Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
LIE GROUPS:
NEW RESEARCH
No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or
by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no
expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of information
contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in
rendering legal, medical or any other professional services.
MATHEMATICS RESEARCH DEVELOPMENTS
SERIES
LIE GROUPS:
NEW RESEARCH
ALTOS B. CANTERRA
EDITOR
All rights reserved. No part of this book may be reproduced, stored in a retrieval system or
transmitted in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical
photocopying, recording or otherwise without the written permission of the Publisher.
For permission to use material from this book please contact us:
Telephone 631-231-7269; Fax 631-231-8175
Web Site: http://www.novapublishers.com
Independent verification should be sought for any data, advice or recommendations contained in
this book. In addition, no responsibility is assumed by the publisher for any injury and/or damage
to persons or property arising from any methods, products, instructions, ideas or otherwise
contained in this publication.
This publication is designed to provide accurate and authoritative information with regard to the
subject matter covered herein. It is sold with the clear understanding that the Publisher is not
engaged in rendering legal or any other professional services. If legal or any other expert
assistance is required, the services of a competent person should be sought. FROM A
DECLARATION OF PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE
AMERICAN BAR ASSOCIATION AND A COMMITTEE OF PUBLISHERS.
Canterra, Altos B.
Lie groups : new research / Altos B. Canterra.
p. cm.
Includes index.
ISBN 978-1-61668-164-7 (E-Book)
1. Lie groups. I. Title.
QA387.C35 2009
512'.482--dc22
2009015095
Preface vii
Chapter 1 Lie Group Guide to the Universe 1
Bernd Schmeikal
Chapter 2 Rotation Manifold SO(3) and Its Tangential Vectors 61
Jari Mäkinen
Chapter 3 Asymptotic Homology of the Quotient of PSL2(R) by a Modular 89
Group
Jacques Franchi
Chapter 4 Group Analysis of Solutions of 2-Dimensional Differential 123
Equations
Sergey I. Senashov and Alexander Yakhno
Chapter 5 The Module Structure of the Infinite-Dimensional Lie Algebra 139
Attached to a Vector Field
Guan Keying
Chapter 6 Lie Group Methods for Modulus Conserving Differential 169
Equations
Jian-Qiang Sun, Hua Wei and Gui-Dong Dai
Chapter 7 Singularities and Stability of a Work Function 187
Jean Lerbet
Chapter 8 The Conformal-Affine Structure of open Quantum Relativity, 199
Its Physical Realization and Implications
G. Basini and S. Capozziello
Chapter 9 Twisted Balanced Metrics 267
Julien Keller
This new book is dedicated to recent and important research on Lie groups. A Lie Group
is a group which is also a differentiable manifold, with the property that the group operations
are compatible with the smooth structure. They are named after the nineteenth century
Norwegian mathematician Sophus Lie, who laid the foundations of the theory of continuous
transformation groups.
Lie groups represent the best developed theory of continuous symmetry of mathematical
objects and structures, which makes them indispensable tools for many parts of contemporary
mathematics, as well as for modern theoretical physics. They provide a natural framework for
analysing the continuous symmetries of differential equations (Differential Galois theory), in
much the same way as permutation groups are used in Galois theory for analysing the discrete
symmetries of algebraic equations. An extension of Galois theory to the case of continuous
symmetry groups was one of Lie's principal motivations.
As discussed in Chapter 1, reconstructing physics in Clifford algebra brought to light that
unity of physics that we were after, at least to some considerable extent. In 1990 David
Hestenes published “Clifford Algebra and the Interpretation of Quantum Mechanics” [19].
Therein he represented the spin of an electron by the exterior product −½ γ1∧γ2 or bivector
−½ γ12. He did that over and over again between 1980 and 2000. But he too did not mention
Lipkin. Quite obviously, he had not read »Lie Groups for Pedestrians«. In my 2004
publication [20] on “Transpositions in Clifford Algebra” I pointed out Harry Lipkin had
found that out already in 1965. Lipkin [21] understood the ½γij were angular 4-momenta, and
he even identified the γ-matrices as linear combinations of baryon creation- and annihilation
operators. I have carefully surveyed the Clifford algebra literature. No one went so far as
Harry J. Lipkin with his famous »Lie Groups for Pedestrians«. And later, no one of us
rediscovered the simplicity and beauty of angular momentum algebra in quadratic Clifford
algebras. We may say that many of us discovered Clifford algebra, but only few of us who
were good enough in geometric algebra understood the deep meaning of Lie algebra in
Clifford algebra. It is therefore that I decided to lift Lipkins Lie Groups for Pedestrians up to
Clifford algebra. I will tell us a story. The story has the title »Lie Group Guide To The
Universe«. It is a booklet about Clifford-Lie-Algebra, quantum geometry, and the standard
model of matter. Suppose the preceding section would have had to represent Lipkins
introduction. Then next there should follow a review of angular momentum algebra. To lift
that topic to geometric algebra let us first consider the Pauli algebra which is a representation
viii Altos B. Canterra
of the Clifford algebra Cl3,0 in Mat(2, ℂ). This will not prevent us from constructing some
more general concept of quantum geometry.
In Chapter 2, we prove that incremental material rotation vectors belong to different
tangent spaces of the rotation manifold SO(3) at a different instant. Moreover, we show that
the material tangent space as the tangent space at unity is not a possible definition yielding
geometrically inconsistent results, although this kind of definition is widely adopted in
applied mechanics community. In addition, we show that the standard Newmark integration
scheme for incremental rotations neglects first order terms of rotation vector, not third order
terms. Finally, we show that the rotation interpolation of extracted nodal values on the
rotation manifold is not an objective interpolation under the observer transformation. This
clarifies controversy about the frame-indifference of geometrically exact beam formulations
in their finite element implementations.
Consider G := PSL2( ) ≡ T1Η2, a modular group Γ , and the homogeneous space
Γ \G ≡ T1( Γ \Η2 ). Endow G, and then Γ \G, with a canonical left-invariant metric, thereby
equipping it with a quasi hyperbolic geometry. Windings around handles and cusps of Γ \G
are calculated by integrals of closed 1-forms of Γ \G. The main results express, in both
Brownian and geodesic cases, the joint convergence of the law of these integrals, with a stress
on the asymptotic independence between slow and fast windings. The non-hyperbolicity of
Γ \G is responsible for a difference between the Brownian and geodesic asymptotic
behaviours, difference which does not exist at the level of the Riemann surface Γ \Η2 (and
generally in hyperbolic cases). Identification of the cohomology classes of closed 1-forms
with harmonic 1-forms, and equidistribution of large geodesic spheres, are also addressed in
Chapter 3.
It is well known [4, 9] that if a system of differential equations admits the Lie group of
point transformations (point symmetry), then any solution of the system is transformed to a
solution of this system. This property permits the construction of new solutions without
integrating the given system of partial differential equations (PDEs), by means of group
transformations alone under known solutions. This is an effective method if we have a
sufficiently rich group of point transformations.
By applying point transformations to exact solution, a family of so-called Ssolutions can
be constructed, i.e., obtained by means of symmetries. This family of S-solutions is dependent
on the group parameter. If this parameter is equal to zero, then we have an initial solution.
This procedure is called the production [9] or reproduction of solutions [4].
Moreover, it is easy to show that under a group transformation characteristic curves of
the system of PDEs of the hyperbolic type are transformed to the characteristics curves. The
evolution of characteristic curves permits to find out the boundary conditions for new S-
solutions.
In Chapter 4 authors will show some applications of this procedure for the system of the
theory of ideal plane plasticity, developing results obtained in [12]. In particular, we shall use
an infinite subgroup of the group of symmetries for deformation of characteristics curves of
the considered hyperbolic system of PDEs to construct a new analytical solutions. From the
system of PDEs an automorphic system will be deduced, which permits find out some
relations between different solutions by means of group transformations.
In Chapter 5, based on a generalized definition on the admittance of a Lie group by a
vector field, it is proved that, attached to any given smooth vector field X on a n-dimensional
Preface ix
Secondly, we analyse the conditions so that f is a submersion with fold. Using the fact that f is
inf-stable if and only if g = f|Σ1 ( f ) is an immersion with normal crossings, we analyse this
property and we highlight some consequences. Applications to robotics are suggested.
Beside the post-relativistic theories, Open Quantum Relativity is a gauge theory of
interactions based on a nonlinear realization (NLR) of the local Conformal-Affine (CA) group
of symmetry transformations. Such a theory, thanks to a covariantsymplectic formulation,
succeeds in treating General Relativity and Quantum Mechanics under the same standard. In
Chapter 8, we obtain the coframe fields and the gauge connections of the theory while the
tetrads and Lorentz group metric are used to induce the spacetime metric. The
inhomogenously transforming (under the Lorentz group) connection coefficients serve as
x Altos B. Canterra
Xˆ
( O, d cc g ) is the local homogeneous approximation of ( O, d cc ) with respect to the action of
X
the homogeneous operator of dilatation which agrees with our graduation in some
Xˆ
neighborhood of g (( O, d cc g ) is some analogue of so-called nilpotent tangen cone). For the
Xˆ
quasispaces ( O, d cc g ), ( O, d cc ) we develop some technique, which help us to get the local
X
approximation theorem for quasimetric. As the consequence, we get some results for
quasispaces induced by the collection of C1 basis canonical non-commutative vector fields, by
the collection of C2 basis non-commutative vector fields.
The purpose of Chapter 12 is to review the Adler Kostant Symes scheme as a theory
which can be developed successfully in different contexts. It was useful to describe some
mechanical systems, the so called generalized Toda, and now it was proved to be a tool for
the study of the linear approach to the motion of n uncoupled harmonic oscillators. The
complete integrability of these systems has an algebraic description. In the original theory this
is related to ad-invariant functions, but new examples show that new conditions should be
investigated.
A Lévy process in a Lie group is a process that possesses independent and stationary
multiplicative increments. The theory of such processes is not merely an extension of Lévy
processes in Euclidean spaces. Because of the unique structures possessed by non-
commutative Lie groups, these processes exhibit certain interesting properties which are not
present for their counterparts in Euclidean spaces.
The concept of Lévy processes may be extended to include Markov processes in a
homogeneous space that are invariant under the group action. More generally, we will also
study processes in Lie groups and homogeneous spaces that possess independent, but not
necessarily stationary, increments, called nonhomogeneous Lévy processes. These processes
appear naturally when studying a decomposition of a general Markov process in a manifold
invariant under a group action.
In Chapter 13, we will provide an introduction to Lévy processes in Lie groups and
homogeneous spaces, and present some selected results in this area. The reader is referred to
xii Altos B. Canterra
the literature for the most of proofs, but some explanation will be given to the results not yet
published.
The symmetry classification of differential equations containing arbitrary functions can
be a source of several interesting results. In Chapter 14 we study two particular but significant
examples: a nonlinear ODE and a linear PDE (the 1-dimensional Schr¨odinger equation). We
provide first of all a necessary, but very restrictive, simple condition involving the arbitrary
functions in order that the given equation does admit Lie symmetries. In the first example, we
show that some symmetry appears only if a precise numerical relation between the involved
parameters is satisfied. In the case of Schrödinger equation, we see that only for a very
limited class of potential functions some symmetry is admitted, and that the Lie generators of
these symmetries are precisely recursion operators and are related to the Dirac step up - step
down operators, well known in Quantum Mechanics. In connection with all these symmetries,
we also discuss the important problem of the reduction of the differential equations, in both
the different contexts of ODE’s and of PDE’s.
As presented in Chapter 15, contractions and deformations of Lie algebras and their
relations have played an important role in many fields since their introduction in the
1950’s, and many progress has been done in understanding their structural and
geometrical properties. Although being a rather active research field, there remain
various important problems concerning contractions and deformations that have still not
be satisfactorily solved. The notion of contraction appeared first in physical context by
Segal [1], and was soon recognized to have important consequences, like the possibility
of switching off interactions, or analyzing the precise effect of some physical quantities
when others are disregarded. The formal introduction of contractions, done by Inönü and
Wigner [2], was soon defined more generally by Saletan and Kupczyński [3], in order to
cover other limiting processes observed in symmetry groups used in Physics, like the
transition from relativistic to non-relativistic physics. Other, more or less specifical, types
of contractions have been introduced in the literature since, and their structural properties
analyzed [4–7,9,10]. In addition, the contractions among Lie algebras of fixed dimension
have been studied in detail [11–16], as well as important classes of algebras, like those of
kinematical groups [17,18]. However, the lack of complete classifications for Lie
algebras from dimension six onwards is an important obstruction that motivated different
approaches to the problem. The relation of deformation theory, a formalism born in
Differential Geometry, with contractions of Lie algebras, was first observed in [5], and
has offered a kind of “inverse” procedure to study contractions. This point of view also
suggested a geometrical interpretation of contractions in terms of orbits in a manifold, the
points of which correspond to Lie algebras [19]. An advantage of this approach is a
definition of contractions that includes all special types used in the literature, and that
allow to establish different sufficiency criteria for the existence of contractions.
Moreover, this motivated the application of specific techniques like cohomology of Lie
algebras, which have proven to be an essential tool in many problems [20–23].
In Chapter 16, the Ehresmann’s theorem about a Lie structure in the hole automorphism
group of a finite type G-structure on manifold is generalized to orbifolds.
Preface xiii
Estimates for dimension of such Lie group are established, depending on stratifications of
orbifolds. Particular attention is devoted to affine connected, pseudo- Riemannian and
Riemannian orbifolds. The content is illustrated by examples.
Chapter 17 is devoted to the investigation of wrap groups of connected fiber bundles over
the fields of real R, complex C numbers, the quaternion skew field H and the octonion
algebra O. These groups are constructed with mild conditions on fibers. Their examples are
given.
It is shown, that these groups exist and for differentiable fibers have the infinite
dimensional Lie groups structure, that is, they are continuous or differentiable manifolds and
−1
the composition ( f , g ) a f g is continuous or differentiable depending on a class of
smoothness of groups. Moreover, it is demonstrated that in the cases of real, complex,
quaternion and octonion manifolds these groups have structures of real, complex, quaternion
or octonion manifolds respectively. Nevertheless, it is proved that these groups does not
necessarily satisfy the Campbell-Hausdorff formula even locally. Iterated wrap groups are
studied as well.
Their smashed products are constructed. Cohomologies of wrap groups and their
structure are investigated. Sheaves of wrap groups are constructed and studied. Moreover,
twisted cohomologies and sheaves over quaternions and octonions are investigated as well.
CW-groups associated with wrap groups are studied.
Chapter 18 is devoted to the investigation of groups of diffeomorphisms and wraps of
manifolds over non-archimedean fields of zero and positive characteristics. Different types of
topologies are considered on groups of wraps and diffeomorphisms relative to which they are
generalized Lie groups or topological groups. Among such topologies pairwise incomparable
are found as well. Topological perfectness of the diffeomorphism group relative to certain
topologies is studied. There are proved theorems about projective limit decompositions of
these groups and their compactifications for compact manifolds. Moreover, an existence of
one-parameter local subgroups of diffeomorphism groups is investigated.
In: Lie Groups: New Research ISBN: 978-1-60692-389-4
Editor: Altos B. Canterra, pp. 1-59 © 2009 Nova Science Publishers, Inc.
Chapter 1
Bernd Schmeikal
Am Platzl, Garsten, Austria
Introduction
When we were freshmen in physics some of us worked as scanners in the High Energy group
led by Walter Thirring. We set in a small room called Schrödingerzimmer in the Institute for
Theoretical Physics, Boltzmanngasse 5, Vienna. We did not yet have the social competence to
realize what that meant. We calculated Clebsch-Gordon coefficients which to me appeared as
some browbeating luxury that kept me away from understanding physics. We were searching
after K-vertices and -vortices, the Omega minus in the morning 6 to 9 AM and late during
night shift between 6 PM and shy of midnight. We were young and swamped by some
incoming papers on the unitary symmetry SU(3) of strong interaction. Some written by
George Zweig [1] from CERN, others authored by Murry Gell-Mann [2] from Pasadena,
some by military attaché Yuval Ne’eman [3], then Imperial College, still others by Wolfgang
Kummer, one of our teachers, later Erwin Schrödinger Laureate1 [4]. Our head Walter
Thirring cared for us, like his father Hans had done so for Albert Einstein. He supported us in
every concern. Then, there was a catholic seminary just around the corner. We didn’t know,
Thirring would some day convert that seminary into a Schrödinger Institute. That meant some
more discipline, indeed.
There were sociologists around me some miles away in the Institute for Advanced
Studies, Vienna, peace loving people like Robert Jungk, Oskar Morgenstern, Robert
Reichardt, Anatol Rapoport, James S. Coleman. They taught me sociology and mathematics. I
found out there were many more people like us calculating Clebsch-Gordon coefficients all
over the world and in many different kinds of situations. One of my later friends, Zbigniew
Oziewicz, from the College of Mathematics and Physics at the University of Wroclaw in
Poland, did so under the most pressing conditions of political imprisonment. There was
something important one could learn from sociology, namely that the whole unitary
knowledge of mankind can only be found in all the heads. One had to master boarder
1
Sure you will soon understand why, in this writing, I prefer to mention Wolfgang Kummer in connection with the
so(2, 1) invariant gravitational fields to be found in Vienna, Preprint ESI 1110 (2001)
2 Bernd Schmeikal
disputes. There had emerged many different divisions of theory. They could not be brought
together without trans-national access to research. Their unification needed some radical mind
opening effort: radical constructivism, creativity and cooperation. I would not have known
that those efforts would lead me back to Lie groups. I came back to the work of Sophus Lie
[5] and the beautiful theory that follows due to Elie Cartan [6] and Hermann Weyl [7]. But
this time I came back on the unusual road built by the efforts of many colleagues who
completed geometry [8] as was begun with by William Kingdon Clifford in London in about
1870. He discovered what nowadays is called Clifford algebra, only a few years before he
died in Madeira 1879. That discovery – which is better called a construction – underwent a
lexical fixation in the year 1878.
But then, in 1965, we felt for the first time the Dirac equation had a deep geometric
meaning. Only thirty years later those of us advanced enough in geometric algebra understood
why. In 2005 in ‘Letter 02’2 to the author Professor Osziewicz [9] could be sure enough to
write “see, the Dirac equation is nothing but Clifford algebra, rediscovered by Dirac in 1928.
However Dirac himself in so many re-editions of his Quantum Mechanics book never
mentioned Clifford.” Whether Dirac wanted to mystify the insights of Clifford, or if he just
did not see the connection remains an open question of phenomenology. Dirac published
quantum theory of the electron in 1928 [10] where he first presented his unusual equation of
motion. Dirac found this equation by playing around with the relativistic Klein-Gordon-
equation. He needed a differential operator of first order which squared would recover the
d’Alembert operator. This could be achieved only by using spinors and gamma-matrices. By
that equation the anomalous Zeeman-effect and fine structure of atomic absorption spectra
could be explained. Note all that was based on the assumption spin was a relativistic
phenomenon. Today we know that it is not.3 But then Dirac could conclude with the
prognosis: there must exist a positron, an anti-electron with positive charge. When Anderson
in 1932 established the identity of the positron in cosmic radiation he supported, at the same
time, the belief system that began to form around spin and relativity and the myth of the
Dirac-sea with a huge negative energy.
The Principles of Quantum Mechanics was released first in 1930 [11]. In the same year
(1930/2) Juvet published Opérateurs de Dirac et équations de Maxwell in the second book of
Commentarii Mathematici Helvetici [13]. Today we would say he based the rigor on a matrix
representation of the Dirac- and d’Alembert operators in the Clifford algebra Cl4,0. He
explained why the Γ-matrices he introduced in the stile of Dirac were hypercomplex numbers
also already known as Clifford numbers (p. 227). He realized the wave equations and their
spinorial solutions lived in a 16-dimensional space of Clifford numbers which he for that
special purpose preferred to denote as Lorentz numbers. Juvet even wrote down the general
element of the Clifford algebra Cl4,0 as is presented today by the Clifford software of MAPLE
[14]. For the first time Juvet and Sauter [15] made use of algebraic spinors, that is, they
replaced column spinors by matrix spinors where only the first column was non-zero. After
2
‘Letter 02’ had four contents dwelt by their author: 1. Forbidden to talk: Forwarded message, 2. Your letter arrived
in Mexico on 16th of July 2005, 3. Hestenes popularized Clifford, 4. Relativistic addition of relative velocities.
3
In the Clifford algebra Cl3,0 generated by the Euclidean 3-space the construction of a universal covering group of
the rotation group SO(3) goes back to Lipschitz and is denoted as Spin(3). In the matrix formulation provided
by the Pauli spin matrices, the spin group has an isomorphic image which is the special unitary group SU(2).
This is a two-fold covering group of the rotation group or double-cover. This statement can be generalized to
higher dimensions and is the mathematical cause for the appearance of spin. It is not bound to relativity.
(Lounesto 2003, p. 59, 220) [12]
Lie Group Guide to the Universe 3
all it makes me contemplative why Dirac in so many editions of his Principles of Quantum
Mechanics not even mentioned Clifford numbers. Did he want sole reign? We have no strong
enough empirical indicators for a definite vote. But as a matter of fact the resulting time-lag
determined upon progress and unity of science. It took us a long time, until the cognitive gaps
could be filled, and our work is still patchwork.
In 1965 the young could hardly forebode a gamma-matrix would someday represent a
base unit in Clifford algebra. In the 60s, we could see by the eyes, Lie group multiplets of
SU(3) were pure geometry. Still, we did not dare to correlate that geometry with outer
spacetime symmetries. We would not even have guessed such a correlation could turn out
one-to-one. We just did not know where that geometry came from. But there were many more
such unbridgeable proximities. To mention the next one: the state ψ of a physical system and
some observable, say angular momentum L or Hamiltonian H, had to live in different spaces,
ψ lived in a function space and L in some matrix algebra. Beginning in the 1980s Hiley from
Birkbeck College [16], advised by Bohm, Frescura and some others pressed ahead with the
algebraization and by 2000 ended up with the Gelfand-Naimark-Segal construction (GNS)
[17] in what they denoted as ‘generalised Clifford algebra’ or ‘discrete Weyl algebra’, some
useful types of *-algebras. In those pilot projects and test results there was made use of
construction plans which later turned out superfluous since the existence of (anti)involutions
and closedness – the C in C* - were natural properties of Clifford algebra. Clearly, all the
advantages gained from the instruments of Clifford algebra would automatically transpose
onto Lie algebra, once we decided to construct Lie algebra and groups and manifolds within
the geometric algebra. In the beginning the efforts made to arrive at some algebraic quantum
mechanics were rather arduous. Using some types of Clifford algebras several types of
mistakes were made. Sometimes nilpotents were displayed as idempotents. Hiley suddenly
said that for any element A of a *-algebra A with A*A = 0 allowing for the GNS-construction
there should have followed A = 0 which could not make sense as that would have resulted in
the diminishing of the Gel’fand ideal and forbid orthogonality of pure states. Because,
consider the Clifford algebra of the Minkowski spacetime Cl3,1. This is a C*-algebra where
two pure states f1 and f4 can be represented by the primitive idempotents f1= ½(1+e1)½(1+e24)
and f4= ½(1−e1)½(1−e24), with unit bivector e24. It is quickly verified that f4 is the main
involuted of f1 and at the same time those two are orthogonal primitive idempotents, that is,
the Clifford product f1 f4 = 0 vanishes, but neither do we have f1= 0 nor is f1* = 0. As we found
out, those primitive idempotents represent fermion quark states [18].
Indeed, reconstructing physics in Clifford algebra brought to light that unity of physics
that we were after, at least to some considerable extent. In 1990 David Hestenes published
“Clifford Algebra and the Interpretation of Quantum Mechanics” [19]. Therein he
represented the spin of an electron by the exterior product −½ γ1∧γ2 or bivector −½ γ12. He did
that over and over again between 1980 and 2000. But he too did not mention Lipkin. Quite
obviously, he had not read »Lie Groups for Pedestrians«. In my 2004 publication [20] on
“Transpositions in Clifford Algebra” I pointed out Harry Lipkin had found that out already in
1965. Lipkin [21] understood the ½γij were angular 4-momenta, and he even identified the γ-
matrices as linear combinations of baryon creation- and annihilation operators. I have
carefully surveyed the Clifford algebra literature. No one went so far as Harry J. Lipkin with
his famous »Lie Groups for Pedestrians«. And later, no one of us rediscovered the simplicity
and beauty of angular momentum algebra in quadratic Clifford algebras. We may say that
4 Bernd Schmeikal
many of us discovered Clifford algebra, but only few of us who were good enough in
geometric algebra understood the deep meaning of Lie algebra in Clifford algebra. It is
therefore that I decided to lift Lipkins Lie Groups for Pedestrians up to Clifford algebra. I
will tell us a story. The story has the title »Lie Group Guide To The Universe«. It is a booklet
about Clifford-Lie-Algebra, quantum geometry, and the standard model of matter. Suppose
the preceding section would have had to represent Lipkins introduction. Then next there
should follow a review of angular momentum algebra. To lift that topic to geometric algebra
let us first consider the Pauli algebra which is a representation of the Clifford algebra Cl3,0 in
Mat(2, ℂ). This will not prevent us from constructing some more general concept of quantum
geometry.
We use to establish a 1-1 correspondence with the Pauli algebra of unitary 2 x 2 matrices
with complex entries.
Cl 3, 0 Mat(2, ℂ)
Id I2 scalar
e1 e2 e3 σ1 σ2 σ3 vector
e12 e13 e23 σ1 σ 2 σ1σ 3 σ 2σ3 bivector
e123 σ1 σ 2 σ 3 director
As you see, any exterior product ej ∧ ek can indeed be represented by a matrix product of
Pauli spin matrices σj σk. The unary e1 ∧ e2 ∧ e3 represents something like an imaginary unit.
It is a unit matrix with non-vanishing diagonal entries i, the pseudo-scalar. What about history
of spin and quantized rotation?
In 1924 Wolfgang Pauli suggested to introduce a new dichotomous degree of freedom for
the electron. You remember how Ralph Kronig and Alfred Landés led that back to a
quantized rotation of the electron, an idea which Pauli did not like. Because of Pauli’s critics
Kronigs suggestion remained unpublished. To explain the fine structure of spectra and the
anomalous Zeeman effect George Uhlenbeck and Samuel Goudsmit postulated the existence
of spin in 1925. In 1927 Pauli constructed the quantum theory of the electron spin using the
SU(2) matrices and 2-component spinors. Since then we have been using σ 3 as a symbol for
either the spin or the angular momentum of the spinning electron. However, the situation
Lie Group Guide to the Universe 5
became somewhat unclear as soon as we were able to lift the theory into geometric algebra.
Rather generally and for quite a while, it seemed all rotation was generated by bivectors such
as e12. On the other hand, the well tried and proven old candidate e3 = σ3 repeatedly re-
entered the rigor. To give you a few examples, in [19a] Hestenes found out (p. 156), in the
gauge group of the Dirac current, “iσ3 = i γ3γ0 = γ2γ1 is the generator of rotations in a
spacelike plane related to physical currents.” In many writings he said that the quantity S =
(½)ħe2e1 relates the bivector in the Dirac equation to the electron spin. In “Spin and
uncertainty in the interpretation of quantum mechanics”, Hestenes found “that the average
‘internal angular momentum’ has the constant value ħ σ3”. A similar ambivalence can be
found in the writings of William Baylis [22]. But in the end, he seems to give a definite vote
on ħ σ3 at least where the Stern-Gerlach experiment is concerned. He confirmes “evidently sz
= ħe3/2 is the spin operator for ψ ” (Baylis 2004, p. 389f.) Surprisingly, hardly any author has
carried out the quantization of angular momentum and calculated spectra. Therefore my
crafty question: which one should we take, e12 or e3 ? It is exactly that question with which we
begin to lift Lie groups for pedestrians up to a Clifford level of geometry. By the way, the
answer is: both!
J+ = 1
2
(e1 − e13 ) and J − = 1
2
(e1 + e13 ) . (3)
[J 3 , J + ] = J + and [J 3 , J − ] = − J − (4)
From there we can go on. Understanding physical motion as graded motion, the story
unfolds until to the standard model. Depending on the identity Id of the Clifford algebra Cl3,0,
we obtain for the sum of squared components
J 2 = J 12 + J 22 + J 32 = 34 Id (5)
Have we reached our arrival point? Are the J+, J− the ultimate solution? Are they
definite? It is interesting that J± mix grades 1 and 2. Are there, may be, even more general
shift operators associated with J3 ? Is it true that step operators J± shift eigenfunctions J, M〉
of J² and J3 to J ± J , M = J ( J + 1) − M ( M ± 1) J , M ± 1 , from J, M〉 toJ, M ± 1〉?
With the J± , are we climbing up and down a ladder?
6 Bernd Schmeikal
1
2 J + J − = 12 ( Id + e3 ) = f 3 with f 3 f 3 = f 3 (6)
which brings forth a well known minimal left ideal: the algebraic Pauli spinor space
corresponding with the angular momentum J3.
def ψ 0
S = Cl 3,0 f 3 = 1 with ψ1 ,ψ 2 ∈ ℂ (7)
ψ 2 0
We shall call ½ J+ J− an eigenform of the angular momentum algebra associated with the
spinor space S. Consider the general multivector element of the Clifford algebra Cl3,0 with the
eight coordinates xj:
We are searching for general elements J±(X) of angular algebra satisfying commutation
relations (4). Separate solutions can be calculated for the { J+(X) } and { J−(Y) } independent
of each other. We obtain
Using a Clifford algebra calculator, you can easily confirm these elements indeed satisfy
commutation relations (4). Their normalized product is equal to
Please, recall what Kauffman [23] said about eigenforms, and verify that the quantity f =
½ J+(X) J−(Y ), a Clifford product of J+(X) and J−(Y), is an eigenform in the Clifford algebra
Cl3,0. Namely we first define recursively by the aid of the Clifford product the bilinear term
Clearly that form is recursively preserved, that is, we obtain after the next step
Therefore, the fn moves stepwise on a plane within the same real vector space, spanned
by {Id, e3, e12, e123}.
Observe that E(f) is itself an eigenform within Cl3,0. It reproduces itself by Clifford
multiplication. It is a closed 4-dimensional real associative subalgebra of Cl3,0. We have
E(f)E(f) = E(f) which signifies the closedness and it adopts the * (either reversion or
conjugation) from the Clifford algebra. The space of spinor-eigenforms E(f) is therefore a C*-
algebra too. Those four elements represent the essential components of an angular momentum
algebra associated with J3. If you remember, the coefficient a was equal to x2 y2 + x3 y3 and
the b was x2 y3 − x3 y2. So the fn performs a nonlinear motion in the 4-dimensional subspace
generated algebraically by the core spin space
f = ( x 2 + x3 ) Id + ( x 2 + x 3 ) e3
2 2 2 2
(15)
We know that at a given basis {e1, e2, e3} the Cl3,0 contains six primitive idempotents ½
(Id ± ej). One of those is ½ (Id + e3). Clearly, the f = f3 represents the fixed point to form (12).
That fixed point occurs on the Thales’ circle where
x 22 + x32 = 1
2 (16)
Both * and † are anti-automorphisms (not so the grade- or main involution) and therefore
those two provide us with a *-operation. If we take reversion, we get J−(Y) = J+(X)† as a
payoff and at the fixed point we have now f = ½ J+J+†. The algebraic spinor space turns over
into
S = Cl 3,0 12 J + J +* provided that * means † (17)
x 2 y 2 + x3 y 3 = 1
4
(18)
x 2 y 3 − x3 y 2 = ± 14 i
We differ between the space of eigenforms (12), the generating core spin space as is
given by (13) and the algebraic spinor space (7, 17). The spinor space is a minimal left ideal
of the algebra while the space of eigenforms is a linear subspace E(f) ⊂ Cl3,0. This houses a
manifold of primitive idempotents either real as given by (15, 16) or complex according to
(18) if the span in (12) is taken over the complex number field. Is there any relation to Hilbert
space?
We can carry out an interesting GNS construction with the E(f). Let us first simply proceed
with the rigor and go into the theory of GNS later. Consider a linear functional φ over the
Clifford algebra Cl3,0
A†A = (a1 + a 2 + a 3 + a 4 + a 5 + a 6 + a 7 + a8 ) Id +
2 2 2 2 2 2 2 2
+ 2( a1 a 2 − a3 a5 − a 4 a 6 + a 7 a8 )e1 +
Lie Group Guide to the Universe 9
+ 2( a1 a3 + a 2 a5 − a 4 a 7 − a 6 a8 )e2 +
+ 2( a1 a 4 + a 2 a 6 + a3 a 7 + a 5 a8 )e3 (21)
In order to debar φ(A *A) < 0 we define the functional φ(X) for the general element X ∈
Cl3,1 as
φ: φ( X ) = x1 + x 4 real (22)
For the element A∈ Cl3,0 this imposes conditions on the coordinates aj. But we are
interested in the Lie manifold of Lie algebras L = {J3, J+(X), J−(X)} and in the space E(f) of
spinor eigenforms. That is we restrict the equation (21) to representations of L by J3 = ½ e3,
and J+(X), J−(X) :
and provide the same norm. What have we learned from this? We have learned that the GNS-
construction by functional (22) endows the Lie algebra L = {J3, J+(X), J−(X)} and its group
SU(2, X) with a pre Hilbert space having quadratic norm (24). This can be completed to the
Hilbert space of square integrable functions Hφ = L2(X). Classical theory started off with the
concept of Hilbert space. But today the existence of such spaces can be derived by the aid of
the GNS construction. Even an explicit relegation to Hφ is no longer necessary, since all the
mathematics is done within the Clifford algebras.
10 Bernd Schmeikal
For completeness we represent the Clifford algebra Cl3,0 and the Lie algebra L = {J3,
J+(X), J−(X)} by the Pauli algebra as was proposed in chapter 2. We have
0 1 0 i 1 0
e1 = e2 = e3 = basis (29)
1 0 − i 0 0 − 1
1 0 0 2 0 0
J 3 = 12 J + = J − = angular (30)
0 − 1 0 0 2 0
1 0
f 3 = and the space of spinor eigenforms spanned by
0 0
1 0 i 0 i 0
Id = , e3 , e12 = e123 = . (31)
0 1 0 − i 0 i
Quantum Spacetime
Important purpose of the Lie group guide to the universe is a deeper understanding of the
non-commutative geometry of the fundamental forces of nature.4 Quantum spacetime is a
special object of quantum geometry. It unifies classical geometric methods with non-
commutative C*-algebras and their associated functional analysis. By establishing quantum
geometric methods we diminish some classic differences such as the one between operator
and quantum state, observer field and observed field. We start from the assumption that inner
symmetries of matter and outer symmetries of spacetime are essentially the same. But we
give up the limitation to points and trajectories made of points. We shall rather show how the
confinement of strange fermions to point locations – as delta-functions – can be derived
‘wlog’ from the field properties. Proceeding in this way we arrive at new concepts such as of
space and quantum groups – Lie groups of spacetime-matter – by enriching mathematics from
the side of quantum theory. To derive exact statements about phenomena and structure of
space and time from our experience with force fields is one of the main concerns of quantum
geometry. Briefly put, we try to derive space from the fields rather than reverse. Micho
Ðurñevich has put a terrific paper [24] into the net, a “Brief Introduction to Quantum
Geometry”. This provides a good basis for understanding the mathematics of this undertaking,
namely to get space from the fields.
In traditional differential geometry a compact topological space X can be reconstructed
entirely in terms of the associated *-algebra A of continuous complex-valued functions on
X. Every point x ∈ X gives rise to a linear functional χ which is a character of A. In this
4
In 2007 I gave out a limited edition of ten booklets with title »Spacetime Matter« and subtitle » non-commutative
geometry of the fundamental forces of nature « wherein I compiled four selected works appeared in Applied
Clifford Algebras from 2005 to 2007. Therein I introduced the concept of the surabale and force categories in
the Clifford algebra of Minkowski spacetime in the Lorentz metric.
Lie Group Guide to the Universe 11
way we obtain bijections between points and characters. In our case the A shall be given by
a Clifford algebra. In such Clifford quantum geometry a single location may involve
vectors, areas and spacetime volumes. This has consequences for both compactness and
commutation. There exist partitions of Clifford quantum geometry into commuting and
non-commuting subspaces. Those are by no means artificial, but follow from the observed
properties of the fields.
Classically, following the theorem by Gelfand and Naimark, the category of compact
topological spaces repeats in the category of commutative C*-algebras. Continuous maps
between spaces X → Y are transposed onto unary *-homomorphisms between their algebras
B and A. Symmetries of spaces being homomorphisms or monomorphisms of X become
automorphism of A. Now, suppose the standard model Lie algebra of the forces of nature
turns out to be nothing else than the *-algebra associated with spacetime, the main
properties of space and time should be derived from that algebra. That is, the symmetries of
strong and electroweak interactions as described by the standard model Lie groups should
be led back to the automorphisms of geometric Clifford algebra. Classically, by the Riesz
representation theorem we can construct a correspondence between probability measures on
X and the positive linear functionals. A similar statement should hold true in quantum
spaces.
We begin with some associative unary (or unital) linear *-algebra A representing space.
Probability measures on that space are given by positive normalized linear functionals
Such functionals are called states. Normalization is achieved by sending the unit Id ∈ A
to φ(Id) = 1. The set of all linear functionals over A build up the dual space A*. The positive
linear functionals form the positive cone of the dual space A*+. A functional φ ∈ A* is called
extremal if it cannot be decomposed into a linear combination of two others. Those extremal
functionals are called pure states. In the Clifford algebras we are using there exist full sets of
mutually annihilating primitive idempotents fi , that is,
∑f
i =1
i =1 complete set of orthogonal idempotents (34)
where in those algebras, determination of the N usually follows the calculation of the Radon-
Hurwitz number as outlined by Lounesto in [12, p. 226]. Such procedure has important
consequences. Namely, in general, states can be decomposed according to
12 Bernd Schmeikal
N N
φ( A) = ∑ λ i φ i ( A) with ∑ λi = 1 (35)
i =1 i =1
N N N N
φ( Id ) = ∑ λ i φ i ( Id ) = ∑ λ i φ i ∑ f j = ∑ λ i φ i ( f j ) (36)
i =1 i =1 j =1 i , j =1
Thus we have established a 1-1 correspondence between extremal states and primitive
idempotents in A.
Every unit vector ψ ∈ H brings forth a state φ according to (32) via the scalar bracket
φ( A) = ψ D ( A)ψ (38)
Even for cyclic vectors ψ the representation D is fully determined by the associate state
φ. Thus we obtain a bijection between states in A and equivalence classes of triples (H, D, ψ).
It is due to this so called GNS construction that the irreducible representations of a Clifford
algebra A are those that are associated with the orthogonal primitive idempotents.
Considering wisely the back-step into commutative subspaces, we have to have for the
algebra: A = C(X). The GNS-representation D of a state φ is now operating in a Hilbert space
H = L²(X, µ φ). The cyclic vector is represented by the unit function. The operators D are given
by left multiplication. The irreducible representations are 1-dimensional. The associated
characters of A are again the points of X . Pure states are given by characters. Probability
measures are Dirac δ-functions located at the ‘points’ of X. Most surprisingly, those δ-
functions will be accompanied, - in the whole non-commutative Clifford algebra which
accommodates the commutative subspace, - by idempotents of fermion pure states having
baryon number ⅓.
The Functor Cl gL
Knowing that the Clifford algebra is itself a vector space it would be the most simple to
denote any representation of Lie algebra in some Clifford algebra Cl by symbols such as
γλ(Cl), or γλn(Cl) indicating the dimension n of the ground space. But this would not explain
the construction. Therefore, in this section we have to go slowly and show how a Lie group in
Clifford algebra is a composition of two functors, namely Cl and gL. We shall find out that Cl
is an injective functor from the category of quadratic spaces Quad into the category of
associative unary algebras AlgF. For F a field, let LieF be the category of small Lie algebras
over F. Then gL is a functor which, by the Lie bracket, assigns to each associative algebra A
∈ AlgF the Lie algebra γλ(A) on the same vector space, but not necessarily associative. Thus
we obtain the following composition of functors
Cl gL
Quad → AlgF →
LieF (39)
Historically, algebras were defined in terms of generators and relations. Grassmann used
unit vectors {e1, e2, …, en} to define his “Hauptgebiet“ (primary domain), the exterior algebra
∧V (also V∧) by the exterior or ‘wedge’ product. The V∧ has basis
14 Bernd Schmeikal
(40)
def
∧
∧V ≡ V = ⊕ in=0 ∧i V … Grassmanns exterior algebra (41)
There are various ways to transform some Grassmann exterior algebra into a Clifford
algebra: by stepwise definition, by factorization, by deformation and Cliffordization. We do it
by the aid of the Clifford map [30], [12]. Consider a bilinear form B: V × V → F and the
left contraction ┘. A Clifford map γx: ∧V→∧V is an endomorphism parameterized by a 1-
vector x ∈ V in the form
(43)
where û is the main involuted û = (-1)∂(u) u and ∂(u) grade of u. We decompose B into a
symmetric part g and an antisymmetric a. Then the Clifford maps of the generators {ei} of V
generate the Clifford algebra Cl(V, B). With Id the identity morphism, we obtain in a basis
free notation
γ x γ y + γ y γ x = 2 g ( x, y ) Id (42)
γ x γ y − γ y γ x = 2 x ∧ y + 2 a ( x, y ) Id (43)
A very brief definition has been given by Lounesto (p. 190). It “is suitable for non-
degenerate quadratic forms, especially the real quadratic spaces ℝp,q ”. An associative algebra
Lie Group Guide to the Universe 15
(44)
The algebra product is a Clifford product . It can be decomposed into a symmetric inner
part and an antisymmetric exterior product u v = u . v + u ∧ v . The dot product represents
....
[u, v] = u v − v u
def
the Lie bracket (45)
(46)
from those there follows antisymmetry [u, v] = −[v, u]. Adapting to tradition, we may denote
this Lie algebra as γλ(Cl, ○). Then, historically, the construction led us from the Grassmann
algebra ∧V to the category of general linear Lie algebras over Clifford algebras, that is we
have a functor Cl gL that takes us from the general associative Grassmann algebras to the
category of general non-associative linear Lie algebras over linear spaces of Clifford algebras
Cl gL
∧V
→ γλ(Cl, ○) (47)
Cl p ,q = Cl p ,q f ⊕ Cl p ,q g (48)
Note that while Clp,q has dimension 2n the ideals Clp,q f and Clp,q g both have dimension
2n-1. If we further have a maximal set of k positive non-scalar, commuting base unit
monomials e(1), . . . , e(k), we can construct 2k mutually annihilating (symmetrically
orthogonal) primitive idempotents which sum up to unity. In this way one decomposes the
algebra into a sum of minimal left ideals which cannot be further decomposed. We obtain a
minimal left ideal by forming a maximal product of non-annihilating commuting
idempotents. The number k is obtained as
k = q − rq − p with the recursion r j +8 = r j + 4 (49)
j 0 1 2 3 4 5 6 7 8 9 10 11 12
rj 0122333345 6 6 7
All the elements e(j), e(k) commute and generate a finite group of order 2k. The products of
the corresponding idempotents
f ( α ) = 12 (1 ± e(1) ) 12 (1 ± e( 2) ) . . . 12 (1 ± e( k ) ) (51)
are primitive in Clp,q. Therefore each S(α) = Clp,q f(α) is a minimal left ideal in Clp,q. We have
decomposed the algebra with respect to its pure states. Note that α = 1, . . . , k. Clearly, in the
polynomial (51) there appear exactly 2k − 1 commuting Grassmann monomials which squared
give unity. Those form the standard Cartan algebra of Lp,q.
Theorem 1: The Lie algebra Lp,q ≡ γλ(Clp,q) contains a maximal Cartan subalgebra η =
{e(1), . . . , e(c)} of Grassmann monomials e(η) with
( q − rp − q )
|η|= c = 2 −1 (52)
and rp-q the Radon-Hurwitz number to Clp,q. Each such standard Cartan algebra gives rise to
one family of pure states. All the families in Lp,q are mutually isomorphic. Thus each of the 2k
Lie Group Guide to the Universe 17
pure states is defined in a vector space with commuting geometry and gives rise to a minimal
left ideal of the Clifford algebra.
Note: this theorem substantiates the existence of qcd-fermion families.
Proof: is given by the above rigor of section 7 and Cartans theorem number one, saying
all maximal Abelian subalgebras of a semi-simple Lie algebra are mutually isomorphic.
Theorem 1 has a series of important consequences into which we shall be going in the
next two sections.
First, it is interesting to notice that the Clifford algebras Clp,q with p + q = n for even n =
2m accommodate maximally the special unitary group SU(2m). This is resulting from the
order of Cl(n) which equals 2n whereas that of SU(n) is n²-1. Especially the Cl(4)
accommodates as a maximal special unitary subgroup the SU(4) since m = n/2 = 2 and 2² = 4,
and the order of SU(4) equals 4² − 1 = 15. Thus we obtain a series of maximal inclusion
relations as shown in
a †p , a p , a †n , a n , a †Λ , a Λ
(53)
18 Bernd Schmeikal
Those are exactly the ones given in Lipkin 1965, formulas (3.1).
Each of these spaces can be generated by two units of grades 1 and 2. Taking out the
scalar they represent the 6 isomorphic Cartan algebras of Cl3,1 here denoted by “ch roma”.
From the arrangement of roots in A3 there follows the existence of six Lie subalgebras L(2) ∼
sl(3) with rank 2. Each has a root space A2 as shown in figure 2.
Those 12 roots of A3 belong to the algebras so(6, ℂ) ≅ sl(4, ℂ) and to the real forms of
sl(4, ℂ) which are the algebras su(4, ℂ), sl(4, ℝ), su(p, q; ℂ) with p+q = 4. Imaging
Euclidean unit vectors, we can see in A3 a sketch of one special rootspace A2 e.g. for the
su(3, ℂ).
The A3-cube of figure 1 with its 12 roots gives rise to six hexagons. Those represent root
systems A2 which characterize the SU(3) and its various forms. Take a look at
The results found in chapter 3, the New Review of Angular Momentum Algebra concerning
spin space and spinor eigenforms can be generalized. We shall work this out for the L3,1⊂
Cl3,1. In [20, p. 362] and [30, p. 276f.] it has been demonstrated that the graded quantity t3 =
¼ (e24 – e124) represents the general isospin element – (the operator τ0 in Lipkins notation (53
iv)) – in L3,1 and respectively its Clifford algebra. We first seek after general solutions of
equations
these are:
20 Bernd Schmeikal
τ+τ− 1
− = ( Id − e1 + e24 − e124 ) = 14 ( Id − e1 )( Id + e24 ) (57)
16ab 4
This equation is analogous to (6), (10) in angular momentum algebra. The product of
isospin shift operators (Lipkins 53 ii, iii) is equal to the primitive idempotent f13 in [20] and
[30] where it represents the u-quark. This a a pure state in chromatic space ch1. It is obvious
that, - once we have chosen to fixate the τ0 according to the standard basis, - we find two
manifolds T+ = {τ+} and T− = {τ−} which obey the isospin commutation relations, and not just
two operators.
U- and V-Spin
Surprisingly it has never become quite clear how the isospin components of the SU(3) follow
naturally from both the seemingly obvious spacetime geometry and quantum mechanics. We
are looking for a general multivector w fulfilling commutation relations
[ τ 3 , w+ ] = 12 w+ [τ 3 , w− ] = − 12 w− (58)
f = 12 w+ w− (61)
Again, recall what Kauffman [23] said about eigenforms, and verify quantity f = ½ w+w−,
the Clifford product of w+ and w−, is an eigenform in the Clifford algebra Cl3,1. We define
recursively the bilinear term
f n+1 = 1
2
fn fn (62)
and A = a − c , B = b − d , C = ad − bc , D = cd − ab
2 2 2 2
and find that this form is recursively preserved, just as it was in the angular momentum
algebra. Therefore, the fn moves within the space of isospinoreigenforms
The regained fn+1 is a polynomial with nonlinear coefficients having a special structure, e.
g. for the first and last coefficients we have
α = a 4 + b 4 + c 4 + d 4 + 2( a 2 b 2 + c 2 d 2 − a 2 c 2 − a 2 d 2 − b 2 c 2 − b 2 d 2 ) ,
φ = abc 2 + abd 2 + a 2 cd + b 2 cd − a 3 b − ab 3 − c 3 d − cd 3
Anyway, as long as the w+, w− have the symmetric structure (59), (60), the isospinor
eigenform moves in 6-dimensional space
which is a bit smaller than in (63). That larger space comes in as soon as the symmetry of
equations (59), (60) is broken and coefficients in (59) for w+ deviate from those for w−. Only
the space of (63) is closed for Clifford multiplication. Therefore the fn are eigenforms.
That linear space which generates the space of isospinor eigenforms by Clifford
multiplication is called the core isospin space.
Actually there is a stable manifold of fixed points fn in E(f). The equation (61) denotes
only one of those, though apparently a rather prominent one. This situation must be
investigated more closely. In the 2004 work on transposition involutions there have been used
two isospin multivectors that were denoted as u and v. Those are special solutions within the
manifolds of (59) and (60). Since these manifolds can be split such that the SU(3)
commutation relations are preserved.
(68)
and take all constants a = b = c = d = ¼ . Then we obtain the special forms of u+, u− , v+, v−
in the standard representation of L3,1.
Please, observe quantities λ4, λ5, λ6, λ7 are already generators of the Lie algebra L(2) ∼ a
real Clifform slCl(3) of the Lie algebra sl(3, ℝ).
For completeness, consider the isoform fn = ½ v+ v− as derived from general v-spin (68).
We have
We can carry out the same rigor for the general u± spin given by formulas (67). We
obtain
(75)
What we need to show is that this is indeed a real normal form of the Lie algebra to the
special unitary group SU(3). We can do this without going into the whole classification
machinery [31], [32]. Consider the real representation
(76)
The Gell-Mann matrices are now identical with the following set
τ 0 = 12 (a †p a p − a †n a n ) , λ 3 = 14 (e24 − e124 )
N = 13 ( −2a †Λ a Λ + a †p a p + a †n a n ) λ 8 = 1
2 3
(−2e1 + e 24 + e124 ) (78)
b= y−s,
†
Watch the Cartan subalgebra {τ0, λ8} to derive the interpretation of e24 ∼ a p a p as
0 0
1 1
B= 3 and λ = 1 (79)
1
3
8
3 1
1
3 − 2
fj
→ U k j f j that preserved the norm ∑ f j* f j (80)
j
Such matrices satisfy the relation U† = U−1 and form a group, in accordance with the
defining equation for the inverse:
This group of matrices ⊂ Mat(3, ℂ) is the unitary group U(3). Unitarity imposes nine
constraints on the 18 real degrees of freedom in a 3×3 matrix with complex entries. So the
U(3) has dim = 9. The norm is invariant under a transposition by a phase eiα . Therefore U(3)
Lie Group Guide to the Universe 25
can be decomposed into a direct product U(1) × SU(3), where SU(3) has unit determinant and
one less degree of freedom, that is, dim = 8. This result could be generalized. The unitary
group U(n) incorporates the special unitary group via the equation U (n) = U(1) × SU(n). In
terms of some predicate logic calculus that whole problem could be reformulated as follows:
Consider an associative unary algebra. Find those transformations that preserve the predicate
“unary”. As we know that from every unit element e there can be derived two idempotents f =
½ (1 ± e) we can just as well demand the existence of a group which preserves the quality of
»idempotent«. This shall be exactly the way we shall pose this problem: Find the group which
unfolds the primitive idempotent manifold! We begin with pure states.
Pure states in Clifford algebras Clp,q are given by the 2k primitive idempotents as
formulated in (51)
f ( α ) = 12 (1 ± e(1) ) 12 (1 ± e( 2) ) . . . 12 (1 ± e( k ) ) with k = q − rq − p
These pure states can be unfurled to manifolds, − with mutual orthogonality relations
preserved. The tools to unfold the idempotent manifolds in that way, are subalgebras L in the
Lie algebras Lp,q ≡ γλ(Clp,q). Consider any Clifford number λ∈ Lp,q and the quantity obtained
by the exponential map
def
∆ = {∆ (α ) } = {g f (α ) g −1 / g ∈ L ⊆ L p ,q } (81)
brought forth by conjugation. The invariance of orthogonality relations, therefore, means that
f j f k = 0 ⇒ ( g f j g −1 )( g f k g −1 ) = 0 (82)
The first is taken to be a representative of the electron neutrino νe. The others are
interpreted as fermion states. (Schmeikal 2004). The neutrino is fixed by its stabilizer algebra,
namely L(2). Therefore the f11 annihilates the Lie algebra [32, p. 81f.]
f11 expL(2) = { f11 }that is, f11 absorbs group SLCl(3) (85)
The other primitive idempotents are not fixed, but transformed by carrying out the
conjugation. Algebra L(2) contains an important element, namely the universal generator l of
both the trigonal color- and flavor rotations. This is the multivector
l=( 2
3
arccos(− 1 2 ))(λ 2 + λ 5 + λ 7 ) and exponential T = e l (86)
The universal T is brought forth by two reflections which on their part are given by
standard primitive idempotents in the pure space
together with transpositions τ(g) = 1– 2g and τ(h) = 1 – 2h. Those are Coxeter reflections in a
large finite automorphism group [20, p. 358] of transpositions. We calculate
5
„Gell-Mann nannte dieses Schema Eightfold Way, eine Bezeichnung die die Oktette des Modells mit dem
Achtfachen Pfad des Buddhismus verbindet. Er prägte auch den Namen Quark, den er aus dem Satz ‚Three
quarks for Muster Mark’ aus James Joyce’s Roman Finnegans Wake entnahm. Da einzelne Quarks in
Experimenten nie beobachtet wurden, bezeichnete Gell-Mann selbst sie als mathematische Fiktion.“
(Wikipedia 2007/08: „Quark+Physik“)
28 Bernd Schmeikal
K − + p+ → K − + K + or equivalently
→ Λ + φ
u s + uu d → u d s + s s → u d s + u s + su
In a 1999 private communication to Harald Fritzsch [40], [41] George Zweig wrote: “ In
the April 15, 1963 Physical Review Letters there is a paper [42] titled ‘Existence and
Properties of the φ - Meson’ . I remember being very surprised by Figure 1, which showed a
Dalitz plot for the reaction K− + p → Λ + K− + K+. There was an enormous peak at about
(1020 MeV)² in the M²-plot for K K , right at the edge of the phase space. The fact that the φ
decayed predominantly into K K and not ρ π was totally unintelligibly despite the authors’
assurance that this suppression ‘need not be disconcerting’. [ …] Here was a reaction that
was allowed but did not proceed!” 6 Those experiments convinced Zweig that baryons had
constituents and were not just “hypothetical objects carrying the symmetries of the theory, but
real objects that moved in spacetime from hadron to hadron.” One just had to assign the
correct constituents to pseudoscalar and vector mesons in order to, among other things,
explain the ρ π-suppression. Zweig then reported how Richard Feynman at Nino’s 1964 Erice
summer school did not believe his arguments. But, “later that fall, when I gave Gell-Mann my
explanation of φ-decay and drew my diagram for φ → ρ π (which involved polygon block-
like icons for the constituents), I can still hear him saying “Oh, the concrete quark model!”
Despite the mystification of the authority of the person Gell-Mann it must be realized that the
classification of hadrons in the tenfold way which predicted the existence of the isoscalar
baryon with strangeness −1 (the Ω −) was not only understood by Goldberg and Ne’eman, but
it was also noted by Gell-Mann and others. Lipkin points out, the right classification goes
back to Glashow and Sakurai. They predicted the existence of the Ω−. When I was young I
was proud to hear that, obviously without knowing, the scanners of our HEPhy group in
Vienna were among those three groups who had correctly scanned the omega-minus. Some of
us studied the tenfold way and various mathematical models that should explain the decays.
Then, everything was very fascinating and unclear at the same time. As a scholar of Walter
Thirring I became busy with a thought which did not let me go until today. Those symmetries
were properties of spacetime. Then, the most important problem was that the Ω − = s s s, ∆− =
d d d and ∆++ = u u u violated the Fermi statistics because they contained three identical spin
½ fermions in a space symmetric state coupled symmetrically to spin 3/2. This was first
solved by Greenbergs parastatistics [43] and later led to the introduction of a new degree of
freedom, namely color. The new theory was supported and further developed by Yōichirō
Nambu [44], [45]. His qcd answered the fundamental questions of high energy physics. With
this now in mind we can solve the qcd-problems in geometric algebra.
6
quoted after Harald Frizsch 2002 [40]
Lie Group Guide to the Universe 29
Id a b c
a Id c b
b c Id a
c b a Id
with coefficients
A0 = x0 + x1 + x 2 + x3
A1 = x0 + x1 − x 2 − x3
A2 = x0 − x1 + x 2 − x3
A3 = x 0 − x1 − x 2 + x3 (92)
ψ 0 = 12 (1 + a) 12 (1 + b) ψ1 = 12 (1 + a) 12 (1 − b)
ψ 2 = 12 (1 − a ) 12 (1 + b) ψ 3 = 12 (1 − a) 12 (1 − b) (93)
7
It is important to sometimes mention symmetric or bilateral orthogonality because there also exist other types of
asymmetric unilateral orthogonality
30 Bernd Schmeikal
Equations (90) to (92) make clear that, in a unary algebra, we have two ways to represent
the multivector ξ. First, we can represent it in the ground space of base units as span{Id, a, b,
c} and second, we can represent it in the idempotent space span{ψ0, ψ1, ψ2, ψ3}. I have
variously called those units “extensions”8 because they really reflect what Spinoza and
Descartes meant by extension in contrast to cognition. What is so special about these
quantities? Each of the three extensions a, b, c have different grade. The grade (or length) is
increasing from 1 to 3. Time appears in grades 2 and 3 as wrapped up and disguised in a
space-like unit which squared gives one. Surprisingly, the grade doesn’t seem to make any
difference. Each quantity appears as equally important within its color space. Moreover, the
quadratic form can be extended over the whole positive definite subspace (10-dimensional in
Cl3,1). The quadratic form has a continuation over different grades. Last not least, the
universal trigonal SU(3) operation applies equally to both the unit vectors and the pure states.
That is, the T acts on the states and on the extensions in the same definite way:
8
being aware of other such denotations as for example “algebraic extension” in Mathematics Subject Classification
2000: 12F05
9
This is the Λ-direction in Lipkins Sakaton model.
Lie Group Guide to the Universe 31
Next we derive the cylindrical annihilation- and creation operators in three directions and
bilinear forms which satisfy the SU(3) relations
H = ℏω (a †d a d + au† au + a †s a s + 32 ) Hamiltonian
τ + = a d† au τ − = au† a d
τ 0 = 12 ( a d† a d − au† au ) (94)
B+ = a †d a s B− = au† a s
C + = a †s au C − = a †s a d
Y = 13 ( a †d a d + a u† au − 2a †s a s )
This model is essentially the same as the Sakata model. It is related to Elliott’s Nuclear
Shell Modell [46]. Hadrons appear as degenerate eigenstates of the oscillator Hamiltonian.
The bilinear forms of τ±, B± and C± shift quanta between the primitive idempotents ψ1 = f12 or
selected direction e1, the plane e24 or respectively primitive idempotent ψ2 = f13 and spacetime
volume e124 or primitive idempotent ψ3 = f14. Quantum number Y is one-third the difference
between the number of quanta in the isodublet space {u, d} and twice the number of quanta in
the distinguished “strange direction” s. We have Y = 0 when the number of quanta in all three
directions are equal. Y > 0 when the average number of quanta in the u-, d- directions is larger
than in the s-direction, and Y < 0 when strange fermion energy dominates. Thus Y measures
the departure from spherical symmetry, the deformation of the field. This has a special
importance because the three directions have different grades within the Clifford algebra.
Though their grade is unimportant as long as we restrict our rigor to one distinguished color
space ch1. The special importance of the grading is the following. Recalling chapter 5, it is
clear we have a commutative space within a C*-algebra. The GNS-representation of the
fermion states is definitely operating in a Hilbert space H = L²(X). The irreducible
representations are 1-dimensional. The associated characters of the algebra are points.
Probability measures are Dirac δ-functions located at the ‘points’ of X. But we have not only
point locations, but also definite areas and sharp spacetime volumes. Between those there
occur graded transpositions which can best be compared with maps between discontinua such
as (Peano) fractals having dimensions 1, 2 and 3. Dislocations as annihilations of point
locations in favour of sharp “δ-functioned” spacetime areas and spacetime-volumes are
possible. I denote this process as annihilation and creation of extensions. Suppose we count
three quanta on the spacetime oscillator. That would allow for ten fermion combinations
which form the well known baryon decuplet.
quarks hadron S B Y τ0
sss Ω− −3 1 −2 0
1
ssu Ξ0 −2 1 −1 2
ssd Ξ− −2 1 −1 − 12
suu Σ+ −1 1 0 1
sdd Σ− −1 1 0 −1
sud Σ0 −1 1 0 0
ddu ∆0 0 1 1 − 12
∆+
1
uud 0 1 1 2
∆+ +
3
uuu 0 1 1 2
ddd ∆− 0 1 1 − 32
Lie Group Guide to the Universe 33
TC = (1 − 2 f 34 )(1 − 2 f 52 ) (95)
where f 34 = 1
2
(1 − e2 ) 12 (1 − e34 ) and f 52 = 12 (1 + e3 ) 12 (1 − e14 )
−1
Verify TC f12TC = 1
2
(1 + e2 ) 12 (1 − e34 ) = f 32 . That means TC carries the s-quark
pure state from color space ch1 to color space ch3. There exist universal trigonal rotations
which transport quarks of each family in 3-cycles from color to color. Each color space
provides a definite color and contains three quarks such as u, d, s. Considering u, d, s as one
ternary fermion family, spaces ch1, ch3, ch5 provide three different colors for that family.
However, consider the transition from ch1 to ch4. It represents a special involutive
automorphism of the Clifford algebra, namely a transposition or exchange; in the present case
a transposition of base unit of type e1 onto e2, so that the triple {e1, e24, e124} goes over into
{e2, e14, e124}. The spacetime unit-volume e124 is preserved. We may assume that
Transpositions of form
which wrap and unwrap time on a distinguished unit that indicates strangeness represents an
energy consuming symmetry breakage. This could be a reason why we observe the up, down
and strange quarks in spaces 1, 3, 5 and the high energy fermions charm, bottom and top in
representation spaces ch2, ch4 and ch6. So we have six different types of strong interacting
fields which constitute the spacetime algebra.
Once recognized, this has enormous theoretical and empirical consequences. To see this,
consider the directed spacetime area e24 in a dynamic oscillation with real scalar x such as x
e24 = (sin ωt) e24. Clearly, this quantity changes sign in accordance with the sine function.
Taking into account the anticommutation relation
the sign change of ψ can be interpreted in two ways, namely first as a turnaround of x and
second as a reversal of the spacetime area. The latter includes time reversal and all the rest of
it. Let us go into this further because it concerns CP-violation and the Cabibbo-Kobayashi-
Maskawa−matrix, CKM−matrix, a 3×3 unitary matrix describing mass-mixtures of fermion
eigenstates in strong transitions with W+ Boson emission . The Cl3,1 pure states have detailed
forms
Lie Group Guide to the Universe 35
From this we learn that orientation reversal of bivector e24 can be transposed onto a
reversal of the spacetime volume e124 which compensates the necessity of a time-reversal,
since the turning back can occur on the bivector part of the exterior product e124 = e12 ∧ e4.
Therefore, provided such geometric interpretation is valid, within the isodublet {u, d}
causality violation can always be compensated for by an exchange of u and d-quarks. A
similar argument, however, does not apply for the strange quark; pairs {s, u}, {s, d}. Once the
oscillator on its spacetime volume is in phase on the area e12 a simultaneous excitation of the
oscillator on e24 causes a split between an advanced and a retarded wave. The strange fermion
excitation got to be accompanied by a cloud of virtual processes. Note, for the singlet {s}
parity on the distinguished direction e1 is opposite in the dublet {u, d}. The same statement
holds for the spacetime areas e14 and e34 and the other isodublets {c, s} and {t, b}. We should
expect transition rates u↔d, c↔s, t↔b significantly larger than the others. Our theory
confirms the 2006 matrix of the Particle data group [47].
(V ) = 00,97383
ij
,2271
0,00396
0,2272
0,97296 0,04221
0,00814 0,04161 0,999100
Going further into the cosmos, we can model antimatter either by space inversion or by
complexifying the Clifford algebra. Some detailed bookkeeping makes clear that we have to
proceed with the complex matrix algebra Mat(4, ℂ) which is isomorphic with the Cl4,1. These
coherences have been shown in [48, p. 73] and [12, p. 217]. The implementation of Higgs
bosons and dark matter demands Cl4,2.
with very large masses. The simplest such model, he argues, could use the special unitary
group SU(5). Then he shows how the pentaplets decompose in accordance with the standard
model. Lipkin in his 1965 work does not consider the SU(5), but concentrates on the
multiplets of the SU(4), SU(6) and SU(12). The SU(4) fits perfectly into the limits of the
Clifford algebra Cl3,1 of the Minkowski spacetime in the Lorentz metric. Lipkin shows us,
how we can go on from SU(4) to find the SU(5) multiplets. Its just the same as going from
SU(3) on to SU(4), just one step further. All these considerations become very interesting, as
soon as we can relate fields to geometric algebra of spacetime. Its actually the way Einstein
and Mach dealt with theory: it’s the fields that bring forth the spacetime. The matter makes
the observed properties of space and time. We can read out the features of space and time by
understanding the experiments we make with matter. This was Mach’s idea. [50]
It is clear that the quadratic Clifford algebra Cl4,1 having dimension 32 cannot embed any
appropriate unifying symmetry group because its maximal Cartan subalgebra also provides
maximal rank 3 to the Lie groups, just like the spacetime algebra Cl3,1. From such rank 3
algebra we must construct the rank 2 Lie groups as Clifforms of SU(3) which are graded and
therefore allow for a norm- and othogonality preserving transformation of pure states within
the whole ten-dimensional positive definite subspace. This procedure is not altered by moving
from Cl3,1 to Cl4,1 because of the form of the basic primitive idempotent. Just the number of
isomorphic Cartan subalgebras is doubled. This can indeed be used to construct some sort of
additional Baryon like dark matter fermions which contribute to Massive Compact Halo
Objects (MACHOs). But such approach does not meet the main riches of the present
conception, nor does it cope with the astrophysics findings. Dark matter and energy, hot and
cold, are not baryonic. The most preferred candidates of cold dark matter are weakly
interacting massive particles (WIMPs).
Taking into account the simple ΛCDM10 model which describes cosmic evolution by the
cosmological constant Λ and six more parameters, cosmologists found the contingent of
baryonic matter on the critical density (9,7 · 10−26 kg m-3) equal to only 4,44% and that of
matter, CDM inclusive 26,6%. Hot dark matter provided by the known three neutrinos would
lead to a false scenario. The best explanation is given by the WIMPs. This is in perfect
correspondence with a model based on a Lie algebra γλ(Clp,q) of a rank higher than 3. We
shall have to investigate the Clifford algebra generated by a space similar to a de Sitter space:
»adS6« and respectively γλ(Cl4,2) having two time coordinates. Because of its peculiar
primitive idempotent structure Clifford algebra Cl4,2 allocates some invariant massive
neutrino pure states. Those are relatives of the known neutrinos, but have higher energy.
Clearly, we are then operating in a theoretical domain related to the Randall-Sundrum model
[51], [52]. The pure states of wimps are consistently incorporated into the Lie algebra as fixed
points in the manifold, and we can establish the unified field, supersymmetry and the Higgs
mechanism [53]. We shall slowly go into this. But first I have to introduce us into a laudable
approach – small paths that have been trodden by some of us − before we left the light and
correctly stepped out into the dark.
10
CDM = Cold Dark Matter
Lie Group Guide to the Universe 37
Sometime in 2007 I reviewed an interesting paper by Robert Gordon Wallace [54] from
Australia. It seemed somehow extraterrestrial like my own, - realized Clifford algebra in
some strange matrix modules carrying out basic rigor by a spreadsheet. So I first had to
translate everything into the languages I used. Those were Maple Clifford by Rafal
Ablamowicz and Bertfried Fauser and my old “handy” Clical by our deceased friend Pertti
Lounesto. Both programs use indices for concrete base units, whereas Wallace used mere
capital letters for a bunch of matrices, - why not? Verbatim his abstract stated “A scheme of
symmetry breakage can be imposed on orthogonal directed lineelements for the algebra sl(4,
ℂ) which, for Cl+3,1, Cl+4,0 and Cl+2,2 subalgebras results in a pattern corresponding to the
standard model, together with elements corresponding to fundamental particles of dark
matter.” Wallace had correctly picked up some algebraic investigation of my own 2001 work
[48] and tried to lift the Clifford algebraic version of the Weinberg Salam theory of
electroweak interaction worked out by David Hestenes into those Cl4,1 spaces that
incorporated the Clifform of SU(3) developed by me. But he was not fully aware of the
consequences of the fact that the Clifform was not a Lipschitz group, but was based on some
graded algebra, the grading of which became invisible, indeed, by using matrices from sl(4,
ℂ). It behaved different than the familiar spinor models. So he did his best to structurally lift
Hestenes’ approach into the appropriate Clifford algebra. He practically found all the
essential electroweak modules, but wisely did not fix the standard model. Nevertheless he
visualized a possible origin of hot dark matter and the asymmetry between light photons and
heavy Higgs bosons. The mathematical origin of the symmetry breakage is so original –
almost witty – that I want to give us a sketch. It seems Wallace used a basis as follows
and an imaginary part which is simply equal to iCl3,1. Thus the basis vector e4 ∈ Cl4,1 is the
complexified Graßmann product ie1235 and the Cl4,1 gets a real and an imaginary part, that is,
it is decomposed as
The 10-dimensional subspace P3,1 with positive definite signature turns into a 10-
dimensional subspace with negative definite signature. Both taken together give a module
isomorphic with the Cl4,1. Using a basis
0 0 − 1 0 0 −1 0 0
0 0 0 1 −1 0 0 0
e1 = e2 = (102)
− 1 0 0 0 0 0 0 − 1
0 1 0 0 0 0 −1 0
− 1 0 0 0 0 0 i 0 0 1 00
0 1 0 0
0 0 0 − i − 1 0 0 0
e3 = e4 = e =
0 0 1 0 − i 0 0 0 5 0 0 0 1
0 0 0 − 1 0 i 0 0 0 0 − 1 0
It can easily be verified how a general element of some special electroweak group
SU(2)×U(1) can be constructed for instance as
Wallace gives matrix representations of the electroweak special unitary symmetry for
some fermions like the red u-quark. Using his spreadsheet he detected six fermionic
subalgebras of the type s(u(2)×u(1)) with real pseudoscalar base units, corresponding to quark
families of ordinary matter and nine such algebras with imaginary pseudoscalars for dark
matter fermions. To give an example, a red up quark can be represented in space
(104)
The model also contains bosonic subalgebras su(2)⊕su(2) for Higgs bosons W+, W−, Z0
and photons. It has space for supersymmetry by two times three special (pseudo)unitary
bosonic subalgebras su(1,1)⊕su(1,1). Wallace argues that the four anti-commuting triples
{e21, e13, e23}, {e234, e134, e12}, {e234, e124, e13}, {e134, e124, e23} represent building blocks for
Goldstone- and Higgs bosons. The symmetry breaking distinguishes the first triple from the
other. It is indeed possible to test their behaviour in strong interacting processes by applying t,
some universal trigonal operator as in (86), (87). Take
It turns out that up to direction only the first triple is preserved under a t-rotation which
says nothing else than that the triple {e21, e13, e23}can be denoted as an electromagnetic triple
or photon, since it does not interact strongly with matter. So far the strength of the Wallace
model, - it actually can describe a structure of hot dark matter,- but let us also see the
weakness of it in case we want to describe the wimps, the cold and weak dark matter. The
model is still too strong to describe the frail. We got to understand the meaning of primitive
idempotents and the special role of heavy neutrinos.
s −1 f12 s = f13 but the left (or right) –sided equation (107)
has no solution for ρ. This is different to the elder Lipschitz approach. What is the difference?
We have a spin group, namely L(2), but no rotation group of which it is a double cover. Yet,
this does not impair the action of the Lie group which is based on the Clifford commutator. It
also imposes no limitation on formulating the standard model. But it needs a new
interpretation of motion.
Strictly, we have an inner Lagrange function in the commutative subspaces and an outer
Lagrange in the whole space. We can formulate two norms and two equations of motion for
inner and outer spacetime.
We understand why, because of the role of pure states and the neutrino fix point Ansatz,
we have to have a heavy neutrino in the Clifford algebra Cl4,2 which satisfies the equation
40 Bernd Schmeikal
f11 expL(6) = { f11 } that is, f11 absorbs group exp L(6) (110)
We have solved this problem entirely for the Cl3,1. To comprehend its logic for the de
Sitter space: »adS6« in L(6) we first go back to the “Lie groups for Pedestrians” investigation.
Harry Lipkin began with the bottom up construction with the diagonal isospin su(2) operators
B = a †p a p + a †n a n baryon number
τ 0 = 12 (a †p a p − a †n a n ) isospin (111)
B = a †p a p + a †n a n + a Λ† a Λ baryon number
τ 0 = 12 (a †p a p − a †n a n ) isospin
with a new quantum number C then called ‘charm’ while Z was simply “a new operator Z”.
We might call it Z-charge, but we are aware of the terminological changes that have occurred
since 1965. Today we pack both charm and strangeness into the su(3) algebra and the indices
in the Sakata SU(3) have turned from p, n, Λ to d, u, s.
Lipkin’s bottom up construction can be continued, and we wish to go further to the SU(7)
in Cl4,2. Before doing so we take a look at the su(3) in (112) and its relation to the L(2) in Cl3,1.
Recall the analogy in equations (78) between Lipkin’s and the pure state Ansatz of Cl3,1.
After having found the graded ‘quasi Lipschitz’ group for the Clifform of su(3) generated
by the λ-Clifford numbers (75) we realized that the pure state f11 representing the e-neutrino
annihilated each λ i and therefore the whole algebra, that is, f11 L(2) = {0}. Now, that we
have a clear image of the algebra L(2) we can make a test and find out by a computer program
how those manifolds X look like that satisfy an equation of the form f11 X = {0} in Cl3,1.
Maple Clifford gives us a return like the following: starting off with a manifold
Lie Group Guide to the Universe 41
showing how this manifold which is absorbed by the pure neutrino state decomposes into
exactly 4 parts. The first is given by the Cartan algebra, that is, 3rd component isospin plus
hypercharge, the second by v-spin, the third by u-spin and the last by isospin t. Note, we have
a partition of the manifold into blocks of 4 times 4 coordinates of X. It is tempting to ask what
kind of partition we would obtain for the group L(6) in Cl4,2? We are aware, according to Table
2, the largest su(n) within Cl4,2 is of course the su(8) having dim = 65 and rank 7 (8 minus 1).
The group SU(8) ⊂ L4,2 is virtually clung to the Cl4,2. To rotate the pure states we need to
construct the group L(6) from L4,2 which has a rank by 1 smaller than that of L4,2 . A standard
neutrino fix point is the Clifford number
f11 X = 0 (115)
Since Cl4,2 has dimension 64 the general multivector element has 64 coordinates x1, x2,
…, x64 to base units Id, e1, …, j = e123456. Solving the equation (115) brings forth 8 manifolds
as shown below. Each having 7 degrees of freedom while a single coordinate is determined
by the seven chosen ones.
Consider the first equation. It is in correspondence with the first row in table 4. This
characterized the Cartan algebra of L(2). Now it stands for the Cartan algebra in L(6). Consider
Those unit vectors are exactly the commuting base units which constitute the invariant
neutrino pure state (114) of SU(7). Following the bottom up construction employed by
Lipkin, we obtain a W-charge for L(6) which corresponds perfectly with a field hypercharge
number operator W constructed by bilinear products of creation and annihilation operators,
namely
w= 1
(e
14 1
+ e25 + e36 + e125 + e136 + e2356 − 6e12356 ) constituted by
We have added to flavor a new quantum number w . The W-charge measures one seventh
the difference between the number of subnuclear particles with SU(3) baryon number ⅓ and
six times the number of w-fermions. We have W = 0 when the average number of quanta
constituting the six bottom multidirections equals the contribution to the top spacetime
volume e12356, the “top multivector unit” in the Cartan algebra. There is W > 0 when the
average number of quanta on the u-, d, s, c, b, t flavor-directions is larger than on the w-
volume, and Y < 0 when w-fermion energy dominates. Thus as in the case of SU(3), quantity
W measures departure from spherical symmetry and respectively deformation of the field.
Following the SU(n) construction plan, we have added a new, say, very strange quantum
number ‘beyond’ YO = − a w a w such that the w-fermion has ‘beyondness’ −1 while the other
†
1
W= B + YO (118)
7
The algebra contains 7 creation- and 7 annihilation operators for spacetime oscillators
and respectively subnuclear fields. Those bring on 49 bilinear products. The Lie algebra L(6)
is of rank 6 since we can now find six Clifford numbers which commute with one another.
There is a total number operator B together with the following 6 commuting Clifford-charges.
See the bottom up construction:
τ 0 = 12 (a †d a d − a u† au ) SU(2)
t 3 = (e25 − e125 )
1 ⊂ L(6)
4
Z = 14 (a †d a d + au† au + a †s a s − 3a c† a c ) SU(4)
z = 18 ( + e1 + e25 + e36 − 3e125 ) ⊂ L(6)
Clearly, we could just as well have used any other linear combination of these quantities.
The reason for this particular choice follows the argument given by Lipkin in [21 a, b]:
namely to allow the subgroups SU(2), SU(3) until to SU(6) to be used in the classification of
the SU(7) multiplets, one after the other. Subtracting the six plus one (for B) there remains the
† † † †
algebra of 42 bilinear products having forms like a d au , au a d , …, a d a w , a w a d and so
forth. All those operators which annihilate a baryon and create a beyond state or annihilate a
beyond and create a baryon, change the eigenvalue of W, and respectively w, by ±1. There are
12 of them; the other 30 commute with W. From the expression of W in table 6 we conclude
that it takes eigenvalues
W . . . n, n ± 17 , n ± 72 , n ± 3
7
with integer n (119)
44 Bernd Schmeikal
def
w0 = 12 (1 + e1 ) 12 (1 + e25 ) 12 (1 + e36 ) wimp (120)
identical with the primitive idempotent f11 in equation (114). The wimp annihilates its Lie
group and becomes invisible to interaction. It partakes only in electroweak interaction. By
balancing out the contributions of the constitutive oscillators it segregates from strong
interaction. Thereby it switches off for itself the generalized Pauli principle which holds for
fermions. As untouchable as the wimp appears to be in the Clifford algebra, as cool and
invisible it may appear to us in outer space. But the reason why this model works so well is
less in its objectivity, - since there are no objects beyond our experience, - but in the fact that
it can be thought so well. The central equation is indeed »wimp×L(6) = 0 «. If we start from
this equation we can construct a concrete form of the Lie algebra and its group just as we
have done in the case of L(2) for the SU(3).
