, discovered within the variational sequence
theory, has not been known in the classical calculus of variations.
13.2.3 Source Forms and Lepage Forms
Classes in the variational sequence can be represented by global differential forms

which play a fundamental role in the calculus of variations and the theory of
differential equations on manifolds. To introduce them we shall need on J r Y the
ideal generated by lifts of contact 1-forms of order one, whose elements take the

form r;1 ! ^ where ! 2 C10 (the annihilator of C1 ) and is of order r. Since
locally they are expressed as ! ^ , where ! D dy y i dxi , 1 6 6 m, are the
basic local contact 1-forms of order 1, we say that forms belonging to this ideal are
f! g-generated.
Definition 13.2.3 ([65]) Let k > 1. By source form of degree n C k and order r we
shall mean a f! g-generated k-contact .n C k/-form on J r Y.
Source .n C1/-forms are called dynamical forms, source .n C2/-forms are called
Helmholtz-like forms.3
The question how to construct examples of source forms is solved by the interior
Euler operator I (see, e.g., [2, 43, 44, 53]). The interior Euler operator reflects in
3
As explained later, for the calculus of variations the most important examples of source forms of
degree n C 1 are EulerLagrange forms, and of source forms of degree n C 2 the Helmholtz forms.
328 O. Rossi
an intrinsic way the procedure of getting a distinguished global representative of a

class 2 rq =qr for q > n by applying to the operator pqn and factorization
by qr .

Let k > 1. For 2 rnCk we have rC1;r D pk C pkC1 C C pkCn ; we set
1 X
r
I ./ D ! ^ .1/jJj dJ i@=@y J pk ; (13.12)
k
jJjD0
where dJ D dj1 : : : djp for jJj D p. It can be shown by means of the partition of unity
arguments that this formula defines a global form I ./.
The operator I W rnCk ! 2rC1 nCk has the following properties:
1. For every 2 rnCk , I ./ is a source form of degree n C k.

2. I . pk / D I ./.
2rC1
3. I ./ belongs to the same class as 2rC1;r , i.e., 2rC1;r I ./ 2 nCk .
4. I 2 D I , up to a canonical projection; precisely, I 2 ./ D 4rC3;2rC1

I ./.
5. Ker I D nCk r
.
6. For k D 1, I ./ is the unique source form representing the class . If k >
1, I ./ is no longer a unique source form equivalent with (precisely, with

2rC1;r ).4 We say that I ./ is a canonical source form for .
By construction, I ./ is a k-contact form, however, in general it is different
from pk . The difference is given by means of the so-called residual operator R as
follows:
pk D I ./ C pk dpk R./ (13.13)
(up to appropriate canonical projectionssee the footnote below). It is worth notice

that R./ is a local strongly contact .n C k 1/-form, and that, contrary to I ./,
the form R./ need not be unique and need not be global.
The operator I enables one to introduce fundamental differential forms as
follows:
Definition 13.2.4 ([53]) Let k > 0. A .n C k/-form on J r Y is called Lepage form
of degree k if
pkC1 d D I .d/ : (13.14)
From the definition we immediately see that

every q-form on Y, q > n, is a Lepage form,
every closed q-form on J r Y, q > n, is a Lepage form,
4
For the sake of simplicity of notations, we shall often omit the pull-back and identify a form with
its lift or projection.
and we can obtain the structure of Lepage forms:

Theorem 13.2.5 ([78]) Equation (13.14) has the solution

rC1;r D pk C d C D pk C pkC1 d C ; (13.15)
where
pk D pk pkC1 R.dpk / ; (13.16)
is an arbitrary at least .k C 1/-contact .n C k 1/-form, and (resp. ) is an

arbitrary at least .k C 2/-contact form.
For every choice of and (resp. ) the Lepage forms (13.15) belong to the

same variational class 2 snCk =nCks
(for a proper s), i.e. s;r pk 2 nCk
s
.
In what follows, we shall denote by fg the family of (possibly local) Lepage
forms containing , as characterized by Theorem 13.2.5.
The form pk is completely determined by its k-contact part pk . It was
discovered in [78]. For k D 0 this is the famous Cartan form [11, 13, 25], hence
we call it Cartan form of degree k. As seen from the defining formula (13.16), the
Cartan form is generically local. However, from the formula (13.15) it follows that
if n D dim X D 1 and r > 0 (mechanics and higher-order mechanics), or, if n > 1
and r D 1 (first order field theory), then for every k > 0 the Cartan form pk is
global.
It is worth notice that an equivalent definition of a Lepage .n C k/-form of order
r obviously is as follows:

rC1;r d D I .d/ C F; where F is at least .k C 2/-contact. (13.17)
In view of (13.15) we also say that is a Lepage equivalent of pk . As proved

in [78], every k-contact .n C k/-form has a global Lepage equivalent of degree k.
For k D 0 (i.e. p0 D h) the definition gives Lepage equivalent of a Lagrangian
h D as introduced by Krupka in [45] and [48]. The dynamical form p1 d D
p1 d is then the EulerLagrange form E D I .d/, and En W ! E is the
EulerLagrange mapping. For k > 0, the most important Lepage forms are Lepage
equivalents of source forms. If k D 1, the definition gives Lepage equivalent of a
dynamical form " (first introduced in [56] for locally variational dynamical forms),
and p2 d D p2 d" is the canonical Helmholtz form H" D I .d"/. The mapping
EnC1 W " ! H" is then the Helmholtz mapping. For k D 2 we get Lepage equivalents
of source forms of degree n C 2 which are called Helmholtz-like forms, and the
corresponding mapping is EnC2 W ! I .d/. This case has been studied for n D 1
in [64] and [74].
We also note that formula (13.15) immediately yields:
Corollary 13.2.6 If n D dim X D 1, then for every k > 0 and every r > 1, any
k-contact .n C k/-form has a unique global Lepage equivalent.
330 O. Rossi
13.2.4 The Inverse Problem in the Variational Sequence
We start with the following definition:

Definition 13.2.7 ([48]) A dynamical form " is called (globally) variational if there
exists a Lagrangian such that " D E (possibly up to a jet projection). " is called
locally variational if the domain of " can be covered by open sets U such that for
every , "jU is variational.
The famous Inverse Problem of the Calculus of Variations is the question under
what conditions a dynamical form " is locally/globally variational. In terms of
the variational sequence this transfers to the question on the image of the Euler
Lagrange mapping, or due to the exactness of the sequence, on the kernel of the
Helmholtz mapping. In this way the inverse problem in its local and global form
extends to each column of the variational sequence, and is solved by the exactness
of the sequence. With help of Lepage forms, the solution of the Inverse Problem is
obtained in full generality and in an elegant way, transferring the whole problem to
the Poincar Lemma and de Rham cohomology.
Before presenting the theorem, recall a convenient modification of the Poincar
homotopy operator, adapted to the contact structure. Denote by P the Poincar
homotopy operator. Then A defined by
A p0 ! D 0; A pk ! D pk1 P!; k > 1 (13.18)

Krupka [48], satisfies rC1;r ! D A d.rC1;r !/ C dA .rC1;r !/, and is adapted to
the decomposition of forms into contact components: if (locally) ! D d, then
X
q

rC1;r DA .rC1;r !/ D A pkC1 ! ; (13.19)
kD0
hence pk D A pkC1 !. Compared to P, the operator A concerns vertical curves

(curves in the fibres over X) only.
Theorem 13.2.8 ([78] General Inverse Variational Problem) Let be a source
form of degree n C k, k > 1. The following conditions are equivalent:
(1) belongs to Ker EnCk .
(2) has a closed Lepage equivalent.
(3) Every Lepage equivalent of satisfies pkC1 d D 0.
If EnCk . / D 0, then there is a local source .n C k 1/-form satisfying
EnCk1 ./ D , and it holds D A .
If moreover HdRnCk
Y D f0g, then there is a global source .n C k 1/-form
satisfying EnCk1 ./ D .
Note that :
For k D 1 condition (1) means that the dynamical form " D p1 is locally
variational.
Condition (3) represents precisely the source constraints, i.e. constraints on a
source form of degree n C k belong to the kernel of the corresponding morphism
in the variational sequence. It is a generalization (to any degree) of the famous
Helmholtz conditions on local variationality of a dynamical form (i.e. necessary
and sufficient conditions of existence of a local Lagrangian for differential
equations represented by the dynamical form).5
Condition (2) is equivalent with (3), hence
P also represents the source constraints.
However, since (up to projection) d D nC1 iD1 pkCi d, condition d D 0 includes
also conditions pkC2 d D 0, . . . , pkCnC1 d D 0 which are dependent, hence
superfluous. In coordinates this gives dependent conditions in form of derivatives
of the Helmholtz conditions (respectively, their higher degree generalizations if
k > 1).
Formula D A for k D 1 (i.e. if is a dynamical form) gives the celebrated
VainbergTonti Lagrangian [89, 90].
13.3 Lepage Manifolds
13.3.1 Lepage Manifolds and Differential Equations
As above, we consider a fibred manifold W Y ! X and its jet prolongations.