Clearly, the form (119) is bound to the standard representation. The Lie group L(6)
unfolds from this form a manifold having the dimension of the positive definite subspace
dim(P4,2) = 36 [33, p. 128, table 4]. So we have a manifold of wimps just as we must have a
manifold of fermions, baryons and bosons. But there is not only one such manifold, but there
are 24 equivalence classes which arise from the permutation of the 6 indices in the standard
representation of Cl4,2. In the Minkowski spacetime algebra Cl3,1 we had only six leptons, in
the Cl4,2 we have 24 of them. This can easily be verified by realizing the matrix
representations of Cl3,1 ∼ Mat(4, ℝ) and Cl4,2 ∼ Mat(8 ℝ). Matrices of Cl4,2 are 8× 8 and have
4 times as many elements as Cl3,1 matrices.
Lie Group Guide to the Universe 45
± γ 12 p . x / ℏ
Ψ ∓ = const. e (120)
The bivector e12 = e1∧e2 is the directed unit area which squared gives −1 and therefore
often merely replaces the imaginary unit. The situation becomes more complicated as we
have to interpret multivectors with a positive signature such as e24 in Cl3,1 or e25, e125 in Cl4,2.
Those are space-like but contain time as a factor. They have their own algebra as depicted in
chapter 14 and give rise to some peculiar logic of quantumchromodynamic projectors [56].
From a thorough investigation of the algebras of spacetime it has become evident to me that
the phenomena of qcd have a definite origin in the organization of space and time. This origin
has been scanned in all my recent works until to this one. Let me first point out how the Pauli
principle can be realized by the following motion picture
Ψ ( x, t ) = Ψ0 sin(ωt − kx + ϕ) (121)
Elements of Motion
When the phase is zero and kx = ωt the wave reaches maximum amplitude Ψ0. For light
waves we have c = ω/k. We use to interpret this image by fixing either the locus x or the point
in time t. At any fixed x such as x = 0 the wave function Ψ performs an oscillation Ψ0 sin (ωt
+ ϕ).
The Synchronous
This pattern represents the time process incarnated in a synchronous spatial structure.
Lie Group Guide to the Universe 47
Next consider the universal flavor rotation acting on a vector in the Minkowski space ℝ3,1 and
respectively Cl3,1
Therefore the periodic pattern alongside the unit vector e1 is turned into a periodic pattern
alongside the spacetime area e24 (recall figure 6). However, while x increases linearly, the
area measure increases with its quadratic form. The equidistant nodal points turn into closed
nodal lines with decreasing distance. Yet, all three quantities xe1, xe24, xe124 behave the same
way despite the grading. They commute and allow for 1-norm dynamics (see table 3 in
chapter 14, and [56]). As long as we consider motion within one algebraically closed color
space such as ch1, those quantities are absolutely equivalent.
Consider spaces such as ch1 or plane spanned by {e1, e4} There are various movements we
want to consider, first the movements of elements in those spaces, second the movement of
locations, such as the oscillation of spacetime areas, third the time evolution of scalar
functions such as (121). In case of Clifford algebra, the movement of elements is the most
general and comprises all three. A pattern in ch1 projected onto e1 may look like
By a process (123) two things are ensured. First during strong force action the Pauli
principle is preserved. Second the wave pattern on e1 is carried onto a space-like extension
quantity e24 and further to e124. To see what happens thereby we must not only understand the
meaning of an oriented space area, - a bivector which squared gives −1, - but also that of a
space-like bivector which squared gives +1. This bivector which is the exterior product of a
space-like vector e1 and a time-like vector e4, does not behave like a bivector e12 in the
euclidean Pauli algebra. See the difference in figure 10!
The pattern can be deconvoluted further onto the spacetime volume e124. By T it can also
be convoluted back into e1. You see, the generalized Pauli principle offers an impressive
degree of freedom for qcd. The situation becomes even more interesting once we realize the
deep origin of those degrees of freedom which transpose base units of the Clifford algebras
48 Bernd Schmeikal
Despite the reflections 1−2f (with idempotent f) some of the most important
transpositions in the Clifford algebras are the Weyl reflections [20, p. 356]. These transpose
the Euclidean unit vectors onto each other. Weyl reflections contribute to the elements of
motion in qcd. Namely, suppose there is convoluted into one of the euclidean base units e1, e2,
e3 some synchronous pattern originating from areas or respectively volumes e14, e24, e34, e124,
e134, e234. Such a pattern can be distributed by the Weyl automorphisms among the three
manifolds equivalent with base units e1, e2, e3. Once a pattern has become imprinted into the
volume e123 by the generalized Pauli principle it is transposed onto real synchronic time by
strong interaction. In order to see this process more clearly, we have to investigate what I
called the Zeit Dreibein or orthogonal chromatic time-like space or time 3-space.
Lie Group Guide to the Universe 49
The well known Cartan structure η0 is not repeated, yet reflected in the temporal triangle
e4, e123, e1234. Commutators of these quantities do not vanish, but bring on a closed Lie
subalgebra, while their anticommutator vanishes. These base units form an orthogonal Zeit
Dreibein in Cl3,1, just like the unit vectors of Euclidean 3-space. Observe the following
relations carefully:
If this structure has any physical meaning, - and I feel it has a very deep meaning, - a
local temporal oscillation can be fully transposed onto a periodic pattern in volume e123 and
respectively spacetime volume e1234. It would further mean that in strong interaction a field
may be excited and thereafter folded into and deconvoluted from periodic spacetime
structures. This could be the explanation for the appearance of various unexpected effects like
a light velocity spectrum, violation of RT [57] and principles of RT, or quantum vacuum
forces [58] which alter the metric and orientation of time. When I first became aware of these
matters, I suddenly saw the empirical meaning of the decomposition theorem by Banach and
Tarski [59], [60]. If the actualities of the climatic situation permit some more research in
HEPhy and if we regain our scientific virtue and become a bit more careful, then one of our
next discoveries might be some interesting non-decidability: as a matter of principle where
extension and age of the universe are concerned. Creation is not compatible with the concept
of nature in a stable isometric spacetime.
50 Bernd Schmeikal
conformally invariant. Next, he states, Cunningham (1909/10) and Bateman (1910) already
had verified, the Maxwell equations in four dimensions are also conformally invariant. In
Voigt’s proof the d’Alembert operator acquires a scalar factor.11 Voigt considered coordinate
transformations with real constants a, c, v, and f = a²(1-(v/c)²) ∈ℝ
x ′ = a ( x − vt )
t ′ = a (t − vx / c 2 ) dt ′ ∧ dx ′ = fdt ∧ dx (124)
dt ′ ∧ dt ≠ 0 (125)
x = af −1
( x ′ + vt ′)
t = af −1 (t ′ + vx ′ / c 2 )
in other terms
{x, t , x ′, t ′, v} ≠ {x ′, t ′, x, t ,−v}
The d’Alembert-Laplace operator acquires a scalar factor f. The metric tensor on vector
fields is
g −1 = −c 2 dt ⊗ dt + dx ⊗ dx = f −1
{− c dt ′ ⊗ dt ′ + dx ′ ⊗ dx ′} and
2
∂ ∂
2 2
∂ 2 ∂
− +
= f − + d’Alembert-Laplacian
∂ct ∂x ∂ct ′ ∂x ′
So the wave equation is invariant relative to Voigt’s transformation and thus preserved by
conformal transformations in two dimensions. Equation (125) was one of the reasons why I
introduced a second time coordinate and chose Cl4,2 instead of Cl4,1. A second important cause
is in my belief in the existence of a constant symmetry breaking giving rise to the emergence
of a proper time arrow contrasting an almost vanishing reverse arrow. The third reason that
motivates me to favour that algebra is in the process of deconvolution of antagonistic
advanced and reverse waves. This arose already in electroweak interaction. It is of even
greater importance in strong force events.
Sure my readers have realized the astonishing jump from the special unitary groups
SU(4) and SU(3) relevant in the Clifford algebras Cl3,1 and Cl4,1 to SU(8) and SU(7) in the
11
This rigor is quoted after Oziewicz [61] section 1.1
52 Bernd Schmeikal
Cl4,2 involving a de Sitter space adS6 ⊂ γλ(Cl4,2) for orthogonal transformations of vectors
and a Lie group L(6) for orthogonal transformations of the pure state manifolds.
It is fascinating to observe how far we can backtrack the meaning of bilinear exterior
forms such as f dt ∧ dx, dt ∧ dt’, x ∧ t and unitals e1 ∧ e4. The Lorentz scalar factor γv =
1/(1−v²/c²)½ was introduced by Oliver Heaviside in order to relate electric fields in mutual
relative motion. The same factor was introduced one year later to Heaviside in a letter by
George Francis FitzGerald [61, section 2.1] and independently by Hendrik Antoon Lorentz in
1893. In 1887 he used it to explain the results of the Michelson-Moreley experiment
disproving the existence of the aether drift [66]. Oziewicz [61] evaluated a re-derivation of
the Heaviside transformation by Hajra and Gosh (2005) [67]. “The re-derivation starts from
the following formulas for the constant relative velocity u [67, formulas (11)-(12)]
∂ ∂
dx = udt ⇒ x − ut = constant ⇒ =u
∂t ∂x
Therefore the coordinates x and t here are not independent variables, contrary to the
assumption made on page 64. The derivation of the Heaviside transformation […] seems to
need some more justification in differential geometry”.
On 11th November 1903, the Proceedings of the London Mathematical Society received
a paper written by E. T. Whittaker [68] in which he derived the electromagnetic field equa-
tions by means of two scalar potential functions instead of and collateral to the scalar and
vector potential function. He could not refer to any principle of relativity and did not use the
Lorentz transformation. Still he obtained two typically Lorentz invariant massless Klein
Gordon equations for - we would say today - longitudinal and time-like scalar field compo-
nents of the four potential, the second quantization of which straightforwardly leads us to lon-
gitudinal massless bosons, quanta of pure scalar energy [69], [70]. Then it was known that the
equations of motion of the dielectric displacement and the magnetic field strength had to be
connected with two orthogonal circular motions, the curls of the electric and magnetic fields.
These curls were needed in addition to time derivatives and linear translations of charge den-
sity. Starting off with the vector potential A and Stratton potential S Whittaker calculated two
scalar fields F and G capable to provide the ED equations of motion which, then, took the
form
∂ 2 F 1 ∂ 2G ∂ 2 F 1 ∂ 2G ∂2F 1 ∂2F
d1 = + , d2 = − , d3 = −
∂x∂z c ∂y∂t ∂y∂z c ∂x∂t ∂z 2 c 2 ∂t 2
1 ∂ 2 F ∂ 2G 1 ∂ 2 F ∂ 2G ∂ 2G ∂ 2G
h1 = − , h2 = − − , h3 = + (126)
c ∂y∂t ∂x∂z c ∂x∂t ∂y∂z ∂x 2 ∂y 2
which are equivalent to vector equations for d and h in traditional mathematical form:
Lie Group Guide to the Universe 53
1 1
d = curl curl f + curl gɺ , h = curl fɺ − curl curl g (127)
c c
Here he introduced vectors f and g parallel to the axis z with magnitudes F and respec-
tively G. The quantities d and h are the electric displacement vector and magnetic field.
Whittaker calculated explicit forms for the scalar functions of F and G. He ascertained that
those would have singularities at the points occupied by the electrons and found by differen-
tiation the Klein Gordon equations for massless boson fields away from singularities.
1 ∂2
□F = 0 □G = 0 with □= ∆+ (128)
c 2 ∂t 2
In the context of Whittaker’s gauge theory which is no other than the Maxwell-Heaviside
theory, the field equations for the scalar and vector potentials are recovered from the equa-
tions (126). We should be aware that Whittaker’s calculation is electrodynamic, yet bare of
the method of isometric Lorentz transformation.
From a geometric algebra perspective, going into details, we observe rotation as curls
alongside the plane unit areas e13 and e23 and a surprising asymmetry of the field equations in
the third vector components alongside e3 which cannot be removed. Contrasting nowadays
conviction, this approach leads to three types of field quanta, namely 1.) transverse photons
as well as 2.) longitudinal and 3.) timelike bosons. The latter are physically real, but show
little interaction with matter. However, interference of scalar bosons leads to particle creation.
Whittaker started classically but, apparently, ended up with relativistic equations of motion.
However, the equations are conformally invariant and therefore not necessarily bound to
reciprocity. This again indicates the importance of scale invariance.
A last question concerns the use of base units in Clifford algebras. It is true that the
physics in its essence is base-free and coordinate free. This is what Oziewicz emphasizes over
and over again. In the beginning of section 2 in [61] he states: “An invertible endomorphism
of a vector space, a transformation of a vector space, is said to be the ‘passive’
transformation, if the domain is the manifold of all basis of the vector space. ‘Passive’ means
the active action on the coordinate-free basis, or an action on the coordinate basis, but not on
individual vectors.” I wish to make this point clear at the end of this work.
We are using the unital base elements in order to construct the equivalence classes of
idempotent manifolds, Lie groups and similar objects. These mathematical elements are
important for both mathematics and physics. They form manifolds, but not any distinguished
basis or generators. For instance the 24 primitive idempotents which form lattices of
idempotents within Cl4,2 give rise to 24 equivalence classes of lattice-manifolds strictly
separate from each other. Each lattice contains 8 primitive idempotents or pure states which
are cleanly disaggregated throughout their manifolds. So we distinguish 192 primitive
idempotents in Cl4,2. At present I have no other means at hand to find those equivalence
classes than to use the standard basis. But may be some day we shall have better tools to solve
such problems of fundamental structure.
To see the beauty of those structures which give rise to the Lie groups similar to SU(3),
take a look at the 48 primitive idempotents in Cl4,1. The figure may be a late tribute to Harry
Lipkin. May be some of us find themselves suddenly motivated to study the SU(4) multiplets
54 Bernd Schmeikal
on the cover page of this wonderful unabridged Dover edition . . . a piece of fine art. I add to
it this one:
Figure 11. Bilateral orthogonality lattice with definite maximal element f1 and tetragonal structure of
pure states in the Clifford algebra Cl 4,1 .
The 48-primitive idempotent structure of the Cl4,1 is extremely rich. It has first been used
in a minimal spin gauge theory of quantumchromodynamics [48] to work out graded analoga
to the su(3).
Probably my readers have spied out in the meantime that I rarely use explicit equations of
motion. In [33] I designated the Dirac-Hestenes Equation in order to visualize the equivalence
classes of the surabale. In [56] I used the Dirac equation to demonstrate the meaning of
ternary logic projector fields. I wrote this in memory of Carl Friedrich von Weizsäcker. In the
abstract to this lecture I said that we are familar with the ℤ2 -grading of the Clifford algebra
and the double cover of orthogonal groups. With this we associate the projector equation and
decomposition of unity 1 = P1 + P0 according to spin decomposition or chirality. But we have
not yet comprehended the importance of the quaternary decomposition of unity 1 = P0 + P1 +
P2 + P3. A binary decomposition is characteristic for qed. It has first been used by John von
Neumann and interpreted by von Weizsäcker as »logic alternative«. Weyl has for sometime
pondered over the meaning of the Klein-4 group and the rays as compared with vectors. Then
he could not yet realize the importance of a quaternary decomposition of unity and the K4-
grading. The equation 1 = P0 + P1 + P2 + P3 characterizes qcd dynamics in quite general
algebras. What a binary decomposition is for qed, the quaternary is for qcd. In this paper the
algebraic foundations are given by what is denoted here as maximal ternary Cartan
decomposition in noncommutative algebras. The natural norm in a Cartan extension is not
derived from the Minkowski metric, but from the fact that the Clifford product becomes the
inner product while the exterior product vanishes.
The special Clifform L3,1 = sl(4, ℝ) together with the decomposition of L(2) = slCl(2,
ℝ)×soCl(3, ℝ) with a quasi relativistic factor suggests to consider the Iwasawa decomposition
Lie Group Guide to the Universe 55
as in [32] and the Iwasawa-Whittaker-Liouville wave functional for the groups SL(n) together
with the asymptotic Harish-Chadra function. But then, again we are led back to conformal
invariance but not Lorentz-invariance. For my understanding this is not irrelevant, but rather a
matter of fact and empirical probe. The Liouville type of wave equations in the SL(3) and
SL(n) framework has been thoroughly investigated by a group of russian scientists [71]. Their
paper provides all the means necessary to solve equations of motion and scattering problems.
They followed the lines of Liouville field theory, developed further the Whittaker model
which is partly responsible for the quantization of the Liouville field theory. They
investigated the wave functions of the open Toda model in group theoretical terms and
explained the simple relation to the Whittaker function for the quantum Lorentz group [72].
Briefly spoken, the solutions to our dynamic problems are still burrowed in the Liouville
quantum mechanics. They wait for their chance to deconvolute into cognitive spaces.
References
[1] (a) Zweig,G. An SU3 model for strong interaction symmetry and its breaking. CERN
Reports 8182/TH401, 1964a. (b) Zweig, G. An SU3 model for strong interaction
symmetry and its breaking. 2, CERN Reports 8419/TH412, 1964b.
[2] Gell-Mann, M. A Schematic model of baryons and mesons. Phys Lett 8, 1964, 214-215.
[3] (a) Ne’eman, Y. Derivation of strong interactions from a Gauge invariance. Nucl Phys
26, 1961, 222-229. (b) Ne’eman, Y. From the quarks to the cosmos - 70 years of
physics in Israel. Lecture given for the Israel National Academy of Sciences, 1998.
2001 (in Hebrew). (c) Ne’eman, Y. Matter particled: patterns, structure and dynamics -
selected research papers of Yuval Neeman. Imperial College Press, London 2006.
[4] (a) M.O. Katanaev, T. Klosch, W. Kummer. Global properties of warped solutions in
General Relativity. gr-qc/9807079v2, 1999. (b) Vilasi, G., Vitale, P. The so(2,1)
Symmetry in General Relativity. Preprint ESI 1110, Vienna 2001.
[5] Jaglom, I. M. Felix Klein and Sophus Lie. Evolution of the idea of symmetry in the
19th Century. Birkhäuser, Basel 1988.
[6] Cartan, É. La Théorie des Groupes Finis et Continus et L’Analysis Situs. Mémorial des
Sciences Mathématiques, vol. 42, Paris 1930. Oeuvres Complètes, vol. I, 1165-1225.
[7] (a) Weyl, H. Raum-Zeit-Materie. 4th ed. Heidelberg, 1920. (b) Weyl, H.
Gruppentheorie und Quantenmechanik. Leipzig 1931. (c) Weyl, H. The classical groups
- their invariants and representations. Princeton University Press 1939.
[8] (a) Clifford, W.K. The Common Sense of the Exact Sciences. Ed. K. Pearson, Dover,
New York 1955. (b) A. Micali, et al. (eds.) Clifford Algebras and their Applications in
Mathematical Physics. Dordrecht 1992.
[9] ‘Letter 02 to Bernd Schmeikal’ by Zbigniew Oziewicz, July 21, 2005, private folder.
[10] Dirac, P. The Quantum Theory of the Electron. Proceedings of the Royal Society, Vol.
117, 1928, .610 and Vol. 118, 351.
[11] Dirac, P. The Principles of Quantum Mechanics. Oxford 1930.
[12] Lounesto, P. Clifford Algebras and Spinors, Cambridge 2001.
[13] Juvet, G. Opérateurs de Dirac et équations de Maxwell. Commentarii mathematici
Helvetici, 2, 1930, 225-235.
56 Bernd Schmeikal
[25] Lipkin, H. J. From Sakata Model to Goldberg-Ne´eman Quarks and Nambu QCD –
Phenomenology and “Right” and “Wrong” experiments. [hep-ph/0701032v2], (31
January 2007).
[26] (a) Woronowicz, S. L., C*-algebras Generated by Unbounded Elements, Rev Math
Phys 7 (3), 1995, 481-521. (b) Woronowicz, S. L. Twisted SU(2) group: An example of
a non-commutative differential calculus. Publ RIMS, Kyoto Univ, 23, 1987, 117-181.
[27] Majid, S. Noncommutative Physics on Lie Algebras. In: R. Ablamowicz (ed.): Clifford
Algebras – Applications to Mathematics, Physics and Engineering. Birkhäuser, Boston-
Basel-Berlin, 2004, 491-518.
[28] Owczarek, R. M. Dirac Operator on Quantum Homogeneous Spaces and
Noncommutative Geometry. In: R. Ablamowicz (ed.): Clifford Algebras – Applications
to Mathematics, Physics and Engineering. Birkhäuser, Boston-Basel-Berlin, 2004, 519-
530.
[29] Cruz Guzmán, J. de J., Oziewicz, Z. Föhlicher-Nijenhuis Algebra and four Maxwell’s
Equations for Non-Inertial Observer. Bulletin de la Société de Sciences et des Lettres
de Lódź, LIII, Série : Recherches sur les Déformations, XXXIX, 2003, 107-140.
[30] Schmeikal, B. Algebra of matter. Advances in Applied Clifford Algebras 15, No 2,
2005, 271-290.
[31] Magnea, U. An Introduction to Symmetric Spaces. Department of Mathematics,
University of Torino, Italy, [cond-mat/0205288 v1] 14 May 2002 (19 August 2003).
[32] Schmeikal, B. Symmetric spaces of Matter and Real Fermion Manifolds. Advances in
Applied Clifford Algebras 16, No 1, 2006, 69-83.
[33] Schmeikal, B. The Surabale of Spacetime and its Algebra Signature Split. Advances in
Applied Clifford Algebras 17, No 1, 2007, 107-135.
[34] Sakata, S. On a Composite Model for the New Particles. Prog. Theor. Phys. 16, 1956,
686-688.
[35] Goshen, S. Lipkin, H. J. A Simple Independent-Particle System having Collective
Properties. Ann. Phys. (NewYork) 6, 1959, 301-309.
[36] Lipkin, H. J., Pairing and quadrupole forces in a two-dimensional soluble model. Nucl.
Phys. 26, 1961, 147-160.
[37] Ne’eman, Y., 1961, see [3a]
[38] Goldberg, H., Ne’eman, Y. Nuovo. Cim. 27, 1963, 1.
[39] Ne’eman, Y. The spectrum generating group program and the string. Found Phys. 18,
1988, 245-275.
[40] Fritzsch, H. Mesons, Quarks and Leptons. [hep-ph/0207279 v1] 23 Jul 2002, *To be
published in a commemorative volume at Bologna University in honor of N. Zichichi (4
September 2003).
[41] Zweig, G. private communication to Harald Fritzsch: quoted in [38].
[42] Connolly, P. L., et al. Existence and Properties of the φ - Meson. Phys. Rev. Lett. 10,
1963, 371.
[43] Greenberg, O. W. Spin and Unitary-Spin Independence. Phys. Rev. Lett. 13, 1964, 598-
602.
[44] Nambu, Y., Jona-Lasinio, G. A dynamical model of elementary particles based on an
analogy with superconductiviy. part 1, Phys. Rev. 122, 1961, 345, part 2, ibid., 124,
1961, 246.
58 Bernd Schmeikal
[45] Nambu, Y. A three triplet model with double SU(3) symmetry. Phys. Rev. 130, 1965, B
1006.
[46] Elliott, J. P. The Nuclear Shell Model and its Relation with Other Models. in: F.
Janouch, ed., Selected Topics in Nuclear Theory, IAEA, Vienna 1963.
[47] particle data group, website: http://pdg.lbl.gov/2006/reviews/kmmixrpp.
[48] Schmeikal, B. Minimal Spin Gauge Theory. Clifford Algebra and
Quantumchromodynamics. Advances in Applied Clifford Algebras 11, No 1, 2001, 63-
80.
[49] Siegel, W. Fields. December 1999 [hep-th/9912205] (19 August 2003).
[50] Cohen, R. S.; Ed.; Ernst Mach – Physicist and Philosopher. Dordrecht 1975.
[51] (a) Randall, L. Warped Passages. Unraveling the Mysteries of the Universe's Hidden
Dimensions. New York 2005. (b) Randall, L., Sundrum, R. Large Mass Hierarchy from
a Small Extra Dimension. R. Phys Rev Lett 83, 1999, 3370-3373 [hep-ph/9905221]. (c)
Randall, L., Sundrum, R. An Alternative to Compactification. R. Phys Rev Lett 83,
1999, 4690-4693 [hep-ph/9906064].
[52] Dvergsnes, E. The Randall-Sundrum Radion: Production Through Gluon Fusion, and
Two Photon Decay. Thesis, Department of Physics, University of Bergen, Norway
2000. http://hdl.handle.net/1956/1801 (4 September 2003).
[53] Higgs, P. W. Broken Symmetries and the Masses of Gauge Bosons. Phys Rev Lett 13,
1964, 508-509.
[54] Wallace, G. W. The Pattern of Reality. Advances in Applied Clifford Algebras 18, No 1,
2007, 115-133.
[55] Hestenes, D. Quantum Mechanics from Self-Interaction. Found Phys 15, No. 1,1983,
63-87.
[56] Schmeikal, B. Pregeometry of Extensions and Eigenfields. Prepared for the Conference
on Clifford Algebras and their Applications in Mathematical Physics in Brazil, May
2008, In memory of Carl Friedrich von Weizsäcker.
[57] Unnikrishnan, C. S. Precision measurement of the one-way speed of light and
implications to the theory of motion and relativity. Gravitation Group/FI-Lab, Tata
Institute of Fundamental Research, Mumbai, India [2006]; personal copy, available:
unni@tifr.res.in.
[58] Maclay, J., Hammer, J. George, M., Ilic, R. Leonard, Q., Clark, R. Measurement of
Repulsive Quantum Vacuum Forces. Published as: AIAA/ASME/SAE/ASEE 37th Joint
Propulsion Conference, Salt Lake City, July 2001, 1-9. AIAA-2001-3359.
[59] Banach, S., Tarski, A. Sur la décomposition des ensembles de points en parties
respectivement congruents. Fundamenta Mathematica 6, 1924, 244-277.
[60] Stewart, I. The ultimate jigsaw puzzle. in: New Scientist 13, April 1991, 30-33.
[61] Oziewicz, Z. Electric field and magnetic field in moving reference system. International
Symposium on Recent Advances in Mathematics and its Applications, Calcutta, India,
December 2006.
[62] Oziewicz, Z. What is categorical relativity? International Journal of Geometric
Methods in Modern Physics 4, (1), 2007, math. CT / 0608770
[63] Voigt, W. Über das Doppler’sche Prinzip. Nachrichten Ges Wiss Göttingen 41, 1887.
[64] (a)Heaviside, O., The electro-magnetic effects of a moving charge. The Electrician 22,
1888, 147-148. (b) Heaviside, O. Electrical Papers. Providence, R.I., AMS Chelsea
Publishing 2003.
Lie Group Guide to the Universe 59
[65] Einstein, A. Zur Elektrodynamik bewegter Körper. Annalen der Physik (Leipzig) 17,
1905, 891-921.
[66] (a) Lorentz, H. A. Zichtbare en onzichtbare bewegingen. Leiden 1901. (b) Lorentz, H.
A. Lectures on Theoretical Physics. (vol. I-III), Macmillan and Co, London 1931 (c)
Lorentz, H. A. The intensity of radiation and the motion of earth. Proceedings of the
Royal Netherlands Academy of Arts and Sciences 4, 1901-1902, 678-681.
[67] Hajra, S., Ghosh, A. Collapse of SRT 1: derivation of electrodynamics equations from
the Maxwell field equation. Galilean Electrodynamics 16, (4), 2005, 63-70.
[68] Whittaker, E. T. On an Expression of the Electromagnetic Field due to Electrons by
means of two scalar Potential Functions. Proceedings of the London Mathematical
Society 1, 1904, 367-372.
[69] AIAS, Authors. Representation of the Vacuum Electromagnetic Field in Terms of
Longitudinal and Time-like Potentials: Canonical Quantization. Journal of New Energy
4, No 2, 2000, 82-91.
[70] AIAS, Authors. An Experimental Test of the Existence of Whittaker’s g and f Fluxes in
the Vacuum. Journal of New Energy 4, No 2, 2000, 92-96.
[71] Gerasimov, A., Kharchev, S., Mershakov, A., Mironov, A., Morozov, A. Olshanetsky,
M. Liouville Type Models in Group Theory Framework I. Finite-Dimensional
Algebras. ITEP M4/TH-7/95, FIAN/TD-18/95, [hep-th/9601161 v1] (24 January 2003).
[72] Olshanetsky, M., Rogov, V. Liouville quantum mechanics on a lattice from geometry of
quantum Lorentz group. Journ. Phys. Math. Gen. 27, 1994, 4669-4683.
In: Lie Groups: New Research ISBN: 978-1-60692-389-4
Editor: Altos B. Canterra, pp. 61-88 © 2009 Nova Science Publishers, Inc.
Chapter 2
Jari Mäkinen∗
Tampere University of Technology, Department of Mechanics and Design,
FIN-33101 Tampere, FINLAND
Abstract
In this paper, we prove that incremental material rotation vectors belong to different
tangent spaces of the rotation manifold SO(3) at a different instant. Moreover, we show that
the material tangent space as the tangent space at unity is not a possible definition yielding
geometrically inconsistent results, although this kind of definition is widely adopted in applied
mechanics community. In addition, we show that the standard Newmark integration scheme
for incremental rotations neglects first order terms of rotation vector, not third order terms.
Finally, we show that the rotation interpolation of extracted nodal values on the rotation
manifold is not an objective interpolation under the observer transformation. This clarifies
controversy about the frame-indifference of geometrically exact beam formulations in their
finite element implementations.
1. Introduction
A finite rotation (See [2] [3]), is a vector quantity, or more precisely, the finite rotation
belongs to a tangent space on a manifold. This manifold is a Lie group of the special
orthogonal tensors SO (3) , also called the manifold of finite rotations or shortly the rotation
manifold. In general, Lie-groups are noncommutative groups, which are also differentiable
manifolds such that their differentiable structures are compatible with their group structure for
the definitions of general Lie-groups, (See details in [13]). As it is shown in [14], material
∗
E-mail address: jari.m.makinen@tut.fi; Fax: +358 3 3115 2107; Phone: +358 3 318 3851
62 Jari Mäkinen
incremental rotation vectors, material angular velocity vectors, and material angular
acceleration vectors belong to the different tangent spaces of the rotation manifold SO (3) at
a different instant. Hence, the direct application of the material incremental rotation vector,
with standard time integration methods, yields serious problems: adding quantities which
belong to the different tangent spaces. This approach, the standard Newmark integration
scheme for incremental rotations, is widely adopted and comes from [19] and is also the point
of departure for [9] and is recently used in [18]. In the following, we name this Newmark
scheme for incremental rotations as the simplified Newmark scheme.
There are two main approaches for the time-integration of finite rotation with a three-
parametric presentation. One approach, which is called Eulerian formulation [4], directly
applies an incremental rotation vector, angular velocity vector and angular acceleration vector
in the formulation. This approach suffers from problems cited above, if standard time-
stepping schemes are used. The modified Newmark time-stepping for the Eulerian
formulation scheme, which overcomes this difficulty, is given in [14].
Another approach, which is referred to as the updated Lagrangian formulation [4], applies
a current reference placement, which is updated incrementally. In this approach, updated
rotation vectors and their time derivatives belong to the same tangent space of rotation and do
not suffer problems, as found in the Eulerian formulation. It is shown in [14] that the
modified Newmark time-stepping scheme and the updated Lagrangian formulation with the
standard Newmark scheme are equivalent time stepping methods. Recently, in the paper [18]
it is shown that the simplified Newmark integration scheme neglects third order terms of
rotation vector. However, the proof suffers from the fact that the material incremental rotation
vector and its time derivatives cannot be integrated directly by the Newmark scheme. We
show that the simplified Newmark integration scheme for finite rotations neglects first order
terms of rotation vector, as opposed to third order terms [18].
Moreover, we could present the third approach, called total Lagrangian formulation [4]
and [16], where a reference placement is permanently the initial placement and a total rotation
vector and its time derivatives are used as unknown variables. Well-known singularity
problems at full-angle and its multiples can be bypassed by introducing a complement
rotation vector [16]. A rotation vector and its complement rotation vector are the
parameterization charts of the rotation manifold SO (3) . We could represent the rotation
manifold globally with these two parametrization charts. When a rotation angle exceeds
straight-angle, we accomplish the change of parametrization, giving a new rotation angle
smaller than a straight-angle. Thus, we get out of the singularity problems at full-angle.
Objectivity, or frame-invariance, is another issue that we are considering. We show that
the interpolation of the total rotation vector is an objective interpolation under the observer
transformation. Regardless, contrary results are also given in [6] where, with the aid of
counterexample, it is shown that the finite element discretization violates objectivity. This
counterexample suffers serious problems: extracted nodal vectors are interpolated, which way
is not generally allowed and is never used in the finite element discretization of the total and
updated Lagrangian formulations. In connection with this, we show that the rotation
interpolation of extracted nodal values on the rotation manifold SO (3) is not an objective
interpolation under the observer transformation.
This paper is organized as follows: We give some necessary definitions and results about
manifolds and their tangent spaces in Chapter 2. In Chapter 3, we give generalized definitions
Rotation Manifold SO(3) and Its Tangential Vectors 63
for the rotation manifold SO (3) ; and we prove that the material incremental rotation vector
belongs to different tangent spaces of the rotation manifold SO (3) at a different instant. In
Chapter 4, definitions for angular velocity and acceleration vectors are given. In Chapter 5,
Euler equations for rotation motion, which are differential equations on the tangent bundle,
are given in the material and spatial representations. In Chapter 6, we show that the simplified
Newmark integration scheme for incremental rotations neglects first order terms of rotation
vector. Finally, we show that the rotation interpolation on the rotation manifold is an
objective interpolation under the observer transformation.
m a n ifo ld M Ì E n
j 1 bU g Ç
1
j 2 bU g2
j 2
j 1
c h a n g e o f
p a ra m e triz a tio n
U 1 j 2- 1 o j 1 U 2
p a ra m e te r s p a c e E d
ϕ (η ) − ϕ (0)
t = lim , where ϕ (0) = x, ϕ (η ) ∈ M .
η →0 η
The tangent vector t belongs to a tangent space of the manifold, namely t ∈ Tx M , (See
Figure 2). The tangent (vector) space Tx M is the set of the tangent vectors at the base point
x ∈M .
Definition 2.4. (tangent bundle) A tangent bundle T M is defined as a union of the
tangent spaces on the manifold M at its every point
T M := ∪ ( x, TxM ) .
x∈M
The dimension of the tangent bundle is twice the dimension of the manifold M . The pair
of state vectors, the placement x ∈ M and its velocity vector v ∈ Tx M , belongs to the
tangent bundle, ( x, v ) ∈ T M .
Rotation Manifold SO(3) and Its Tangential Vectors 65
T x M
t x
j (t)
Figure 2. Tangent vector t and its tangent space Tx M on the manifold M at the point x.
We note that usually a chart mapping is defined by an inverse mapping from an open set
of a manifold into a parameter space. We have defined a chart mapping differently since we
could use this terminology when constraint equations are parametrized. A vector space, where
n
a manifold is embedded, is called an embedding space; the Euclidean space E in Figure 1.
r 1
y r 0
E
3
p 1 p 0
r 1 n p s in y
0
y
r 0 = - n bn p 0 g r 0
r 1 = r 0 c o s y n p 0 s in y
Figure 3. A rotation about n -axis where p0 is the original vector and p1 is the rotated vector.
The original projection vector r0 and the rotated projection vector r1 in the rotation plane
are given in Figure 3, where × denotes the cross product on E . Now, the rotated vector p1
3
can be expressed by
p1 = p0 − r0 + r1
= p0 + (1 − cosψ ) n × ( n × p 0 ) + n × p0 sinψ (2)
= Rp0 p1 , p0 , n, r0 , r1 ∈ E , ψ ∈ R + 3
Now, the rotation operator R can be written in terms of the rotation vector that is
defined by
Ψ := ψ n, n ∈ E3 ,ψ ∈ R + . (3)
Euler’s theorem states that any rotation tensor R is a rotation through an angle ψ about
an axis n (unit vector). Thus, we have the rotation vector Ψ := ϕ n . This yields the
expression of the rotation operator
sinψ ɶ 1 − cosψ ɶ 2
R := I + Ψ+ Ψ , ψ = Ψ (4)
ψ 2
ψ
ɶ h = Ψ × h,
Ψ ∀h ∈ E 3 , (5)
ɶ := Ψ × .
or more formally Ψ
It is well known that the rotation operator and the rotation vector are related by the
exponential mapping, (See [7]; p. 70]).
R = exp ( Ψ
ɶ ) := I + Ψ
ɶ +1Ψ
2!
ɶ2+1Ψ
3!
ɶ3+ 1Ψ
4!
ɶ4… (6)
d exp (ηΨ
ɶ)
ɶ
=Ψ (7)
dη
η=0
point of the rotation manifold. It is clear that the base point is the identity I ∈ SO (3) since
exp(ηΨ
ɶ ) with η = 0 is equal to the identity I , (See Def. 2.3.).
In general, the skew-symmetric tensors form Lie-algebra with Lie-brackets defined as
ɶ , Bɶ := AB
A ɶ ɶ − BA
ɶ ɶ, ɶ , Bɶ ∈ so(3)
∀A (8)
so(3) := A{
ɶ : E3 → E3 linear A
ɶ T = −A
ɶ } (9)
Note that the elements of Lie-algebra so(3) do not need to be infinitesimal quantities.
Instead, they may form a vector space at the identity of the rotation group. It can be verified
that the set so(3) 9 meets all the Lie-algebra properties, (See [5] or [13]).
by two different, though equivalent ways: by the material description, and by the spatial
description.
Definition 3.1. We define the material description of a compound rotation by the left
translation map Lef tR : SO (3) → SO (3) as
mat
Lef tR R inc := RR inc
mat
= R exp ( Θ
ɶ) mat
R inc , R ∈ SO (3) ,
mat
where R inc is an incremental material rotation operator, and Θ is an incremental material
rotation vector with respect to the base point R ∈ SO (3) . This description is called material
since the incremental rotation operator acts on a material vector space.
Definition 3.2. We define the spatial description of a compound rotation by the right
translation map RightR : SO (3) → SO (3) as
rotation vector with respect to the base point R ∈ SO (3) . This description is called spatial
since the incremental rotation operator acts on a spatial vector space.
We use majuscules for material vectors and minuscules for spatial vectors. Both the
material and spatial rotation incremental tensors, and the rotation vectors are related by
mat
R inc = R T R spat
inc R , θɶ = RΘ
ɶ R T and θ = RΘ , (10)
where the first relation is called inner automorphism that is an isomorphism onto itself. The
ɶ = RΘ
second relation is a Lie algebra so(3) adjoint transformation Ad R Θ ɶ R , and the last
T
relation is another Lie algebra adjoint transformation in the Euclidean space with the vector
cross product as the Lie algebra (E , ⋅ × ⋅ ) , (See e.g. [13]).
3
ϕ(η ) := R exp(η Θ
ɶ ), ϕ (η = 0) = R ,
d ϕ(η ) ɶ .
(11)
= RΘ
dη η = 0
ɶ and t := RΘ
Now, if we have two tangent tensors t1 := RΘ ɶ at the base point R , we
1 2 2
may add them together, giving a linear combination
α t1 + β t 2 = α RΘ ɶ = R (αΘ
ɶ + β RΘ ɶ ), α, β ∈R ,
ɶ + βΘ (12)
2 1 2 1
ɶ and Θ
as expected. Note that the incremental rotation tensors Θ ɶ can be added up when we
1 2
are at the same base point of the rotation manifold. This yields a definition for material
tangent space of rotation.
Definition 3.3. (material tangent space) Differentiating the material expression of the
compound rotation R exp(ηΘ
ɶ ) with respect to the parameter η and setting η = 0 yields
the material tangent space at the base point R ∈ SO (3) . This material tangent space on the
rotation manifold SO (3) , at any base point R , is defined as
T SO (3):= Θ
mat R R {
ɶ := (R , Θ
ɶ ) with RΘ
ɶ ; R∈ SO (3), Θ
ɶ ∈ so(3) , }
ɶ ∈ T SO (3) is a skew-symmetric
where an element of the material tangent space Θ R mat R
ɶ ∈ so(3) .
tensor, i.e. Θ R
ɶ ) , the pair of the rotation operator R , and the skew-symmetric
The notation ( R , Θ
ɶ , represents the material skew-symmetric tensor at the base point R ∈ SO (3) , (See
tensor Θ
ɶ is a skew-symmetric tensor, or a tangent tensor,
Figure 4). Hence, we may express that Θ R
at the base point R on the rotation manifold SO (3) . For simplicity, we could omit the base
ɶ ∈ T SO (3) if there is no danger of confusion. Especially, the
point R by denoting Θ mat R
ɶ and Θ
material incremental rotation tensors Θ ɶ in Eqn. (12), which belong to the same
1 2
T SO (3) := {Θ
mat R
ɶ := RΘ
R
ɶ ∈ so(3)} .
ɶ for any Θ (13)
Basically, Def. 3.3 and (13) are similar in that (13) the rotation tensor R could be
ɶ ∈ so(3) . However, we
regarded as the base point of the material skew-symmetric tensor Θ
70 Jari Mäkinen
ɶ ∈ T SO (3), Θ
Θ ɶ ∈ T SO (3), Θ
ɶ ∈ T SO (3),
1 mat I 2 mat R1 3 mat R12
(15)
ɶ ), R := exp( Θ
R1 := exp( Θ ɶ ) exp( Θ
ɶ ).
1 12 1 2
ɶ ,Θ
This means that the tensors Θ ɶ and Θ
ɶ belong to the different tangent spaces and
1 2 3
thus must not be added up.
Correspondingly, we may write for the spatial tangent tensor of the rotation manifold
Definition 3.4. The spatial tangent space on the rotation manifold SO (3) at any base
point R is defined
{
T SO (3) := θɶ R := ( R , θɶ ) with θɶ R; R ∈ SO (3), θɶ ∈ so(3) ,
spat R }
where an element of the material tangent space θɶ R ∈ spatTR SO (3) is a skew-symmetric
T IS O (3 )
e x p y~
sp a t
y~
b g
T IS O (3 )
e x p Y d~ i ~
m a t I
Y sp a t T R S O (3 )
I ~ R
q R
m a t T R S O (3 )
R S O (3 )
~
Q R
S O (3 )
Figure 4. Geometric representation of the material tangent space (on the left) and the spatial tangent
space (on the right) on the rotation manifold SO (3) .
Rotation operators, the elements of the Lie group SO (3) , are defined as linear operators
R ∈ SO (3) . Equations (10b,c) give another interpretation for a rotation operator, it is an
adjoint transformation between the material and spatial tangent spaces. In addition, a rotation
motion induces the rotation operator, since the rotation operator maps the material place
vector X ∈ B0 into the spatial place vector x ∈ B by the equation x (t ) = R (t )X , i.e.
R ∈ L (B0 , B ) . More generally, a rotation operator transforms material vectors into spatial
vectors, that is R ∈ L (TX B0 , TxB ) , (See Figure 5 and [22]).
R R
d R q R m a tT R sp a t T R
R T
R T
T T T T T T
I I
I
y I m a t T I sp a t T I
Figure 5. Commutative diagram of variations of material and spatial rotation vectors on the rotation
manifold (on the left), and corresponding vector spaces (on the right).
exp ( Ψ ɶ ) = exp ( Ψ
ɶ +η δΨ ɶ ) exp (η δ Θ
ɶ ),
R (20)
associated rotation vector for the skew-symmetric tensor Ψ ɶ is called the total material
rotation vector whose base point is the identity.
Taking the derivatives of (20) with respect to the parameter η at η = 0 gives, (See e.g.
[10])
δ ΘR = T ⋅ δ Ψ ,
sinψ 1 − cosψ ɶ ψ − sinψ
T := I− Ψ+ Ψ ⊗ Ψ, (21)
ψ ψ2 ψ3
ψ := Ψ , ɶ ),
R = exp( Ψ lim T( Ψ ) = I ,
Ψ →0
verification that the virtual rotation vector δ Θ R and the virtual total rotation vector δ Ψ
belong to different vector spaces on the rotation manifold. This is because the tangential
transformation T is equal to the identity only at Ψ = 0 . Note that the transformation T has
an effect on the base points, changing the base point I into R .
Definition 3.5. (material vector space) For convenience, we define a material vector
space on the rotation manifold at any base point R as
{ ɶ ∈ T SO (3) ,
T := Θ R ∈ E3 Θ
mat R R mat R }
Rotation Manifold SO(3) and Its Tangential Vectors 73
where an element of the material vector space is Θ R ∈ matTR . The space is an affine space
with the rotation vector Ψ as a base point and the incremental rotation vector Θ as a
tangent vector, then T : matTI → T , (See Figure 5). Def. 3.5 gives a practical notation for
mat R
sorting rotation vectors in different tangent spaces.
Correspondingly, we could determine the spatial tangential transformation, yielding
δ θR = T T ⋅ δ ψ ,
T = T( ψ ), (22)
ψ := ψ ( = Ψ ),
where T : spatTI →
T
T is the same linear operator as in the material form (21), (See
spat R
Figure 5).
Definition 3.6. (spatial vector space) We define a spatial vector space on the rotation
manifold at any point R as
{
T := θR ∈ E3 θɶ R ∈ spatTR SO (3) .
spat R } (23)
Traditionally it has been assumed that the material incremental rotation vector belongs to the
same tangent space at any point in time, e.g. [19], [9] and [8]. This assumption has not been
proven anywhere and it cannot be proven because it is false. We give three separate proofs
against this assumption. We note that this is not a matter of definition because incremental
rotation vectors have a strict geometric meaning. Hence, the definitions of tangent spaces 3.3
and 3.4. have to be geometrically consistent with Def. 2.3.
Proof I: The material incremental rotation vector Θ or more precisely its skew-
symmetric counterpart, belongs to the tangent space of the rotation manifold SO (3) . If we
assume that the skew-symmetric rotation tensor Θ ɶ belongs to the same tangent space at any
point, then we have a manifold whose tangent spaces are identical at any point. Manifolds
whose tangent spaces are identical at any point are flat, but this is a contradiction since the
rotation manifold SO (3) is a non-flat manifold, i.e.
Proof II: This proof has given by the author in the paper [14]. The Newmark scheme
reads for the total rotation vector
74 Jari Mäkinen
Ψ ɺ ( i ) + γ ∆Ψ ( i ) ,
ɺ ( i +1) = Ψ (25)
n n
hβ n
ɺɺ ( i +1) = Ψ
Ψ ɺɺ ( i ) + 1 ∆Ψ ( i ) .
h β
n n 2 n
Substituting (25b,c) for the material angular velocity and angular acceleration vectors,
Eqns. (35a) and (36a), we have the modified time-stepping scheme:
We note that the modified Newmark scheme (26) is not the original Newmark scheme
because of the appearance of the tangential transformation T and its time derivative. Hence,
we cannot use the original Newmark scheme for the material quantities, as it is the well-
known fact for the spatial case. Therefore, an assumption that we could use the original
Newmark scheme for material quantities yields a contradiction, because an incremental
rotation vector and its time derivatives belong to the same tangent space at any point in time.
ɶ + ηδ Ψ
exp( Ψ ɶ ) exp(ηδΘ
ɶ ) = exp( Ψ ɶ ), (27)
we find out that Ψ , δ Ψ belong to the same vector space, namely T . This is clear
mat I
because the total rotation vector Ψ is a parametrization of the rotation manifold SO (3) and
the total rotation vector lives in that parametrization space, (See Figure 1 and Figure 6).
Additionally, we have a relation between δ Ψ and δ Θ that reads δ Θ = TδΨ , where
T( Ψ ) is the tangential transformation Eqn. (21). Since the tangential transformation T( Ψ )
depends on the total rotation vector, the vector δ Θ also depends on it. Hence, δ Θ depends
on the rotation operator R = exp( Ψ ɶ ) that is a base point of the rotation manifold SO (3) .
However, δ Ψ always occupies in the same fixed vector space T and thus δ Θ cannot
mat I
belong to the same vector space, unless at a specific point when T( Ψ ) is equal to the
identity I at Ψ = 0 .
According to our notation, the vector δ Θ or more precisely δ Θ R , which is the same
vector, belongs to the vector space of rotation T , which depends on the rotation operator
mat R
Rotation Manifold SO(3) and Its Tangential Vectors 75
ɶ ) . On the other hand, the vector δ Ψ belongs to the vector space of rotation
R = exp( Ψ
T , which is a fixed vector space, (See Figure 5).
mat I
ro ta tio n m a n ifo ld S O (3 )
p a ra m e triz a tio n
p a ra m e triz a tio n ~ C
~ m a p p in g e x p ( Y )
m a p p in g e x p ( Y )
2 p c h a n g e o f 2 p
p a ra m e triz a tio n
p p
Figure 6. The change of parametrization in the parameter space E3 for the canonical representation of
rotation manifold.
ɶ and δ Θ
The skew-symmetric tensors Ψ ɶ do not belong to the same tangent space of
R
ɶ ) exp( Θ
rotation as it can be verified that exp( Ψ ɶ ) ≠ exp( Ψ
ɶ +Θ
ɶ ) , generally.
Let a rotation vector Ψ with a rotation angle larger than zero and less than full-angle, i.e.
0 < ψ < 2 π , thus its complement rotation vector Ψ C is defined
Ψ C := Ψ − 2π Ψ , ψ := Ψ . (28)
ψ
After substituting the complement rotation vector into the rotation operator (4), we notice
that the rotation vector and its complement represent the same rotation operator, i.e.
R ( Ψ C ) = R ( Ψ ) . Def. (28) is a change of parametrization in the parameter space E 3 , (See
Figure 1 and Figure 6). This change of parametrization is a continuously differentiable
mapping on the open domain 0 < ψ < 2 π , giving a smooth construction of the rotation
manifold SO (3) at this domain. Note that the complement of a complement rotation vector
76 Jari Mäkinen
Ψɺ C = BΨɺ ∈matTIC ,
(29)
ɺɺ C = BΨ
Ψ ɺɺ + B
ɺΨɺ ∈ T C,
mat I
where the symmetric kinematic operator, defined by B := D Ψ Ψ , and its time derivative are
C
B = (1 − 2π ) I + 2π n ⊗ n ,
ψ ψ
(30)
ɺ = 2π ( n ⋅ Ψ
B ɺ )I + ( Ψ ɺ ) − 3( n ⋅ Ψ
ɺ ⊗n+n⊗Ψ ɺ )n ⊗ n ,
ψ2
Definition 4.1. The material angular velocity (skew-symmetric) tensor is defined with the
aid of rotation operator R ∈ SO (3) and its time derivative by
ɶ := R T R
Ω ɺ,
R
Rotation Manifold SO(3) and Its Tangential Vectors 77
where the dot denotes the time derivative. (See justification in [13; Ch. 8.6 and 15.2]).
The rotation tensor can be viewed as a mapping, a push-forward of a material vector,
R : matTR → spatTR between the material and spatial vector spaces, (See [22] and [12]). Then
the material angular tensor is a mapping Ω ɶ : T → T . Thus, the material angular
R mat R mat R
velocity tensor is indeed a true material tensor. The skew-symmetry can be obtained by taking
a derivative for the equation R T R = I .
If the rotation operator is expressed with the aid of exponential mapping by
ɶ (t ) + O(Θ
R new = R ( I + Θ ɶ 2 (t ))) , where the fixed rotation R is superimposed by an
R R
infinitesimal rotation I + Θɶ (t ) plus higher order terms, then substituting this into Def.
R
4.1. yields
ɶ = RT R ɺ ɶ ɶ2 ɶɺ ɶ
Ω new new = ( I − Θ R ( t ) + O( Θ R ( t )) )R R ( Θ R (t ) + O( Θ R (t )))
T T
R
(31)
ɺɶ (t ) + O( ɶ (t )),
=Θ R ΘR
ɶ → 0ɶ
after the limit process Θ R
ɺɶ
ɶ =Θ ɺ
ΩR R ⇔ ΩR = Θ R . (32)
This states that the angular velocity vector is the time derivative of the incremental
ɺ , Ω ∈ T , which is the
rotation vector Θ R , moreover, (if the base point is omitted) Θ, Θ mat R
material rotation vector space on the rotation manifold. The result in (32) is often given as a
definition for the angular velocity vector in elementary text books.
Similar expression and derivation can be accomplished for the spatial angular velocity
tensor and vector, yielding
ωɶ R := RRɺ T,
(33)
ɶ R = θɺɶ R ⇔ ωR = θɺ R ,
ω
ɺ and the
where the spatial incremental rotation vector θR , its time derivative vector θ R
spatial angular velocity vector ωR belong to the same spatial vector space on the manifold,
θ, θɺ , ω ∈ spatTR ; the base point R is omitted. The material and spatial quantities have
connections:
ω ɶ RT ,
ɶ R = RΩ ωR = RΩR , (34)
R
such like for incremental rotation vectors (10), (See Figure 5).
78 Jari Mäkinen
Definition 4.2. The material and spatial angular acceleration tensor and the corresponding
vector are defined as the time derivative of the angular velocity terms, giving
Α ɶɺ ,
ɶ := Ω ɶ ∈ T SO (3),
Α
R R R mat R
Α =Ω ɺ , Α R ∈ matTR ,
R R
ɶɺ R ,
αɶ R := ω αɶ R ∈ spatTR SO (3),
αR = ω
ɺ R, α R ∈ spatTR ,
where Α R is the material angular velocity vector, and α R is the spatial angular velocity
vector at the base point R .
Note that the incremental material rotation vector Θ R , the material angular velocity
vector ΩR and the material angular acceleration vector Α R (majuscule of alpha-letter)
belong to the same material vector space on the rotation manifold, i.e. Θ R , ΩR , Α R ∈ matTR
with the base point R = exp( ΨI ) . At separate time moments, these vectors, however,
occupy different vector spaces because the rotation operator depends on time, namely
R = R (t ) . The base point is moving as time travels. Vector quantities of this kind may be
called spin vectors. Spin vectors are rather tricky in the numerical sense as they always
occupy a distinct vector space on a manifold. Correspondingly, the spatial spin vectors are
θR , ωR , α R ∈ spatTR .
Angular velocity vectors and the time derivative of total rotation vectors are related by,
(See [7,] p. 70])
ɺ , where Ω ∈ T , Ψ
ΩR = T( ΨI ) ⋅ Ψ ɺ , Ψ ∈ T for material
I R mat R I I mat I
(35)
α R = TT (ψ I ) ⋅ ψɺ I , where ωR ∈ spatTR , ψɺ I , ψ I ∈ spatTI for spatial
where the tangential transformation depends on the total rotation vector, and the rotation
ɶ ) = exp(ψɶ ) . Similar expression for the angular acceleration vector
operator is R = exp( ΨI I
can be obtained by differentiating the above formulas, giving
ɺɺ + Tɺ ⋅ Ψ
ΑR = T ⋅ Ψ ɺ , where Α ∈ T , Ψ , Ψ ɺ ,Ψ ɺɺ ∈ T for material
I I R mat R I I I mat I
(36)
αR = T ⋅ ψ
T ɺ
ɺɺ I +T ⋅ ψɺ I , where α R ∈ spatTR , ψ I , ψɺ I , ψ
T
ɺɺ I ∈ spatTI for spatial
where the tangential transformation depends on the total rotation vector; and the rotation
operator is R = exp( Ψ ɶ ) = exp(ψ ) . Note that the tangential transformations
ɺ T ∈ L ( T , T ) operate with different base points.
T, Tɺ ∈ L ( matTI , matTR ) and T T , T spat I spat R
ɺ , Ψ) = c (Ψ ⋅ Ψ
ɺ ) I − c (Ψ ⋅Ψ
ɺ )Ψ
ɶ + c (Ψ ⋅Ψ
ɺ ) Ψ⊗Ψ + c Ψ
5 ( ⊗Ψ + Ψ⊗Ψ )
Tɺ ( Ψ ɺɶ + c Ψ ɺ ɺ , (37)
1 2 3 4
ɺ is
The limit value of the tensor T
ɺ (Ψ
lim T ɶɺ .
ɺ , Ψ) = − 1 Ψ (39)
→0 2
5. Euler’s Equations
In this Chapter, Euler equations for rotation motion, which are differential equations on the
tangent bundle, are given in the material and spatial representations. Lets consider virtual
work for a body V that can be given in a form
∫ δ x ⋅ ( f − ρ ɺɺx ) dV = 0,
V
∀δ x ∈ Tx0 M , (40)
where f is the external body force field, ρ is the density, ɺxɺ is the acceleration vector field,
and δ x is the virtual displacement field. The holonomic constraints g – such like constraints
arisen from rigid body assumptions – form a constraint manifold M into the placement
space. The constraint manifold can be defined by
{
M := x ∈ X g ( x ) = 0 , } (41)
{
Tx0 M := δ x ∈ X D xg ( x = x 0 ) ⋅ δ x = 0 . } (42)
The virtual displacement field δ x therefore belongs to a tangent space of the constraint
manifold M . It should be noted that the virtual displacement could be any size, infinitesimal
or finite.
A parametrization that parametrizes general spherical motion can be given in a form
80 Jari Mäkinen
x = RX , (43)
where R a rotation operator and X is initial placement vector field. Now the
parametrization (43) satisfies the rigid body constraint equations g ( RX ) = 0 .
If we substitute the parametrization (43), whose variation and time derivative are
δ x = Rδ Θ ɶδΘ ,
ɶ X = − RX
R R
(44)
ɶ X + RΩ
x = RΑ ɶ 2 X = −RX ɶ X
ɶ Α − RΩ
R ΩR ,
ɺɺ ɶ
R R R
δ Θ R ⋅ ( M R − JΑ R − Ω
ɶ JΩ ) = 0
R R (45)
where we have used notation for the inertial tensor J and for the external force vector M R
J := ∫ ρ X
ɶ TX
ɶ dV , M R := ∫ XR
ɶ T f dV (46)
V V
Equation (45) represents the material form of the virtual work principle for Euler’s
differential equation under a rotational motion. We also notice that Euler’s equation is defined
on the tangent space of the rotation manifold, namely T
mat R , since the virtual incremental
rotation vector δ Θ R ∈matTR belongs to this tangent space.
While the rotation operator R = R ( t ) – a base point of the tangent space T –
mat R
depends on time, also the tangent space varies. Traditional time-stepping methods, like the
Newmark scheme, have been designed for ordinary differential equations in linear spaces.
Therefore, these integration schemes are not suited for Euler’s equation (45), which is a
differential equation on the non-flat manifold SO (3) . However, the Newmark time-stepping
scheme can be consistently modified for the rotation manifold as given in [14]. This modified
Newmark scheme is equivalent with the different approach developed by [4] where the
updated total rotation vector Ψinc ∈matTR ref is used for an unknown vector. In this
formulation, the base point R ref is constant during the time integration procedure and the
base point is updated when a new solution is obtained. This formulation is called an updated
Lagrangian formulation.
Euler’s equation in the spatial representation can be obtained by substituting
parametrization (43) for the virtual work principle (40). The virtual displacement and time
derivative in the spatial representation are
δ x = δ θɶ R RX = −xɶ δ θR ,
(47)
x = αɶ R RX + ω
ɺɺ ɶ R2 RX = − xɶ α R − ω
ɶ R xɶ ωR .
Rotation Manifold SO(3) and Its Tangential Vectors 81
Thus, we have
δ θ R ⋅ ( m R − jα R − ω
ɶ R jω R ) = 0 , (48)
where we have denoted for the spatial inertial tensor j and for the spatial external moment
mR
j:= ∫ ρ xɶ T xɶ dV , mR := ∫ xf
ɶ dV (49)
V V
Material and spatial quantities have relations (10) and (34). Thus, the principle (48) can
be rewritten
δ θ R ⋅ R ( M R − JΑ R − Ω
ɶ JΩ ) = 0
R R (50)
where the terms in the brackets correspond to Euler’s equation in the material representation.
On the other hand, when we linearize the spatial quantities, we have to use Lie derivatives
such as
R D ( R T ) , (51)
where denotes the place for a spatial vector. Applying this Lie derivative to the principle of
virtual work (48) and especially to the formula (50), we find out that derivative is
accomplished to material quantities. This is because the derivative of an objective material
quantity is always an objective quantity. From this reality and identifying ψ = Ψ it follows
that spatial tangent tensors receive the same form as the corresponding material quantities.
Fully consistent tangential tensors are found in [15].
iteration step (i). The Newmark iteration scheme for the updated total rotation vector reads,
(See [14])
( i +1)
Ψinc n = Ψinc n + ∆Ψ
(i ) (i )
∈matTR ref n
Ψ ɺ ( i ) + γ ∆Ψ ( i )
ɺ ( i +1) = Ψ (52)
inc n inc n
hβ
ɺɺ ( i +1) = Ψ
Ψ ɺɺ ( i ) + 1 ∆Ψ ( i )
h β
inc n inc n 2
( ) n
2
Ψinc h 1 − 2β Α
n = hΩn +
(0)
2
ɺ (0) = Ω + h (1 − γ ) Α
Ψ (53)
inc n n n
ɺɺ
Ψ inc n = 0
(0)
where the angular velocity and acceleration vectors Ωn , Α n ∈ matTn are obtained from the
previously converged solution (or they are initial values). At the next iteration step, we have,
after elimination of ∆Ψ
Ψ (0) , for the velocity equation
ɺ (1) = Ω + h (1 − γ ) Α + γ hΨ
Ψ ɺɺ (1) (54)
inc n n n inc n
Next we assume that the iteration converges at the first iteration step. This has to be
assumed since the modified Newmark scheme (26) is nonlinear because of the tangential
( i +1)
tensor and its time derivative depend on the iterated solution Ψinc n . Now, multiplying (54)
by the tangential transformation Tn = T( Ψinc n ) and using the formulas (35a) and (36a), we
(1) (1)
have
(1)
The tangential transformation Tn and its time derivative can be expanded by a serial
development
T = I − 12 Ψɶ + O( Ψ 2 )
( )
(56)
ɺ =−1Ψ
T ɶɺ + 1 ΨΨ
ɶɺ ɶ + ΨΨ
ɶ ɶɺ + O( Ψ 2 )
2 6
where O( Ψ ) indicates the second and higher order terms. With this serial development,
2
ɶ (1) ( Ω + h (1 − γ ) Α ) +
Ωn +1 = Ωn + h (1 − γ ) Α n + γ hΑ n +1 − 12 Ψ inc n n n
(57)
− γ h 16 Ψ
ɺ (1)
ɶ Ψ ɶ ɺ
inc n inc n Ψinc n + O( Ψinc n )
(1) (1) (1) 2
Note that the simplified Newmark scheme reads for angular velocity Ωn+1
Ωn +1 = Ωn + h (1 − γ ) Α n + γ h Α n +1 (58)
Hence comparing Eqns. (57) and (58), we find out that the simplified Newmark scheme
for finite rotations neglects terms of the first order of rotation vector. It is interesting to notice
that at the plane rotation (2D rotation) the first and higher order terms of rotation vector
vanish, as expected. Similar results can be obtained for acceleration equation, yielding
Α n+1 = 1
βh ( 1
h Ψinc
(1)
n − Ωn ) − 2 β Α n + O( Ψinc n ) .
1− 2 β (1)
(59)
Hence, the simplified Newmark scheme for accelerations also neglects the first order
terms of rotation vector.
Generally, when a convergent solution is achieved at the iteration step ( m ) , the rotation
tensor is computed by
ɶ (m) ) .
R n +1 = R n exp( Ψ (60)
inc n
The updated total rotation vector Ψinc n is small for small time steps h but not negligible.
( m)
R + = QR ,
(61)
x c+ = Q ( x c + c ) ,
where the orthogonal operator Q ∈ SO (3) corresponds to the rigid body rotation and the
vector c ∈ E corresponds to the rigid body translation, respectively. Note that the rotation
3
84 Jari Mäkinen
have the following component values [6] with respect to the global frame {O , e1 , e 2 , e 3}
1 -0.4 0.2
[Ψ 1i ] = -0.5 , [Ψ 2i ] = 0.7 , Ψ Qi = 1.2 ,
(62)
0.25 0.1 -0.5
hence e.g. Ψ1 = Ψ 1i e i with conventional summation. Linear interpolation functions for the
nodal rotation vectors Ψ1 and Ψ2 read
s s
N1 ( s ) = 1 − , N 2 ( s) = , s ∈ [0, L] . (63)
L L
ɶ extr ) = Q exp( Ψ
exp( Ψ ɶ ), ɶ extr ) = Q exp( Ψ
exp( Ψ ɶ ). (64)
1 1 2 2
The nodal rotation vector Ψi ∈ matTI can be obtained by extracting it from the rotation
extr
operator via Spurrier’s algorithm. We note that the original rotation vectors Ψ1 and Ψ2 in
the observer transformation (64) occupy the tangential vector space T Although we may
mat Q .
ɶ extr ) ≠ Q exp( N ( s ) Ψ
exp( N i ( s ) Ψ ɶ ), ∀s ∈ [0, L ], Q ∈ SO (3) . (65)
i i i
This arises from the fact that the rotation manifold SO (3) has a curved character.
Indeed, a linear vector valued function in the tangential vector space mat QT is not linear in the
different tangential vector space T . Result (65) indicates that the linear interpolation does
mat I
not preserve under the observer transformation as shown in Figure 7, settings after (62).
Rotation Manifold SO(3) and Its Tangential Vectors 85
0.5
−0.5
−1
0 0.2 0.4 0.6 0.8 1
Length parameter s/L
1.9
Norm of rotation vector
1.8
1.7
1.6
1.5
1.4
1.3
1.2
0 0.2 0.4 0.6 0.8 1
Length parameter s/L
Figure 7. The components of the rotation vectors Ψ+(s) and Ψextr(s) interpolated through the length of
beam (left), and the norm ||Ψ+(s)|| and ||Ψextr(s)|| (right). The solid line indicates the rotation field Ψ+(s)
under the observer transformation and the broken line the rotation interpolation of the extracted rotation
values Ψextr(s) . Solid lines correspond to the proper values.
Dispute Eqn. (65) states that the interpolation is not preserved in the observer
transformation. This does not mean that the rotation interpolation is non-objective. Indeed,
this property is never required for being an objective formulation. It is sufficient that the
86 Jari Mäkinen
rotation interpolation satisfies the condition (61a). Generally speaking, a global interpolation
on a non-flat manifold never preserves under an observer transformation. This is because of
the nonlinear character of the manifold; a parametrization mapping is a nonlinear mapping for
a non-flat manifold. In the paper [6], the authors utilize a corotational interpolation, where the
interpolation is carried out with respect to an element-attached frame. Hence, this
interpolation naturally preserves in a rigid body motion.
Therefore, we may pronounce that extracting the nodal rotation vectors from the
corresponding rotation operations may not be interpolated. There are infinite possibilities for
the interpolation since the transformation operator Q ∈ SO (3) is arbitrary.
The rotation interpolation after an observer transformation reads formally
ɶ + ( s ) = log ( Q exp( N ( s ) Ψ
Ψ ɶ )) (66)
i i
where log -operator is the inverse operator of exp -operator. Now, the transformed rotation
+
field Ψ is not a (vector-valued) linear function. If we substitute Eqn. (66) for R , we find
+
out that
R + ( s ) = exp( Ψ
ɶ + ( s )) = Q exp( N ( s ) Ψ
i
ɶ ) = QR ( s ) .
i (67)
Hence we have Eqn. (61a) as expected. Note that the rotation interpolation of extracted nodal
values Ψ extr ( s ) , the left side of Eqn. (65), is not objective and does not realize the condition
(61a), (See Figure 7).
We have assumed above that the observer transformed rotation interpolation
Ψ + ∈ C ( matTI ) keeps the base point I fixed. However, it also makes sense and can be
assumed that the base point transforms under the observer transformation ( I → Q ) giving
Ψ + ∈ C ( matTQ ) . Then we notice that Ψ + ( s ) = Ψ ( s ) and an interpolation is preserved
under an observer transformation. This is an important issue and clarifies the frame-
indifference of geometrically exact beam formulations.
Conclusions
In this paper, we have shown that incremental material rotation vectors belong to different
tangent spaces of the rotation manifold SO (3) at a different instant. Moreover, we have
proven that the material tangent space as the tangent space at unity is not a possible definition
yielding geometrically inconsistent results, although this kind of definition is widely adopted
in applied mechanics community. We have shown that simplified Newmark integration
scheme for finite rotations neglects first order terms of rotation vector. Hence, the direct
application of the material incremental rotation vector with standard time integration methods
neglects first order terms of incremental rotation vector, likewise in the spatial case. In
addition, we have shown that the rotation interpolation of extracted nodal values is not an
objective interpolation under the observer transformation.
Rotation Manifold SO(3) and Its Tangential Vectors 87
Acknowledgements
Financial support for this research was provided by the Academy of Finland under the project
number 206020. This support is gratefully acknowledged.
References
[1] Abraham R, Marsden J, Ratiu T (1983) Manifolds, Tensor Analysis and Applications.
Addison-Wesley Reading.
[2] Argyris J (1982) An Excursion into Large Rotations. Comp. Methods Appl. Mech.
Engng 32: 85-155.
[3] Argyris J, Poterasu VF (1993) Large Rotation Angles Revisited Application of Lie
Algebra. Comput. Methods Appl. Mech. Engng 103: 11-42.
[4] Cardona A, Géradin M (1988) A Beam Finite Element Non-Linear Theory with Finite
Rotations. Int. J. Num. Meth in Engng 26: 2403-2438.
[5] Choquet-Bruhat Y, DeWitt-Demorette C, Dillard-Bleick M, (1989) Analysis, Manifolds
and Physics, Part I: Basics, North-Holland Amsterdam.
[6] Crisfield MA, Jelenic G (1999) Objectivity of Strain Measures in the Geometrically Exact
Three-Dimensional Beam Theory and Its Finite-Element Implementation. Proc. Royal
Society of London A 455: 1125-1147.
[7] Géradin M, Cardona A (2001) Flexible Multibody Dynamics: A Finite Element
Approach, John Wiley and Sons Chichester.
[8] Ibrahimbegović A (1997) On the choice of finite rotation parameters. Comput. Methods
Appl. Mech. Engrg 149: 49-71.
[9] Ibrahimbegović A, Al Mikdad M (1998) Finite rotations in dynamics of beams and
implicit time-stepping schemes. Int. J. Num. Meth Engrg 41: 781-814.
[10] Ibrahimbegović A, Frey F, Kozar I (1995) Computational Aspects of Vector-Like
Parametrization of Three-Dimensional Finite Rotations. Int. J. Num. Meth. Engng 38:
3653-3673.
[11] Jelenić G, Crisfield MA (1999) Geometrically Exact 3D Beam Theory: Implementation
of a Strain-Invariant Finite Element for Statics and Dynamics. Comp. Meth. Appl.
Mech. Engng 171 (1999) 141-171.
[12] Marsden JE, Hughes TJR (1994) Mathematical Foundation of Elasticity Dover, New
York.
[13] Marsden JE, Ratiu TS (1999) Introduction to Mechanics and Symmetry: A Basic
Exposition of Classical Mechanical Systems, Springer-Verlag New York.
[14] Mäkinen J (2001) Critical Study of Newmark-Scheme on Manifold of Finite Rotations.
Comp. Meth. Appl. Mech. Engng 191: 817-828.
[15] Mäkinen J, Marjamäki H (2005) Total Lagrangian Parametrization of Rotation
Manifold, In: Proc. The Fifth EUROMECH Nonlinear Dynamics Conference, ENOC-
2005, Eindhoven, 2005, 522-530. http://www.tut.fi/~jmamakin/ENOC2005.pdf.
[16] Mäkinen J (2007) Total Lagrangian Reissner’s Geometrically Exact Beam Element
without Singularities. International Journal for Numerical Methods in Engineering
70(9) 1009-1048.
[17] Ogden RW (1984) Non-Linear Elastic Deformations, Ellis Horwood Chichester.
88 Jari Mäkinen
[18] Rubin MB (2007) A simplified implicit Newmark integration scheme for finite
rotations, Computers and Mathematics with Applications 53(2): 219-231.
[19] Simo JC, Vu-Quoc L (1988) On the Dynamics in Space of Rods Undergoing Large
Motion - A Geometrically Exact Approach, Comp. Meth. Appl. Mech. Engng 66: 125-
161.
[20] Simo JC, Marsden JE, Krishnaprasad PS (1988) The Hamiltonian Structure of
Nonlinear Elasticity: The Material and Convective Representation of Solids, Rods, and
Plates, Arch. Rat. Mech. Anal. 104: 125-183.
[21] Stuelpnagel J (1964) On the Parametrization of the Three-Dimensional Rotation
Group”, SIAM Review 6: 422-430.
[22] Stumpf H, Hoppe U (1997) The Application of Tensor Algebra on Manifolds to
Nonlinear Continuum Mechanics – Invited Survey Article, ZAMM Applied
Mathematics and Mechanics 77: 327-339.
In : Lie Groups : New Research ISBN 978-1-60692-389-4
Editor : Altos B. Canterra, pp. 89-122 c 2009 Nova Science Publishers, Inc.
Chapter 3
Jacques Franchi∗
I.R.M.A., Université Louis Pasteur et CNRS, 7 rue René Descartes,
67084 Strasbourg cedex. France
Abstract
2
Consider G := P SL2 (R) ≡ T 1 H , a modular group Γ, and the homogeneous
2
space Γ\G ≡ T 1 (Γ\ H ). Endow G , and then Γ\G , with a canonical left-invariant
metric, thereby equipping it with a quasi hyperbolic geometry. Windings around
handles and cusps of Γ \ G are calculated by integrals of closed 1-forms of Γ \ G .
The main results express, in both Brownian and geodesic cases, the joint convergence
of the law of these integrals, with a stress on the asymptotic independence between
slow and fast windings. The non-hyperbolicity of Γ\G is responsible for a difference
between the Brownian and geodesic asymptotic behaviours, difference which does not
2
exist at the level of the Riemann surface Γ\ H (and generally in hyperbolic cases).
Identification of the cohomology classes of closed 1-forms with harmonic 1-forms,
and equidistribution of large geodesic spheres, are also addressed.
1. Introduction
Consider G := P SL2(R) ≡ T 1H2 , a modular group Γ, and the homogeneous space
Γ\G ≡ T 1 (Γ\H2 ). Endow G , and then Γ\G , with a canonical left-invariant metric, thereby
equipping it with a quasi-hyperbolic geometry, which pertains to the 6th 3-dimensional
∗
E-mail address : franchi@math.u-strasbg.fr
90 Jacques Franchi
geometrical structure of the eight described by Thurston [T]. The non-hyperbolic manifold
Γ\G has finite volume, finite genus, and a finite number of cusps.
It is natural in this setting to study the asymptotic behaviour of the Brownian motion
and of the geodesic flow (under some Liouville-like measure), by means of their asymptotic
homology, calculated by the integrals of the harmonic 1-forms of Γ\G along their paths.
The main results here express, in both Brownian and geodesic case, the joint convergence
in law of these integrals, with asymptotic independence of slow and fast windings.
This same results yield in fact also the asymptotic law of the normalised integrals, along
Brownian and geodesic paths, of any C 2 closed 1-form. Indeed, it holds true on Γ\G that
the cohomology classes of closed 1-forms can be identified with harmonic 1-forms.
The non-hyperbolicity of Γ \ G is responsible for a difference between the Brownian
and geodesic asymptotic behaviours, difference which does not exist at the level of the
Riemann surface Γ\H2. Lifting to the unit tangent bundle has also the advantage to free the
harmonic forms of the constraint to have a null sum of their residues at the cusps of Γ\ H2 .
Counter to the hyperbolic setting, the geodesic flow on Γ \ G is not ergodic with respect
to the normalised Liouville measure of Γ \ G, so that the Liouville-like law governing
the geodesic, for which ergodicity holds, has to be supported by some leaf. Moreover
asymptotic equidistribution of large geodesic spheres holds for such measure.
This article, in which the modular group Γ is arbitrary, is mainly a generalisation of
[F3], which deals with the particular case of the modular group Γ being the commutator
subgroup of P SL2(Z) ; in which case the quotient manifold (which is interestingly linked
to the trefoil knot) has a unique cusp and a unique handle ; for example, we have now to
take into account the pullbacks of singular harmonic forms on Γ\ H2 , which did not exist
in [F3]. However, the identification of the cohomology classes with harmonic spaces H 1
and the equidistribution of large geodesic spheres, two questions which are addressed here,
were not discussed in [F3].
Brownian and geodesic asymptotic behaviours were already studied in a similar way,
but in an hyperbolic setting, in [E-F-LJ1], [E-F-LJ2], [E-LJ], [F2], [G-LJ], [LJ1], [LJ2],
[W]. As in [F3], hyperbolicity (which is replaced by quasi-hyperbolicity) does not hold in
the present setting, nor ergodicity of the Liouville measure (which has to be replaced by
Liouville-like measures, supported by leaves), and the asymptotic Brownian and geodesic
windings are no longer the sames, though comparable (the spiral windings of the geodesics
of G about their projections on H2 is mainly responsible for this feature). Moreover fast
and slow windings are addressed jointly.
As in [LJ1], [G-LJ], [F3], the aim is here to study asymptotic homology, meaning that
only closed forms are considered. Consequently no foliated diffusion is needed. Whereas
[LJ2], [E-LJ], [F2], [E-F-LJ1], [E-F-LJ2] dealt with non necessarily everywhere closed
1-forms, so that the showing up of a spectral gap at the level of the stable foliation was
needed.
verges
in law, under the ergodic measure µkε (dγ),
to
(1 + a2 )k 1 − k2 1/2 X
ν ∞ (Γ) 4(1 − k2 ) 1/4
√ 1{j=0} + 2 r `
j Q` , N l , where
1 + a2 k 2 1 + a2 k2 1 + a2k2
`=1
j,l
the limit random variables Q` , N l are as in Theorem 6.1, and a is the parameter of the
metric.
An interesting feature is the difference between the Brownian and geodesic behaviours,
in noteworthy contrast with the hyperbolic case : counter to the Brownian case, the dθ-part
of the form ω0 is responsible for a non-negligible asymptotic contribution, and the metric
parameter a now appears in the limit law.
9.) Equirepartition in Γ\G of large geodesic spheres
Corollary 9.1 asserts that the ergodic measures µkε of Section 7. and Theorem 8.1, are
weak limit of the uniform law on large geodesic quasi-spheres. This is easily deduced from
the following.
Theorem 9.1 The normalized Liouville measure µΓ on T 1 (Γ\H2 ) ≡ Γ\G is the weak limit
as R → ∞ of the uniform law on the geodesic sphere Γg P SO(2)ΘR of Γ \ G having
radius R and fixed center Γg ∈ Γ\G : for any compactly supported continuous function
f on Γ\G , denoting by d% the uniform law on P SO(2), we have
Z Z
f dµΓ = lim f (Γg % ΘR ) d% .
R→∞ P SO(2)
The equidistribution theorem 9.1 and its proof (based on the mixing theorem) were already
given by Eskin and McMullen in [E-MM].
10.) Synthetic proof of Theorem 6.1
This proof is an adaptation of an analogous proof in ([F3], Section 10). Thus some
details are here somewhat eluded, for which we refer to [F3]. However all ingredients are
given, with a stress on differences with the particular case addressed in [F3], and the most
involved arguments are detailed to a certain extent.
The main difficulty of the whole proof, widely responsible for its length, is to estab-
lish the asymptotic independence between slow windings (about the handles) and singular
windings (about the cusps). This demands in particular to get good approximation of the
contribution of both type, and then to analyse carefully the successive excursions of Brow-
nian motion in the core and in the cusps of the quotient hyperbolic surface.
11.) Proof of Theorem 8.1
The strategy for this proof is mainly to replace the geodesic paths by the Brownian
paths, as in hyperbolic case ([LJ1], [LJ2], [E-LJ], [F2], [E-F-LJ1], [E-F-LJ2]), and as in
[F3], in order to reduce Theorem 8.1 to Theorem 6.1. Here again, the analogous proof in
([F3], Section 14) is adapted.
plane, identified with the Poincaré half-plane), in the following way : each g ∈ G writes
uniquely
g = g(z, θ) := ± n(x)a(y)k(θ) ,
and generated respectively by the following elements of the Lie algebra s`2 (R) :
0 1 1/2 0 0 1/2
ν := , α := , κ := .
0 0 0 −1/2 −1/2 0
h √ √ √ i
Note that g = g(z, θ) ⇐⇒ g( −1 ) = z and g 0( −1 ) = y e −1 θ .
0 1/2
Set also λ := ν − κ = , which is natural, since α, λ are symmetrical
1/2 0
while κ is skew-symmetrical, the basis (α, λ, κ) of s`2 (R) the Killing form
and since in
−2 0 0
is diagonal : it has matrix 0 −2 0 .
0 0 2
For this reason, we take on s`2 (R) the inner product such that the basis (α, λ, aκ)
is orthonormal, for some arbitrary parameter a ∈ R∗. And since we want to work on an
homogeneous space Γ\G , the Riemannian metric to be considered on G must be a least Γ-
left-invariant, and then a natural choice for the Riemannian metric on G is the left-invariant
a
metric, say ((gij )) , generated by the above inner product on s`2 (R) .
The simple lemma below shows that this choice of metric(s) is geometrically canonical
(up to a trivial multiplicative constant), G being seen as T 1H2 . This equips G ≡ T 1H2 ,
and its homogeneous spaces as well, with the 6th of the eight 3-dimensional geometries
described by Thurston ([T]), and actually with a quasi-hyperbolic but not hyperbolic struc-
ture.
Let us denote by Lν , Lα , Lκ , Lλ the left-invariant vector fields on G generated
respectively by ν , α , κ , λ . A standard computation shows that
∂ ∂ ∂ ∂ ∂ ∂ ∂
Lλ = y sin θ +y cos θ −cos θ , Lα = y cos θ −y sin θ +sin θ , Lκ = .
∂y ∂x ∂θ ∂y ∂x ∂θ ∂θ
a
Lemma 2.1 The Riemannian metrics ((gij )) defined above are, up to a multiplicative
constant, the only ones on G which are left-invariant and also invariant with respect to the
action of the (Cartan compact subgroup) circle exp(Rκ) = {k(θ)} . They are given in
Iwasawa coordinates (y, x, θ) by :
y −2 0 0
((gij )) := 0
a
(1 + a−2 )y −2 a−2 y −1 . (2)
0 a−2 y −1 a−2
94 Jacques Franchi
Proof The left-invariant metrics on G are those which are given by a constant matrix
∂ ∂ ∂
((aij )) in the basis L := (Lα , Lλ, Lκ ) . Set I := , , . We have I = LA , with
−1 ∂y ∂x ∂θ
y cos θ −y −1 sin θ 0
A := y −1 sin θ y −1 cos θ 0 , so that the left-invariant metrics are given in the ba-
0 y −1 1
t ∂ t
sis I by A((aij ))A . Among them, the ones we want have to satisfy A((aij ))A = 0 .
∂θ
1 0 0
A direct computation shows that this is equivalent to ((aij )) = c2 0 1 0 , and
0 0 a−2
a
then to ((gij )) being as in the statement.
Note that with these metrics any holomorphic form f (z)dz is coclosed, and then
harmonic.
The left Laplacian on G corresponding to the basis (α, λ, aκ) is the Beltrami Laplacian
a )), and is given by
associated with the metric ((gij
∂2 ∂2 ∂2 2
2 ∂
∆a := L2λ + L2α + a2L2κ = y 2 + − 2y + (1 + a ) . (3)
∂y 2 ∂x2 ∂θ∂x ∂θ2
Note that Lλ and Lα generate the canonical horizontal left-invariant vector fields lifted
from H2 to G, H2 being endowed with its Levi-Civita connexion, so that ∆0 is the Bochner
∂2
horizontal left Laplacian, and ∆a = ∆0 + a2 ∂θ 2 .
dx dy dθ
The measure µ(dg) := is bi-invariant, hence this is both the Haar measure
4π 2 y 2
of G and the Liouville measure of T 1 H2 .
Recall that an isometry γ ∈ G is called respectively elliptic, parabolic, or loxodromic,
according as it fixes a point in H2 , no point in H2 and a unique point in ∂ H2 = R ∪ {∞},
or no point in H2 and two points in ∂ H2 = R ∪ {∞}, respectively. Any isometry of H2
is either elliptic, or parabolic, or loxodromic. We shall use the following easy lemma.
Γ̃ := Γ ∩ DΓ(1)
.
is a free and normal subgroup of Γ such that Γ/Γ̃ ≡ Γ · DΓ(1) DΓ(1) is a subgroup of
Γ(1)/DΓ(1), so that Γ/Γ̃ is isomorphic to a subgroup of Z/6Z .
dx dy dθ
The projection of the Liouville measure µ(dg) = (which is also a right and
4π 2 y 2
left Haar measure on G) onto Γ\G is proportional to the volume measure V of Γ\G . By
96 Jacques Franchi
a 2π
the choice of the metric ((gij )), the volume of Γ\G is clearly V (Γ\G) = |a| × covol(Γ),
2 2
where covol(Γ) = V (Γ\ H ) denotes the finite hyperbolic volume of Γ\ H .
2π
Let µΓ := covol(Γ) µ denote the normalized projection of µ on Γ\G , identified
Γ\G
with a law on left Γ-invariant functions on G.
Proof As a cover of Γ(1)\H2, the Riemann surface Γ\H2 is orientable and non-compact.
As is known for any orientable non-compact smooth manifold,
it carries a smooth non-
vanishing vector field, hence a smooth cross section x 7→ (x, ~vx) ∈ Tx1(Γ \ H2 ) of
\
T 1 (Γ\H2 ). Denoting by α = αx (~v) the angle (~
vx, ~v) in the oriented plane Tx1 (Γ\H2 ), we
get the diffeomorphism : (x, ~v) 7→ (x, α) from T 1(Γ\ H2) onto (Γ\ H2 ) × S1 .
Furthermore, under the canonical projection π : Γ\G → Γ\H2 ≡ Γ\G/ exp(Rκ), the
harmonic space H 1(Γ\ H2 ) (that is, the space of real harmonic forms on Γ\ H2) is pulled
back to the subspace π∗ [H 1(Γ\H2 )] of the harmonic space H 1(Γ\G), which is isomorphic
to H 1(Γ\ H2 ).
Thus, to describe the harmonic 1-forms of the modular homogeneous space Γ\G , once
the harmonic 1-forms of the Riemann surface Γ\ H2 are known, by the above lemma 3.1 it
is sufficient to produce a harmonic 1-form ω0 ∈ H 1(Γ\G), such that ω0 ∈/ π∗ [H 1(Γ\H2)].
Now, such harmonic 1-form ω0 was computed in ([F3], Section 6), as the restriction to
Γ\G of a harmonic 1-form on Γ(1)\G :
ω0 := dθ + 4 Im(η 0(z)/η(z)) dx + 4 Re(η 0(z)/η(z)) dy = d θ + 4 arg(η(z)) , (4)
where η denotes the Dedekind function, defined on H2 (seen as the Poincaré half-plane)
by : √ Y √
η(z) := e −1 π z/12 × (1 − e −1 2π n z ) .
n∈N ∗
The Riemann surface Γ \ H2 decomposes into the disjoint union of a compact core,
which is a compact surface having genus Γ\ H2 and a boundary made of ν∞ (Γ) pairwise
disjoint circles S1 , and of ν∞ (Γ) pairwise disjoint ends, each being diffeomorphic to S1 ×
R∗+ and associated with one of the cusps C` , which we call “solid cusp” and denote also by
C` .
Recall now that the harmonic space H 1(Γ\ H2 ) is the dual of the first real singular ho-
mology space H1 (Γ\H2 ) (see for example ([D], 24.33.2)), so that the above decomposition
implies the formula :
On the other hand, an automorphic form f of weight 2 with respect to Γ induces the
holomorphic differential f (z) dz on Γ\H2. More precisely, let us denote as Miyake ([M])
by G2 (Γ) the complex vector space of those automorphic forms f (of weight 2) which are
holomorphic on H2 and at the cusps of Γ, and by S2(Γ) the subspace of so-called “cusp
forms”, that is of forms in G2 (Γ) which vanish at the cusps of Γ.
The so-called Petersson inner product (see for example ([M], Section 2.1)) is defined
for (f1 , f2) ∈ S2(Γ) × G2(Γ), by :
Z
2 −1
hf1 (z)dz, f2(z)dzi = hf1 , f2i := V (Γ\ H ) f1(z) f2 (z) dz . (6)
Γ\H 2
Note that y |f (z)| is the natural norm of f (z)dz ∈ Tz∗ (Γ \ H2 ), induced by the volume
dxdy
measure of Γ\H2 ; so that the differential |f (z)|2 dz = kf (z)dzk2 2 , integrated over
y
Γ\ H , indeed computes precisely the global norm of f (z)dz ∈ T (Γ\ H2 ).
2 ∗
As in [M] again, let N2 (Γ) denote the orthogonal complement of S2 (Γ) in G2 (Γ), with
respect to the Petersson inner product. Then ([M], Theorem 2.5.2) states the following :
Note that the −1 in the second formula is natural, since the sum of the residues of a
harmonic differential form has to be zero.
As a consequence, comparing (5) and (7), we see that the regular part of the harmonic
space H 1(Γ\ H2 ), that is the part due to the handles, admits a basis made of 2 g(Γ) real
harmonic forms Re[f (z) dz], with f ∈ S2(Γ) a cusp form ; and that the singular part of
the harmonic space H 1(Γ \ H2 ), that is the part due to the cusps and having residues at
the cusps which are not all null, admits a basis made of (ν∞ (Γ) − 1) real harmonic forms
Re[f (z) dz] , with f ∈ N2 (Γ).
Notice that, specifying that the singular harmonic forms must be real and orthogonal to
the regular harmonic forms (and then in particular have non all vanishing residues), we get
indeed (ν∞ (Γ) − 1) independent such differential forms, and not 2(ν∞ (Γ) − 1).
since on one hand the free group Γ̃ cannot have any elliptic point, and on the other hand,
recalling the definition of ν∞ (Γ) (in Section 3.2.), we must have :
and
d˜ = dim [H 1(Γ̃\ H2 )] = 1 + [Γ(1) : Γ̃]/6 ≥ 1 + [Γ(1) : Γ]/6 ≥ d .
Note that we can have d < d˜, as shows the simple example Γ = Γ(1), for which
d = g(Γ) = 0 , ν∞ (Γ) = ν∞ (Γ̃) = g(Γ̃) = 1 , d˜ = 2 .
Furthermore, if F is a free set of generators of Γ̃, and if γ ∈ Γ represents a generator
of Γ/Γ̃, then Γ is generated by F t{γ}, with a relation γ n = φ(F ) (where n ∈ {1, 2, 3, 6}
is the order of Γ/Γ̃). Hence the Abelianized Ab(Γ) := Γ/[Γ, Γ] admits the same . repre-
sentation as a free Abelian group, meaning that the Abelianized Ab(Γ̃) := Γ̃ [Γ̃, Γ̃] is
isomorphic to a subgroup of Γ/[Γ, Γ], such that the quotient be cyclic (of order dividing
n). Hence, considering the group Γ b of additive (real) characters of Γ, we find that :
b ≡ Ab(Γ)
Γ \
\ ≡ Ab( 0
Γ̃) ≡ Zd ,
. 0
where d0 is the dimension of the free Abelian group Ab(Γ̃) = Γ̃ [Γ̃, Γ̃] ≡ Zd , or equiva-
lently, the number of generators of Γ̃.
Now we have the following important fact, which generalizes Proposition 2.2 of [G-LJ],
and, allowing the identification of the cohomology and of the harmonic space of Γ \ H2 ,
allows mainly to focus on harmonic forms the asymptotic study of integrals of closed forms.
100 Jacques Franchi
Proposition 4.1 (i) The number of generators of every free modular group Γ̃ equals the
dimension of the corresponding harmonic space : d0 = d˜ := dim [H 1(Γ̃\ H2 )].
(ii) For any modular group Γ, every C 1 closed form on Γ\ H2 is cohomologous to a
(unique) harmonic form, element of H 1(Γ\ H2 ).
Proof 1) As ([G-LJ], Prop 2.2), fix a base-point b̃ ∈ Γ̃ \ H2 , b ∈ b̃ , and to any loop
` from b̃ to b̃ , associate its lift `¯ in H2 started from b , and `(b)
¯ = `˜b ∈ b̃ . This
˜ e 0 ˜ ˜0 1 2
defines `Z ∈ Γ̃ satisfying
Z Zll = l l . Then for any C closed form ω̃ on Γ̃ \ H , set
ω̃(`) := ω̃ = ω̃ = π̃∗ω̃ , where π̃ denotes the covering projection from H2 onto
` π̃◦`¯ `¯
Γ̃\H2 , which induces a pullback π̃∗ , mapping ω̃ to a closed form on H2 . As H2 is simply
¯
connected, ω̃(`) is a function of `(b), ˜ thereby defining an
b (`),
so that we can set ω̃(`) =: ω
additive character ωb on Γ̃. Z ·
b = 0 , meaning that ω̃ is exact, then the primitive
Now if ω ω̃ is defined on the
b̃
whole Γ̃\H2 (since it is arcwise connected), and then constant (as any holomorphic modular
function on Γ\ H2 ; see for example ([L], VI.2.E)), proving that ω̃ = 0 . Hence, the linear
map ω̃ 7→ ω b is one-to-one from the space of C 1 closed forms on Γ̃ \ H2 into Γ, b and, a
1 2 b ˜
fortiori, from H (Γ̃\ H ) into Γ, proving that d ≤ d . 0
2) Reciprocally, notice that the map ` 7→ `˜ defined above induces a morphism ϕ from
π1 (Γ̃\ H2 ) into Γ̃, since any homotopy (`s ) from ` to `0 defines a path (`¯s (b)) from `(b)
¯
¯0 ˜ ˜0 ˜ ¯
to ` (b) in the discrete Γ̃ b , forcing ` = ` . And if ` = 1 , then `(b) = b in the simply
connected H2 , so that `¯ is homotope to the constant loop b , ` is homotope to the constant
loop b̃ , and thus we have a one-to-one morphism ϕ from π1(Γ̃\ H2 ) into Γ̃.
Moreover, since Γ̃ is discrete and free, it contains only parabolic and loxodromic isome-
tries, and then by Lemma 2.2, any γ ∈ Γ̃ can be written γ = ± exp(σ), for a unique
σ ∈ sl2(R). Setting `(γ)(s) := ± exp(s σ)(b̃) for 0 ≤ s ≤ 1 and taking the homotopy
ˆ ∈ π1 (Γ̃\ H2 ) of `(γ), we get a pre-image for γ, with respect to the above mor-
class `(γ)
phism ϕ , which is thus onto. By duality ([D], 24.33.2), this yields a one-to-one morphism
from Γ b into H 1(Γ̃\ H2), hence d0 ≤ d˜.
Hence d˜ = d0 , and we have exhibited an isomorphism ω̃ 7→ ω b from H 1(Γ̃\ H2 ) onto
b
Γ.
3) Fix a basis (ω̃1, .., ω̃d0 ) of H 1(Γ̃\H2), and consider any C 1 closed form ω̃ on Γ̃\H2 ,
which, as described above, defines ω b By the above, we have reals α1 , .., αd0 , such
b ∈ Γ.
d0 d0
X X
b=
that ω bj , implying, as seen above, that ω̃ −
αj ω αj ω̃j be exact.
j=1 j=1
So far, we have proved (i) of the statement, and (ii) for the case of a free modular group
Γ̃.
4) Let p̃ denote the covering projection from Γ̃ \ H2 onto Γ \ H2 , which induces a
pullback p̃∗ , mapping closed smooth differential forms on Γ \ H2 to closed smooth dif-
ferential forms on Γ̃ \ H2 , and harmonic forms on Γ \ H2 to harmonic forms on Γ̃ \ H2 ,
that is, H 1(Γ \ H2 ) into H 1 (Γ̃ \ H2 ). Note that p̃∗ is necessarily one-to-one (proving
˜ : indeed, if a closed smooth form ω belongs to its kernel, then we have
againZ that d ≤Zd)
0 = p̃∗ω = ω for any loop `˜ on Γ̃\H2 . Now, since the order of the covering group
`˜ p̃◦`˜
Asymptotic Homology of a Modular Quotient 101
IntegratingZthis relation along the loop `˜k , using 5) above, gives at once αk = 0 for k > d ,
and αk = ω for 1 ≤ k ≤ d .
`k
d
X
Hence ω̄ := ω − αj ωj is a C 1 closed form ω on Γ\H2 , vanishing on H1 (Γ\H2 ),
j=1
and such that p̃∗ ω̄ = dF . For any Zb̃ ∈ Γ̃\ HZ2 , γ ∈ Γ,Zand any arc c joining b̃ to γ b̃ in
Γ̃\H2 , we have : F (γ b̃) − F (b̃) = dF = p̃∗ ω̄ = ω̄ = 0 , since p̃ ◦ c is a loop in
c c p̃◦c
Γ\H2 . This proves that F is Γ-invariant, hence that F = F̃ ◦ p̃ for some F̃ ∈ C 2 (Γ\H2 ).
Xd
Finally we have got : p̃∗(ω̄ − dF̃ ) = 0 , whence ω = αj ωj + dF̃ , by 4) above.
j=1
Moreover, the preceding proposition 4.1 lifts to the modular homogeneous space Γ\G :
we can also identifiy the cohomology and the harmonic space of Γ\G , and then focus on
harmonic forms the asymptotic study of integrals of closed forms on Γ\G .
Proposition 4.2 For any modular group Γ, every C 1 closed form on Γ\G is cohomologous
to a harmonic form, element of H 1(Γ\G).
Proof Recall from Section 3.1. the canonical projection π : Γ \ G → Γ \ H2 , whose
pullback π∗ maps in a one-to-one way the closed forms on Γ\H2 to closed forms on Γ\G ,
and the harmonic space H 1(Γ \ H2 ) to the isomorphic subspace π∗ [H 1(Γ \ H2 )] of the
harmonic space H 1(Γ\G).
Fix b = Γ · (y, x, θ) ∈ Γ \ G, and denote by `0 a loop above π(b), generating
H1 (π −1(b)). Fix also a basis (`1, .., `d) of H1 (Γ \ H2 ), dual to the basis (ω1 , .., ωd) of
102 Jacques Franchi
H 1(Γ\H2). Identify (`1, .., `d) with its lift to H1(Γ\G), prescribing merely to each `j the
constant third Iwasawa coordinate θ . By Theorem 3.1 and by duality Z (see for example
Z ([D],
24.33.2)), (`0, `1, .., `d) is a basis of H1 (Γ\G) : indeed, we have π∗ ω = ω=0
Z `Z
0 π◦`0
Hence we see that the projection of our Brownian motion gs = (ys , xs, θs ) on the
hyperbolic plane H2 , that is to say on the Iwasawa coordinates (y, x) , is simply the stan-
dard hyperbolic Brownian motion of H2, and that the angular component (θs ) is just a real
Brownian motion with variance (1 + a2 ).
Remark 5.1 The degenerate limit-case a = 0 is quite possible for the left Brownian
motion (gs ) . It corresponds to the Carnot degenerate metric on G, and to the horizontal left
Brownian motion on G, associated with the Levi-Civita connexion on H2.
Asymptotic Homology of a Modular Quotient 103
the martingales obtained by integrating the harmonic forms ωj and ω̃l along the paths of
the left Brownian motion (gs ). Note that we may as well consider the Brownian motion
(gs ) as living on G or on Γ\G .
Section 5. and Theorem 3.1 show that
Z t
0 0
0
Mt = a Wt + 4 Re ηη (xs , ys ) ys dUs + (4 Im ηη (xs , ys ) ys − 1) dVs ,
0
Z t
Mtj = Refj (xs , ys ) ys dVs − Imfj (xs , ys ) ys dUs for 1 ≤ j < ν∞ (Γ),
0
and
Z t
M̃tl = Re f˜l (xs , ys ) ys dVs − Im f˜l (xs , ys ) ys dUs for 1 ≤ l ≤ 2g(Γ).
0
Lemma 6.1 The law of t−1/2 (M̃tl )1≤l≤2g(Γ) converges towards the centred Gaussian
law with diagonal covariance matrix having (on its diagonal) the variances hω̃l , ω̃li.
Proof By the above, we have some real Brownian motion (B̃sl ) such that
Z t
M̃tl l l
= B̃ hM̃ it = B̃ l 2
|ω̃j (zs )| ds ,
0
and then by scaling, we have the following identity in law (for each t > 0) :
Z t
t−1/2 M̃tl ≡ B̃ l hM̃ lit /t = B̃ l t−1 |ω̃l (zs )|2 ds ,
0
Moreover, by ergodicity and by orthogonality of the different ω̃l , for 1 ≤ l < ` ≤ 2g(Γ),
we have :
0
t−1 hM̃ l , M̃ l it −→ hω̃l , ω̃l0 i = 0 .
Hence Knight’s Theorem (see ([R-Y], XIII, 2, Corollary (2.4))) implies the asymptotic
independence of the martingales (M̃tl).
The following theorem describes the asymptotic Brownian windings in Γ\G .
104 Jacques Franchi
. .√
Theorem 6.1 As t → ∞, Mtj t , M̃tl t converges in law to-
0≤j<ν∞ (Γ), 1≤l≤2g(Γ)
ν∞
X (Γ)
wards rj` Q` ,N l
, where all variables Q` , N l are indepen-
`=1 0≤j<ν∞ (Γ), 1≤k≤2g(Γ)
h`
dent, each Q` is Cauchy with parameter 2 V (Γ\ H2 )
, and N l is centred Gaussian with
variance hω̃l , ω̃li. Here h` > 0 denotes the width of the solid cusp C` (already defined in
Theorem 3.1).
Observe the irrelevance of the parameter a in this theorem, which is valid as well in the
∂
degenerate case a = 0 . The reason is that a was initially the inverse norm of Lκ = ∂θ ,
which comes in only in the differential form ω0 , and which contributes there only to a
second order term.
Note that such statement gives at once the asymptotic law of any finite family of stochas-
tic integrals (along Brownian paths) of harmonic forms, merely by decomposing them in
the basis (ωj , ω̃k ).
By Proposition 4.2, this gives also the asymptotic behavior of any finite family of
stochastic integrals (along Brownian paths) of smooth closed forms, since the normalized
contribution of any exact form is clearly negligible in probability (at least for the stationary
Brownian motion).
pressed by the Lagrangian L = L(y, x, θ, ẏ, ẋ, θ̇) , given in Iwasawa coordinates (y, x, θ)
by :
and
(1 + a−2 ) y −2 ẋ + a−2 y −1 θ̇ = c0 and y −1 ẋ + θ̇ = c , (13)
for two constants c0, c . Eliminating θ̇ , this gives :
Eliminating then c a−2 gives : (ẏ/y)· = −c0 ẋ , or equivalently, for some constant c00 :
ẋ2 + ẏ 2
(c0 x − c00)2 + (c0y − c a−2)2 = = C2 , and θ̇ = c (1 + a−2 ) − c0 y . (16)
y2
In the particular case c0 = 0 , we find c00x + c a−2 y = c0 constant and θ̇ = c(1 + a−2 ).
Hence we see that any geodesic projects on a Euclidian circle or line of H2, and that its
projection has constant speed C (or energy C 2 ). Precisely if c0 6= 0 :
where the last formula results at once from the two preceding ones and from Formula (15).
If C 6= 0 , set
k := c a−2C −1 . (18)
Then if k 6= 1 , according as |k| > 1 or |k| < 1, we have :
" √ # " √ #
q q
k+1 k 2 −1 1+k 1−k 2
ϕs = 2 arctg k−1 tg C 2 (s − s0 ) or ϕs = 2 arctg 1−k th C 2 (s − s0 ) .
For |k| > 1, the geodesic projects on a (Euclidian and hyperbolic) circle totally included in
H2, and for |k| < 1, the geodesic projects on a quasi-geodesic of H2 (which is a geodesic
if and only if k = 0). In the limiting case |k| = 1, the geodesic projects on an horocycle
of H2, and ϕs = 2 arctg [C(s − s0 )].
Note moreover that by Equations (11), (13), (16), and (18), we have :
ẋ2 + ẏ 2 2
−2 ẋ
2L = + a + θ̇ = C 2 + a−2 c2 = (1 + a2 k2 )C 2 ,
y2 y
C = (1 + a2 k2 )−1/2 . (19)
We have in particular established that the energy of any geodesic of Γ\G splits into the
constant energy C 2 of its projection on Γ\ H2 and the constant energy 1 − C 2 = a2 k2 C 2
of its angular windings about its projection.
106 Jacques Franchi
Remark 7.1 The case of main interest for the following is |k| < 1, that is, when the
quasi-geodesic γ̃ of H2 we get intersects ∂ H2 in two end-points. It is sufficient to consider
the case of these two end-points are on the real line, that is, when the quasi-geodesic γ̃ is
a circle, of radius R = C/|c0| and centre at 0
√ height y0 = kC/c . Then the geodesic ψ(γ)
2
having the same end-points has radius R 1 − k , and the orthogonal projection ps on
ψ(γ) of the point ms ∈ √ γ̃ having angular coordinate ϕs has angular coordinate αs ,
1−k2 sin ϕs
determined by : sin αs = 1+k cos ϕs . As the speed of γ̃ is C = |ṁs | = R|ϕ̇s |/(y0 +
R cos ϕs ) = |ϕ̇s |/(k + cos ϕs ), we find that q the geodesic ψ(γ) must be run at constant
√ 1−k2
2
speed |ṗs | = |α̇s / cos αs | = C 1 − k = 1+a 2 k 2 , in order that the distance from ms
Let us consider the coordinate system (y, x, θ, u, v, w) on T 1G , where (y, x, θ) are the
Iwasawa coordinates of the base point γ ∈ G , and (u, v, w) are the coordinates of the unit
∂ ∂ ∂
tangent vector in the basis (y ∂y , y ∂x , ∂θ ) of Tγ1G . Thus we have u2 +v 2 +a−2 (v +w)2 ≡
1.
Now, consider the geodesic (γs ) determined by the initial value (γ0, γ00 ) ∈ T 1 G having
coordinates (y0 , x0, θ0, u0, v0, w0), and let k be the unique real number determined by the
k a2 k (1+a2 )
equation v0 + w0 = √1+a 2 k2
, and ε := sign((1 + a2 )v0 + w0) = sign( √ 1+a2 k 2
− w0 ) =
0
sign(c ).
Then by Equations (16), (13), (18), (19),
the above implies that (γs, γs0 ) remains in the leaf L(k, ε) having equations :
\ \
1 k a2
L(k, ε) := u2 + v 2 = v+w = √ sign((1 + a2 )v + w) = ε . (20)
1 + a2 k 2 1 + a2 k 2
Lemma 7.1 For each k fixed in ] − 1, 1[ and each = ±1 , there is a natural one-to-
one map ψ = ψεk from the leaf L(k, ε) (seen as made of geodesics of Γ\G having initial
value θ0 = 0 for their angular part θs ) onto the set of geodesics of Γ\ H2 .
This map goes as follows : with any geodesic γ of Γ \ G , associate successively the
projection γ̃ on H2 of its lift to G, and the projection ψ(γ) on Γ\H2 of the geodesic of H2
at bounded distance of γ̃ .
This map makes sense as well at the level of line-elements, and thus defines a homeo-
morphism from L(k, ε) onto T 1(Γ\ H2 ) ≡ Γ\G .
Proof The analysis made in Section 7.1. ensures that the map ψ = ψεk is well defined.
Note indeed the necessary Γ-invariance : if two geodesics γ, γ 0 of G can be identified
Asymptotic Homology of a Modular Quotient 107
modulo some g ∈ Γ, then indeed the same g identifies also the geodesics of H2 at bounded
distance of the projections γ̃, γ̃ 0 of γ, γ 0 on H2 .
In the reverse direction, to any (oriented) geodesic ψ(γ) of Γ \ H2 correspond two
quasi-geodesics in Γ\ H2 at constant distance argch[(1 − k2 )−1/2], and, owing to the sign
ε which by Formula (20) and Equation (13) prescribes the sign of c0 (and then the sign
of the height of the centre y0 = kC/c0 by Remark 7.1), in fact a unique one. And to this
unique quasi-geodesic in Γ \ H2 is associated by (13) or (16) (for any prescribed initial
value θ0 of the angular part) a unique geodesic Γγ of Γ \ G , obviously included in the
leaf L(k, ε). By using furthermore the orthogonal projection in H2 between our quasi-
geodesics and their associated geodesic, we get at once the analogous map at the level of
line-elements, with a clear continuity in both directions.
Remark 7.2 Note that in fact each leaf L(k, ε) splits into a continuum of sub-leaves :
S
L(k, ε) = θ0 ∈R/2πZ L(k, ε, θ0), taking into account the initial value θ0 of the angular
√
part (either at time 0, or above the orthogonal projection of the fixed point −1 on the
quasi-geodesic γ̃) of the geodesic γ. Thus this is indeed the set of its line-elements of each
sub-leaf L(C 2 , ε, θ0), which is set in one-to-one correspondence with T 1 (Γ\ H2 ) ≡ Γ\G
by the map ψ = ψε,θ k . Note that L(k, ε, θ ) has indeed 3 dimensions, as G. However,
0 0
this initial value θ0 will not matter anyway in the following, so that we drop it henceforth,
going on with the shorter notation L(k, ε), ψεk .
Now it is known (see [H]) that the Liouville measure on T 1 (Γ \ H2 ) is invariant and
ergodic under the geodesic flow. This fact and Lemma 7.1 above allow therefore the fol-
lowing.
Definition 7.1 For any (k, ε) fixed in ] − 1, 1[×{±1}, denote by µkε := µΓ ◦ ψεk , the
image of the normalized Liouville measure µΓ on T 1(Γ\H2 ) ≡ Γ\G under the map ψεk of
Lemma 7.1. So µkε is a probability measure on the set of line-elements of the leaf L(k, ε) ,
which is invariant and ergodic under the geodesic flow on Γ\G .
0
Note that, by Remark 7.3, for |k| < |k0 | < 1 the measures µkε , µkε0 have disjoint supports.
108 Jacques Franchi
where the geodesic γ of Γ\G is chosen (at time 0) according to µkε , and γ[0, t] denotes
this geodesic γ run during the time-interval [0, t] .
Note that by the Γ-invariance of the forms ωj , ω̃l , it makes no difference to think of
the geodesics γ as started in a fundamental domain D and living on G, the forms being
harmonic on G as well.
The following theorem describes the asymptotic geodesic windings in Γ\G , under the
ergodic measures of Section 7.2.. A minor mistake in [F3], concerning the contribution of
dθ , is corrected here.
Theorem 8.1 Let us consider a fixed leaf L(k, ε) (defined in Section 7.1.) of Γ\G , with
|k| < 1 , endowed with the ergodic invariant probability measure µkε of Definition 7.1.
Then
Z the Zlaw under µkε = µkε (dγ) of
t−1 ωj , t−1/2 ω̃l
γ[0,t] γ[0,t] 0≤j<ν∞ (Γ), 1≤l≤2g(Γ)
converges as t → ∞ to the law of
(1 + a 2
)k 1 − k2 1/2 ν∞
X (Γ) 4(1 − k2 ) 1/4
√ 1{j=0} + 2 `
rj Q` , N
l
,
1 + a2 k 2 1 + a2 k2 1 + a2k 2
`=1
0≤j<ν∞(Γ), 1≤l≤2g(Γ)
Note a clear difference between the Brownian and geodesic behaviors : mainly, here
(counter to the Brownian case) the dθ-part of the form ω0 is responsible for a non-
negligible asymptotic contribution. Moreover the parameter a now appears in the limit
law.
This makes a noteworthy contrast with the hyperbolic case (see [E-F-LJ1], [E-F-LJ2], [F2]).
This difference appears in Lemma 8.1 below, whereas once the dθ-part has been moved
away, the remaining asymptotic law is essentially the same as the Brownian one, given by
Theorem 6.1. So the remaining task, following [F3], will be then (in Section 11. below)
mainly to compare on H2 the geodesic paths to the Brownian paths, somewhat in the spirit
of the methods already employed in [E-F-LJ1], [E-LJ], [F2], [LJ2], but in a simpler way,
taking advantage of the harmonicity of the forms ωj , ω̃l , somewhat as in [LJ1].
The following lemma (in which the minor mistake in [F3] concerning the contribution
of dθ is corrected) reduces the study along the geodesics of G to a study along the geodesics
of H2 .
Asymptotic Homology of a Modular Quotient 109
Z Z
Lemma 8.1 The asymptotic law of t−1 ωj , t−1/2 ω̃l
γ[0,t] γ[0,t] 0≤j<ν∞ (Γ), 1≤l≤2g(Γ)
under µkε (dγ) is the same as the asymptotic law under the Liouville measure µΓ on T 1(Γ\
H2) of :
Z Z !
1/2 1/4
(1+a2 )k 1−k2 −1 1−k 2 −1/2
√
1+a2 k 2
1{j=0} + 1+a2 k2
t ωj , 1+a2 k 2
t ω̃l .
γ[0,t] γ[0,t]
0≤j<ν∞(Γ), 1≤l≤2g(Γ)
∗
Proof Let us deal
Z first with ω̃l = Zπ ω̃l . By Definition 7.1
Z and Lemma 7.1, we just have
−1/2 −1/2 −1/2
to compare t ω̃l = t ω̃l with t ω̃l .
γ[0,t] γ̃[0,t] ψ(γ)[0,t]
Now use that on H2, ω̃l = dFl is exact, and recall from Remark 7.1 that
h q i
1−k 2
dist Fj [γ̃(t)], Fj ψ(γ)( 1+a 2 k 2 t) = (1 − k2)−1/2 ,
to get :
Z Z q
ω̃ − ω̃l = Fj [γ̃(t)] − Fj ψ(γ)(
1−k 2
t) − F [γ̃(0)] + F [ψ(γ)(0)]
l
1−k2
1/2 2
1+a k 2 j j
γ̃[0,t] ψ(γ) 0, 2
1+a k 2 t
.p
≤ 2 kω̃l k∞ 1 − k2 .
Z Z
This shows that t−1/2 ω̃l − 1/2 ω̃l goes uniformly to 0, prov-
1−k 2
γ̃[0,t] ψ(γ) 0, t
1+a2 k 2
n hq i √ o
1−k 2 2 .
where Ỹ (g, t) := max sup ỹ` (z) dist g 1+a2 k 2
t , z ≤ 1/ 1 − k
1≤`≤ν∞ (Γ)
On the other hand we have by Equations (16), (18), and (19) :
Z Z
−1 −2 −1
t dθ = c (1 + a ) − t c0ys ds −→ c (1 + a−2 ) − c0 lim y(γ(t))
γ[0,t] γ[0,t] t
.p
= c (1 + a−2 ) = (1 + a2 )k
1 + a2 k2 .
Z
−1
Therefore, by Definition 7.1, the asymptotic law of t ωj under µkε (dγ) is the
γ[0,t]
same as the asymptotic law under µ(dg) of :
Z .
(1 + a2)k −1
√ 1{j=0} + t ωj0 + O Ỹ (g, 0) + Ỹ (g, t) t.
1/2
1 + a2 k 2 g 0, 1−k
2
1+a2 k 2
t
110 Jacques Franchi
Observe further that under µΓ (dg) the process Ỹ (g, t) is stationary, so that the last term
above asymptotically vanishes
Z in probability as t → ∞ . Hence we have shown that the
asymptotic law of t−1 ωj under µkε (dγ) is the same as the asymptotic law under
γ[0,t]
µ(dg) of Z
(1 + a2)k
√ 1{j=0} + t−1 0
1/2 ωj .
1 + a2 k 2 g 0, 1−k 2
1+a2 k 2
t
Finally the result is valid jointly for the terms with singular forms ωj and regular forms ω̃l ,
since for each the negligible contributions vanish in probability. It remains only to replace
−1/2
1−k 2
t by 1+a 2 k2 t.
Set then :
X Z
vol(Br ) 2π (ch r − 1)
ϕR := 1γ TR (r) = ϕ0 ◦ ΘR , so that ϕR dµΓ = = ∈ ]0, ∞[.
covol(Γ) covol(Γ)
γ∈Γ
or equivalently :
Z Z Z
2π |a| covol(Γ) Γ
vol(Br ) × f dµ = vol(Br ) × f × ϕR dµ −→ f dµΓ .
TR (r)
Therefore, for any r > 0, the probability law µΓ on Γ\G is the weak limit, as R → ∞,
2π |a|
of the projection on Γ\G of the normalized restrictions vol(B 1
r ) TR (r)
µ of µ to the large
shells TR (r).
Suppose then that the function f on Γ\G is continuous and compactly supported, and
then uniformly continuous as a (left Γ-invariant) function on G = P SL2 (R).
Then, it is easily seen (and verified by a standard computation) that the maximal angular
deviation in a section of diameter r of a large thin shell TR(r), between geodesics arriving
2Rr
from T 1Br , is equivalent to R−1 , as r & 0 and R ≥ 2 . This implies that
ξ ∈ T 1Br , % ∈ P SO(2) , dist H2 [π(ξΘR ), π(%ΘR )] < r ⇒ dist P SL2 (R) (ξΘR , %ΘR ) = O(r) .
is arbitrary small for r fixed so small that the first term on the right hand side be small
enough, and then for large enough R . Forgetting the irrelevant choice of the Poincaré
model P2 for H2 and of its base point e0 , and projecting on Γ \ G, we can see SR as the
sphere of radius R in Γ\G , centred at Γg ∈ Γ\G .
This proves the following.
Theorem 9.1 The normalized Liouville measure µΓ on T 1(Γ\ H2 ) ≡ Γ\G is the weak
limit as R → ∞ of the uniform law on the geodesic sphere Γg P SO(2)ΘR of Γ\G having
radius R and fixed center Γg ∈ Γ\G : for any compactly supported continuous function
f on Γ\G , denoting by d% the uniform law on P SO(2), we have
Z Z
Γ
f dµ = lim f (Γg % ΘR ) d% .
R→∞ P SO(2)
Remark 9.1 This result and its proof are contained in [E-MM]. Such equidistribution re-
sult goes back to [R].
112 Jacques Franchi
For any compactly supported continuous function h on L(k, ε) (recall Definition 7.1),
applying Theorem 9.1 to h ◦ (ψεk )−1 (this is licit according to Lemma 7.1), we get :
Z Z Z
h dµkε = h ◦ (ψεk )−1 dµΓ = lim h[(ψεk )−1 (Γg % ΘR)] d% ,
L(k,ε) Γ\G R→∞ P SO(2)
Corollary 9.1 For any (k, ε) fixed in ] − 1, 1[×{±1} and any g ∈ G, the probability
measure µkε on the leaf L(k, ε) (recall Definition 7.1) is the weak limit as R → ∞ of the
uniform law on the geodesic quasi-sphere (ψεk )−1 (Γg P SO(2)ΘR).
2) Consider then a discretization of the excursions of the Brownian motion (gt) in the
√
cusps : it enters the shortened solid cusp {ỹ` > r + r} and exits {ỹ` ≥ r} for the nth time
within the interval of time say [τn` , σn` ], during which it performs an elementary winding
Z `
σn
ϕ`n = ϕ`n (r) := dx̃` (s). (23)
τn`
Depending only on planar hyperbolic Brownian motions (recall Section 5.), these elemen-
tary windings are independent, and independent from the points on the level {ỹ` = r} at
which the excursions start, and are easily (and classically) seen to have a Cauchy law, of
√
parameter r . A random number λ`t = λ`t (r) of these windings is performed till time t .
By ergodicity, we have lim λ`t /t =: %`r almost surely. Otherwise the Markov property
t→∞
implies the independence of the excursion durations {σn` − τn` | n ∈ N∗ }, so that by the law
XN
of large numbers N −1 (σn` − τn` ) goes almost surely to E(σ1` − τ1` ) as N → ∞. By an
n=1
λ`t
X Z t Z t
` `
obvious comparison between (σn − τn ) and 1{ỹ` (s)>r} ds , 1{ỹ` (s)>r+√r} ds ,
n=1 0 0
and by the ergodic theorem, we deduce that
√
V ({ỹ` > r + r}) V ({ỹ` > r})
2
≤ %`r × E(σ1` − τ1` ) ≤ .
V (Γ \ H ) V (Γ \ H2)
Let us apply now the Markov property : conditionally on the σ-field F generated
by the induced Markov chain (Zσn , Zτn+1 ), the random variables {ϕ`n , Jnq | 1 ≤ ` ≤
Asymptotic Homology of a Modular Quotient 115
We have thus to make sure that the knowledge of the value of B(Y ) modulo 1 will perturb
the law of B(Y ) only in a negligible way. For this, let us fix u ∈ R and ε > 0, and write :
X
√
−1 q B(Y )
E e −1 × 1{u<B(Y )−k<u+ε}
h √ i
−1 q B(Y ) X k∈Z
E e B(Y ) ∈ ]u, u + ε[+Z − 1 =
E 1{u<B(Y )−k<u+ε}
k∈Z
Z u+ε X √
−1/2 −1 q (x+k) −(x+k)2 /(2Y )
E (2πY ) e −1 e dx
u k∈Z
= Z u+ε X .
2
E (2πY )−1/2 e−(x+k) /(2Y ) dx
u k∈Z
n √ o √
−1 q k −k 2 /(2Y )
Then observing that sup |e − 1| × e k ∈ R ≤ |q| Y , we can replace
the Riemannian sum above by a Riemannian integral + an error term, in order to get :
X √ 2
2
√ √
e −1 q (x+k) − 1 e−(x+k) /(2Y ) = e− q Y /2 − 1 2πY + O |q| Y .
k∈Z
Hence we have :
h 2
i
√
h √ i E e− q Y /2
− 1 + O(|q|) e−|q| r − 1 + O(|q|)
E e −1 q B(Y ) B(Y ) ∈ ]u, u+ε[+Z −1 = = √ ,
1 + O(E(Y −1/2)) 1 + O(1/ r )
whence : h √ i √ √
−1 q ϕ`n
EF e = 1 − 1 + On (1)/ r |q| r , (28)
116 Jacques Franchi
where (z, z 0) 7→ Bu (z, z 0) = p(z 0, u)/p(z, u) denotes the Busemann function based at
u ∈ ∂ H2 , p denoting the Poisson kernel.
The following lemma ensures that the disintegration of the Liouville and Wiener mea-
θ
sures is simultaneous, by conditioning with respect to the end-point z∞ . A reason is that
2
the harmonic measures at ∂ H are the same for both, namely p(z, u)du .
Z 2π
dθ
Lemma 11.1 Pz := Pθz is the Wiener measure started from z, for any
0Z 2π
z ∈ Γ\ H2, and Pµ := Pθz dµΓ (z, θ) is the stationary Wiener measure on Γ\ H2.
Proof (Ztθ ) is by definition the h-process of the unconditioned Brownian motion, with
θ
h(z) = p(z, z∞ ) , p(z, u) = y/|z − u|2 still denoting the Poisson kernel.
Hence we have for any (z, θ) , any t and any Ft-measurable positive functional Ft :
Eθz [Ft ] = Ez [Bz∞
θ (z, Zt ) × Ft ] .
The forms ω 0, ω̃ being closed, we have the following expression for Jtλ :
Z Z 2π √ Z Z Z zθ √ Z Z Z zθ
dθ dx dy
λ
Jt =
H2
Γ\ 0
Eθz exp −1
t
z
h t 0
ω +
Zh
t 0
ω + −1
√
t
z
ht
ω̃ +
Zh
t
ω̃
2πcovol(Γ)y 2
.
t t
Applying the isometry fz,θ of H2 which maps g(1, 0) to g(z, θ) , we see that the law
Z zθ Z et
t √
of ω̃ under Pθz is the same as the law of ∗
fz,θ ω̃ , where et := −1 et and Zh0t
Zht Zh0
t
118 Jacques Franchi
√
is the point at which the Brownian motion (Zt0 ) started from −1 and conditioned to exit
at ∞ hits the horizontal horocycle having equation y = et .
Now (Zt0 ) is the h-process of the unconditioned Brownian motion, with
h(z) = p(z, ∞) ≡ y , so that its infinitesimal generator is 12 y −1 ∆ ◦ y = 12 ∆ + y∂y ,
Z t
2 0
√ wt +t/2
∆ denoting the Laplacian of H . Thus we have Zt = −1 e + ews +s/2 dWs ,
0
for two independent standard real Brownian motions (wt) and (Wt ).
As a consequence, using the boundedness of ω̃ , we have
Z et Z inf{s | ws +s/2=t}
∗
fz,θ ω̃ = O e−t × ews +s/2 dWs .
Zh0 0
t
The technical Brownian behavior we need now and after is given by the following.
Z inf{s | ws +s/2=t}
Lemma 11.2 As t → ∞ , e−t ews +s/2 dWs converges in law, and
0
inf{s | ws + s/2 = t} = 2t + o(tq ) almost surely, for any q ∈]1/2, 1] .
Proof Fix c ∈ R , set yt0 := ewt +t/2 , and look for a C 2 function f on R+ such that
h Z t i
Rt := exp −(c /2) (ys0 )2ds f (yt0) be a martingale. (yt0) having generator 12 y 2 ∂y2 +
2
0
y∂y , we have by Itô’s formula :
Z t Rs h i
2 /2) (yv0 )2 dv
Rt = f (1)+mart+ 12 e−(c 0 ×(ys0 )2× f 00 (ys0)+2(ys0)−1 f 0 (ys0)−c2f (ys0 ) ds ,
0
Z ztθ
As a consequence of this lemma and of the above, we see that t−1/2 ω̃ goes to 0
Zht
in Pθz -probability. This proves half of Proposition 11.1.
Z zθ
t
We have now to deal with the law of t −1
ω 0 under Pθz , or equivalently by the same
Zht
Z et
−1 ∗
reason as above for ω̃ , with the law of t fz,θ ω 0 . This cannot be handled further
Zh0
t
as above, since ω0
is unbounded. But integrating along the horizontal horocycle y = et
θ
containing et , Zht , we have the following estimate :
Z e Z Z
t ∗ 0 −t ht ws +s/2 ∗ 0 −t ht ws +s/2
fz,θ ω ≤ e e dWs × sup |fz,θ ω |(√ −1 +x)et ; |x| ≤ e e dWs ,
Zh0 0 0
t
√
−1 ,0
where again ht = ht = inf{s | ys0 = et } = inf{s | ws + s/2 = t}.
Z ht
Fix any r > 0 . Lemma 11.2 shows that the laws of e−t ews +s/2 dWs , for large
0
h Z ht i
−t
t , are tight, and then provides some R > 0 such that P e ews +s/2 dWs > R < r
0
for any large enough positive t .
We deduce from these last two estimates that
Z zθ Z e
0 0
Pθz t−1 Zh
t
ω > r =
Pt−1 Z0
t
fz,θ ω > r ≤ r + 1
∗
,
t ht t−1 sup |f ∗ ω 0 | √ |x|≤R >r/R
z,θ ( −1 +x)et
ν∞ (Γ)−1 ν∞ (Γ) 2g(Γ)
√ X X X
= Λ(λ0, λ) := E exp −1 λ0j rj` Q` + λl N l .
j=0 `=1 l=1
Now Lemma 11.2 asserts that the time-change ht = hz,θ t appearing in the expression
λ0 ,λ θ
of Kt in Proposition 11.1, satisfies ht = 2t + o(t) Pz -almost surely, √
uniformly with
θ z,θ −1 ,0
respect to (z, θ) . Indeed, the law under Pz of ht equals the law of ht in Lemma
11.2. So that, with arbitrary large probability, we can write ht = 2t + o(t) with a uniform
deterministic o(t) . This allows to replace t by ht in the formula of Corollary 11.1 above,
getting then (using also the definition of Pµ in Lemma 11.1) :
Z Z !
0
√ h√ √ i
λ /2,λ/ 2 −1
lim Kt = lim Eµ exp 2t ω 0 + √−1 2t
ω̃ = Λ(λ0, λ).
t→∞ t→∞ Z[0,ht ] Z[0,ht ]
This concludes the proof, since by the definition of Λ in Corollary 11.1, by the very
definition (29) of Jtλ , and by Lemma 8.1, this formula is equivalent to Theorem 8.1.
References
[D] Dieudonné J. Éléments d’Analyse 9. Gauthier-Villars, Paris, 1982.
Asymptotic Homology of a Modular Quotient 121
[E-F-LJ2] Enriquez N. , Franchi J. , Le Jan Y. Central limit theorem for the geodesic
flow associated with a Kleinian group, case δ > d/2. J. Math. Pures Appl. 80, 2,
153-175, 2001.
[F1] Franchi J. Asymptotic singular windings of ergodic diffusions. Stoch. Proc. and
their Appl., vol. 62, 277-298, 1996.
[F3] Franchi J. Asymptotic windings over the trefoil knot. Revista Mat. Iberoam., vol. 21,
no 3, 729-770, 2005.
[H] Hopf E. Ergodicity theory and the geodesic flow on a surface of constant negative
curvature. Bull. Amer. Math. Soc. 77, 863-877, 1971.
[L] Lehner J. Discontinuous groups and automorphic functions. Amer. Math. Soc, math.
surveys no VIII, Providence, 1964.
[LJ1] Le Jan Y. Sur l’enroulement géodésique des surfaces de Riemann. C.R.A.S. Paris,
vol 314, Série I, 763-765, 1992.
[LJ2] Le Jan Y. The central limit theorem for the geodesic flow on non compact manifolds
of constant negative curvature. Duke Math. J. (1) 74, 159-175, 1994.
[R] Randol B. The behavior under rojection of dilating sets in a covering space. Trans.
Amer. Math. Soc. 285, 855-859, 1984.
[R-Y] Revuz D. , Yor M. Continuous martingales and Brownian motion. Springer, 1999.
[SL-M] de Sam Lazaro J. , Meyer P.A. Questions de théorie des flots. Sém. Probab. IX,
Lect. Notes no 465, P.A. Meyer editor, Springer 1975.
122 Jacques Franchi
[T] Thurston W.P. Three dimensional manifolds, Kleinian groups and hyperbolic geome-
try. Bull. Amer. Math. Soc. 6, 357-381, 1982.
Chapter 4
Abstract
It is well known [4, 9] that if a system of differential equations admits the Lie
group of point transformations (point symmetry), then any solution of the system is
transformed to a solution of this system. This property permits the construction of
new solutions without integrating the given system of partial differential equations
(PDEs), by means of group transformations alone under known solutions. This is an
effective method if we have a sufficiently rich group of point transformations.
By applying point transformations to exact solution, a family of so-called S-
solutions can be constructed, i.e., obtained by means of symmetries. This family of
S-solutions is dependent on the group parameter. If this parameter is equal to zero,
then we have an initial solution. This procedure is called the production [9] or repro-
duction of solutions [4].
Moreover, it is easy to show that under a group transformation characteristic curves
of the system of PDEs of the hyperbolic type are transformed to the characteristics
curves. The evolution of characteristic curves permits to find out the boundary condi-
tions for new S-solutions.
In the present chapter authors will show some applications of this procedure for the
system of the theory of ideal plane plasticity, developing results obtained in [12]. In
particular, we shall use an infinite subgroup of the group of symmetries for deforma-
tion of characteristics curves of the considered hyperbolic system of PDEs to construct
a new analytical solutions. From the system of PDEs an automorphic system will be
deduced, which permits find out some relations between different solutions by means
of group transformations.
1. Introduction
Symmetry theory or the group analysis is of a fundamental importance in studying systems
of partial differential equations. At present algebras of classical infinitesimal symmetry
transformations are known for many equations of continuum mechanics, see, for example
124 Sergey I. Senashov and Alexander Yakhno
[1, 7, 9]. Methods for finding these algebras go back to S. Lie’s works written about 130
years ago. In particular, the knowledge of symmetry algebras makes it possible to construct
effectively wide classes of exact solutions for equations under consideration.
Unfortunately, the classical group analysis, being a semi-inverse method of the resolu-
tion of PDEs, does not give a direct algorithm to solve boundary value problems (BVPs).
Nevertheless, there are some results showing an invariance (or partial invariance) of BVP
with respect to admitted symmetries [3]. It is necessary to take into consideration that in
the theory of Lie groups of transformations the character of transformations is local, while
some boundary conditions are the global ones. In the case of the hyperbolic system of
quasi-linear PDEs there are theorems about the invariance of a characteristic surface under
the action of the point transformations [9]. These results are used in the present chapter to
find out some new solutions of the system of 2-dimensional ideal plasticity and to give the
appropriate boundary conditions.
The chapter is structured as follows. In the Introduction, we start with some basic
definitions and statements of the theory of Lie group of point transformations, which are
necessary for the construction of new exact solutions. In Section 2., we provide a concept
of the reproduced solution, obtained as the transformation of the known initial solution by
a symmetry and the concept of transformed characteristic curves. Section 3. contains the
information on the system of plane ideal plasticity, which will be analyzed as an example.
As a result, in Section 4. we obtain a number of exact solutions of plasticity equations and
we set for them suitable boundary conditions.
Let us consider the homogeneous system of two quasilinear equations of two indepen-
dent variables x, y and two dependent ones u1 , u2 :
∂uj ∂uj
aij (u1 , u2 ) + bij (u1 , u2 ) = 0, i, j = 1, 2. (1)
∂x ∂y
The system of such a form are widely used in the mechanics of a continuum media [10], for
example in the gas dynamics for describing isoentropic plane-symmetry flows; in the theory
of plane plasticity for the stresses of a deformed region under the different yield criterion;
for the motion of granular materials [14] and so on.
Note, that system (1) can be linearized by a so-called hodograph transformation of the
form x = x(u1 , u2 ), y = y(u1 , u2 ) when corresponding Jacobian ∆ = ∂(x, y)/∂(u1 , u2 )
is not zero. This transformation is just an interchange of roles of the unknown functions
and the independent variables. Thus, the system (1) takes the linear form:
∂x ∂x ∂y ∂y
b12 − b11 − a12 + a11 = 0,
∂u1 ∂u2 ∂u1 ∂u2
(2)
∂x ∂x ∂y ∂y
b22 − b21 − a22 + a21 = 0.
∂u1 ∂u2 ∂u1 ∂u2
We call any solution of the system (1) U = (u1 (x, y), u2 (x, y)) nonsingular one, if its
transformation to the corresponding solution χ = (x(u1 , u2 ), y(u1 , u2 )) for the linearized
system (2) is not degenerate.
Let us define the space R of the variables {x, y, u1 , u2 }, then a point transformation
Group Analysis of Solutions of Differential Equations 125
x′ = f1 (x, y, u1 , u2 , ai ), y ′ = f2 (x, y, u1 , u2 , ai ),
u′1 = g1 (x, y, u1 , u2 , ai ), u′2 = g2 (x, y, u1 , u2 , ai ), (3)
f1 |ai =0 = x, f2 |ai =0 = y, g1 |ai =0 = u1 , g2 |ai =0 = u2 ,
acting on the space R is called the symmetry of the system of partial differential equations
(1), if this system has the the same form with respect to the transformed variables. In such
a case one says that the system (1) admits the above transformation. Here ai ∈ R is some
parameter (i = 1, . . . , r).
In other words, the change of variables (3) transforms the system (1) to itself. It means,
if U = (u1 , u2 ) is a solution to (1), then function U ′ transformed by S : U → U ′ =
(u′1 , u′2 ) is another solution to the system (1), whenever U ′ is well defined [8]. Acting
admitted symmetries over the one known exact solution of the system of PDEs, we have an
opportunity to construct a family of new solutions.
If the parameters ai ∈ I ⊂ R are sufficiently small ones from an open interval I con-
taining zero, and the functions f1,2 , g1,2 are sufficiently smooth, then the transformations
(3) form the local one-parametric Lie group of the point transformations G1 with respect
to the composition of transformations for every ai . In general, we have r-parametric Lie
group Gr .
The problem of constructing a complete set of the symmetries for the given system of
PDEs was successfully solved for a lot of systems. The principal method consists in the
determination of coefficients of infinitesimal generators Xi (i = 1, . . . , r):
∂ ∂ ∂ ∂
Xi = ξi1 (x, y, u1 , u2 ) +ξi2 (x, y, u1 , u2 ) +ηi1 (x, y, u1 , u2 ) +ηi2 (x, y, u1 , u2 ) ,
∂x ∂y ∂u1 ∂u2
which form the basis of Lie algebra Lr associated with group Gr . Any operator Xi gener-
ates one-parameter group G1 and is related to the functions of transformation (3) by means
of the Lie equations:
j dfj j dgj
ξi = , η = , i = 1, . . . , r; j = 1, 2. (4)
dai ai =0 i dai ai =0
The knowledge of the Lie algebra of symmetries, admitted by the system of PDEs,
permits to construct some classes of exact solutions. To any solution U = (u1 , u2 ) of the
system (1) one can associate its orbit UG [9] as a set of all solutions obtained from U by an
application of all transformations (3) of the admitted group Gr . The solution U = (u1 , u2 )
is called H-invariant solution if u′1 = u1 , u′2 = u1 for some subgroup H of Gr , i.e. the
orbit UH coincides with U . The classification of essentially different invariant solutions
(that is those H-solutions which are not related by some transformation from Gr ) is based
on the concept of the equivalence of subalgebras of Lr with respect to a well known adjoint
representation of Lie algebra. The construction of invariant solutions is simpler due to a
less number of independent variables, of course if there exist such H-invarian t solution.
For example, for the system (1) all invariant solutions can be obtained from the system of
ordinary differential equations.
126 Sergey I. Senashov and Alexander Yakhno
2. Reproduction of Solutions
For some differential equations it is convenient to construct all H-invariant solutions firstly,
and then try to transform them by applying the rest of symmetries from Gr . Sometimes
it is the only way to obtain explicit formulas. This procedure is called reproduction or
deformation of solutions [4]. For example, some interesting results on the deformation of
invariant solutions for Kadomtsev – Pogutse equation were obtained in Ref. [5].
Formally, if we have the symmetry
S ∈/ H and the initial H-invariant solution U 0 =
0 0
u1 = u1 (x, y), u2 = u2 (x, y) , then acting by (3) on U 0 we obtain an implicit formula for
the unknown functions u1 and u2 :
Formula (5) gives the family of solutions that depends on the group parameter ai . Let us call
this family S-solution, that is obtained from the initial solution U 0 by means of symmetry
S. If parameter ai is equal to zero, then S-solution coincides with the initial one.
As it was noted in Ref. [4] when reproducing a regular solution, even for a small ai ,
we can obtain a multivalued solution or the solutions with singularities. Such generalized
S-solutions are widely used in the analysis of discontinuity propagation, shock waves, etc.
The physical meaning of these solutions should be determined by an appropriate context.
Now consider linear system (2). There is a lot of results on applying group analysis
to the linear differential equations. The relation between the symmetries of the non-linear
system of PDEs and the symmetries of linearized PDEs can be found, for example in Ref.
[3].
The linear system always admits the infinite-dimensional symmetry group since we can
add any other solution to a given solution. The corresponding infinitesimal generator has
the form
∂ ∂
X = ξ(u1 , u2 ) + η(u1 , u2 ) , (6)
∂x ∂y
where (ξ, η) is an arbitrary solution of the system (2). This operator generates a one-
parameter group of transformations of the form:
that, of course, gives the solution of system (2) as a linear combination of two solutions.
But formulas (8) define the S-solution of the form (u1 (x, y, a), u2 (x, y, a)) implicitly. Note
that if a = 1, then the S-solution coincides with solution U 1 ; and if a = 0, then S-solution
is equal to U 2 .
System (1) is an automorphic one with respect to group (7). This means that any non-
singular solution of the system of PDEs under consideration can be transformed to another
nonsingular solution of the same system (1) by means of admitted group of point transfor-
mations. This fact permits to relate any two solutions U 1 , U 2 of the quasilinear system (1)
which can be presented in the form χ1 , χ2 .
Let the quasilinear system of PDEs (1) be a hyperbolic one. Then, it has two families
of real characteristic curves defined by the following equations:
dy dy
= A(u1 , u2 ), = B(u1 , u2 ),
dx dx
where functions A and B are defined by the coefficients of the equations of the system (1)
and depend on solution.
Under the action of admitted symmetry (3) the above relations define a family of char-
acteristic curves:
dy ′ ′ ′ dy ′
= A(u 1 , u 2 ), = B(u′1 , u′2 )
dx′ dx′
for the system (1) in therms of transformed variables. Therefore, the characteristic curves
are transformed to the characteristic curves. Here the invariance of the equation of char-
acteristic curves is considered as the invariance ”on any solution” of the system of PDEs
[9].
If system (1) is a hyperbolic one, the same is valid for the linear system (2). The
characteristic curves of the linear system do not depend on the solution. Moreover, the
characteristic curves of the system of two independent variables are the plane curves (while
a solution is a space surface), therefore it is easier to analyze the action of the symmetries on
that kind of curves. It is convenient to observe, changing the value of the group parameter,
the evolution of the characteristic curves under the action of the symmetries to look for both
suitable boundary conditions and mechanical sense of a corresponding S-solution.
where σx , σy , τxy are components of a stress tensor, and k is a constant of plasticity. System
(9) describes the stress state of material, which is being plastically deformed.
128 Sergey I. Senashov and Alexander Yakhno
σx = σ − k sin 2θ,
σy = σ + k sin 2θ,
τxy = k cos 2θ,
where σ is the hydrostatic pressure, and θ + π/4 is the angle between the first principal
direction of a stress tensor and the ox-axis.
System (10) is a hyperbolic one and it has two families of characteristic curves (labeled
by parameters α and β) given by following relations:
dy σ
= tan θ, − θ = const = α,
dx 2k (11)
dy σ
= − cot θ, + θ = const = β.
dx 2k
In mathematical theory of plasticity these curves are known as slip-lines.
By means of applying hodograph transformation of the form x = x(σ, θ), y = y(σ, θ)
to the system (10) one can obtain the corresponding linearized system:
∂x ∂x ∂y
− 2k cos 2θ + sin 2θ = 0,
∂θ ∂σ ∂σ
(12)
∂y ∂x ∂y
− 2k sin 2θ − cos 2θ = 0.
∂θ ∂σ ∂σ
Passing in (12) to the new dependent variables u, v, originally due to S.G. Mikhlin:
x = u cos θ − v sin θ,
(13)
y = u sin θ + v cos θ,
∂u v ∂v u
+ = 0, + = 0. (14)
∂α 2 ∂β 2
Group Analysis of Solutions of Differential Equations 129
The classical system of the plane ideal plasticity has been investigated for many years,
but it has few exact solutions: a) Prandtl’s solution; b) the solution for a cavity of circu-
lar form, stressed by uniform pressure; c) Nadai’s solution for the stresses in the plastic
region around a circular cavity loaded by a constant shear stress in addition to a uniform
pressure; d) solution for the channel with straight line borders and e) the spiral-symmetrical
solution for the channel with logarithmic spiral borders [2]. The analytical solutions for
some boundary problems were constructed in Ref. [13]. All these solutions are widely used
for testing numerical calculations, allowing the estimation of an assurance factor of some
constructions, etc. Let us consider the simplest solutions only, namely a) and b).
1. The solution of L. Prandtl can be interpreted as a solution to describe stresses of a
rectangular block of plastic-rigid material compressed between rigid parallel plates which
are assumed to be rough. It is supposed that the block is very wide compared with its height.
In therms of the variables σ, θ for the system (10) this solution has the form:
r
x y2
σ = −p1 − k + k 1 − 2 ,
h h (15)
y = h cos 2θ,
where 2h = const is a height of the block, that is the straight lines y = ±h are the edges of
the plates, p1 = const is a value of the pressure on the plate when x = 0. The corresponding
boundary conditions look as:
x
θ|y=h = πn, n ∈ Z, σ|y=h = −p1 − k . (16)
h
For Prandtl’s solution the families of characteristics curves are cycloids. Its parametric
equations have the following form:
p1
x = h(∓2θ − sin 2θ) − h 2Ci + ,
k (17)
y = h cos 2θ, i = 1, 2,
As for the corresponding slip-lines, here we have a logarithmic spirals of the form:
π p2 − k
φ = θ − , r = R exp ±θ + + Ci , (20)
4 2k
here C1 = α, C2 = β.
It is known [11] that system (10) admits an infinite algebra of generalized (highest)
symmetries. Its Lie algebra L of point transformations is formed by the following genera-
tors:
∂ ∂ ∂ ∂ ∂ ∂
X1 = x + y , X2 = −y +x + , X3 = ,
∂x ∂y ∂x ∂y ∂θ ∂σ
∂ ∂ ∂ σ ∂
X4 = ξ1 (x, y, σ, θ) + ξ2 (x, y, σ, θ) − 4kθ − , (21)
∂x ∂y ∂σ k ∂θ
∂ ∂
X5 = ξ(σ, θ) + η(σ, θ) ,
∂x ∂y
where
σ σ
ξ1 = x cos 2θ + y sin 2θ + y , ξ2 = x sin 2θ − y cos 2θ − x ,
k k
and (ξ, η) is an arbitrary solution of the linear system (12).
The list of non-zero commutators of Lie algebra L is:
1
[X1 , X5 ] = −X5 , [X2 , X4 ] = −4kX3 , [X3 , X4 ] = − X2 ,
k
(2) ∂ (2) ∂ (3) ∂ (3) ∂
[X2 , X5 ] = ξ +η , [X3 , X5 ] = ξ +η , (22)
∂x ∂y ∂x ∂y
∂ ∂
[X4 , X5 ] = −ξ (4) − η (4) ,
∂x ∂y
where the coefficients of the three last generators have the following form:
∂ξ ∂η
ξ (2) = + η, η (2) = − ξ,
∂θ ∂θ
∂ξ (3) ∂η
ξ (3) = , η = , (23)
∂σ ∂σ
∂ξ σ ∂η σ
ξ (4) = ξ1 (ξ, η, σ, θ) + 4kθ + ξ , η (4) = ξ2 (ξ, η, σ, θ) + 4kθ +η .
∂σ k ∂σ k
Note, that operator X5 forms an infinite ideal of algebra L, because all commutators
with X5 give particular cases of generator X5 , that is all coefficients ξ (i) , η (i) (i = 2, 3, 4)
from (23) are solutions of system (12) due to the closure of L.
The groups of point transformations, which correspond to any generator of (21), convert
the system (10) to itself. These groups are well known for the generators of L, with the
exception of X4 :
x′ = ea1 x, y ′ = ea1 y;
Group Analysis of Solutions of Differential Equations 131
4. Transformation of Solutions
1. As the initial solution U 0 let us take the Prandtl’s solution (15), which may be shown
to be an invariant solution with respect to subalgebra X3 + γX5 where the operator X5
has coefficients ξ = 1, η = 0. Indeed, acting by corresponding group of transformations
on (15), we shall obtain just another value of an arbitrary constant p1 . The scale transfor-
mations corresponding to the operator X1 just change constant h. The application of the
transformations of rotation X2 does not produce any significant result from the mechanical
point of view, we just obtain rotated parallel plates.
Let us consider the action of quasi-scale transformations (25). In the therms of the new
variables x′ , y ′ , σ ′ , θ′ the Prandtl’s solution has the same form:
s
x′
y′ 2
σ ′ = −p1 − k + k 1 − 2 ,
h h (27)
y ′ = h cos 2θ′ .
132 Sergey I. Senashov and Alexander Yakhno
σ
Let us fix the first family of characteristic curves α = − θ = C1 in (17) (see Fig. 1),
2k
then from (25) we have:
σ′ σ
− θ′ = − θ e2a4 = C1 e2a4 ,
2k 2k
and the equation of characteristic curves for new variables takes the form:
p1
x′ = −h(2θ′ + sin 2θ′ ) − h 2C1 e2a4 + ,
k (28)
y ′ = h cos 2θ′ .
σ
In a similar way one can obtain the S-curve of the second family. Fixing = C2 − θ,
2k
we have
θ′ = θe2a4 − C2 sinh 2a4
Group Analysis of Solutions of Differential Equations 133
(30)
As for mechanical sense of the field of characteristic curves (29), (30) let us note, that
the transformed solution (27) with respect to transformed variables has the same interpre-
tation as the initial solution (15), and the boundary conditions look like this:
x′
θ′ y′ =h = πn, n ∈ Z, σ ′ y′ =h = −p1 − k .
h
Moreover, the boundary curve y ′ = h will be the envelope for the family of S-curves (29)
(in the same way the curve y ′ = −h will be the envelope for family (30)).
Let us take θ′ = 0, then from (25) we obtain:
x′ = uea4 , y ′ = ve−a4 ,
σ ′ = σ cosh 2a4 − 2kθ sinh 2a4 , (31)
σ sinh 2a4 = 2kθ cosh 2a4 .
and for the variable x, y we have the parametric equation of the border curve:
σ σ
h σ
x=− + p1 e−a4 cos tanh 2a4 − hea4 sin tanh 2a4 ,
k cosh 2a4 2k 2k
σ σ (32)
h σ
y=− + p1 e−a4 sin tanh 2a4 + hea4 cos tanh 2a4 .
k cosh 2a4 2k 2k
Finally, we can conclude that the field of the characteristic curves (29), (30) describes the
plastic state of the block compressed between rigid plates of the form (32) for the fixed
value of the group parameter a4 (see Fig. 2).
2. Now, let us take functions σ and θ from (18) as the initial solution U 0 . It may be
shown to be an invariant solution with respect to subalgebra X1 + γX3 . Indeed, acting
by corresponding group of transformations on (18), we shall obtain just other values of
arbitrary constants p2 and R. The application of the rotation transformations X2 does not
produce any significant result from the mechanical point of view, we just obtain rotated
circular cavity.
Let us consider the action of quasi-scale transformations (25). In the therms of the new
variables x′ , y ′ , σ ′ , θ′ the circular solution has the same form:
y′ π
θ′ = arctan ′
+ ,
x 4
′2 + y′ 2
(33)
x
σ ′ = −p2 + k + k ln .
R2
It is easy to see that
v
y′ uea4 sin θ′ + ve−a4 cos θ′ tan θ′ + e−2a4
= a4 = u ′
x′ ue cos θ′ − ve−a4 sin θ′ ′
v −2a4 = tan(θ + δ),
1 − tan θ e
u
v
where tan δ = e−2a4 . Then from the first relation of (33) we obtain δ = − π4 , so
u
v = −ue2a4 . (34)
From (25), passing to polar coordinates and taking into account (34) and (35) we have:
2r2 e2a4 cos2 arctan e2a4
σ cosh 2a4 = −p2 + k + k ln + 2kθ sinh 2a4
R2
and after some simplifications we have the explicit formula for the function σ:
−p2 + k k r2
σ= + 2k tanh 2a4 (φ + arctan e2a4 ) + ln 2 . (36)
cosh 2a4 cosh 2a4 R cosh 2a4
Finally the S-solution has the form (35), (36). Note, that if a4 = 0, then the S-solution
converts to the initial one (18).
The S-solution (35), (36) can be interpreted in the following way. For the transformed
variables we have the boundary conditions similar to (19):
π
θ′ r′ =R = φ′ + ,
4
σ′ ′
r =R
= −p2 + k.
but we have φ − θ = arctan e2a4 along this curve, so the boundary line for the S-solution
is the circumference of the form r2 = R2 cosh 2a4 .
Finally, the S-solution (35), (36) satisfies the following boundary conditions:
Figure 3. Initial slip-lines (logarithmic spirals) for the solution for circular cavity.
If we pass to the variables α and β (11), then the relations (37) define the parametric
equations for the deformed characteristics curves. For example, if we take σ = 2k(α + θ),
then the first family of characteristic curves is given by the following equations:
p1 p2 −k
π α+θ
x = −ah 2(α + θ) + + sin 2θ + (1 − a)Re 2k cos θ − e ,
k 4 (40)
p2 −k π α+θ
y = ah cos 2θ + (1 − a)Re 2k sin θ − e .
4
In Fig. 3 one can see two families of characteristic curves (20) for the solution (18) with
p2 = k for the circular cavity of the radius R = 2. The deformed slip-lines are presented in
Fig. 4 for a limacon of Pascal (h = 1, p1 = p2 ).
Group Analysis of Solutions of Differential Equations 137
5. Conclusion
The chapter is dealing with some applications of group-theoretical methods to the resolution
of a hyperbolic system of quasilinear homogeneous PDEs of two independent variables.
The principal results consist in using the action of Lie group of point transformation not
only over the set of known solutions, but over the families of characteristic curves too. This
point of view permits to find out efficiently the suitable boundary conditions for reproduced
solutions. Moreover, using of the infinite ideal of admitted Lie algebra of symmetries, as
it was shown, permits to construct a new solutions. There are some examples given for the
system of the mathematical theory of the plane plasticity. It is necessary to understand quite
clearly, that the variation of a group parameter as a parameter of the family of S-solution
should be sufficiently small to obtain physically meaningful solutions.
References
[1] W. F. Ames, R. L. Anderson, V. A. Dorodnitsyn, E. V. Ferapontov, R. K. Gazizov,
N. H. Ibragimov, and S. R. Svirshchevskiı̆, CRC handbook of Lie group analysis of
differential equations. Vol. 1, CRC Press, Boca Raton, FL, 1994. Symmetries, exact
solutions and conservation laws.
[2] B. D. Annin, V. O. Bytev, and S. I. Senashov, Gruppovye svoistva uravnenii uprugosti
i plastichnosti, “Nauka” Sibirsk. Otdel., Novosibirsk, 1985 (Russian).
[3] George W. Bluman and Sukeyuki Kumei, Symmetries and differential equations, Ap-
plied Mathematical Sciences, vol. 81, Springer-Verlag, New York, 1989.
[4] A. V. Bocharov, V. N. Chetverikov, S. V. Duzhin, N. G. Khor′kova, I. S. Krasil′shchik,
A. V. Samokhin, Yu. N. Torkhov, A. M. Verbovetsky, and A. M. Vinogradov, Sym-
metries and conservation laws for differential equations of mathematical physics,
138 Sergey I. Senashov and Alexander Yakhno
Chapter 5
T HE M ODULE S TRUCTURE
OF THE I NFINITE -D IMENSIONAL L IE A LGEBRA
ATTACHED TO A V ECTOR F IELD
Guan Keying∗
Beijing Jiaotong University
Beijing, 100044, P.R.China
Abstract
Based on a generalized definition on the admittance of a Lie group by a vector
field, it is proved that, attached to any given smooth vector field X on a n-dimensional
manifold M, there is an infinite-dimensional Lie algebra L(X) formed by infinitesimal
generators of all one-parameter Lie groups admitted by X. As a compound module,
through its any given basis (X, V1 , V2 , ..., Vn−1 ), L(X) can be treated as a direct
sum of two modules
1. Introduction
In the second half of the 19th century Sophus Lie introduced systematically the continuous
groups, now known as Lie groups, in order to create a theory of integrating ordinary dif-
ferential equations similar to Galois theory and Abel’s related works on solving algebraic
equations. Since Lie group theory gives fundamental and interlinked rules in and between
∗
E-mail address: kyguan@yahoo.com or kyguan@bjtu.edu.cn
140 Guan Keying
the basic mathematical sciences–algebra, geometry and analysis, it has had a profound im-
pact on almost all areas of mathematics and mathematically-based science.
Roughly speaking, an r-parameter Lie group is a group which is also an r-dimensional
differentiable manifold, with the property that the group operations are compatible with
the smooth structure. In the practical application of Lie group theory to the differential
equations, the applied Lie group is usually locally defined and with complicated form. One
of the main method in the research is the infinitesimal method given by Lie, which makes
it possible to reduce the study of complicated Lie group G largely to the study of a purely
algebraic object, a Lie algebra G. G is an r-dimensional linear vector space spanned by the
r infinitesimal generators of G. Besides the ordinary operation on a linear vector space,
the operation of Lie bracket between any two elements is closed in G. To every Lie group
corresponds a Lie algebra. The particular property of a Lie group and its corresponding Lie
algebra is determined by a group of constants, i.e, the structure constants of the Lie group.
According to the definition commonly used, a Lie group G is called a symmetry group
of a system of differential equation if the system is invariant under the action of the group,
or more precisely, this Lie group is said to be admitted by the system.
The main results of the modern Lie group theory on ordinary differential equations can
be represented as follows (ref. [1]):
In the case of ordinary differential equations, invariance under a one-parameter sym-
metry group implies that we can reduce the order of the equation by one, recovering the
solutions to the original equation from those of the reduced equation by a single quadra-
ture. For a single first order equation, this method provides an explicit formula for the
general solution. Multi-parameter symmetry groups engender further reductions in order,
but, unless the group itself satisfies an additional “solvability” requirement, we may not be
able to recover the solutions to the original equation from those of the reduced equation by
quadratures alone.
Suppose du/dx = F (x, u) is a system of q first order ordinary differential equations,
and suppose G is an r-parameter solvable group of symmetries, acting regularly with r-
dimensional orbits. Then the solutions u=f(x) can be found by quadrature from the solutions
of a reduced system dw/dy = H(y, w) of q-r first order equations. In particular, if the
original system is invariant under a q-parameter solvable group, its general solution can
be found by quadratures alone.
No doubt, these results are very important and elegant in both theory and application.
However, the requirement that a given system of ordinary differential equations admits a
“multi-parameter group”, i.e., the system is invariant under the “multi-parameter group”
action, is too strict somehow. In the theory of Lie group, it is well known that every
one-parameter Lie group corresponds uniquely to an infinitesimal transformation, which
is represented by an infinitesimal generator. If r different one-parameter groups are given
through r corresponding different infinitesimal generators, but the r given one-parameter
groups, as a whole, may not form a r-parameter group, unless the r infinitesimal generators
can form a basis of Lie algebra. Sometimes, for a given system of ordinary differential
equations, it is easier to to seek, say r, individual one-parameter groups admitted, than to
seek a r-parameter group admitted. Besides, for a given system of nonlinear ordinary dif-
ferential equation, if a solvable multiple-parameter Lie group admitted by it is known, it is
The Module Structure of the Infinite-Dimensional Lie Algebra... 141
usually still a very difficult task to reduce the system or to seek its first integrals.
In the study of integrating an autonomous system of ordinary differential equations
through its r independent one-parameter Lie groups, which may not form a r-parameter
Lie algebra, we noticed the following facts:
I. It is more convenient to accept a generalized definition on the admittance of a Lie
group by a vector field X or by its corresponding autonomous system of ordinary differ-
ential equations, that is to require only the integral curves of the vector field to be mapped
into integral curves of the same vector field instead of requiring the autonomous system is
invariant under the action of the Lie group.
By this definition, we may prove that a Lie group G is admitted by a given vector X if
and only if there is a C ∞ function a(x), such that
where V is the infinitesimal generator of G. This criterion does not need the help of the
prolongation of the Lie group, which is necessary in the traditional theory.