Definition 13.3.1 Let r > 0. By a Lepage manifold of order r we shall mean a
fibred manifold W Y ! X endowed with a Lepage .n C 1/-form on J r Y such
that p1 is a dynamical form.
A Lepage manifold of order r will be denoted by .r ; /.
By definition, a Lepage manifold .r ; / is endowed with a (uniquely deter-
mined) global dynamical form " D p1 of order r C 1. In fibred coordinates
" D E ! ^ !0 (13.20)
where the coefficients E are functions on a coordinate domain in J rC1 Y. The

dynamical form represents globally and intrinsically differential equations of order
r C 1. More precisely, the equations arise as equations for sections of the fibred
5
Helmholtz conditions (k D 1) in the form (3) were first presented in [57]. The coordinate form of
the Helmholtz conditions has been first obtained by Helmholtz [37] for second order ODEs, and
by Anderson and Duchamp [3] and Krupka [47, 48] for higher order PDEs.
For k D 2 (and mechanics) the explicit coordinate form of (3) can be found in [64] and [74].
332 O. Rossi
manifold such that
E J rC1 D 0: (13.21)
Indeed, in coordinates, this is a system of m D dim Y dim X differential equations

(ODEs if dim X D 1 and PDEs if dim X > 1) of order r C 1 for the components of
sections of .
Since is a Lepage equivalent of ", Theorem 13.2.8 tells us that if p2 d D 0
then the dynamical form " is locally variational (and vice versa), meaning that
the Helmholtz form H" D EnC1 ."/ vanishes, and around every point, " is the
EulerLagrange form of a local Lagrangian . Hence the corresponding differential
equations (13.21) are EulerLagrange equations of ; in such a case, we say that
Eq. (13.21) are variational (as they stand). Moreover by Theorem 13.2.8 (2), local
variationality of " is equivalent with the existence of a closed Lepage equivalent
of " (which, however, need not be global). In other words, on a Lepage manifold
.r ; / such that p2 d D 0 but d 0, we can locally always find a closed Lepage
form 0 such that p1 0 D p1 D ".
On the other hand, given a Lepage manifold .r ; / such that p2 d 0, it holds
H" 0, the Lepage family fg cannot be (even locally) represented by a closed
form, and the corresponding equations (13.21) are not variational (as they stand).
From this analysis we can see that the concept of Lepage manifold concerns in
fact three columns in the variational sequence: In the variational case and in the
non-variational case, respectively, we have
En EnC1
! E ! 0
EnC1 EnC2
" ! H" ! 0
with help of source forms, or
d d
f g ! fd g D fE g ! 0
d d
f" g ! fd" g D fH" g ! 0
with help of Lepage forms.

Apart from the dynamical form ", every Lepage manifold is canonically endowed
with the exterior differential system D generated by n-forms
i where runs over all r -vertical vector fields on J r Y: (13.22)

We have the following theorem (cf. [63]):

Theorem 13.3.2 Let .r ; / be a Lepage manifold, " D p1 . Let be a section of
. The following conditions are equivalent:
(1) is a solution of differential equations (13.21) for ".
(2) J r is an integral section of D , i.e.
J r i D 0 for every r -vertical vector field on J r Y: (13.23)
The set of holonomic solutions of D does not depend on the choice of an

element of the Lepage family fg.
Proof All the assertions follow easily from

J r i D J rC1 i . p1 / D J rC1 .E !0 / D .E J rC1 / . / !0
(13.24)
if we take for the s the r-th, resp. .rC1/-prolongation of a local basis ./ D @=@q

(hence ./ D ). t
u
Remark 13.3.3 Consider the associated system DO of , i.e. the exterior differential
system locally generated by n-forms i where runs over all vector fields on J r Y.
Then D DO . However, D and DO have the same holonomic integral sections.
This follows from the fact that for 1 6 i 6 n,

J r i@=@xi D J rC1 i@=@xi . p1 / D .E y i / J rC1 !0 ; (13.25)
hence equations for holonomic integral sections of D and DO take the same form:
E J rC1 D 0, 1 6 6 m.
Remarkably, given a Lepage manifold .r ; /, one can consider all (not only
holonomic) solutions of D . The corresponding equations are then equations for
components of sections of the fibred manifold r W J r Y ! X and take the form
i D 0 for every r -vertical vector field on J r Y: (13.26)
Equivalently they can be written in the form
i@=@y j :::j D 0; 1 6 6 m; 0 6 k 6 r; 1 6 j1 6 6 jk 6 n : (13.27)

1 k
We note that (13.27) are first order differential equations (ODEs or PDEs depending
on the dimension of X), and (generally) they are not equivalent with (13.23), i.e.,
with Eq. (13.21). However, if a section of is a solution of (13.21) then J r is a
solution of (13.27).
Equation (13.27) can be given another geometric interpretation in terms of
Lepage manifolds. Namely, given a Lepage manifold .r ; / of order r > 1, we
334 O. Rossi
can consider as a Lepage form of order zero for the fibred manifold r W J r Y ! X.
In this way we relate with .r ; / the Lepage manifold ..r /0 ; / of order zero. Let
us denote the i-contactization operator with respect to the projection r by pQ i . The
associated dynamical form
"Q D pQ 1 (13.28)
is then defined on J 1 .J r Y/ and called the Hamilton form of ". Note that is, indeed,
the Lepage equivalent of "Q (more precisely, the Lepage equivalent of "Q is projectable
onto J r Y, and its projection is ). The corresponding exterior differential system is
DQ D D and we can see that integral sections of D are exactly those sections
of r which are solutions of the (first order) differential equations related with the
dynamical form "Q.
Definition 13.3.4 We call DQ D D the Hamiltonian system, and the correspond-
ing first order differential equations the Hamilton equations of .
Remark 13.3.5 As we have seen, if p2 d D 0 then H" D 0 and " is locally
variational. This means that the corresponding .rC1/-order equations for sections
of are variational as they stand, and they come as EulerLagrange equations from
local Lagrangians D A ". Since generally pQ 2 d 0, H"Q 0, and the Hamilton
form "Q need not be locally variational. If, however, pQ 2 d D 0 then "Q D pQ 1 is locally
variational, and the Hamilton equations are first order EulerLagrange equations of
local first order Lagrangians Q D A "Q.
It is worth note that by Theorem 13.2.8 condition p2 d D 0 guarantees that
" D p1 has (at least a local) closed Lepage equivalent 0 . Thus, in the variational
case, we can always find (at least locally) variational Hamilton equations. Note
that D and D0 are different (the Hamilton equations are different), however, the
holonomic solutions of D and D0 are the same (indeed, p1 D p1 0 D ").
Remark 13.3.6 (Calculus of Variations) If, in particular, one has a Lepage manifold
.r ; / where is closed, then both " and "Q are locally variational. This means
that we have an r-th order variational principle over the fibred manifold for
" (Lagrange theory) and an associated zero order variational principle over the
fibred manifold r for "Q (Hamilton theory). (If r D 0 they, of course, coincide.)
The relationship between solutions of the EulerLagrange equations determined by
" and Hamilton equations determined by "Q then can be characterized as follows:
holonomic solutions of the Hamilton equations coincide with r-jet prolongations of
extremals.
The case of closed will be discussed in more detail in Sect. 13.5.
13.3.2 Regular Lepage Manifolds
From the definition of D we can see that at each point jrx 2 J r Y, the rank of D is
less or equal to the dimension of the bundle Vr of r -vertical vectors.
Definition 13.3.7 A Lepage manifold .r ; / (resp. a Lepage .n C 1/-form on

J r Y) is called regular if
rank D D dim Vr D the fibre dimension over X : (13.29)
Proposition 13.3.8 .r ; / is regular if and only if the map
Vr 3 ! i 2 n .J r Y/ (13.30)
mapping r -vertical vector fields to n-forms on J r Y, is injective.

Proof If is regular then the n-forms J D i@=@y J , where 1 6 6 m and
J D f j1 : : : jk g, 1 6 j1 6 6 jk 6 n, 0 6 k 6 r, generating D , are linearly
independent at each point of J r Y. Hence i D J J D 0 means that J D 0 for all
and J, proving that D 0.
Conversely, if the mapping (13.30) is injective, then J J D 0 implies that

J D 0 for all and J, 1 6 6 m and J D f j1 : : : jk g, 1 6 j1 6 6 jk 6 n,
0 6 k 6 r. This means that the n-forms J are linearly independent, so that
rank D D dim Vr . t
u
It is worth compare regular Lepage manifolds with symplectic and multisym-
plectic manifolds6:
The Cauchy distribution of a regular Lepage .n C 1/-form need not be (and in
applications usually is not) trivial (see also Example 13.3.1 below).
Every multisymplectic manifold is a regular Lepage manifold of order zero.
If dim X D 1, r > 0, and dim J r Y is odd, then every Lepage manifold .r ; /
with closed and regular is fibrewise symplectic (i.e., restricted to each fibre
r1 .x/ is a symplectic form).
Example 13.3.1 ([36]) Consider the fibred manifold Rn Rm ! Rn where m is
even, and the .n C 1/-form
1
D B dy ^ !0 C B1 dy ^ dy ^ !1 ; B1 D B1 ; det B1 0:
2
(13.31)

Since rank D rank B ; B1 ; 0; : : : ; 0 Dm, .0 ; / is a regular Lepage manifold.
The corresponding dynamical form is " D B C B1 y1 dy ^!0 . Computing i D
0 for a vector field
@ @
D N i C ; (13.32)
@xi @y
6
A differential form ! of degree q > 2 on a manifold M is called multisymplectic if it is closed
and nondegenerate (meaning that the Cauchy distribution of ! is zero, i.e. the condition i ! D 0
implies D 0) [8].
336 O. Rossi
and taking into account that !11 D 0, we get the components of ,
D N 1 M B ; 1 6 6 m; N 2 D D N n D 0; (13.33)
where M is the inverse matrix to B1 , and the condition B D 0. However,

the last condition with as above takes the form of identity, since M B B is
identically zero. Summarizing, we have obtained that the regular form has a one-
dimensional kernel spanned by the everywhere non-zero vector field
@ @
1
M B : (13.34)
@x @y
13.3.3 Helmholtz Symmetries
We shall close this section with an important result which yields a relationship
between non-variational and variational Lepage manifolds (and hence between the
corresponding non-variational and variational equations).
Definition 13.3.9 Let .r ; / be a Lepage manifold, " D p1 the corresponding
dynamical form. A vector field is called Helmholtz symmetry if is a contact
symmetry, and the flow of leaves invariant the Helmholtz form H" .
Immediately from the definition it follows that a contact symmetry is a
Helmholtz symmetry if and only if
L H" D 0 : (13.35)
The meaning of Helmholtz symmetries comes out by the following theorem:

Theorem 13.3.10 ([54]) Let " be a dynamical form such that H" 0. Let be a
contact symmetry such that L H" D 0. Then L " is a locally variational dynamical
form.
If .r ; / is a non-variational Lepage manifold, i.e. such that p2 d 0,
then Eq. (13.35) where " D p1 possibly provides us with nontrivial Helmholtz
symmetries. By the above theorem, every such a symmetry assigns to the Lepage
manifold .r ; / another Lepage manifold, and the new Lepage manifold is
variational.
More precisely, we can state the following:
Theorem 13.3.11 Let .r ; / be a Lepage manifold, and assume that p2 d 0.
Let be a -projectable vector field on Y such that D J rC1 leaves the form
p2 d invariant, i.e., let L p2 d D 0. Then .r ; 0 / where 0 D LJ r , is a Lepage
manifold, and p2 d 0 D 0.
Proof We shall use the fact that the Lie derivative along a contact symmetry
preserves the decomposition of differential forms into contact components (cf.
formula (13.2)). By the preceding theorem, "0 D L p1 is a locally variational

dynamical form. The .n C 1/-form 0 D LJ r satisfies p1 0 D p1 LJ r D
LJ rC1 p1 D LJ rC1 ", which is a dynamical form. Moreover, p2 d 0 D p2 dLJ r D
p2 LJ r d D LJ rC1 p2 d D 0, proving that .r ; 0 / is a variational Lepage
manifold. t
u
13.4 Hamiltonian Systems
13.4.1 Lepage Manifolds of Order Zero
Definition 13.4.1 By a Hamiltonian system we shall mean a Lepage manifold of

order zero.
By definition, a Hamiltonian system is determined by a .n C 1/-form defined
on the total space Y of a fibred manifold W Y ! X over an n-dimensional base
X. Hamilton equations are then first order differential equations for sections of
arising from the dynamical form " D p1 on J 1 Y (called the Hamilton .n C 1/-form
in this context), or as integral sections of the exterior differential system D on Y,
locally generated by n-forms D i@=@y , 1 6 6 m.
We note that a Hamiltonian system .0 ; / is regular if rank D D m D dim Y
dim X.
The local structure of Hamiltonian systems has been clarified by Hakov and
Krupkov in [36]:
Denote by .xi ; y / local fibred coordinates on Y. Then can be expressed in the
form
1 i1 1
D B dy ^ !0 C B 1 dy ^ dy1 ^ !i1 C Bi 1i21 2 dy ^ dy1 ^ dy2 ^ !i1 i2
2 3
1
CC Bi1 :::in dy ^ dy1 ^ ^ dyn ^ !i1 :::in ; (13.36)
.n C 1/ 1 :::n
where the components are totally antisymmetric in both the lower and the upper
indices; note that this implies that they are also totally symmetric with respect to
j :::jp :::jq :::jk j1 :::jq :::jp :::jk
the interchange of pairs of indices as follows: B 11 ::: p :::q :::k D B 1 :::q :::p :::k . The
corresponding Hamilton .n C 1/-form then reads:
" D E ! ^ !0 ; (13.37)
where
E D B C Bi 11 yi11 C Bi 1i21 2 yi11 yi22 C C Bi 1:::i1 :::

n
y1 : : : yinn ;
n i1
(13.38)
338 O. Rossi
and we note that the functions E satisfy the antisymmetry conditions
@E @E
D : (13.39)
@yi @yi
Using formula (13.38) we can express in terms of components of " [36]:
X
n
1 @k E

1;0 D E ! ^ !0 C 1
1 k
k ! ^ ! ^ ^ ! ^ !i1 :::ik :
kD1
k.k C 1/ @y i1
: : : @y ik
(13.40)
Formula (13.38) yields the Hamilton equations:

B C Bi 11 yi11 C Bi 1i21 2 yi11 yi22 C C Bi 1:::i1 :::
n
y 1
n i1
: : : y n 1
in J D 0 ; (13.41)
where 1 6 6 m, and the functions B satisfy the antisymmetry conditions

mentioned above. Note that the most general Hamilton PDEs are polynomial in the
first derivatives.
The associated exterior differential system D is generated by the n-forms
i@=@y
Xn
1 i1 :::ik (13.42)
D B !0 C B 1 :::k dy1 ^ ^ dyk ^ !i1 :::ik ; 1 6 6 m:
kD1
k
Obviously, we have the following explicit characterization of regular Hamilto-

nian systems:
Proposition 13.4.2 ([36]) Regularity of is equivalent with any of the following
conditions:
(1) The n-forms , 1 6 6 m, are linearly independent at each point of their
domain of definition.
(2) The rank of the matrix

B Bi 11 Bi 1i21 2 : : : Bi 1:::i1 :::
n
n
(13.43)
where labels rows and the other sets of indices label columns, is maximal and
equal to m D dim Y dim X at each point of Y.
(3) The rank of the matrix
!
@E @2 E @n E
E : : : (13.44)
@yi11 @yi11 @yi22 @yi11 : : : @yinn
where labels rows and the other sets of indices label columns, is maximal and
equal to m D dim Y dim X at each point of Y.
Corollary 13.4.3 ([36]) For be regular any of the following conditions is suffi-
cient:
!
@k E
rank D m; 1 6 k 6 n ; 1 6 i1 6 i2 6 ik 6 n : (13.45)
@yi11 : : : @yikk
Also for the following theorem and its proof we refer to [36]:
Theorem 13.4.4 ([36]) If is regular, then the ideal generated by D is closed.
Let us now assume that a Hamiltonian system .0 ; / satisfies p2 d D 0. Then
as we know, the Hamilton equations are variational, i.e. the Hamilton form " D p1
is locally variational: its components E satisfy the first order Helmholtz conditions
@E @E @E @E @E

C D 0; C dj D 0 : (13.46)
@yj @yj @y @y @yj
Theorem 13.2.8 also guarantees that there exists a closed .n C 1/-form giving rise
to ". However, in the zero order case a stronger result holds true, saying that every
variational Hamiltonian system is closed. More precisely:
Theorem 13.4.5 Given a Hamiltonian system .0 ; / on Y, the following condi-
tions are equivalent:
(1) p2 d D 0
(2) is closed.
The theorem means that if an .nC1/-form on Y satisfies p2 d D 0 then also all
the other contact components are zero, pk d D 0 for 3 6 k 6 n C 1. This property
comes from the uniqueness of a projectable (onto Y) Lepage equivalent of " D p1 ,
proved in [36]:
Theorem 13.4.6 ([36]) Let " be a dynamical form on J 1 Y such that " D p1 for
some .n C 1/-form on Y. The following conditions are equivalent:
(1) " is locally variational.
(2) the .n C 1/-form on Y is unique and satisfies d D 0.
(3) the components of " satisfy the (first order) Helmholtz conditions, and is given
by formula (13.36), resp. (13.40).
By the above theorem, variational Hamiltonian systems come from local first
order Lagrangians
D A " D hA D h (13.47)
where is a local n-form on Y such that d D .

340 O. Rossi
13.4.2 Legendre Coordinates and Poisson Structures
There is an important case of Hamiltonian systems represented by that is at most

2-contact. Hence,
1 @E
1;0 D E ! ^ !0 C ! ^ ! ^ !i ; (13.48)
2 @yi
where the antisymmetry relation (13.39) holds, so that
1
D B dy ^ !0 C Bi dy ^ dy ^ !i (13.49)
2
where Bi D Bi . Since in this case E D B C Bi yi , the Hamilton equations

are quasilinear first order PDEs:

B C Bi yi J 1 D 0 ; (13.50)
and D is generated by the n-forms
D B !0 C Bi dy ^ !i ; (13.51)
where 1 6 6 m, and the functions Bi satisfy the antisymmetry conditions

mentioned above.
Assume that
m
dD (13.52)
nC1
is an integer, i.e., assume that the base and fibre dimensions satisfy the relation
d C dn D m. Next, assume that around each point y 2 Y there is a fibred chart
.xi ; yK ; piK /, where 1 6 i 6 n, 1 6 K 6 d, such that
p2 D Ki ^ ! K ^ !i ; (13.53)
(summation over K D 1; : : : ; d) where ! K and Ki are the corresponding basic

contact forms on J Y, i.e., in the associated coordinates x ; y ; pK ; yj ; pK;j on J 1 Y,
1 i K i K i
! K D dyK yKj dx j ; Ki D dpiK piK;j dx j : (13.54)
We shall call coordinates of this kind Legendre coordinates on the Lepage manifold
.0 ; /.
In Legendre coordinates we have

1;0 D AK ! K C AKi Ki ^ !0 C Ki ^ ! K ^ !i ; (13.55)
that is,
D HK dyK ^ !0 HiK dpiK ^ !0 C dpiK ^ dyK ^ !i (13.56)
(the choice of the minus signs is a matter of convention), where

j
HK D AK pKj ; HiK D AKi C yKi : (13.57)
We can see that:

Proposition 13.4.7 If admits Legendre coordinates, then is regular, and
p3 d D 0 : (13.58)
Proof p3 d D 0 is obvious from (13.55). If is a vertical vector field then D

K @=@yK C Ki @=@piK , and the condition i D 0 yields K D 0, Ki D 0, i.e. D 0,
proving that is regular. t
u
Computing D we obtain Hamilton equations in Legendre coordinates as
follows:
j
@yK @pK
D HiK ; D HK : (13.59)
@xi @xj
It is worth note that p3 d D 0 does not mean that should be closed. Therefore
Legendre coordinates are defined also for a class of non-variational Hamiltonian
systems.
Denote D C dpiK ^ dyK ^ !i ; the splitting on the right-hand side is generally
not invariant, i.e. is a local .n C 1/-form. In case that is closed we have d D 0,
hence D dH ^ !0 for a (local) function H (called Hamilton function), and the
Hamilton equations (13.59) take the familiar form
j
@yK @H @pK @H
D i ; D K : (13.60)
@xi @pK @xj @y
A covering of .0 ; / by Legendre coordinates induces a multi-Poisson struc-

ture: precisely, on Y there arises the following tensor field, to be called the Poisson
bivector 1-form:
@ @
D ^ K dxi : (13.61)
@pK @y
i
With help of we can introduce a directional bracket of (possibly local)

functions on Y as follows: Let v be a nowhere zero vector field on X, and denote
by vQ a lift of v to Y. Given f ; g 2 00 we set:
f f ; ggv D .df ; dg; v/

Q : (13.62)
342 O. Rossi
The bracket obviously does not depend on the lift of v and it holds

@ @
f f ; ggv D ^ dx i
.df ; dg; v/
@piK @yK
(13.63)
@f @g @g @f
D v i D .ivQ /.df ; dg/
@piK @yK @piK @yK
Theorem 13.4.8 .00 ; f:; :gv / is a Poisson algebra.