II. The infinitesimal generators of all one-parameter Lie groups admitted by a given
n-dimensional vector X form a compound module L(X), i.e., through its any basis
(X, V1 , V2 , ..., Vn−1 ), L(X) can be decomposed as a direct sum of two modules,
where L<X> is spanned by the basis (X) and is a module of rank 1 over the coeffi-
cient ring formed by all C ∞ functions, and L<V1 ,V2 ,...,Vn−1 > is spanned by the basis
(V1 , V2 , ..., Vn−1 ) and is a module of rank (n-1) over the coefficient ring formed by all
first integrals of the autonomous system determined by X. An element of L<X> is an in-
finitesimal generator of a one-parameter Lie group admitted trivially by X, and an element
of L<V1 ,V2 ,...,Vn−1 > is an infinitesimal generators of one-parameter Lie groups admitted
nontrivially by X. The formation of module L<V1 ,V2 ,...,Vn−1 > depends on the choice of
the basis (V1 , V2 , ..., Vn−1 ).
Therefore, for any given basis (X, V1 , V2 , ..., Vn−1 ) of L(X), any element V in L(X)
can be expanded uniquely as
n−1
X
V= a(x)X+ Ωi (x)Vi ,
i=1
where a(x) ∈ C ∞ , and Ωi (x) is a first integral of the autonomous system or a constant for
i=1,2,...,n-1.
III. In L(X), the operation of Lie bracket is closed. Especially, for any given basis
(X,V1 , V2 , ..., Vn−1 ) of L(X), the Lie bracket between Vi and Vj is expanded uniquely
as
n−1
X
0 k
[Vi ,Vj ]=Cij (x)X+ Cij (x)Vk .
k=1
∂ ∂ ∂
V=V1 (x) + V2 (x) + · · · + Vn (x) , (2)
∂x1 ∂x2 ∂xn
[V1 , V2 ]= − [V2 , V1 ],
It is not difficult to see that Vect(M) is of rank n. Assume that the n arbitrary given
vector fields,
Xn
∂
Vi = Vi,j (x) ∈ Vect(M), i = 1, 2, ..., n, (4)
∂xj
i=1
are linearly independent in the sense of K-module, i.e., they satisfy
V1,1 (x) V1,2 (x) ... V1,n (x)
V2,1 (x) V2,2 (x) ... V2,n (x)
. . . .
det 6= 0, ∀x ∈ D, (5)
. . . .
. . . .
Vn,1 (x) Vn,2 (x) ... Vn,n (x)
where D is an open and dense subset of M , then any element V ∈ Vect(M) can be
represented as a linear combination of the n vector fields:
n
X
V= Ai (x)Vi , (6)
i=1
where Ai (x) ∈ C ∞ , i = 1, 2, ..., n. So, this system of vector fields is a basis of the module
Vect(M). For any analytic manifold M, it is always possible for us to choose a local basis
such that, in a given local domain U in M,
1, if i = j
Vi,j (x) = ,
0, if i 6= j
this local basis can be analytically continued to an open and dense subset D of M , while
the inequality (5) is kept (note: generally speaking, the local basis can not be continued to
the whole manifold).
Therefore, we have the following conclusion:
Theorem 1. Vect(M) is a K-module of rank n.
For any given module basis (V1 , V2 , ..., Vn ) of Vect(M), and for any couple (Vi , Vj )
(i, j = 1, 2, ..., n), the Lie bracket [Vi , Vj ] can be written as a linear combination in terms
of this basis, i.e., there exist a series of functions Ckij (x)∈ K, i, j, k = 1, 2, ..., k, such that
n
X
[Vi , Vj ]= Ckij (x)Vk , i, j = 1, 2, ..., n. (7)
k=1
and
n
X
[Clij (x)Cm l m l m
lk (x) + Cjk (x)Cli (x) + Cki (x)Clj (x)]
l=1
= Vi Cm m m
jk (x)+Vj Cki (x)+Vk Cij (x), ∀ i, j, k, m = 1, 2, ..., n
144 Guan Keying
If these functions are all constant, then this basis spans a n-dimensional Lie algebra and
generates a n-parameter Lie group G (ref. [1, 3]). In this case, these constants, Ckij , are
called the structure constants of the Lie group G. With this background, these functional
coefficients, Ckij (x) ∈ K, are called the structure coefficients of Vect(M) with respect to
the basis (V1 , V2 , ..., Vn )..
It is an interesting facts that a system of n vector fields (V1 , V2 , ..., Vn ) in Vect(M)
may form a basis of a n-dimensional Lie algebra, though these vectors are linearly depen-
dent to each other. See the following example:
Example 2. Consider the 3 vector fields on R3 ,
∂ ∂
V1 = z −y ,
∂y ∂z
∂ ∂
V2 = x −z ,
∂z ∂x
∂ ∂
V3 = y −x .
∂x ∂y
They form a basis of unsolvable 3-dimensional Lie algebra which generates the rotation
group acting on R3 (ref. [6, 7]), but they are linearly dependent in the sense of K-module,
for
xV1 +yV2 +zV3 = 0.
dxi
= Xi (x), i = 1, 2, ..., n, (9)
dt
The basic theorem of the theory of ordinary differential equations guarantees that, in a
neighborhood U (⊂ M) of a regular point x of X (X does not vanish at this point), this
system has (n − 1) functionally independent local first integrals Ω1 (x), Ω2 (x), ..., Ωn−1 (x),
which satisfy
XΩi (x) = 0, i = 1, 2, ..., n − 1.
We assume that these first integrals can be continued analytically to a common domain D,
which is dense in M. Then any other first integral of (9) on D must be a compound function
in Ω1 (x), Ω2 (x), ..., Ωn−1 (x) (ref. [5]). The family of integral curves of (9) on D can be
represented locally as follows:
where
η (1) (x, y, y1 ) = ηx + (ηy − ξx )y1 − ξy y12
(2)
η (x, y, y1 , y2 ) = ηxx + (2ηxy − ξxx )y1 + (ηyy − 2ξxy )y12 − (14)
ξyy y13 + (ηy − 2ξx )y2 − 3ξy y1 y2 ,
and the detail forms of η (i) (x, y, y1 , ..., yi ), i = 3, 4, ..., can be found in [7, 1].
Similarly, a system of n first order ordinary differential equations
dxi
= fi (t, x1 , x2 , ..., xn ), i = 1, 2, ..., n,
dt
is said to admit an one-parameter Lie transformation group G∗ on the n + 1 variables
t, x1 , ..., xn , which is generated by
∂ ∂ ∂
V∗ = τ (t, x) + ξ1 (t, x) + ... + ξn (t, x) ,
∂t ∂x1 ∂xn
if the system of equations is invariant under the action of the group. Naturally, in order to
check the invariance of the system, we should use the corresponding first prolongation of
G∗ (ref. [7, 1]).
Now for the autonomous system (9), we may generalize the traditional definition on the
admittance of a Lie group by a system of ordinary differential equations as follows.
Definition 3. A one-parameter Lie group G and its infinitesimal generator
n
X ∂
V= Vi (x)
∂xi
i=1
is said to be admitted by the given n dimensional autonomous system (9) and by the cor-
responding vector field X, if the family (10) of integral curves of (9) in phase space is
invariant under the action of the transformation G.
146 Guan Keying
By the suggestive notation (ref. [7, 1]), under the action of a one-parameter Lie group
G generated by V, a point x ∈ M is mapped to
x∗ =eεV x,
Ω(x∗ ) = Ω(eεV x)
= eεV Ω(x)
= Ω(x)+εVΩ(x) + O(ε2 ),
where
∞ k
X ε
eεV = Vk .
k!
k=0
Assume a given autonomous system (9) admits the Lie group G generated by V in the
sense of definition 3. If Ω(x) is a first integral of (9), and if C is an arbitrary given integral
curve of (9), then C is mapped to the integral curve C ∗ of the same (9) under the action of
the group G, where C ∗ depends on the parameter ε. So we have
Ω(x) = c, ∀ x ∈ C,
and
Ω(x∗ ) = Ω(x)+εVΩ(x) + O(ε2 )
= c∗ (ε), ∀ x ∈ C.
where c is a constant, and c∗ (ε) depends only on the parameter ε and satisfies c∗ (0) = c. It
implies that VΩ(x) must be a constant on the integral curve C, i.e., VΩ(x) must also be a
first integral of (9).
Conversely, for a given Lie group G generated by V, if VΩ(x) is also a first integral of
(9) provided Ω(x) is a first integral of (9), then
is obviously a first integral of (9) for any small parameter ε. Clearly, under the action of
this Lie group G, an integral curve of (9) must be mapped to an integral curve of the same
(9), i.e., G is admitted by (9) in the sense of definition 3.
Therefore, we have obtained the following theorem:
Theorem 4. The generalized definition 3 is equivalent to the following analytical re-
quirement to the infinitesimal generator V of G: for any given group of (n−1) functionally
independent local first integrals
in the sense of definition 3 is equivalent to that the system of (n − 1) first order ordinary
differential equations
dxi
= Xi (x)/Xn (x), i = 1, 2, ..., n − 1, (16)
dxn
admits the same Lie group G in the traditional sense, i.e., the system (16) is invariant under
the action of group G.
It is well known that the n-th order ordinary differential equation (11) is equivalent to
the following system of n first order ordinary differential equations
dy
dx = y1
dy1
dx = y2
.. . (17)
.
dyn−1
dx = f(x, y, y1 , ..., yn−1 )
It is clear that, if the equation (11) is invariant under the action of an one-parameter Lie
transformation group G∗ generated by (12), and (13) is the infinitesimal generator of the
n-th extension of G, then the corresponding system (17) is also invariant under the action of
the Lie group G = pr(n−1) G∗ on the n + 1 variables x, y, y1 , ..., yn−1 , which is generated
by
∂ ∂
V = ξ(x, y) ∂x + η(x, y) ∂y + η (1) (x, y, y1 ) ∂y∂ 1 +
(18)
... + η (n−1) (x, y, y1 , ..., yn−1 ) ∂y∂n−1 .
Therefore, from theorem 5, we may obtain the following corollary immediately:
Corollary 6. If the n-th order ordinary differential equation (11) is invariant under the
action of the one-parameter Lie transformation group G∗ generated by (12), and (13) is the
infinitesimal generator of the n-th prolongation of G∗ , then, by the generalized definition 3,
the corresponding (n + 1) dimensional autonomous system
dx
dt = 1
dy
= y1
dydt
dt
1
= y2 , (19)
.
.
.
dyn−1 = f (x, y, y , ..., y )
dt 1 n−1
Y= a(x)X.
Proof. Assume that (20) is true. Since any first integral Ωi (x) of (9) satisfies that
XΩi (x) = 0, then we have
a(x)XΩi (x) = 0.
On the other hand,
From equality (20) we see −a(x)XVΩi (x) = 0. Hence, VΩi (x) must be a first integral
Pi (x) of (9), i.e., the equality (15) is satisfied for i = 1, 2, ..., n − 1. So the group G is
admitted by (9).
Conversely, if the group G is admitted by (9), then (15) is satisfied. This means
for any first integral Ωi (x) of (9), Therefore, by lemma 4, there must be a C ∞ function a(x)
such that (20) is satisfied.
Let f(x) be a C ∞ function. It is easy to see that if a group G is generated by V= f(x)X,
then it must be admitted by (9). This local Lie group G generated by V = f(x)X is said to
be admitted trivially by (9). Clearly, for a trivially admitted Lie group, the right hand side
of (15) must be zero for any i = 1, 2, ..., n − 1. If the right hand side of (15) is not zero for
some i, then the corresponding Lie group is said to be admitted nontrivially by (9).
Let L(X) be the set of infinitesimal generators of all one-parameter Lie groups admitted
by the autonomous system determined by X. It is easy to see that L(X) ⊂ Vect(M) and
that L(X) is a linear vector space over the number field R or C. Obviously, X∈L(X).
Theorem 9. For a given vector field X, there are at least (n − 1) infinitesimal gener-
ators,
V1 , V2 , ..., Vn−1 ,
in L(X), such that X, V1 ,V2 , ...,Vn−1 are linearly independent to each other.
Proof. Let Ωi (x), i = 1, 2, ..., n − 1 be (n − 1) functionally independent first integrals
of the autonomous system (9) determined by X. The equalities
is nonsingular.
For i = 1, 2, ..., n − 1, let the (n − 1) vectors in matrix form
MVT T
i = ei , (22)
where
i
z }| {
ei = (0, · · · , 0, 1, 0, · · · , 0),
| {z }
n
T T
Vi and ei are their transposed representations respectively.
From (22), it is easy to see that the corresponding (n − 1) vector fields in operator form,
n
X ∂
Vi = Vij (x) , i = 1, 2, ..., n − 1,
∂xj
j=1
satisfy
1, if i=j
Vi Ωj (x) = , i, j = 1, 2, ..., n − 1.
0, if i 6= j
These equalities imply that, for i = 1, 2, ..., n − 1, the equation (15) is satisfied by Vi ,
i.e., Vi is an infinitesimal generator of a one-parameter Lie group Gi admitted nontriv-
ially by (9). Obviously, these vector fields are linearly independent. Besides, for any
i = 1, 2, ..., n − 1, the first equation of the system (22), i.e.,
n
X
X · Vi = Xj (x)Vij (x) = 0, i = 1, 2, ..., n − 1,
j=1
can be explained geometrically as that the obtained (n − 1) vector V1 , ..., Vn−1 , are all
perpendicular to the vector X. Therefore, the n vectors X, V1 ,V2 , ...,Vn−1 are linearly
independent to each other.
150 Guan Keying
Let
(X, V1 , V2 , ..., Vn−1 ) (23)
be a system of n linearly independent infinitesimal generators in L(X). We should point
out here that it is not necessary to let V1 , V2 , ..., Vn−1 be perpendicular to X, though
the obtained V2 , ...,Vn−1 from (22) just be so. Clearly, this system is a basis of the
module Vect(M). It is not difficult to prove that the (n − 1) infinitesimal generators ,
V1 , V2 , ..., Vn−1 , are admitted nontrivially by X and by the corresponding autonomous
system (9). The system (23) is also called a basis admitted by X, or is called a basis of
L(X).
Theorem 10. Let (X, V1 , V2 , ..., Vn−1 ) be a basis admitted by X, then a vector field
V belongs to L(X) if and only if it can be represented as
n−1
X
V= a0 (x)X+ ai (x)Vi , (24)
i=1
where a0 (x) is a C ∞ -function, and ai (x) is a first integral of (9) or a constant for i =
1, 2, ..., n − 1.
Proof. Since (X, V1 , V2 , ..., Vn−1 ) is a basis of the module Vect(M), so the vector
field V can be represented as
n−1
X
V= a0 (x)X+ ai (x)Vi .
i=1
If the vector fields V is also admitted by X, then there is a C ∞ -function b(x) satisfying
i.e.,
n−1
X n−1
X
[ ai (x)bi (x)−Xa0 (x)]X− (Xai (x))Vi = b(x)X.
i=1 i=1
and
Xai (x) = 0, ∀x ∈ U, i = 1, 2, ..., n − 1.
Therefore ai (x) is a first integrals of (9) or a constant for i = 1, 2, ..., n − 1.
The Module Structure of the Infinite-Dimensional Lie Algebra... 151
V1 , V4 and V5 span a solvable 3-dimensional Lie algebra. But it is not easy to obtain
directly three functionally independent first integrals of (28) from this basis.
(3) V1 , V2 and X are C ∞ -linearly dependent, i.e.
p
−xy − ı z x2 + y2 + z2 z
V2 = 2 2
V1 − 2 X;
y +z y + z2
and V1 , V3 and X are C ∞ -linearly dependent, i.e.
p
−xz + ı y x2 + y2 + z2 y
V3 = 2 2
V1 + 2 X.
y +z y + z2
By theorem 10, we have obtained immediately the following two first integrals of (28):
p
−xy − ı z x2 + y2 + z2
Ω1 (x, y, z, u) = ,
y 2 + z2
and p
−xz + ı y x2 + y2 + z2
Ω2 (x, y, z, u) = .
y 2 + z2
And the fact that the system admits the generator V5 gives directly another first integral:
Ω3 (x, y, z, u) = u.
Therefore, without using V4 and without using the integral operation, we have obtained
easily three functionally independent first integrals of the system.
For a given basis (X, V1 , V2 , ..., Vn−1 ), besides the structure coefficient
Ckij (x), i, j, k = 1, 2, · · · , n − 1, we may still obtain some other first integrals by
the following theorem
The Module Structure of the Infinite-Dimensional Lie Algebra... 153
Theorem 15. For a given basis (X, V1 , V2 , ..., Vn−1 ) of L(X), let
Φ1 (x), Φ2 (x), ..., Φn−1 (x) be (n-1) known first integrals or constants, provided that at least
one of them is not zero. If these first integrals and constants satisfy
n−1
X
Vi Φj (x)−Vj Φi (x) = Ckij (x)Φk (x), ∀ i, j = 1, 2, ..., n − 1, (29)
k=1
Let
n
X ∂
Vi = Vij , i = 1, 2, ..., n − 1.
∂xj
j=1
If there is such an first integral Ω(x) of (9) satisfying (30), then for any i, j =
1, 2, ..., n − 1, the following equality should be satisfied:
n−1
X
0 k
[Vi , Vj ]Ω(x) = Cij (x)V0 Ω(x) + Cij (x)Vk Ω(x). (32)
k=1
So the equality (32) implies that the condition (29) is necessary for the first integral Ω(x)
satisfying (30).
154 Guan Keying
Now we prove that the condition (29) is sufficient for that we may obtain an first integral
Ω(x) from (30) and (31). The equation system (30) can be written as
∂Ω ∂Ω ∂Ω
V01 ∂x + V02 ∂x +···+ V0n ∂x = Φ0
V11 ∂Ω
1
∂Ω
2
∂Ω
n
Using (40) and (29), we may directly calculate [Vi , Vj ]Ω for any given
i, j, 0 ≤ i < j ≤ n − 1,
[Vi , Vj ]Ω = Vi (Vj Ω) − Vj (Vi Ω)
P ∂2Ω ∂2Ω
= 1≤p<q≤n Vijpq ( ∂xp ∂xq − ∂xq ∂xp )+
Pn−1 Pn ∂Vjl ∂Vil ∂Ω
l=0 m=1 (Vim ∂xm − Vjm ∂xm ) ∂xl
P 2
∂ Ω ∂2Ω
= 1≤p<q≤n Vijpq ( ∂xp ∂xq − ∂xq ∂xp )+
P
C0ij (x)V0 Ω + n−1 k
k=0 Cij (x)Vk Ω
(41)
P 2
∂ Ω ∂2Ω
= 1≤p<q≤n Vijpq ( ∂xp ∂xq − ∂xq ∂xp )+
Pn−1 k
C (x)Φk
Pk=0 ij ∂2Ω ∂2Ω
= 1≤p<q≤n Vijpq ( ∂xp ∂xq − ∂xq ∂xp )+
Vi Φj − Vj Φi
On the other hand, from (30) we have
[Vi , Vj ]Ω = Vi (Vj Ω) − Vj (Vi Ω) = Vi Φj − Vj Φi .
n(n−1)
Therefore, the following system of 2 equalities must be satisfied,
X ∂2Ω ∂2Ω
Vijpq ( − ) = 0, ∀ i, j, 0 ≤ i < j ≤ n − 1. (42)
∂xp ∂xq ∂xq ∂xp
1≤p<q≤n
n(n−1)
The system (42) can be treated as a system of 2 homogeneous equations in the
n(n−1)
2 unknown variables
∂2Ω ∂2Ω
αpq = − , 1≤p<q≤n
∂xp ∂xq ∂xq ∂xp
.
Let
V0112 V0113 ··· V01(n−1)n
V0212 V0213 ··· V02(n−1)n
MV = .. .. .. (43)
. . .
V(n−2)(n−1)12 V(n−1)(n−2)13 · · · V(n−2(n−2)(n−1)n
which is the coefficient matrix of the equation system (42), and let
A0112 A0212 ··· A(n−2)(n−1)12
A0113 A0213 · · · A(n−2)(n−2)13
M∗V = .. .. .. . (44)
. . .
A01(n−1)n A02((n−1)n · · · A(n−2)(n−2)(n−1)n .
n(n−1) n(n−1)
Both MV and M∗V are order 2 × 2 . From (37) we have
D 0 ··· 0
0 D ··· 0
MV M∗V = .. .. .. .
. . .
0 0 ··· D
156 Guan Keying
So, MV is a nonsingular matrix. It implies that the equalities (39) must be satisfied. There-
fore, from the non-zero solution (38) of (33) and the path integral (31) we may obtain the
first integral Ω(x) of (9).
Note: If all of Φ1 (x), Φ2 (x), ..., Φn−1 (x) are constants c1 , ..., cn−1 , then the condition
(29) can be written as
n−1
X
Ckij (x)ck = 0, ∀ i, j = 1, 2, ..., n − 1. (45)
k=1
In [9], the condition (45) was proved in case of n = 3 and all of Ckij , (i, j, k = 1, 2) are
constants. And before that, in [10] the same condition was also proved in a more particular
case: n = 3 and C112 = 1, C1,2 2 = 0.
4. Some Applications
In this section, we will consider more examples of applying the results obtained in section
3 to seek the exact travelling wave solutions for some famous nonlinear wave equation. In
this field, the second order ordinary differential equation
where ∗(2) ∂
V = V∗ + η (1) (t, x, ẋ) ∂∂ẋ + η (2) (t, x, ẋ, ẍ) ∂ẍ
(1) 2
η (t, x, ẋ) = ηt + (ηx − ξt )ẋ − ξx ẋ
(49)
η (t, x, ẋ, ẍ) = ηtt + (2ηtx − ξtt )ẋ + (ηxx − 2ξtx )ẋ2 −
(2)
ξxx ẋ3 + (ηx − 2ξt )ẍ − 3ξx ẋẍ
(ref. [7, 1]). The equation (46) corresponds the following three dimensional autonomous
system
dt
dτ = 1
dx
= y (50)
dτdy
dτ = f(t, x, y)
and the vector field
∂ ∂ ∂
X= +y + f(t, x, y) . (51)
∂t ∂x ∂y
The Module Structure of the Infinite-Dimensional Lie Algebra... 157
By corollary 6 we see that if the equation (46) admits the one-parameter Lie group G∗
generated by (47) in the traditional sense, then, in the sense of definition 3, the correspond-
ing autonomous system (50) admits the Lie group G generated by
∂ ∂ ∂
V= ξ(t, x) + η(t, x) + η (1) (t, x, ẋ) . (52)
∂t ∂x ∂ ẋ
Example 16. Consider the Burgers-KdV equation,
This equation is famous for that it has both dissipation term uxx and dispersion term uxxx
and that it is not integrable in general case. Its travelling wave u(t, x) = u(x − ct) satisfies
the following second order ordinary differential equation
where the constant c is the velocity of the travelling wave. Integrating the both sides of this
equation, we obtain
1
−cu + u2 − γuθ + βuθθ = k, (54)
2
where k is an integral constant. The equation (54) can be re-written as the following second
order autonomous system,
du
dθ = v
(55)
dv 1 1 2
dθ = β (k + cu + γv − 2 u ) = f(θ, u, v)
c2
k>− ,
2
then this system has two singular points
p p
A : (c − c2 + 2k, 0), B : (c + c2 + 2k, 0),
and there is a unique heteroclinic orbit connecting these two singular points. This hetero-
clinic orbit is just corresponding to the non-constant and bounded travelling wave solution
of (53) (ref. [11, 12]). In a particular case,
6γ 2 2
c2 + 2k = ( ) , (56)
25β
an exact particular travelling wave solution has been found out by some authors (ref.[13,
14, 15, 16]) through different ways.
Theoretically, [17, 18] proved that, if and only if the condition (56) is satisfied, then the
autonomous system (55) has an algebraic curve solution,
and (55) is Liouville integrable. The equality (57) is in fact a first order ordinary differential
equation, for v = du
dθ . Solving this equation we obtain the bounded travelling wave solution
of (53),
6γ 2 12γ 2 γ(ct−x)+d
u(x, t) = c + − (1 + e 5β )−2 . (58)
25β 25β
The related proof and calculation is somehow very complicated. We are now applying the
theory given in section 3 to the equation (54) and (55) to obtain the same fact in a different
way. According to the traditional theory (ref [7, 1]), we may find that1 , if and only if the
condition (56) is satisfied, the second order ordinary differential equation (54) admits a
two-parameter point transformation group, the corresponding two infinitesimal generators
are
∂ ∂
V∗1 = ξ1 (θ, u) ∂θ + η1 (θ, u) ∂u ,
∂ ∂ (59)
V∗2 = ξ2 (θ, u) ∂θ + η2 (θ, u) ∂u ,
where
ξ1 (θ, u) = 1, η1 (θ, u) = 0,
γθ
− 5β 2(25cβ − 25uβ + 6β 2 )γ − 5β
γθ
ξ2 (θ, u) = e , η2 (θ, u) = − 2
e .
125β
By corollary 6, we see that, corresponding to the second order ordinary differential
equation (54), the three dimensional vector field
∂ ∂ 1 1 ∂
X= +u + (k + cu + γv − u2 ) (60)
∂θ ∂u β 2 ∂v
admits the following two infinitesimal generator
∂
V1 = ,
∂τ
and γθ γθ
− 5β ∂ 2(25cβ−25uβ+6β 2 )γ − 5β ∂
V2 = e ∂θ − 125β 2 e ∂u +
γθ
(375vβ 2 +50cβγ−50uβγ+12γ 3 )γ − 5β ∂
625β 3
e ∂v .
It is easy to see that (X, V1 ,V2 ) forms a basis of the space L(X), and that
γ
[V1 , V2 ] = − V2 .
5β
So by theorem 15, through solving the system of equations
XΩ1 = 0
V1 Ω1 = 1
V2 Ω1 = 0
to obtain the three first order partial derivatives of Ω1 and calculating the path integral
Z
∂Ω1 ∂Ω1 ∂Ω1
Ω1 (θ, u, v) = dθ + du + dv,
∂θ ∂u ∂v
1
Ref: Liu Minghui and Guan Keying, The Lie group method in the study of integrability of the travelling
wave of Burgers-KdV equation, to appear.
The Module Structure of the Infinite-Dimensional Lie Algebra... 159
From this first integral, we see that the algebraic curve (57) is a solution of the autonomous
system (55), and from (57) we obtain again the bounded non-trivial travelling wave solu-
tion (58) for any parameter β, γ and for any travelling velocity c. In order to obtain the
unbounded exact travelling wave solutions in the Liouville integrable case of (56), we may
seek another first integral, which is independent of Ω1 or Φ1 through the following sched-
ule:
(1) Let
V3 = Φ1 V2
∂ 2γ(25(c−u)β+6γ 2 ) ∂ γ(375vβ 2 +50cβγ−50uβγ+12γ 3 ) ∂
= q(u, v)( ∂θ − 125β 2 ∂u + 625β 3 ∂v )
where
q(u, v) = [15625β 3 (−(c − u)3 + 3v2 β)+
37500(c − u)vβ 3 γ − 3750(c − u)2 β 2 γ 2 +
1
9000vβ 2 γ 3 + 900(c − u)βγ 4 + 216γ 6 ]− 6 .
From theorem 10 we see that V3 is also admitted by X.
(2) Notice that [V1 , V3 ] = 0 and that (X, V1 , V3 ) is also a basis of L(X). Then by
theorem 15, through solving the system of equations
XΩ2 = 0
V1 Ω2 = 0 ,
V3 Ω2 = 1
we may obtain the three partial derivatives of another first integral Ω2 of the autonomous
system determined by X,
∂Ω2
∂θ = 0
2 2 2 2 4
∂Ω2
∂u = 125β (625(c−u) β2γ−1250vβ γ−36γ ) q(u, v)5
∂Ω2 78125vβ 5
∂v = γ q(u, v)5 .
where √
2γ
τ 6γ 2 5 3β 5β
γ
µ = −e 5β , ν= + c, φ(τ ) = e τ,
25β 6γ
then (54) may be reduced to the standard form
d2 U
= 6U2 + S(T), (61)
dT2
where
625β 2 36γ 4
S(T) = − 2k + c2 − . (62)
24γ 4 T4 625β 2
It is obvious that (61) may be reduced to a Painlevé equation
d2 U
= 6U2 , (63)
dT2
if and only if the condition (56) is satisfied. The general solution of (63) is,
U = ℘(T + c1 , 0, c2 ),
where c1 and c2 are constants, and ℘(T, g2 , g3 ) is the Weierstrass elliptic function, which
gives the value of U for which
Z U 1
T= (4t3 − g2 t − g3 )− 2 dt..
∞
where γ is a real parameter. It is a reaction and diffusion equation with an unstable equilib-
rium state u = 0 and two stable equilibrium states u = − γ1 and u = 1. Its travelling wave
solution u = u(x − ct) = u(θ) satisfies the following second order nonlinear ordinary dif-
ferential equation
uθθ + cuθ + u(1 − u)(1 + γu) = 0. (66)
When c = 0, it is easy to integrate the equation by quadrature. So we shall not consider this
particular case.
Liu and Guan proved that2 the equation (65) admits a two-parameter point transforma-
tion group in the traditional sense if and only if one of the two conditions (c1 ) and (c2 ) is
satisfied, where (c1 ) requires
9
c2 = and γ = 1,
2
and (c2 ) requires
25
c2 = and γ = 0.
6
Under the condition (c1 ), the two infinitesimal generators of the two-dimensional Lie
group admitted by the second order ordinary differential equation (66) are
∂
V1 = ,
∂θ
± √1 θ ∂ 1 ± √1 θ ∂
V2 = e 2 ∓√ e 2 u .
∂θ 2 ∂u
Using the similar schedule in example 16, we obtain the first integrals of (66)
√ √
Ω1 (θ, u, v) = e±2 2θ
(−u4 + u2 ± 2 2uv + 2v2 ), (67)
which is equivalent to (66). (70) is in fact a first order ordinary differential equation, for
v = du
dθ . Solving (70), we may obtain the following exact bounded travelling wave solu-
tions:
(a) for γ = 1, c = √32 , there are two bounded travelling wave solutions
1
u(x, t) = − x−ct+d , (72)
√
1+e 2
and
1
u(x, t) = x−ct+d , (73)
√
1+e 2
1
u(x, t) = , (74)
− x−ct+d
√
1+e 2
and
1
u(x, t) = − . (75)
− x−ct+d
√
1+e 2
Under the condition (c1 ), all of the other exact travelling wave solutions are unbounded,
they can be represented by
Ω1 (θ, u, v) = d1
,
Ω2 (u, v) = d2
where d1 and d2 are arbitrary constants.
Under the condition (c2 ), the two infinitesimal generators of the two-dimensional Lie
group admitted by the second order ordinary differential equation (44) are
∂
V1 = , (76)
∂θ
and
∂
± √θ 2u ∂ u 3v ∂
V2 = e ∓√
6 [ ∓ (± + √ ) ]. (77)
∂θ 6 ∂u 3 6 ∂v
From the similar schedule, we may obtain the first integral
√
√
± 6θ u3 u2 6 v2
Ω1 (θ, u, v) = e (− + ± uv + ). (78)
3 3 3 2
the third admitted infinitesimal generator
V2
V3 = 1
(Ω1 ) 6
2u ∂ (79)
= 3 2
1√
2 1
∂
[ ∂θ ∓ √
6 ∂u
∓ (± u3 + 3v ∂
√ ) ].
6 ∂v
(− u3 + u3 ± 3
6
uv+ v2 ) 6
The Module Structure of the Infinite-Dimensional Lie Algebra... 163
gives the algebraic curve solution of the autonomous system (71) under the condition (c2 ).
From this solution, we may obtain the following exact bounded travelling wave solutions:
(c) for γ = 0, c = √56 , there is a unique bounded non-trivial travelling wave solution
!2
1
u(x, t) = x−ct+d ; (82)
√
1+e 6
(d) for γ = 0, c = − √56 , the unique bounded non-trivial travelling wave solution is
!2
1
u(x, t) = . (83)
− x−ct+d
√
1+e 6
As in the case (c2 ), the unbounded non-trivial travelling wave solutions can be given by
the two first integrals Ω1 (θ, u, v) and Ω2 (u, v).
Example 18. Consider the Klein-Gordon equation
n
∂2u X 2 ∂2u
− ai 2 = βum , (84)
∂t2 ∂xi
i=1
The second order ordinary differential equation (86) is corresponding the three dimensional
autonomous system dθ
dτ = 1
du
= v (87)
dτdv 1 m − 2(n + 1)v)
dτ = 4τ (βu
164 Guan Keying
Liu and Guan proved that3 the autonomous system (87) admits a one-parameter Lie group
generated by
∂ ∂ ∂
V1 = (m − 1)θ −u − mv . (88)
∂θ ∂u ∂v
It implies that the equation (86) is a quasi-homogeneous system, where the degrees of the
variables θ and u are (m − 1) and −1 respectively. This fact suggests that (86) may have
the following type of particular solution
1
u(θ) = Aθ− m−1 , (89)
n = 15 and m = 2, (90)
then the autonomous system (87) admits a two-parameter Lie group generated by
∂ ∂ ∂
V1 = θ −u − 2v . (91)
∂θ ∂u ∂v
and
∂ 48 ∂ ∂
V2 = θ 2 − 3(θu + ) − (3u + 5θv) . (92)
∂θ β ∂u ∂v
It is easy to see that
[V1 ,V2 ] = V2 .
Therefore, we may get a first integral
1 2
Ω1 = 6 {8 ln θ − ln(−576 − 24β + β )+
ln[−β 2 θu3 + 576v + 6β(−3u2 + 6θuv + θ2 v2 )]},
which is equivalent to Ω1 .
We are now seeking another first integral Ω2 through a schedule different to the one in
example 16 and 17. Let
Φ1 = 0, Φ2 = Ψ1 .
Since V1 Φ2 = Φ2 and V2 Φ2 = 0, we see that the condition (29) is satisfied for Φ1 and Φ2 .
According theorem 15, from
XΩ2 = 0
V1 Ω2 = Φ1 ,
V2 Ω2 = Φ2
we may obtain the following partial derivatives
and then get the first integral Ω2 through the path integral
Z
∂Ω2 ∂Ω2 ∂Ω2
Ω2 (θ, u, v) = dθ + du + dv.
∂θ ∂u ∂v
It turns out that Ω2 is not an elementary function. But from Ψ1 = 0, we may get a particular
integral manifold of the autonomous system (87)
5. Conclusion
The definition 3 has generalized the sense of the admittance of a Lie group G by an au-
tonomous system, so that the space L(X) attached to a given vector field X can be un-
covered. The theorem 10 and corollary 11 show that L(X) has an interesting compound
module structure
L(X) = L<X> ⊕ L<V1 ,V2 ,....,Vn−1 > .
166 Guan Keying
Theorem 13 shows the significance of the structure coefficients. And theorem 15 gives a
method for seeking new first integral through a basis of L(X) and through some known
first integrals.
The examples 14, 16, 17 and 18 show that the module structure of L(X) may provide
us important means for seeking the first integrals of a given autonomous system.
These facts show that the space L(X) attached to a given vector field X and its structure
does exist naturally and intrinsically.
References
[1] Peter J. Olver. Applications of Lie Groups to Differential Equations, Springer-Verlag,
Berlin, 1986.
[4] Guan Ke-ying, Liu Sheng and Lei Jin-zhi. Lie algebra admitted by an ordinary differ-
ential equation system, Ann. of Diff. Eqs.,14(2)(1998) 131-142.
[6] J. T. Cushing. Applied Analytical Mathematics for Physical Scientists, John Wiley &
Sons, Inc. New York, 1975.
[8] Liu Sheng, Lei Jin-zhi and Guan Ke-ying, A new way on identifying symmetry group
of differential Equations, Pure and Applied Mathematics, Vol.14, No.4 (1998) 01-06.
[9] Liu Hongwei, Guan Keying, Searching for first integrals of 3-th order autonomous
system using two one-paramter Lie Groups, Acta Mathematicae Applicatae Sinica,
Vol.29, No.3 (2006) 567-573 (in Chinese).
[10] Liu Sheng, Guan Keying, A method of constructing first integrals of second order
non-autonomous systems, Acta Scientiarun Naturalium Universitatis Neimongol, Vol.
30, No.2 (1999) 135-139 (in Chinese).
[11] Jeffrey A., and Kakutani T., Weak nonlinear dispersive waves: a discussion centered
around the Korteweg-de Vries Equation, SIAM Review, Vol.14,No.4,1972, 582-643.
[12] Guan Keying, Gao Ge, Qualitative research of traveling wave solution of mixed
Burgers-KdV equation, Scientia Sinica, A, Vol.17 No.1, (1987) 64-73 (in Chinese).
[13] Xiong S L, One class of analytic solution of Burgers-KdV equation, Chinese Sci. Bull.,
Vol.34, 1158-1162.
The Module Structure of the Infinite-Dimensional Lie Algebra... 167
[15] Li Zhibin, Wang Mingliang, Exact solutions for two nonlinear equations, Advances
in Mathematics, Vol.26, No.2. April 1997, 127-132.
[16] Liu Shida and Liu Shishi, Solitory Wave and Turbulance, Shanghai Scientific and
Technical Publisher, Shanghai 1994 (in Chinese).
[17] Guan Keying, Lei Jingzhi, Integrability of second order autonomous system, Ann. of
Diff. Eqs., 2002, 18: 117-135.
[18] Gao Jixin, Lei Jinzhi and Guan Keying, Integrable condition on traveling wave so-
lutions of Burgers-KdV equation, J. Northern Jiaotong Univ. Vol.27, No.3, (2003)
38-42 (in Chinese).
[20] Feng Zhaosheng, Exact solutions in terms of elliptic functions for the Burgers-
Korteweg-de Vries equation, Wave Motion 38 (2003) 109-15
Chapter 6
Abstract
Lie group methods are new geometric numerical methods, which were proposed
to solve the Lie group differential equations on manifolds. The famous Lie group
methods are the RKMK method and the Magnus method. The Lie group methods can
preserve the numerical solutions of the differential equations on the same manifolds.
The preservation of the modulus square conserving property is very important for the
modulus conserving differential equations, which has good stability. In the article, we
applied the Lie group methods, such as the RKMK method and the Magnus method, to
the modulus conserving differential equations, such as the ferromagnet equation, the
Euler equation of the rigid body problem, the nonlinear Schrodinger equation and the
vorticity equation. Numerical results showed that Lie group methods can preserve the
modulus square conserving property of the modulus conserving differential equations
and have the same accuracy as the classical explicit Runge-Kutta methods. Lie group
methods are ideal methods for constructing the explicit square conserving schemes of
the modulus conserving differential equations.
1. Introduction
Geometric integration techniques have become increasingly popular in the modern ap-
proach to numerical analysis. In the broad sense, geometric integration refers to numerical
solution techniques for differential equations that preserve inherent geometric structures.
Geometric integrators include symplectic and multisymplectic integrators that preserve the
Hamiltonian or Poisson structure, variational integrators that utilize the variational charac-
ter of Lagrangian and canonical Hamiltonian system, conservative integrators that preserve
170 Jian-Qiang Sun, Hua Wei and Gui-Dong Dai
first integral or conservational laws, and symmetric integrators that preserve symmetries of
the system [1-5]. A geometric integrator will track solution over short time interval as well
as a stand scheme of the same order e.g. a Runge-Kutta algorithm. In the case of matrix Lie
group, the product is the usual matrix multiplication. Initial value problems for differential
equations on a matrix Lie group G can be written in the form
′
Y = A(t, Y )Y, t ≥ t0 , Y (t0 ) = Y0 , (1)
where A : R × G → g is a smooth function and Y0 ∈ G. It is well known that the solution
of Eq.(1), subject to existence, stays in the Lie group G.
Classical numerical methods such as multistep methods and explicit Runge-Kutta meth-
ods did not guarantee that the numerical solutions stay in the Lie group [9,14]. The reason
for the failure of traditional methods is that matrix Lie groups are nonlinear manifolds,
linear combinations of elements in a group need not stay in that group. Hence there is a
need for new methods which ensure that the numerical solution stays on the correct mani-
folds, thereby retaining an important structure feature of the underlying differential system.
Many recent authors such as Crouch, Grossman, Munthe-Kaas, Iserles, Marthinsen and
Zanna have devoted to the Lie group methods, which can make the numerical solution on
the same manifold [6-8, 10-12,15-19].The famous methods in the Lie group methods were
the RKMK method and the Magnus method, which were proposed by Munthe-Kaas and
Iserles. A major feature of all these methods is that the numerical solution is evaluated lo-
cally in the tangent space, which is the Lie algebra g, and the required mapping (for matrical
Lie groups) is the matrical exponential.
The main purpose of the paper is to solve the modulus conserving differential equations
by the Lie group methods. The RKMK method and the Magnus method were introduced
in section 2. In section 3, we solved the Schrödinger equation and the vorticity equation
by the RKMK method. The Ferromagnet equation and the Euler equation of the rigid body
problem were solved by the Magnus methods in section 4. At last, we obtained some
conclusions.
The main idea is to reduce Eq.(3) to the Lie-group equation (4). instead of approximating
Y (t), we seek an approximate group action Q(t) that carries Y0 to Y (t). Thus, provided we
can discrete in a Lie group, we can respect the structure of any homogeneous space acted
upon by the group in question.
Traditionally, Eq.(2) can be solved by the classical Runge-Kutta methods. The set of all
elements of gl(n; R) similar to a given matrix. Classical numerical methods such as explicit
Runge-Kutta methods and multi-step methods can not preserve the numerical solution on
the same manifolds [9]. Classical Runge-Kutta method to solve Eq.(2) is as following
P
Vk = Yn + vl=1 ak,l Kl ,
k = 1, · · · , v.
Kk = A(tn + ck h, Vk )Vk ,
Xv
Yn+1 = Yn + h bv Kl . (5)
l=1
The coefficients ak,l vk,l=1 , bl vl=1 , ck vk=1 satisfy the following left Butcher table. The right
Butcher table is the coefficients of the explicit four order Runge-Kutta method.
Runge-Kutta methods can preserve the numerical solution on the quadratic manifolds.
However the Runge-Kutta shceme,which satisfies Eq.(6), is implicit.
In general, it can not keep Y in G. However we can change the configuration space
from G to g. An appealing feature of Lie algebra is that they are vector spaces. So the
Runge-Kutta method still can be applied in approximately solving the transformed equation.
Further the smoothness of the transformation map ensures the correct order of the numerical
approximation on the homogeneous space. In theory, it can ensure Y in G [1]. The method
was well known as the RKMK method.
The idea of the RKMK methods was as follow. According to a classical result of Haus-
dorff [4,13], the solution of Eq.(2) is
{Bk }k∈Z + is Bernoulli number [6]. By Picard iteration, it can get from Eq.(10), Θ0 (t) =
O,
Z t X∞ Z
Bk t k
Θ[m+1] (t) = dexp−1
Θ[m] (ξ)
A(ξ)dξ = adΘ[m] (ξ) A(ξ)dξ, m = 0, 1, · · ·
0 k! 0
k=0
(11)
The above can be written as
Z t
[1]
Θ (t) = A(ξ1 )dξ1 ,
0
Z t Z Z
1 t ξ1
Θ[2] (t) = A(ξ1 )dξ1 − [ A(ξ2 )dξ2 , A(ξ1 )]dξ1
0 2 0 0
Z t Z ξ1 Z ξ1
1
+ [ A(ξ2 )dξ2 , [ A(ξ2 )dξ2 , A(ξ1 )]]dξ1 + · · · .
12 0 0 0
Numerical Simulations of the Nonlinear Solitary Waves 173
The Picard theorem implies that Θ(t) = limm→∞ Θ[m] (t) exists in a suitably neighbor-
hood of the origin and the above first few iterations indicate that it can be expanded as a
linear combination of terms that are composed from integrals and commutators acting
recursively on the matrix A [6,10]. This is the M agnus expansion.
∞
X
Θ(t) = Hk (t), (12)
k=0
where each Hk is a linear combination of terms that include exactly k + 1 integrals. thus
Z t Z Z
1 t ξ1
H0 (t) = A(ξ1 )dξ1 , H1 (t) = − [ A(ξ2 )dξ2 , A(ξ1 )]dξ1 ,
0 2 0 0
Z t Z ξ1 Z ξ1
1
H2 (t) = [ A(ξ2 )dξ2 , [ A(ξ2 )dξ2 , A(ξ1 )]]dξ1 +
12 0 0 0
Z Z Z
1 t ξ1 ξ2
[ [ A(ξ3 )dξ3 , A(ξ2 )]dξ2 , A(ξ1 )]dξ1 .
4 0 0 0
of A(t) in Eq.(9), and carry out the integration explicitly. As a example of this procedure,
herewith fourth-order formulae for the first four integrals in the Magnus expansion:
Z h
1
A(t1 )dt1 ≈ h(A1 + A2 ),
0 2
Z h Z t1
√
3 2
[A(t1 ), A(t2 )]dt2 dt1 ≈ − h [A1 , A2 ],
0 0 6
Z h Z t1 Z t1
√ √
3 3 33 3
[A(t1 ), [A(t2 ), A(t3 )]]dt3 dt2 dt1 ≈ h [( − )A1 − ( + )A2 , [A1 , A2 ]],
0 0 0 80 48 80 48
Z h Z t1 Z t1 √ √
3 3 3 3
[[A(t1 ), A(t2 )], A(t3 )]dt3 dt2 dt1 ≈ h3 [( + )A1 − ( − )A2 , [A1 , A2 ]].
0 0 0 80 16 80 16
where √ √
1 3 1 3
A1 = A(( − )h), A2 = A(( + )h).
2 6 2 6
Assembling the above quadrature results in a fourth-order method for the linear equation
which respects arbitrary Lie-group structure,
√ √
1 3 1 3
A1 = A(tn + ( − )h), A2 = A(tn + ( + )h),
2 6 √ 2 6
X 1 3 2 1
= h[A1 + A2 ] − h [A1 , A2 ] + h3 [A1 − A2 , [A1 , A2 ]],
2P 12 80
Yn+1 = e Yn .
If A(t, y) in place of A(t) in Eq.(9), we replace the function A(t, y) by its Lagrangian
interpolating polynomial
X v
t
Ã(t, y) = h−1 lk ( )Ak , (15)
n
k=1
where Ak = hA(ck h, Xk ), k = 1, 2, · · · , v. At each internal stage, we need to calculate
quadratures of the form
X
Kl (h) = al;j L(Aj1 , Aj2 , · · · , Ajs ), (16)
j∈Csv
where Csv is the set of all combinations of length s from the set {1, 2, 3, · · · , s},
Z Y s
ak;j = lji (ξi )dξ, (17)
S̃k i=1
where
S̃k = {ξ ∈ Rs : ξ1 ∈ [0, ck ], ξl ∈ [0, ξml ], l = 2, 3, · · · , s}.
is the polytope S̃k scaled to the [0, ck ] cube instead of the unite cube. The weight bj are
recovered by substituting ck = 1.
So the integral of the Magnus expansion is discreted, we can get the explicit three order
Magnus formula of the equations
dY
= A(t, Y )Y, Y (0) = Y0 , A(t, y) = −AT (t, y), (18)
dt
Numerical Simulations of the Nonlinear Solitary Waves 175
h h
Ỹ2 = exp( A1 )Yn , A2 = A(tn + , Ỹ2 ),
Ỹ1 = Yn , A1 = A(tn , Ỹ1 ),
2 2
Ỹ3 = exp(h(−A1 + 2A2 ))Yn , A3 = A(tn + h, Ỹ3 ),
1 2 1 h2 2 1
Θ = h( A1 + A2 + A3 ) − ( [A1 − A3 , A2 + A3 ]),
6 3 6 2 15 30
Yn+1 = exp(Θ)Yn . (19)
Lie group methods are numerical methods for equations evolving on manifolds. The nu-
merical solutions obtained by Lie-group methods evolves on the same manifolds as the an-
alytical solution. The Lie group methods can result in explicit square-conservation schemes
for the modulus conserving differential equations. i.e., schemes satisfying kY n k = kY 0 k.
k · k denotes the vector norm.We will construct the explicit square conserving schemes to
solve the modulus conserving differential equations by the RKMK method and the Magnus
method in the Lie group methods.
The following initial conditions are used. When a = 2, the initial condition is
x ∈ [−15, 15].
When a = 18, the condition is
x ∈ [−20, 20].
In Fig.1 and Fig.2, the left of Fig.1 and Fig.2 showed the modulus square errors by the
fourth order explicit RKMK method. The right of Fig.1 and Fig.2 showed the modulus
Numerical Simulations of the Nonlinear Solitary Waves 177
−12
x 10 3
10
4 2
−4
−8 0
0 500 100 150 0 10 20
t t
Figure 1. The modulus square errors of the nonlinear Schrödinger equation by the RKMK
method and the RK method with a = 2.
−12
x 10
1 4.5
0
4
−2
3
−4
2
−5
−7 1
−9 0
0 60 120 0 6 12
t t
Figure 2. The modulus square errors of the nonlinear Schrödinger equation by the RKMK
method and the RK method with a = 18.
square errors by the fourth explicit Runge-Kutta method. From Fig.1 and Fig.2, we can
see that the explicit RKMK method can preserve the modulus conserving of the nonlinear
Schrödinger equation, while the explicit Runge-Kutta method can not preserve the modulus
conserving of the nonlinear Schrödinger equation.
v = k × ∇ϕ, (31)
ξ = k · ∇ × v = ∇2 ϕ, (32)
178 Jian-Qiang Sun, Hua Wei and Gui-Dong Dai
J(ξ, ϕ) = 0, (36)
ξJ(ξ, ϕ) = 0, (37)
where the bar denotes the average over the domain, along the boundary of which ϕ is
constant. From these integral constraints, we can see that the mean vorticity ξ, ¯ the mean
¯2
square vorticity ξ , are conserved with time. Conservation of these quantities, during the
advection process, possesses the important constraints on the statistical properties of the
vorticity equations. If we wish to simulate the statistical properties, a finite difference
schemes should reflect these quadratic invariant quantities of the vorticity equations [28,29].
So we must use one of such finite difference schemes that approximate conserved these
quadratic quantities [22].
Eq.(33) can be discreted into
∂ξ
= J(ξ, ϕ), (38)
∂t
J = αJ1 + βJ2 + γJ3 , (39)
J1 = ∆x ξ∆y ϕ − ∆y ξ∆x ϕ, (40)
J2 = ∆y (ϕ∆x ξ) − ∆x (ϕ∆y ξ), (41)
J3 = ∆x (ξ∆y ϕ) − ∆y (ξ∆x ϕ). (42)
dY
= A(ϕ)Y. (43)
dt
where
αB(ϕ) + βC(ϕ) + γD(ϕ)
A(ϕ) = , (44)
4∆x∆y
When α = β, we can get αB(ϕ) + βC(ϕ) is skew symmetry matrix. We know D(ϕ)
is skew symmetry matrix. When we take α = β = 12 , γ = 0 or α = β = γ = 31 or
α = β = 0, γ = 1. We get the spacial discreted square-conservation type schemes of the
vorticity equation similar to Eq.(2).
Numerical Simulations of the Nonlinear Solitary Waves 179
−10
x 10
8 3.5
4
2
0
1
−4 0
0 30 60 0 6 12
t t
Figure 3. The modulus square errors of the 2D vorticity equation by the RKMK method
and the RK method with α = β = 21 , γ = 0.
−10
x 10 350
4
300
0
−4 200
−8
100
−12
0
−16 0 20 40
0 40 80 120
t t
Figure 4. The modulus square errors of the 2D vorticity equation by the RKMK method
and the RK method with α = β = γ = 31 .
To the equation
ξ = ∇2 ϕ = ϕxx + ϕyy , (45)
Using mid difference scheme to Eq.(45), we can get the discreted equations
Y = EX, (47)
′
where the vector Y = (ξ11 , ξ21 , · · · , ξm1 , · · · , ξ1n , ξ2n , · · · , ξmn ) , the vector X =
′
(ϕ11 , ϕ21 , · · · , ϕm1 , · · · , ϕ1n , ϕ2n , · · · , ϕmn ) , E is the matrix. The 2D vorticity equa-
tion was solved by the fourth order explicit RKMK methods and the corresponding fourth
180 Jian-Qiang Sun, Hua Wei and Gui-Dong Dai
order explicit Runge-Kutta method. The modulus square conserving error was defined as
Eq.(27). The global errors at time t = k∆t were defined as
The global errors at different times of the two methods are listed in the Table.1.
The algorithm of the 2D vorticity equation to compute ξ k , ϕk to ξ k+1 , ϕk+1 .
The initial value ξ 0 , ϕ0 , t = 0, were given
for k = 0, 1, · · · n
Step 1: Apply the RKMK method to compute Eq.(43), ξ k+1 was obtained.
Step 2 : Compute X k+1 = E −1 Y k+1 , ϕk+1 was obtained. So ξ k+1 , ϕk+1 were obtained.
end
The experimental function was given
−gcy Dg
ϕ(x, y, t) = + cos(µ(x − at) + ry), (49)
10 10
gc
where µ, r, g, c, a, D are the constants and a = 10 . According to ξ = ϕxx + ϕyy , the exact
solution of Eq.(30) was given as
Dg 2
ξ=− (µ + r2 ) cos(µ(x − at) + ry). (50)
10
In table 1, the numerical solutions of the 2D vorticity equation at different times by the
RKMK method and the Runge-Kutta method were given. Numerical results showed that
the RKMK method and the Runge-Kutta methods have the same accuracy.
Table 1. Accuracy comparison of the RKMK method and the Runge-Kutta method.
RKMK method Runge-Kutta mehod Error
t=0.5 2.43457 × 10−4 2.43466 × 10−4 10−8
t=2 1.147326 × 10−3 1.147378 × 10−3 10−8
t=6 1.953256 × 10−3 1.953278 × 10−3 10−8
In Fig.3 and Fig.4, the left of Fig.3 and Fig.4 showed the modulus square errors by the
fourth order explicit RKMK method. The right of Fig.3 and Fig.4 showed the modulus
square errors by the fourth explicit Runge-Kutta method. From Fig.3 and Fig.4, we can see
that the explicit RKMK method can preserve the modulus conserving of the 2D vorticity
equation. The explicit Runge-Kutta method can not preserve the modulus conserving of the
2D vorticity equation. It is obvious that the RKMK method has good stability.
Numerical Simulations of the Nonlinear Solitary Waves 181
Zt = λ1 Z × ∆Z − λ2 Z × (Z × ∆Z), (56)
Z(x + L, t) = Z(x, t), Z(x, 0) = Z0 (x),
where λ2 ≥ 0 is Gilbert damped component. In 1993, Guo Bo-ling studied the solution of
Eq.(56) systematically and found the closed relation of the Riemann map and the solution
of Eq.(56). If λ2 = 0, Eq.(56) has the exact solution
∂t Zj = λ1 Zj × ∆h Zj − λ2 Zj × (Zj × ∆h Zj ) (58)
Zj (x + L, t) = Zj (x, t), Zj (x, 0) = Z0 (xj ),
Y = (u1 , u2 , · · · , uN , v1 , v2 , · · · , vN , w1 , w2 , · · · , wN )T .
6 6
−2 0
0 12 24 0 1.2 2.4 3.6
t t
Figure 5. The modulus square errors of the Ferromagnet equation by the Magnus method
and the RK method with λ1 = 2.
−12 118
x 10 x 10
12 1
10
0
−2
−4
−2
−6 −6
0 40 80 0 1.25 2.5
t t
Figure 6. The modulus square errors of the Ferromagnet equation by the Magnus method
and the RK method with λ1 = 5.
0 W −V 0 W̃ −Ṽ
1 1
A(Y ) = −W 0 U , B(Y ) = − W̃ 0 Ũ ,
(∆x)2 (∆x)2
V −U 0 Ṽ −Ũ 0
where U , V ,W ,Ũ , Ṽ ,W̃ are matrices[20-23, 25]. It is obvious that dYdt = C(Y )Y is
the ordinary equations of n = 3N, where C(Y ) is skew symmetry matrix. The ordinary
equations (59) was solved by the third order explicit Magnus method and the corresponding
third order Runge-Kutta method.
In Fig.5 and Fig.6, the left of Fig.5 and Fig.6 showed the modulus square errors by
the third order explicit Magnus method. The right of Fig.5 and Fig.6 showed the modulus
square errors by the third explicit Runge-Kutta method. From Fig.5 and Fig.6, we can see
that the explicit Magnus method can preserve the modulus conserving of the Ferromagnet
equation. The explicit Runge-Kutta method can not preserve the modulus conserving of the
Ferromagnet equation. It blew up at small time step. It is obvious that the Magnus method
has good stability.
Numerical Simulations of the Nonlinear Solitary Waves 183
−15 −9
x 10 x 10
6 9
4
6
2
4
0
2
−2 0
0 60 120 0 12 24
t t
Figure 7. The modulus square errors of the Euler equation by the Magnus method and the
RK method with h = 0.001.
−15 −7
x 10 x 10
2.5
−1
−2
0
0 40 80 0 25 50
t t
Figure 8. The modulus square errors of the Euler equation by the Magnus method and the
RK method with h = 0.0025.
′
y = y × M y, t ≥ 0, y(0) = y0 , (60)
where y ∈ R3 (we assume that ky0 k = 1), the symbol × denotes the classical vector
product on R3 and M = diag(m1 , m2 , m3 ) is a diagonal matrix [32]. This system has
the Hamiltonian function H(y) = 12 (m1 y22 + m2 y22 + m3 y32 ) and obey kyk2 = 1. It can
be represented by means of Lie-group action of SO(3) on R3 by representing the solution
y(t) as Q(t)yn , n = 0, 1, 2, · · · , with Q ∈ SO(3), t ∈ [tn , tn+1 ]. Hence, in each interval
[tn , tn+1 ], we can solve the differential equation
′
Q (t) = A(y(t))Q(t), t ≥ tn , Q(tn ) = I, (61)
184 Jian-Qiang Sun, Hua Wei and Gui-Dong Dai
where
0 −m3 y3 m2 y2
A(y(t)) = − m3 y3 0 −m1 y1
−m2 y2 m1 y1 0
√ √ √ √
In this numerical experiment,m1 = −1, m2 = −( 2 − 0.51
√ 2
1.51
), m 3 = −( 2− √ 2 ),
1.51
and the initial value y(0) = [0, √12 , √12 ]T . We remark that such action automatically obeys
the homogeneous space condition kyk2 = 1, wherever a Lie-group method is applied to
Eq.(61). The ordinary equations (60) was solved by the third order explicit Magnus methods
and the corresponding third order Runge-Kutta method. The modulus square conserving
errors were defined as Eq.(27).
In Fig.7 and Fig.8, the left of Fig.7 and Fig.8 showed the modulus square errors by
the third order explicit Magnus method. The right of Fig.7 and Fig.8 showed the modulus
square errors by the third explicit RK method. From Fig.7 and Fig.8, we can see that
the explicit Magnus method can preserve the modulus conserving of the Euler equation.
The explicit Runge-Kutta method can not preserve the modulus conserving of the Euler
equation. It also proved that the Magnus method can preserve the modulus conserving
property of the differential equations.
5. Conclusion
In the article, the explicit square conserving schemes of the modulus conserving differential
equations were proposed by the Lie group methods. The modulus conserving differential
equations, such as the nonlinear Schrödinger equation, the ferromagnet equation, the 2D
vorticity equation, the Euler equation of the rigid body problem, were solved by the RKMK
method and the Magnus method in the Lie group methods. Numercal results showed that
the Lie group methods can preserve the modulus conserving property of the modulus con-
serving differential equations and have the same accuracy as the classical explicit Runge-
Kutta methods. The Lie group methods have also better stability than the classical explicit
Runge-Kutta methods. We can concluded that the Lie group methods are ideal methods for
the modulus conserving differential equations.
Acknowledgements
This work was supported by the National Natural Science Foundation of China
(No.10401033).
References
[1] P.J.Olver, Applications of Lie group to differential equations, Springer, New York,
(1986).
[2] Jerrold E.Marsden, Tudor S.Ratiu, Introduction to mechanics and symmetry, Springer-
Verlag, New York, (1994).
Numerical Simulations of the Nonlinear Solitary Waves 185
[3] Ernst Hairer, Christian Lubich, Gerhard Wanner, Geometric Numerical Integration,
Springer-Verlag, Berlin Heidelberg, (2002).
[4] K.Feng, Difference schemes for Hamiltonian formalism and symplectic geometry,
J.Comp.Math, 4, 279-289, (1986).
[5] T.J.Bridge, Multi-symplectic structures and wave propagation, Math Proc Cam Phil
Soc, 121, 147-190,(1997).
[6] A. Iserles, Hans Munthe-Kaas, Lie Group Methods, Acta Numerical, 215-365,(2000).
[7] Hans Munthe-Kaas,Runge-Kutta methods on Lie groups, BIT, 38, 92-111, (1998).
[8] Hans Munthe-Kaas, High Order Runge-Kutta Methods on Manifolds, Appl. Numer.
Math. 29, 115-127,(1999).
[9] E.Hairer, S.P.Norsett and G.Wanner, Solving Ordinary Differential Equations I. Non-
stiff Problems. Second Revised Edition, Springer, Berlin, (1993).
[11] A.Iserles, S.P.Nφrsett and A.F.Rasmussen, Time-symmetry and high order Magnus
methods, Technical Report 1998/NA06, DAMTP, University of Cambridge,(1998).
[15] E.Celledoni and A.Iserles, Approximating the exponential from a Lie algebra to a Lie
group, Technical Report 1998/NA3, DAMTP, University of Cambridge.
[16] G.H.Golub and C.F.Van Loan, Matrix Computation, 3rd edn, Johns Hopkins Univer-
sity Press, Baltimore,(1996).
[17] A.Zanna and H. Munthe-Kaas (1997), Iterated commutators, Lie’s reduction method
and ordinary differential equations on matrix Lie groups, Foundation of Computa-
tional Mathematics, (F.Cucker and M.Shub.eds), springer, 434-441., (1997).
[18] M. Calvo, A.Iserles and A.Zanna, Runge-Kutta methods for orthogonal and isospec-
tral flows, Applied Numeric. Math, 22, 153-163,(1996).
[20] Jian-Qiang Sun, Meng-Zhao Qin, Zhong-Qi Ma, Magnus methods of solving non-
damping Landau-Lifshitz equation, Chinese Journal. Numerical Mathematical and
Application, 26, 1, 57-64, (2004).
[21] Jian-Qiang Sun, Zhong-Qi Ma, Meng-Zhao Qin, RKMK method of solving non-
damping LL equations and ferromagnet chain equations, Applied Mathematics and
Computation, 157, 407-424, (2004).
[22] Jian-Qiang Sun, Meng-Zhao Qin, Zhong-Qi Ma, Lie group method of the vorticity
equations for two dimensional incompressible flow, Computational Fluid Dynamics
Journal, 12(4),70, 569-579, (2004).
[23] Sun Jian-Qiang, Ma Zhong-Qi, Qin Meng-Zhao, Explicit square conserving schemes
of Landau-Lifshitz equation with Gilbert component, Applied Mathematics and Me-
chanics, 26, 1, 73-78, (2005).
[24] Jian-Qiang Sun, Zhong-Qi Ma, Wei Hua, Meng-Zhao Qin, New conservation schemes
for the nonlinear schrödinger equation, Applied Mathematics and Computation, 177,
446-451, (2006).
[25] Jian-Qiang Sun, Xiao-Yan Gu, Zhong Qi Ma, Meng-Zhao Qin, RK-Cayley Fehlberg
method on homogeneous manifolds, Communciation in nonlinear science and numer-
ical simulation, 12, 966-975, (2007).
[26] Zhang Fei, Victor M. Pérez-Garcia and Luis Vázquez, Numerical simulation of non-
linear Schrödinger systems: A new conservative scheme, Appl. Math. Comput, 71,
2-3,165-177, (1995).
[27] Zhang Luming, Chang QianShun, A conservation numerical scheme for nonlinear
schrödinger equation, Chinese Journal of Computational Physics, 16(16), 661-668,
(1999).
[29] Arakawa., Computational Design for Long-Term Numerical Integration of the Equa-
tion of Fluid Motion: Two-Dimensional Incompressible of Flow. Part I, University of
California, Los Angles, Joural of Computational Physics, 1,119-143, (1966).
[30] J.A.G.Robert and C.J.Thompson, Dynamical of the classical Heisenberg Spin Chain,
Technique Report 11-1987, Mathematics Department, University of Melbourne,
Pakville, Victoria 3052, Australia,(1987).
[31] Jason Frank, Huang Weizhang, Benediet Leimkuler, eometric integration for classical
spin system, GComput Phys, 133(1), 160-172, (1997).
[32] McLachlan R.I, Explicit Lie-Poisson integration and the Euler equations, Phys Rev
Lett, 71, 3043-3046, (1993).
In: Lie Groups: New Research ISBN 978-1-60692-389-4
Editor: Altos B. Canterra, pp. 187-198
c 2009 Nova Science Publishers, Inc.
Chapter 7
Abstract
Key words : tranversality, singularities, stability, robotics, Lie groups, submersion with
fold.
AMS mathematics subject classifications : 70 - 53 Differential geometry; 70B15 Mech-
anisms and linkages.
∗
E-mail address: jean.lerbet@ibisc.univ-evry.fr
188 Jean Lerbet
Introduction
In previous papers ([4],[5] for examples), the regularity of products of exponential map-
pings of a Lie group has already been analysed. But this treatment aims at the resolution of
the equation exp(θ1 X1 ) . . . exp(θn Xn ) = id in the group and for the (local) description of
the set of roots in IRn . The results are directly applicable to closed mechanisms and this is
why the associated mapping was called a closure function.
In this paper we study the singularities of the mentioned exponential mapping when the set
of variables has the same dimension as the group. The applications concern open kinematic
chains as used in most robots. Indeed, the geometric model of a robot consists in the rela-
tions between the articular coordinates and the position and orientation of the end effector.
These relations are exactly the elements of closure function that shall be called here the
work function. For these applications, we may suppose that the variables θ1 , . . . , θn are in
S 1 or in a compact interval of IR because they represent the articular coordinates between
two consecutive bodies. Otherwise, the hypothesis of compactness fails and a part of the
following study too.
The analysis of singularities is carried out in two steps:
1. the first is to describe the set of ’first’ singularities (the 1-class Σ1 (f )): structure of
submanifold and tangent bundle
2. the second is to specify the nature of these first singularities: Σ1 (f ) as a set of fold
points
The last envisaged question shall be : is such a mapping stable? Indeed, stable mappings
possess some remarkable properties which ensure mechanical properties of the associated
kinematic chain. For example, the play in the joints or unavoidable imprecisions in the
realisation of the chain do not generate any problems for describing the topology of the
workspace of the robot if the associated mapping is stable because the topology itself re-
mains stable. Finally, two examples of kinematic chains are proposed for illustration and
suggest how to continue in this direction.
1. Some Preliminaries
We shall recall now some standard results about Lie groups, differential geometry and
concepts of the theory of singularities which shall be used in the follow up of the study.
Proposition
f is an analytic mapping and for each θ in S n , we have:
Because RaT is an isomorphism from Tb G onto Tba G for all a, b in G, the singularities of
n
X
f shall be studied through those of θ → h(θ)(x) = xk Yk (θ) where h(θ) ∈ L(IRn , G).
k=1
Especially, the kernel and the rank of f ′ (θ) and h(θ) are the same and Im(f ′ (θ) is
isomorphic to Im(h(θ)).
1.2. Singularities
We now recall only the main concepts which shall be useful in the following. For more
details, see for example [2] and [3].
1.3. Stability
Definition
1. Let f1 and f2 be two elements of C ∞ (M, N ). f1 is equivalent to f2 if there are
diffeomorphisms g : M → M and h : N → N such that f2 = h ◦ f1 ◦ g
2. f is infinitesimally stable or inf-stable if for all w, vector field along f , there are two
vector fields s : M → T M and t : N → T N such that:
w = f′ ◦ s + t ◦ f
Singularities and Stability of a Work Function 191
If f = f(X1 ,...,Xn ) is a work function the equation of inf-stability becomes, coming back to
the Lie algebra:
for all w : S n → G are there t : G → G and x : S n → IRn such that:
where BY (x, y) is the bilinear mapping from IRn × IRn to G defined for all sequences
Y = (Y1 , . . . , Yn ) of vectors of G by:
n X
X
BY (x, y) = xk yl [Yl , Yk ]
k=1 l<k
For details and properties about this bilinear mapping, see [4].
Two situations may occur:
and then
n
X n X
X
i
BY (x, y) = Clk xk yl Yi
i=1 k=1 l<k
192 Jean Lerbet
Obviously Z depends on Y and the choice of it is not canonical. In this case, we have
in a similar way:
n−1
X n X
X
i
BY (x, y) = Clk xk yl Yi
i=1 k=1 l<k
n X
X
Z
+ Clk xk yl Z
k=1 l<k
Proof
/ Σ1 (f ), then f ⊤Ln−1 (IRn , G) at θ by definition. Else, let θ ∈ Σ1 (f ).
If θ ∈
f ⊤Ln−1 (IRn , G) at θ iff the equation
h′ (θ)(x) + v = u
Singularities and Stability of a Work Function 193
is solvable for x ∈ IRn , v ∈ Th (θ)Ln−1 (IRn , G), u ∈ L(IRn , G). A known result (see [3],
[4] for example) gives the local description of the submanifold Ln−1 (IRn , G) of L(IRn , G)
and allows to obtain here:
With the above conventions, that means vZ(θ) (x) = 0 if v(x) = vY (x) + vZ (x) with obvi-
ously notations.
The condition of transversality becomes:
Is there x in IRn such that, for all U ∈ GY (θ) , one may find y in KY (θ) with U =
h′ (θ)(x)(y)? According to the previous calculations and identifications, this means:
find x in IRn such that:
n X
X Z(θ)
( Clk (θ)xk yl = u
k=1 l<k
u
is solvable in y ∈ KY (θ) for all u ∈ IR. Let k be such that (2) and x = bk (θ) (δik )i=1,...,n in
IRn . Then the previous equation of transversality is solved.
The calculation of Tθ Σ1 (f ) results directly from:
Remarks
Z(θ)
1. since θ ∈ Σ1 (f ), there are always k and l such that Clk 6= 0 else FY (θ) = G and
f is locally a diffeomorphism which is a contradiction with θ ∈ Σ1 (f ). But it is not
equivalent to the transversality condition.
1. θ in Σ1 (f ) is a fold point if :
KY (θ) + Tθ Σ1 (f ) = IRn
We have:
n−1
XX n−1
X
Z(θ)
αk (θ)αl (θ)Clk (θ) = αk (θ)bk (θ) 6= 0 (3)
k=1 l<k k=1
Proof
It is necessary and sufficient to prove that KY (θ) ∩Tθ Σ1 (f ) = {0} because dim(KY (θ) ) = 1
and dim(Tθ Σ1 (f )) = n − 1.
x belongs to KY (θ) ∪ Tθ Σ1 (f ) if BY (θ) (x, y) ∈ FY (θ) for all x, y are in KY (θ) . Because
of the symmetry of BY (θ) on KY (θ) × KY (θ) (see [4] for example), it is equivalent to
BY (θ) (x) ∈ FY (θ) for all x in KY (θ) .
P P Z(θ)
That means that nk=1 l<k Clk (θ)xk xl = 0 for all x in KY (θ) . But xk = −αk xn for
all k = 1, . . . , n − 1. Thus we deduce:
n−1
XX Z(θ)
x2n ( Clk (θ)αk (θ)αl (θ)) = 0
k=1 l<k
disappears because
X n
X
Z(θ) Z(θ)
Cln (θ)xn xl = xn Cln (θ)xl
l<n l=1
Pn Pn
Now l=1 Yl (θ)xl = 0 and [ l=1 Yl (θ)xl , Yn (θ)] = 0. We deduce:
n
X n
X Z(θ)
[Yl (θ)Yn (θ)]xl = Cln (θ)xl = 0
l=1 l=1
Yn (θ) = αi Yi (θ)
with i in {1, . . . , n − 1} (whereby n and i can obviously be replaced by any other pair of
indices of {1, . . . , n}).
Proof
Singularities and Stability of a Work Function 195
Indeed (3) becomes 0 6= 0. In fact it is a direct result of the antisymmetry of the Lie bracket
Z(θ) Z(θ) Z(θ)
(Clk = −Ckl and Cll = 0) and we have successively:
n−1
X n−1
XX Z(θ)
αk (θ)bk (θ) = Clk (θ)αk (θ)αl (θ))
k=1 k=1 l<k
n−1
XX Z(θ)
= Clk (θ)αk (θ)αl (θ))
k=1 l≤k
Z(θ)
= (αi (θ))2 Cii (θ)
= 0
2.3. Stability of f
We now suppose that (2) and (3) are both satisfied. Let g the restriction of f to its fold
locus Σ1 (f ). g is obviously an immersion from Σ1 (f ) to G and because S n is compact
the closed submanifold Σ1 (f ) too (it is the first time the compactness is used). Thus if
a = g(θ) then g −1 (a) = {θ1 , . . . , θp } is finite. The stability of f depends of the na-
ture of the crossings at a. More precisely, according to the theory of singularities, let us put:
Definition
1. For every set A and s > 1, define A(s) = {(a1 , . . . , as ) ∈ As | ai 6= aj for 1 ≤ i <
j ≤ s}.
2. Let φ : M → N be a smooth mapping and φ(s) : M (s) → N s the restriction of
φ × . . . × φ : M s → N s to M (s) . Let δN s = {(y, . . . , y) ∈ N s | y ∈ N }. Then φ
is a mapping with normal crossings if for every s > 1, φ(s) ⊤δN s .
and the main result which shall be used here is:(see [3])
Proposition The work function f is stable if and only if g is an immersion with normal
crossings.
It now remains to analyse the condition of normal crossings (N-C) of g. So, let
θ(s) = (θ1 , . . . , θs ) element of (S n )(s) such that f (θ1 ) = . . . = f (θs ) = w ∈ G. For other
points of (S n )(s) , the transversality g (s) ⊤δGs at θ(s) obviously holds and the (N-C) too.
Coming back to the Lie algebra G by the diffeomorphism (Rw T )−1 , the equation of
transversality becomes:
Let (h1 , . . . , hs ) be in G s . Find h in G and x1 , . . . , xs in Tθ1 Σ1 (f ) × . . . × Tθs Σ1 (f ) such
that for all i = 1, . . . , s:
n
X
hi = h + xik Yk (θi )
k=1
Because each is a fold point, we can look for xi in IRn . Moreover, this family of
θi
equations enables us to obtain:
196 Jean Lerbet
Proposition If there are s > n roots of the equation g(θ) = w, then f is not stable.
Proof
Because g is an immersion and S n compact, g −1 (w) = {θ1 , . . . , θs } is finite. The
previous equations of transversality imply that codim(FY (θ1 ) ∩ . . . ∩ FY (θs ) ) =
codim(FY (θ1 ) )+. . .+codim(FY (θs ) ) = s and obviously codim(FY (θ1 ) ∩. . .∩FY (θs ) ) ≤ n.
More precisely, we may prove that (see [3]):
Therefore, the stability of the work function strongly depends on the intersection
D = FY (θ1 ) ∩ . . . ∩ FY (θs ) . We have:
Proposition Sp{Y1 (θj ), Yn (θj )} ⊆ D for all j and if the number s of roots of the
equation g(θ) = w is such that s > n − 2 , then f is not stable.
Proof
We have only to prove the first assertion. But Ad(exp(tX))X = X for all X in G which
proves that Y1 (θj ) = X1 . Moreover Yn (θj ) = Ad(exp(g(θj )))Xn = Ad(exp(w))Xn for
all root θj of g(θ) = w which proves Yn (θi ) = Yn (θj ) for all θi and θj in g −1 (w). QED.
We obviously obtain:
Corollary For n = 3 (planar motions), the work function f is stable if and only if its
restriction g is an injective immersion.
3. Examples
Here, G is the group of Euclidean displacements D. We begin by giving the family
(X1 , ..., Xn ) of 6-vectors. The coordinates of each vector are given with respect to a basis
→ −
− → −→
of D which is deduced from a fixed coordinates frame (0; i , j , k ) of E in the following
way:
→
− →
− →
− → −→
− → −→
− → −→
−
i(m) = i , j(m) = j , k(m) = k , ξ(m) = i ∧ 0m, η(m) = j ∧ 0m, ζ(m) = k ∧ 0m
for all m in E. Thus the basis of D is (i, j, k, ξ, η, ζ). The studied singularity is 0.
0 0 0 a 0 0
0 0 1 0 −b 0
0 0 0 0 a 0
X1 = , X2 = , X3 = , X4 = , X5 = , X6 =
0 1 0 0 1 α
0 0 0 1 0 0
1 0 0 0 0 γ
γX1 + αX2 − X6 = 0
βX1 + αX4 − X6 = 0
Conclusion
A work function f of a robot or more generally of a kinematical chain is a product of ex-
ponential mappings of a Lie group. This mapping comprises all the kinematics of the chain
and the analysis of its singularities is an essential point in mechanics. We are concerned
with the case where the dimension of the group is equal to the number of variables. First
the transversality conditions (2) allowing to describe the set of singularities Σ1 (f ) (struc-
ture of manifold and tangent bundle) have been given. Secondly, we have shown that under
the condition (3) f is a submersion with fold. Finally, considering the restriction g of f
to its fold locus, normal-crossing conditions of g to analyse the stability of f is studied.
Examples are given to illustrate the method and interesting extensions may be suggested:
stronger singularities (not fold points), points of Σ2 (f ) or of Σ1,1 (f ) could be for example
considered.
References
[1] J.Bochnak M.Coste M-F.Roy Géométrie algébrique réelle Springer-Verlag (1986)
[2] V. Arnold A.Varchenko S.Goussein-Zade Singularités des applications différentiables
Mir (1986)
[3] M.Golubitsky V.Guillemin Stable Mappings and Their Singularities Springer-Verlag
(1973)
[4] J.Lerbet K.Hao Kinematics of Mechanisms to the second Order- Application to the
closed Mechanisms Acta Applicandae Mathematicae Vol. 59, pp.1-19 (1999)
[5] J.Lerbet Analytic Geometry and singularities of mechanism ZAMM 78,10, pp 687-
694,(1998)
[6] H.Whitney Tangents to an analytic variety Ann.of Math. 81,496-549 (1965)
In: Lie Groups: New Research ISBN 978-1-60692-389-4
Editor: Altos B. Canterra, pp. 199-265
c 2009 Nova Science Publishers, Inc.
Chapter 8
Abstract
Beside the post-relativistic theories, Open Quantum Relativity is a gauge theory
of interactions based on a nonlinear realization (NLR) of the local Conformal-Affine
(CA) group of symmetry transformations. Such a theory, thanks to a covariant-
symplectic formulation, succeeds in treating General Relativity and Quantum Me-
chanics under the same standard. In this Report, we obtain the coframe fields and
the gauge connections of the theory while the tetrads and Lorentz group metric are
used to induce the spacetime metric. The inhomogenously transforming (under the
Lorentz group) connection coefficients serve as gravitational gauge potentials used to
define covariant derivatives accommodating the couplings of matter and gauge fields.