Proof We have to prove that f:; :gv satisfies the following conditions:
(1) f f ; ggv D fg; f gv (antisymmetry),
(2) f f ; ghgv D gf f ; hgv C hf f ; ggv (Leibniz rule),
(3) ff f ; ggv ; hgv C ffh; f gv ; ggv C ffg; hgv ; f gv D 0 (Jacobi identity),
for all f ; g; h 2 00 .
The antisymmetry is obvious, and the Leibniz rule easily follows from the
definition. Let us show that also the Jacobi identity holds true. We have:
!
@f f ; ggv @h @h @f f ; ggv
ff f ; ggv ; hgv D j vj
@pL
j @y L
@pL @yL
!
@2 f @g @f @2 g @2 g @f @g @2 f @h
D C i j i
j i @yK @pK @y @pL @pL @pK @y
j
@pK @y @pL @yL
j
@pL @pK K i K K
2 !
@h @ f @g @f @2 g @2 g @f @g @2 f
j C i vi vj
@pL @yL @pK @yK @pK @yK @yL @yL @piK @yK @piK @yK @yL
i
Cycling f ; g; h and summing up then yields zero, as desired. t

u
Definition 13.4.9 We shall call the bracket f:; :gv Poisson bracket in the direction
of v.
Given (any) 1-form on Y, there arises a vector field X;v , defined by
X;v .g/ D .; dg; v/ for every g 2 00 : (13.64)
We shall call X;v the Hamiltonian vector field of in the direction of v. We can see
that X;v is vertical and

@ @
X;v D Ki K K i v i ; (13.65)
@y @pK
where Ki and K are components of at dpiK and dyK , respectively. In particular,
for s of the form p1 D dV f , we get Hamiltonian vector fields associated with
functions in 00 :

@f @ @f @
Xf ;v D i @yK
K i vi : (13.66)
@pK @y @pK
We note that
f f ; ggv D Xf ;v .g/ : (13.67)
13.5 CartanLepage Manifolds in Field Theory
Many important examples of Lepage manifolds appear in physics, namely in

mechanics and field theories. In this section we shall present the Lepage manifolds
setting applicable to electromagnetism (Maxwell equations), gravity (Einstein
equations), gauge theories, and others. We refer also to [81] for other aspects and
further references.
13.5.1 Closed Lepage Forms
Equations of motion of classical fields and gravity are second order PDEs affine
in the second derivatives, and variational. This leads us to consider closed Lepage
manifolds of order one.
The local structure of Lepage .n C 1/-forms we are interested in, is as follows.
Theorem 13.5.1 ([70]) Let be a Lepage .n C 1/-form on J 1 Y such that " D p1
is a dynamical form affine in the second derivatives, and d D 0. Then in fibred
coordinates may be written as
D " C (13.68)
where is a closed at least 2-contact form, and where " is closed and completely
determined by ". An explicit expression for " is
@E @E
2;1 " D E ! ^ !0 C 1
! ^ ! ^ !j C ! ^ !i ^ !j C
2
@yj @yij
1 @n E
C ! ^ ! 1 ^ ^ ! n ^ !j1 jn
n.n C 1/ @yj11 @yjnn
1 @n E
C ! ^ ! 1 ^ ^ ! n1 ^ !pn ^ !j1 jn ;
.n/2 @yj11 @yjn1
n1
@yjnnp
(13.69)
344 O. Rossi
where the E satisfy Helmholtz conditions
@E @E
D 0;
@yij @y ij
@E @E @E
C 2dk D 0; (13.70)
@yi @yi @y ik
@E @E @E @E

C dj dj dk D 0 :
@y @y @yj @yjk
Poincar Lemma then yields:

Theorem 13.5.2 For a Lepage .n C 1/-form as above, local Lagrangians are
given by the VainbergTonti formula D hA D A ". In coordinates the set of all
local Lagrangians takes the form
Z

1
LDy E xi ; uy ; uyj ; uyjk du C dj
j ; (13.71)
0
where
j are arbitrary functions. The restriction of to a suitable open set U
satisfies
jU D d C d; (13.72)
where is the PoincarCartan form of a local Lagrangian for ", and is an at

least 2-contact n-form.
In view of these properties we can regard our Lepage manifold as a fibred
manifold, equipped with a class of locally equivalent Lagrangians. We therefore
2
S fL g of 2Lagrange functions, each defined on an open set U J Y,
have a family
such that U D J Y, and whenever U \ U ;, around every point of the
intersection we have L D L C dj ' j for some functions ' j . Note that in some
cases we can find a global Lagrangian in the family, but in general such a global
Lagrangian (even of higher order) does not exist. The obstructions come from the
topology of Y (see [3]).
13.5.2 De DonderHamilton Equations on Lepage Manifolds
Formula (13.72) shows that there is an important family of Lepage manifolds,

namely those with as short as possible, or, as close to the Cartan form as
possible. Since d is at most 2-contact, this concerns s which are at most
2-contact (leading to the choice of 2-contact and such that p3 d D 0).7 Requiring
be a first order 2-contact form, we are led to the following definition:
Definition 13.5.3 We say that .1 ; / is a CartanLepage manifold if is closed,
at most 2-contact and f! g-generated.8;9
Notice that in fibred coordinates we have

1 @E
2;1 D E ! ^ !0 C 2 dk f ! ^ ! ^ !j
j;k
@yj
(13.73)
@E
C f ! ^ !i ^ !j ;
i;j
@yij
i;j j;i j;i

where E satisfy the Helmholtz conditions, f D f D f are some first order
functions such that d D 0, and the condition
i;j i;p
@ @E @E @f @f
D C (13.74)
@ypj @yi @y i @yp @yj
(meaning that is projectable onto J 1 Y) holds true.

Keeping notations as in the previous sections, we shall consider the exterior
differential system D , and equations for its holonomic integral sections (Euler
Lagrange equations) as well as equations for all its integral sections (Hamilton
equations). We recall that in general Hamilton and EulerLagrange equations are
not equivalent, as there might exist Hamilton extremals that are not prolongations
of extremals.
It is worth notice that on a Lepage manifold both the EulerLagrange equations
and the Hamilton equations are independent of a choice of a concrete Lagrangian
for E.
We can see that is regular if rank D m C nm, and that in coordinates the
regularity condition now reads as follows:
Theorem 13.5.4 ([60]) is regular if and only if at each point of J 1 Y

@E
det f i;j 0: (13.75)
@yij
7
In general we cannot take simply d since need not be global.
8
As we shall see, in this case the Hamilton equations become of De Donder type.
9
This choice of in the Lepage class fg is the most simple. As we know, more generally we can
add to (any) at least 2-contact form, and/or the derivative of a 1-contact form, and the dynamical
form ", hence the EulerLagrange equations, remain the same. However, the Hamilton equations
then may no longer be of De Donder type.
346 O. Rossi
We have the following important theorem on canonical form of a CartanLepage

.n C 1/-form.
Theorem 13.5.5 Let .1 ; / be a CartanLepage manifold. Then around every
point in J 1 Y there is a neighbourhood U and functions H and p defined on U
j
such that
jU D dH ^ !0 C dp j ^ dy ^ !j : (13.76)
j
If, moreover, is regular, then the functions p are independent, meaning that
!
j
@p
rank D max D mn: (13.77)
@yk
Proof The theorem is proved with help of the Poincar Lemma. Since d D 0
we have locally D d, and we can set D A where A is the contact
homotopy operator. Then D hA is the VainbergTonti Lagrangian for the
locally variational dynamical form " D p1 . Explicitly, if we denote the components
of " by
E D A C Bij yij ; Bij D Bji D Bij ; (13.78)
we have by (13.73)

D A 2F i yi dy ^ !0 C F ij y i dyj ^ !0
(13.79)
C F i dy ^ dy ^ !i C F ij dy ^ dyj ^ !i ;
where

1 @A @A
F i D F
i
D 2dN j f i;j ; (13.80)
4 @yi @y i
F ij D F
ji
D Bij f i;j ; (13.81)
and dN i denotes the cut total derivative dN i D di y ij @y@ . Later we shall need the
j
following identities yielded by d D 0:
!
@F i 1 @F
ik
@F
ik ij
@F @F
ik
C D 0; D : (13.82)
@yk 2 @y @y @yk @yj
Let j1x 2 J 1 Y. Using the definition of A , we obtain in a neighbourhood U of j1x

Z 1 Z 1 Z 1

ij
DA D y .A /du 2y yi i
F u du C y i yj F u du !0
0 0 0
Z 1 Z 1
i ij
2y F u du C yj F u du dy ^ !i
0 0
Z 1
ij
C y F u du dyj ^ !i ;
0
(13.83)
where W u; xi ; y ; y j ! xi ; uy ; uy j for u 2 0; 1.
We shall show that one can express in the form
D H!0 C pi dy ^ !i C d
(13.84)
for some functions H, pi , and .n 1/-form

. Shrinking U if necessary, we consider
the map

N W 0; 1 U 3 v; xi ; y ; y j ! xi ; y ; vy j 2 U J 1 Y; (13.85)
and put
Z 1 Z 1
ij
f D i
yj y F u du N dv C ' i .xp ; y /; (13.86)
0 0
where ' i are arbitrary functions. Using (13.82) we obtain that the f i satisfy
Z 1
@f i ij
D y F u du : (13.87)
@yj 0
In this way we obtain:
0 D A d. f i !i / D H!0 C pi dy ^ !i (13.88)
where
Z 1 Z 1

i ij @f i
pi D 2y F u du yj F u du
0 0 @y
Z 1 Z 1
H D y .A /du C 2y yi F i u du (13.89)
0 0
Z 1 ij @f i
y i yj F u du C i :
0 @x
348 O. Rossi
The n-form 0 satisfies
d0 D dH ^ !0 C dpi ^ dy ^ !i D dA D ; (13.90)
as desired. t
u
Definition 13.5.6 We call (13.76) canonical form of , and the functions H and pi
Hamiltonian and momenta of , respectively.
Note that
h0 D hA hd. f i !i / (13.91)
is a family of local first order Lagrangians for " (i.e. equivalent with the second
order VainbergTonti Lagrangian hA D A "). The Lagrangians are determined
up to hd.' i !i / D di ' i !0 .
For 0 D h0 we have 0 D L0 !0 where
L0 D H C pi y i ; (13.92)
and
@L0
pi D I (13.93)
@y i
the latter follows from formulas for H and pi (13.89) if we use (13.82). Moreover,
since
@pi @E
D F D B f D
ij ij i;j
f i;j ; (13.94)
@yj @yij
the regularity condition (13.75) takes the form

! !
@pi @2 L0
det D det 0: (13.95)
@yj @y i @yj
We can see that if 0 and N 0 are two Lagrangians defined by the canonical form
of (13.76) then locally N 0 D 0 C hd', where 0 D h0 for some 0 as above, and
' is a local horizontal .n 1/-form on Y. We note that if one of these Lagrangians
is regular (in the sense of the regularity condition (13.95)) then all the Lagrangians
in the family are regular (and this is if and only if is regular). Thus the functions
' i in (13.86) play the role of admissible gauge functions.
Remark 13.5.7 It is worth notice that for n D dim X > 1, not every first order
Lagrangian for E has the pleasant properties as have the Lagrangians 0 related
with the canonical form of . Namely, it may happen that for a local first order
Lagrangian D L!0 the momenta . pL /i D @L=@y i and Hamiltonian HL D L C

. pL /i y i for no ' are equal to pi and H, defined by (13.89). Moreover, even if is
regular, L need not satisfy the condition
!
@2 L
det 0: (13.96)
@y i @yj
For discussion on equivalent first order Lagrangians such that one satisfies (13.96)
and the other does not we refer to [87].
We have seen that for regular , the map xi ; y ; y j ! xi ; y ; p is a local
j
coordinate transformation on J 1 Y. In view of Sect. 13.4.2 the new coordinates are

Legendre coordinates.
Summarizing, we have obtained:
Theorem 13.5.8 Let .1 ; / be a CartanLepage manifold. Then in a neighbour-
hood of every point in J 1 Y there exists a first order Lagrangian for " D p1 .
Moreover, if H and pi are a Hamiltonian and momenta of defined by the
canonical form of then locally there exists a first order Lagrangian for " such that
H D L C pi y i and p i D @y@L . Any such a Lagrangian is of the form D 0 C hd',
i
where 0 D h0 and ' is a local horizontal .n 1/-form on Y.
If, moreover, is regular, then there exists a covering of J 1 Y by Legendre
coordinates, and a Poisson tensor on J 1 Y; they are defined by the canonical
form (13.76) of . Moreover, every Lagrangian D h0 C hd' as above satisfies
the regularity condition (13.95).
If we now compute the generators of the exterior differential system D in
Legendre coordinates, we obtain the Hamilton equations of in the form
@.y / @H @. pi / @H
D i ; D : (13.97)
@xi @p @xi @y
These equations are known also as De DonderHamilton equations.

The meaning of the regularity condition is as follows.
Theorem 13.5.9 On a Lepage manifold .1 ; / where is closed, at most
2-contact, f! g-generated and regular, the EulerLagrange and Hamilton
equations are equivalent. Explicitly, if is an extremal then J 1 is a Hamilton
extremal; conversely, every Hamilton extremal is of the form J 1 where is an
extremal.
An easy proof of this theorem comes from analysing the solutions of the exterior
differential system D for (13.73) satisfying (13.75).
Note that, by construction, explicit formulas for the Hamiltonian and momenta
come from an integration procedure using the Poincar Lemma and are determined
by rather than by a particular Lagrangian.
350 O. Rossi
13.5.3 Dual Jet Bundles
Our aim now is to provide a dual picture, analogously to symplectic mechanics

where one considers the cotangent bundle T M equipped with the canonical
symplectic form. The present setting is due to [10] (see also [83], and [67] for higher
order mechanics).
Given a fibred manifold W Y ! X we shall explore that 1;0 W J 1 Y ! Y is an
affine bundle, modelled on the vector bundle VX Y T X ! Y, where VX Y ! Y
is the bundle of vectors tangent to Y and vertical over X. For every fixed y 2 Y, the
fibre Jy1 Y of 1;0 over y is an affine space of dimension nm, hence it is associated
with a vector space of dimension .nm C 1/, its extended dual, whose elements are
the real-valued affine maps on Jy1 Y. Collecting the extended duals and taking the
union over y 2 Y we obtain a manifold of dimension n C m C nm C 1, denoted by
J Y, and called the extended dual of J 1 Y.

If
2 J Y and
W Jy1 Y ! R, we put 1;0 .
/ D y. Then 1;0 W J Y ! Y becomes
a vector bundle.

Each fibre Jy Y is a vector space containing a distinguished subspace of constant
maps. The quotient space will be denoted by Jy Y, and the union over y will be
denoted by J Y and called the reduced dual of J 1 Y. We have dim J Y D dim J Y
1 D dim J 1 Y.
Next we denote by W J Y ! J Y the quotient map, and by 1;0
W J Y ! Y the
induced vector bundle projection.
Let f.y/ denote the restriction of a function to the fibre
over y. Given a chart
.U; .xi ; y // on Y and y 2 U, the affine functions 1.y/ ; y i.y/ form a basis for the

vector space Jy Y. Denote by P.y/ ; Pi .y/ the dual basis of Jy Y ; in this way, we

i
get smooth functions P, Pi locally defined i on J Yi such that x ; y ; P; P are local
i
coordinates on the extended dual. Then x ; y ; P are coordinates on the reduced

dual J Y.
It can be shown that with a choice of a volume element on X, J Y is diffeomor-
phic to the bundle of n-forms on J 1 Y, locally generated by .!0 ; dy ^ !i / (see [81]
for the proof). The global .n C 1/-form
D dP ^ !0 C dPi ^ dy ^ !i (13.98)
then can be regarded as a canonical multisymplectic form on J Y. Now, any local

section h of the projection W J Y ! J Y gives rise to a local closed .n C 1/-form
h D h D dH ^ !0 C dPi ^ dy ^ !i ; (13.99)
on J Y, where we have introduced the function H D . P h/ defined on

dom h J Y (an abstract Hamiltonian). With h the manifold J Y becomes a
regular CartanLepage manifold of order zero.
We can consider the exterior differential system Dh D fi h g where runs over

vector fields on J Y, vertical over X, so that integral sections W X ! J Y of Dh
satisfy the De DonderHamilton equations

j
@.y / @H @ P @H
D ; D : (13.100)
@xi @Pi @xj @y
From (13.99) it is clear that h induces a Poisson structure on J Y defined by

the Poisson tensor
@ @
D ^ dxi : (13.101)
@Pi @y
So far we have constructed a universal (abstract) Hamiltonian bundle, canon-

ically associated with any jet bundle 1 W J 1 Y ! X. Now we shall establish
a connection between an abstract Hamiltonian system and a concrete variational
system on a Lepage manifold.
So, assume that our jet bundle 1 is a CartanLepage manifold, equipped with
a Lepage .n C 1/-form . As above, " D p1 will be the corresponding dynamical
form (EulerLagrange form). On the extended dual J Y we have the canonical
symplectic form corresponding to a choice of volume form on X. Both the sides
of the picture can be related by means of a map Leg W J 1 Y ! J Y fibred over the
identity of Y defined by
Leg D : (13.102)
We call Leg an extended Legendre map of and Eq. (13.102) a duality equation.
We may also construct a composite map leg W J 1 Y ! J Y by setting leg D
Leg . This will be called the reduced Legendre map.
The map Leg is not unique; all such maps are parametrized by gauge functions
'.xi ; y /. A similar remark applies to leg .
In the regular case we can choose Legendre coordinates on J 1 Y. Using these
Legendre coordinates, and the canonical coordinates on J Y, the map leg is
represented by the identity mapping. We immediately obtain the following result.
Theorem 13.5.10 ([83]) If is regular, then every extended Legendre map is an
immersion and every corresponding reduced Legendre map is a local diffeomor-
phism.
We say that is hyper-regular if there is an extended Legendre map Leg which
is defined globally and which has the property that the corresponding reduced
Legendre map leg is a diffeomorphism. Any such map gives rise to a global
Hamiltonian section h by setting h D Leg leg1 . If is regular rather than hyper-
regular, then we may construct local sections h in this way. The negative coordinate
representative of such a section h will be the function H D . Ph/, a Hamiltonian
352 O. Rossi
function corresponding to . Corresponding to a choice of the extended Legendre

map we get a family of Hamiltonian functions parametrized by '.
With help of the definition of the Legendre maps, and using Legendre coordi-
nates, we easily obtain the following result.
Theorem 13.5.11 ([83]) For a hyper-regular on J 1 Y the following assertions are
true:
(1) Leg D leg h D :
(2) rank h D rank Dh D rank D rank D D m C nm:
(3) leg Dh D D .
(4) If W X ! J Y is an integral section of Dh , then leg1
D J 1 where is a
section of W Y ! X, and it is an integral section of D .
(5) Every integral section of D is of the form J 1 , and D leg J 1 is an
integral section of Dh .
13.6 Lepage Manifolds in Mechanics: The Metric,

Symplectic and Poisson Structures
In the last section we shall consider a fibred manifold W Y ! X over a base

of dimension 1. In this case equations for sections of are ordinary differential
equations, and Lepage .n C 1/-forms are 2-forms. We emphasize that for n D
1 there occurs a substantial simplification: every dynamical form has a unique
(global) Lepage equivalent. This means that we have a one-to-one correspondence
between dynamical forms and Lepage manifolds. Moreover, there is one-to-one
correspondence between Lepage manifolds defined by closed forms and variational
equations (precisely speaking, locally variational dynamical forms):
Theorem 13.6.1 ([56]) Let .r ; / be a Lepage manifold and assume that is
closed. Then " D p1 is locally variational. Moreover, is a unique Lepage
equivalent of ".
Conversely, let " be a locally variational dynamical form on J s Y, s > 0. Then
there exists a unique Lepage manifold .r ; ) such that " D p1 . It holds r D s 1
and d D 0.
Explicit formula for can be found in [56] (see also [59]).
In what follows we shall be interested in Lepage manifolds which are a
background for second order ordinary differential equations, and in particular, for
geometric mechanics. As usual in this situation, we shall take Y D R M, where
M is a smooth manifold, dim M D m, and X D R. Then J 1 Y D R TM,
J 2 Y D RT 2 M, etc. We shall use fibred coordinates adapted to the product structure
of Y, and a global coordinate on R. This means that the transformation rules between
.t; q / and .Nt; qN / satisfy Nt D t and @Nq =@t D 0. The associated coordinates on
J 1 Y and J 2 Y will be denoted by .t; q ; qP / and .t; q ; qP ; qR /, respectively. We set
! D dq qP dt, !P D d qP qR dt.
Apart from the general case, we shall be interested also in time-independent

mechanical systems. In order to define this concept we note that the global vector
field @=@t on R trivially lifts to R TM and its prolongations. We call a differential
form time-independent, if L@=@t D 0. We obtain that is time independent if and
only if " is time-independent, and this means that @E =@t D 0 for all .
Due to uniqueness of the Lepage equivalent in mechanics, Lepage manifolds
for second order ordinary differential equations are completely characterized by the
following theorem due to Krupkov and Prince:
Theorem 13.6.2 ([65])
(1) Every dynamical form " on J 2 Y has a unique Lepage equivalent. It is defined
on J 3 Y, and takes the coordinate form