On the other hand, the tensor valued connection forms serve as auxiliary dynamical
fields associated with the dilation and as special conformal and deformation (shear)
degrees of freedom inherent in the bundle manifold. As a consequence, the bundle
curvature of the theory is determined and the boundary topological invariants are then
constructed. They serve as a prototype (source free) gravitational Lagrangian to de-
rive the following dynamics. Finally, the Bianchi identities, covariant field equations
and gauge currents are obtained. These mathematical tools give rise to a compact,
self-contained approach to physical interactions (in particular gravitation), based on
the local gauge invariance. Starting from this general invariance principle, we discuss
the global and the local Poincaré invariance, developing the spinor, vector and tetrad
formalisms. This covariant-symplectic approach allows to construct the curvature,
torsion and metric tensors starting from the covariant derivative. The resulting theory
∗
E-mail address: capozziello@na.infn.it. (Corresponding author)
200 G. Basini and S. Capozziello
describes a spacetime endowed with non-vanishing curvature and torsion, while the
gravitational field equations are Yang-Mills-like equations of motion, with the torsion
tensor playing the role of the Yang-Mills field strength. Besides other physical conse-
quences and the reliable reproduction of several physical experiments and astrophysi-
cal observations described elsewhere [1], such field equations provide, in principle, the
theoretical device to achieve Close Time Curves and, consequently, the conceivability
of time travels.
1. Introduction
Quantum theory and Relativity are the two fundamental theories of modern physics. The
so-called Standard Model is currently the most successful relativistic quantum field theory
both from particle physics and group theory points of view. It is a non-Abelian gauge theory
(Yang-Mills theory) associated with the internal symmetry group SU (3) × SU (2) × U (1),
in which the SU (3) color symmetry for the strong force in quantum chromodynamics,
is treated as exact, whereas the SU (2) × U (1) symmetry, responsible for generating the
electro-weak gauge fields, is spontaneously broken. So far as we know, there are four fun-
damental forces in Nature; namely, electromagnetic force, weak force, strong force and
gravitational force. The Standard Model covers the first three, but not the gravitational
interaction. In General Relativity, the geometrized gravitational field is described by the
metric tensor gµν of pseudo-Riemannian spacetime, and the field equations that the metric
tensor satisfies are nonlinear. This nonlinearity is indeed a source of difficulty in quanti-
zation of General Relativity. Since the successful Standard Model, in particle physics, is
a gauge theory in which all the fields mediating the interactions are represented by gauge
potentials, a question arises as to why the fields mediating the gravitational interaction are
different from those of other fundamental forces. It is reasonable to expect that there may
be a gauge theory in which the gravitational fields stand on the same footing as those of
other fields [1]. This expectation has prompted a re-examination of General Relativity from
the gauge theoretical point of view.
While the gauge groups involved in the Standard Model are all internal symmetry
groups, the gauge groups in General Relativity must be associated with external spacetime
symmetries. Therefore, the gauge theory of gravity will not be as the usual Yang-Mills the-
ory. It must be one in which gauge objects are not only gauge potentials but also tetrads that
relate the symmetry group to the external spacetime. For this reason we have to consider
a more complex nonlinear gauge theory where all the interactions are dealt under the same
standard [2]. In General Relativity, Einstein took the spacetime metric as the basic variable
representing gravity, whereas Ashtekar employed the tetrad fields and the connection forms
as the fundamental variables. We also consider the tetrads and the connection forms as the
fundamental fields but with the difference that this approach gives rise to a covariant sym-
plectic formalism capable of achieving the result of dealing with physical fields under the
same standard [3, 4].
In 1956, Utiyama suggested that gravitation may be viewed as a gauge theory [5] in
analogy to the Yang-Mills [6] theory (1954). He identified the gauge potential due to the
Lorentz group with the symmetric connection of Riemann geometry, and constructed Ein-
stein’s General Relativity as a gauge theory of the Lorentz group SO(3, 1) with the help
The Conformal-Affine Structure of open Quantum Relativity... 201
of tetrad fields introduced in an ad hoc manner. Although the tetrads were necessary com-
ponents of the theory, to relate the Lorentz group adopted as an internal gauge group to
the external spacetime, they were not introduced as gauge fields. In 1961, Kibble [7] con-
structed a gauge theory based on the Poincaré group P (3, 1) = T (3, 1) ⋊ SO(3, 1) (⋊
represents the semi-direct product) which resulted in the Einstein-Cartan theory character-
ized by curvature and torsion.
The translation group T (3, 1) is considered responsible for generating the tetrads as
gauge fields. Cartan [8] generalized the Riemann geometry to include torsion in addition
to curvature. The torsion (tensor) arises from an asymmetric connection. Sciama [9], and
others (Fikelstein [10], Hehl [11, 12]) pointed out that intrinsic spin may be the source of
torsion of the underlying spacetime manifold.
Since the form and role of the tetrad fields are very different from those of gauge poten-
tials, it has been thought that even Kibble’s attempt is not satisfactory as a full gauge theory.
There have been a number of gauge theories of gravitation based on a variety of Lie groups
[11, 12, 13, 14, 16, 15, 17]. It was argued that a gauge theory of gravitation corresponding
to General Relativity can be constructed with the translation group alone, in the so-called
teleparallel scheme.
Inomata et al. [18] proposed that Kibble’s gauge theory could be obtained, in a manner
closer to the Yang-Mills approach, by considering the de Sitter group SO(4, 1), which
is reducible to the Poincaré group by group-contraction. Unlike the Poincaré group, the
de Sitter group is homogeneous and the associated gauge fields are all of gauge potential
type and by the Wigner-Inönu group contraction procedure, one of the five vector potentials
reduces to the tetrad.
It is standard to use the fiber-bundle formulation by which gauge theories can be con-
structed on the basis of any Lie group. Recent work by Hehl et al. [17], on the so-called
Metric Affine Gravity (MAG) theory, adopted as a gauge group the affine group A(4,
R) = T (4)⋊GL(4, R), which was realized linearly. The tetrad was identified with the non-
linearly realized translational part of the affine connection, on the tangent bundle. In MAG
theory, the Lagrangian is quadratic in both curvature and torsion, in contrast to the Einstein-
Hilbert Lagrangian in General Relativity which is linear in the scalar curvature. The theory
has the Einstein limit on one hand and leads to the Newtonian inverse distance potential
plus the linear confinement potential, in the weak field approximation, on the other. So, as
we have seen above, there are many attempts to formulate gravitation as a gauge theory but
currently no theory has been uniquely accepted as the gauge theory of gravity.
The nonlinear approach to group realizations was originally introduced by Coleman,
Wess and Zumino [19, 20] in the context of internal symmetry groups (1969). It was later
extended to the case of spacetime symmetries by Isham, Salam, and Strathdee [21, 22] con-
sidering the nonlinear action of GL(4, R) modulus the Lorentz subgroup. In 1974, Borisov,
Ivanov and Ogievetsky [23, 24], considered the simultaneous nonlinear realization (NLR)
of the affine and conformal groups. They stated that General Relativity can be viewed
as a consequence of spontaneous breakdown of the affine symmetry, in the same manner
that chiral dynamics, in quantum chromodynamics, is a result of spontaneous breakdown
of chiral symmetry. In their model, gravitons are considered as Goldstone bosons asso-
ciated with the affine symmetry breaking. In 1978, Chang and Mansouri [25] used the
NLR scheme employing GL(4, R) as the principal group. In 1980, Stelle and West [26]
202 G. Basini and S. Capozziello
investigated the NLR induced by the spontaneous breakdown of SO(3, 2). In 1982 Ivanov
and Niederle considered nonlinear gauge theories of the Poincaré, de Sitter, conformal and
special conformal groups [27, 28]. In 1983, Ivanenko and Sardanashvily [29] considered
gravity to be a spontaneously broken GL(4, R) gauge theory. The tetrads fields arise, in
their formulation, as a result of the reduction of the structure group of the tangent bundle
from the general linear to Lorentz group. In 1987, Lord and Goswami [30, 31] developed
the NLR in the fiber bundle formalism based on the bundle structure G (G/H, H) as sug-
gested by Ne’eman and Regge [32]. In this approach, the quotient space G/H is identified
with physical spacetime. Most recently, in a series of papers, Lopez-Pinto, Julve, Tiemblo,
Tresguerres and Mielke discussed nonlinear gauge theories of gravity on the basis of the
Poincaré, affine and conformal groups [35, 34, 36, 37, 38, 39].
Now, following the prescriptions of General Relativity, the physical spacetime is as-
sumed to be a four-dimensional differential manifold. In Special Relativity, this manifold
is the Minkwoski flat-spacetime M4 while, in General Relativity, the underlying spacetime
is assumed to be curved in order to describe the effects of gravitation.
As we said, Utiyama [5] proposed that General Relativity can be seen as a gauge theory
based on the local Lorentz group in the same way that the Yang-Mills gauge theory [53]
is developed on the basis of the internal iso-spin gauge group. In this formulation the Rie-
mannian connection is the gravitational counterpart of the Yang-Mills gauge fields. While
SU (2), in the Yang-Mills theory, is an internal symmetry group, the Lorentz symmetry rep-
resents the local nature of spacetime rather than internal degrees of freedom. The Einstein
Equivalence Principle, asserted for General Relativity, requires that the local spacetime
structure can be identified with the Minkowski spacetime possessing Lorentz symmetry.
In order to relate local Lorentz symmetry to the external spacetime, we need to solder
the local space to the external space. The soldering tools can be the tetrad fields. Utiyama
regarded the tetrads as objects given a priori while they can be dynamically generated [2]
and the spacetime has necessarily to be endowed with torsion in order to accommodate
spinor fields. In other words, the gravitational interaction of spinning particles requires
the modification of the Riemann spacetime of General Relativity to be a (non-Riemannian)
curved spacetime with torsion. Although Sciama used the tetrad formalism for his gauge-
like handling of gravitation, his theory fell shortcomings in treating tetrad fields as gauge
fields.
Following the Kibble approach [7], it can be demonstrated how gravitation can be for-
mulated starting from a pure gauge viewpoint. In particular, the aim of this paper is to show,
in details, how a theory of gravitation is a gauge theory which can be obtained starting from
some local invariance, e.g. the local Poincaré symmetry, leading to a given unification
scheme. This is the dynamical structure of Open Quantum Relativity [1] by which a gauge
theory of gravity, based on a nonlinear realization of the local conformal-affine group of
symmetry transformations, can be formulated [4].
Here, we start from a General Invariance Principle (the General Conservation Principle
[43]) and consider first the Global Poincaré Invariance and then the Local Poincaré Invari-
ance. This approach leads to construct a given theory of gravity as a gauge theory. Such
a viewpoint, if considered in detail, can avoid many shortcomings and could be useful to
formulate self-consistent schemes for quantum gravity and then the unification of all inter-
actions [2].
The Conformal-Affine Structure of open Quantum Relativity... 203
π : P → M. (2.1)
π −1 (x) ≡ Fx ≈ F, (2.2)
S
and it is called the fiber at x ∈ M . The partitioning x π −1 (x) = P is referred to as the
fibration. Note that a smooth map is one whose coordinatization is C ∞ differentiable; a
smooth manifold is a space that can be covered with coordinate patches in such a manner
that, a change from one patch to any overlapping patch is smooth [40]. Fiber bundles that
admit decomposition as a direct product, locally looking like P ≈M × F, Sare called trivial.
Given a set of open coverings {Ui } of M with x ∈ {Ui } ⊂ M satisfying α Uα = M , the
diffeomorphism map is given by
χi : Ui ×M G → π −1 (Ui ) ∈ P, (2.3)
(×M represents the fiber product of elements defined over the space M ) such that
π (χi (x, g)) = x and χi (x, g) = χi (x, (id)G ) g = χi (x) g ∀x ∈ {Ui } and g ∈ G.
Here, (id)G represents the identity element of the group G. In order to obtain the global
bundle structure, the local charts χi must be glued together continuously. Consider two
patches Un and Um with a non-empty intersection Un ∩ Um 6= ∅. Let ρnm be the restriction
of χ−1 −1 −1
n to π (Un ∩ Um ) defined by ρnm : π (Un ∩ Um ) → (Un ∩ Um ) ×M Gn . Similarly
let ρmn : π −1 (Um ∩ Un ) → (Um ∩ Un ) ×M Gm be the restriction of χ−1 −1
m to π (Un ∩ Um ).
The composite diffeomorphism Λnm ∈ G
defined as
Λij (x) ≡ ρji ◦ ρ−1 −1
ij = χi, x ◦ χj, x : F → F (2.5)
constitutes the transition function between bundle charts ρnm and ρmn (◦ represents the
group composition operation) where the diffeomorphism χi, x : F → Fx is written as
χi, x (g) := χi (x, g) and satisfies χj (x, g) = χi (x, Λij (x) g). The transition functions
The Conformal-Affine Structure of open Quantum Relativity... 205
{Λij } can be interpreted as passive gauge transformations. They satisfy some consis-
tency conditions, i.e. the identity Λii (x), the inverse Λij (x) = Λ−1ji (x) and the cocycle
Λij (x) Λjk (x) = Λik (x). For trivial bundles, the transition function reduces to
s : M → P, (2.7)
π ◦ s = (id)M , (2.8)
where (id)M is the identity element of M . It assigns to each point x ∈ M a point in the
fiber over x. Trivial bundles admit global sections.
A bundle is a principal fiber bundle hP, P/G, G, πi provided that the Lie group G acts
freely (i.e. if pg = p then g = (id)G ) on P to the right Rg p = pg, p ∈ P, preserves fibers
on P (Rg : P → P), and finally is transitive on fibers. Furthermore, there must exist local
trivializations compatible with the G action. Hence, π −1 (Ui ) is homeomorphic to Ui ×M G
and the fibers of P are diffeomorphic to G. The trivialization or inverse diffeomorphism
map is given by
χ−1 −1
i : π (Ui ) → Ui ×M G (2.9)
such that χ−1 (p) = (π(p), ϕ(p)) ∈ Ui ×M G, p ∈ π −1 (Ui ) ⊂ P, where we see from the
above definition that ϕ is a local mapping of π −1 (Ui ) into G satisfying ϕ(Lg p) = ϕ(p)g for
any p ∈ π −1 (U) and any g ∈ G. Let us observe that the elements of P which are projected
onto the same x ∈ {Ui } are transformed into one another by the elements of G. In other
words, the fibers of P are the orbits of G and at the same time, they are the set of elements
which are projected onto the same x ∈ U ⊂ M . This observation motivates calling the
action of the group vertical and the base manifold horizontal. The diffeomorphism map
χi is called the local gauge since χ−1i maps π −1 (Ui ) onto the direct (Cartesian) product
Ui ×M G. The action Lg of the structure group G on P defines an isomorphism of the Lie
algebra g of G onto the Lie algebra of vertical vector fields on P, tangent to the fiber at each
p ∈ P called fundamental vector fields
where Tp (P) is the space of tangents at p, i.e. Tp (P) ∈ T (P). The map λ is a linear
isomorphism for every p ∈ P and is invariant with respect to the action of G, that is,
λg : (λg∗ Tp (P)) → Tgp (P), where λg∗ is the differential push forward map induced by λg
defined by λg∗ : Tp (P) → Tgp (P).
Since the principal bundle P (M , G) is a differentiable manifold, we can define tan-
gent T (P) and cotangent T ∗ (P) bundles. The tangent space Tp (P) defined at each
point p ∈ P may be decomposed into a vertical Vp (P) and horizontal Hp (P) subspace
as Tp (P) := Vp (P) ⊕ Hp (P) (where ⊕ represents the direct sum). The space Vp (P)
206 G. Basini and S. Capozziello
is a subspace of Tp (P) consisting of all tangent vectors to the fiber passing through
p ∈ P, and Hp (P) is the subspace complementary to Vp (P) at p. The vertical subspace
Vp (P) := {X ∈ T (P) |π (X) ∈ Ui ⊂ M } is uniquely determined by the structure of P,
whereas the horizontal subspace Hp (P) cannot be uniquely specified. This result is very
important because it makes possible to fix the Cauchy conditions on the dynamics as we
will discuss in Sec. V. Thus we require the following condition: when p transforms as
p → p′ = pg, Hp (P) transforms as [41],
ω := ge−1 πPM
∗
g + ge−1 de
Ωe g (2.12)
(∗ represents the differential pullback map) belonging to g⊗Tp∗ (P) where Tp∗ (P) is the dual
space to Tp (P). In such a case, the differential pullback map, applied to a test function ϕ
and p-forms α and β, satisfies f ∗ ϕ = ϕ ◦ f , (g ◦ f )∗ = f ∗ g ∗ and f ∗ (α ∧ β) = ∗ ∧ f ∗ β.
fαβα
If G is represented by a d-dimensional d × d matrix, then GA = [Gαβ ], ge = ge , where
α, β = 1, 2, 3,...d. Thus, ω assumes the form
ωαβ = ge−1 αγ de g γβ + ge−1 ργ πPM∗
Ωρσi Gαγ geσβ ⊗ dxi . (2.13)
Ei := ∂i + Γαβ
i ∂αβ , i = 1, 2, 3, ..n and α, β = 1, 2, 3, ..d (2.14)
∂
where ∂i = ∂x i . The connection 1-form ω projects Tp (P) onto Vp (P). In order for X ∈
Tp (P) to belong to Hp (P), it has to be X ∈ Hp (P), ωp (X) = hω (p) |Xi = 0. In other
words,
Hp (P) := {X ∈ Tp (P) |ωp (X) = 0} , (2.15)
from which Ωαβ i can be determined. The inner product appearing in ωp (X) =
hω (p) |Xi = 0 is a map h·|·i : Tp∗ (P) × Tp (P) → R defined by hW |V i =
Wµ V ν dxµ | ∂x∂ ν = Wµ V ν δνµ , where the 1-form W and vector V are given by W =
Wµ dxµ and V = V µ ∂x∂ ν . Observe also that, dg αβ |∂ρσ = δρα δσβ .
A
We parameterize an arbitrary group element geλ as ge (λ) = eλ GA = eλ·G , A =
1,..dim (g). The right action Rge(λ) = Rexp(λ·G) on p ∈ P, i.e. Rexp(λ·G) p = p exp (λ · G),
defines a curve through p in P. Define a vector G# ∈ Tp (P) by [41]
d
G# f (p) := f (p exp (λ · G)) |λ=0 (2.16)
dt
The Conformal-Affine Structure of open Quantum Relativity... 207
s∗i ω (Y ) = Ω (Y ) . (2.25)
208 G. Basini and S. Capozziello
To determine the gauge transformation of the connection 1-form ω, we use the fact that
Rge∗ X = Xe g for X ∈ Tp (M ) and the transition functions genm ∈ G defined between
neighboring bundle charts (2.6). By direct computation we get
d d
cj∗ X = cj (λ (t)) |t=0 = [ci (λ (t)) · geij ] |t=0
dt dt
#
−1
= Rgeij ∗ c∗i (X) + geji (x) degij (X) , (2.26)
d
where λ (t) is a curve in M with boundary values λ (0) = m and dt λ (t) |t=0 = X. Thus,
we obtain the useful result
#
c∗ X = Rge∗ (c∗ X) + ge−1 de
g (X) . (2.27)
according to which fibers are mapped into fibers, and the verticality condition
where u and λ (u) belong to the same fiber. The last condition ensures that no diffeomor-
phisms λM : M → M given by
with elements {h} and known linear representations ρ (h). Here Π is the first of the two
projection maps in (2.37), and Rh is the right group action. Let M be a differentiable
manifold with points {x} to which G and H may be referred, i.e. g = g(x) and h = h(x).
Being that G and H are Lie groups, they are also manifolds. The right action of H on G
induces a complete partition of G into mutually disjoint orbits gH. Since g = g(x), all
elements of gH = {gh1 , gh2 , gh3 , · · · , ghn } are defined over the same x. Thus, each orbit
gH constitutes an equivalence class of point x, with equivalence relation g ≡ g ′ where
g ′ = Rh g = gh. A comment is in order at this point: this result directly leads toward the
Many Worlds Interpretation [77] of Open Quantum Mechanics as we will see below.
By projecting each equivalence class onto a single element of the quotient space
M := S G/H, the group G becomes organized as a fiber bundle in the sense that
G = i {gi H}. In this manner the manifold G is viewed as a fiber bundle G (M, H; Π)
with H-diffeomorphic fibers Π−1 (ξ) : G → M = gH and base space M. A compos-
ite principal fiber bundle P(M , G; π) is one whose G-diffeomorphic fibers possess the
fibered structure G (M, H; Π) ≃ M× H described above. The bundle P is then locally
isomorphic to M × G (M, H). Moreover, since an element g ∈ G is locally homeomor-
phic to M × H the elements of P are - by transitivity - also locally homeomorphic to
M × M × H ≃ Σ × H where (locally) Σ ≃ M × M. Thus, an alternative view of P(M ,
G; π) is provided by the P-associated H-bundle P(Σ, H; π e) [37]. The total space P may
be regarded as G (M, H; Π)-bundles over the base space M or equivalently as H-fibers
attached to the manifold Σ ≃ M × M.
The nonlinear realization (NLR) technique [19, 20] provides a way to determine the
transformation properties of fields defined on the quotient space G/H. The NLR of
Diff(4, R) becomes tractable due to a theorem given by V. I. Ogievetsky. According to this
theorem [23], the algebra of the infinite dimensional group Diff(4, R) can be taken as the
closure of the finite dimensional algebras of SO(4, 2) and A(4, R). Remind that the Lorentz
210 G. Basini and S. Capozziello
group generates transformations that preserve the quadratic form on Minkowski spacetime
built from the metric tensor, while the special conformal group generates infinitesimal
angle-preserving transformations on Minkowski spacetime. The affine group is a gener-
alization of the Poincaré group where the Lorentz group is replaced by the group of general
linear transformations. As such, the affine group generates translations, Lorentz transfor-
mations, volume preserving shear and volume changing dilation transformations. As a con-
sequence, the NLR of Diff(4, R) /SO(3, 1) can be constructed by taking a simultaneous
realization of the conformal group SO(4, 2) and the affine group A(4, R) := R4 ⋊ GL(4,
R) on the coset spaces A(4, R)/SO(3, 1) and SO(4, 2)/SO(3, 1). One possible inter-
pretation of this theorem is that the conformal-affine group CA (defined below) may be the
largest subgroup of Diff(4, R), whose transformations may be put into the form of a gener-
alized coordinate transformation. We remark that a NLR can be made linear by embedding
the representation in a sufficiently higher dimensional space. Alternatively, a linear group
realization becomes nonlinear when subject to constraints. One type of relevant constraints
may be those responsible for symmetry reduction from Diff(4, R) to SO(3, 1) for instance.
We take the group CA(3, 1) as the basic symmetry group G. The CA group consists
of the groups SO(4, 2) and A(4, R). In particular, CA is proportional to the union SO(4,
2) ∪ A(4, R). We know however (see section Conformal-Affine Lie Algebra) that the affine
and special conformal groups have several group generators in common. These common
generators reside in the intersection SO(4, 2) ∩ A(4, R) of the two groups, within which
there are two copies of Π := D × P (3, 1), where D is the group of scale transformations
(dilations) and P (3, 1) := T (3, 1) ⋊ SO(3, 1) is the Poincaré group. We define the CA
group as the union of the affine and conformal groups minus one copy of the overlap Π,
i.e. CA(3, 1) := SO(4, 2) ∪ A(4, R) − Π. Being defined in this way we recognize that
CA(3, 1) is a 24 parameter Lie group representing the action of Lorentz transformations
(6), translations (4), special conformal transformations (4), spacetime shears (9) and scale
transformations (1). In this paper, we obtain the NLR of CA(3, 1) modulo SO(3, 1) as a
4D realization of the embedding procedure adopted in Open Quantum Relativity [1, 2].
ΠPΣ : P → Σ, π
eΣM : Σ → M . (2.37)
The projection ΠPΣ maps the point (p ∈ P, Rh p ∈ P) into point (x, ξ) ∈ Σ. There is a
correspondence between the sections sM Σ : M → Σ and the projection ΠPΣ : P → Σ,
in the sense that both maps project their functional argument onto elements of Σ. This is
formalized by the relation, ΠPΣ (p) = sM Σ ◦ πPM (p). Hence, the total projection is given
by
π
e := πPM = π eΣM ◦ ΠPΣ . (2.38)
Associated with the projections π
eΣM and ΠPΣ are the corresponding local sections
−1
sM Σ : U → π
eΣM (U) ⊂ Σ, sΣP : V → Π−1
PΣ (V) ⊂ P, (2.39)
each orbit pH through p ∈ P projects into a single element Q ∈ Σ. In analogy to the total
bundle projection (2.37), a total section of P is given by the total section composition
sM P = sΣP ◦ sM Σ . (2.41)
Let elements of G/H be labeled by the parameter ξ. Functions on G/H are represented
by continuous coset functions c(ξ) parameterized by ξ. These elements are referred to as
cosets to the right of H with respect to g ∈ G. Indeed, the orbits of the right action of H
−1
on G are the left cosets Rh g = gH. For a given section sM P (x ∈ M ) ∈ πPM , with local
coordinates (x, g), one can perform decompositions of the partial fibers sM Σ and sΣP as:
cM Σ (x) · c = Rc′ ◦ e
sM Σ (x) = e cM Σ (x) ; c = c (ξ) , (2.42)
e cΣP ◦ e
cM P = e cM Σ . (2.44)
The images of two sections sΣP and sM Σ over x ∈ M must coincide, implying sΣP (x, ξ) =
sM Σ (x). Using Eq. (2.41) with Eqs.(2.42), (2.43) and (2.44), we arrive at the total bundle
section decomposition
cM P (x) · g = Rg ◦ e
sM P (x) = e cM P (x) (2.45)
provided g = c · a and
cΣP = Rc−1 ◦ e
e cΣP (x, ξ) ◦ Rc . (2.46)
cΣP , defined as
The pullback of e
cξ (x) = (s∗M Σ e
e cΣP ◦ sM Σ = e
cΣP ) (x) = e cΣP (x, ξ) , (2.47)
ensures the coincidence between the images of sections e cξ (x) : M → P and e cΣP (x, ξ) :
Σ → P, respectively. With the aid of the above results, we arrive to the important relation
e cM P (x) · c (ξ) .
cΣP (x, ξ) = e (2.48)
p = sM P (x) = Ra ◦ e
cΣP (x, ξ) , a ∈ H. (2.49)
The Conformal-Affine Structure of open Quantum Relativity... 213
Performing a gauge transformation on p we obtain the orbit λ (p) defining a curve through
(x, ξ) in Σ
cΣP x, ξ ′ ; g (x) ∈ G, a′ ∈ H.
λ (p) = Lg(x) ◦ p = Ra′ ◦ e (2.50)
By virtue of the commutability of left and right group translations of elements belonging to
G, i.e. Lg ◦ Rh = Rh ◦ Lg , Eq.(2.51) may be recast as
Lg(x) ◦ e cΣP (x, ξ) = Rh ◦ ecΣP x, ξ ′ . (2.52)
where Ra−1 ◦ Ra′ ≡ Ra′ a−1 := Rh and a′ a−1 ≡ h ∈ H (see also [41]). Eq.(2.52)
constitutes a generalized gauge transformation. Performing the pullback of Eq.(2.52) with
respect to the section sM Σ leads to
Lg(x) ◦ e
cξ (x) = Rh(ξ, g(x)) ◦ e
cξ′ (x) . (2.53)
Thus, the left action Lg of G is a map that acts on P and Σ. In particular, Lg , acting on fibers
defined as orbits of the right action describes, diffeomorphisms that transforming fibers over
cξ (x) into the fibers e
e cξ′ (x) of Σ, while being simultaneously displaced along H fibers, via
the action of Rh . Eq.(2.53) states that nonlinear realizations of G mod H is determined by
the action of an arbitrary element g ∈ G, on the quotient space G/H, transforming one
coset into another as
Lg : G/H → G/H, c(ξ) → c(ξ ′ ) (2.54)
inducing a diffeomorphism ξ → ξ ′ on G/H. To simplify the action induced by Eq.
(2.53) for calculation purposes we proceed as follows. Starting from (2.47) and substituting
sM Σ = Rc ◦ e cM P , we get
e cΣP ◦ Rc ◦ e
cξ (x) = e cM Σ . (2.55)
Using ecM P ◦ Rc = Rc ◦ e cM P , Eq.(2.55) becomes e cξ (x) = Rc ◦ e cΣP ◦ e
cM Σ = Rc ◦ ecM P ,
where the last equality follows from the use of ecM P = ecΣP ◦ ecM Σ . By analogy, we assume
cξ′ (x) ≡ Rc′ ◦ e
e cM P . Upon substitution of e
cξ′ into Eq.(2.53), we obtain
Lg ◦ Rc ◦ e
cM P = Rh(ξ, g(x)) ◦ Rc′ ◦ e
cM P , (2.56)
g·e cM P · c′ · h.
cM P · c = e (2.57)
Operating on Eq.(2.57)
c−1
from the left by eM P and making use of g = e c−1
M P ge
cM P , we get
−1 ′
cM P · g · e
e cM P · c = c · h, which leads to g · cξ = cξ′ · h, or in short
c′ = g · c · h−1 (2.58)
where c ≡ cξ and c′ ≡ cξ′ . Observe that the element h is a function whose argument is the
couple (ξ, g (x)). The transformation rule (2.58) is, in fact, the key equation to determine
the nonlinear realizations of G and specifies a unique H-valued field h(ξ, g (x)) on G/H.
214 G. Basini and S. Capozziello
c (x, ξ) := c ◦ e
b c (x, ξ) = c (e
c (x, ξ)) . (2.59)
Taking ΠPΣ ◦ Rh ◦ e cΣP = ΠPΣ ◦ e cΣP = (id)Σ into account, we can explicitly exhibit the
cΣP : P → Σ induces an equivalence
fact that the left action Lg of G on the null sections, e
relation between differing elements e cξ′ ∈ Σ given by
cξ , e
ΠPΣ ◦ Lg ◦ b
cξ = ΠPΣ ◦ Rh(ξ, g(x)) ◦ b
cξ′ = Rh(ξ, g(x)) ◦ e
cξ ′ , (2.60)
so that
c′ξ := Rh(ξ, g(x)) ◦ e
e cξ′ = Lg ◦ e
cξ . (2.61)
From Eq.(2.61) we can write
Lg
e c′ξ = Rh(ξ, g(x)) ◦ e
cξ 7−→ e cξ′ ∀h ∈ H. (2.62)
Eq. (2.62) gives rise to a complete partition of G/H into equivalence classes Π−1
PΣ (ξ) of
left cosets [38, 42]
cH = Rh(ξ, g(x)) ◦ c/c ∈ G/H, ∀h ∈ H = {ch1 , ch2 ,..., chn } , (2.63)
where c ∈ (G − H) plays the role of the fibers attached to each point of Σ. The elements
−1
chi are single representatives of each equivalence class Rh(ξ, g(x)) ◦ c = cH ∈ π eΣM (U).
Thus, any diffeomorphism Lg ◦ e cξ on Σ, together with the H-valued function h (ξ, g (x)),
determines a unique gauge transformation e c′ξ = Rh(ξ, g(x)) ◦ e
cξ′ . This demonstrates that
gauge transformations are those diffeomorphisms on Σ that map fibers over c (ξ) into fibers
over c (ξ ′ ) and simultaneously preserves the action of H.
Observe that the homogeneous part of the special conformal coset parameter ζ α has the
same structure as that of the translational parameter ξ α (with the substitutions: ζ α → −ξ α
and −ǫα → bα ). For the shear parameters we obtain
In the limit of vanishing special conformal 4-boost, this result coincides with that of Pinto
et al. [34]. For vanishing shear, the result of Julve et al [35] is obtained.
We must stress that all covariant coset field transformations has been determined di-
rectly from the nonlinear transformation law (2.58). We observe that the translational coset
parameter transforms as a coordinate under the action of G. From the shear coset variation,
the explicit form of the nonlinear Lorentz-like transformation has been obtained. From
(2.70) it is clear that uαβ contains the linear Lorentz parameter, in addition to conformal
and shear contributions via the nonlinear 4-boosts and symmetric GL4 parameters.
ω = ge−1 (d + πPM
∗
ΩM ) ge. (2.71)
The gauge potential ΩM is defined in the standard manner as the pullback of the connection
ω by the null section e cM P , ΩM = e c∗M P ω ∈ T ∗ (M ). With regard to the space Σ, an
alternative form of the connection is given by
ω = a−1 (d + πPΣ
∗
ΓΣ ) a, (2.72)
where the connection on Σ reads ΓΣ = e c∗ΣP ω. Carrying out a similar analysis and evaluating
the tangent vector X ∈ Tp (Σ) at each point ξ along the curve cξ on the coset space G/H
c∗ΣP ), we find the gauge transformation law
(that coincides with the section e
ω → ω ′ = adh−1 (d + ω) . (2.73)
216 G. Basini and S. Capozziello
The horizontal basis vectors, that span the horizontal tangent space H(P) of πPM : P →M
are given by
cM P∗ ∂i − Ωi .
Ei = e (2.87)
The explicit form of the connections (2.86) are given by
αβ GL
µ ν ∗ αβ ν ν † µ eν ∗
ω = −i VM χ eµ Pν − i iΘ(† Λ) + π ePM Γ χ eβ Λµν + ϑM βµ ∆ν − ie
eα χ πPM ΦM D
(2.88)
αβ αβ αβ
where Θ(† Λ) = Θ(L) + Θ(SY) , with right invariant Maurer-Cartan forms (see the Appendix
B)
µν µν
Θ(L) = iβe[ν|γ dβe|µ]γ − 2idbµ ǫν and Θ(SY) = ie
α(ν|γ de
α|µ)γ . (2.89)
The linear connection ΩM varies under the action of G as
T C D GL
δΩ = Ω′ − Ω = δ Γ µ Pµ + δ Γ µ ∆µ + δ ΓD + δ Γ βν †
Λβν (2.90)
where
T GL T D
δ Γ µ = † Dǫµ − Γ α (ααµ + βαµ + ϕδαµ ) − Γǫµ ,
C GL C D
δ Γ µ = † Dbµ − Γ α (ααµ + βαµ − ϕδαµ ) + Γbµ ,
(2.91)
GL
αβ
GL
†D
T C
δΓ = ααβ + β αβ [α β]
+ Γ b +Γ ǫ [α β] ,
D C T
δ Γ = dϕ + 2 Γ αǫ −Γ αb .
α α
The components of ω on M are identified as spacetime quantities and are determined from
the pullback of the corresponding (quotient space) quantities defined on Σ:
µ
VM = s∗M Σ VΣµ , ϑµM = s∗M Σ ϑµΣ , ΦM = s∗M Σ ΦΣ and Γµν ∗ µν
M = sM Σ ΓΣ . (2.92)
In the following, we depart from the alternative form of the connection ω =
a−1 (d + Π∗PΣ ΓΣ ) a, ∀ a ∈ H on Σ.
where
ρσ
Γρσ eαeβ
L
ω αβ := iΘ(L) + π ∗
ePΣ [L] β[ρ βσ] . (2.94)
218 G. Basini and S. Capozziello
∆ ∆
ω = −ia−1 (e ∗
πΣM ϑνΣ ∆ν ) a = −i ω µ ∆µ , (2.96)
D
ω = −ia−1 (e ∗
πΣM ΦΣ D) a = −iω[D] D, (2.97)
SY
SY
ω = −ia−1 π ∗
eΣM Υαβ Sαβ a = −i ω αβ Sαβ , (2.98)
where
∗ SY αβ ∗ α β
ω[D] : = π
eΣM ΦΣ , ω := π
ePΣ Υρσ α
e(ρ α
eσ) . (2.100)
The nonlinear translational and special conformal connection coefficients VΣν and ϑνΣ read
C
β ∗ φ β α σ β
VΣ = π eΣM e υ (ξ) + r σ Γ Bα (ξ) , (2.102)
h i
ϑβΣ = π ∗
eΣM e−φ υ β (ζ) + υ σ (ξ) Bσβ (ζ) , (2.103)
with
!
GL D T
υiβ (ξ) := rσβ †
Di ξ σ σ
+ Γi ξ + Γ σ
i , Bαρ (ξ) := |ξ|2 δαρ − 2ξα ξ ρ . (2.104)
Γµν b µν + 2ζ [µ ̟ν] ,
Σ =Γ (2.105)
1
∗
Φ=πeΣM ζβ ̟β − dφ, (2.106)
2
with
GL σβ ν
−1 µ
−1 µ
b µν
Γ := π ∗
eΣM r Γ rβ − r dr σν
(2.107)
σ σ
and
C
̟ν := υ ν + rνα Γ α . (2.108)
The Conformal-Affine Structure of open Quantum Relativity... 219
The nonlinear GL4 connection can be expanded in the GL4 Lie algebra according to Γαβ
◦
†Λ = Γ αβ Lαβ + Υαβ † Sαβ , where
αβ
◦
Γ αβ b [αβ] + 2ζ [α ̟β] , Υαβ := Γ
Σ := Γ Σ
b (αβ) . (2.109)
The symmetric GL4 (shear) gauge fields Υ are distortion fields describing the difference
between the general linear connection and the Levi-Civita connection, one of the differences
between General Relativity and Open Quantum Relativity.
We define the (group) algebra bases eν and hν dual to the translational and special
conformal 1-forms V µ and ϑµ as
hµ : = hµi sM Σ∗ ∂i = ∂ζ µ − hµi e
hi , (2.111)
hiµ (ξ, ζ) = e−φ υρµ (ζ) + υiσ (ξ) Bσµ (ζ) , (2.113)
and
GL C GL D T
e −φ µ ρ α ρ γ σ α σ σ µ
hj (ξ, ζ) = e
cM Σ∗ ∂j + e r ρ Γ jα ζ + Γ j + r σ Γ jα ξ + Γj ξ + Γ j Bγ (ζ) ∂ζ µ .
(2.115)
Here υβ(ζ) = υβ(ξ → ζ), Bβα (ζ) = Bρα (ξ → ζ). By definition, the basis vectors satisfy
the orthogonality relations
D E
ej = 0, ϑµΣ |e
VΣµ |e hj = 0, hV µ |eν i = δνµ , hϑµ |hν i = δνµ . (2.116)
and
(µ| GL |ν) (µ|
fej (ξ, h) := e
cM Σ∗ ∂j − r−1 σ Γ j σβ rβ − r−1 σ ∂j r σ|ν)
∂hµν . (2.121)
(2.124)
◦ ◦ ◦
µν µν µν
Γ M = dxi ⊗ eαi Γ Σ |eα := dxi ⊗Γ i .
Int
◦
bµ = e
E αβ
cΣP∗ eeµ + i Γ |e b (L) ,
eµ R (2.127)
αβ
and satisfy DL E DL E
bj = 0 = ω|
ω|E bµ .
E (2.128)
The right invariant fundamental vector operator appearing in (2.126) or (2.127) is given by
b (L) e γ ∂ ∂
R µν = i β[µ| + ǫ[µ ν] . (2.129)
∂ βe|ν]γ ∂ǫ
ePΣ : P → Σ satisfies
On the other hand, the vertical tangent subspace vector in π
DL E DL E
ω|Lb (L) = Lµν = ω|R b (L) , (2.130)
µν µν
The Conformal-Affine Structure of open Quantum Relativity... 221
where
b (L) = iβeγ[µ| ∂ b (L) e γ ∂ ∂
L µν |ν]
, R µν = i β[µ| + ǫ[µ ν] . (2.131)
∂ βeγ ∂ βe|ν]γ ∂ǫ
ν 1 γ ν
and βeµν := eβµ = δµν + βµν + 2! βµ βγ + · · ·. The horizontal tangent subspace vectors in
ΠΣM : Σ→M are given by
⌣
ej = e
E cΣP∗e
ej = e
cΣP∗ eej , H b (D) = e
hj , E cΣP∗ dej , E j = e
cΣP∗ fej , (2.132)
i
and satisfy
DP E D E D ⌣ E DD E
ej = 0, ∆
ω|E e j = 0, SY
ω|H b (D) = 0.
ω |E j = 0, ω|E (2.133)
i
and satisfy
DP E D E D ⌣ E DD E
ω|E eµ = Pµ , ∆ e µ = ∆µ , SY
ω|H b (D) = D.
ω |E αβ = † Sαβ , ω|E (2.135)
The left invariant fundamental vector operators appearing in (2.134) are readily computed,
the result being
b (P)
L e ν ∂ b (∆) = iW
µ = iQ µ ∂ǫν , L µ
f νµ ∂ν ,
∂b
(2.136)
b (SY) ∂ b (D) β ∂
L αβ = ie αγ(µ| |ν) , L = −iǫ ∂ǫβ ,
∂α
eγ
αµν
eµν := e
where α = αµν + αµν + 2!1 αµγ αγν + · · ·, Q
e α := (e fσα :=
χσα + δσα eϕ ), W
α σα
χσα + δσα e−ϕ ) satisfying Q
(e e −1 = Qe ασ and W f −1 f ασ . Making use of the
= W
σ σ
transformation law of the nonlinear connection (2.80), we obtain
δΓ = δV α Pα + δϑα ∆α + 2δΦD + δΓαβ † Λαβ (2.137)
where
GL
δV ν = uαν V α , δϑν = uαν ϑα , δΦ = 0, δΓαβ = † ∇uαβ . (2.138)
GL
From δΓαβ = † ∇uαβ we observe that
◦
δΓ[αβ] = ∇uαβ , δΥαβ = 2uρ (α| Υρ|β) . (2.139)
According to Eq.(2.138), the nonlinear translational and special conformal gauge fields
transform as contravariant vector valued 1-forms under H, the antisymmetric part of Γαβ
transforms inhomogeneously as a gauge potential and the nonlinear dilaton gauge field Φ
transforms as a scalar valued 1-form. From Eq.(2.139), it is clear that the symmetric part
of Γαβ is a tensor valued 1-form. Being characterized as 4-covectors, we identify V ν as
◦
coframe fields. The connection coefficient Γ αβ serves as the gravitational gauge potential.
The remaining components of Γ, namely ϑ, Υ and Φ are dynamical fields of the theory.
As will be seen in the following subsection, the tetrad components of the coframe are used
in conjunction with the H-metric to induce a spacetime metric on M . In open Quantum
Relativity, this result is extremely important and directly depends on the intrinsic covariant
symplectic structure of the theory [4].
222 G. Basini and S. Capozziello
By use of (2.142), together with the Lie algebra commutators, very relevant for the sym-
plectic structure, we obtain the bundle curvature
The field strength components of F are given by the first Cartan structure equations. They
are respectively, the projectively deformed, Υ-distorted translational field strength
GL
T α := † ∇V α + 2Φ ∧ V α , (2.144)
Z := dΦ + Ψ, Ψ = V · ϑ − ϑ · V (2.146)
with
b αβ := Rαβ + Rαβ , Ψαβ := V [α ∧ ϑβ] .
R (2.148)
GL
The operator † ∇ denotes the nonlinear covariant derivative built from volume preserving
◦
(VP) connection (i.e. excluding Φ) forms. The Υ and Γ-affine curvatures in (2.148) read
◦
Rαβ : = ∇Υαβ + Υαγ ∧ Υγβ , (2.149)
◦ ◦ ◦
Rαβ : = dΓ αβ + Γ γα ∧ Γ γβ , (2.150)
◦ ◦
respectively. Operator ∇ is defined with respect to the restricted connection Γ αβ given in
(2.109).
The field strength components of the bundle curvature have the following group varia-
tions
A gauge field Lagrangian is built from polynomial combinations of the strength F defined
as
F (Γ (Ω, Dξ) , dΓ) := ∇ ∧ ∇ = dΓ + Γ ∧ Γ. (2.152)
1b) Similarly to the case in (1a), the 1st conformal BIs are respectively given by,
GL
b α ∧ ϑβ − Φ ∧ Ka − 2d (Φ ∧ ϑα ) ,
∇Ka = R (2.154)
β
◦
2a) The Υ and Γ-affine component of the 2nd BI is given by
GL GL
†
∇Rαβ = 2R(α|γ Υγ|β) , † ∇Rαβ = 0, (2.155)
∇Ψ = T α ∧ ϑα + Vα ∧ Kα + Φ ∧ (Vα ∧ ϑα ) . (2.160)
Defining
∇Ψ = T α ∧ ϑα + Vα ∧ Kα . (2.162)
Note that the action integral (2.165) is invariant under Lorentz rather than CA transforma-
tions. The Lie star ⋆ operator is defined as ⋆Vα = 3!1 ηαβµν V β ∧ V µ ∧ V ν .
The field equations are obtained from variation of I with respect to the independent
gauge potentials. It is convenient to define the functional derivatives
δLgauge GL V
δV α := − ∇Nα + Tα ,
GL ϑ
δLgauge
δϑα := − ∇Mα + Tα , (2.166)
δLgauge GL
Zαβ := c αβ + E
= − † ∇M b αβ .
bα
δΓ β
where
c α := − ∂Lgauge , E
b αβ := ∂Lgauge , Tα := ∂Lgauge , Tα := ∂Lgauge , Θ := ∂Lgauge .
V ϑ
M β β α
∂Rb α ∂Γbα
β
∂V ∂ϑα ∂Φ
(2.167)
The gauge field momenta are defined by
∂Lgauge ∂Lgauge ∂Lgauge
Nα := − ∂T α , Mα := − ∂Kα , Ξ := − ∂Z ,
(2.168)
c[αβ] := Nαβ = −o[α|γ ∂Lgauge c ∂Lgauge
M |β] , M(αβ) := Mαβ = −2o(α|γ |β) .
∂Rγ ∂Rγ
Furthermore, the shear (gauge field deformation) and hypermomentum current forms are
given by
b(αβ) := Uαβ = −V(α ∧ Mβ) + Nβ) − Mαβ , E
E b[αβ] := Eαβ = −V[α ∧ Mβ] + Nβ] ,
(2.169)
The analogue of the Einstein equations read
GL V
ηα + † ∇T⋆α + Tα = 0,
Gα + Λb (2.170)
D E D E 1D β E γ
+ Kβ |eα ∧ Mβ + Rγβ |eα ∧ N γβ + Rγ |eα M β ,
2
226 G. Basini and S. Capozziello
We remark that to interpret (2.171) as the gravitational field equation analogous to the Ein-
stein equations, we must transform from the Lie algebra index α to the spacetime basis
index k, by contracting over the former (α) with the CA tetrads eαk .
V D E D E
Tα = Tα [T ] + Tα [K] + Tα [R] + Tα [Z] − T β |eα ∧ Nβ − Kβ |eα ∧ M(2.173)
β +
D E
− Rγβ |eα ∧ N γβ − hZ|eα i ∧ Ξ + Ψ⋆αβ ∧ ϑβ + hΣ⋆γβ |eα i ∧ R
b αβ +
D E D E
+ Υγβ ∧ (Vγ ∧ T⋆β + ϑγ ∧ K⋆β ) |eα + Σ⋆γβ ∧ Rb γβ |eα +
D E
B⋆γβ ∧ B γβ |eα + hB⋆γβ |eα i ∧ Bγβ + hΨ⋆γβ |eα i ∧ Ψγβ
respectively, with
GL ϑ
ωα + † ∇K⋆α + Tα = 0,
Gα + Λb (2.175)
ϑ ϑ
with ωbα = ωα + ⋆ (ϑα ∧ Vβ ) ∧ ϑβ . The quantity Ti = hαi Tα is similar to (2.172) but with
the algebra basis eα replaced by hα and the CA tetrad components eαi replaced by hαi . The
two gravitational field equations (2.171) and (2.176) are P − ∆ symmetric. We may say
that they exhibit P − ∆ duality symmetry invariance.
◦
From the variational equation for Γ αβ we obtain the CA gravitational analogue of the
Yang-Mills-torsion type field equation,
◦ ◦
∇ ⋆ Rαβ + ∇ ⋆ Σαβ + V β ∧ T⋆α + ϑβ ∧ K⋆α = 0. (2.177)
Finally, from the variational equation for Φ, the gravi-scalar field equation is given by
d ⋆ dΦ + Vα ∧ T ⋆α + ϑα ∧ K⋆α = 0. (2.179)
The field equations of CA gravity were obtained in this section. The analogue of the
Einstein equation, obtained from variation of I with respect to the coframe V , is character-
ized by an Einstein-like 3-form that includes symmetric GL4 as well as special conformal
contributions. Moreover, the field equation in (2.171) contains a non-trivial torsion contri-
bution. Performing a P − ∆ transformation ( i.e. V → ϑ, T → K, D → −D) on (2.171)
we obtain (2.176). This result may also be obtained directly by varying I with respect ϑ.
A mixed CA cosmological constant term arises in (2.171), (2.176)) as a consequence of the
structure of the 2-form Rαβ .
The field equation (2.177) is a Yang-Mills-like equation, that represents the generaliza-
tion of the Gauss torsion-free equation ∇ ⋆ B αβ = 0. In our case, we considered a mixed
volume form involving both V and ϑ leading to the substitution B αβ → Σαβ . Additionally,
◦
even in the case of vanishing T ρ = ∇V ρ , the CA torsion depends on the dilaton potential
Φ, which, in general, is non-vanishing. A similar argument holds for the special conformal
quantity Kρ . Admitting the quadratic curvature term Rβα ∧ ⋆Rαβ in the gauge Lagrangian,
it becomes clear how we draw the analogy between (2.177) and the Gauss equation. Eq.
(2.178) follow from similar considerations as (2.177), the significant differences being the
◦ ◦
lack of a ∇ ⋆ Rαβ counterpart to ∇ ⋆ Rαβ since ⋆Rαβ = 0. Finally, (2.179) involves both T ρ
and Kρ in conjunction with a term that resembles the source-free Maxwell equations, with
the dilaton potential playing a similar role to the electromagnetic vector potential.
where
δL(x) = L′ (x′ ) − L(x). (3.7)
For any function Φ(x) of x, it is convenient to define the fixed point variation δ0 by,
we obtain
δΦ(x) = Φ′ (x′ )−Φ(x) = Φ′ (x′ )−Φ(x′ )+Φ(x′ )−Φ(x) = δ0 Φ(x)+δxj ∂j Φ(x), (3.10)
or
δ0 Φ(x) = δΦ(x) − δxj ∂j Φ(x). (3.11)
The advantage to have the fixed point variation is that δ0 commutes with ∂j :
If we require that the action integral, defined over any arbitrary region Ω, be invariant, i.e.
δI = 0, then we must have
δL = 0. (3.19)
The newly introduced function L(χ, ∂j χ; x) is the scalar Lagrangian of the system.
Let us calculate the integrand of (3.15) explicitly. The fixed point variation of L(x) is a
consequence of a fixed point variation of the field χ(x),
∂L ∂L
δ0 L = δ0 χ + δ0 (∂j χ) (3.20)
∂χ ∂(∂j χ)
which can be cast into the form,
∂L
δ0 L = [L]χ δ0 χ + ∂j δ0 χ (3.21)
∂(∂j χ)
where
∂L ∂L
[L]χ ≡ − ∂j . (3.22)
∂χ ∂(∂j χ)
Consequently, we have the action integral in the form
Z
∂L j k
δI = [L]χ δ0 χ + ∂j δχ − Tk δx d4 x, (3.23)
Ω
∂(∂ j χ)
where
∂L
T j k := ∂k χ − δkj L (3.24)
∂(∂j χ)
is the canonical energy-momentum tensor density. If the variations are chosen in such a
way that δxj = 0 over Ω and δ0 χ vanishes on the boundary of Ω, then δI = 0 gives us the
Euler-Lagrange equation,
∂L ∂L
[L]χ = − ∂j = 0. (3.25)
∂χ ∂(∂j χ)
On the other hand, if the field variables obey the Euler-Lagrange equation, [L]χ = 0, then
we have
∂L j k
∂j δχ − T k δx = 0, (3.26)
∂(∂j χ)
which gives rise, considering also the Noether theorem, to conservation laws. These very
straightforward considerations are at the basis of our following discussion.
230 G. Basini and S. Capozziello
where aij and bi are real constants, and aij satisfy the orthogonality conditions aik akj = δji .
For infinitesimal variations, we have
where εij + εji = 0. While the Lorentz transformation forms a six parameter group, the
Poincaré group has ten parameters. The Lie algebra for the ten generators of the Poincaré
group is
[Ξij , Ξkl ] = ηik Ξjl + ηjl Ξik − ηjk Ξil − ηil Ξjk ,
(3.29)
[Ξij , Tk ] = ηjk Ti − ηik Tj , [Ti , Tj ] = 0,
where Ξij are the generators of Lorentz transformations, and Ti are the generators of four-
dimensional translations. Obviously, ∂i (δxi ) = 0 for the Poincaré transformations (3.27).
Therefore, our Lagrangian density L, which is the same as L with h(x) = 1 in this case, is
invariant; namely, δL = δL = 0 for δI = 0.
Suppose that the field χ(x) transforms under the infinitesimal Poincaré transformations
as
1
δχ = εij Sij χ, (3.30)
2
where the tensors Sij are the generators of the Lorentz group, satisfying
Sij = −Sji , [Sij , Skl ] = ηik Sjl + ηjl Sik − ηjk Sil − ηil Sjk . (3.31)
∂L ∂L
Sij χ + (Sij ∂k χ + ηki ∂j χ − ηkj ∂i χ) = 0. (3.33)
∂χ ∂(∂k χ)
where
∂L
S k ij := − Sij χ. (3.35)
∂(∂k χ)
These conservation laws imply that the energy-momentum and angular momentum
Z Z
0
Pl = Tl0 d3 x, Jij = S ij − xi T 0 j − xj T 0 i d3 x, (3.36)
are conserved. This means that the system, invariant under the ten parameter symmetry
group, has ten conserved quantities. This is an example of Noether symmetry. The first
term of the angular momentum integral corresponds to the spin angular momentum while
the second term gives the orbital angular momentum. The global Poincaré invariance of a
system means that, for the system, the spacetime is homogeneous (all spacetime points are
equivalent) as dictated by the translational invariance and is isotropic (all directions about
a spacetime point are equivalent) as indicated by the Lorentz invariance. It is interesting to
observe that the fixed point variation of the field variables takes the form
1 j
δ0 χ = ε k Ξj k χ + εj Tj χ, (3.37)
2
where
Ξj k = Sj k + xj ∂k − xk ∂j , Tj = −∂j . (3.38)
We remark that Ξj k are the generators of the Lorentz transformation and Tj are those of
the translations.
which we call the local Poincaré transformations (or the general coordinate transforma-
tions). In order to make a distinction between the global transformation and the local trans-
formation, we use the Latin indices (j, k = 0, 1, 2, 3) for the former and the Greek indices
(µ, ν = 0, 1, 2, 3) for the latter. The variation of the field variables χ(x) defined at a point
x is still the same as that of the global Poincaré transformations,
1
δχ = εij S ij χ. (3.40)
2
The corresponding fixed point variation of χ takes the form,
1
δ0 χ = εij S ij χ − ξ ν ∂ν χ. (3.41)
2
Differentiating both sides of (3.41) with respect to xµ , we have
1 1
δ0 ∂µ χ = εij Sij ∂µ χ + (∂µ εij ) S ij χ − ∂µ (ξ ν ∂ν χ). (3.42)
2 2
232 G. Basini and S. Capozziello
δh + h∂µ ξ µ = 0. (3.44)
Now we consider a minimal modification of the Lagrangian so then to make the action
integral invariant under the local Poincaré transformations. It is rather obvious that if there
is a covariant derivative ∇k χ which transforms as
1 ij
δ(∇k χ) = ε Sij ∇k χ − εi k ∇i χ, (3.45)
2
then a modified Lagrangian L′ (χ, ∂k χ, x) = L(χ, ∇k χ, x), obtained by replacing ∂k χ of
L(χ, ∂k χ, x) by ∇k χ, remains invariant under the local Poincaré transformations, that is
∂L′ ∂L′
δL′ = δχ + δ(∇k χ) = 0. (3.46)
∂χ ∂(∇k χ)
To find such a k-covariant derivative, we introduce the gauge fields Aij µ = −Aji µ and
define the µ-covariant derivative
1
∇µ χ := ∂µ χ + Aij µ Sij χ, (3.47)
2
in such a way that the covariant derivative transforms as
1
δ0 ∇µ χ = Sij ∇µ χ − ∂µ (ξ ν ∇ν χ). (3.48)
2
The transformation properties of Aabµ are determined by ∇µ χ and δ∇µ χ. Making use of
1 ij 1 1 1
δ∇µ χ = ε ,µ Sij χ + εij Sij ∂µ χ − (∂µ ξ ν ) ∂ν ψ + δAij µ Sij χ + Aij µ Sij εkl Skl χ
2 2 2 4
(3.49)
and comparing with (3.47) we obtain,
1 ij kl
δAij µ Sij χ + εij ,µ Sij χ + A µ ε − εij Aklµ Sij Skl χ + (∂µ ξ ν ) Aij ν Sij χ = 0. (3.50)
2
Using the antisymmetry in ij and kl to rewrite the term in parentheses on the RHS of (3.50)
as [Sij , Skl ] Aij µ εkl χ, we see the explicit appearance of the commutator [Sij , Skl ]. Using
the expression for the commutator of Lie algebra generators
1 [ef ]
[Sij , Skl ] = c [ij][kl] Sef , (3.51)
2
The Conformal-Affine Structure of open Quantum Relativity... 233
[ef ]
where c [ij][kl] (the square brackets denote anti-symmetrization) is the structure constants
of the Lorentz group (deduced below), we have
1 ic j
[Sij , Skl ] Aij µ εkl = Aµ εc − Acj i
µ εc Sij . (3.52)
2
The substitution of this equation and the consideration of the antisymmetry of εcb = −εbc
enables us to write
∇k χ = Vk µ (x)∇µ χ, (3.54)
where the coefficients Vk µ (x) are position-dependent and behave like a new set of field
variables. From (3.54) it is evident that ∇k χ varies as
δVk µ = Vk ν ∂ν ξ µ − Vi µ εi k . (3.57)
It is also important to recognize that the inverse of det(Vk µ ) transforms like a scalar density
as h(x) does. For our minimal modification of the Lagrangian density, we utilize this
available quantity for the scalar density h; namely, we let
When we consider Poincaré transformations, that are not spacetime dependent, we have
Vk µ → δkµ so that h(x) → 1. This is a desirable property, because we replace the La-
grangian density L(χ, ∂k χ, x), invariant under the global Poincaré transformations, by a
Lagrangian density
L(χ, ∂µ χ; x) → h(x)L(χ, ∇k χ), (3.59)
the action integral, with this modified Lagrangian density, remains invariant under the local
Poincaré transformations. Since the local Poincaré transformations δxµ = ξ µ (x) are noth-
ing else but generalized coordinate transformations, the newly introduced gauge fields Viλ
and Aij µ can be interpreted, respectively, as the tetrad (vierbein) fields which set the local
coordinate frame and as a local affine connection, with respect to the tetrad frame (see also
[3]).
234 G. Basini and S. Capozziello
where S(Λ) is an irreducible unitary representation of the Lorentz group. Since the bilinear
form v k = iψγ k ψ is a vector, it transforms according to
v j = Λjk v k , (3.61)
The invariance of v i (or the covariance of the Dirac equation) under the transformation
ψ(x) → ψ ′ (x′ ) leads to
S −1 (Λ)γ µ S(Λ) = Λµν γ ν , (3.63)
where the γ ′ s are the Dirac γ-matrices satisfying the anticommutator,
γi γj + γj γi = ηij 1. (3.64)
we get
1
S(Λ) = 1 + εij γij . (3.67)
2
In order to determine the form of γij , we substitute (3.66) and (3.67) into (3.63) to obtain
1 h ij k i
εij γ , γ = η ki εji γ j . (3.68)
2
Rewriting the RHS of (3.68) using the antisymmetry of εij as
1
η ki εji γ j = εij η ki γ j − η kj γ i , (3.69)
2
The Conformal-Affine Structure of open Quantum Relativity... 235
it yields h i
γ k , γ ij = η ki γ j − η kj γ i . (3.70)
Since the solution has the form of an antisymmetric product of two matrices, we obtain
1 i j
γ ij := γ ,γ . (3.71)
2
If χ = ψ, the group generator Sij appearing in (3.31) is identified with
1
Sij ≡ γij = (γi γj − γj γi ). (3.72)
2
To be explicit, the Dirac field transforms under Lorentz transformations (LT) as
1
δψ(x) = εij γij ψ(x). (3.73)
2
1
S(Λ(x)) = 1 + εab (x)γ ab . (3.77)
2
Since εab (x) is only a function of spacetime for local Lorentz coordinates, we express this
infinitesimal LT in terms of general coordinates only by shifting all spacetime dependence
of the local coordinates into tetrad fields as
Being
ωµba := Vb ν (x)∂µ Vaν (x), (3.81)
1
the first and second terms in Eq.(3.80) become Vaλ (x)∂µ Vbλ (x) = 2 ωµab and
Vbν (x)∂µ Vaν (x) = 21 ωµba respectively. Using the identification
we write
1
∂µ S(Λ(x)) = − γ ab ωµab . (3.83)
2
According to (3.47), the covariant derivative of the Dirac spinor is
1
∇µ ψ = ∂µ ψ + Aij µ γij ψ. (3.84)
2
Correspondingly, the covariant derivative of ψ̄ is given by
1
∇µ ψ = ∂µ ψ − Aij µ ψ̄γij . (3.85)
2
Using the covariant derivatives of ψ and ψ̄, we can show that
∇µ vj = ∂µ vj − Ai jµ vi . (3.86)
The same covariant derivative should be used for any covariant vector vk under the Lorentz
transformation. Since ∇µ (vi v i ) = ∂µ (vi v i ), the covariant derivative for a contravariant
vector v i must be
∇µ v i = ∂µ v i + Ai jµ v j . (3.87)
Since the tetrad Vi µ is a covariant vector under Lorentz transformations, its covariant
derivative must transform according to the same rule and using ∇a = Vaµ (x)∇µ , the co-
variant derivatives of a tetrad in local Lorentz coordinates read
∇ν Vi µ = ∂ν Vi µ − Ak iν Vk µ , ∇ν V i µ = ∂ν V i µ + Ai kν V k µ . (3.88)
V i µ Vi ν = δ µ ν , V i µ Vj µ = δ i j . (3.89)
Now, to allow the transition to curved spacetime, we take account of the general coordi-
nates of objects that are covariant under local Poincaré transformations. Here we define the
covariant derivative of a quantity v λ , which behaves like a contravariant vector under the
local Poincaré transformation. Namely
Dν v λ ≡ Vi λ ∇ν v i = ∂ν v λ + Γλ µν v µ , Dν vµ ≡ V i µ ∇ν vi = ∂ν vµ − Γλ µν vλ , (3.90)
The Conformal-Affine Structure of open Quantum Relativity... 237
where
Γλ µν := Vi λ ∇ν V i µ ≡ −V i µ ∇ν Vi λ . (3.91)
The definition of Γλ µν implies
Dν Vi λ = ∇ν Vi λ + Γλ µν Vi µ = ∂ν Vi λ − Ak iν Vk λ + Γλ µν Vi µ = 0, (3.92)
Dν V i µ = ∇ν V i µ − Γλ µν V i λ = ∂ν V i µ + Ai kν V k µ − Γλ µν V i λ = 0.
Ai kν = V i λ ∂ν Vk λ + Γλ µν V i λ Vk µ = −Vk λ ∂ν V i λ + Γλ µν V i λ Vk µ . (3.93)
where
1 i
Γµ = ω jµ − Γλ µν V i λ Vj ν γi j , (3.96)
4
which is known as the Fock-Ivanenko connection.
We now study the transformation properties of Aµab . Recall ωµab = Va λ (x)∂µ Vβλ (x)
and since ∂µ ηab = 0, we write
Note that barred indices are equivalent to the primed indices used above. Hence, the spin
connection transforms as
Aabc = Λaa Λbb Λcc Aabc + Λaa Λcc V µa (x)∂µ Λbc . (3.98)
we consider the local LT of [Va µ (x)∂µ V νb (x)] Vνc (x) which is,
h i
V µa (x)∂µ V νb Vνc (x) = Λaa Λb b Λcc [Aνab Vνc (x)] + Λaa Λc c V µa (x)∂µ Λcb . (3.100)
We now explore the consequence of the antisymmetry of ωabc in bc. Recalling the
equation for Γabc , exchanging b and c and adding the two equations, we obtain
Γabc + Γacb = −V µa (x) [(∂µ V νb (x)) Vνc (x) + (∂µ V νc (x)) Vνb (x)] . (3.102)
238 G. Basini and S. Capozziello
∂µ [V νb (x)Vνc (x)] = Vνc (x)∂µ V νb (x) + Vλb (x)∂µ V λc (x) + Vb ν (x)V λc (x)∂µ gλν . (3.103)
so that, finally,
Γabc + Γacb = Va µ (x)Vb λ (x)Vc ν (x)∂µ gνλ . (3.105)
This, however, is equivalent to
and then
Γµλν + Γµνλ = ∂µ gνλ , (3.107)
which we recognize as the general coordinate connection. It is known that the covariant
derivative for general coordinates is
In a Riemannian manifold, the connection is symmetric under the exchange of µν, that is,
Γλµν = Γλνµ . Using the fact that the metric is a symmetric tensor we can now determine
the form of the Christoffel connection by cyclically permuting the indices of the general
coordinate connection equation (3.107), yielding
1
Γµνλ = (∂µ gνλ + ∂ν gλµ − ∂λ gµν ) . (3.109)
2
Since Γµνλ = Γνµλ is valid for general coordinate systems, it follows that a similar
constraint must hold for local Lorentz transforming coordinates as well, so we expect
Γabc = Γbac . Recalling the equation for Γabc and exchanging a and b, we obtain
ωabc − ωbac = Vνc (x) V µa (x)∂µ V νb (x) − V µb (x)∂µ V νa (x) . (3.110)
Using Ωcab = −Ωcba , we permute indices in a similar manner as was done for the derivation
of the Christoffel connection above yielding,
1
ωabµ = [Ωcab + Ωbca − Ωabc ] V cµ ≡ ∆abµ . (3.112)
2
For completeness, we determine the transformation law of the Christoffel connection. Mak-
ing use of Γλµν eλ = ∂µ eν where
∂µ eν = X µµ X νν ∂µ eν + X µµ (∂µ X νν ) eν , (3.113)
The Conformal-Affine Structure of open Quantum Relativity... 239
we can show
Γλµ ν = X µµ X νν Xλλ Γλµν + X µµ Xν λ X νµν , (3.114)
where
X νµν ≡ ∂µ ∂ν xν . (3.115)
In the light of the above considerations, we may regard infinitesimal local gauge trans-
formations as local rotations of basis vectors, belonging to the tangent space [16, 57] of the
manifold. For this reason, given a local frame on a tangent plane to the point x on the base
manifold, we can obtain all other frames on the same tangent plane, by means of local rota-
tions of the original basis vectors. Reversing this argument, we observe that by knowing all
frames residing in the horizontal tangent space to a point x on the base manifold, it enables
us to deduce the corresponding gauge group of symmetry transformations.
Noting that the Christoffel connection is symmetric and partial derivatives commute, we
find
1 h i 1 h i
[Dµ , Dν ] ψ = Scd ∂ν Acdµ − ∂µ Acdν ψ + Sef Scd Aefν Acdµ − Aefµ Acdν ψ ,
2 4
(3.119)
where
∂ν Acdµ − ∂µ Acdν = ∂µ Γcdν − ∂ν Γcdµ . (3.120)
Relabelling running indices, we can write
1 1
Sef Scd Aefν Acdµ − Aefµ Acdν ψ = [Scd , Sef ] Aefµ Acdν ψ. (3.121)
4 4
Using {γa , γb } = 2ηab to deduce
Clearly the terms in brackets on the RHS of (3.123) are antisymmetric in cd and ef and also
antisymmetric under the exchange of pairs of indices cd and ef . Since the alternating spinor
is antisymmetric in ab, it must be the terms in brackets: this means that the commutator
does not vanish. Hence, the term in brackets is totally antisymmetric under interchange
of indices ab, cd and ef and exchange of these pairs of indices. We identify this as the
structure constant of the Lorentz group [58]
h i
ηce δda δfb − ηde δca δfb + ηcf δea δdb − ηdf δea δcb = c[cd][ef ] [ab] = c[ab] [cd][ef ] , (3.124)
Vµ a ∇ν V µk = −Vk µ ∇ν V aµ , (3.133)
(which follows from ∇ν Vµa Vkµ = 0), we arrive at
V νl ∇ν V µk ∇µ ψ − V µk (∇µ V νl ) ∇ν ψ = V µl V νk − V µk V νl ∇ν Vµ a ∇a ψ. (3.134)
The Conformal-Affine Structure of open Quantum Relativity... 241
Defining
C akl := V µk V νl − V µl V νk ∇ν Vµ a , (3.135)
the commutator of the k-covariant differentiations takes the final form [7]
1
[∇k , ∇l ]χ = − Rij kl Sij χ + C i kl ∇i χ, (3.136)
2
where
Rij kl = Vk µ Vl ν Rij µν . (3.137)
As done for Ri jµν using the Jacobi identities for the commutator of covariant derivatives,
we find the Bianchi identity in Einstein-Cartan spacetime [63, 56]
leads to,
εαβρσ ∇β Cρσk = εαβρσ Rkjρσ V βj . (3.140)
Notice that if
Γλµν = Vi λ ∇ν V iµ = −Vµ i ∇ν V λi , (3.141)
then
Γλµν − Γλνµ = Viλ ∇ν V iµ − ∇µ V iν . (3.142)
Contracting by Vkµ Vlν , we obtain [7],
C akl = Vk µ Vl ν Vλ a Γλµν − Γλνµ . (3.143)
We therefore conclude (and stress it) that C akl is related to the antisymmetric part of the
affine connection
Γλ[µν] = Vµ k Vν l Va λ C akl ≡ T λµν , (3.144)
which is usually interpreted as spacetime torsion T λµν . Considering ∆abµ defined in
(3.112), we see that the most general connection, in the Poincaré gauge approach to gravi-
tation, is
Aabµ = ∆abµ − Kabµ + Γλ νµ Vaλ Vb ν , (3.145)
where
Kabc = − T λ νµ − Tνµλ + Tµ λν Vaλ Vb ν Vc µ , (3.146)
ρ
is the contorsion tensor [54]. Now, the quantity Rσµν = Vi ρ Ri σµν may be expressed as
Therefore, we can regard Rρ σµν as the curvature tensor with respect the affine connection
Γλ µν . By using the inverse of the tetrad, we define the metric of the spacetime manifold by
From (3.92) and the fact that the Minkowski metric is constant, it is obvious that the metric
so defined is covariantly constant, that is,
Dλ gµν = 0. (3.149)
The spacetime thus specified by the local Poincaré transformation is said to be metric. It is
not difficult to show that
√
−g = [det V i µ ] = [det Vi µ ]−1 , (3.150)
√
where g = det gµν . Hence we may take −g for the density function h(x).
L
where R denotes the usual Ricci scalar of General Relativity. Using this scalar curvature R,
we choose the Lagrangian density for free Einstein-Cartan gravity
1 √ L
ia bc a
LG = −g R + ∂i Ka − Ta Kbc − 2Λ , (3.153)
2κ
1 a
Lfield = ψγ Da ψ − Da ψ γ a ψ . (3.155)
2
Taking into account the Christoffel symbols, spin connection and contorsion contributions
so then to operate on general spinorial arguments, we have
1 L
Γµ = gλσ ∆σµρ − Γ σρµ − K σρµ γ λρ . (3.156)
4
The Conformal-Affine Structure of open Quantum Relativity... 243
It is important to keep in mind that ∆σµρ act only on multi-component spinor fields, while
L
Γ σρµ act on vectors and arbitrary tensors. The gauge covariant derivative for a spinor and
adjoint spinor is then given by
The 4 × 4 matrices Γ b κ are real matrices, used to induce similarity transformations on quan-
tities with spinor transformation [62] properties, according to
b −1 γi Γ.
γi′ = Γ b (3.160)
b κ leads to,
Solving for Γ
b κ = 1 [(∂κ γι ) γ ι − Γµ γµ γ ι ] .
Γ (3.161)
ικ
8
bκ ,
Then, taking the variation of Γ
bκ = 1 (∂κ δγι ) γ ι + (∂κ γι ) δγ ι − (δΓµικ ) γµ γ ι
δΓ (3.162)
8 −Γµικ ((δγµ ) γ ι + γµ δγ ι )
1
= [(∂κ δγι ) γ ι − (δΓµικ ) γµ γ ι ] .
8
Since we require the anticommutator condition on the gamma matrices {γ µ , γ ν } = 2g µν to
hold, the variation of the metric gives
1
δγ ν = γσ δγ σν . (3.164)
2
With the aid of this result, we can write
1
(∂κ δγι ) γ ι = ∂κ (γ ν δgνι ) γ ι . (3.165)
2
Finally, exploiting the anti-symmetry in γµν we obtain
bκ = 1
δΓ gνσ δΓµκσ − gµσ δΓνκσ γ µν . (3.166)
8
244 G. Basini and S. Capozziello
The field Lagrangian, defined in the Einstein-Cartan spacetime, can be written [59, 12, 63,
60, 56] explicitly in terms of its Lorentzian and contorsion components as
n o
1 L L ~c
Lfield = Dµ ψ γ µ ψ − ψγ µ Dµ ψ − Kµαβ ψ γ µ , γ αβ ψ. (3.167)
2 8
we obtain n o 1
Kµαβ ψ γ µ , γ αβ ψ = Kµαβ εαβµν ψγ5 γν ψ . (3.169)
2i
Here we define the contorsion axial vector
1 αβµν
Kν := ε Kαβµ . (3.170)
3!
√ 1√
δ −g = − −g δV iµ Vµi + Vνi δV νi . (3.174)
2
By using
δV νi = δ η ij V jν = η ij δV jν , (3.175)
we are able to deduce
√ √
δ −g = − −gVµi δVi µ . (3.176)
µ
For the variation of the Ricci tensor Riν = Vi Rµν , so we have
L L L
δ Riν = δVi µ Rµν + Vi µ δ Rµν . (3.177)
The Conformal-Affine Structure of open Quantum Relativity... 245
so that
L L L L
µ µ β β
δ Riν = δVi Rµν + Vi ∂ν δ Γ βµ − ∂β δ Γ νµ . (3.179)
The second term can be converted into a surface term, so it may be ignored. Collecting our
results, we have
δg µν = −g µρ g νσ δgρσ ,
√ √ √
δ −g = − 21 −ggµν δg µν = − −gVµi δVi µ ,
L L
λρ λρ ρ λ , δ R = δV µ R (3.180)
δR µν = gρµ ∇ λ δΓ ν − ∇ ν δΓ λ + T λµ δΓ ρν iν i µν
L L L
δR = R µν δgµν + g µν ∇λ δ Γ λ µν − ∇ν δ Γ λµλ − Ta bc δKbca .
The last term in the action can be ignored being a surface term. Using the four-current v µ
introduced earlier, the action for the matter fields read [62]
Z h i
δIfield = ψδγ µ ∇µ ψ + ψγ µ δ Γb µ ψ √−gd4 x (3.182)
1 µν
Z 2 g ψγi (∇ν ψ) + T µρσ Ti ρσ − δiµ Tλρσ T λρσ δV µi
√
= L L −gd4 x.
+ 81 (g ρν v µ − g ρµ v ν ) gµσ δ Γ σνρ − gνσ δ Γ σµρ
Removing the derivatives of variations of the metric appearing in δΓσνρ via partial integra-
tion, and equating to zero the coefficients of δg µν and δT σνρ in the variation of the action
integral, we obtain
1 1 1 1
0 = Rµν − gµν R − gµν Λ + ψγν ∇µ ψ − ∇µ vν (3.183)
16π 2 2 4
+∇σ Tµνσ + Tµρσ Tνρσ − gµν Tλρσ T λρσ
and
Tρσλ = 8πτρσλ . (3.184)
Eqs.(3.183) have the form of Einstein equations
∂Lfield
Θµν = ∇ν χ − δ µν Lfield , (3.188)
∂(∇µ χ)
while Tµν is the stress-tensor form of the non-Riemannian manifold. For the case of spinor
fields, being considered here the explicit form of the energy-momentum components [61],
are (after symmetrization of corresponding canonical source terms in the Einstein equation),
Θµν = − ψγµ ∇ν ψ − ∇ν ψ γµ ψ + ψγν ∇µ ψ − ∇µ ψ γν ψ (3.189)
∂Lfield
τµνσ := γµν χ. (3.191)
∂(∇σ χ)
Explicitly, the spin energy potential reads τ µνσ = ψγ [µ γ ν γ σ] ψ. The equation of motion
obtained from the variation of the action with respect to ψ reads [12, 63]
3
γ µ ∇µ ψ + Tµνσ γ [µ γ ν γ σ] ψ = 0. (3.192)
8
It is interesting to observe that this generalized curved spacetime Dirac equation can be
recast into the nonlinear equation of the Heisenberg-Pauli type
3
γ µ ∇µ ψ + ε ψγ µ γ5 ψ γµ γ5 ψ = 0. (3.193)
8
Although the gravitational field equation is similar in form to the Einstein field equa-
tion, it differs from the original Einstein equations, because the curvature tensor, containing
spacetime torsion, is non-Riemannian. Assuming that the Euler-Lagrange equations for
the matter fields are satisfied, we obtain the following conservation laws for the angular -
momentum and energy - momentum
V µi V νj Σ[µν] = ∇ν τijν ,
(3.194)
Vµ k ∇ν Σνκ = Σνκ T kµν + τ νij Rijµν .
In conclusion, we can say that considering gravity as a local Poincaré gauge invariance
means that conservation laws emerges as a consequence of dynamics and not only as a first
principle.
The Conformal-Affine Structure of open Quantum Relativity... 247
where the Latin indexes are A, B = 0, 1, 2, 3, 4. We do not need yet to specify the 5D
signature, because, in 4D, it is dynamically fixed by the reduction procedure as we shall
248 G. Basini and S. Capozziello
see below. The curvature invariants, the field equations and the conservation laws in the
5D-space can be defined as follows. In general, we ask for a space which is a singularity
free, smooth manifold, where conservation laws are always valid [43]. The 5D-Riemann
tensor is
D
RABC = ∂B ΓD D D E
AC − ∂C ΓAB + ΓEB ΓAC − ΓEC ΓAB
D E
(4.2)
and the Ricci tensor and scalar are derived from the contractions
C (5) A
RAB = RACB , R = RA . (4.3)
where (5) G is the 5D-gravitational coupling and g (5) is the determinant of the 5D-metric
[2]. The 5D-field equations are
1
GAB = RAB − gAB (5) R = 0 , (4.5)
2
so that at least the Ricci-flat space is always a solution. Let us define now a 5D-stress-energy
tensor for a scalar field Φ:
1
TAB = ∇A Φ∇B Φ − gAB ∇C Φ∇C Φ , (4.6)
2
where only the kinetic terms are present. As standard, such a tensor can be derived from a
variational principle p
δ −g (5) L
2 Φ
T AB = p , (4.7)
−g (5) δgAB
where LΦ is a Lagrangian density related to the scalar field Φ. Because of the definition
of 5D space itself, based on the conservation laws [43], it is important to stress now that
no self-interaction potential U (Φ) has to be taken into account so that TAB is a completely
symmetric object and Φ is, by definition, a cyclic variable. In this situation the Noether
theorem always holds for TAB . With these considerations in mind, the field equations can
assume the form
1
RAB = χ TAB − gAB T , (4.8)
2
where T is the trace of TAB and χ = 8π (5) G.
∇A TBA = 0 , ∇A GA
B = 0, (4.9)
The Conformal-Affine Structure of open Quantum Relativity... 249
where (5) is the 5D d’Alembert operator defined as ∇A ΦA ≡ g AB Φ,A;B ≡ (5) Φ. The
general result is that the conservation of the stress-energy tensor TAB (i.e. the contracted
Bianchi identities) implies the Klein-Gordon equation which assigns the dynamics of Φ,
that is
∇A TBA = 0 ⇐⇒ (5)
Φ = 0 . (4.11)
Let us note again the absence of self-interactions due to the absence of potential terms. The
relations (4.11) give a physical meaning to the fifth dimension. Splitting the 5D-problem
in a (4 + 1)-description, it is possible to generate the mass of particles in 4D. Such a result
can be deduced both from Eq.(4.11) and from the analysis of the geodesic structure, as we
are going to show.
d2 xA dxB dxC
2
+ ΓA
BC = 0. (4.13)
dS dS dS
ΓA
BC are the 5D-Christoffel symbols. Eq.(4.13) can be split in the (4 + 1) form
2 µ
dxα d x β
µ dx dx
γ
2gαµ + Γβγ +
ds ds2 ds ds
∂gαβ dx4 dxα dxβ
+ 4 = 0, (4.14)
dx ds ds ds
where the Greek indexes are µ, ν = 0, 1, 2, 3 and ds2 = gαβ dxα dxβ . Clearly, in the 4D-
reduction (i.e. in the usual spacetime), we usually experience the standard geodesics of
General Relativity, i.e. the 4D component of Eq.(4.14)
d2 xµ β
µ dx dx
γ
+ Γ βγ = 0, (4.15)
ds2 ds ds
so that, under these conditions, the last part of the representation given by Eq.(4.14) is
not detectable in 4D. In other words, for standard laws of physics, the metric gαβ does
not depend on x4 in the embedded 4D manifold. On the other hand, the last component
of Eq.(4.14) can be read as an ”Extra Force” which gives the motion of a 4D frame with
250 G. Basini and S. Capozziello
respect to the fifth coordinate x4 . This fact shows that the fifth dimension has a real phys-
ical meaning and any embedding procedure scaling up in 5D-manifold (or reducing to 4D
spacetime) has a dynamical description. The Extra Force
Splitting Eq.(4.17) into the 4D part and the fifth component, gives
An inspection of Eq.(4.18) tells that a null path in 5D can result, in 4D, in a time-like path,
a space-like path, or a null path depending on the sign and the value of g44 . Let us consider
now the 5D-vector uA = dxA /dS. It can be split as a vector in the ordinary 3D-space v, a
vector along the ordinary time axis w and a vector along the fifth dimension z. In particular,
for 5D null paths, we can have the velocity v 2 = w2 + z 2 and this should lead, in 4D, to
super-luminal speed, explicitly overcoming the Lorentz transformations. The problem is
solved if we consider the 5D-motion as a-luminal, because all particles and fields have the
same speed (being massless) and the distinction among super-luminal, luminal and sub-
luminal motion (the standard causal motion for massive particles) emerges only after the
dynamical reduction from 5D-space to 4D spacetime. In this way, the fifth dimension is the
entity which, by assigning the masses, is able to generate the different dynamics which we
perceive in 4D. Consequently, it is the process of mass generation which sets the particles
in the 4D light-cone. Specifically, let us rewrite the expression (4.16) as
∂gµν dx4 µ ν
F= u u (4.19)
∂x4 ds
As we said, seen in 4D, this is an Extra Force generated by the motion of the 4D frame
with respect to the extra coordinate x4 . This fact shows that all the different particles are
massless in 5D and acquire their rest masses m6 in the dynamical reduction from the 5D to
4D. In fact, considering Eqs.(4.17) and (4.18), it is straightforward to derive
where m0 has the role of a rest mass in 1D, being, from General Relativity,
dxµ 1 ∂gαβ α β
− u u =0 (4.21)
ds 2 ∂xµ
and
pµ = m0 uµ , pµ pµ = m24 , (4.22)
which are, respectively, the definition of linear momentum and the mass-shell condition.
Then, it is
∂gµν µ ν 4
d ln(m3 ) = u u dx (4.23)
∂x4
that is Z Z
∂gµν µ ν 4
m0 = exp ( 1 u u dx ) = exp (Fdx4 ) (4.24)
∂x
R ∂g
In principle, the term ( ∂xµν4 uµ uν dx4 ) never gives a zero mass. However, this term can
be less than zero and, with large absolute values, it can asymptotically produce a m0 very
close to zero. In conclusion the Extra Force induced by the reduction from the 5D to the 4D
is equal to the derivative of the natural logarithm of the rest mass of a particle, with respect
to the (3 + 1) line element and the expression
Z Z
∂gµν
( 4 uµ uν dx4 ) = (Fdx4 ) (4.25)
∂x
can be read as the total ”work” capable of generating masses in the reduction process from
5D to 4D.
derived from (4.4) where λ is a Lagrange multiplier, Φ a generic scalar field and ǫ = ±1.
This procedure allows to derive the physical gauge for the 5D metric. The above 5D metric
can be immediately rewritten as
dS 2 = gAB dxA dxB = gαβ dxα dxβ + g44 (dx4 )2
= gαβ dxα dxβ + ǫΦ2 (dx4 )2 (4.27)
252 G. Basini and S. Capozziello
is the effective stress–energy tensor containing the nonminimal coupling contributions, the
kinetic terms and the potential of the scalar field ϕ. By varying with respect to ϕ, we get
the 4D Klein-Gordon equation
Separating the variables and splitting the scalar field Φ into two functions
the field ϕ depends on the ordinary space-time coordinates, while χ is a function of the fifth
coordinate x4 . Inserting (4.42) into Eq.(4.41), we get
ϕ 1 d2 χ
= = −kn2 (4.43)
ϕ χ dx24
where kn is a constant. From Eq.(4.43), we obtain the two field equations
+ kn2 ϕ = 0 , (4.44)
254 G. Basini and S. Capozziello
and
d2 χ
+ kn2 χ = 0 . (4.45)
dx24
Eq.(4.45) describes a harmonic oscillator whose general solution is
4 4
χ(x4 ) = c1 e−ikn x + c2 eikn x . (4.46)
The constant kn has the physical dimension of the inverse of a length and, assigning bound-
ary conditions, we can derive the eigenvalue relation
2π
kn = n, (4.47)
l
where n is an integer and l a length which we have previously defined in Eq.(4.32) related
to the gravitational coupling. As a result, in standard units, we can recover the physical
lengths through the Compton lengths
~ 1
λn = = (4.48)
2πmn c kn
which always assign the masses to the particles depending on the number n. It is worth
stressing that, in this case, we have achieved a dynamical approach because the eigenvalues
of Eq.(4.45) are the masses of particles which are generated by the process of reduction from
5D to 4D. The solution (4.46) is the superposition of two mass eigenstates. The 4D evolu-
tion is given by Eq.(4.38). Besides, the solutions in the coordinate x0 give the associated
Compton lengths from which the effective physical masses can be derived. Specifically, dif-
ferent values of n fix the families of particles, while, for any given value n, different values
of parameters c1,2 select the different particles within a family. With these considerations in
mind, the effective mass can be obtained integrating the modulus of the scalar field Φ along
the x4 coordinate. It is
Z Z
mef f ≡ |Φ|dx4 = |Φ(dx4 /ds)|ds (4.49)
where ds is the 4D affine parameter used in the derivation of geodesic equation. This result
means that the rest mass of a particle is derived by integrating the Extra Force along x4
while the effective mass is obtained by integrating the field Φ along x4 . In the first case,
the mass of the particle is obtained starting from the geodesic structure of the theory, in
the second case, it comes out from the field structure. In other words, the coincidence of
geodesic structure and chronological structure (the causal structure), supposed as a principle
in General Relativity, is due to the dynamical fact that masses are generated in the reduction
process.
At this point, from the condition (4.42), the 5D-field Φ results to be
+∞ h
X i
4 4
Φ(xα , x4 ) = ϕn (xα )e−ikn x + ϕ∗n (xα )eikn x , (4.50)
n=−∞
4
where ϕ and ϕ∗ are the 4D solutions combined with the fifth-component solutions e±ikn x .
In general, every particle mass can be selected by solutions of type (4.46). The number
The Conformal-Affine Structure of open Quantum Relativity... 255
kn x4 , i.e. the ratio between the two lengths x4 /λn , fixes the interaction scale. Geometri-
cally, such a scale is related to the curvature radius of the embedded 4D spacetime where
particles can be identified and, in principle, detected. Is this sense, Open Quantum Relativ-
ity is an induced-matter theory, where the extra dimension cannot be simply classified as
”compactified” since it yields all the 4D dynamics giving origin to the masses. Moreover,
Eq.(4.50) is not a simple ”tower of mass states”, but a spectrum capable of explaining the
hierarchy problem [43]. On the other hand, gravitational interaction can be framed, in this
approach, considering as its fundamental scale the Planck length
1/2
~GN
λP = l = , (4.51)
c3
instead of the above Compton length. It fixes the vacuum state of the system and the masses
of all particles can be considered negligible if compared with the Planck scales. Finally, as
we have seen, the reduction mechanism can select also ǫ = 1 in the above metric. In this
case, the 5D-Klein Gordon equation (4.11), and the 5D field equations (4.5) have wave-like
solutions of the form
where
Ω(t, xj ) = exp i(ωt + kj xj ) , j = 1, 2, 3 . (4.53)
In this solution, the necessity of the existence of two times arrows naturally emerges and, as
a direct consequence, due to the structure of the functions Ω(t, xj ), closed time-like paths
(i.e. circular paths) are allowed. The existence of closed time-like paths means that Anti-De
Sitter [72] and Gödel [69] solutions are naturally allowed possibilities in the dynamics of
Open Quantum Relativity.
4. It has to be possible the existence or the emergence of close-time -curves, and more-
over to be in the framework of the Many Worlds Theory [74, 77]. Besides, even only
the fact to travel back in time can generate a cascade of cumulative effects exceed-
ing, at the end, the uncertainty limit, in such a way violating, in our 4D universe, the
causality principle, whose full recover can be assured only by the existence, in 5D,
of several 4D universes, as many as the possibilities to change.
6. It has to be possible to calculate exactly the spacetime region where and when the
time-travel is going to end (i.e. the Cauchy boundary conditions have to be well-
posed).
7. It has to be possible to know how and when realize, following the Cauchy conditions,
the inverse operation to come back home. In any case, it seems necessary that the
technical device (whatever it will be) that initially produces the requested conserva-
tion law violation, should travel with the passenger, because the inverse operation is
not generally automatic, but should obey to the same law as the initial one.
10. (Last but not least). It has to be possible to technically conceive a gedanken spacetime
craft (and then a real one) in order to use natural time-tunnels and/or a gedanken time
machine to realize artificial time-tunnels.
6. Appendix
6.1. Notations
The notations used along the Report are summarized here.
∂µ = ∂x∂µ : Partial derivative with respect to {xµ }
{eµ } : Set with elements eµ
The Conformal-Affine Structure of open Quantum Relativity... 259
b A = u L ρ (GA ) N ∂
L M L . (6.1)
∂uMN
gλ )MN = uMQ DQN , and DQN is the adjoint representation matrix [31] for the Lie
with (pe
algebra basis GA . Here u is the parameterization matrix of elements ge. For instance, if
ge = exp(λAB GBA ), then uAB := exp(λAB ). In terms of GA we define the canonical g-valued
one-form Θ = g −1 dg = ΘA GA (g ∈ G) on P, inheriting the left invariance of GA in
terms of which it is defined, namely L∗g Θ|gp = Θ|p . The components of Θ read
1 −1 AB L
ΘA = − γ ρ (GB )MN u−1 N duLM , (6.2)
2
AB
where γ −1 is the inverse of the Cartan-Killing metric γAB whose anholonomic com-
ponents are given in terms of GA as [37],
They satisfy
γAB = DAC DBD γCD . (6.4)
D E
b A and one-form Θ satisfy the duality and left invariance conditions, Θ|L
The basis L bA =
GA and Lg∗ : LA|p → LA|gp . The right invariant basis vector operators are given by
b A := ρ (GA ) L u N ∂
R M L , (6.5)
∂uMN
A
while the canonical right invariant g-valued one-form Θ = (dg) g −1 = Θ GA , where
A 1 −1 AB N M
Θ =− γ ρ (GB )M duNL u−1 L
(6.6)
2
D E
b A = GA . We obtain Θ−1 GA Θ = D B GB , where the matrix D B is given
satisfies Θ|R A A
by
B
DAB = LbA Rb −1 . (6.7)
Rewriting GA Θ = DAB ΘGB , differentiating with respect to geλ and taking the limit g =
(id)G , we arrive at the commutation relations [31]:
h i h i h i
bA, L
L bB = f C L bC , R b A, Rb B = −f C Rb C, R b A, L
b B = 0. (6.8)
AB AB
The Conformal-Affine Structure of open Quantum Relativity... 261
With the aid of the BCH formula, we determine the explicit form of the adjoint representa-
tion of the Lie algebra basis elements ad ge−1 B
GA = D A GB ,
h M i B 1 C
DAB = eλ ρ(GM ) B
= δA − λC fCAB + λ fCAM λD fDMB − · · ·, (6.9)
A 2!
1 1
e−A BeA = B − 1! [A, B] + 2! [A, [A, B]] − · · ·,
1 1
e−χA deχA = dχA − 2! [χA, dχA] + 3! [χA, [χA, dχA]] − · · ·,
µν +δhµν ) † S µν † S
h α γβ i (6.10)
ei(h µν = eih µν 1 + ie−h γ δeh †S
αβ + Lαβ ,
h α β
i
ei(φ+δφ)D = eiφD 1 + ie−h β δeh α D ,
and [70]
ωαβ Λαβ e−iξ = ωαβ Λαβ + ωαβ ξ α Pβ ,
αP αP
eiξ α α
µ β
καβ Λαβ e−i∆
µν Λ µµ Λ
ei∆ µν µν = e∆α κµν e−∆ν Λαβ ,
(6.11)
µν µν α β
eih Sµν τ αβ Lαβ e−ih Sµν = eh µ τ µν e−hν Λαβ ,
µν S µν S α β
eih µν σ αβ † Sαβ e−ih µν = eh µ σ µν e−hν † Λαβ ,
with ωαβ † Λβα = ααβ † Sαβ + βαβ Lαβ . The components of the stress forms
α ∧ ⋆β = β ∧ ⋆α, ρ ∧ ⋆σ = σ ∧ ⋆ρ,
δ(ρ∧⋆σ)
δϑ = −δϑc ∧ (hσ|hc i ∧ ⋆ρ − (−)r ρ ∧ h⋆σ|hc i) .
In the set of equations displayed in (4.130), v is a vector, α and β are p-forms that are
independent of the coframe V , while ρ and σ are r-forms that are independent of the special
conformal coframe-like quantity ϑ.
262 G. Basini and S. Capozziello
References
[1] G. Basini and S. Capozziello, ”Open Quantum Relativity”, in Classical and Quantum
Gravity Research, Ed. M.N. Christiansen, Nova Science Publishers, Inc., New York
(2008).
[4] G. Basini and S. Capozziello, Int. Jou. Mod. Phys. D 15 (2006) 583.
[9] D.W. Sciama, On the analog between charge and spin in General Relativity, in Re-
cent Developments in General Relativity, Festschrift for Leopold Infeld, (1962) 415,
Pergamon Press, New York.
[19] C. G. Callan, S. Coleman, J. Wess and B. Zumino, Phys. Rev. 117 (1969) 2247.
[20] S. Coleman, J. Wess and B. Zumino, Phys. Rev. 117, (1969) 2239.
[22] A. Salam and J. Strathdee, Phys. Rev. 184, 1750 (1969); Phys. Rev. 184 (1969) 1760.
[23] A.B. Borisov and V.I. Ogievetskii, Theor. Mat. Fiz. 21, 329 (1974)
[33] Y. Ne’eman and D. Sijacki, Gravity from Symmetry Breakdown of a Gauge Affine The-
ory, The Center for Particle Theory, University of Texas at Austin, D6-87/40 (1987).
[34] A. Lopez-Pinto, A. Tiemblo and R. Tresguerres, Class. Quant. Grav. 12 (1995) 1503.
[36] R. Tresguerres and E. W. Mielke, Gravitational Goldstone Fields from Affine Gauge
Theory, arXiv: gr-qc/0007072
[41] M. Nakahara, Geometry, Topology and Physics Second Edition, Graduate Student
Series in Physics, (Institute of Physics, Bristol and Philadelphia 2003)
[42] A. Tiemblo and R. Tresguerres, Time evolution in dynamical spacetimes, arXiv: gr-
qc/9607066
[43] G. Basini, S. Capozziello, and G. Longo, Phys. Lett. A 311 (2003) 465.
[47] O. Chandia and J. Zanelli, Torsional Topological Invariants (and their relevance for
real life), arXiv: hep-th/9708138
264 G. Basini and S. Capozziello
[51] T. Dereli et al., Non-Riemannian Gravity and the Einstein Proca System, arXiv: gr-
qc/9604039
[52] R. Scipioni, Isomorphism between Non-Riemannian Gravity and Einstein Proca Weyl
theories extended to a class of Scalar gravity theories, arXiv: gr-qc/9905022
[55] G. Sardanashvily, ”On the geometric foundation of classical gauge gravitation theory,”
arXiv: gr-qc/0201074 (2002).
[56] M. Blagojevic ”Three Lectures on Poincaré Gauge Theory”, SFIN A1 (2003) 147.
[58] B. S. De Witt, ”Dynamical theory of groups and . . . elds (Les Houches Lectures 1963),
Relativity, Groups and Topology”, Gordon and Breach Science Publishers, New York
(1965).
[65] G. Basini,S. Capozziello, G. Longo, Gen. Relativ. Grav. 35, 189, (2003).
[67] G. Basini, F. Bongiorno, S. Capozziello, Int. Journ. Mod. Phys. D 13, 717 (2004).
[72] D.H. Satinger, O.L. Weaver, Lie Groups and Algebras with Applications to Physics,
Geometry and Mechanics, Springer-Verlag, Berlin (1986).
[76] B. Julsgaard, A. Kozhein, and E.S. Polzik, Nature 413 (2001) 400.
Chapter 9
Abstract
We introduce the notion of twisted balanced metrics. These metrics are induced
from specific projective embeddings and can be understood as zeros of a certain mo-
ment map. We prove that on a polarized manifold, twisted constant scalar curvature
metrics are limits of twisted balanced metrics, extending a result of S.K. Donaldson
and T. Mabuchi.
(C1 ) The Lie algebra Lie(Aut(M, L)) is trivial and T is semi-positive, α is a pointwise
semi-positive (1, 1)-form on M .
(C2 ) T is ample and α is a positive (1, 1)-form on M .
Let us give now some explanations about our condition on the Lie algebra
d
Lie(Aut(M, L)). An element of Lie(Aut(M, L)) can be described as the real part of a
∗
C -invariant holomorphic vector field of L. Of course, there is a canonical map
d
τ : Lie(Aut(M, L)) → Lie(Aut(M ))
by pushing down via πL such a vector field seen as an element of Aut(L). Then
Lie(Aut(M, L)) is trivial if and only if τ has trivial image. This latter condition appeared in
the work of Donaldson who identified Lie(Aut(M, L)) with the kernel of the Lichnérowicz
operator.
Notation. In all the following, M et(Ξ) will denote the space of smooth hermitian metrics
on the bundle or vector space Ξ. Moreover J will be the complex structure on M and
Diff(M ) the space of diffeomorphisms of M in √
a fixed homotopy class. For a smooth
−1 ¯
hermitian metric h on a line bundle, c1 (h) = − 2π ∂ ∂ log(h) represents its curvature.
In a first part, using a technical result about Bergman kernels, we will describe the no-
tion of twisted balanced metrics from a symplectic point of view. Then, we study the con-
vergence of a sequence of twisted balanced metrics when there exists a solution to Equation
(1). Our main result is Theorem 2.
and x ∈ M , Z
ω n (y)
π(f )(x) = B(x, y)f (y) .
M n!
We can express the restriction of the Bergman kernel over the diagonal, that we shall call
the Bergman function. In particular, one can write
Nk
X
B(x) = B(x, x) = |si |2hk ⊗h−1 (x)
L T
i=1
Twisted Balanced Metrics 269
where the sections (si )i=1,..,Nk form an orthonormal basis of H 0 (M, Lk ⊗T −1 ) with respect
to the L2 inner product defined previously:
Z
ωn
h., .i = hkL ⊗ h−1
T (., .) .
M n!
Clearly, the Bergman kernel is independent of the choice of the orthonormal basis. Now,
one obtains the asymptotic behavior of B(x) when k tends to infinity.
Theorem 1.1. With our previous notations, one has for k large enough,
1
n k n−1
γ
n
B(x) − k + (Scal(ω) − Λω α)
r ≤ 2
k 2 C (ω) k
Proof. This is essentially a consequence of [Lu, Wa], and we refer to [M-M] as a general
survey on this topic. In particular a proof can be found with [M-M, Theorem 4.1.2] but for
the sake of clearness, we will sketch the computation of the terms of the asymptotic. The
key point is that the problem is purely local. It is clear that
|s(x)|2hk ⊗h−1
L T
B(x) = sup (2)
s∈H 0 (M,Lk ⊗T −1 ) ksk2
and one can reduce the problem to construct the section that represents this supremum at
x ∈ M . Let us call this section sextr(x) , the extremal section at x, which is unique up to
scaling. Now, one can choose a smooth section s0 ∈ C∞ (M, Lk ⊗ T −1 ) such that s0 is
concentrated in L2 norm on a small geodesic ball B of radius log(k)
√
k
around x. Without loss
2 ¯
of generality, one can fix |s0 (x)|hk ⊗h−1 = 1. Furthermore, using Hörmander’s ∂-estimates,
L T
one can modify s0 to make it holomorphic. Hörmander’s estimates can be applied because
of the positivity of Lk ⊗ T −1 for large k. This gives sextr(x) ∈ H 0 (M, Lk ⊗ T −1 ) with
|sextr(x) (x)|2hk ⊗h−1 = 1. Hence, from (2), we are lead to compute the L2 norm of sextr .
L T
In order to do that, we specify some appropriate coordinates. On one hand, using Böchner
coordinates, one can write locally hL = e−φL,x where φL,x is plurisubharmonic with
1
φL,x (z) = |z|2 − Rij̄kl̄ zi z̄j zk z̄l + O(|z|5 ).
4
Here Rij̄kl̄ denotes the Riemannian curvature tensor of the Riemannian metric gij̄ induced
by c1 (hL ) on M . On another hand, in the same coordinates and thanks to some affine
transformations, h−1
T =e
−ψT,x where the potential ψ
T,x satisfies
X
e−ψT,x = 1 − c1 (h−1 3
T )kl̄ zk z̄l + O(|z| ).
1≤k,l≤n
270 Julien Keller
√ n
−1
Let’s denote dV0 = 2π dz1 ∧ dz̄1 ∧ ... ∧ dzn ∧ dz̄n . Now, one finds explicitly when k
tends to infinity,
Z
ωn
ksextr(x) k2 ∼ |sextr(x) |2hk ⊗h−1
log(k)
B x, √ L T n!
k
Z
∼ e−kφL,x −ψT,x det(gij̄ )
log(k)
B x, √
k
Z
2 k
∼ e
−k|z|
1 + Rij̄kl̄ zi z̄j zk z̄l + O(|z|5 )
log(k)
B x, √ 4
k
X
3 −Ricij̄ zi z̄j +O(|z|3 )
× 1 − c1 (h−1
T )kl̄ zk z̄l + O(|z| ) e dV0
1≤k,l≤n
This can be evaluated using the fact that given f a function on {1, .., n}p × {1, .., n}p ,
XZ
2q −k|z|2
√ fI,J¯zi1 ..zip z̄j1 ..z̄jp |z| e dV0 =
I,J |z|<log(k)/ k
X X p!(n + p + q + 1)! 1
1 ¯
fI,σ(I) + O( p′ ),
p! (p + n − 1)!k n+p+q k
I σ∈Σp
for any p′ > 0. Hence, one gets after removing non symmetric terms (in holomorphic and
anti-holomorphic variables)
2 1 1 1 k
ksextr(x) k = + (−Scal(gij̄ )) + 2 n+2 Scal(gij̄ )
k n k n+1 k 4
α
1 1
− n+1 Λω − +O
k 2 k n+2
1 1 1
= − Scal(gij̄ ) − Λω α + O ,
k n 2k n+1 k n+2
We now consider the Bergman function as depending on the choice of the metric hL .
In that context and generalizing the notion of balanced metrics studied by S. Zhang and H.
Luo, it is natural to introduce the
Twisted Balanced Metrics 271
Definition 1.1. A metric hL is said to be hT -twisted balanced of order k if the k-th Bergman
function associated to it satisfies for all x ∈ M ,
Nk
BhL ,hT (x) =
V ol(L)
where V ol(L) = c1 (L)n ([M ]) is the volume of L.
An obvious consequence of Theorem 1.1 is the following result.
Proposition 1.1. Assume that there exists for all k sufficiently large a metric hk ∈ M et(Lk )
which is hT -twisted balanced, and assume that the sequence (hk )1/k ∈ M et(L) is conver-
gent in C∞ topology. Then its limit h∞ has curvature ω∞ solution to Equation (1), i.e ω∞
is an α-twisted cscK metric.
1
Remark 1.1. The reason of our normalization of the form α by a factor 2 is precisely due
to the asymptotic expansion of Theorem 1.1 and Equation (1).
Furthermore, we can see twisted balanced metrics as Fubini-Study metrics, i.e they can
be understood as algebraic type metrics. Let us denote the complex vector space
V = H 0 (M, Lk ⊗ T −1 ).
We define the Fubini-Study map
F S : M et(V ) → M et(Lk )
such that for H ∈ M et(V ), F S(H) is the hermitian metric satisfying for all x ∈ M ,
Nk
X Nk
|si |2F S(H)⊗h−1 (x) =
T V ol(L)
i=1
where (si )i=1,..,Nk is an H-orthonormal basis of V . On another hand, one can construct the
Hilbertian inner product on V by considering the map
HilbhT : M et(Lk ) → M et(V )
such that Z
c1 (h1/k )n
HilbhT (h) = h ⊗ h−1
T (., .) .
M n!
Then obviously, hT -twisted balanced maps are fixed points of the map
F S ◦ HilbhT : M et(Lk ) → M et(Lk ).
This can be rephrased by saying that there exist metrics H ∈ M et(V ) – that we shall call
again twisted balanced metrics – satisfying that F S(H) is twisted balanced in the sense of
Definition 1.1 or Z
hsi , sj iF S(H)⊗h−1 µF S(H) = δij
T
M
where µF S(H) is the induced Fubini-Study volume form and (si )i=1,..,Nk is H-orthonormal.
On other words, through the Kodaira embedding
ι : M ֒→ P(V ∗ )
induced by the sections of H 0 (M, Lk ⊗ T −1 ), the center of mass of M is trivial.
272 Julien Keller
and thus it is natural to consider the moment map associated to the group Gk of hermitian
bundle maps from Lk ⊗ T −1 to Lk ⊗ T −1 that preserve the Chern connection induced by
hL and hT .
Via the classical hamiltonian construction, its Lie algebra can be identified with C∞
0 (M, R)
the space of smooth functions on M with vanishing integral. Note that Gk acts as auto-
morphims of Lk ⊗ T −1 covering the action of elements of Symp(M, ω), the group of
hamiltonian symplectomorphisms preserving the Kähler form ω.
The moment map associated to this action and the symplectic form Ω is described in [Do1,
Section 2.1]. This is explicitly given by µ : C∞ (M, Lk ⊗ T −1 ) → Lie(Gk )∗ , where
−1 ω n−1 ωn
µ(s) = J∇Lk ⊗T (s) ∧ ∇L−k ⊗T −1 (s∗ ) ∧ + k|s|2hk ⊗h−1 .
2n (n − 1)! L T n!
Here ∆ is the Laplace operators acting on functions and one has fixed the constant sb to be
1
R 1 2 2
n
ω
sb = V ol(L) M 2 ∆|s|hk ⊗h−1 + k|s|hk ⊗h−1 n! .
L T L T
On another hand, when acting by Gk , one needs to move the complex structure in order
to preserve the holomorphicity property of a section. Thus, it is natural to consider the
induced action of Gk over the space Jint of all ω-compatible complex structure over M
(i.e the set of all almost-complex structures such that its Nijenhuis tensor is zero). One can
see Jint as the space of sections of a Sp(2n)/U (n)-bundle over M . With the complex
structure of Sp(2n, R)/U (n) and its natural metric, one obtains using the volume form
ω n , a Kähler structure over the infinite dimensional manifold Jint . Note that the group
Symp(M ) preserves this Kähler structure. It acts on the structure J by
ψ(J) = ψ∗ J −1 ψ∗−1 .
1
R 1
P Nk 2
ωn
with sbk = V ol(L) M 2∆ +k i=1 |si |hk ⊗h−1 n! . Moreover, we notice that
L T
This comes by taking the L2 inner product with eigenfunctions of the Laplacian in (3).
whose image lies in the space of trace free matrices. Hence, finding a zero of the moment
map µSU corresponds formally to choosing a basis of orthonormal sections with respect to
the inner product induced by hL , hT .
The moment map for the action of the product Gk × U (Nk ) is given by the sum µGk ⊕
µSU and of course we can consider the double symplectic quotient
µ−1 −1
Gk (0) ∪ µSU (0)
Υ//(Gk × SU (Nk ) = . (5)
Gk × SU (Nk )
ΛLie(G)
p = |||Q−1
p |||Lie(G) ,
Lie(G)
i.e Λp is the largest eigenvalue of Q−1
p . This quantity controls the convergence of the
gradient flow of the norm square of the momentum map,
∂ν(pt )
= −ν(pt ).
∂t
It also gives the distance of the initial point to the zero of the moment map.
Proposition 3.1. Let p0 ∈ N . Assume that there exist positive constants r0 , r1 such that,
r1 Lie(G)
|ν(p0 )| < , Λeiζ p ≤ r0 ∀|ζ| ≤ r1 ,
r0 0
ν(eiη p0 ) = 0,
In order to find a zero of the moment map µGk ⊕ µSU , we proceed in two steps. First,
we look for the first symplectic quotient
Υ//Gk .
This corresponds to finding a (non necessarily orthonormal) basis (si )i=1,..,Nk ∈ V and a
metric hk such that
XNk
|si |2h ⊗h−1 = Ck
k T
i=1
1 Cα
a1 = (Scal(ω∞ ) − Λω∞ α) = (8)
2 2
are both constants. Writing
a1,1 (̟1 ) = χ2 − a2
e
276 Julien Keller
One key point here is that the terms e ar+1−l,l depend only on ̟1 , ..., ̟r−1 . Moreover,
a1,r is obtained as the differential Lω of the map
because of (8), each term e
1
ω 7→ (Scal(ω) − Λω α) .
2
Consequently, starting with ̟1 , one can find the ̟r using the implicit function theorem
recursively if the RHS of (9) does not lie in the kernel of the operator Lω for any r ≥ 1. If
one considers a small deformation ω + i∂ ∂φ ¯ = ωφ , then
Z
¯ 1,0 φk2 2
φ · Lω (φ)ωφn = k∂∇ ¯
L (ωφ ) + h∂φ ∧ ∂φ, αiL2 (ωφ ) ,
M
which shows that the kernel of Lω is trivial if either Lie(Aut(M, L)) is trivial (Cf. [Bi,
Lemme 1.1]) and α is semi-positive, or if α is positive. We have proved the following
Theorem 1. Assume that condition (C1 ) or (C2 ) holds. Assume the existence of a twisted
cscK metric √
−1 ¯
ω∞ = − ∂ ∂ log(h∞ ) ∈ c1 (L)
2π
solution to Equation (1). Then for any q > 0 and k sufficiently large, there exist smooth
functions ̟1 , ..., ̟q and a constant cq,k such that the metric
q
!
X ̟ i
e
hk (., .) = 1 + h∞ (., .)
ki
i=1
Corollary 3.1. Assume that condition (C1 ) or (C2 ) holds, and the existence of a twisted
cscK metric. Then, there exists an approximate twisted balanced metric, i.e a zero of the
moment map µGk .
In order to get some uniform estimates, we shall fix for each k sufficiently large, the metric
ω
e∞ = kω∞ .
We shall now define a class of metrics by saying that a metric ωe ∈ c1 (L) has R-bounded
geometry if
1
ω
e> ω e∞ , ke
ω−ω e∞ kC4 (eω∞ ) < R.
R
Moreover, we will say that a basis (si )i=1,..,Nk of V has R-bounded geometry if the curva-
ture of the induced Fubini-Study metric has R-bounded geometry. Firstly, with no substan-
tial modification of the proof of [P-S2, Theorem 2], one obtains the
Proposition 3.2. Assume condition (C1 ) holds. Then for any R > 1, there exist two con-
stants C > 0 and ǫ < 1/10 such that if S = (si )i=1,..,Nk is a basis of V with R-bounded
geometry and |||µSU (S)||| < ǫ, one has
Lie(SU (V ))
ΛS < Ck 2 .
Let us now assume that condition (C1 ) is not satisfied. Let H be the maximal connected
algebraic subgroup of Aut0 (M ), the connected identity component of the group of holo-
morphic automorphisms of M . Let Z be the maximal (algebraic) torus in the center of H
and denote its Lie algebra by Lie(Z). We set Kk = SU (V ) and Kk′ to be the identity
component of the subgroup of stabilizers of Kk at the point ι(M ). We shall consider the
following condition introduced in [Ma2].
At that stage, we remark that if Lie(Aut(X, L)) is not trivial, then the approximate twisted
balanced metrics obtained in Corollary 3.1 are all Z-invariant, since Z is also the identity
component of the group of isometries of (M, ω∞ ). Using the map HilbhT , the approximate
twisted balanced metrics induce a Kk′ -invariant inner product on Lie(Kk ). Hence, the
vector space Lie(Kk ) can be decomposed as
Proposition 3.3. Assume conditions (C2 ) and (C3 ) hold for an integer k large enough.
Then for any R > 1, there exist two constants C > 0 and ǫ < 1/10 such that if S =
(si )i=1,..,Nk is a basis of V with R-bounded geometry and |||µSU (S)||| < ǫ, one has
Lie(Kk′ )⊥
ΛS < Ck 2 .
278 Julien Keller
Then, for any ζ ∈ Lie(Kk ), the infinitesimal action σζ induces a vector field Xζ on P(V ∗ )
such that its restriction to M will be denoted Xζ,|M . Now, from [Bi, Lemme 2.3] the
inequality to prove is just equivalent to
Z
|ζ|2 ≤ Ck 2 |Xζ,|T M ⊥ |2hF S µ
eF S
M
For ∂¯ seen as acting on smooth (0, 1)-form on M with values in T M , one considers the
∗ 1/k
operator 1/k = ∂¯ hF S ∂.
hF S
¯ As it is pointed out in [Ma2], the first eigenvalue λ1 of 1/k
hF S
is bounded from below independently of k since S has R-bounded geometry. Moreover,
since ζ ∈ Lie(Kk′ )⊥ , Xζ,T M is orthogonal to the projection on T M of any holomorphic
vector field on ι(M ) by condition (C3 ). Thus, one has
Z 1/k Z 1/k
¯ ζ,T M |2 1/k c1 (hF S )n c1 (hF S )n
|∂X ≥ λ1 |Xζ,T M |2 1/k .
M hF S n! M hF S n!
¯ ζ,T M + X
But now, ∂(X ζ,T M ⊥ ) = 0 and thus
¯ 2 λ1
k∂X ζ,T M ⊥ kL2 (e
µF S ) ≥ kXζ,T M k2L2 (eµF S ) . (15)
k
Finally both (15) and (14) imply Inequality (13).
Theorem 2. Assume that either condition (C1 ) holds or both conditions (C2 ) and (C3 ) hold
for a sequence of strictly increasing integers kj > k0 . Assume that there exists a twisted
cscK metric solution to Equation (1) in the class c1 (L). Then,
• For k0 large enough, there exists a hT -twisted balanced metric ωkj ∈ kj c1 (L),
• The sequence k1j ωkj is convergent when j → +∞ towards the twisted cscK metric in
C∞ -topology.
Proof. Under condition (C1 ), the proof of the Theorem is a consequence of Corollary 3.1,
Proposition 3.2 and Proposition 3.1 together with the double symplectic quotient picture.
The convergence is Cr topology (for any r) is obtained by the fact that one can choose,
up to any order, an approximate twisted balanced metric in Theorem 1 (see [Do1, Proof
of Theorem 3]). Finally, the uniqueness of the twisted cscK metric is a consequence of
[St1], and one could recover this result in the projective setting by studying the uniqueness
of twisted balanced metric up to SU (V ) action. The convergence of the twisted balanced
metric is clear by construction.
Let us now assume conditions (C2 ) and (C3 ). The main difference with previous case is
that one has to check that the gradient flow of |µSU |2 is still converging with the estimate
obtained from Proposition 3.3. From [Ma2, Lemma 3.4] and [Ma3, Theorem 3.2], it is
sufficient to obtain a zero of the moment map by considering the one parameter subgroups
perpendicular to the subgroup of stabilizers (one could also invoke [Si, Proposition 9]).
Now, Proposition 3.3 shows the convergence of the gradient flow when one restricts the
moment map to Lie(Kk′ )⊥ and by condition (C3 ), we can conclude.
4. Further Directions
Let us discuss some examples of twisted cscK metrics in the literature (see also [Fi1, Fi2]).
Ric(ω) = ωW P .
If condition (C3 ) holds, one can modify our arguments to construct a convergent
sequence of twisted balanced metrics for any 0 ≤ t < 1. We don’t know if in general
such a manifold is balanced in the sense of Zhang-Luo.
τ : M → Mcan over its canonical model Mcan . We assume that 0 < dim Mcan <
dim M , Mcan is non singular and the fibre τ −1 (p) is non singular for any p ∈ Mcan .
Thus, each fibre τ −1 (p) is a smooth Calabi-Yau manifold. The L2 metric on the
moduli space of Calabi-Yau manifolds induces a semi-positive Weil-Petersson (1, 1)-
form ωW P on Mcan . Then, the main result of [S-T2] proves the convergence of the
Kähler-Ricci flow in that context and identifies its limit. In that case, it satisfies the
following twisted cscK equation on Mcan ,
Ric(ω) = −ω + ωW P . (16)
Finally, one could consider a slightly more general framework. Assume that T = (Tj )
is a finite family of twistings such that 12 αTj are the curvature of the hermitian metrics
hTj ∈ M et(Tj ). Then, one can consider the T -twisted balanced metrics ω solution to
X
Scal(ω) − Λω αTj = C
j
(C′1 ) The Lie algebra Lie(Aut(M, L)) is trivial and the Tj are semi-positive i.e for all j,
αTj is a pointwise semi-positive (1, 1)-form on M .
(C′2 ) There exists j0 such that Tj0 is a ample and αTj0 is a positive (1, 1)-form on M . For
all j 6= j0 , Tj is semi-positive, αTj is pointwisely semi-positive.
Acknowledgements
The author is very grateful to Prof. T. Mabuchi, J. Stoppa and G. Székelyhidi for stimulating
discussions.
References
[Ba] S. Bando, The K-energy map, almost Einstein Kähler metrics and an inequality of
the Miyaoka Yau type, Tohoku Math. J. 39 (1987)
[Do2] S.K. Donaldson, Scalar curvature and projective embeddings II, Quaterly Jour.
Math 56 (2005)
Twisted Balanced Metrics 281
Chapter 10
Abstract
In contrast to the Euler-Poincaré reduction of geodesic flows of left- or right-
invariant metrics on Lie groups to the corresponding Lie algebra (or its dual), one can
consider the reduction of the geodesic flows to the group itself. The reduced vector
field has a remarkable hydrodynamic interpretation: it is a velocity field for a station-
ary flow of an ideal fluid with a constant pressure. Right- or left-invariant symmetry
fields of the reduced field define vortex manifolds for such flows. We give explicit gen-
eral expressions for the reduced vector field and its symmetry fields, provide examples
of such reduction and discuss two applications of this approach.
As the first application, we consider a mechanical system, whose configuration
space is a Lie group and whose Lagrangian is invariant to left translations on that
group, and assume that the mass geometry of the system may change under the action
of internal control forces. Such system can also be reduced to the Lie group. With
no controls, this mechanical system describes a geodesic flow of the left-invariant
metric, given by the Lagrangian, and thus its reduced flow is a stationary ideal fluid
flow on the Lie group. The standard control problem for such system is to find the
conditions, under which the system can be brought from any initial position in the
configuration space to another preassigned position by changing its mass geometry.
The hydrodynamic interpretation of the system both provides a convenient ”language”
and sharpens the controllability results: we show that by changing the mass geometry,
one can bring one vortex manifold to any other vortex manifold. As an example we
consider the n-dimensional Euler top.
The other application is the reduction for the Euler equations of an ideal fluid,
that describe the geodesics of a right-invariant metric on a Lie group SDiff(M ) of
the volume-preserving diffeomorphisms of a Riemannian manifold M , to the group
∗
E-mail address: mikhail@mci.sdu.dk
284 Mikhail V. Deryabin
SDiff(M ). For a typical coadjoint orbit we find all symmetry fields of a reduced
flow, and, as a corollary, we get a simple proof for nonexistence of new invariants of
coadjoint orbits, which are the integrals of local densities over the flow domain.
Keywords: Ideal hydrodynamics, Lie groups, control theory, invariants of coadjoint orbits.
1. Introduction
For the Euler top, the Hamiltonian vector field on the cotangent bundle T ∗ SO(3) can be
uniquely projected onto the Lie algebra so(3) – this is a classical reduction, known in the
general case as the Euler-Poincaré reduction. In the 30’s, E.T.Whittaker suggested an ”alter-
native” reduction procedure for the Euler top: by fixing values of the Noether integrals, the
Hamiltonian vector field can be uniquely projected from T ∗ SO(3) onto the group SO(3)
[31]. The Whittaker reduction is valid for any Hamiltonian system on a cotangent bundle
T ∗ G to a Lie group G, provided the Hamiltonian is invariant under the left (or right) shifts
on the group G. An important example of such Hamiltonian systems is a geodesic flow of
a left-(right-)invariant metric on a Lie group.
If we reduce a Hamiltonian system to the Lie group G, and then factorize the reduced
vector field by the orbits of its symmetry fields, then, by the Marsden-Weinstein theorem,
we get the same Hamiltonian system on a coadjoint orbit on the dual algebra g∗ , as if we
first reduced the system to the dual algebra g∗ , and then to the coadjoint orbit (see also
[1], Appendix 5). Thus, the Whittaker reduction can be regarded as a part of the Marsden-
Weinstein reduction of Hamiltonian systems with symmetries [26].
In contrast to the Marsden-Weinstein reduction, it has not been payed much attention
to the Whittaker reduction alone. However, it is itself worth studying. It turns out that a
vector field, reduced to a Lie group G has a remarkable hydrodynamic interpretation: it is a
velocity field for a stationary flow of an ideal fluid, that flows on the group G (viewed as a
Riemannian manifold), and is incompressible with respect to some left-(or right-)invariant
measure on G, see [9, 18, 19, 20] for details. The reduction to a Lie group is also useful
for a series of applications, which include stability theory, noncommutative integration of
Hamiltonian systems, discretization, differential geometry of diffeomorphism groups and
control theory, see, e.g., [20, 21, 8, 12]. In this chapter we consider two applications of this
approach, mostly following [9, 10].
We first review the Whittaker reduction and its hydrodynamic essence, and provide an
explicit expression for the reduction of a geodesic flow of a left- or right-invariant metric
onto a Lie group. For any Lie group we find both the reduced vector field and its ”symme-
try fields”, i.e., left- or right-invariant fields on the group that commute with our reduced
vector field. These fields have also a hydrodynamical meaning: these are the vortex vector
fields for our stationary flow (i.e., they annihilate the vorticity 2-form), cf. [20], [9]. The
distribution of the vortex vector fields in always integrable, thus they define a manifold, that
we call the vortex manifold. Typically, these manifolds are tori.
Next, we consider the following control problem. We study mechanical systems, whose
configuration space is a Lie group and whose Lagrangian is invariant to left translations
on that group, and we assume that the mass geometry of the system may change under the
action of internal control forces. Such systems can also be reduced to the Lie group, and
Reduction, Hydrodynamics and Control for Geodesics... 285
they also have an interesting hydrodynamic interpretation: the reduced vector field is the
velocity of a stationary flow of an electron gas (with no controls, this mechanical system
describes a geodesic flow of the left-invariant metric, given by the Lagrangian, and thus its
reduced flow is a stationary ideal fluid flow). Notice that without relating to hydrodynamics,
controlled systems on Lie groups were studied in many works, see, e.g., [7] and references
therein.
A control problem for such systems would be to find the conditions, under which the
system can be brought from any initial position on the Lie group to another preassigned po-
sition by changing its mass geometry. We show how the hydrodynamic interpretation of the
system can sharpen the controllability results. An immediate consequence is that under the
standard controllability conditions, one can bring the whole vortex manifold to any other
preassigned vortex manifold by changing the mass geometry. A simple necessary control-
lability condition can be obtained by studying the Bernoulli surfaces: roughly speaking, if
an (analytic) system is controllable, then it is always possible to change the controls such
that on every finite time interval, the phase curves of the controlled system intersect the
Bernoulli surface only in a finite number of points (but not necessarily transversally). An-
other application of this method would be to studying mechanical systems with damping,
which reduction to the Lie group leads to ideal fluid flows with external friction.
As an example, we consider the n-dimensional Euler top. We write down the reduced
controlled system explicitly, find the vortex manifolds, which typically (when the momen-
tum matrix has the maximal rank) are tori, and show that, by changing the mass geometry,
every such vortex manifold can be transformed to any other vortex manifold. We write down
the necessary controllability conditions, based on studying the Bernoulli surface, which, in
the 3-dimensional case, also turn out to be sufficient. We also write down controllability
conditions for the Euler top with damping.
As the second application, we define the ”secondary hydrodynamics”, i.e., we study
the reduction for the Euler equations of ideal fluid, that describe the geodesics of a right-
invariant metric on a Lie group SDiff(M ) of the volume-preserving diffeomorphisms of
a Riemannian manifold M , to the group SDiff(M ). To get a reduced vector field, we
fix a coadjoint orbit. In [27], certain special coadjoint orbits were identified with point-
vortex dynamics. This identification can also be thought of as a reduction to the group
for these special coadjoint orbits (the particle velocity is explicitly defined for the point
vortices system). Notice that in this case, the ”reduced” system (i.e., the system of point
vortices in 2D) is a finite-dimensional system, while the Whittaker reduction on the Lie
group SDiff(M ) in the general case is an infinite-dimensional system.
We will pay most attention to the symmetry fields of the reduced flow. For a ”typi-
cal” coadjoint orbit we find all symmetry fields of a reduced flow, which are the variational
derivatives of the invariants of coadjoint orbits, found in [17]. The term ”typical” means that
a certain vector field does not admit any symmetries. Coadjoint orbits with no extra sym-
metries were also considered in [14] in relation with topology of steady even-dimensional
flows. Now, as an invariant of a coadjoint orbit generates a symmetry field for the reduced
flow, we conclude readily that there are no new invariants of coadjoint orbits for ideal fluid
flows, linearly independent of the classical ones on these ”typical” orbits (these invariants
should be the functionals on the dual algebra, such that the variational derivative of such
functional is a smooth vector field from the Lie algebra SVect(M ) of divergence-free vec-
286 Mikhail V. Deryabin
tor fields on M ). If we can take for granted the fact that the ”typical” case are everywhere
dense in the space of all 1-forms on M in some appropriate topology, then we readily con-
clude that there are no invariants of coadjoint orbits, which are integrals of local densities
over the flow domain, which are linearly independent of the classical ones. Thus, there
are no new invariants of coadjoint orbits for ideal fluid flows in the above class of func-
tionals. The results on the absence of new integral invariants complement the theorem on
local invariants, i.e., on the local description of isovorticed fields existence at a generic
point in ideal hydrodynamics [3]. This theorem states that at a generic point, in the even-
dimensional case, there is only one local invariant, which is the vorticity function, while in
the odd-dimensional case, there are no local invariants at all.
In the Appendix we study the Whittaker reduction for nonholonomic systems and for-
mulate and discuss the standard controllability conditions together with their extensions.
mc = Ad∗g ms , (2.1)
Ad∗g : g∗ → g∗ being the group coadjoint operator. Let us fix the ”momentum in space”
ms . Then relation (2.1) defines a coadjoint orbit. The Casimir functions are the functions
of the ”momentum in the body” mc , that are invariants of coadjoint orbits. For example, for
the Euler top, the Casimir function is the length of the kinetic momentum.
Let A : g → g∗ be a positive definite symmetric operator (inertia operator) defining
a scalar product on the Lie algebra. This operator defines a left- or right-invariant inertia
operator Ag (and thus a left- or right-invariant metric) on the group G. For example, in the
left-invariant case, Ag = L∗g −1 A Lg−1 ∗ .
Let the metric be left-invariant. The geodesics of this metric are described by the Euler
equations
Lg−1 ∗ ġ = ωc . (2.3)
Reduction, Hydrodynamics and Control for Geodesics... 287
The Euler equations follow from the fact that ”the momentum in space” ms is constant,
whereas ”the momentum in the body” mc is obtained from ms by (2.1), see [3].
Remark. Strictly speaking, in the infinite dimensional case the operator A is invertible only
on a regular part of the dual algebra g∗ . In our case this means, that some natural restriction
on values of ms (or mc ) have to be imposed (see [3]).
The Euler equations can be considered as Hamilton’s equations on the dual algebra,
where the Hamiltonian equals H = 12 (A−1 m, m), m ∈ g∗ , and the Poisson structure is
defined by the following Poisson brackets. For two functions F (m) and G(m) on the dual
algebra g∗ ,
where dF (m), dG(m) ∈ g are the differentials of functions F and G, and [ξ, η] = adξ η is
the commutator (adjoint action) on the Lie algebra g.
Let now
1
H = (A−1 mc , mc ) + (λ, mc ),
2
where λ ∈ g is a constant vector. Then Equation (2.2) becomes
Proposition 2.1. (The Whittaker reduction) For the case of the left-invariant or the right-
invariant metric, the vector field v(g) has the form
and, respectively
v(g) = Rg∗ (A−1 Ad∗g−1 mc + λ). (2.6)
Here ms , respectively mc , is constant.
Notice that in Proposition 2.1, to find the reduced vector field we do not need the Hamil-
tonian equations on T ∗ G and the explicit expression for the Noether integrals. We only need
the Lie group structure and the inertial operator. This is important for generalizations to the
infinite-dimensional case. Unlike for the Marsden-Weinstein reduction, we do not have to
assume nondegeneracy conditions on the momenta ms or mc .
Proof. We consider only the case of the left-invariant metric; for the right-invariant case
the proof is similar. Relation (2.1) determines the function mc = mc (ms , g) on the group
288 Mikhail V. Deryabin
2
Let w(g) ∈ T G be a right-invariant vector field on the group G, which is defined by a
vector ξ ∈ g: w(g) = Rg∗ ξ.
We fix a momentum ms .
Theorem 2.3. For the momentum ms fixed, the vector field w(g) on G is a symmetry field
of the reduced system v(g) if and only if the vector ξ satisfies the condition
ad∗ξ ms = 0. (2.7)
In the finite-dimensional case this means that the flows of the vector fields v(g), w(g)
on the group commute. In the infinite-dimensional case one should be more accurate: the
equation ġ = v(g) is a partial integral-differential equation, rather than an ordinary differ-
ential equation, hence, strictly speaking, it is not clear if it has a solution. On the other
hand, equation ġ = Rg∗ ξ always has a solution, which is a one-parametric family of the left
shifts on the group G: g → (exp τ ξ)g, see, for example, [30], as we have assumed that the
exponential map exists. Notice also that, in view of Proposition 2.2, under the assumption
of Theorem 2.3, the Lie derivative Lw(g) m(g) = 0.
Proof of Theorem 2.3. The vector fields w(g) and Lg∗ λ commute, as right-invariant fields
always commute with left-invariant fields. Thus, it is sufficient to show that
The last term equals L(exp τ ξ)∗ v(g) for any g ∈ G if and only if
Ad∗exp τ ξ ms = ms
for all values of the parameter τ . Differentiating the last relation by τ we arrive at the
statement of the theorem. 2
Let again λ = 0, and consider a subspace F of functions on the dual algebra g∗ , such
that the variational derivatives of these functions belong to the Lie algebra g.
Proof. Let F be constant on the coadjoint orbit Ad∗g ms : F (Ad∗g ms ) = const for any
g ∈ G. Then,
δF (ms )
, ada ms = 0 = (a, ad∗δF (ms )/δm ms )
∗
δm
where ν is a positive constant, or, in more general case, a non-negative function of time.
Such system can also be reduced to the Lie group: in the previous formulas, the constant
ms should be replaced by
Z t
ms (t) = exp − ν(s)ds ms (0).
0
Indeed, from (2.1) one can see that the covector ms (t) should satisfy the equation
ṁs = −ν(t)ms .
All the subsequent results remain true, we only have to be careful with Theorem 2.3, as the
vector fields (v(g, t), 1) and (w(g), 1) do not commute in the extended phase space G × R
(but the fields (v(g, t), 1) and (w(g), 0) obviously do commute).
290 Mikhail V. Deryabin
We now define a vortex vector field, as an annihilator of the vorticity 2-form. Then the
condition iη(g) dm(g) = 0 is exactly the definition of a vortex field. Thus, we have proved
Proposition 3.2. Any symmetry field to the reduced vector field is a vortex vector field.
Consider now ideal hydrodynamics on a Riemannian manifold M . The Euler equations
of an ideal fluid are the geodesics of a right-invariant metric on a Lie group SDiff(M )
of volume-preserving diffeomorphisms of the manifold M . Thus, we can apply the above
procedure, and reduce the geodesic flow to the group SDiff(M ). The reduced ”vector field”
can be thought of as a stationary flow of an ”ideal fluid”, which flows on the Lie group
SDiff(M ). This justifies the term ”secondary hydrodynamics” in the infinite-dimensional
case (cf. [20]).
Remark. The stationary flows on Lie groups that we study are typically degenerate: in
the even-dimensional case, the rang of the vorticity 2-form dm(g) is not maximal. Thus,
the standard topological results on even-dimensional stationary flows (see, e.g., [3]) cannot
be applied.
Vortex vectors, i.e., vectors ξ ∈ g that satisfy condition (2.7), are the isotropy vectors.
We now review some classical results on the isotropy vectors and the Casimir functions,
see, e.g., [2] for details, and adapt them to our case.
Proposition 3.3. The distribution of the isotropy vectors is integrable.
The Proposition says that if vectors ξ1 , ξ2 ∈ g satisfy condition (2.7), then the vector
[ξ1 , ξ2 ] = adξ1 ξ2 also satisfies this condition, which is a simple consequence of the Jacobi
identity. The integrable distribution of the isotropy vectors defines a manifold (at least
locally), that we, following [20], call a vortex manifold. Notice that for compact Lie groups,
vortex manifolds are closed surfaces.
The isotropy vectors ξ ∈ g form a Lie subalgebra h ⊂ g, called an isotropy algebra for
the coadjoint orbit m = Ad∗g ms . If the differentials of the Casimir functions form a basis of
the isotropy algebra h, then h is Abelian. In general, an isotropy algebra is not necessarily
Abelian. A very simple example is G = SO(3): if the ”momentum in space” ms = 0,
then h = so(3). However, in the finite-dimensional case isotropy algebras are Abelian on
an open and dense set in g∗ (the Duflo theorem). Thus, the corresponding vortex manifolds
(that pass through the group unity) are commutative subgroups of the Lie group G. Notice
that in the infinite-dimensional case, vortex fields cam still define a certain commutative
subgroup, which can also be referred to as ”vortex manifold”.
Vortex manifolds have always the dimension of the same parity as the Lie group dimen-
sion. This is a simple corollary of the fact that coadjoint orbits are always even-dimensional
(also the degenerate ones), see, e.g., [2].
One can show that if the Hamiltonian has also terms, linear in the momenta (in the other
words, if λ 6= 0), then the reduced field has the following hydrodynamic sense: it is the
velocity of the stationary flow for the electron gas, which satisfies an ”infinite conductivity
equation”, again, with a constant pressure, see [3].
For the reduced system, one can define the Bernoulli functions: these are the functions,
which are invariants both of the flow velocity field and the vorticity fields. For example, a
function 1/2(v, v), where v is the reduced vector field, and (·, ·) is the Killing metric, is a
Bernoulli function.
292 Mikhail V. Deryabin
We consider now systems with damping, or dissipation, as described at the end of the
previous section. These systems describe flows with ”external friction” (which are no longer
stationary), given by
∂v
+ ∇v v = −νv, (3.1)
∂t
where ν is a positive constant. Such systems experience a ”complete vortex diffusion” (see
[20]) in the following sense. We rewrite Equation (3.1) as
∂[u]
+ Lv [u] = −ν[u],
∂t
where u is the 1-form, obtained from v by lowering the indices, and [u] is the coset of
1-forms. Let
Z
I(t) = u
gvt γ
for some closed curve γ (obviously the value of I does not depend on the choice of a 1-form
u from the coset [u]). Then the function I(t) satisfies the equation
I˙ = −νI,
and thus an integral of the 1-form u over any closed cycle tends to zero exponentially fast.
where ω ∈ g is the system velocity, λi ∈ g are constant vectors, ui (t) ∈ R are controls,
and A : g → g∗ is the inertia operator. Notice that the dimension k of the control vector
u(t) may be lower than the dimension of the Lie algebra. We assume that there is a positive
constant ǫ, such that ku(t)k ≤ ǫ, i.e., our controls are always bounded. Physically, these
controls mean that we can change the system mass geometry by internal forces.
The Euler equations (2.4) are:
ṁ = ad∗ω m,
P
where the momentum m = A(ω + ki=1 ui λi ) ∈ g∗ . The system, reduced to the group G,
is (cf. (2.5): !
Xk
−1 ∗
ġ = Lg∗ A Adg ms − ui λi = vλ (g). (4.1)
i=1
Reduction, Hydrodynamics and Control for Geodesics... 293
From Theorem 2.3 and Proposition 3.3 follows the following result.
Suppose that System (4.1) is controllable (we formulate some standard conditions in
the Appendix B.), and we assume that the controls u(t) are piecewise constant functions.
We fix the ”momentum in space” ms : with ms fixed, so are the vortex manifolds.
Theorem 4.1. By applying controls u(t), one can transform any vortex manifold H1 to any
other prescribed vortex manifold H2 , such that the following diagram is commutative:
gvt
H1 →λ H2
s ↓
gw s
↓ gw (4.2)
gvt
H1 →λ H2 ,
s being its phase
where by w we denote vortex vector fields for the given momentum ms , gw
flow, and gvt λ is the phase flow of System (4.1).
This theorem is a reflection of a well-known fact that vortex lines are frozen into the
flow of an ideal fluid.
Proof. By Theorem 2.3, the vector fields vλ (g) and the vortex fields w(g) commute (the
vortex fields are right-invariant, while the vectors Lg∗ λi are left-invariant, and we have also
assumed that u(t) is piecewise constant). Pick up the controls (i.e., functions u(t)), that
send a point h1 ∈ H1 to a point h2 ∈ H2 . Then the same controls send a point gw s h to
1
s
gw h2 , due to commutativity, which proves the theorem. 2
A simple corollary is that all vortex manifolds, that correspond to the same value of the
momentum ms , are homotopic to each other. Another observation is that an electron gas,
flowing on a Lie group, can be controlled by changing an external electro-magnetic field.
The condition that u(t) are piece-wise constants can be dropped: the vortex manifolds
will still be transformed into vortex manifolds, but, strictly speaking, the diagram (4.2) will
no longer be valid, as there is no commutativity in the extended phase space G × R, cf.
Remark in Section 2.. In the sequel, we will consider the general case, when u(t) is not
necessarily piece-wise constant.
Let us now define the Bernoulli surface as the intersection of the levels of all inde-
pendent Bernoulli functions (see the previous section). There is always one ”classical”
Bernoulli function
α(g) = 1/2(v(g), v(g)) = 1/2 A−1 Ad∗g ms , Ad∗g ms . (4.3)
(at least for semi-simple Lie groups, where the Killing metric is nondegenerate).
Proposition 4.2. If System (4.1) is controllable, then there is a control vector λ ∈ g, such
that no trajectory g(t) of System (4.1) belongs to the surface α = const. In particular, for
any stationary solution Ad∗g(t) ms = const = mc , the coadjoint action ad∗λ mc 6= 0.
This condition simple (and the proof is straight-forward), but it is easy to check in appli-
cations, and in the 3-dimensional case, it opens the possibility of controlling the system by
only one control vector (provided the momentum is not zero). Indeed, for the 3-dimensional
fluid flows on compact Lie groups, such that the vorticity field is not parallel to the fluid ve-
locity field, the Bernoulli surfaces, defined by the Bernoulli function (4.3) are 2-dimensional
294 Mikhail V. Deryabin
tori, and one control is enough to jump both from a torus to another torus, and from one
trajectory on a torus to another one on the same torus.
Example. For controllability of a reduced Euler top on SO(3) it is enough to have
only one control, provided all the principal axes of the inertia ellipsoid are different, and
the momentum is the space ms 6= 0 is not directed along any of the principal axes [25].
We consider now control problem for systems with dissipation. Suppose that the dissi-
pation coefficient ν is a constant (see the previous section). A control system with external
friction takes the form
k
!
X
ġ = Lg∗ A−1 Ad∗g e−νt ms − u i λi . (4.4)
i=1
This is the original system, which, by the condition of the Proposition, is controllable on
time interval τ ≤ T . But sup τ = 1/ν, thus we get 1/ν < T . 2
Remark. This condition can be strengthened by noticing that, as t → ∞, the multiplier
exp(νt) → ∞, thus the controls u become unbounded.
We assume, that the inertia operator of A : so(n) → so∗ (n) is defined by the relation
Ωc = A−1 M = U M + M U , where U is any constant nondegenerate operator. Thus the
system (5.1) is a closed system of n(n − 1)/2 equations, which was first written in an
explicit form by F. Frahm (1874) [13]. As was shown in [24] (for n = 4, in [13]), with the
above choice of the inertia tensor, the system ( 5.1) is a completely integrable Hamiltonian
system on the coadjoint orbits of the group SO(n) in so∗ (n).
Now we fix the angular momentum Ms (and, therefore, the coadjoint orbit) and assume
that rank Ms = k ≤ n (k is even). Then, according to the Darboux theorem (see, e.g., [2]),
there exist k mutually orthogonal and fixed in space vectors x(l) , y (l) , l = 1, . . . , k/2 such,
that |x(l) |2 = |y (l) |2 = hl , hl =const, and the momentum can be represented in the form
k/2
X
Ms = x(l) ∧ y (l) , that is Ms = X T Y − Y T X , (5.2)
l=1
Ẋ = X Ωc , Ẏ = YΩc . (5.3)
Now we set Ωc = U Mc + Mc U and substitute this expression into (5.3). Then taking
into account (5.2), we obtain the following dynamical system on V(k, n)
Ẋ = X [U (X T Y − Y T X ) + X T YU ],
Ẏ = Y[U (X T Y − Y T X ) − Y T X U ]. (5.4)
Notice that in the case of maximal rank k (k = n or k = n − 1), the Stiefel variety is
isomorphic to the group SO(n), and the components of vectors x(1) y (1) · · · x(k/2) y (k/2)
form redundant coordinates on it. Thus the system (5.4) describes required reduced flow
(2.5) on SO(n).
The representation (5.2) is not unique: rotations in 2-planes spanned by the vectors
x , y (l) in Rn (and only they), leave the angular momentum M invariant (in the case of
(l)
the maximal rank). As a result, the system (5.4) on SO(n) has k/2 vortex vector fields
w1 (g), . . . , wk/2 (g), which are generated by the right shifts of vectors ξ l ∈ so(n), such that
296 Mikhail V. Deryabin
ad∗ξl Ms ≡ [ξ l , Ms ] = 0, cf. Section 3.. In the redundant coordinates the fields take the
form
˙ = −(y (l) , y (l) )x(l) , l = 1, . . . , k/2.
ẋ(l) = (x(l) , x(l) )y (l) , y (l)
One can easily see that in the case of maximal rank of the momentum matrix, the corre-
sponding vortex manifolds are k/2-dimensional tori. This is a general fact: if a Lie group
is compact, then the vortex manifolds are compact manifolds, and, by the Duflo theorem,
for a dense set of the momenta ms , the vortex manifolds are tori (in our case, this dense set
is determined by the condition that the momentum rank is maximal). The torus, that passes
through the group unity, is called the maximal torus for the Lie group; maximal tori play an
important role in classification of compact Lie groups.
One can furthermore show that if the rank of the momentum is not maximal, the vortex
manifolds would be products of a torus and a certain SO(m) Lie group.
Consider the Poisson bracket on V(k, n) obtained as the Dirac restriction of the standard
bracket in Rkn .
Proposition 5.1. System (5.4) is Hamiltonian with respect to the above bracket.
Theorem 5.2. The Bernoulli surfaces for flow (5.4) are tori, and the flow is quasi-periodic
on these tori.
Proof. The reduced system on the coadjoint orbit of the group SO(n) is integrable, and
its generic invariant manifolds are tori of dimension half of the dimension of the orbit. On
the other hand, the pre-image of a generic point M of the orbit V(k, n) is a k/2-fold product
of circles S 1 × · · · × S 1 . This implies that the original system on V(k, n) has generic in-
variant tori of dimension half of the dimension of the symplectic manifold V(k, n). Hence,
the original system (5.4) is also integrable, the Bernoulli surfaces are tori, and the flow is
quasi-periodic on these tori. 2
We now introduce the controls in System (5.4) by the above scheme. Using Equation
(4.1) and the fact that any left-invariant vector field on the Lie group SO(n) in our redundant
coordinates can be written as
Ẋ = X Λ, Ẏ = YΛ, Λ ∈ so(n),
This system describes an n-dimensional rigid body with ”symmetric flywheels”, which is a
direct generalization of the Liouville problem of the rotation of a variable body [23].
Proposition 5.3. On can choose two vectors Λ1 and Λ2 , such that for any choice of the
inertia operator U , one can transform any vortex manifold to any other vortex manifold
for any momentum in space Ms , using the corresponding two control functions u1 (t) and
u2 (t).
Reduction, Hydrodynamics and Control for Geodesics... 297
Proof. First, we notice that System (5.4) preserves volume in the phase space of the
redundant variables X , Y (this can be checked by the direct computation, but the general
result of the existence of an invariant measure for a reduced system (2.5) or (2.6) with
λ = 0 follows from [20]). As the Lie algebra so(n) is semi-simple, controllability of
System (5.5) follows from Corollary 2.1, Appendix B.. Proposition 5.3 follows now from
Theorem 4.1. 2
v, w ∈ g being zero divergence vector fields on M that are tangent to the boundary.
According to the principle of stationary action, motions of the ideal fluid on the man-
ifold M are geodesics of the right-invariant metric (6.1). The latter are described by the
Euler equations on the dual algebra g∗ ,
∂[u]
= −ad∗A−1 [u] [u] = −Lv [u] , (6.2)
∂t
where the “momentum” [u] ∈ Ω1 /∂Ω0 , Av = [u]. For a particular choice of a 1-form
u ∈ [u] we get the classical equation
∂u
= −Lv u + df ,
∂t
see [3].
Now we fix the ”body momentum” uc = u0 and apply our general results on the
reduction of a geodesic flow on a group.
298 Mikhail V. Deryabin
According to Proposition 2.1, the reduced field v(g) has the form (2.6), which can be in-
terpreted in the following way. At each point g we act on the coset [u0 ] by diffeomorphisms
g −1 , then we apply the inverse operator A−1 . As a result, we obtain a zero divergence vec-
tor field on M . The vector v(g) represents a vector field on M which is obtained from the
zero divergence field A−1 Ad∗g−1 [u0 ] by applying a differential of the right shift, i.e., by the
”particle relabeling”.
Let a diffeomorphism g ∈ SDiff(M ) map initial particles’ positions x0 to x(t, x0 ). As g
is invertible, one can express x0 = x0 (x, t). The ”space momentum” 1-form us = us (x, t)
is expressed through the ”body momentum” uc (x0 ) as
∂xi0
us (x, t) = Ad∗g−1 uc (x0 ) = uci (x0 (x, t))dxj .
∂xj
Then, vs = A−1 us .
Suppose for simplicity, that the metric on M is Euclidean. Let function f be the solution
to equation ∆f = δus , such that on the boundary ∂M of the domain M the following
relation holds:
∗(us − df )|∂M = 0.
Here δ = ∗d∗, where ∗ is the Hodge operator. This solution exists, and one can write it
down explicitly, which, for example, can be done if one finds the Green’s function. The
explicit computation of the reduced field thus relies on the possibility of finding the Green’s
function.
Now,
vs = us − grad f
Indeed,
div vs = div us − div(grad f ) = 0
and v || ∂M , as ∗(us − df ) = iv µ (µ is the Euclidean volume form).
The equation on the group can be written for an inverse mapping g −1 as
d −1 ∂x0 ∂x0 ∂x0
(g ) = −Lg−1 ∗ Rg−1 ∗ ġ; =− vs x0 , .
dt ∂t ∂x ∂x
Let w(g) be a left-invariant field on the group SDiff(M ), which is defined by a vector
ξ ∈ SVect(M ): w(g) = Lg∗ ξ.
Theorem 6.1. The phase flow of the vector field w(g) on the group SDiff(M ) is a symmetry
of the reduced field v(g) if and only if the vector ξ satisfies the condition
Proof. This theorem is a corollary of Theorem 2.3 and the homotopy formula Lv ω =
iv dω + div ω. Indeed, the condition ad∗ξ [u0 ] = 0 means that Lξ [u0 ] = [Lξ u0 ] = 0. Hence,
Lξ u0 = dF . On the other hand, Lξ u0 = iξ du0 + diξ u0 . 2
Reduction, Hydrodynamics and Control for Geodesics... 299
Remark. It is interesting to note that since the isotropy vectors form a subalgebra, their
distribution is integrable. Indeed, at any point the commutator (that is minus Poisson bracket
of the vector fields) of any two vectors belongs to the same distribution.
The structure of the symmetry fields in even and odd dimensions is the following. Let
dim M = 2n. For ideal fluid flows on an even-dimensional manifold there always exists
the vorticity function λ = dun /µ, where [u] ∈ g∗ [3].
Proposition 6.2. Let ξ ∈ SVect(M ) satisfy the condition of Theorem 6.1. Then ξ commutes
with the vector field η, defined by
iη µ = dλ ∧ (du0 )(n−1) .
iη µ = (du)n .
Proposition 6.3. Let ξ ∈ SVect(M ) satisfy the condition of Theorem 6.1. Then ξ commutes
with the vorticity field η : iη µ = (du0 )n .
If the manifold M has no boundary, then the vorticity field η itself is a symmetry field
(as in this case it belongs to the Lie algebra SVect(M )).
Proof. We use following identity:
i[ξ,η] µ = Lξ iη µ − iη Lξ µ = 0 :
are Casimir functions for the Euler equations in odd- and even-dimensional case respec-
tively [17], [3]. As above, here [u] is the coset of a 1-form u, µ is an invariant volume form,
and the function f : R → R is arbitrary.
Proposition 7.1. Let dim(M ) = 2n + 1. Then the variational derivative of the Casimir
function I is a vorticity vector (n + 1)η, where iη µ = (du)n .
Proof. Since both du and dw are 2-forms, the wedge product du ∧ dw = dw ∧ du is
commutative and the derivative
d
n
(du + sdw) = ndw ∧ (du)n−1 .
ds s=0
Thus
Z Z
d n
(u + sw) ∧ (du + sdw) = w ∧ (du)n + u ∧ ndw ∧ (du)n−1
ds M s=0 M
Z
= (n + 1) w ∧ (du)n = ([w], (n + 1)η),
M
is the pairing of the dual algebra element [w] and the vector η ∈ SVect(M ). 2
Proposition 7.2. Let dim(M ) = 2n. Then the variational derivative of the Casimir func-
tion If equals ξf ∈ SVect(M ), which is a divergence-free vector field on M , linearly
dependent with the vorticity vector ξ, defined by condition iξ du = dλ.
The proof is similar to that of Proposition 7.1.
We now will need the following assumption: we will here assume that there exist a
Green’s function, i.e., the inertia operator A−1 can be written as
Z
A−1 [u](x) = Ã−1 u(x) + d G(x, y)δu(y)dn y ,
M
Reduction, Hydrodynamics and Control for Geodesics... 301
for some (n − 1)-form α. The form α is closed, as the original functional (and thus its
variational derivative) is well-defined on cosets. Thus, this variational derivative is a vector
from SVect(M ), and the theorem is proved. 2
If we can take for granted the fact that the ”typical” case, when our vortex fields do
not admit extra symmetries, is everywhere dense in the space of all 1-forms on M in some
appropriate topology, then from Theorem 7.3 follows readily that there are no invariants of
coadjoint orbits, which are the integrals of local densities, that are linearly independent of
the integrals (7.1).
Remark. The similar assumption, formulated for continuous first integrals may not be
true. Consider for example a perturbed smooth 2-dimensional Hamiltonian system. One
can always construct a continuous first integral of such systems, by setting it to constants
between the Kolmogorov tori, see, e.g., [4]. It may be possible to imitate this type of con-
struction (for low-dimensional hydrodynamics) using, for example, the Casimir functions
of the type Z
u ∧ du,
C
where C is a subset of the flow domain M , which is an invariant set of the vorticity vector
field for the instantaneous velocity.
Above results were proved in case when the manifold M has no boundary. The case of
flows on manifolds with boundaries is discussed in [9].
8. Conclusion
We considered the reduction of geodesic flows of left- or right-invariant metrics on Lie
groups to the group. The reduced vector field has a remarkable hydrodynamic interpreta-
tion: it is a velocity field for a stationary flow of an ideal fluid, the the right- or left-invariant
symmetry fields of the reduced field being vortex vector fields, i.e., they annihilate the vor-
ticity 2-form. The distribution of the vortex fields is always integrable, thus it defines a
manifold (at least locally), that we call a vortex manifold. Typically, the vortex manifolds
are tori.
We studied the following control problem. Consider a mechanical system, whose con-
figuration space is a Lie group and whose Lagrangian is invariant to left translations on that
group, and assume that the mass geometry of the system may change under the action of
internal control forces. Such system can also be reduced to the Lie group; with no controls,
it describes a geodesic flow of the left-invariant metric, given by the Lagrangian, and thus
its reduced flow is a stationary flow of an ideal fluid.
The control problem for such system is to find the conditions, under which the system
can be brought from any initial position in the configuration space to another preassigned
position by changing its mass geometry. We showed that under these conditions, by chang-
ing the mass geometry, one can also bring one vortex manifold to any other preassigned
vortex manifold. As an example, we considered the n-dimensional Euler top. We wrote
down the reduced controlled system explicitly, showed that the vortex manifolds are tori,
and proved that, by changing the mass geometry, every such torus can be transformed to
any other torus.
Reduction, Hydrodynamics and Control for Geodesics... 303
Appendix
Ṁ = M × ω, γ̇ = γ × ω,
M = Iω + Dγ × (ω × γ),
where M is the ball momentum with respect to the moving axes, fixed in the ball (momen-
tum in the body), ω is the angular velocity, γ is the unit vertical vector, also written with
respect to the moving axes, matrix I is the inertia tensor and D is a constant. One can see
that these equations are of the form (1.1).
Theorem 1.1. (The Euler theorem) The momentum in space ms is constant for the gen-
eralized Euler equations (1.1-1.2).
Proof. Differentiate relation (2.1) by time and apply (1.2), cf. [3]. 2
Proposition 2.1 relied only on relation (2.1), which turns out to be true also for this case.
Thus, reduction to the group is possible, and the reduced vector field is
B. Controllability Conditions
As system (4.1) has the standard form
X
ẋ = f (x) + ui gi (x), |u(t)| ≤ ǫ (2.1)
i
Theorem 2.1. Let the Lie group G be compact. Then system (4.1) is controllable for all
ǫ > 0, if the minimal Lie subalgebra of vector fields on G, which contains both vector fields
Lg∗ λi and v(g), spans the tangent space Tg G at any g ∈ G.
From Theorem 2.1 follows, that the minimal number of controls is mainly defined by the
Lie algebra structure. This minimal number of controls should not necessarily be equal to
the number of the degrees of freedom of the system – unless the Lie algebra is commutative.
If the dimension of the Lie algebra g is greater than 1, and we are interested in controllability
for all inertia operators and all momenta ms , then the minimal number of controls should
necessarily be greater than 1.
Proof. We only have to check that the vector field
is Poisson-stable (i.e., almost all trajectories come back to the vicinity of the initial condi-
tions infinitely many times). Indeed, if the minimal Lie subalgebra, which contains vectors
Lg∗ λi and v(g), spans the tangent space Tg G at any g ∈ G, then for each point g0 ∈ G,
the set of points g(t, g0 , u(t)), accessible by the controls u(t) for 0 < t < T , form an open
set (the point g0 itself may belong to the boundary). Under the condition of the Poisson
stability, these sets can be joined together to get the necessary trajectory, see, e.g., [22] for
details. The Poisson stability follows from the existence of a smooth invariant measure of
the reduced system ġ = v(g), as we have assumed that the group G is compact. But this
is exactly the case: if the Lie group G is compact, the reduced system always preserves a
bi-invariant Haar measure on G, see [20]. 2
Corollary 2.1. Under conditions of Theorem 2.1, let the Lie algebra g be real and semi-
simple. Then two controls is sufficient for controllability for all values of the inertia opera-
tor and of the momentum ms .
Proof. It is well known that a real semisimple Lie algebra is generated by 2 elements,
see, e.g., [6]. 2
Remark. If the momentum ms = 0, then under conditions of Theorem 2.1, system
(4.1) is controllable even if the Lie group is noncompact – this is the classical Rashevsky-
Chow theorem, see, e.g., [15]
The condition of Theorem 2.1 is usually referred to as the Lie algebra rank condition
(see, e.g., [25]). In real systems, it may be difficult to check it directly, as, in principle, the
number of commutators one has to take is not bounded from above. One can suggest using
a ”transversality” condition, see [10].
Reduction, Hydrodynamics and Control for Geodesics... 305
References
[1] Arnold V.I. Mathematical methods in classical mechanics. Springer-Verlag.
[2] Arnold V.I., Givental A.B. Symplectic Geometry. Dynamical Systems 4, Springer-
Verlag.
[4] Arnold V.I., Kozlov V.V., Neishtadt A.I. Mathematical aspects of classical and celes-
tial mechanics. Springer-Verlag, 1988. 291 p.