@E 1 @E d @E @E
D E ! ^ dt C

! ^ !
@Pq 4 @Pq dt @Rq @Rq

1 @E @E
C
C ! ^ !P : (13.103)
2 @Rq @Rq
(2) If the second order ODEs are affine in the second derivatives, i.e. such that
E D A .t; q ; qP / C g .t; q ; qP /Rq (13.104)
then the Lepage equivalent is projectable onto J 2 Y and reads

1 @A @A @g @g
D A ! ^ dt C 2 2 P
q ! ^ !
4 @Pq @Pq @t @q
C g. / ! ^ d qP C g qR ! ^ dt

1 @g @g @g
C 2 qR ! ^ ! (13.105)
4 @Pq @Pq @Pq
where . /, resp. denotes symmetrization, resp. antisymmetrization in the

indicated indices.
(3) The Lepage equivalent of (13.104) is projectable onto J 1 Y if and only if
@g @g
g D g ; D : (13.106)
@Pq @Pq
In this case

1 @A @A
D A ! ^ dt C ! ^ ! C g ! ^ d qP : (13.107)
4 @Pq @Pq
354 O. Rossi
Definition 13.6.3 Dynamical forms characterized by property (3) of the above

theorem are called semi-variational.
Relations (13.106) are integrability conditions for a function f .t; q ; qP / such that
@2 f
g D ; (13.108)
@Pq @Pq
and which is determined up to a linear term. Moreover, it can be shown [59, 61] that
every regular semi-variational dynamical form " splits canonically into a sum of a
variational dynamical form "g , coming from a Lagrangian D T dt determined by
the g , and a first order dynamical form
(a force), so that the semi-variational
equations take the form
@T d @T

D
: (13.109)
@q dt @Pq
This means that splits to
D g C
(13.110)
where g is the Lepage equivalent of the EulerLagrange form of T, and

is the
Lepage equivalent of
. Note that (due to variationality) dg D 0.
If the matrix .g / is regular and time and velocity independent, it determines a
metric g on M. Then (up to the sign) T is the kinetic energy of the metric g, and
Eq. (13.109) take the (contravariant) form
qR C qP qP D g
; (13.111)
where is the Levi-Civita connection of g. If g is regular, time independent and

0-homogeneous in the velocities, it is a Finsler metric, and the corresponding
equations with
D 0 are equations for geodesics on the Finsler manifold M. In
a general case, we call a regular morphism g.t; q ; qP / satisfying the integrability
conditions (13.106) a variational metric (more about this see [61] and [80]).
We remind that " is variational if and only if p2 d H" D 0 (Helmholtz
conditions), or, equivalently, if is closed. Then the force
D
! ^ dt appearing
in Eq. (13.109) is a first-order locally variational form (i.e. satisfies the first-order
Helmholtz conditions). It can be shown that this is if and only if
takes the form of
Lorentz force (see, e.g., [61] or [59] for details).
As we have seen above, second order ordinary differential equations are modelled
on Lepage manifolds of order 2, or, if they are semi-variational or variational, on
Lepage manifolds of order 1.
In the sequel we shall consider a Lepage manifold .1 ; / over a fibred manifold
W R M ! R. Let us summarize what we already know: .1 ; / is endowed
with
the dynamical form " D p1 which is semi-variational or variational depending

on whether d 0 or d D 0,
the exterior differential system D , generated by the co-distribution 0 on RTM
spanned by 1-forms i where runs over all 1 -vertical vector fields on J 1 Y,
and having the same holonomic solutions as ".
Since is a 2-form, we also have
the characteristic distribution (and its annihilator 0 ) of on R TM. Recall
that
D spanf j i D 0g; (13.112)
and its annihilator consists of 1-forms i where runs over all vector fields on
R TM. Thus, , and both the distributions have the same holonomic
solutions which coincide with the solutions of the ODEs defined by ".
Moreover, if is closed (" is locally variational), then has a local canonical
form
D dH ^ dt C dp ^ dq (13.113)
where the functions H, p are defined as components in the canonical basis of the
Cartan form of a local Lagrangian D hA , i.e. D A D Hdt C p dq .
From now on, let us assume that .1 ; / is a regular Lepage manifold, i.e.
rank D dim V1 D 2m. Then the distribution has rank one and coincides
with the characteristic distribution . Moreover, we get the following additional
structures:
Variational metric g [61] defined by
@E
g D : (13.114)
@Rq
Indeed, by the regularity assumption on this matrix is regular. Consequently, on

every fibre of R M, g induces a velocity dependent metric. In case that is time
independent, g becomes time-independent and projects onto a velocity dependent
metric on M. This may be, for example, a Finsler metric if g is 0-homogeneous
in velocities, or a Riemannian metric on M if g does not depend on velocities.
A first order dynamical form
(a force), defined by the canonical splitting " D
"g C
; the force is variational or non-variational, depending on whether is
closed or not.
Global vector field on RTM, spanning the distribution D . It is defined
by i D 0 and dt. / D 1. is a semispray and its integral sections identify
with solutions of the equations determined by ". In coordinates,
@ @ @
D C qP g A : (13.115)
@t @q @Pq
356 O. Rossi
A splitting of the tangent bundle to R TM associated with (e.g. [84, 85, 88])
T.R TM/ D spanf g H V1;0 (13.116)
where V1;0 is the bundle of the very vertical vectors (vertical over R M) and

@ 1 @ @
H D span
C (13.117)
@y 2 @Pq @Pq
where D g A . This splitting induces a splitting of the cotangent bundle

which gives an adapted basis of 1-forms (dual basis to (; H ; @=@Pq /) as follows:
.dt; ! ; /, where
1 @
D d qP ! dt : (13.118)
2 @Pq
In the adapted basis the Lepage 2-form takes a simple expression
D g ! ^
: (13.119)
If is closed, time independent and regular: The splitting of (13.113) is

invariant and defines the symplectic form ! D dp ^ dq , and hence also the
Poisson structure D @p@ ^ @q@ on TM. Obviously, then splits into a vector
field on R and a time independent vector field on TM which is the Hamilton
vector field for !.
More generally, on a regular Lepage manifold with closed (locally variational
dynamical form ", corresponding to variational equations), apart from a (general-
ized) metric structure g, we have also fibrewise symplectic and Poisson structures.
Considering the dual jet bundle, we get in the same way as in the case of field
theory in the previous section, the duality between the Lepage manifold .1 ; /
and the bundle .J Y; h /. For a detailed exposition we refer to [67].
Lepage manifolds arise wherever differential equations of dynamical type (i.e.
equations defined by dynamical forms) are concerned. In particular, they are behind
any system of first-order ODEs (a vector field, dynamical system) and regular
higher-order ODEs (a semispray, a spray), a symplectic structure, or behind any
extremal principle (e.g. kinetic energy, Finsler function, etc.). Let us discuss the
latter two cases in more detail:
Lepage Manifolds in Riemann and Finsler Geometry Let .M; g/ be a Rie-
mannian manifold (actually, the signature of the metric is inessential). Denote by
.xi /, 1 6 i 6 m, local coordinates on M. We have on TM the induced kinetic
energy, T D 12 gij xP i xP j , which is a first-order Lagrangian for the fibred manifold
W RM ! R. The EulerLagrange equations, that are equations for the geodesics
on M, are represented by a second-order vector field (semispray) on R TM. If
we denote by .dt; ! i ; i / the basis of one forms on R TM, dual to the natural
splitting of R TM induced by , we get the Lepage 2-form D gij ! i ^ j ,

which is closed. We easily check that spans the kernel of , so that .1 ; /, where
1 W R TM ! R, is the Lepage manifold naturally related with .M; g/.
A similar situation arises in Finsler geometry. Denote by T o M the slit tangent
bundle of M (i.e. let T o M be TM with the zero section excluded). If F is a Finsler
function on T o M, and g is the corresponding Finsler metric tensor
@2 F 2
gij D ; (13.120)
@Pxi @Px j
the equations for Finsler geodesics are EulerLagrange equations of the Lagrangian
F 2 . Let be the semispray on R T o M defined by the EulerLagrange equations,
and .dt; ! i ; i / the basis of one forms on R T o M, dual to the natural splitting
of R T o M induced by . The Lepage manifold associated with .M; F/ is .1 ; /
where 1 W R T o M ! R, and D gij ! i ^ j . Again, is closed.
Notice that, in both the cases, invariantly splits to a sum dH ^ dt C !, where
! is the related symplectic form.
Acknowledgements Research supported by grant No. 14-02476S Variations, Geometry and

Physics of the Czech Science Foundation, and by the IRSES project LIE-DIFF-GEOM (EU FP7,
nr 317721).
References
1. R. Abraham, J.E. Marsden, Foundations of Mechanics, 2nd edn. (The Benjamin/Cummings