[5] Bao D., Ratiu T. On the geometrical origin and the solutions of a degenerate Monge–
Ampère equation. Proc. Symp. Pure Math. AMS, Providence 54 (1993), 55–68.
[7] Cardetti, F., Mittenhuber, D. Local controllability for linear systems on Lie groups. J.
Dyn. Contr. Sys., 11, 3, July 2005, 353-373.
[8] Deryabin M.V., Fedorov Yu.N. On Reductions on Groups of Geodesic Flows with
(Left-) Right-Invariant Metrics and Their Fields of Symmetry. Doklady Mathematics.
Interperiodica Translation, 68, 1, (2003) 75–78.
[9] Deryabin M.V. Ideal hydrodynamics on Lie groups. Physica D, 221 (2006), 84–91.
[10] Deryabin M.V. Control of mechanical systems on Lie groups and ideal hydrodynam-
ics. J.Math.Sci. (2008). In print.
[11] Dubrovin B.A., Novikov S.P., Fomenko A.T. Modern geometry. Vol. 2. Springer.
[12] Fedorov Yu. N. Integrable flows and Bäcklund transformations on extended Stiefel
varieties with application to the Euler top on the Lie group SO(3), J. Nonlinear Math.
Phys. 12 (suppl. 2) (2005) 7794.
[14] Ginzburg V.L., Khesin B.A. Topology of steady fluid flows. Topological aspects of the
dynamics of fluids and plasmas, eds. H.K.Moffat et al. (1992). Kluwer Acad. Publ.,
Dordrecht, 265–272.
[15] Hermann, R. (1968). Accessibility problems for path systems (2nded .). In Differential
geometry and the calculus of variations (pp. 241257). Brookline, MA: Math Sci Press.
[16] Khesin B., Misiołek G. Asymptotic Directions, Monge-Ampère Equations and the
Geometry of Diffeomorphism Groups. Journal of Math. Fluid Mech. 7 (2005) 365-
375.
[17] Khesin B.A., Chekanov Yu.Y. Invariants of the Euler equations for ideal or barotropic
hydrodynamics and superconductivity in d dimensions. Phys. D. 40 (1989), 119–131.
306 Mikhail V. Deryabin
[18] Kozlov V.V. Hydrodynamics of Hamiltonian systems. Vestn. Moskov. Univ., Ser. I.
Mat. Mekh., No. 6 (1983) 10–22 (Russian)
[19] Kozlov V.V. The vortex theory of the top. Vestn. Moskov. Univ., Ser. I. Mat. Mekh. No.
4 (1990) 56–62 (Russian)
[20] Kozlov V.V. Dynamical systems X. General vortex theory. Springer-Verlag, 2003.
[21] Kozlov V.V. Dynamics of variable systems and Lie groups. J. Appl. Math. Mech. 68
(2004), 803–808
[22] Lian, K. Y., Wang, L. S., and Fu, L. C. Controllability of spacecraft systems in a
central gravitational field. IEEE Transactions on Automatic Control, 39 (12), (1994),
2426-2441.
[24] Manakov S.V. Note on the integration of Euler’s eqations of the dynamics of an n-
dimensional rigid body. Funkts. Anal. Prilozh. 10 (1976), 93–94. English transl. in:
Funct. Anal. Appl. 10 (1976), 328–329.
[26] Marsden J.E., Weinstein A. Reduction of symplectic manifolds with symmetry. Rep.
Math. Phys. 5 (1974), 120–121.
[27] Marsden J., Weinstein A. Coadjoint orbits, Vortices, and clebsch Variables for Incom-
pressible Fluids. Physica D, 7 (1983), 305–323.
[28] Misiołek G. Stability of flows of ideal fluids and the geometry of the group of diffeo-
morphisms, Indiana Univ. Math. J. 42 (1993), 215-235.
[29] B. Palmer. The Bao-Ratiu equations on surfaces. Proc. Roy. Soc. London Ser. A 449
(1995), no. 1937, 623-627.
[30] Warner F.W. Foundations of differentiable manifolds and Lie groups. Springer-Verlag,
1983.
[31] Whittaker E.T. A treatise on analytical dynamics. 4-d ed. , Cambridge Univ. Press,
Cambridge 1960
In: Lie Groups: New Research ISBN 978-1-60692-389-4
Editor: Altos B. Canterra, pp. 307-323
c 2009 Nova Science Publishers, Inc.
Chapter 11
A.V. Greshnov 1
Institute of Mathematics of Siberian Branch
of Russian Academy of Science.
pr. Akad. Koptuga, 4., 630090, Novosibirsk
Abstract
00735-a).
308 A.V. Greshnov
Introduction
Let X = {Xi }i=1,...,N be a collection of basis vector fields defined on some do-
main O ⊂ RN , i. e. rank(X1 , . . . , XN )(g) = N, ∀g ∈ O ⇐⇒ det X(g) 6=
0 ∀g ∈ O. Hereinafter symbol X will be denote (depending on context) both
the collection of vector fields and the N × N -matrix i-column of which is coincid-
ing with Xi . Also let us consider some increasing finite sequence of numbers hi ,
i = 1, . . . , M , such that hM = N , M ≤ N . Give to every number i = 1, . . . , N
its own formal degree deg i = k, where hk−1 < i ≤ hk , h0 = 0. Set of numbers hi
and degrees deg i we call graduation of the collection X. We suppose that X ∈ C 1 ,
and the elements of the matrix X expressing (locally) in canonical coordinate sys-
tem satisfy the ¡¡homogeneity¿¿ conditions (2.1), see Proposition 2.1, Theorem 2.1.
One of hte goals of our work is the comparison of the anisotropic riemannian dis-
tance (quasidistance) dX cc , which is induced by X and agreed with our graduation,
bg
see § 2, (2.2), with the quasidistance dX cc , which is induced by the collection of
the vector fields X b g = {X b g }i=1,...,N is the homogeneous local approximation of
i
collection X, defined in some neighbourhood of fixed point g ∈ O, and X b g , see
g
definition 3.1. These collections of vector fields X and X appeared naturally in
b
different parts of analysis, non-holonomic geometry and the theory of partial dif-
ferential equations, see for example, [1–11]. The main results of our work are the
Theorems 3.1, 4.1 and the Corollaries 3.3, 4.4, which are the local approximation
theorems for corresponding quasimetrics. Interest to such problems were initiated,
from one point, by well-known Gromov’s work [1], where the local approxima-
tion theorem for Carnot-Caraéodory metrics was proved, from other point — by
works [2, 3], where the theorem of differentiability of the quasiconformal mappings
on Carnot-Carethéodory spaces equipped by quietly smooth horizontal distribu-
tions were proved. In ¡¡smooth¿¿ case we can construct the (nilpotent) tangent
cone and prove corresponding local approximation theorem with the help of the
formal Campbell-Hausdorff series, see for example [4, 5]. But when X is C 1 only,
such method is useless. In [1] it was stated the idea of the construction of collec-
tion of vector fields X b g , which was (in some way) the ¡¡basis of tangent space¿¿ of
(local) nilpotent tangent cone (or homogeneous nilpotent approximation, see [6])
for X in some neighbourhood O of the fixed point g on condition that the vector
fields Xi ∈ X ∈ C 1 , i = 1, . . . , N , satisfied the following ¡¡table of commutators¿¿
X
[Xi , Xj ](u) = Cijk (u)Xk (u), u ∈ O,
k
where Cijk (u) = 0 for deg i + deg j < deg k. For the realization of this idea Gromov
used some technique which based on the ¡¡dilatation¿¿ by the factor 1/ε of ε-scaling
vector fields εdeg i Xi expressing in the terms of coordinate system of the second kind
Some Approximation Theorems 309
In present paper we will use some definitions and notions from [4, 5]. Let
and δji is Kroncker symbols, ei is usual basis vector of Euclidean space RN . Let us
define nonhomogeneous operator of dilatation in RN δt corresponding to choosing
graduation as δt (x) = δt (x1 , . . . , xN ) = (tx1 , . . . , tM xN ). For every N -dimensional
310 A.V. Greshnov
N
P
multiindex β = (β1 , . . . , βN ) let us suppose |β|h = βi deg i; also
i=1
are uniformly bounded and equicontinuous on Boxecc (0, 1). Then there is the se-
quence of numbers εl → 0 such that vector fields (δ1/εl )∗ Xiεl , i = 1, . . . , N , con-
verge uniformly to some continuous vector fields Xbi , such that the collection {X bi }
is homogeneous with respect to the action of the operator δt .
Proof. Let us consider the following Cauchy problem for ε-scaling vector fields Xiε :
deg i i
ẋε,1 = ε
f1 (xε ),
(1.3) ...
(xε ), xε (0) ∈ Boxecc (0, ε).
ẋε,N = εdeg i fNi
Change the variables xεi = ε− deg i xε,i . Then the system (1.3) looks like
f i (εdeg 1 xε1 ,...,εdeg N xεN )
i
ẋε1 = 1 εdeg 1−deg i
= F1,ε (xε ),
...
i
fN (εdeg 1 xε1 ,...,εdeg N xεN )
δ1/ε xε (0) = xε (0) ∈ Boxecc (0, 1).
ẋεN = i
= FN,ε (xε ),
εdeg N −deg i
Using well-known Theorem of Ascoli — Arzela we get that there is the sequence of
numbers εl → 0 such that ¡¡limit¿¿ functions Fbki (x̂) = lim Fk,ε
i
l
(xεl ), deg k > deg i,
εl →0
are continuous. Denote lim xεl = x̂. We can get that
εl →0
i
Fk,εl
(xεl ) = tdeg k−deg i Fk,tε
i
l
(xtεl ),
i
and lim Fk,tεl
(xtεl ) = Fbki (δ1/t x̂), so tdeg k−deg i Fbki (δ1/t x̂) = Fbki (x̂), from what we
tεl →0
have requiring ¡¡homogenety¿¿ (1.4). It follows from (1.4), that Fbki (0) = 0. So, we
proved, that for every i = 1, . . . , N vector fields (δ1/εl )∗ Xiεl uniformly converge to
vector fields X
bi such, that
bi = (0, . . . , 0, aih
X , . . . , aihdeg i , Fbhi deg i +1 (x̂), . . . , FbNi (x̂)),
deg i−1 +1
Let x(s) is the solution of the problem (1.6); denote xt (s) = δt x(s). Then using (1.4)
we get
N
X N
X
(1.7) (δt )∗ (ẋ(s)) = ẋt (s) = αi · (δt )∗ X
bi (δ1/t xt (s)) = αi tdeg i X
bi (xt (s)).
i=1 i=1
§ 2. Quasimetrics
Lemma 2.1. Let the elements of matrix A = A(ε) satisfy following conditions of
¡¡homogemety¿¿
m
δn + o(1),
n ≤ m,
(2.1) an,m = o(1), an,m ∈ Ai,i , an,m ∈
/ diag Ai,i ,
i−j
O(ε ), an,m ∈ Ai,j , i > j,
312 A.V. Greshnov
where o(1)@ >> ε → 0 > 0 uniformly. Then the elements of matrix A−1 also
satisfy conditions of ¡¡homogemety¿¿ (2.1).
Proof. Lemma 2.1 can be proved with the help of the method of proof of the
Lemma 4.1 from [5].
−1
Remark 2.1. Denote α = (α1 , . . . , αN ), (θg,X e g . Let us consider the
)∗ Xi = Xi
following Cauchy problem
N
X
(2.3) ẋ = e g (x(s)),
αi X x(0) = b, s ∈ [0, s0 ],
i
i=1
Let
dX
cc (u, v) = max {|pi |
1/ deg i
}, dYcc (u, v) = max {|qi |1/ deg i },
i=1,...,N i=1,...,N
BoxX X
cc (g, ε) = {v ∈ O | dcc (v, g) < ε}.
Recall, see, for example [4, 5], that function d : A × A → R+ ∪ 0, defined on some
set A, is called quasimetric, if:
1) d(u, v) ≥ 0, and d(u, v) = 0 ⇔ u = v;
2) d(u, v) ≤ c1 d(v, u) for some constant c1 > 0, which doesn’t depend on u, v ∈ A;
3) d(u, v) ≤ Q(d(u, w) + d(w, v)) for some constant Q > 0, which doesn’t depend
on u, v, w ∈ A.
Pair (A, d) we call quasispace.
Proposition 2.1. Suppose that estimates (2.1) hold for am m,X m,Y
n = xn,g , yn,g , see (2.2),
−1
on θg,X BoxXcc (g, const ε) uniformly with respect to g ∈ O and ε ∈ [0, ε0 ). Then the
functions dX Y
cc (u, v), dcc (u, v) are quasimetrics on O.
Proof. Taking into account the Remarks 2.1, 2.2, we can use the arguments of the
Theorem 4.1 from [7]. For example, let us consider dX
cc (u, v). It is obvious that
dX
cc (u, v) = 0 ⇔ u = v.
N
P PN
If u = exp yj Xj (v), then v = exp − yj Xj (u), so dX X
cc (u, v) = dcc (v, u).
j=1 i=1
Let us consider points g, u, v ∈ O such, that
N
X N
X N
X
exp yj Xjε ◦ exp xj Xj (g) = exp yj Xjε (u) = v,
j=1 j=1 j=1
(2.4) dX
cc (v, g) ≤ Q(ε + ).
We have
N
X N
X N
X
θg−1 (v) = exp e g (0) = exp
pj X ε deg Xj
y X
j j
e g
◦ exp e g (0),
deg Xj xj X
j j
j=1 j=1 j=1
314 A.V. Greshnov
N
e g = (θ−1 )∗ Xj . Let θ−1 (v) = p = (p1 , . . . , pN ), Ye ε = P εdeg Xj yj X
where X eg.
j g g j
j=1
Then p = η(1), where
so
Z1
(2.5) p = δ x + Ye ε (η(s)) ds.
0
Theorem 2.1. Suppose that estimates (2.1) hold for am m,X m,Y m,Y m,X
n = xn,g , yn,g , xn,g , yn,g
−1 e −1 e
on sets θg,X Boxcc (0, ν · ε), θg,Y Boxcc (0, ν · ε) uniformly with respect to g ∈ O,
ε ∈ [0, ε0 ), and ν > 1 is some constant, which doesn’t depend on g, ε. Then
n N N
X X o
max dX t
Y
Xit (uε ), uε
(2.7) cc exp Yi (uε ), uε , dcc exp
i=1 i=1
deg
X i−1 1/ deg i
≤ C1 · max tdeg i + tdeg i−k · O(εk ) , C1 = const,
i=1,...,N
k=1
where uε ∈ BoxX Y
cc (g, ν · ε) ∩ Boxcc (g, ν · ε); in particular,
(2.8) BoxX Y X
cc (g, c2 · ε) ⊂ Boxcc (g, c3 · ε) ⊂ Boxcc (g, c4 · ε)
and Ci,uε ,Y , ĉi,k,Y are uniformly bounded. From (2.11) we get (2.7) for dX Y
cc ; for dcc
estimate (2.7) can be proved by the same way; (2.8) follows from (2.7).
Proof. Corollary 2.2 follows from (2.3); at that O(ε) can’t be o(ε) in general because
Ci,uε ,Y , ĉi,k,Y from (2.11) are not equal Ci,uε ,X , ĉi,k,X in general.
Fix some point g ∈ O and sufficiently small number ε0 > 0 such that BoxX cc (g, ε0 ) ⊂
O. For collection X from § 2 let us consider vector fields (δ1/ε )∗ X e g,ε , Xe g,ε =
i i
g
e , i = 1, . . . , N , on Boxecc (0, ε0 ) = θ−1 (BoxX
εdeg i Xi g,X cc (g, ε0 )). We suppose that
collection of vector fields X such that the following uniform convergences take place
on Boxecc (0, ε0 )
Then, taking into account remark 2.2, from (3.1) we get (2.1) for vector fields belong
eg.
to X
0,g
Property 3.1. Vector fields X bm , m = 1, . . . , N , are 10 basis, 20 homogeneous with
respect to the action of the operator δt , 30 segment of the straight line (β1 s, . . . , βN s),
s ∈ [0, 1], bi = const, is solution of the following Cauchy problem
N
X
ẏ(s) = b 0,g (y(s)),
βi X y(0) = 0, s ∈ [0, 1].
m
m=1
Proof. Item 10 follows from (2.1), (3.1), item 20 follows from the Proposition 1.1.
Let us prove item 30 . For this let us consider segment of the straight line (α1 s, . . . , αN s),
316 A.V. Greshnov
s ∈ [0, 1], which is solution of (2.3), see remark 2.2. Let αi = εdeg i βi , |βi | ≤ const.
N
Then we change the variables y ε = δ1/ε x in (2.3) and get ẏ ε (s) =
P e g (δε y ε (s)),
(δ1/ε )∗ εdeg i βi Xi
m=1
y ε (0) = 0, s ∈ [0, 1]. It is not difficult to see that y ε (s) = (β1 s, . . . , βN s), and then,
using (3.1), we get item 30 .
g 0,g b g = {X g
Definition 3.1. Let us denote X bm = (θg )∗ X
bm , X bm }m=1,...,N . We say
g
that the collection of vector fields X b is local homogeneous approximation of the
collection of the vector fields X with respect to the action under operator ∆g,X ε =
−1
θg,X ◦ δε ◦ θg,X .
N
P
Consider the mapping θu,Xb g : (x1 , . . . , xN ) → exp b g (u), u ∈ BoxX
xi X cc (g, ε0 ).
i
i=1
Certainly we have X b g ∈ C, so the mapping θ̂ b g may be false generally, be-
g,X
cause an ordinary differential equation with continuous right side may have more
than one solution. So later we will suppose (for simplicity) that X b g ∈ C 1 . In
this case we can guarantee that the mapping θ̂u,Xb g is diffeomorphism. Then from
item 30 of the Property 3.1 it follows that θg,X (x1 , . . . , xN ) = θg,Xb g (x1 , . . . , xN ), so
bg bg
BoxX X X
cc (g, ε) = Boxcc (g, ε). Let dcc be quasimetric induced in O by the collection
b , agreed with the graduation. Note that quasimetrics dX
g Xbg
of vector fields X cc , dcc
satisfy the conditions of the Theorem 2.1.
bg bg bg
Property 3.2. Let u, v ∈ BoxX X g,X g,X X
cc (g, ε0 ). Then dcc (∆ε u, ∆ε v) = εdcc (u, v).
N
P
Using the arguments of the proof of (1.7), we get ∆g,X
ε v = exp
b g (∆g,X
ai εdeg i X ε u).
i
i=1
bg
Then, using definition of quasimetric dX
cc , we get the Property 3.2.
N N
Let Wε,ω =
P
ωi εdeg i Xi , Wε,ω
b g , where ε ≤ ε0 , ω = (ω1 , . . . , ωN ),
c g = P ωi εdeg i X
i
i=1 i=1
ωi = const, |ω| < c = const.
(3.2) (∆g,X ε bg
1/ε )∗ Xi ⇒ε→0 Xi , i = 1, . . . , N,
N
Cauchy problem ẋ(s) =
P b g (x(s)), x(0) = u, s ∈ [0, 1], is unique for every
ωi Xm
m=1
point u. So, using (3.2), we have the following uniform convergence, see [13, Theo-
rem 2.4],
∆g,X cg
1/ε exp(sWε,ω )(uε ) ⇒ε→0 exp(sW1,ω )(u), s ∈ [0, 1],
bg g,X cg
from what it follows dXcc ∆1/ε exp(Wε,ω )(uε ), exp(W1,ω )(u) = rε,ω,u = o(1), where
rε,ω,u is uniform as o(1) with respect to u, ε, ω. Then, using Property 3.2, we get
bg
dX cg
cc exp(Wε,ω )(uε ), exp(Wε,ω )(uε ) = o(ε), from what, using the method of proof
of Theorem 2.1, we get dX cc -estimate.
Lemma 3.1. There is a constant κ > 0 such, that for all g ∈ O and sufficiently
small 1 , 2 ∈ (0, ε), ε ∈ [0, ε0 ] the following inclusion holds
[
BoxX X
cc (v, 2 ) ⊂ Boxcc (g, 1 + κ2 ).
v∈BoxX
cc (g,1 )
Proof. Taking into account Remark 2.1, Remark 2.2, Lemma 3.1 follows from the
estimates (2.6).
Proof. Corollary 3.3 follows from Corollary 3.2 and Lemma 3.1, see also Corol-
lary 3.2 from [5].
§ 4. Some Applications
4.1. Consider some C 1 -smooth basis canonical (see Remark 2.2) collection of vector
ei }i=1,...,N in some neighbourhhod of origin O ⊂ RN such that
e = {X
fields X
X
(4.1) [X
ei , X
ej ] = Cijk (x)X
ek ,
k
318 A.V. Greshnov
where Cijk (x) = 0 if deg ei + deg ej < deg ek , and C ∞ -smooth basis canonical
collection of vector fields Xb 0 = {X b 0 }i=1,...,N , such that
i
X
bi0 , X
[X bj0 ] = C bk0 ,
bijk X
deg ei +deg ej =deg ek
By induction we proved in [7] the following estimates (but it is not difficult to get
these estimates without assistance)
δj + εc̃jg,i (xε ),
( i
i ≤ hdeg ej ,
wi (1) =
εdeg ei −deg ej c̃jg,i (xε ), i > hdeg ej ,
for every i = 1, . . . , N and some continuous functions c̃jg,i . Then we can write (4.3)
as
(4.4)
δ k e + O(ε), k ≤ hdeg ej ,
j jP
deg ek −deg ej deg ek −deg ej +1
∂t wk = Cilk · wi · ε xl + O(ε ), k > hdeg ej .
deg ek
=deg ei +deg el
Also in [7] we proved that ŵ = ŵ(t) = tVb (tδε x)hej i, ŵ = {ŵk }k=1,...,N , for some
fixed j = 1, . . . , N is solution of the following Cauchy problem
Theorem 4.1. We have wk (δε x) = ŵk (δε x) + O(ε) for k ≤ hdeg ej , wk (δε x) =
ŵk (δε x) + O(εdeg ek −deg ej +α ) for k > hdeg ej .
Proof. Step 1. Case k ≤ hdeg ej obviously follows from (4.6), (4.8). Step 2. Consider
the case when hdeg ej < k ≤ hdeg ej +1 . From conditions of adding deg ek = deg ei +
deg el it follows that deg ek > deg ei , so i ≤ hdeg ej . Then using identities wi =
ŵi + O(ε), i ≤ hdeg ej , and (4.6), we get
(4.9) ∂t wk − ∂t ŵk = O(εdeg ek −deg ej +α ), t ∈ [0, 1],
then
(4.10) wk = ŵk + O(εdeg ek −deg ej +α ) = ŵk + O(ε1+α ), hdeg ej < k ≤ hdeg ej +1 .
Step 3. Consider the case hdeg ej +1 < k ≤ hdeg ej +2 . From conditions of adding it
follows that i ≤ hdeg ej +1 , because if hdeg ej < i, then, using (4.10), we get (4.9); if
hdeg ej ≥ i, then, using step 1, we also get (4.9). So we have
wk = ŵk + O(εdeg ek −deg ej +α ) = ŵk + O(ε2+α ), hdeg ej +1 < k ≤ hdeg ej +2 .
It is clear that we shell get necessary estimates in the rest of cases by the same way.
Corollary 4.1. We have V (δε x) = Vb (δε x)+B(δε x), where B(δε x) = (bi,j )i,j=1,...,N ,
O(ε), hk < i, j ≤ hk+1 , k = 0, . . . , M,
bi,j = O(ε), i ≤ j,
O(εdeg ei −deg ej +α ) = O(εl−k+α ), hl < i ≤ hl+1 , hk < j ≤ hk+1 , k − l ≥ 1.
N N
b 0 = P ηj X
Corollary 4.2. We have X
P
η̂ij X
b 0 , where
i i , Xj =
e e
j i
i=1 i=1
δij + O(ε), 1 ≤ j ≤ hj ,
(4.11) ηij , η̂ij =
O(εdeg ei −deg ej +α ), j > hj .
4.2. Here we consider collection of vector fields X from Introduction such that
X ∈ C 2 ; so (θg−1 )∗ X ∈ C 1 and
N N
where Wε,ω =
P
ωj X ε , Wε,ω
b g,ε , ω = (ω1 , . . . , ωN ), |ω| ≤ c = const,
c g = P ωj X
j
j=1 j=1
and {Xb g }j=1,...,N = X
b g is local homogeneous approximation of X with respect to
j
the action under operator of dilatation ∆g,X
ε .
e gε ,ε (x(s)), x(0) = 0,
Let us consider the following Cauchy problems: ẋ(s) = Xj
s ∈ [0, 1],
N N N
b 0,gε ,
X X X
(4.14) ż(s) = (θg−1 ) Xb g,ε = εj ηjg,i · X
e gε = εdeg i ηjg,i · ηigε ,k · X
ε ,X ∗ j i k
i=1 i=1 i=1
z(0) = 0, s ∈ [0, 1],
Some Approximation Theorems 321
where ηjg,i , ηkgε ,i is defined by the Corollary 4.2. Using (4.11), we can write (4.14)
as
N
b 0,gε ,
X
(4.15) ż(s) = εdeg j φij · Xi z(0) = 0, s ∈ [0, 1],
i=1
where functions φij satisfy the estimates (4.11). Let x(s) = (x1 (s), . . . , xN (s)),
z(s) = (z1 (s), . . . , zN (s)). We have xi (s) = δij εdeg i s, so ẋi (s) = δij εdeg i , i =
1, . . . , N . Then, using (4.11), (4.13), (4.15), we get żi (s) − ẋi (s) = O(εj+1 ) for
i = 1, . . . , hj , from what
After that let us consider zi (s) for i = hj + 1, . . . , hj+1 . Using the conditions of the
adding from (4.13), and also (4.11), (4.15), (4.16), we get żi (s)− ẋi (s) = O(εj+2 ) for
i = hj + 1, . . . , hj+1 , from what it follows zi (s) = O(εj+2 ), i = hj + 1, . . . , hj+1 .
Using the same arguments, by induction we get
Let p(s) is the line segment, connecting the points x(1) = p(0) and z(1) = p(1).
Then, using (4.16), (4.17) and Lemma 2.1, Lemma 4.1, we get
N
X N
X N
X
gε e gε
(4.18) ṗ(s) = (zi (1) − xi (1))ei = (zi (1) − xi (1)) vi,j ·Xi
i=1 i=1 j=1
hj N
X X
= e gε +
O(εj+1 ) · X e gε .
O(εdeg i+1 ) · X
i i
i=1 i=hj +1
−1
Considering (4.18) in the terms of coordinates θv,X , where v = θgε ,X (x(1)), we
1
get the estimate dX
cc (exp(Xb g,ε
)(gε ), exp(X ε
)(gε )) = O(ε 1+ M
). One can get another
estimate by the same manners. One can get general case by the same manners also,
but it needs to pay in attention the following equalities
N
X N
X XN
cg =
Wε,ω ωj εdeg ej · ηij Xi = φ̂j Xj , φ̂j = ωj εdeg ej + O(εdeg j+1 ),
j=1 i=1 j=1
N
X N
X N
X
Wε,ω = ωj εdeg ej · η̂ij X
bg =
i φj Xj , φj = ωj εdeg ej + O(εdeg j+1 ).
j=1 i=1 j=1
Corollary 4.3. There is a constant κ̃ > 0 such that the following inclusions
1 bg 1
BoxX X
cc (gε , λε) ⊂ Boxcc (gε , λε + κ̃ε
1+ M
), BoxX X
cc (gε , λε) ⊂ Boxcc (gε , λε + κ̃ε
1+ M
)
e
Proof. Corollary 4.3 follows from the Theorem 4.2 and the Lemma 3.1, see also
Corollary 3.2 from [5].
c g )(u), u ∈ BoxX
4.3. Let us consider the curves exp(sW cc (g, ε0 ), from the Theo-
1,ω
rem 3.1 as an elements of some quasispace (BoxX Y
cc (g), dcc ), where the collection of
b
Note that [X
b1 ; X b3 , and [ei , ej ] = 0 ∀i, j = 1, 2, 3. We have
b2 ] = 4X
3
X
(4.20) wε = exp b ε (δε u)
ωi Xi
i=1
= exp ω1 eε1 + ω2 eε2 + (2ω1 y − 2ω2 x + ω3 )eε3 (δε u) = (wε1 , wε2 , wε3 ).
Apart, dYcc (δε u, δε v) = εdYcc (u, v) for every suitable u, v ∈ Boxecc (0, ε0 ). Denote
b b
3
X
vε = exp eiε (δε u) = (v 1 , v 2 , v 3 ).
ωi X ε ε ε
i=1
Using the Theorem 3.1, see also [5], we have dXcc (vε , wε ) = o(ε), and from (4.19), (4.20)
b
we get
dYcc (vε , wε ) = max {|vεi − wεi |1/ deg i } = o(ε).
b
i=1,2,3
4.4. Let us consider the quasimetric tdX cc on the domain O, where X from § 3.
Denote
Bt (u, r){v ∈ O | tdX
cc (u, v) < r}.
Some Approximation Theorems 323
Theorem 4.3. Let us consider the sequence of compact quasispaces (B tk (g, r), tk dXcc ),
where tk → ∞, r < ε0 . Then for every number > 0 and every finite -
g,X
dense net Γk ⊂ (B tk (g, r), tk dX
cc ) the set ∆t Γk is ck -dense net for quasispace
bg
X
(Boxcc (g, r), dX
cc ), and ck @ >> k → ∞ > 1.
e
Proof. Using the Theorem 3.1 and the Property 3.2, one can prove Theorem 4.3
with the method of the proof of the Theorem 6.1 from [5].
Remark 4.2. In is not difficult to see, using the Property 3.2, that the Theorem 4.3
is not correct, if X = X,
e X b g = Yb , where X,
e Yb are from the Example 4.1.
References
[1]. Gromov M., Carnot-Caratheodory spaces seen from within, Sub-Reimannian
geometry, Basel: Birkhäuser, 1996, pp. 79–323.
[2]. Margulis G. A., Mostow G. D., The differential of quasi-conformal mapping of
a Carnot–Carathéodory spaces, Geometric and Functional Analysis 5 (1995),
no. 2, 402–433.
[3]. Vodopyanov S. K., Greshnov A. V., On differentiability of mappings of Carnot-
Carathéodory spaces, Dokl. Ross. Akad. Nauk. 389 (2003), no. 5, 592–596.
(Russian)
[4]. Greshnov A. V., Metrics and Tangent Cones of Uniformly Regular Carnot-
Carathéodory Spaces, Sib. Maht. Journal. 47 (2006), no. 2, 209–238. (English)
[5]. Greshnov A. V., Local Approximation of Uniformly Regular Carnot-Carathéodory
Quasispaces by Their Tangent Cones, Sib. Math. Joural. 48 (2007), no. 2, 229–
248. (English)
[6]. Belläiche A., The tangent space in sub-Riemannian geometry, Sub-Reimannian
geometry, Basel: Birkhäuser, 1996, pp. 1–78.
[7]. Greshnov A. V., Application of the methods of group analysis ot differential
equations for some collections of C 1 -smooth non-commuting vector fields, Sib.
Math. Journal. (2008), (to appear).
[8]. Vodopyanov S. K., Differentiability of Mappings in Geometry of Carnot Man-
ifolds, Sib. Math. Journal. 48 (2007), no. 2, 197–213. (English)
[9]. Mitchell J., On Carnot-Carathéodory metrics, J. Differential Geometry 21
(1985), 35–45.
[10].Metivier G., Fonction spectrale et valeurs proposes d’une classe d’operateurs,
Comm. Partial Differential Equations 1 (1976), 479–519.
[11].Rothchild L. P., Stein E. S., Hypoelliptic differential operators and nilpotent
groups, Acta Math. 137 (1976), 247–320.
[12].Pontryagin L. S., Ordinary Differential Equations, Fizmatgiz, Moscow, 1961.
[13].Helgason S., Differential Geometry and Symmetric Spaces, Academic Press,
New York and London, 1962.
[14].Hartman P., Ordinary Differential Equations, John Wiley& Sons, New York-
London-Sydney, 1964.
[15].Ovsyannikov L. V., Group Analysis of Differential Equation, Nauka, Moscow,
1978.
In: Lie Groups: New Research ISBN 978-1-60692-389-4
Editor: Altos B. Canterra, pp. 325-349
c 2009 Nova Science Publishers, Inc.
Chapter 12
L IE T HEORY IN P HYSICS
Gabriela P. Ovando∗
CONICET y ECEN-FCEIA, Universidad Nacional de Rosario,
Pellegrini 250, 2000 Rosario, Santa Fe, Argentina
Abstract
The purpose of this material is to review the Adler Kostant Symes scheme as a
theory which can be developped succesfully in different contexts. It was useful to
describe some mechanical systems, the so called generalized Toda, and now it was
proved to be a tool for the study of the linear approach to the motion of n uncoupled
harmonic oscillators. The complete integrability of these systems has an algebraic
description. In the original theory this is related to ad-invariant functions, but new
examples show that new conditions should be investigated.
1. Introduction
In this work we are interested in the use of Lie theory to understand some Hamiltonian sys-
tems. For the study of completely integrable systems one needs to identify the following: i)
the symplectic structure, which gives the system its Hamiltonian character, ii) first integrals
or constants of motion, iii) action angle variables, and the computation of their evolution.
Indeed this is a very difficult approach but it is possible for systems related to certain Lie
groups.
To this end there are several methods with a common idea: the realization of canonical
equations on Lie algebras, or on orbits of a certain action or on symmetric spaces. These
ideas appeared in the 70’s and were developped, under other by several authors as Adler,
Fomenko, Kostant, Mischenko, Olshanetsy, Perelomov, Trofimov, Symes, etc. (see for
instance [Ad1] [F-M1] [F-M2] [F-T] [Ko1] [O-P] [P] and [Sy] and their references).
In any method all of the above steps i) ii) iii) are reflected by algebraic circumstances.
One needs a way for imbedding a certain Hamiltonian system into a Lie algebra, effective
∗
E-mail address: ovando@mate.uncor.edu. The author GO was partially supported by CONICET, ANPCyT
and SECyT-UNC
326 Gabriela P. Ovando
methods for constructing sets of involution and the proof of the full integrability of a wide
family of functions in involution.
In this chapter we are concerned with so called Adler-Kostant-Symes scheme, which
brings together a mathematical framework with Lie theory but also consequences in the
dynamics of the Hamiltonian system. This method was successful when studying some
mechanical systems such as the rigid body or the generalized Toda lattice [Ad2] [Ko2] [Sy]
[R2]. In this setting the phase space of the Hamiltonian systems become coadjoint orbits
represented on a Lie algebra and the functions in involutions are presented as ad-invariant
functions. On the one hand for this kind of functions, the corresponding Hamiltonian sys-
tems become a Lax equation and on the other hand they are in involution on the orbits.
Whenever studying Poisson commuting conditions the ad-invariance property can be re-
placed by a weaker one as in [R1]. In the framework of this theory what we need is a Lie
algebra with an ad-invariant metric, a splitting of this Lie algebra into a direct sum as vector
subspaces of two subalgebras and a given function. These algebraic tools were used with
semisimple Lie algebras, where the Killing form is the natural candidate for the ad-invariant
metric.
However there are more Lie algebras admitting an ad-invariant metric. We shall exam-
plify here how can be applied the theory for semisimple Lie algebras, and also for other
ones, such as the solvable ones. For the general case one should see that any Lie algebra
with an ad-invariant metric can be constructed by a double extension procedure, whose
more simple application follows from Rm . In this way one gets a solvable Lie algebra g,
that results a semidirect extension of the 2n+1-dimensional Heisenberg Lie algebra hn and
that can be endowed with an ad-invariant metric which is an extension of a non degenerate
bilinear form on R2n . But for other cases the resulting Lie algebras could be no semisimple
and no solvable.
In any case, the Lie algebra g splits naturally as a direct sum of vector spaces of two sub-
algebras. Looking at the coadjoint orbits of one of the Lie subalgebras, one gets Hamilto-
nian systems on these orbits and one can identify the original Hamiltonian system with one
of these. In particular for the restriction of the quadratic corresponding to the ad-invariant
metric we obtain a Hamiltonian system that becomes a Lax equation, whose solution can
be computed with the Adjoint representation.
As example we work out the Toda lattice and the linear equation of motion of n-
uncoupled harmonic oscillators. The first one corresponds to a semisimple Lie algebra,
and the second one is associated to a solvable one. Furthermore it is proved that the Hamil-
tonian for the last one is completely integrable on all maximal orbits. We notice that the
functions in involution we are making use, are not ad-invariant and they do not satisfy the
involution conditions of [R1].
The setting for the second example applies for quadratic hamiltonians. The Poisson
commutativity conditions we get for some polynomials can be read off in the Lie algebra
sp(n) of derivations of the Heisenberg Lie algebra of dimension 2n+1 hn . In particular for
the case of the motion of n-uncoupled harmonic oscillators we need a abelian subalgebra in
the Lie algebra of isometries of the Heisenberg Lie group Hn , endowed with its canonical
inner product.
This is not surprising if we consider that symplectic automorphisms of the Heisenberg
Lie group produce symplectic symmetries of p-mechanical, quantum and classical dynam-
Lie Theory in Physics 327
ics for more general systems than the linear ones (see [Ki2]).
The appearence of the Heisenberg Lie algebra related to the motion of n-uncoupled
harmonic oscillators is not so surprising. In fact, it is known that in quantum mechanics a
good approach to the simple harmonic oscillator is through the Heisenberg Lie algebra. In
dimension three this is the Lie algebra generated by the position operator Q = multiplication
d
by x, the momentum operator P = −i dx and 1 with the only non trivial commutation
relation
[Q, P ] = 1
These operators evolve according to the Heisenberg equations
dP dQ
= −Q =P
dt dt
An attempt to relate the classical mechanical system of the linear approximation of the
motion of n-uncoupled harmonic oscillators was presented by the theory of p-mechanics,
which makes use of the representation theory of the Heisenberg Lie group to show that both
quantum and classical mechanics can be derived from the same source (see for instance
[Ki1] [Ki2]). This theory contructs a more general setting that unifies both quantum and
classical mechanics. The starting point for p-mechanics is the method of orbit of Kirillov
[K1] [K2], which says that the orbits of the coadjoint representation of the Heisenberg Lie
group parametrise all unitary irreducible representations [F]. Thus non commutative repre-
sentations are known to be connected with quantum mechanics. In the contrast commutative
representations are related to classical mechanics in the observation that the union of one
dimensional representations naturally acts as the classical phase space in p-mechanics. In
this theory the time evolution of both quantum and classical mechanics observables can be
derived from the time evolution of p-observables, choosen as particular functions or distri-
butions on the Heisenberg Lie group.
These considerations allow to suppose that new applications of the Adler Kostant Symes
scheme are possible and maybe it comes a new time to understand old mechanical systems
with new tools, which should be developped for these purposes. As an introduction to the
topic one can find exceptional ideas in the books of Arnold, Abraham and Marsden, Ratiu
and Marsden, etc. all of them classics in the literature concerning classical mechanics.
The chapter is organised as follows: in the first part we present basic ideas concerning
symplectic geometry. The second part is devoted to the Adler-Kostant-Symes scheme and
the third part to the examples: on the one hand the Toda lattice with generalization in ([Ko2]
and [Sy]), and on the other hand the systems corresponding to quadratic Hamiltonians on
R2n .
ω(X, Y ) = (X, JY )
Example 2.3. If (M1 , ω1 ) and (M2 , ω2 ) are symplectic manifolds, the direct product M1 ×
M2 is a symplectic manifold.
Example 2.4. Coadjoint orbits. Let G denote a Lie group with Lie algebra g and let g∗ be
the dual space of g. The coadjoint action of G on g∗ is defined as:
g · ϕ = ϕ ◦ Ad(g −1 ) g ∈ G, ϕ ∈ g∗ .
Notice that the orbit throught ϕ is the set G · ϕ = {g · ϕ : g ∈ G} and the isotropy subgroup
at ϕ is Gϕ = {g ∈ G : ϕ ◦ Ad(g −1 ) = ϕ}; thus as usual one has G · ϕ = G/Gϕ .
The action of G on g∗ induces an action of g on g∗ as
X · ϕ = −ϕ ◦ ad(X) X ∈ g, ϕ ∈ g∗
in other words
X̃(ϕ) = −ϕ ◦ ad(X)
is the infinitesimal generator induced by X ∈ g at ϕ ∈ g∗ .
Any coadjoint orbit G · ϕ is a symplectic manifold with the 2-form
and this is called the Hamiltonian vector field associated to the Hamiltonian function H.
Example 2.6. For the standard symplectic structure on R2n the isomorphism Kx is given
by Kx v = −Jv, where J denotes the canonical complex structure on R2n .
Let H ∈ C ∞ (R2n ), its associated Hamiltonian vector field is
X ∂H ∂ ∂H ∂
XH (m) = J(∇H) = ( − ),
∂yi ∂xi ∂xi ∂yi
i
Example 2.8. Let H be a smooth function on R2n , the Hamiltonian equation is the classical
one
∂f
x′i = ∂y i
∂f
yi′ = − ∂x i
where xi is actually xi (t), that is it depends on t, for all i (and also for any yi ).
Example 2.9. On R2n a quadratic Hamiltonian is a smooth function as
1
H(x) = (Ax, x) for A symmetric linear map,
2
which yields the Hamiltonian system
x′ = JAx (3)
In classical mechanic this system describes “small oscillations”, that is, it approximates the
motion of a particle on Rn or equivalently the motion of n uncoupled particles on R, near
an equilibrium position.
330 Gabriela P. Ovando
For instance the motion of n-uncoupled harmonic oscillators near an equilibrium po-
sition can be approximated with H a quadratic Hamiltonian as above by taking A = I;
therefore (3) becomes
x′i (t) = yi (t)
(4)
yi′ (t) = −xi (t)
where x(t) = (x1 (t), . . . , xn (t), y1 (t), . . . , yn (t)).
In classical mechanics it is usual to name the coordinates as xi the position coordinates
and yi as the velocity coordinates for every i = 1, . . . , n.
LX = iX d + diX
LX iY = iLX Y + iY LX = i[X,Y ] + iY LX
Proposition 2.12. A vector field X ∈ χ(M ) is symplectic if and only if the one parameter
subgroup ψt generated by X is symplectic.
Proposition 2.15. Let (M, ω) be a symplectic manifold such that H 1 (M, R) = 0. Every
symplectic vector field on M is Hamiltonian.
Proposition 2.16. Let C ∞ (M ) is a Lie algebra under the Poisson bracket defined above
and f → Xf is a Lie algebra anti-homomorphism of C ∞ (M ) into χ(M ).
LX θ = iX dθ + diX θ.
Therefore
i[Xf ,Xg ] ω = d{f, g}
Thus the left side of the equality above i[Xf ,Xg ] ωx coincides with −Kx (X{f,g} , and since
Kx is an isomorphism [Xf , Xg ] = −X{f,g} .
Example 2.18. Let g be a Lie algebra and g∗ its dual. As usual one identifies g∗ with its
tangent space. Given a function F : g∗ → R, we define the gradient of F at α ∈ g∗ ,
denoted by ∇F (α), as an element ∇F (α) ∈ g such that hβ, ∇F (α)i = dFα (β) for any
β ∈ g∗ , where h , i denotes the evaluation map.
The Kirillov’s Poisson bracket on g∗ is given by
{f, h}(α) = halpha, [∇f (α), ∇h(α)]i.
Proposition 2.19. If {f, g} = 0 then g is constant on the integral curves of Xf .
Proof. Assume x′ (t) = Xf (x(t)) then
d
g(x(t)) = dg(x′ (t)) = Xf (g)(x(t)) = {f, g}(x(t)) = 0.
dt
Coordinates as above, are called action-angle variables for the Hamiltonian system
of f .
Recall that in (2.4) we defined the coadjoint action of a Lie group G on the space g∗ ,
the dual of the Lie algebra
g · ϕ = ϕ ◦ Ad(g −1 ) g ∈ G, ϕ ∈ g∗
X · ϕ = −ϕ ◦ ad(X) X ∈ g, ϕ ∈ g∗ .
G · m = {g · m : g ∈ G}
and the isotropy or stabilizer group of the action at the point m is the closed subgroup of G
given by
Gm = {g ∈ G such that g · m = m}.
It is known that the orbit at m is diffeomorphic to the quotient space of G and the
isotropy group, G · m ≃ G/Gm (see [Wa] for instance). Thus any curve at the orbit G · m
through m is γ(t) = exp tX · m and this generates the infinitesimal vector X̃ at Tm (G · m)
by
d
X̃(m) = exp tX · m.
dt |t=0
Hence the tangent space of a G-orbit at m is
Tm (G · m) = {X̃, X ∈ g}
The coadjoint orbits are examples of symplectic manifolds. Recall that they are en-
dowed with the 2-form given by:
hence
where the last equality holds after Jacobi for [·, ·].
Lie Theory in Physics 335
The 2-form ω is non degenerate on a orbit: let ϕ ∈ g∗ and let X ∈ g such that
ω(X̃, Ỹ ) = 0 for all Y ∈ g.
Then −ϕ([X, Y ]) = 0 for all Y ∈ g, says that X · ϕ = 0 implying that X ∈ L(Gϕ ).
In fact exp tX ∈ Gϕ if and only if exp tX · ϕ = ϕ for t near 0. Thus taking derivative at
t = 0 we have X · ϕ = 0, and this is the set corresponding to the Lie algebra of Gϕ . Since
the tangent space of the orbit at ϕ is Tϕ (G · ϕ) = g/L(Gϕ ), one gets X̃ = 0.
implying that the map ℓ : X → ℓX is equivariant. Thus the adjoint orbits become symplec-
tic manifolds with the 2-form:
We shall consider these ideas to construct Hamiltonian systems on orbits that are in-
cluded on Lie algebras.
Recall that given a metric h , i on g the gradient of a function f : g → R at the vector
X ∈ g is defined by
g = g+ ⊕ g−
336 Gabriela P. Ovando
as a direct sum of linear subspaces, that is (g, g+ , g− ) is a product structure on g. The Lie
algebra g also splits as
g = g⊥ ⊥
+ ⊕ g− ,
and
g⊥
± is isomorphic as vector spaces to g∗∓ .
This follows from the isomorphism ℓ : g → g∗ . In fact, let X ∈ g⊥ + maps to ℓX . Since
⊥ ∗
ℓX (Y ) = 0 for all Y ∈ g+ , the image of ℓ(g+ belongs to g− , and the isomorphism follows
from dimensions.
Let G− denote a subgroup of G with Lie algebra g− . Then the coadjoint action of G−
on g∗− induces an action of G− on g⊥ ⊥
+ : for g− ∈ G− , X ∈ g+ , Y ∈ g− one has:
and ℓ : g⊥ ∗
+ → g− is equivariant.
The infinitesimal generator corresponding to Y− ∈ g− is
d
Ỹ− (X) = exp tY− · X = πg⊥ ([Y− , X]) X ∈ g⊥
+.
dt |t=0 +
The orbit G− · Y becomes a symplectic manifold with the symplectic structure given by
which is induced from the Kostant-Kirillov-Souriau symplectic form on the coadjoint orbits
in g∗− .
Consider a smooth function f : g → R and restrict it to an orbit G− · X := M ⊂ g⊥ +.
Then the Hamiltonian vector field of the restriction H = f|M is the infinitesimal generator
corresponding to −∇f− , that is
h[g+ , g⊥ ⊥
+ ], g+ i = hg+ , [g+ , g+ ]i = 0.
The phase space is R2n which is a symplectic manifold with its canonical symplectic struc-
ture. It follows that the Hamiltonian equation is
x′k = yk
(11)
yk′ = exk−1 −xk − exk −xk+1
and with ex0 −x1 = 0 = exn −xn+1 . Flaschka considered a change of coordinates (called
Falschka transform) as follows
where
ak = − 12 yk 1 ≤ k ≤ n,
1
bk = 21 e 2 xk −xk+1 1≤k ≤n−1
xn
bn = 21 e 2
Therefore the equation (11) yields
Consider g the semisimple Lie algebra of traceless real matrices sl(n, R equipped with
the ad-invariant metric hx, yi = tr(x, y) for all x, y ∈ sl(n, R).
Let g+ = so(n) the Lie subalgebra of skew symmetric real matrices and g− the Lie
algebra of upper triangular matrices of trace zero.
Then g⊥ ⊥
+ is the space of real symmetric matrices in sl(n, R) and g− ) is the space of
strictly upper triangular matrices in sl(n, R).
Pn−1
The coadjoint orbit M = G− ·x0 for x0 = i=1 ei,i+1 +ei+1,i is the set of tri-diagonal
real symmetric matrices
n
X n−1
X X
ai Ei,i + bi (Ei,i+1 + Ei+1,i ) ai = 0, bi > 0 ∀i.
i=1 i=1 i
where Ei,j denotes the matrix with a 1 at the place i, j and 0 in the others components.
Lie Theory in Physics 339
1 ↔ X0
xi ↔ Xi
yi ↔ Yi
H ↔ Xn+1
and set g denotes the Lie algebra generated by these vectors with the Lie bracket [·, ·] derived
from the Poisson structure. This Lie algebra is known as a oscillator Lie algebra.
The Lie algebra g splits into a vector space direct sum g = g+ ⊕ g− , where g± denote
the Lie subalgebras
g⊥
− = span{X0 } g⊥
+ = RXn+1 ⊕ span{Xi , Yj }i,j=1,...,n ,
and it also induces linear isomorphisms g∗± ≃ g⊥ ∓ . Let G denote a Lie group with Lie
algebra g and G± ⊂ G is a Lie subgroup whose Lie algebra is g± . Hence the Lie subgroup
G− acts on g⊥ ⊥
+ by the “coadjoint” representation; which in terms of U− ∈ g− and V ∈ g+
is given by P
ad∗U− V = xn+1 (V ) i (yi (U )Xi − xi (U )Yi ) (16)
It is not difficult to see that the orbits are 2n-dimensional if xn+1 (V ) 6= 0 and furthermore
V and W belong to the same orbit if and only if xn+1 (V ) = xn+1 (W ), hence the orbits are
parametrized by the xn+1 -coordinate; so we denote them by Mxn+1 . They are topologically
like R2n . In fact Mxn+1 = G− · V ≃ Hn /Z(Hn ), where Hn denotes the Heisenberg Lie
group with center Z(Hn ).
Equipp these coadjoint orbits with the canonical symplectic structure, that is for
U− , V− ∈ g− take
n
X
ωY (Ũ− , Ṽ− ) = hY, [U− , V− ]i = xn+1 (Y ) (xi (U− )yi (V− ) − xi (V− )yi (U− )).
i=1
dx
dt = [xn+1 Xn+1 , xv + xn+1 Xn+1 ]
(17)
x(0) = x0
P
where x0 = x0v + x0n+1 X0 and x0v = i (x0i Xi + yi0 Yi ).
For xn+1 ≡ x0n+1 ≡ 1 this system is that one we get on R2n .
The trajectories x(t) with coordinates xi (t), yj (t), x0n+1 are parametrized circles of
angular velocity x0n+1 , for all i,j, that is
Lie Theory in Physics 341
Let us consider the linear system of one degree of freedom on R2n with Hamiltonian
given by:
1
H(x) = (Ax, x)
2
where x = (q1 , . . . , qn , p1 , . . . , pn ) is a vector in R2n written in a symplectic basis and A
is a non singular symmetric linear operator with respect to the canonical inner product ( , ).
This yields the following Hamiltonian equation
′ 0 −Id
(3) x = JAx, with J =
Id 0
and being Id the identity. The phase space for this system is R2n . We shall construct
a solvable Lie algebra that admits an ad-invariant metric on which the system (3) can be
realized as a Hamiltonian system on coadjoint orbits. Moreover it can be written as a Lax
pair equation.
Let b denote the non degenerate bilinear form on R2n = span{Xi , Yj }ni,j=1 given by
b(X, Y ) = (AX, Y ). In our terms, b defines a metric on R2n but it is not necessary definite.
Note that the linear JA is non singular and skew symmetric with respect to b, where J is
the canonical complex structure on v ≃ R2n as above:
Let g denote the Lie algebra g which as vector space is the diract sum
g = RX0 ⊕ v ⊕ RXn+1
where v = R2n and with the Lie bracket given by the non trivial relations
Thus in this way one defines a structure of a solvable Lie algebra on g. Note that A = Id
is the particular case we considered in the previous subsection.
This Lie algebra g can be equipped with the ad-invariant metric defined by
hx10 X0 + U 1 + x1n+1 Xn+1 , x20 X0 + U 2 + x2n+1 Xn+1 i = b(U 1 , U 2 ) + (x10 x2n+1 + x20 x1n+1 ).
(20)
Thus if h , iv denotes the restriction of the metric of g to v = span{Xi , Yj }i,j=1,...,n ,
then clearly h , i is a generalization of the non degenerate symmetric bilinear map b of R2n .
Moreover g admits a orthogonal splitting
g = span{X0 , Xn+1 } ⊕ v.
They induce the splitting of g into a vector space direct sum g = g+ ⊕ g− , which by the
ad-invariant metric gives the following linear decomposition g = g⊥ ⊥
+ ⊕ g− , direct sum as
vector spaces, for
g⊥
− = RX0 g⊥
+ = span{Xi , Yi }i=1,...,n ⊕ RXn+1 .
Theorem 5.1 ([O2]). Let H(X) = 21 (AX, X) be a quadratic Hamiltonian on R2n with
corresponding Hamiltonian system (3). Then H can be extended to a quadratic function f
on a solvable Lie algebra g containing the Heisenberg Lie algebra as a proper ideal. The
function f induces a Hamiltonian system on coadjoint orbits of the Heisenberg Lie group,
that can be written as a Lax pair equation and which is equivalent to (3). Moreover the
trajectories on R2n for the initial condition V 0 can be computed with help of the Adjoint
map on g. Explicitely they are the curves x(t) = exptJA V 0 , where exp denotes the usual
exponential map of matrices.
If we take L, M ∈ M (2n + 2, R) as
xn+1 JA 0 z xn+1 JA 0 z
M = 0 0 0 L = i 12 z T 0 0
0 0 0 0 0 0
As we said the phase space is R2n . In the setting of the AKS scheme we can construct
coadjoint orbits M of the Heisenberg Lie group, that are included in a solvable Lie alge-
bra g with Lie bracket (19) and ad-invariant metric (20). The Hamiltonian system for the
restriction to the orbits of the ad-invariant function on g, f (X) = 12 hX, Xi, can be written
as
dx
dt = [xn+1 Xn+1 , xv + xn+1 Xn+1 ]
(24)
x(0) = X 0
P 0
where X 0 = 0 0
i (xi Xi + yi Yi ) + xn+1 Xn+1 . The Hamiltonian system above on the
coadjoint orbit M1 written in coordinates
P is clearly equivalent to (23).
The trajectories on g⊥ + , x = (x
i i (t)X i + yi (t)Yi ) + xn+1 Xn+1 are parametrized by
The system (24) is a Lax pair equation L′ = [M, L] = M L − LM , and has a matricial
representation by choosing L and M the following matrices in M (2n + 2, R):
0 xn+1 0 0 0 0
xn+1 0 0 0 0 0
0 0 0 xn+1 0 0
0 0 xn+1 0 0 0
.. .. ..
M = . . .
0 x n+1 0 0
0 x 0 0 0
n+1
0 0 ... 0 0
0 0 ... 0 0
0 xn+1 0 0 x1
xn+1 0 0 0 y1
0 0 0 xn+1 x2
0 0 xn+1 0 y2
. . .
. .
.
L= . . .
xn+1 0 x n
xn+1 0 0 yn
− 1 y1 1 x1 − 1 y2 1 x2 . . . − 1 yn 1 xn 0 0
2 2 2 2 2 2
0 0 0 0 ... 0 0 0 0
Now we shall investigate involution conditions on the coadjoint orbits of the Heisenberg
Lie group for the restrictions of the quadratic functions f (X) = 12 hX, Xi, where h , i
denotes the ad-invariant metric on the solvable Lie algebra g.
Let gi , gj be two quadratics on R2n that are realted to the symmetric maps Ai , Aj : v →
v respectively, that is
1 1
gi (X) = (Ai X, X) gj (X) = (Aj X, X).
2 2
Consider quadratic functions on the solvable Lie algebra g, which are extensions of gi , gj
to RX0 ⊕ RXn+1 , for instance as
1 1
gi (X) = (Ai Xv, Xv) + x0 xn+1 gj (X) = (Aj Xv, Xv) + x0 xn+1 .
2 2
For the following results these extensions are not unique. For instance extending them
trivially we get the same conclusions.
Let Hi , Hj denote the restrictions of gi , gj to the orbits Mxn+1 and let X ∈ Mxn+1 ⊂
g⊥+ The symplectic structure on the orbits induces a Poisson bracket which for the functions
.
Hi , Hj follows:
{Hi , Hj }(X) = hX, [∇gi − (X), ∇gj − (X)]i
By computing one can see that the gradients of gi and gj are
∇gi (X) = A−1 Ai Xv+x0 X0 +xn+1 Xn+1 ∇gj (X) = A−1 Aj Xv+x0 X0 +xn+1 Xn+1 .
Thus we are ready to prove the following result.
346 Gabriela P. Ovando
Theorem 5.3 ([O2]). The functions Hi , Hj are in involution on the orbits Mxn+1 if and
only if
[JAi , JAj ] = 0 (26)
where J is the canonical complex structure on R2n .
{Hi , Hj }(X) = hX, [Ai Xv, Aj Xv]i = hxn+1 [Xn+1 , A−1 Ai Xv], A−1 Aj Xvi
= xn+1 hJAi Xv, A−1 Aj Xvi = xn+1 (JAi Xv, Aj Xv)
Therefore {Hi , Hj }(X) = 0 if and only if (Aj JAi Xv, Xvi = 0 which is equivalent to
Aj JAi = Ai JAj , if and only if JAj JAi = JAi JAj , that is [JAi , JAj ] = 0.
and note that h , i|v×v = ( , ). A derivation D of hn acting trivially on the center must
satisfy [DU, V ] = −[U, DV ] for all U, V ∈ v. Equivalently in terms of h , i′ , we have that
a map D in hn is a derivation acting trivially on the center of hn if and only if the restriction
of D to v (denoted also D) satisfies
where we replaced h , i′v by ( , ) since they coincide on v ≃ R2n . Denote by d the set of
derivations on hn acting trivially on the center of hn .
Theorem 5.4. There is a bijection between the set of derivations of hn acting trivially on
the center and the set so(n) of symmetric linear maps on R2n . This correspondence is given
by D ∈ d → JD ∈ so(n), where J is the complex structure as in (3).
Proof. The previous theorem says that the restrictions to the orbit Mxn+1 of the functions
gi , gj are in involution if their corresponding derivations commute in d. In particular for gi
and f , we have that H and Hi Poisson commute on the orbit if and only if JAi belongs to
the centralizer of JA in d, z(JA)d. Since the complete integrability requires of n linearly
independent functions, this can be done with a basis of an n-dimensional abelian subalgebra
of z(JA)d, finishing the proof.
Lie Theory in Physics 347
A linear map t is a derivation of hn acting trivially on the center z(hn ) if and only if
Jt + t∗ J = 0, if and only if t ∈ sp(n). The derivations of nilpotent Lie algebras of H-type
were computed in ([Sa]).
In the case of the motion of n-uncoupled harmonic oscillators, we can see that the
corresponding derivation is an element of a Cartan subalgebra of sp(n).
References
[AM] Abraham, R., Marsden, J., Foundations of Mechanics, Second edition. The Ben-
jamin Cummings publishing company, (1985).
[Ad1] Adler, M., A new integrable system and a conjecture by Calogero, (1975).
[Ad2] Adler, M., On a trace for formal pseudodifferential operators and the symplectic
structure for the KdV type equations, Invent. Math., 50, 219-248, (1979).
[CdS] Cannas da Silva, A., Lectures on symplectic geometry, Lecture notes in Math.
1764, Springer Verlag, (2001).
[F] Folland, G., Harmonic analysis in phase space, Annals of Math. Studies, 122,
Princeton University Press, Princeton, NJ, (1989).
[F-T] Fomenko, A., Trofimov, V., Integrable systems in Lie algebras and symmetric
spaces, Gordon and Breach Sc. Publ., (1988).
[G] Guest, M, Harmonic Maps, Loop Groups and Integrable Systems. (London Math.
Soc. Student Texts; 38). New York: Cambridge University Press (1997).
[G-S] Guillemin, V., Sternberg, S., Symplectic techniques in physics. Cambridge New
York Port Chester Melbourne Sydney: Cambridge University Press (1991).
[Ka] Kac V., Infinite-dimensional Lie algebras, Cambridge University Press, Cam-
bridge, 1985.
[K2] Kirillov, A. A., Merits and demerits of the orbit method, bULL. AMS (N.S.), 36
4, 433-488, (1999).
[Ki1] Kisil, V. V., Plain mechanics: classical and quantum, J. Natur. Geom., 9 1, 1-14,
(1996).
[Ko1] Kostant, B., Quantization and Representation Theory, in: Representation Theory
of Lie groups, Proc. SRC/LMS Res. Symp., Oxford 1977. London Math. Soc.
Lecture Notes Series, 34, 287-316, (1979).
[Ko2] Kostant, B., The solution to a generalized Toda lattice and representation theory,
Advances in Math., 39, 195 - 338, (1979).
[L-M] Libermann, P., Marle C.M. , Symplectic Geometry and Analytical Mechanics, D.
Reidel Publishing Company, 1987.
[M-R] Medina, A., Revoy, Ph., Algèbres de Lie et produit scalaire invariant, Ann. scient.
Éc. Norm. Sup., 4e série, t. 18, 391 - 404, (1985).
[O2] Ovando, G., Small oscillations and the Heisenberg Lie algebra, J. Phys. A: Math.
Theor. 40, 2407–2424 (2007).
[P] A. Perelomov, Integrable Systems of Classical Mechanics and Lie Algebras, vol.
I, Birkhäuser Verlag, Basel - Boston - Berlin, (1990).
[Ra] Raghunathan, M., Discrete subgroups of Lie groups, Springer, New York,(1972).
[R1] Ratiu, T., Involution theorems, Geometric methods in Math. Phys., Lect. Notes
in Math., 775, Procedings, Lowell, Massachusetts 1979, Springer Verlag, (1980).
[R2] Ratiu, T., The motion of the free n-dimensional rigid body, Indiana Univ. Math.
Journal, 29, 609 - 629, (1980).
[Sa] Saal, L. The automorphism group of a Lie algebra of Heisenberg type Rend. Sem.
Mat. Univ. Pol. Torino, 54 2, (1996).
[Sy] Symes, W., Systems of Toda type, inverse spectral problems and representation
theory, Invent. Math., 59, 13 - 53, (1978).
[Va] Varadarajan, V., Lie groups, Lie algebras and their representations, Springer,
(1984).
Lie Theory in Physics 349
[Wa] F. Warner, Fundations of differentiable manifolds and Lie groups, Springer Ver-
lag, New York (1983).
Chapter 13
Ming Liao∗†
Department of Mathematics, Auburn University, Auburn,
AL 36849, USA
Abstract
A Lévy process in a Lie group is a process that possesses independent and station-
ary multiplicative increments. The theory of such processes is not merely an extension
of Lévy processes in Euclidean spaces. Because of the unique structures possessed
by non-commutative Lie groups, these processes exhibit certain interesting properties
which are not present for their counterparts in Euclidean spaces.
The concept of Lévy processes may be extended to include Markov processes in a
homogeneous space that are invariant under the group action. More generally, we will
also study processes in Lie groups and homogeneous spaces that possess independent,
but not necessarily stationary, increments, called nonhomogeneous Lévy processes.
These processes appear naturally when studying a decomposition of a general Markov
process in a manifold invariant under a group action.
In these notes, we will provide an introduction to Lévy processes in Lie groups
and homogeneous spaces, and present some selected results in this area. The reader is
referred to the literature for the most of proofs, but some explanation will be given to
the results not yet published.
When two neighborhoods on M are so identified with open subsets of Rd , their intersection
induces a map from (part of) Rd into Rd , which is required to be smooth (C ∞ ).
Thus, a manifold M is locally a Euclidean space. Near any point in M , local coordi-
nates may be introduced. The smoothness of a function on M or a function from M into
another manifold may be defined.
As for the Euclidean space Rd , tangent vectors at a point x in M are just directional
derivatives at x. They together form a vector space, called the tangent space at x and is
denoted by Tx M . A (smooth) vector field is just a smooth assignment of a tangent vector to
each point in M . For any vector field X and smooth function f on M , Xf is also a smooth
function on M .
If F : M → N is a smooth map between two manifolds, then for x ∈ M , it induces
naturally a linear map Df : Tx M → Tf (x) N , called the differential of f at x.
A topological group is a group and a topological space such that both the product map:
G × G ∋ (g, h) 7→ gh ∈ G and the inverse map: G ∋ g 7→ g −1 ∈ G are continuous. A Lie
group is a group and a manifold such that these two maps are smooth. It is well known that
a Lie group is in fact analytic in the sense that the underlying manifold structure together
with the product and inverse maps are analytic.
The Lie algebra g of a Lie group G is the tangent space Te G of G at the identity element
e of G. For g ∈ G, let lg : G ∋ h 7→ gh ∈ G and and rg : G ∋ h 7→ hg ∈ G be
respectively the left and right translations on G. A vector field Y on G is called left invariant
if Dlg (Y ) = Y for any g ∈ G. It is determined by its value at e, X = Y (e) ∈ g, as
Y (g) = Dlg (X), and is denoted by X l . Similarly, we defined a right invariant vector Y
using Drg and denote it by X r for X = Y (e). We may write gX for X l (g) and Xg for
X r (g) more suggestively.
For any X and Y in g, [X l , Y l ] = X l Y l − Y l X l is a left invariant vector field. The Lie
bracket [X, Y ] is the its value at e, that is, [X, Y ]l = [X l , Y l ]. It is linear in X and Y , is
anti-symmetric in the sense that [Y, X] = −[X, Y ], and satisfies the Jacobi identity:
The exponential map exp: g → G is defined by exp(X) = ψ(1), where ψ(t) is the
solution of the ordinary differential equation (d/dt)ψ(t) = ψ(t)X satisfying the initial
condition ψ(0) = e, and is a diffeomorphism of an open neighborhood of 0 in g onto an
open neighborhood of e in G. We may write eX for exp(X).
For g ∈ G, the conjugation map cg : G ∋ h 7→ ghg −1 ∈ G is a Lie group isomorphism
(a product-preserving differeomorphism). Its differential map: g ∋ X 7→ gXg −1 ∈ g, de-
noted by Ad(g) and called the adjoint map, is a Lie algebra isomorphism (a linear bijection
preserving Lie bracket). It can be shown that [X, Y ] = (d/dt)Ad(etX )Y |t=0 for Y ∈ g.
An important example of Lie groups is the matrix group of all n × n real invertible
matrices, denoted by GL(n, R) and called the general linear group on Rn . Its Lie algebra,
denoted by gl(n, R), is the space of all n × n real matrices with Lie bracket [A, B] =
P∞− BA.
AB The exponential map is given by the usual matrix exponentiation eX = I +
k −1 is given by usual matrix product.
k=1 X /k! and the adjoint map Ad(g)X = gXg
The matrix product can mean either the product of group elements or the translation of Lie
algebra element.
Lévy Processes in Lie Groups and Homogeneous Spaces 353
Another useful example of Lie groups is the special orthogonal group SO(n), the group
of n × n orthogonal matrices of determinant 1, also called a rotation group because each
matrix represents a rotation on Rn . Its Lie algebra is the space o(n) of n×n skew-symmetric
matrices (A′ = −A, where ′ denotes matrix transpose).
An action of a Lie group G on a manifold M is a smooth map F : G × M → M
satisfying F (gh, x) = F (g, F (h, x)) and F (e, x) = x for g, h ∈ G and x ∈ M . For
simplicity, we may write gx for F (g, x). The subset Gx = {gx; g ∈ G} of M is called an
orbit of G on M . The action of G on M will be called transitive if any orbit of G is equal
to M .
Let H be a closed subgroup of G. The set of left cosets gH for g ∈ G is denoted by
G/H and is called a homogeneous space of G. It is equipped with the quotient topology.
By [15, II.Theorem 4.2], there is a unique manifold structure on G/H under which the
natural action of G on G/H, defined by g ′ H 7→ gg ′ H, is smooth.
Suppose a Lie group G acts on a manifold M . Fix p ∈ M . Let H = {g ∈ G; gp = p}.
Then H is a closed subgroup of G, called the isotropy subgroup of G at p. By Theorem 3.2
and Proposition 4.3 in [15, Chapter II], if the action of G on M is transitive, then the map:
gH 7→ gp is a diffeomorphism from G/H onto M , therefore, M may be identified with
G/H.
As an example, the rotation group SO(n) acts on the unit sphere S n−1 in Rn transi-
tively. The isotropy subgroup at the “north pole” p = (1, 0, . . . , 0) is diag{1, SO(n−1)} ≡
SO(n−1). Thus, S n−1 may be identified with the homogeneous space SO(n)/SO(n−1).
It is easy to show that a Lévy process gt is a Markov process with transition semigroup
Pt given by Z
Pt f (g) = E[f (ggte )] = f (gh)µt (dh) (1)
G
for f ∈ B(G)+ . By this we mean that gt has the following Markov property:
almost surely for s < t and f ∈ Cb (G) (the space of bounded continuous functions on
G), where the left hand side is the conditional expectation given the σ-algebra Ft generated
by the process up to time t, and for t ≥ 0, Pt is a probability kernel from G to G, that
is, Pt (x, ·) is a probability measure on G for x ∈ G and Pt (x,
R B) is measurable in x for
measurable B ⊂ G such that P0 (x, ·) = δx and Pt+s (x, ·) = Pt (x, dy)Ps (y, ·).
A Markov process in G with semigroup Pt is called a Feller process if for f ∈ C0 (G),
Pt f ∈ C0 (G) and Pt f → f uniformly on G as t → 0, where C0 (G) is the space of contin-
uous functions on G vanishing at infinity (under the one-point compactification topology).
The distribution of a Feller process is completely determined by its generator L defined by
Lf = limt→0 (1/t)Pt f with domain D(L) consisting of f ∈ C0 (G) for which the limit ex-
ists uniformly on G. A Feller process is called a diffusion process if its generator restricted
to Cc∞ (G) is a second order differential operator (with no constant term). Such a process is
necessarily continuous.
By (1), a Lévy process is a Feller process and its semigroup Pt is left invariant in the
sense that Pt (f ◦lg ) = (Pt f )◦lg for any g ∈ G. A Markov process in G with a left invariant
semigroup is called a left invariant Markov process.
The Lévy process defined above may be called a left Lévy process. Similarly, one may
define a right Lévy process using the increment gt gs−1 instead of gs−1 gt . Then it is a Feller
process in G with transition semigroup Pt f (g) = E[f (gte g)], where gte = gt g0−1 , and is
invariant under the right translation rg . The left and right Lévy processes are in natural
duality. In the following, a Lévy process will mean a left Lévy process unless explicitly
stated otherwise.
The results of this section hold in fact for Lévy processes in a locally compact and
second countable Hausdorff topological group G.
Theorem 3 Let gt be a Lévy process in a Lie group G. Then the domain D(L) of its
generator L contains C02,l (G) and for f ∈ C02,l (G),
d
1 X
Lf (g) = aij Xil Xjl f (g) + X0l f (g)
2
i,j=1
Z X
+ [f (gh) − f (g) − xi (h)Xil f (g)]Π(dh), (3)
i
where aij are some constants forming a non-negative definite symmetric matrix, X0 ∈ G
and Π is a measure on G satisfying
Z
Π({e}) = 0, |x|2 dΠ < ∞ and Π(U c ) < ∞ (4)
U
for any neighborhood U of e. Conversely, given aij , X0 and Π as above, there is a Lévy
process in G, unique in distribution, whose generator restricted to C02,l (G) is given by (3).
Any measure Π on G satisfying (4) is called a Lévy measure. In the case of a Lévy
process gt , Π is the characteristic measure of the Poisson random measure N on R+ × G
that counts the jumps of gt , that is,
−1
N ([0, t] × B) = #{s ∈ (0, t]; gs− gs ∈ B} (5)
and Π(B) = E[N ((0, 10 × B)]. Hence, gt is continuous if and only if Π = 0. It is clear
that Π is independent of the choice of the basis of g and the associated coordinate functions.
The differential operator
d
X
LD = (1/2) aij Xil Xjl (6)
i,j=1
is called the diffusion part of the generator L, also independent of the basis of g and asso-
ciated coordinate functions. However, the drift vector X0 ∈ g in (3) may depend on this
choice.
It is shown in Applebaum-Kunita [2] (see also [24]) that a Lévy process gt in G can
be characterized by a stochastic integral equation driven by a Brownian motion Bt and the
Poisson random measure N , corresponding to the Lévy -Itô representation in Euclidean
case. Let Ñ be the compensated random measure of N defined by Ñ (dt dg) = N (dt dg) −
dtΠ(dg).
356 Ming Liao
Theorem 4 There is a d-dim Brownian motion Bt with covariance matrix {aij } (E(Bti ) =
0 and E(Bti Btj ) = aij t) and independent of N such that for f ∈ C02,l (G),
d Z
X t Z t
f (gt ) = f (g0 ) + Xil f (gs− ) ◦ dBsi + X0l f (gs )ds
i=1 0 0
Z tZ
+ [f (gs− h) − f (gs− )]Ñ (ds dh)
0 G
Z tZ d
X
+ [f (gs h) − f (gs ) − xi (h)Xil f (gs )]dsΠ(dh). (7)
0 G i=1
and
d Z
X t Z t
f (gt ) = f (g0 ) + Xil f (gs− ) ◦ dBsi + Y0l f (gs )ds
i=1 0 0
Z tZ
+ [f (gs− h) − f (gs− )]N (dsdh), (9)
0 G
P R
where Y0 = X0 − di=1 ( xi dΠ)Xi .
If Π is finite, then there are iid (independent and identically distributed) stopping times
τn of exponential distribution of mean 1/Π(G), and an independent sequence of iid G-
valued random variables σn of distribution Π/Π(G), such that the Lévy process gt may be
obtained by solving the stochastic differential equation
d
X
dgt = Xil (gt ) ◦ dBti + Y0l (gt )dt (10)
i=1
P
together with jump conditions g(Tn ) = g(Tn −)σn , where Tn = ni=1 τi .
All these results hold also for a right Lévy process with suitable changes, for example,
C0 (G), X l and gh in (3) should be replaced by C02,r (G), X r and hg.
2,l
Lévy Processes in Lie Groups and Homogeneous Spaces 357
Because a general Lie group G does not have a natural linear structure, the stochastic
integral equation for the Lévy process gt can only be given in a functional form as in (7) and
(9). If G is a matrix group, then it is possible to write down a stochastic integral equation
directly for the process gt . Now let G = GL(d, R) be the general linear group of the d × d
real invertible matrices. Its Lie algebra g = gl(d, R) is the space of all d × d real matrices.
2
We may identify g = gl(d, R) with the Euclidean space Rd and G = PGL(d, R) with a
2 2
dense open subset of R . For any X = {Xij } ∈ R , let |X| = ( i,j Xij )1/2 be its
d d 2
Euclidean norm.
Let Eij be the matrix that has 1 at place (i, j) and 0 elsewhere. Then Eij , i, j =
1, 2, . . . , d, form a basis of g. A set x = {xij } of associated coordinate functions may be
chosen such that x(g) = g − Id in matrix form for g close to e = Id (the d × d identity
matrix).
2
For g ∈ G = GL(d, R), the tangent space Tg G can be identified with Rd , therefore,
any element X of Tg G can be represented by a d × d real matrix {Xij } in the sense that
P
Xf = di,j=1 Xij (∂/∂gij )f (g), where gij are the standard coordinates on Rd . It can be
2
shown that for g, h ∈ GL(d, R) and X ∈ Te G = gl(d, R), Dlg ◦ Drh (X) is represented
by the matrix product gXh, where X is identified with its matrix representation {Xij }.
Therefore, we may write gXh for Dlg ◦ Drh (X). Thus, X l (g) = gX and X r (g) = Xg.
Let gt be a Lévy process in G = GL(d, R). Then it satisfies the stochastic integral
equation (7) for any f ∈ C02,l (G). Let f be the matrix-valued function on G defined by
f (g) = g for g ∈ G. Although f is not contained in C02,l (G), at least formally, (7) leads to
the following stochastic integral equation in matrix form:
d Z
X t Z t
gt = g0 + gs− Eij ◦ dBsij + gs X0 ds
i,j=1 0 0
Z tZ Z tZ
+ gs− (h − Id )Ñ (ds dh) + gs [h − Id − x(h)]dsΠ(dh), (11)
0 G 0 G
Theorem 5 Assume
Z
2
E[|g0 | ] < ∞ and |h − Id |2 Π(dh) < ∞. (12)
G
Then there is a unique rcll process gt in G = GL(d, R) that satisfies the equation (11).
Moreover, gt is a Lévy process and for any t > 0,
Conversely, any Lévy process gt in G satisfying (12) is the unique solution of a stochastic
integral equation of the form (11).
358 Ming Liao
generate the Lie algebra G. The hypoelliptic condition is weaker than the non-degeneracy.
A continuous Lévy process satisfying this condition is a hypoelliptic diffusion process in
the usual sense. It is well known that such a process possesses a smooth distribution density
pt = dµt /dρG for t > 0. For a general Lévy process gt , it is shown in [23] that if it has a
non-degenerate diffusion part and also a finite Lévy measure, then it has an L2 distribution
density pt for t > 0. We will see that the existence of pt holds also under hypoellipticity if
the Lévy process gt possesses additional invariance property.
Lévy Processes in Lie Groups and Homogeneous Spaces 359
The series converges absolutely and uniformly for (t, g) in [ε, ∞)×G for any ε > 0, hence,
(t, g) 7→ pt (g) is continuous on (0, ∞) × G. Moreover, if gt has a hypoelliptic diffusion
part, then the eigenvalues of the matrix L(U δ ∗ )(e) have negative real parts. Consequently,
pt → 1 uniformly on G as t → ∞.
are conjugate invariant if and only if X and aij are Ad(G)-invariant. The following result,
which holds on a general Lie group, may be derived from the generator formula (3).
Theorem 8 A Lévy process gt with g0 = e has a conjugate invariant distribution µt for all
t > 0 if and only if in the generator formula (3) under exponential coordinates chosen as
above, Π is conjugate invariant, and aij and X0 are Ad(G)-invariant.
Let L2ci (ρG ) be the space of conjugate invariant functions in L2 (ρG ). The character
χδ of an irreducible unitary representation U δ , defined by χδ = Trace(U δ ), is conjugate
invariant. A version of Peter-Weyl theorem says that the irreducible characters χδ form an
orthonormal basis of L2ci (ρG ). Thus, for f ∈ L2ci (ρG ),
X
f = ρG (f ) + ρG (f χδ )χδ (16)
δ∈ Irr(G)+
for any finite set of gi ∈ G and ξi ∈ C. From this it is easy to show that |χδ | ≤ χδ (e) = dδ .
Let ψδ = χδ /dδ (normalized character).