Publishing Company, Reading, 1978)
2. I. Anderson, The Variational Bicomplex, Utah State University, Technical Report (1989)
3. I. Anderson, T. Duchamp, On the existence of global variational principles. Am. J. Math. 102,
781867 (1980)
4. V.I. Arnold, V.V. Kozlov, A.I. Neishtadt, Mathematical Aspects of Classical and Celestial
Mechanics (Springer, Berlin, 2006)
5. D.E. Betounes, Extension of the classical Cartan form. Phys. Rev. D 29, 599606 (1984)
6. G.E. Bredon, Sheaf Theory (McGraw-Hill, New York, 1967)
7. R. Bryant, P. Griffiths, D. Grossmann, Exterior Differential Systems and Euler-Lagrange
Partial Differential Equations. Chicago Lectures in Mathematics (University of Chicago Press,
Chicago, 2003)
8. F. Cantrijn, L.A. Ibort, M. de Len, Hamiltonian structures on multisymplectic manifolds.
Rend. Sem. Mat. Univ. Pol. Torino 54, 225236 (1996)
9. C. Carathodory, ber die Variationsrechnung bei mehrfachen Integralen. Acta Szeged Sect.
Scient. Mathem. 4, 193216 (1929)
10. J.F. Cariena, M. Crampin, L.A. Ibort, On the multisymplectic formalism for first order field
theories. Diff. Geom. Appl. 1, 345374 (1991)
11. . Cartan, Lecons Sur Les Invariants Intgraux (Hermann, Paris, 1922)
12. M. Crampin, D.J. Saunders, The Hilbert-Carathodory and PoincarCartan forms for higher-
order multiple-integral variational problems. Houston J. Math. 30(3), 657689 (2004)
13. M. de Len, P.R. Rodrigues, Generalized Classical Mechanics and Field Theory (North-
Holland, Amsterdam, 1985)
358 O. Rossi
14. M. de Len, P.R. Rodrigues, Methods of Differential Geometry in Analytical Mechanics

(North-Holland, Amsterdam, 1989)
15. M. de Len, J.C. Marrero, D.M. de Diego, A new geometric setting for classical field theories.
Banach Cent. Publ. 59, 189209 (2003)
16. P. Dedecker, Calcul des variations, formes diffrentielles et champs godsiques, in Coll.
internat. du C.N.R.S., Strasbourg, 1953 (C.N.R.S., Paris, 1954), pp. 1734
17. P. Dedecker, On the generalization of symplectic geometry to multiple integrals in the calculus
of variations, in Lecture Notes in Mathematics, vol. 570 (Springer, Berlin, 1977), pp. 395456
18. P. Dedecker, Sur le formalisme de HamiltonJacobiE. Cartan pour une intgrale multiple
dordre suprieur. C. R. Acad. Sci. Paris Sr. I 299, 363366 (1984)
19. P. Dedecker, Existe-t-il, en calcul des variations, un formalisme de Hamilton-Jacobi-. Cartan
pour les intgrales multiples dordre suprieur? C. R. Acad. Sci. Paris Sr. 1 298, 397400
(1984)
20. A. Echeverria-Enriquez, M.C. Muoz-Lecanda, N. Romn-Roy, Multivector field formulation
of hamiltonian field theories: equations and symmetries. J. Phys. A Math. Gen. 32, 84618484
(1999)
21. A. Echeverria-Enriquez, M.C. Muoz-Lecanda, N. Romn-Roy, On the multimomentum
bundles and the Legendre maps in field theories. Rep. Math. Phys. 45, 85105 (2000)
22. C. Ehresmann, Les prolongements dune varit diffrentiable: 1. calcul des jets, prolongement
principal. C. R. Acad. Sci. Paris 233, 598600 (1951)
23. C. Ehresmann, Les prolongements dune space fibr diffrentiable. C. R. Acad. Sci. Paris 240,
17551757 (1955)
24. M. Ferraris, M. Francaviglia, On the global structure of the Lagrangian and Hamiltonian
formalisms in higher order calculus of variations, in Proceedings of the International Meeting
on Geometry and Physics, Florence, 1982, ed. by M. Modugno (Pitagora, Bologna, 1983), pp.
4370
25. P.L. Garcia, The Poincar-Cartan invariant in the calculus of variations. Symp. Math. 14, 219
246 (1974)
26. P.L. Garcia, J. Muoz, On the geometrical structure of higher order variational calculus, in
Modern Developments in Analytical Mechanics I: Geometrical Dynamics, ed. by S. Benenti,
M. Francaviglia, A. Lichnerowicz. Proceedings of the IUTAM-ISIMM Symposium, Torino,
1982 (Accad. delle Scienze di Torino, Torino, 1983), pp. 127147
27. G. Giachetta, L. Mangiarotti, G. Sardanashvily, New Lagrangian and Hamiltonian Methods in
Field Theory (World Scientific, Singapore, 1997)
28. G. Giachetta, L. Mangiarotti, G. Sardanashvily, Covariant Hamilton equations for field theory.
J. Phys. A Math. Gen. 32, 66296642 (1999)
29. H. Goldschmidt, S. Sternberg, The Hamilton-Cartan formalism in the calculus of variations.
Ann. Inst. Fourier 23, 203267 (1973)
30. M.J. Gotay, A multisymplectic framework for classical field theory and the calculus of
variations, I. Covariant Hamiltonian formalism, in Mechanics, Analysis and Geometry: 200
Years After Lagrange, ed. by M. Francaviglia, D.D. Holm (North Holland, Amsterdam, 1990),
pp. 203235
31. M.J. Gotay, J.M. Nester, G. Hinds, Presymplectic manifolds and the DiracBergmann theory
of constraints. J. Math. Phys. 19, 23882399 (1978)
32. M.J. Gotay, J.A. Isenberg, J.E. Marsden, R. Montgomery, Momentum Mappings and the
Hamiltonian Structure of Classical Field Theories with Constraints (Springer, New York,
1992)
33. K. Grabowska, Lagrangian and Hamiltonian formalism in field theory: a simple model. J.
Geom. Mech. 2, 375395 (2010)
34. P.A. Griffiths, Exterior Differential Systems and the Calculus of Variations. Progress in
Mathematics, vol. 25 (Birkhuser, Boston, 1983)
35. D.R. Grigore, On a generalization of the PoincarCartan form in higher-order field theory, in
Variations, Geometry and Physics (Nova Science Publishers, New York, 2008), pp. 5776
36. A. Hakov, O. Krupkov, Variational first-order partial differential equations. J. Differ. Equ.
191, 6789 (2003)
37. H. Helmholtz, Ueber die physikalische Bedeutung des Prinzips der kleinsten Wirkung. J. fr
die reine u. angewandte Math. 100, 137166 (1887)
38. H.A. Kastrup, Canonical theories of Lagrangian dynamical systems in physics. Phys. Rep. 101,
1167 (1983)
39. J. Kijowski, A finite-dimensional canonical formalism in the classical field theory. Commun.
Math. Phys. 30, 99128 (1973)
40. J. Kijowski, W. Szczyrba, Multisymplectic manifolds and the geometrical construction of
the Poisson brackets in the classical field theory, in Geometrie symplectique et physique
mathematique. Colloques internationaux du Centre national de la recherche scientifique, vol.
237 (C.N.R.S., Paris, 1975), pp. 347378
41. J. Kijowski, W.M. Tulczyjew, A Symplectic Framework for Field Theories. Lecture Notes in
Physics, vol. 107 (Springer, Berlin, 1979)
42. I. Kolr, Some geometric aspects of the higher order variational calculus, in Geometrical
Methods in Physics, ed. by D. Krupka. Proceedings of the Conference on Differential Geometry
and its Applications. Nov Mesto na Morave, Sept. 1983, vol. 2 (J. E. Purkyne University, Brno,
1984), pp. 155166
43. M. Krbek, J. Musilov, Representation of the variational sequence by differential forms. Rep.
Math. Phys. 51(23), 251258 (2003)
44. M. Krbek, J. Musilov, Representation of the variational sequence by differential forms. Acta
Appl. Math. 88(2), 177199 (2005)
45. D. Krupka, Some geometric aspects of variational problems in fibred manifolds. Folia Fac. Sci.
Nat. Univ. Purk. Brunensis, Physica, Brno (Czechoslovakia) 14, 65 pp. (1973). ArXiv:math-
ph/0110005
46. D. Krupka, A map associated to the Lepagean forms of the calculus of variations in fibred
manifolds. Czechoslov. Math. J. 27, 114118 (1977)
47. D. Krupka, On the local structure of the Euler-Lagrange mapping of the calculus of variations,
in Proceedings of the Conference on Differential Geometry and its Applications, Nov Mesto na
Morave (Czechoslovakia), 1980 (Charles University, Prague, 1982), pp. 181188. ArXiv:math-
ph/0203034
48. D. Krupka, Lepagean forms in higher order variational theory, in Modern Developments
in Analytical Mechanics I: Geometrical Dynamics, ed. by S. Benenti, M. Francaviglia, A.
Lichnerowicz. Proceedings of the IUTAM-ISIMM Symposium, Torino, 1982 (Accad. Sci.
Torino, Torino, 1983), pp. 197238
49. D. Krupka, Variational sequences on finite order jet spaces, in Differential Geometry and
Its Applications, ed. by J. Janyka, D. Krupka. Conference Proceeding, Brno, 1989 (World
Scientific, Singapore, 1990), pp. 236254
50. D. Krupka, Variational sequences in mechanics. Calc. Var. 5, 557583 (1997)
51. D. Krupka, Global variational theory in fibred spaces, in Handbook of Global Analysis
(Elsevier Science B. V., Amsterdam, 2008), pp.773836
52. D. Krupka, Introduction to Global Variational Geometry (Atlantis Press, Amsterdam, 2015)
53. D. Krupka, J. edenkov, Variational sequences and Lepage forms, in Differential Geometry
and Its Applications, ed. by J. Bure, O. Kowalski, D. Krupka, J. Slovk. Conference
Proceeding, Prague, August 2004 (Charles University, Prague, 2005), pp. 617627
54. D. Krupka, O. Krupkov, G. Prince, W. Sarlet, Contact symmetries of the Helmholtz form.
Differ. Geom. Appl. 25, 518542 (2007)
55. D. Krupka, O, Krupkov, D. Saunders, The Cartan form and its generalizations in the calculus
of variations. Int. J. Geom. Meth. Mod. Phys. 7, 631654 (2010)
56. O. Krupkov, Lepagean 2-forms in higher order Hamiltonian mechanics, I. Regularity. Arch.
Math. (Brno) 22, 97120 (1986)
57. O. Krupkov, On the inverse problem of the calculus of variations for ordinary differential
equations. Math. Bohem. 118, 261276 (1993)
360 O. Rossi
58. O. Krupkov, A geometric setting for higher-order DiracBergmann theory of constraints. J.

Math. Phys. 35, 65576576 (1994)
59. O. Krupkov, The Geometry of Ordinary Variational Equations. Lecture Notes in Mathematics,
vol. 1678 (Springer, Berlin, 1997)
60. O. Krupkov, Hamiltonian field theory. J. Geom. Phys. 43, 93132 (2002)
61. O. Krupkov, Variational metric structures. Publ. Math. Debr. 62, 461495 (2003)
62. O. Krupkov, Lepage forms in the calculus of variations, in Variations, Geometry and Physics
(Nova Science Publishers, New York, 2008), pp. 2956
63. O. Krupkov, Variational equations on manifolds, in Advances in Mathematics Research, vol.
9 (Nova Science Publishers, New York, 2009), pp. 201274
64. O. Krupkov, R. Malikov, Helmholtz conditions and their generalizations. Balkan J. Geom.
Appl. 15(1), 8089 (2010)
65. O. Krupkov, G.E. Prince, Lepage forms, closed 2-forms and second-order ordinary differential
equations. Russ. Math. (Iz. VUZ) 51(12), 116 (2007)
66. O. Krupkov, D.J. Saunders (eds.), Variations, Geometry and Physics (Nova Science Publish-
ers, New York, 2009), p. 360
67. O. Krupkov, D.J. Saunders, Affine duality, and Lagrangian and Hamiltonian systems. Int. J.
Geom. Methods Mod. Phys. 8, 669697 (2011)
68. O. Krupkov, D. Smetanov, On regularization of variational problems in first-order field
theory, in Proceedings of the 20th Winter School Geometry and Physics (Srn, 2000). Rend.
Circ. Mat. Palermo (2), Suppl. No. 66 (2001), pp. 133140
69. O. Krupkov, D. Smetanov, Legendre transformation for regularizable Lagrangians in field
theory. Lett. Math. Phys. 58, 189204 (2001)
70. O. Krupkov, D. Smetanov, Lepage equivalents of second order EulerLagrange forms and
the inverse problem of the calculus of variations. J. Nonlin. Math. Phys. 16, 235250 (2009)
71. B. Kupershmidt, Geometry of jet bundles and the structure of Lagrangian and Hamiltonian
formalisms, in Lecture Notes in Mathematics, vol. 775 (Springer, Berlin, 1980), pp. 162217
72. T. Lepage, Sur les champs godsiques du calcul des variations. Bull. Acad. Roy. Belg. Cl. des
Sci. 22, 716729 (1936)
73. P. Libermann, C.-M. Marle, in Symplectic Geometry and Analytical Mechanics. Mathematics
and Its Applications (D. Reidel, Dordrecht, 1987)
74. R. Malkov, On a generalization of Helmholtz conditions. Acta Math. Univ. Ostrav. 17, 1121
(2009)
75. L. Mangiarotti, M. Modugno, Fibred spaces, jet spaces and connections for field theories,
in Proceedings of the Meeting Geometry and Physics, 1982, Florence (Pitagora, Bologna,
1982), pp. 135165
76. J.E. Marsden, S. Pekarsky, S. Shkoller, M. West, Variational methods, multisymplectic
geometry and continuum mechanics. J. Geom. Phys. 38, 253284 (2001)
77. P.J. Olver, Equivalence and the Cartan form. Acta Appl. Math. 31, 99136 (1993)
78. M. Palese, O. Rossi, E. Winterroth, J. Musilov, Variational sequences, representation
sequences and applications in physics, 61 pp. ArXiv:1508.01752v1 [math-ph]; SIGMA 12,
045 45 pp. (2016)
79. H. Poincar, Les mthodes nouvelles de la Mcanique cleste, vol. 3 (Gauthier-Villars, Paris,
1899)
80. O. Rossi, Homogeneous differential equations and the inverse problem of the calculus of
variations. Publ. Math. Debr. 84, 165188 (2014)
81. O. Rossi, Geometry of variational partial differential equations and Hamiltonian systems, in
Geometry of Jets and Fields, Banach Center Publications, vol. 110 (Inst. of Math., Polish of
Academy of Sciences, Warszawa, 2016), pp. 219237
82. O. Rossi, D.J. Saunders, Dual jet bundles, Hamiltonian systems and connections. Diff. Geom.
Appl. 35, 178198 (2014)
83. O. Rossi, D. Saunders, Lagrangian and Hamiltonian duality. J. Math. Sci. 218, 813816 (2016)
84. W. Sarlet, Geometrical structures related to second-order equations, in Differential Geometry

and Its Applications, ed. by D. Krupka, A. vec. Conference Proceedings, Brno, 1986 (D.
Reidel, Dordrecht, 1986), pp. 279288
85. D.J. Saunders, Jet fields, connections and second-order differential equations. J. Phys. A Math.
Gen. 20, 32613270 (1987)
86. D.J. Saunders, The Geometry of Jet Bundles. London Mathematical Society Lecture Note
Series, vol. 142 (Cambridge University Press, Cambridge, 1989)
87. D.J. Saunders, The regularity of variational problems. Contemp. Math. 132, 573593 (1992)
88. D.J. Saunders, A new approach to the nonlinear connection associated with second order (and
higher-order) differential equation fields. J. Phys. A Math. Gen. 30, 17391743 (1997)
89. E. Tonti, Variational formulation of nonlinear differential equations I, II. Bull. Acad. Roy. Belg.
Cl. Sci. 55, 137165, 262278 (1969)
90. M.M. Vainberg, Variational Methods in the Theory of Nonlinear Operators (GITL, Moscow,
1959) (in Russian)
91. A.M. Vinogradov, A spectral sequence associated with a non-linear differential equation, and
algebro-geometric foundations of Lagrangian field theory with constraints. Soviet Math. Dokl.
19, 144148 (1978)
92. A. Weinstein, Lectures on Symplectic Manifolds. CBMS Regional Conference Series in
Mathematics, vol. 29 (American Mathematical Society, Providence, 1977), 48 pp.
93. R.O. Wells, Differential Analysis on Complex Manifolds (Springer, Berlin, 1980)
94. E.T. Whittaker, A Treatise on Analytical Dynamics of Particles and Rigid Bodies (The
University Press, Cambridge, 1917)

3319621807

Uploaded by

Copyright:

Available Formats

3319621807

Uploaded by

Document Information

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

3319621807

Uploaded by

Copyright:

Available Formats

UNIPA Springer Series

Lie Groups, Differential

ISSN 2366-7516 ISSN 2366-7524 (electronic)

Library of Congress Control Number: 2017952063

Springer International Publishing AG 2017

Printed on acid-free paper

This Springer imprint is published by Springer Nature

Palermo, Italy, February 11, 2016

universal enveloping algebra of a relatively free Sabinin algebra on two generators,

Palermo, Italy Giovanni Falcone

1 A Short Survey on Lie Theory and Finsler Geometry.. . . . . . . . . . . . . . . . . 1

10 Reduction of Some Semi-discrete Schemes

M. Abashidze Batumi Shota Rustaveli State University, Batumi, Georgia

M.Z. Menteshashvili Muskhelishvili Institute of Computational Mathematics of

Giovanni Falcone, Karl Strambach, and Ming Xu

A Lie algebra over a field k is a k-vector space g, endowed with a non-associative

Springer International Publishing AG 2017 1

is a Lie algebra. In particular, the algebra of endomorphisms of a k-vector space

fulfilling the Lie brackets

x; y 2 I; for all x 2 I and y 2 g:

For instance, sl.n; k/ is an ideal of gl.n; k/.

Every finite-dimensional Lie algebra h over a field k is isomorphic to a Lie

and a representation is said to be faithful if its kernel is trivial, i.e., it is injective.

g.k/ D g.k1/ ; g.k1/ ;

and the descending central series

1.1.1 Representations, Derivations, Killing Form

n W sl.2; k/ ! sl.n; k/ 6 gl.n; k/ (1.3)

defined, for 0 6 k 6 n, by the following action on the vectors fe0 ; : : : ; en1 g of a

n .X0 /ek D .n 2k 1/ek I n .X1 /ek D kek1 I n .X2 /ek D .n k 1/ekC1 :

Lx ; Ly D iLz ; Ly ; Lz D iLx ; Lz ; Lx D iLy : (1.4)

The three-dimensional real Lie algebra generated by Lx , Ly , Lz is isomorphic

adx .y/ D x; y;

adx .y; z/ D adx .y/; z C y; adx .z/:

More generally, a map W h ! h such that y; z D y; z C y; z (and,

.x; y/ D Tr.adx ady /;

1.1.2 The Weyl Algebra of Differential Operators

As mentioned in the opening, the composition of differential operators is a non-

polynomial coefficients in n variables. Its generators xk (acting as f .x/ 7! xk f .x/)

It was introduced by Hermann Weyl in his mathematical formulation of quantum

on the algebra of homogeneous polynomial of degree n 1 in two variables x1 and

0 .ek / D .n 2k 1/ek ; 1 .ek / D kek1 ; 2 .ek / D .n k 1/ekC1 :

1.1.3 Lie Groups, Exponential Map

A Lie group is a group defined over a finite-dimensional smooth manifold, such

with  2 R, which, by the way, are written in complex form as

confirming, once again, the self-consistency of Lie theory.

Three-dimensional Lie groups, .2n C 1/-dimensional Heisenberg groups, and

1.1.4 Compact Lie Algebras

As such, it is the real form of the group of complex matrices

Putting alternatively x3 D x4 D 0, x2 D x4 D 0, and x2 D x3 D 0, we obtain

.x; y/ D Tr.adx ady /

of a semisimple real Lie algebra g determines, by Sylvesters inertia, an index, which

is closed under transposition, the Cartan decomposition is just the decomposition

1.2 Non-associative Structures

1.2.1 Loops and Geometry

the left translations of a proper topological loop, then G is homeomorphic to a

1.2.2 Gerstenhaber Algebras

.x; y; z/ D .x; z; y/:

x; y; z C z; x; y C y; z; x D

.x; y; z/ C .y; z; x/ C .z; x; y/ .y; x; z/ .x; z; y/ .z; y; x/:

homotopy Gerstenhaber algebras found a place in the investigations of Kontsevich

0 .ek / D .n 2k 1/ek ; 1 .ek / D kek1 ; 2 .ek / D .n k 1/ekC1 :

with 2 R, which, by the way, are written in complex form as

where x 2 Ak , y 2 A. A slight generalization of the operator defines Batalin

where Dx denotes the derivation operator with respect to x. Of course, S .R/

and D 12 .Q2 C P2 C 1/ is essentially self-adjoint on S .R/.

.U .g.; ; //f /.x/ WD exp.i.x C //f .x C /; x 2 R:

This implies that T0 0 (otherwise, 0 D 0) and