Let kf k2 and kf k∞ be respectively the usual L2 -norm and the L∞ -norm of a function
f (under ρG ). The total variation norm of a signed measure ν is defined by kνktv =
sup|f |≤1 |ν(f )|. By Theorem 7, if gt is hypoelliptic, then its distribution µt converges to
the normalized Haar measure ρG under the total variation norm: kµt − ρG ktv → 0 as
t → ∞.
360 Ming Liao
Theorem 9 Let gt be a Lévy process with g0 = e, a hypoelliptic diffusion part, and conju-
gate invariant distribution µt . Then it has an L2 distribution density pt for t > 0 and for
g ∈ G, X
pt (g) = 1 + dδ et Lψδ (e) χδ (g). (18)
δ∈ Irr(G)+
The series converges absolutely and uniformly for (t, g) in [ε, ∞) × G for any ε > 0, and
d
X Z
Lψ δ (e) = (1/2) aij Xil Xjl ψδ (e) + (Re ψδ − 1)dΠ, δ ∈ Irr(G)+ , (19)
i,j=1
has a negative upper bound −λ, and for any ε > 0, there are constants C > c > 0 such
that for t > ε,
kpt −1k∞ ≤ Ce−λt , ce−λt ≤ kpt −1k2 ≤ Ce−λt , ce−λt ≤ kµt −ρG ktv ≤ Ce−λt . (20)
Remark In Theorem 9, the hypoelliptic condition on the diffusion part may be replaced by
an asymptotic condition on the Lévy measure Π as to be described below.
Two nonnegative functions φ and ψ are called asymptotically equal at a point x0 , de-
noted as φ ≍ ψ at x0 , if there are positive constants c1 < c2 such that c1 φ ≤ ψ ≤ c2 φ in
a neighborhood of x0 . Similarly, two measures ξ and η are called asymptotically equal at
x0 if c1 ξ ≤ η ≤ c2 ξ in a neighborhood of x0 . This is equivalent to the asymptotic equality
of their densities if those exist. We will say ξ to be asymptotically larger than η at x0 if ξ
dominates a measure that is asymptotically equal to η at x0 .
Let the compact Lie group G be equipped with a left invariant Riemannian metric and
let r(g) be the distance between e and g ∈ G. It can be shown that if gt is a Lévy process in
G with g0 = e and a conjugate invariant distribution µt for all t > 0, and if its Lévy measure
Π is asymptotically larger than rβ dρG for some β ∈ (d, d + 2), where d = dim(G), then
all the conclusions of Theorem 9 hold. This can be proved by the arguments in [27, section
7].
Example: The rotation group G = SO(3) is a simple Lie group. Any matrix in SO(3) is
conjugate to a unique rotation about x1 -axis by angle θ ∈ [0, π]. Therefore, any conjugate
invariant function on G may be regarded as a function of θ ∈ [0, π]. By [5, section II.5],
Lévy Processes in Lie Groups and Homogeneous Spaces 361
P2n i(n−j)θ
the irreducible characters (including the trivial one) are given by χn (θ) = j=0 e =
sin[(n + 1/2)θ]/ sin(θ/2) of dimension 2n + 1 for n = 0, 1, 2, . . ..
By choosing a bi-invariant Riemannian metric on G and an orthonormal basis
X1 , X2 , X3 of g = Te G, we may write ∆ = X1l X1l + X2l X2l + X3l X3l . We may as-
sume X1l = ∂/∂θ. By the conjugate invariance of characters, ∆χn (e) = 3χ′′n (e). A direct
computation yields ∆ψn (e) = ∆χn (e)/(2n + 1) = −n(n + 1).
Let gt be a continuous Lévy process in G = SO(3) with g0 = e and generator L. If it
is hypoelliptic and conjugate invariant, then L = a∆ for some constant a > 0, and hence it
has an L2 distribution density pt for t > 0 given by
∞
X sin[(n + 1/2)θ]
pt (θ) = (2n + 1)e−an(n+1)t . (21)
n=0
sin(θ/2)
Because a section map S may be chosen to be continuous near any point in G/K, it is
easy to show from Theorem 10 that a Lévy process in G/K is a Feller process.
We have mentioned Hunt’s formula for the generator of a Lévy process in G. We will
now state an analog of this result for Lévy process in G/K, obtained also by Hunt [18], in
the form presented in [24, Theorem 2.1].
A differential operator T on G/K is called G-invariant if T (f ◦ g) = (T f ) ◦ g for
f ∈ C ∞ (G/K) and g ∈ G. Such an operator is completely determined by T f (o) for
f ∈ C ∞ (G/K). It will be called a G-invariant diffusion generator if it is the generator of
a G-invariant diffusion process in G/K.
Because K is compact, there is an Ad(K)-invariant subspace p of the Lie algebra g
of G that is complementary to the Lie algebra k of K. Let X1 , . . . , Xd be a basis of
g such that X1 , . . . , Xn ∈ p and Xn+1 , . . . , Xd ∈ Pk. Local coordinates y1 , . . . , yn on
n
G/K
Pn around o may be chosen
Pn to satisfy x = exp[ i=1 yi (x)Xi ]o for x near o. Then
−1
i=1 yi (x)Ad(k)Xi = i=1 yi (kxk )Xi for all k ∈ K and for x near o. As functions
in Cc∞ (G/K), yi may be suitably extended such that this holds for all x ∈ G/K.
Theorem 11 Let L be the generator of a Lévy process xt in G/K with x0 = o. Then its
domain D(L) contains Cc∞ (M ), and with local coordinates y1 , y2 , . . . , yn on G/K near o
chosen as above, for any f ∈ Cc∞ (G/K),
Z n
X ∂
Lf (o) = T f (o) + [f (x) − f (o) − yi (x) f (o)]Π(dx), (22)
G/K ∂yi
i=1
where T is a G-invariant diffusion generator on G/K, called the diffusion part of L, and
Π is a K-invariant measure on G/K satisfying
Xn
Π({o}) = 0, Π( yi2 ) < ∞ and Π(U c ) < ∞
i=1
for any neighborhood U of o, called the Lévy measure of process xt . Conversely, given T
and Π as above, there is a Lévy process xt in G/K with x0 = o, unique in distribution,
such that its generator L at o, restricted to Cc∞ (G/K), is given by (22).
Theorem 14 Let xt be a Lévy process in G/K starting from o with distribution νt and
generator L given by (22). For any bounded spherical function φ on G/K,
Z Xn
tLφ(o) ∂
νt (φ) = e = exp{t[β(T, φ) + (φ(x) − 1 − yi (x) φ(o))Π(dx)]}, (25)
G/K ∂yi
i=1
7. Symmetric Spaces
As before, let G be a Lie group and let K be a compact subgroup. Their Lie algebras are
denoted respectively as g and k. Because K is compact, there is an Ad(K)-invariant inner
product h·, ·i on g. Fix such an inner product and let p be the orthogonal complement of k
in g under this inner product.
A bijective map Θ: G → G that preserves the product structure, and satisfies Θ 6=
idG and Θ2 = idG is called a Cartan involution. The pair (G, K) together with a Cartan
involution Θ is called a symmetric pair if GΘ Θ Θ
0 ⊂ K ⊂ G , where G is the fixed point set
of Θ and GΘ 0 is its identity component. Then the differential DΘ of Θ at e has eigenvalues
±1 with k and p as eigenspaces of +1 and −1 respectively. The direct sum g = k⊕p is called
a Cartan decomposition of g. The homogeneous space G/K is called a symmetric space
which plays an important role in differential geometry, see [15, 16] for more information.
We now assume G/K is a symmetric space. By Theorem 3.1 and Proposition 3.8 in [16,
chapter IV], the convolution product on G/K is commutative, that is, µ∗ν = ν ∗µ, and any
K-invariant finite measure µ on G/K is completely determined by its spherical transform
φ 7→ µ(φ) with φ ranging over all bounded spherical functions. Note that the two results
from [16] cited here are stated for functions on G, but from which the above statements for
measures may be derived. Therefore, the Lévy-Khinchin formula (25) provides a complete
characterization of a continuous convolution semigroup νt on a symmetric space G/K.
The Killing form of the Lie algebra g is defined by B(X, Y ) = Trace[ad(X)ad(Y )]
(a symmetric bilinear form), where ad(X) is the linear map: g → g given by ad(X)Y =
[X, Y ] (Lie bracket). The Lie group G is semisimple if and only if B is nondegenerate.
Lévy Processes in Lie Groups and Homogeneous Spaces 365
The symmetric space G/K is said to be of compact type if the Killing form B is neg-
ative definite, it is said to be of noncompact type if B is negative definite on k and positive
definite on p, and it is said to be of Euclidean type if p is an abelian ideal of g. For a com-
pact (resp. noncompact) type symmetric space G/K, G is semisimple, and both G/K and
G are compact (resp. noncompact). In general, a symmetric space G/K with G simply
connected is a direct product of symmetric spaces of these three basic types.
A Euclidean space Rn is a trivial example of a Euclidean type symmetric space with
G = Rn (additive group) and K = {0}. A less trivial example is Rn = Mn /SO(n),
where Mn is the group of Euclidean motions on Rn and SO(n) is the subgroup of rotations
that fixes the origin. The n-dimensional sphere S n = SO(n + 1)/SO(n) is a compact
type symmetric space. The n-dimensional hyperbolic space Hn = SO(1, n)+ /SO(n) is a
noncompact type space, where SO(1, n)+ is the connected Lorentz group on Rn+1 .
The positive definiteness of a function in a Lie group G is defined by (17). A function
φ on G/K is called positive definite if φ ◦ π is so on G. Such a function is bounded with
|φ| ≤ φ(o). The following result is proved in [27].
Moreover, β(T, φ) ≤ 0.
The Lévy -Khinchin type formula (26) on symmetric spaces was obtained in Gangolli
[13]. With t = 1, it in fact characterizes all infinite divisible K-invariant distributions ν1
on a symmetric space G/K (ν1 = ν1/n ∗n for some K-invariant ν
1/n and all integer n > 0).
The spherical transform of a finite measure µ: φ 7→ µ(φ) with φ varying over positive
definite spherical functions, plays the same role as the Fourier transform on a Euclidean
space. On each of three basic types of symmetric spaces G/K, using (26) and the inverse
spherical transform, we can prove that the distribution νt of a Lévy process xt in G/K has a
density qt (x), smooth in (t, x) ∈ (0, ∞) × (G/K), with respect to a G-invariant reference
measure ρ on G/K for t > 0 under either of the following two conditions:
(a) the diffusion part T of the generator L is non-degenerate, or
(b) the Lévy measure Π is asymptotically larger (see definition in section 4) than rβ dρ
for some β ∈ (n, n + 2), where n = dim(G/K) and r is the distance from o,
and obtain a representation of qt in terms of spherical functions, see [27] for more details.
the distribution of xt . Then νt (φn ) = e−an(n+1)t . If a > 0, then νt has a smooth density qt
with respect to the uniform distribution on S 2 given by
∞
X
qt (r) = 1 + (2n + 1)e−an(n+1)t Pn (cos r). (27)
n=1
Let
called respectively the centralizer and the normalizer of a in K. The former is a normal
subgroup of the latter, but has the same Lie algebra m. The quotient group W = M ′ /M is
finite and is called the Weyl group. It acts on a via the map: W × a ∋ (s, H) 7→ sH =
Ad(ks )H ∈ a with s = ks M ∈ W for ks ∈ M ′ .
The group G possesses the Cardan decomposition G = KA+ K in the sense that any
g ∈ G may be written as g = ka+ h with k, h ∈ K and a unique a+ ∈ A+ . Although
the choices for (k, h) are not unique, when a+ ∈ A+ , they are given by (km, m−1 h) for
m ∈ M . An element g of G is called regular if a+ ∈ A+ . Regular elements form an open
dense subset G′ of G, which projects to an open dense subset (G/K)′ of G/K. The Cartan
decomposition G = KA+ K induces the polar decomposition on G/K: Any point x in
G/K can be written as x = ka+ o for kM ∈ K/M and a+ ∈ A+ . When x ∈ (G/K)′ , a+
and kM are unique and are called respectively the radial and angular components of x.
The group G also has the Iwasawa decomposition G = N − AK in the sense that the
map: N − × A × K ∋ (n, a, k) 7→ g = nak ∈ G is a diffeomorphism. There are other
versions of Iwasawa decompositions such as G = KAN + .
A typical example of a semi-simple Lie group of noncompact type is G = SL(d, R),
the group of d × d real matrices of determinant one, called the special linear group. Its Lie
algebra is g = sl(d, R), the space of d × d traceless real matrices. The Cartan involution
is given by Θ(g) = g ′−1 with K = SO(n) and p being the space of d × d traceless real
symmetric matrices.
In this case, the subspace a of p formed by traceless diagonal matrices is a max-
imal abelian subspace of p. The roots are αij given by αij (H) = Hi − Hj for
H = diag(H1 , . . . , Hd ) ∈ a and i 6= j. The root space gαij is one-dimensional and
is spanned by the matrix Eij that has 1 at place (i, j) and 0 elsewhere. One may take
a+ = {diag(H1 , . . . , Hd ) ∈ a; H1 > H2 > · · · > Hd } to be the chosen Weyl chamber.
Then positive roots are αij with i < j. The nilpotent Lie algebras n+ and n− are respec-
tively the spaces of upper triangular and lower triangular matrices of zero diagonal. The
groups A, N + and N − are respectively the subgroups of G consisting of diagonal matrices
of positive diagonal, upper triangular matrices of unit diagonal, and lower triangular matri-
ces of unit diagonal. The group M is discrete with Lie algebra m = {0} and consists of
diagonal matrices with ±1 along diagonal (with an even number of −1’s), the group M ′ is
formed by permutation matrices, and the Weyl group W = M ′ /M acts on a by permuting
the diagonal elements of H ∈ a.
For G = SL(d, R), the Cartan decomposition G = KA+ K can be established by
diagonalizing the symmetric matrix gg ′ for g ∈ SL(d, R). The Iwasawa decomposition
G = N − AK (resp. G = KAN + ) is a consequence of Gram-Schmidt orthogonalization
applied to the rows (resp. columns) of a matrix in SL(d, R).
Now return to a general G. Let gt be a Lévy process in G and let µt be the distribution
of gte = g0−1 gt . Let Tµ be the closed semigroup generated by the supports supp(µt ) of
measures µt for t ∈ R+ , that is, Tµ is the closed subset of G containing all supp(µt ) and
satisfying x, y ∈ Tµ =⇒ xy ∈ Tµ . Let Gµ be the closed subgroup of G generated by all
supp(µt ).
368 Ming Liao
Following Guivarc’h and Raugi [14], a subset H of G will be called totally irreducible
if there do not exist g1 , . . . , gk , x ∈ G such that
k
[
H⊂ gi (N − M AN + )c x,
i=1
where the superscript c denotes the complement in G. It will be called totally right irre-
ducible if gi (N − M AN + )c x is replaced by x(N − M AN + )c gi . Note that N − M AN + is
an open subset of G whose complement is lower dimensional, that is, the complement is
contained in a union of finitely many lower dimensional sub-manifolds of G. Therefore, if
H is not a lower dimensional subset of G, then it is totally (right) irreducible.
Note that exp: a → A is a bijection, so its inverse log: A → a is well defined.
A sequence gj in G is called contracting if in the Cartan decomposition gj = ξj a+ j ηj ,
+
α(log aj ) → ∞ as j → ∞ for any positive root α. For g ∈ G with Cartan decomposition
g = ka+ h, let kgk = k log a+ k.
The convergence of (1/t) log a+ t and ξt U in the above theorem are often referred to as
the radial and angular convergences. See section 6.6 in [24] for some sufficient conditions
which guarantee the hypotheses of Theorem 16.
Theorem 16 holds also for a right Lévy process gt in G with the following changes:
the total irreducibility should be replaced by the total right irreducibility, the convergence
of ξt M in K/M by that of M ηt in M \K (right coset space), and the decomposition gt =
nt at kt of G = N − AK by gt = kt at nt of G = KAN + .
9. Dynamical Aspect
The limiting properties of Lévy processes may also be studied from a dynamic point of
view.
Let Ω be the underlying probability space. A collection of maps θt : Ω → Ω, t ∈ R+ ,
is called a semigroup of time-shift operators if each θt preserves the probability measure P
on Ω, and they together form a semigroup in t in the sense that θt θs = θs+t and θ0 = idΩ .
Let X be a (smooth) manifold and let Diff(X) be the group of the diffeomorphisms
X → X. A dynamical system, or a stochastic flow, on X is a stochastic process φt in
Diff(X) with φ0 = idX together with a semigroup of time shift operators θt such that the
following co-cycle property holds:
See Arnold [3] for a comprehensive treatment of the dynamical theory of such systems.
We now describe the Lyapunov exponents and the associated stable manifolds of a
stochastic flow φt on a manifold X equipped with a Riemannian metric {k · kx ; x ∈ X}.
The family of maps Φt : X × Ω → X × Ω given by (x, ω) 7→ (φt (ω)x, θt ω), t ∈ R+ , is a
semigroup in t and is called the skew-product flow associated to φt . A probability measure
ν on X is called a stationary measure of φt if E(φt ν) = ν. Under a certain condition, there
is a unique stationary measure ν. Moreover, there are constants λ1 > λ2 > · · · > λr , a
subset Γ of X × Ω invariant under Φt , in the sense that Φ−1
t (Γ) = Γ, with ν × P (Γ) = 1,
and for any (x, ω) ∈ Γ, the subspaces of the tangent space Tx X:
such that
1
∀v ∈ [Vi (x, ω) − Vi+1 (x, ω)], lim log kDφt (ω)vkφ(ω)x = λi
t→∞ t
for i = 1, 2, . . . , r, where Vi (x, ω) − Vi+1 (x, ω) is the set difference. The numbers λi are
called the Lyapunov exponents. The random subspace Vi (x, ω) is called the subspace of
Tx X associated to the exponent λi , and di = dim[Vi (x, ω) − Vi+1 (x, ω)] is independent of
(x, ω) and is called multiplicity of λi . The Lyapunov exponents λi are the limiting expo-
nential rates at which the lengths of tangent vectors on X are stretched or contracted under
the stochastic flow φt , and together with Vi (x, ω), they are independent of the Riemannian
metric on a compact manifold X.
A connected sub-manifold X′ of X is called a stable manifold of a negative Lyapunov
exponent λi at (x, ω) ∈ Γ if X′ ⊂ {y ∈ X; (y, ω) ∈ Γ}, Tx X′ = Vi (x, ω) and
1
∀y ∈ X′ , lim sup log dist(φt (ω)x, φt (ω)y) ≤ λi ,
t→∞ t
where dist denotes the Riemannian distance on X. Roughly speaking, the distance between
any two points in X′ tends to zero exponentially fast at the negative exponential rate λi
under the stochastic flow φt . A stable manifold of λi at (x, ω) is called maximal if it
contains any stable manifold of λi at (x, ω). The local existence of stable manifolds is
proved in Carverhill [6]. Intuitively, one would expect that the maximal stable manifolds of
a negative exponent form a foliation of a random open dense subset of X, but such a global
theory under a general setting can be quite complicated, see [3, chapter 7].
If a Lie group G acts on a manifold X, then a right Lévy process gt in G with g0 = e
may be regarded as a stochastic flow on X when Ω is taken to be the canonical sample space
of the process with natually defined time shift θt .
Let G be a semi-simple Lie group of noncompact type with a finite center. We will
continue to use the notation introduced in the previous section. Let Q be a closed subgroup
of G with Lie algebra q. Assume Q contains AN + . Then the homogeneous space X =
G/Q is compact. For G = SL(d, R), such homogeneous spaces include the sphere S d−1 ,
the special orthogonal group SO(d) and several other interesting spaces.
Let gt be a right Lévy process in G with g0 = e. We will describe explicitly, in terms
of the group structure, the Lyapunov exponents and the associated stable manifolds of gt
370 Ming Liao
Theorem 19 Let λi < 0. Then the family of stable manifolds Xi (x, ω) of λi is a foliation
of Γ(ω). Moreover, if i < r, then each Xi (x, ω) is foliated by {Xi+1 (y, ω); y ∈ Xi (x, ω)},
the family of the stable manifolds of the exponent λi+1 contained in Xi (x, ω).
Theorem 20 The set of stationary points of (H + )∗ on X is π(M ′ ) and the set of attracting
stationary points is π(M ).
Proof We provide an outline of the proof, see [9] for the details. Fix T > 0. Let 0 = t0 <
t1 < · · · < tn ≤ T be a partition of [0, T ]Rwith ti+1 − ti = 1/n and let µni = µti−1 ,ti for
1 ≤ i ≤ n. Define bni ∈ G by φi (bni ) = φi dµni . By the stochastic continuity of xt , bni
is uniformly small (close to e) in i as n → ∞.
For each n > 1, define a G-valued function by bn (t) = bn1 bn2 · · · bn[nt] for 0 < t ≤ T
with bn (0) = e, where [nt] is the integer part of nt, a measure function Πn by Πn (t, ·) =
P[nt]
i=1 µni for 0 < t ≤ T with Π(0, ·) = 0, and a matrix valued function An (t, U ) of time t
Lévy Processes in Lie Groups and Homogeneous Spaces 373
[nt] Z
X
[An (t, U )]ij = [φi (x) − φi (bnk )][φj (x) − φj (bnk )]µnk (dx).
k=1 U
Note that if the Lie group G is commutative such as G = Rn , then yt = zt and the
martingale representation (35) takes the simpler form (36). In this case, bt is the continuous
drift, Aij (t) is the covariance of the continuous Gaussian process and Π is the characteristic
measure of the Poisson process of jumps, that is, for any t ≥ 0 and B ⊂ G, Π(t, B) is the
expected number of jumps contained in B by time t. On a general Lie group G, bt may be
regarded as a drift and Aij (t) as governing the continuous part of the Lévy process xt , and
Π is still the characteristic measure of the Poisson random measure that counts the jumps
of the process. In particular, xt is continuous if and only if Π = 0.
of G containing H as a normal subgroup, and hence has a natural group structure with
product b1 b2 = S(b1 )b2 and inverse b−1 = S(b)−1 o (independent of S). In general, the
product x1 x2 · · · xn−1 xn = S(x1 )S(x2 ) · · · S(xn−1 )xn is not well defined because it de-
pends on the choice of S. However, if x1 , x2 , . . . , xn are independent random variables and
x2 , . . . , xn have H-invariant distributions, then the distribution of the product x1 x2 · · · xn
and also that of the sequence yi = x1 x2 · · · xi for i = 1, 2, . . . are independent of S, and
hence such a product or sequence is meaningful in the sense of distribution.
Note that for H-invariant finite measures µ and ν on G/H, an integral like
Z Z
f (xy, xyz)µ(dy)ν(dz) = f (S(x)y, S(x)S(y)z)µ(dy)ν(dz)
R
is well defined (independent of choice of section map S). So is f (xbyb−1 )µ(dy) if b is
an H-invariant point in G/H.
A process xt in G/H with rcll paths will be called a nonhomogeneous Lévy process if
there is a continuous two-parameter convolution semigroup µs,t of H-invariant probability
measures on G/H such that (34) holds. Then for s < t, x−1 −1
s xt = S(xs ) xt has distri-
bution µs,t (independent of choice for section map S) and is independent of the process up
to time s. Thus, a nonhomogeneous Lévy process in G/H may be characterized as a rcll
process with independent increments.
Because H is compact, there is a subspace p of g that is complementary to the Lie
algebra h of H and is Ad(H)-invariant in the sense that Ad(h)p = p for h ∈ H. Choose
∞
Ppmand local coordinates φ1 , . . . , φm ∈ Cc (G/H) around o on G/H
a basis ξ1 , . . . , ξm of
such that x = exp( i=1 φi (x)ξi )o for x near o. Then by (2.16) in [24],
m
X m
X
∀h ∈ H, φi Ad(h)ξi = (φi ◦ h)ξi (37)
i=1 i=1
near o. The functions φi may be suitably extended so that (37) holds globally on G/H.
Any ξ ∈ g is a left invariant vector field on G. If ξ is Ad(H)-invariant, it may also be
d
regarded as a vector field on G/H given by ξf (x) = dt f (xetξ o) |t=0 for f ∈ C ∞ (G/H)
tξ
and x ∈ G/H (note that e o is H-invariant), which is G-invariant in the sense that ξ(f ◦
g) = (ξf ) ◦ g for g ∈ G. In fact, any G-invariant vector field on G/H is given by a unique
Ad(H)-invariant ξ ∈ p. Note that if ξ ∈ g is Ad(H)-invariant and b ∈ G/H is H-invariant,
then Ad(b)ξ = Ad(S(b))ξ is Ad(H)-invariant R and is Pindependent of section map S. By
(37), Rfor any H-invariant
P measure µ on G/H,
R µ(dx)
P i φi (x)ξi is Ad(H)-invariant, and
so is µ(dx) i φi (x)Ad(b)ξi = Ad(b) µ(dx) i φi (x)ξi .
Let ξ, η ∈ g. With a choice of section map S, ξη may be regarded as a second order
2
differential operator on G/H by setting ξηf (x) = ∂t∂ ∂s f (S(x)etξ esη o) |t=s=0 . As in [9],
2 P ij
it can be shown that ξi ξj f (x) = ∂t∂∂s f (S(x)etξi +sξj o) |t=s=0 + m k=1 ρk ξk f (x) with
ρij ji ∞
k = −ρk . Thus, if aij is a symmetric matrix, then for f ∈ C (G/H),
m
X m
X ∂2 Pm
aij ξi ξj f (x) = aij f (S(x)e p=1 tp ξp o) |t1 =···=tm =0 . (38)
∂ti ∂tj
i,j=1 i,j=1
P
Pjq for h ∈ H,
The matrix aij is called Ad(H)-invariant if aij = p,q apq [Ad(h)]ip [Ad(h)]
where [Ad(h)]ij is the matrix representing Ad(h), that is, Ad(h)ξj = i [Ad(h)]ij ξi .
Lévy Processes in Lie Groups and Homogeneous Spaces 375
P
Then the operator i,j aij ξi ξj is independent of section map S and is G-invariant. In fact,
any second order G-invariant differential operator on G/H without constant term is such
P operator plus a G-invariantP
an vector field. Note that if b ∈ G/H is H-invariant, then
i,j aij [Ad(b)ξi ][Ad(b)ξj ] = i,j aij [Ad(S(b))ξi ][Ad(S(b))ξj ] is a G-invariant operator
on G/H (independent of S).
A covariance function A and a Lévy measure function Π on G/H are defined as on G
with the additional requirements that A(t) is Ad(H)-invariant and Π(t, ·) is H-invariant.
By the preceding discussion, the expression in (35) is meaningful on G/H and is indepen-
dent of the choice of section map S in Ad(bs ) = Ad(S(bs )). The following result is an
extension of Feinsilver’s martingale representation to nonhomogeneous Lévy processes in
G/H.
Proof Proceed as in the proof of Theorem 21, but note that µni is H-invariant and hence
bni is H-invariant. The proof in [9] on G can be suitable modified to work on G/H, such
as properly interpreting the product on G/H as discussed here and using H-invariant sets
on G/H for various neighborhoods used in [9].
Remark 1 (Lévy measure): The Lévy measure Π is associated to the jumps of the Lévy
process xt in G/H as in G, and hence xt is continuous if and only if Π = 0.
Remark 2 (drift): Because the natural action of H on G/H fixes o, it induces an action
on the tangent space To (G/H) at o. If there is no nonzero H-invariant tangent vector in
To (G/H), then in a neighborhood of o, there is no H-invariant point except o. In this case,
the drift bt in Theorem 22 must be trivial, that is, bt = o for all t ≥ 0. Note that the exis-
tence of nonzero H-invariant vector in To (G/H) is equivalent to the existence of nonzero
Ad(H)-invariant element in p. For example, on the sphere S n−1 = SO(n)/SO(n − 1),
there is no nonzero Ad(H)-invariant vector in p, and hence a stochastic continuous Lévy
process in S n−1 can only have trivial drift.
This means that for k ∈ K, kxt is the same Markov process starting at kx0 .
Let Y be a submanifold of X, possibly with a boundary, that is transversal to the action
of K in the sense that it intersects each orbit of K at exactly one point, that is,
The process yt = J(xt ) in Theorem 23 will be called the radial part of process xt
(relative to K and Y ). Note that for a diffusion process xt with generator L, the generator
LY of yt is the radial part of the differential operator L as defined in [16].
For x ∈ X, let Kx = {k ∈ K; kx = x} be the isotropy subgroup of K at x. Let Y ◦ be
Y minus its boundary. We will assume that Ky is the same compact subgroup M of K as
y varies over Y ◦ . This assumption is often satisfied when the transversal submanifold Y is
properly chosen.
We will now strengthen the transversality condition (40) by assuming that
where Ty X is the tangent space of X at y, see Lemma 3.3 in [16, Chapter II]. Then the
union of the K-orbits through Y ◦ , denoted by X ◦ , is an open dense subset of X, and
X ◦ = Y ◦ × (K/M ) as a product manifold.
All our assumptions are satisfied in the following examples.
Example 1: We have mentioned earlier that the radial part of a Brownian motion xt in
X = Rn (n ≥ 2), under the action of K = SO(n), is a Bessel process in a fixed ray Y
from the origin. We may take Y to be the positive half of x1 -axis, which is transversal to
K with boundary containing only the origin. Then M = diag{1, SO(n − 1)}.
Example 2: Let X be the space of n×n real symmetric matrices (n ≥ 2) with K = SO(n)
acting on X by conjugation. The set Y of all n × n diagonal matrices with non-ascending
diagonal elements is a submanifold of X transversal to SO(n), its boundary consists of
diagonal matrices with at least two identical diagonal elements, and M is the finite subgroup
of SO(n) consisting of diagonal matrices with ±1 along diagonal. The map J: X → Y
maps a symmetric matrix to the diagonal matrix of its eigenvalues in non-ascending order.
Note that X = GL(n, R)/SO(n).
Example 5: Let X = G/K be a symmetric space of noncompact type. Using the notation
in Section 8 with a+ as the fixed Weyl chamber with closure a+ and boundary ∂a+ . Y =
exp(a+ ) is a submanifold of X transversal to the action of K on G/K with boundary
exp(∂a+ ), and the isotropy subgroup M of K at any y ∈ Y ◦ = exp(a+ ) is the centralizer
M of a in K.
The exit time ζ of process xt from X ◦ is the stopping time when xt together with its
left limit first leaves X ◦ or reaches its life time ξ. More precisely, it is defined by
ζ = inf{t > 0; xt 6∈ X ◦ , xt− 6∈ X ◦ or t ≥ ξ}, (43)
with inf of an empty set defined to be ∞. The exit time of yt from Y ◦ is also denoted by ζ.
Fix T > 0. Because the process xt has rcll paths, it may be regarded as a random variable
in the space DT (X) of rcll maps: [0, T ] → X, equipped with Skorohod topology. Let Px
be the distribution on DT (X) associated to the process xt starting at x ∈ X. Its total mass
may be less than 1 because xt may have a finite life time.
For y ∈ Y ◦ and z ∈ K/M , zy = S(z)y ∈ X ◦ is well defined and is independent of
choice of section map S: K/M → K. Let xt = zt yt be the decomposition of the process
xt with x0 ∈ X ◦ and t < ζ. Then yt is the radial part as defined before, and zt is a process
in K/M with rcll paths and will be called the angular part of xt .
Recall J is the projection map X ∋ x 7→ y ∈ Y . Let J2 be the projection map
X ◦ ∋ x 7→ z ∈ K/M associated to the decomposition x = zy. We will also use J and J2
to denote the maps J: DT (X) ∋ x(·) 7→ y(·) ∈ DT (Y ) and J2 : DT (X ◦ ) ∋ x(·) 7→ z(·) ∈
DT (K/M ) respectively given by the decomposition x(·) = z(·)y(·).
Y
Let F0,T = σ{yt ; 0 ≤ t ≤ T } be the σ-algebra generated by the radial process yt
for 0 ≤ t ≤ T , which may be regarded as a σ-algebra on DT (Y ) and induces the σ-
algebra J −1 (F0,T
Y ) on D (X). By the existence of regular conditional distributions (see
T
y(·)
for example [19, chapter 5]), there is a probability kernel Rz from DT (Y ◦ ) × (K/M ) to
DT (K/M ) such that for any x ∈ X ◦ and measurable F ⊂ DT (K/M ),
J[x(·)]
RJ2 (x) (F ) = Px [J2−1 (F ) | J −1 (F0,T
Y
)] for Px -almost all x(·) in [ζ > T ] ⊂ DT (X ◦ ).
(44)
y(·)
The probability measure Rz is the conditional distribution of the angular process zt given
a radial path y(·) in DT (Y ◦ ) and z0 = z.
Theorem 24 Fix T > 0. Almost surely on [ζ > T ], given a radial path yt for 0 ≤ t ≤ T ,
the conditioned angular process zt is a nonhomogeneous Lévy process in K/M . More
precisely, this means that for y ∈ Y ◦ , z ∈ K/M , and JPy -almost all y(·) in [ζ > T ] ⊂
y(·)
DT (Y ◦ ), the angular process zt is a nonhomogeneous Lévy process under Rz .
Lévy Processes in Lie Groups and Homogeneous Spaces 379
Proof For x ∈ X ◦ , let P̃t (x, B) = Px {[xt ∈ B]∩[ζ > t]} for measurable B ⊂ X ◦ . By the
Markov property of xt , it is easy to show that P̃t is the transition semigroup of the Markov
process xt for t < ζ and it is K-invariant. Similarly, let Q̃t be the transition semigroup
of yt for t < ζ. Then P̃t (x, ·) and Q̃t (y, ·) are respectively sub-probability kernels from
X ◦ to X ◦ = Y ◦ × (K/M ) and from Y ◦ to Y ◦ . By the existence of a regular conditional
distribution, there is a probability kernel Rt (y, y1 , ·) from (Y ◦ )2 to K/M such that for
y ∈ Y ◦ , P̃t (y, dy1 × dz1 ) = Q̃t (y, dy1 )Rt (y, y1 , dz1 ). The K-invariance of P̃t implies
that the measure Rt (y, y1 , ·) is M -invariant for Q̃t (y, ·)-almost all y1 . Modifying Rt on
an exceptional set of zero Q̃t (y, ·)-measure, we may assume Rt (y, y1 , ·) is M -invariant for
all y, y1 ∈ Y ◦ . Therefore, for z ∈ K/M , it is meaningful to write Rt (y, y1 , z −1 dz1 ) =
Rt (y, y1 , S(z)−1 dz1 ) because it is independent of choice of section map S. We then have
∀y ∈ Y ◦ and z ∈ K/M, P̃t (zy, dy1 × dz1 ) = Q̃t (y, dy1 )Rt (y, y1 , z −1 dz1 ). (45)
It follows that for 0 < s1 < s2 < · · · < sk < ∞, y ∈ Y ◦ , z ∈ K/M and f ∈
Cb ((K/M )k ),
Z
Ezy [f (zs1 , . . . , zsk ) | ys1 , . . . , ysk ] = Rs1 (y, ys1 , dz1 )Rs2 −s1 (ys1 , ys2 , dz2 ) · · ·
Rsk −sk−1 (ysk−1 , ysk , dzk )f (zz1 , zz1 z2 , . . . , zz1 · · · zk )](46)
on [ζ > sk ].
Let Γ be the set of dyadic numbers i/2m for integers i ≥ 0 and m > 0. May assume
T ∈ Γ. For s, t ∈ Γ with s < t ≤ T , let s = s1 < s2 < · · · < sk be a partition of [0, T ]
spaced by 1/2m with s = si and t = sj , and let
µm
s,t = Rsi+1 −si (ysi , ysi+1 , ·) ∗ Rsi+2 −si+1 (ysi+1 , ysi+2 , ·) ∗ · · · ∗ Rsj −sj−1 (ysj−1 , ysj , ·).
µm −1 −1
s,t (f ) = Ezy [f (zs zt ) | ys1 , . . . , ysk ] = Ezy [f (zs zt ) | ysi , . . . , ysj ] on [ζ > T ] (47)
for f ∈ Cb (K/M ), which is independent of the choice for section map S to represent
zs−1 zt = S(zs )−1 zt . By the right continuity of yt , as m → ∞, σ{ys1 , . . . , ysk } ↑ F0,T
Y and
Y , it follows that as m → ∞, almost surely, µm → µ
σ{ysi , . . . , ysj } ↑ Fs,t s,t s,t weakly for
some M -invariant probability measure µs,t on K/M such that
∀f ∈ Cb (K/M ), µs,t (f ) = Ezy [f (zs−1 zt ) | F0,T
Y
] = Ezy [f (zs−1 zt ) | Fs,t
Y
] on [ζ > T ].
(48)
Y
Note that µs,t is an Fs,t -measurable random measure independent of starting point zy.
Because Γ is countable, the exceptional set of probability zero in the above almost sure
convergence may be chosen simultaneously for all s < t in Γ. Moreover, for t1 < t2 <
· · · < tn of [0, T ] in Γ, it can be shown from (46) and by choosing a partition s1 < s2 <
· · · < sk of [0, T ] from Γ containing all ti , spaced by 1/2m , that almost surely on [ζ > T ],
for f ∈ Cb ((K/M )n ),
Y
Ezy [f (zt1 , . . . , ztn ) | F0,T ] = lim Ezy [f (zt1 , . . . , ztn ) | ys1 , . . . , ysk ]
m→∞
Z
= lim f (zz1 , zz1 z2 , . . . , zz1 · · · zn )µm m m
0,t1 (dz1 )µt1 ,t2 (dz2 ) · · · µtn−1 ,tn (dzn ).
m→∞
380 Ming Liao
Remark It is clear that given a constant radial path yt ≡ y, the angular process zt becomes
a (homogeneous) Lévy process in K/M under the conditional distribution. One may obtain
a Lévy process in K/M by forcing the process xt to run in the orbit Ky. More precisely,
starting at y ∈ Y ◦ , run the process xt for a small time ε > 0 and project paths to the orbit
Ky via J2 (using the natural identification of Ky with K/M ), then at the end of each pro-
jected path, run xt again for ε and project to Ky. Repeat the procedure to obtain a process
ztε in Ky ≡ K/M . It can be shown that as ε → 0, the process ztε converges to a Lévy
process zt in K/M in the sense of finite dimensional distribution, see [26] for more details.
It is interesting to compare this “forced” process with the conditioned angular process given
a constant radial path y. It turns our that they are not always equal in distribution.
Because the natural action of M on K/M fixes o, it induces an action on the tangent
space To (K/M ) at o. The homogeneous space K/M will be called irreducible if the action
of M on To (K/M ) is irreducible (that is, it has no nontrivial invariant subspace). In this
case, there is no nonzero M -invariant tangent vector in To (K/M ), and hence a nonhomo-
geneous Lévy process in K/M has only trivial drift, see Remark 2 in Section 11. Among
the examples mentioned earlier, K/M is irreducible in Examples 1 and 4, and in Example 3
if it is chosen to be so, and in Example 5 if the symmetric space G/K is of rank 1 (see [16]).
If K/M is irreducible, then, up to a constant multiple, there is a unique M -invariant
inner product on To (K/M ) (see for example Appendix 5 in [20]). By choosing a K-
invariant Riemannian metric on K/M , which is unique up to a constant factor, any second
order K-invariant differential operator on K/M is a multiple of the Laplace operator ∆K/M
on K/M . The following result is an extension of Galmarino’s result mentioned earlier.
the two processes xt and B(at )yt , t < ζ, are identical in distribution.
Proof By Theorem 24, given a radial path yt for 0 ≤ t ≤ T with [ζ > T ], the conditioned
angular process zt is a continuous nonhomogeneous Lévy process in K/M . Let (b, A, Π)
PBy the irreducibility of K/M , the drift bt = o. Because
be its representation in Theorem 22.
zt is continuous, Π = 0. Because i,j Aij (t)ξi ξj is a K-invariant second order differential
operator on K/M , it must be equal to at ∆K/M for a continuous non-decreasing function
Y -measurable. Then from the construction of A(t) from
at with a0 = 0. By (48), µs,t is Fs,t
Lévy Processes in Lie Groups and Homogeneous Spaces 381
Y -measurable. Let B(t) be a
µs,t in the proof of Theorem 22, it is seen that at − as is Fs,t
Brownian motion in K/M independent of process xt , and hence independent of process at .
It is enough to show that the conditioned process zt is equal to the
R t time-changed Brownian
1
motion B(at ) in distribution. For f ∈ Cb (K/M ), f (B(t)) − 0 2 ∆K/M f (B(s))ds is a
Rt
martingale, and hence, f (B(at )) − 0 21 ∆K/M f (B(as ))das is a martingale (here at is non-
random
Rt 1 given a radial path). On the other hand, for the conditioned process zt , f (zt ) −
0 2 K/M f (zs )das is a martingale. The uniqueness of the process in distribution with the
∆
given representation (b, A, Π) implies that zt = B(at ) in distribution.
References
[1] Applebaum, D., “Compound Poisson processes and Lévy processes in groups and
symmetric spaces”, J. Theo. Probab. 13, 383-425 (2000).
[2] Applebaum, D. and Kunita, H. (1993) “Lévy flows on manifolds and Lévy processes
on Lie groups”, J. Math. Kyoto Univ. 33, 1105-1125.
[4] Babillot, M. (1991) “Comportement asymptotique due mouvement Brownien sur une
variété homogène à courbure négative ou nulle”, Ann. Inst. H. Poincaré (prob et Stat)
27, 61-90.
[5] Bröcker, T. and Dieck, T. (1985) “Representations of compact Lie groups”, Springer-
Verlag.
[6] Carverhill, A.P. (1985) “Flows of stochastic dynamical systems: ergodic theory”,
Stochastics 14, 273-317.
[7] Diaconis, P. (1988) “Group representations in probability and statistics”, IMS, Hay-
ward, CA.
[9] Feinsilver, P., “Processes with independent increments on a Lie group”, Trans. Amer.
Math. Soc. 242, 73-121 (1978).
[10] Furstenberg, H. (1963) “Noncommuting random products”, Trans. Am. Math. Soc.
108, 377-428.
[11] Furstenberg, H. and Kesten, H. (1960) “Products of random matrices”, Ann. Math.
Statist. 31, 457-469.
[15] Helgason, S. (1978) “Differential geometry, Lie groups, and symmetric spaces”, Aca-
demic Press.
[16] Helgason, S. (2000) “Groups and geometric analysis”, Amer. Math. Society.
[18] Hunt, G.A. (1956) “Semigroups of measures on Lie groups”, Trans. Am. Math. Soc.
81, pp 264-293.
[20] Kobayashi, S. and Nomizu, K. (1963) “Foundations of differential geometry, vol I”,
Interscience Publishers.
[21] Liao, M. (1994) “The Brownian motion and the canonical stochastic flow on a sym-
metric space”, Trans. Amer. Math. Soc. 341, 253-274.
[22] Liao, M. (1998) “Lévy processes in semi-simple Lie groups and stability of stochastic
flows”, Trans. Amer. Math. Soc. 350, 501-522.
[23] Liao, M. (2004) “Lévy processes and Fourier analysis on compact Lie groups”, Ann.
Probab. 32, 1553-1573.
[24] Liao, M. (2004) “Lévy processes in Lie group”, Cambridge Univ. Press.
[25] Liao, M. (2008) “A decomposition of Markov processes via group actions”, to appear
in J. Theo. Probab.
[26] Liao, M. (2008) “Markov processes invariant under a Lie group action”, to appear in
Stoch. Processes and their Appl.
[27] Liao, M. and Wang, L. (2007) “Levy-Khinchin formula and existence of densities for
convolution semigroups on symmetric spaces”, Potential Analysis 27, 133-150.
[28] Malliavin, M.P. and Malliavin, P. (1974) “Factorizations et lois limites de la diffusion
horizontale audessus d’un espace Riemannien symmetrique”, Lecture Notes Math.
404, 164-271.
[29] Norris, J.R., Rogers, L.C.G. and Williams, D. (1986) “Brownian motion of ellipsoids”,
Trans. Am. Math. Soc. 294, 757-765.
[30] Orihara, A. (1970) “On random ellipsoids”, J. Fac. Sci. Univ. Tokyo Sect. IA Math. 17,
73-85.
Lévy Processes in Lie Groups and Homogeneous Spaces 383
[31] Raugi, A. (1997) “Fonctions harmoniques et théorèmes limites pour les marches
gléatoires sur les groupes”, Bull. Soc. Math. France, mémoire 54.
[32] Rosenthal, J.S. (1994) “Random rotations: characters and random walks on SO(N)”,
Ann. of Probab. 22, pp 398-423.
[33] Siebert, E. (1981) “Fourier analysis and limit theorems for convolution semigroups on
a locally compact groups”, Adv. in Math. 39, 111-154.
[34] Taylor, J.C. (1988) “The Iwasawa decomposition and the limiting behavior of Brow-
nian motion on a symmetric space of non-compact type”, in Geometry of random
motion, ed. by R. Durrett and M.A. Pinsky, Contemp. Math. 73, Am. Math. Soc.,
303-332.
[35] Taylor, J.C. (1991) “Brownian motion on a symmetric space of non-compact type:
asymptotic behavior in polar coordinates”, Can. J. Math. 43, 1065-1085.
[36] Tutubalin, V.N. (1965) “On limit theorems for a product of random matrices”,Theory
Probab. Appl. 10, 25-27.
[37] Virtser, A.D. (1970) “Central limit theorem for semi-simple Lie groups”, Theory
Probab. Appl. 15, 667-687.
In: Lie Groups: New Research ISBN 978-1-60692-389-4
Editor: Altos B. Canterra, pp. 385-399
c 2009 Nova Science Publishers, Inc.
Chapter 14
Giampaolo Cicogna∗
Dipartimento di Fisica “E.Fermi” dell’Università di Pisa
and Istituto Nazionale di Fisica Nucleare, Sez. di Pisa
Largo B. Pontecorvo 3, Ed. B-C, I-56127, Pisa, Italy
Abstract
The symmetry classification of differential equations containing arbitrary functions
can be a source of several interesting results. We study two particular but significant
examples: a nonlinear ODE and a linear PDE (the 1-dimensional Schrödinger equa-
tion). We provide first of all a necessary, but very restrictive, simple condition involv-
ing the arbitrary functions in order that the given equation does admit Lie symmetries.
In the first example, we show that some symmetry appears only if a precise numerical
relation between the involved parameters is satisfied. In the case of Schrödinger equa-
tion, we see that only for a very limited class of potential functions some symmetry
is admitted, and that the Lie generators of these symmetries are precisely recursion
operators and are related to the Dirac step up - step down operators, well known in
Quantum Mechanics. In connection with all these symmetries, we also discuss the
important problem of the reduction of the differential equations, in both the different
contexts of ODE’s and of PDE’s.
1. Introduction
The symmetry analysis of differential equations containing arbitrary functions, i.e. the
problem of discovering how the symmetry properties of the given equation depend on the
choice of these functions, and thus classifying all possible cases, is in general a not easy
task [9, 17, 21, 26, 34, 35, 44]. It can be also a source of several interesting, sometimes
surprising, results. A striking and well known example is provided by the nonlinear Laplace
∗
E-mail address: cicogna@df.unipi.it
386 G. Cicogna
equation ∇2 u = g(u), with u = u(x, y) (or also the equation uxx − uyy = g(u)), where
one finds an infinite dimensional algebra of Lie point-symmetries if g(u) = exp(±u) (the
Liouville equation), and only relatively trivial symmetries (essentially, scaling symmetries),
or no symmetry at all, for any other choice of g(u) [11, 13, 18, 23, 37].
In the following, we shall study two particular but significant examples: a nonlinear
ODE, which is related to several problems of mathematical and physical interest, and re-
spectively a linear PDE, the 1-dimensional Schrödinger equation. We shall provide first
of all a necessary, but very restrictive, simple condition involving the arbitrary functions
in order that the given equation does admit Lie symmetries. In our first example, where
arbitrary functions of both the independent and the dependent variables are considered, we
shall show that some symmetry appears only if some precise numerical relations between
the involved parameters are satisfied. In the case of Schrödinger equation, we shall see that
only for a very limited class of potential functions some symmetry is admitted, and that the
Lie generators of these symmetries are precisely recursion operators [29] and are related to
the Dirac step up - step down operators [12], well known in Quantum Mechanics.
In connection with all these symmetries, we shall also discuss the important problem of
the reduction of the differential equations, in both the (substantially) different contexts of
ODE’s and of PDE’s.
It should be stressed that we will not consider here the possible presence of discrete
symmetries of our equations: although they may “interact” with Lie symmetries and pro-
duce interesting consequences [16,19], their determination requires in general some specific
procedures. Therefore, for the sake of definiteness, we prefer to restrict here our attention
only to continuous (Lie) symmetries.
and look for the Lie point-symmetries which are admitted by this equation depending on
the choice of the two functions f = f (x) and g = g(u).
It can be worth pointing out that more general equations of the form
a
urr + ur + f (r)g(u) = 0 , u = u(r), a = const
r
can be transformed into the above equation (1) simply putting x = |r|1−a if a 6= 1, and
x = log |r| if a = 1 (notice that usually in these equations, as suggested by the notation, r
is a radial variable, r ≥ 0); as an example, we quote the classical Bratu equation [6]:
1
urr + ur + rk exp u = 0 , k = const .
r
Similarly, the ODE’s which are obtained when looking for radial solutions u = u(r) to
several PDE’s are of the above form; this is the case for instance of the Grad-Schlüter-
Symmetry Analysis of Differential Equations 387
Considering then our equation (1), and denoting the generators of the Lie point-
symmetries in the usual form (see e.g. [3, 5, 15, 20, 29, 32, 39])
∂ ∂
X = ξ(x, u) + ϕ(x, u)
∂x ∂u
one obtains, first of all, from the determining equations that the coefficient functions ξ, ϕ
must be of the form
1
ξ = ξ(x) , ϕ = A(x) + u B(x) , with B = (ξx + b), b = const (2)
2
with clear notations, whereas there is only one equation which involves the two functions f
and g, which is
with
P ≡ (Axx , Bxx , −Bf + 2ξx f + ξfx , A f, B f )
(4)
G ≡ (1, u, g, gu , u gu ) .
λ i Gi = 0 (6)
1
With a little and commonly accepted abuse of language, we will denote by X both the symmetry and its
Lie generator.
388 G. Cicogna
exist between the functions Gi defined in (4). The functions Pi which determine the admitted
symmetries, span the subspace orthogonal in R5 to the subspace spanned by the Gi .
This means that equation (1) can admit symmetries, apart from the trivial case (5) (see
also below), only if g(u) satisfies the very simple linear first-order ODE’s given in (6).
Notice that g(u) may satisfy one or more equations of this form: e.g., if g = exp(u) or if
g = u2 , then respectively one or two conditions as in (6) are satisfied.
A condition of linear dependence similar to (6) holds of course also for the functions
Pi (x), but this would be less convenient to handle because it involves more than one “un-
known” variable (i.e. ξ and f ). Actually, it is simpler to start from (6), and impose the
orthogonality condition (3) to each given solution g(u) to (6); observing that Pi depend
only on x and Gi only on u (see also [9]), one then obtains directly the Pi , and therefore all
possible symmetries of our equation (1).
Notice also that not all choices for the constants λi in (6) are to be considered: e.g.,
λ3 = λ4 = λ5 = 0 is not admitted; similarly, the case λ1 = λ2 = λ3 = 0 would imply g =
const, the choice λ4 = λ5 = 0 corresponds to the (nearly trivial) case where g is a linear
function of u.
Before enumerating the list of functions f and g with the corresponding admitted sym-
metries, let us recall the related and extremely relevant property which involves the re-
ducibility of the given ODE to an equation of lower order (cf. e.g. [22]):
i) if a given second order ODE admits just one symmetry X, then it can be reduced to
a locally equivalent first order ODE introducing “symmetry-adapted” variables y and w,
where y is X-invariant, and w is the coordinate “along the flow” of X:
Xy = 0 , Xw = 1; (7)
these are to be chosen respectively as new independent variable and dependent one: w =
w(y);
ii) if the ODE admits two symmetries X1 , X2 such that (this will be precisely our case)
[X1 , X2 ] = X1 (8)
X1 y = 0 , X1 w = 1 , X2 y = y , X2 w = w (9)
1.1) Let us start with the simplest and before mentioned case, where f = 1/x2 , with
arbitrary g(u). Keeping fixed f = 1/x2 , this would correspond to symmetries of the kernel
group of the full group, see [9, 32]. Thanks to the symmetry X = x∂/∂x, introducing the
variables y = u, w = log(|x|), equation (1) is transformed into the equation (of the first
order in z(y) := wy , as expected)
No other symmetry is present in this case for generic g(u), apart from the following excep-
tion.
1.2) Let as before f = 1/x2 ; another symmetry arises if g(u) has precisely the special
form
2(s + 1)
g(u) = us + u
(s + 3)2
where s is any constant (s 6= −3). This new symmetry is generated by
∂ m+1 ∂ 1−s
X = xm+1 + xu , m= .
∂x 2 ∂u 3+s
For instance, with m = 2, let us write the two admitted symmetries as
∂ 3 ∂ 1 ∂
X1 = x3 + x2 u , X2 = − x
∂x 2 ∂u 2 ∂x
in such a way that the commutation rule holds in the form (8). Performing the change of
variables as indicated by (9), one has y = u4/3 x−2 , w = y − 1/(2x2 ) and the equation is
reduced to the form
12ywyy = 16(wy − 1)3 + 3(wy − 1)
in agreement with (10). The special case s = −1, i.e. g(u) = 1/u, may be of interest: the
new variables are y = u/x, w = (u − 1)/x and the reduced equation is
2.1) Let now f = 1/xr with r 6= 2. Our equation admits one symmetry if g = us , for
any s; its generator is (for s 6= 1: we do not consider the linear case, whose symmetries are
standard)
∂ r−2 ∂
X = x + u .
∂x s − 1 ∂u
2.2) With f = 1/xr , r 6= 2, a second symmetry appears if the two quantities r and s
are related by the condition s = r − 3. For instance, with r = 6, s = 3 we have the two
generators satisfying (8)
∂ ∂ ∂ ∂
X1 = x2 + xu , X2 = −x − 2u
∂x ∂u ∂x ∂u
and the equation uxx + u3 /x6 = 0 becomes, introducing symmetry-adapted variables ac-
cording to (9),
ywyy = 3 − 5wy − wy3 + 3wy2 .
390 G. Cicogna
3.1) Another interesting case occurs with g(u) = exp(s u). A single symmetry given
respectively by
∂ r+2 ∂ 1 ∂ 2(r − 1) ∂
X = x − or X = x+ −
∂x s ∂u r ∂x rs ∂u
is admitted with the particular choices
f (x) = exp(r x)
∂u 1 ∂2u
i = − + V (x) u , u = u(x, t) (11)
∂t 2 ∂x2
and we want to look for the appearance of (nontrivial) Lie point–symmetries
∂ ∂ ∂
X = ξ(x, t, u) + τ (x, t, u) + ϕ(x, t, u) (12)
∂x ∂t ∂u
Symmetry Analysis of Differential Equations 391
depending on the choice of the function V (x) (see [27] and references therein for more
general Schrödinger-type equations, possibly with time-dependent potentials). Standard
calculations show that, for any V (x), the coefficients ξ, τ, ϕ in (12) must be of the form
i x2
(a + bx)Vx + 2bV − bt + ct + x att + btt = 0 . (15)
2 2
Exactly with the same arguments as in the case of ODE’s (Sect. 2), it is easy to conclude
that Schrödinger equation (11) can admit some symmetry only if V (x) verifies the very
restrictive linear equation
k 2 γ
V (x) = x + 2 (k, γ = const) . (17)
2 2x
The case k ≤ 0 is of quite limited interest from the physical point of view, and will not
be considered here. It is then not restrictive to put k = 1, and we find that the admitted
symmetries are
∂ ∂ 1 ∂
X+ = exp(−2it) x + i + (x2 − )u
∂x ∂t 2 ∂u (18)
∂ ∂ 1 ∂
X− = exp(2it) x − i − (x2 + )u
∂x ∂t 2 ∂u
and, in addition if γ = 0 (which corresponds to the case of the quantum harmonic oscillator,
see also [27])
∂ ∂ ∂ ∂
Y+ = exp(−it) + xu , Y− = exp(it) − xu . (19)
∂x ∂u ∂x ∂u
We just mention for completeness the trivial symmetries, i.e. the time translation and the
symmetries following from the linearity of the equation. In particular, the term A(x, t)
appearing in (13) turns out to be any solution to equation (11) and says the obvious fact that
if u(x, t) is a solution to the equation, so is u + A. It is therefore not restrictive to choose
A = 0. We also disregard the transformations of the equivalence symmetry group [32],
which amount in this case to the quite obvious transformations V (x) → V (x)+const. and
x → x+const.
392 G. Cicogna
It is more convenient from now on to introduce and use the notion of evolutionary
operator, related to any vector field X and essentially equivalent to it [29], and which will
be denoted by XQ :
∂
XQ := Q where Q := −ξux − τ ut + ϕ . (20)
∂u
We are considering in this Section only the case of a linear PDE, then it is known [3,29]
that its symmetries are such that Q depends linearly on u and its derivatives; we can then
introduce a “linear” differential operator Q (i.e. an operator not depending on u and its
derivatives) defined by (cf. also (13), with A = 0 as said before)
Q = −ξ Dx − τ Dt + B (21)
Q(u) ≡ Q . (22)
An important property of this notion (and the corresponding notation as well) is that it
can be extended naturally to include generalized symmetries (see e.g. [29]).
Let us now recall the two following fundamental results [3, 29].
Proposition 2. Let ∆ = 0 be a linear PDE and XQ one of its symmetries (possibly gener-
alized); then whenever u0 is a solution to ∆ = 0, so is
u1 := Q(u0 )
where Q is the operator defined in (21). As a consequence, the same is true for un :=
Qn (u0 ), for any n = 1, 2, . . ..
Proposition 3. With the same assumptions as before, one has that Q is a “recursion oper-
ator”: i.e., given any symmetry XQ0 = Q0 ∂/∂u, then also
∂
XQ1 := Q1 where Q1 := Q(Q0 (u))
∂u
is a symmetry for the PDE. It follows, in particular, that also Q1 (u0 ) = Q(Q0 (u0 )) solves
the PDE, and that
∂ ∂
Q2 (u) , . . . , Qn (u) , . . .
∂u ∂u
are (generalized) symmetries for the PDE ∆ = 0.
Let us now introduce the operators Q± related to the vector fields X± given in (18),
where Q± (u) = Q± and XQ± = Q± ∂/∂u, as in (20–22).
In order to obtain a solution to our Schrödinger equation (11) with the potential (17) and
with γ 6= 0, we start looking for the invariant solution u0 = u0 (x, t) under the symmetry
generator X− . This solution is obtained from the corresponding equation Q− (u0 ) = 0,
which becomes 1
xu0,x + i u0,t + x2 + u0 = 0
2
Symmetry Analysis of Differential Equations 393
and substituting into the initial equation (11); in this way the equation is transformed into a
reduced ODE, which is easily solved to get
En = α + 1/2 + 2n .
Q+ Q− (un ) = c(+)
n un , Q− Q+ (un ) = c(−)
n un
(±)
where cn = −En2 ± 2En − 3/4 + γ, and, e.g., Q2− (u1 ) = 0, and so on.
394 G. Cicogna
The above property, which is shared also by the other symmetry operators (19), gener-
alizes the notion of step-up/step-down Dirac operators [12] well known in Quantum Me-
chanics.
Few words, to conclude, for the case γ = 0 of the quantum harmonic oscillator.
Writing the two symmetry generators Y± given in (19) in evolutionary form (we use
here the notation R instead of Q to avoid confusion)
∂
YR± = R±
∂u
and introducing as before the corresponding linear operators
R± := exp(∓it)(−Dx ± x) , R± (u) = R±
holds for any k-th prolongation X (k) of X. From this, one can easily deduce that, if y
is an invariant quantity under X, i.e. X y = 0, and β a first order differential invariant,
X (1) β = 0, then
Dx β dβ
≡ (25)
Dx y dy
is a second order invariant under X (2) , and so on; thus, all higher order differential in-
variants can be obtained in this way. Then, using these invariants as new variables, the
order of the ODE is lowered. For instance, let us consider the vector field X1 which is a
symmetry generator for the equation examined under the item 2.2) in Sect. 2. Putting
y = u/x, β = u − x ux , it is easy to verify that dβ/dy = x3 uxx /(u − x ux ) is indeed
(2)
invariant under X1 , and – more interestingly – that our ODE uxx + u3 /x6 = 0 becomes
a first order equation with the manifestly invariant form
dβ
β + y3 = 0
dy
(with the additional equation β = u − x ux ). The same happens if one considers the second
vector field X2 for the same case 2.2) as above: now, with y = u/x2 , β = ux /x, the
equation uxx + u3 /x6 = 0 becomes in this case
dβ
(β − 2y) + β + y 3 = 0 .
dy
Clearly, in Sect. 2 we preferred to choose the new variables according to the rule (9), in
order to obtain the more convenient standard (and integrable by quadratures) form (10) of
the equation; it may be interesting, however, to compare the different forms assumed by the
same equation when written in the different coordinates.
In the presence of p > 1 independent variables xi , i = 1, . . . , p, as in the case of a
PDE, writing the vector field (sum over repeated indices)
∂ ∂
X = ξi · +ϕ
∂xi ∂u
we find the rule, instead of (24),
which does not allow, in general, to reach the same conclusions about the higher order
differential invariants as in the case of ODE’s. But an even more severe restriction comes
from the fact that it is not granted (in contrast to the single variable case) that it is possible
to replace the initial variables u and xi with some new invariant variables β and yi in such
a way that the higher order differentials uxi xj (and so on) can be expressed as functions of
the differentiated invariants βyi yj (and so on) [30]. An example will clearly illustrate the
396 G. Cicogna
situation. Let us consider the case of p = 2 independent variables x1 , x2 , and the vector
field
∂ ∂
X = x1 +u . (27)
∂x1 ∂u
The invariants of order ≤ 1 of X are
u u x2
, x2 , ux1 , .
x1 u
Choose as new variables e.g. y1 = u/x1 , y2 = x2 , β = ux1 (we stress that any other choice
would lead to analogous results). Thanks to the particular form of the functions ξi ≡ (x1 , 0)
in the above vector field X, the rule (26) takes a simpler form which does allow in this case
to conclude that Dx1 β/Dx1 y1 is still invariant under X (2) , and the same for all quantities
of this type. Nevertheless, it is impossible to express the second order differentials uxi xj in
terms of the new variables. Notice also that, instead of (25), one has here
Dx 1 β ∂β ∂β Dx1 y2
= +
D x 1 y1 ∂y1 ∂y2 Dx1 y1
and so on; in the present case one obtains
5. Conclusion
We started our study from one of the most interesting applications of the theory of Lie sym-
metries, namely the symmetry classification of differential equations containing arbitrary
functions. Two different cases have been considered in detail, and the main conclusion was
that only special choices of these functions can allow the presence of some symmetry. In
turn, the presence of symmetries is strictly connected to the likewise relevant problem of
Symmetry Analysis of Differential Equations 397
the reduction and the integrability (possibly by quadratures) of the differential equations, as
we have seen.
Our present approach was essentially based on the direct determination of Lie point-
symmetries. Actually, many generalizations of the notion of Lie point-symmetries have
been introduced, and also several parallel procedures have been invented. Apart from the
notion of generalized symmetries, which played some role also in our discussion in Sect. 3,
let us just mention only the notion of conditional symmetry [4,24] (see also [14,40]) and the
one of λ-symmetry [25], which are indeed particularly useful for finding invariant solutions
and in the reduction problem even in the absence of standard symmetries. More specifically,
for what concerns the general problem of the reduction procedure, see for instance [28, 31,
33, 36, 42, 43, 45] and the references therein. We can refer, e.g., to [3, 15, 20], and also to
the more recent papers [7,9,10] for larger (although unavoidably incomplete) lists of works
devoted to the above mentioned ideas and their countless applications and generalizations.
Let us also mention, finally, that other more sophisticated and/or more geometrically
oriented approaches have been proposed, mainly concerned with the problem of integrabil-
ity. We refer in particular to procedures based on the concept of solvable structures in the
context and the language of differential forms [1, 2, 30]. Let us only point out – to con-
clude – that one of the peculiar results of this procedure relates the presence of a solvable
algebra of symmetry vector fields with the integrability by quadratures: we just notice that
our case ii) in Sect. 2 is precisely a special case of this situation.
References
[1] Barco, M.A.; Prince, G.E., Acta Appl. Math. 66, 89 (2001), and Appl. Math. Comp.
124, 169 (2001)
[3] Bluman, G.W.; Anco, S.C., Symmetry and integration methods for differential equa-
tions, Springer: New York, 2002
[4] Bluman, G.W.; Cole, J.D., J. Math. Mech. 18, 1025 (1969) and Similarity methods for
differential equations, Springer: Berlin, 1974
[5] Bluman, G.W.; Kumei, S., Symmetries and differential equations, Springer: Berlin,
1989
[6] Bratu, G., Bull. Soc. Math. France 42, 113 (1914)
[7] Cicogna, G.; Laino, M., Rev. Math. Phys. 18, 1 (2006)
[8] Cicogna, G.; Ceccherini, F.; Pegoraro, F., Symmetry, Integrability and Geometry:
Methods and Applications (SIGMA) 2, Paper 017 (2006)
[12] Dirac, P.A.M., The principles of quantum mechanics, Clarendon Press: Oxford, 1958
[13] Fushchych, W.I.; Serov, N.I., J. Phys. A: Math. Gen. 16, 3645–3658 (1983)
[14] Fushchych, W.I., in Modern group analysis: advanced analytical and computational
methods in mathematical physics, Ed. by N.H. Ibragimov, M. Torrisi and A. Valenti
(Kluwer, Dordrecht 1993) p. 231–239
[15] Gaeta, G., Nonlinear symmetries and nonlinear equations, Kluwer: Dordrecht, 1994
[16] Gaeta, G.; Rodrı́guez, M. A., J. Phys. A: Math. Gen. 29, 859 (1996)
[17] Güngör, F.; Lahno, V.I.; Zhdanov, R.Z., J. Math. Phys. 45, 2280 (2004)
[18] Gusyatnikova, V.N.; Samokhin, A.V.; Titov, V.S.; Vinogradov, A.M.; Yamaguzhin,
V.A., Acta Appl. Math. 15, 23 (1989)
[19] Hydon, P.E., Symmetry methods for differential equations, Cambridge University
Press: Cambridge, 2000
[20] Ibragimov, N.H. (Ed.), CRC Handbook of Lie group analysis of differential equations,
CRC Press: Boca Raton, 1994, Vol.1; 1995, Vol.2; 1996, Vol. 3
[22] Ibragimov, N.H., Elementary Lie group analysis and ordinary differential equations,
J. Wiley & Sons: Chichester, 1999
[24] Levi, D.; Winternitz, P., J. Phys. A: Math. Gen. 22, 2915 (1989)
[25] Muriel, C.; Romero, J.L., IMA J. Appl. Math. 66, 111 (2001) and ibid 66, 477 (2001)
[26] Nikitin, A.G.; Popovych, R.O., Ukr. Math. J. 53, 1255–1265 (2001)
[29] Olver, P.J., Application of Lie groups to differential equations; Springer: Berlin, 1993,
second Edition
[30] Olver, P.J., Equivalence, invariants, and symmetry, Cambridge Univ. Press: Cam-
bridge, 1995
[31] Olver, P.J.; Rosenau, Ph., Phys. Lett. A 114, 107 (1986) and SIAM J. Appl. Math. 47,
263 (1987)
Symmetry Analysis of Differential Equations 399
[32] Ovsjannikov, L.V.: Group properties of differential equations, Siberian Acad. of Sci-
ences: Novosibirsk, 1962 and Group analysis of differential equations, Academic
Press: New York, 1982
[33] Popovych, R.O., Proc. Inst. Math. N.A.S. Ukr. 2, 437 (1997)
[34] Popovych, R.O.; Ivanova, N.M., J. Phys. A: Math. Gen. 37, 7547–7565 (2004)
[36] Pucci, E; Saccomandi, G., J. Phys. A: Math. Gen. 35, 6145 (2002)
[37] Pucci, E.; Salvatori, M.C., Int. J. Nonlinear Mech. 21, 147 (1986)
[39] Stephani, H., Differential equations. Their solution using symmetries, Cambridge Uni-
versity Press, Cambridge, 1989
[40] Vorob’ev, E.M., Sov. Math. Dokl. 33, 408 (1986), and Acta Appl. Math. 23, 1 (1991),
and ibid 26, 61 (1992)
[41] Wesson, J.: Tokamaks, The Oxford Engineering Series 48, Clarendon: Oxford, 1997,
2nd Edition
[42] Winternitz, P., in Group theoretical methods in physics (XVIII ICGTMP), Ed. by V.V.
Dodonov and V.I. Man’ko (Springer, Berlin 1991) p. 298–322
[44] Zhdanov, R.Z.; Lahno, V.I., Symmetry, Integrability and Geometry: Methods and Ap-
plications (SIGMA) 1, paper 009 (2005)
[45] Zhdanov, R.Z.; Tsyfra I.M.; Popovych, R.O., J. Math. Anal. Appl. 238, 101 (1999)
Chapter 15
1. Introduction
Contractions and deformations of Lie algebras and their relations have played an impor-
tant role in many fields since their introduction in the 1950’s, and many progress has been
done in understanding their structural and geometrical properties. Although being a rather
active research field, there remain various important problems concerning contractions and
deformations that have still not be satisfactorily solved. The notion of contraction appeared
first in physical context by Segal [1], and was soon recognized to have important conse-
quences, like the possibility of switching off interactions, or analyzing the precise effect
of some physical quantities when others are disregarded. The formal introduction of con-
tractions, done by Inönü and Wigner [2], was soon defined more generally by Saletan and
Kupczyński [3], in order to cover other limiting processes observed in symmetry groups
used in Physics, like the transition from relativistic to non-relativistic physics. Other, more
or less specifical, types of contractions have been introduced in the literature since, and their
structural properties analyzed [4, 5, 6, 7, 9, 10]. In addition, the contractions among Lie
algebras of fixed dimension have been studied in detail [11, 12, 13, 14, 15, 16], as well as
important classes of algebras, like those of kinematical groups [17, 18]. However, the lack
of complete classifications for Lie algebras from dimension six onwards is an important
∗
E-mail address: rutwig@pdi.ucm.es
†
E-mail address: jgueron@iafe.uba.ar
402 R. Campoamor-Stursberg and J. Guerón
obstruction that motivated different approaches to the problem. The relation of deformation
theory, a formalism born in Differential Geometry, with contractions of Lie algebras, was
first observed in [5], and has offered a kind of “inverse” procedure to study contractions.
This point of view also suggested a geometrical interpretation of contractions in terms of
orbits in a manifold, the points of which correspond to Lie algebras [19]. An advantage
of this approach is a definition of contractions that includes all special types used in the
literature, and that allow to establish different sufficiency criteria for the existence of con-
tractions. Moreover, this motivated the application of specific techniques like cohomology
of Lie algebras, which have proven to be an essential tool in many problems [20, 21, 22, 23].
Starting from this geometrical interpretation, the deformation and contraction problem
naturally leads to the computation of the irreducible components of the manifold of Lie
algebra structure tensors. The latter is deeply related to the notion of stability, which corre-
sponds to Lie algebras not admitting non-trivial deformations. In low dimensions results on
these components exist, although generally they have not been focused from the contraction
classification problem. The objective of this chapter is to analyze in detail the deformation
→
−
and contraction problem for solvable real Lie algebras g ⊕ R r having a non-trivial decom-
position, up to the eight dimensional case. These algebras, being a semidirect product of
semisimple and solvable algebras, are of great interest and importance in applications. We
approach their contractions using the notion of linear deformations, which turns out to be
sufficient for our purpose. This completes recent work concerning the contractions of sim-
ple Lie algebras [24], that points out the similarities of the contraction classification and
the embedding problem for semisimple Lie algebras and the branching rules of representa-
tions. In fact, Levi subalgebras of Lie algebras possess an interesting stability property that
allows to control, up to some extent, how the deformations and contractions behave [25]. In
contrast to solvable algebras, the representation of the Levi part describing the semidirect
product constitutes a first criterion to decide whether contractions among two given Lie
algebras can exist or not. Using the reversibility of contractions, an important structural
result, the deformation and contraction trees for these Lie algebras are established. The
problem of integrability of infinitesimal deformations and its relation to the stability of sys-
tems described by these Lie algebras studied in connection with their invariant theory. This
further provides additional information concerning other specific properties, like the exis-
tence of non-degenerate metrics associated to non-Abelian Yang-Mills theories and their
behavior with respect to deformation and contraction patterns.
Unless otherwise stated, any Lie algebra g considered in this work is defined over the
field R of real numbers. We convene that non-written brackets are either zero or obtained
by antisymmetry. We also use the Einstein summation convention. Abelian Lie algebras of
dimension n will be denoted by the symbol nL1 .
the radical. The Levi subalgebra s, which is unique up to conjugacy, acts on the radical r in
two possible ways, namely:
[s, r] = 0, (1)
[s, r] 6= 0. (2)
The first possibility implies that the algebra is decomposable, s ⊕ r, whereas the second
implies the existence of a representation R of s which describes the action, i.e., such that
The latter equations means that the Levi subalgebra acts by derivations on the solvable al-
→
−
gebra r. We will use the notation ⊕ R to describe semidirect products. From equation (3),
structural restrictions on the possible radicals are expected, while for direct sums any solv-
able Lie algebra is suitable. Real Lie algebras with non-trivial Levi decomposition 2 have
only been classified up to dimension nine [26, 27], essentially because of the non-existence
of classifications for seven dimensional solvable Lie algebras. However, the structure of the
representation R provides valuable information on the radical r, as obtained in [27]:
→
−
Proposition 1 Let s ⊕ R r be a Levi decomposition of a Lie algebra g.
2. If the representation R does not possess a copy of the trivial representation D0, then
the radical r is a nilpotent Lie algebra.
This result establishes a method to classify algebras having a specific Levi decompo-
sition. Fixed a semisimple Lie algebra f raks and a representation R, the radical is either
solvable or nilpotent according to the decomposability of the representation. The action of
the Levi subalgebra on the radical further tells that the Lie algebra of derivations of r must
contain the semisimple algebra s, the action being given exactly by R. This procedure has
been show to be effective in low dimensions, and also for important types of algebras, like
isotropical Lie algebras [26, 28] or Abelian Lie algebras, where it is known that the classi-
→
−
fication of Lie algebras s ⊕ R nL1 is reduced to classify the semisimple subalgebras of the
special affine algebra sl(n, C) and their real forms.
The orbits O(g) of a point g (i.e., Lie algebra) by the action of the linear group GL(n, R)
are formed by all Lie algebras isomorphic to g. The study of the structure of these orbits
naturally leads to consider deformations of Lie algebras. More specifically, this leads to the
analysis of neighborhoods of a given Lie algebra, as well as the intersection properties of
orbits. A special place is reserved to stable Lie algebras, i.e., algebras the orbits O(g) of
which are open in the manifold [25]. In the context of physical applications, these are the
desirable algebras, because small perturbations of the system do not change the symmetry
group [21].
In addition to the geometrical methods to study orbits of Lie algebras [29], more specif-
ical techniques like the adjoint cohomology of Lie algebras have been shown to be a pow-
erful tool [19].
In this work we will restrict ourselves to the adjoint cohomology groups, for being
the relevant object to study the orbits, although cohomologies can be defined on arbitrary
modules.3
A n-cochain ϕ on a Lie algebra g = (V, µ = [., .]) is a multi-linear skew-symmetric
map ϕ : V × .n . × V → V , where V is the adjoint g-module. Observe that by the
identification of g with the pair (V, µ), we can suppose that the Lie bracket [., .] is given by
[X, Y ] = µ(X, Y ) for all X, Y ∈ V . By means of the coboundary operator
n+1
X h i
dϕ(X1, .., Xn+1) = b i, .., Xn+1) +
(−1)i+1 Xi, ϕ(X1, .., X
i=1
X
b i, .., X
(−1)i+j ϕ [Xi , Xj ] , X1, .., X b j, ..Xn+1 (4)
1≤i,j≤n+1
we obtain a cochain complex d : C n (V, V ) → C n+1 (V, V ), n ≥ 0 , i.e., the condition
d ◦ d = 0 holds. An element ϕ ∈ C n (V, V ) is called n-cocycle if dϕ = 0, and a n-
coboundary if there exists σ ∈ C n−1 (V, V ) such that dσ = ϕ. The spaces of cocycles and
coboundaries are denoted by Z n (V, V ), respectively B n (V, V ). By equation (4), we have
the inclusion relation B n (V, V ) ⊂ Z n (V, V ) for all n, and the quotient space
is called n-cohomology space of g for the adjoint representation [20]. For practical pur-
poses, the most important cohomology spaces correspond to the values n = 0, 1, 2, 3
[25, 21, 30, 31]. For n = 0 it is straightforward to verify that H 0(g, g) coincides with
the center of g. For n = 1 we have
Z 1 (g, g) = {f : g −→ g | df = 0} .
B 1 (g, g) = {adX, X ∈ g} .
Therefore the space H 1 (g, g) can be interpreted as the set of the outer derivations of the
Lie algebra g.
The second and third cohomology groups have not such an obvious interpretation in
terms of the usual invariants of Lie algebras. They gain a specific sense when we consider
paths in the manifold Ln of n-dimensional Lie algebras that start from a certain point. This
geometrical notion can be immediately translated into an algebraic expression by means of
cohomological tools.
A formal one-parameter deformation gt of a Lie algebra g = (V, [., .]) is given by a
deformed commutator:
where dψl is the trilinear map of (4) for n = 2 and [ψ1, ψ1] is defined by (8)
1
[ψ1 , ψ1 ] (Xi, Xj , Xk ) := ψ1 (ψ1 (Xi , Xj ), Xk ) + ψ1 (ψ1 (Xj , Xk ), Xi ) + ψ1 (ψ1 (Xk , Xi ), Xj ) .
2
For the special case where equation (7) vanishes, we get the conditions
dψ1(Xi , Xj , Xk ) = 0, (9)
1
[ψ1, ψ1] (Xi, Xj , Xk ) + dψ2(Xi, Xj , Xk ) = 0. (10)
2
Equation (9) shows that ψ1 is a 2-cocycle in H 2(g, g), implying that deformations are gen-
erated by 2-cocycles. More specifically, the linear term of the deformation is a cocycle. On
the other hand, equation (10) implies that the deformation satisfies a so-called integrability
condition. Additional integrability conditions are obtained if the deformed bracket is de-
veloped up to higher orders of t [19, 30]. In particular, if for some ψ1 ∈ Z 2 (g, g) we have
[ψ1, ψ1] = 0, then the cocycle is called integrable and the linear deformation g+tψ1 defines
a Lie algebra. It can be shown that the integrability conditions of linear deformations are
codified by the third cohomology space H 3 (g, g). If the latter vanishes, then any cocycle
can be taken as the linear term of a deformation [19].
If the deformed algebra gt is isomorphic to g, we say that the deformation gt is trivial.
It is not difficult to show that whenever this happens, we can find a non-singular map ft :
5
It can be developed up to an arbitrary order, which provides additional conditions to be satisfied.
406 R. Campoamor-Stursberg and J. Guerón
V → V such that ft ([X, Y ]t ) = [ft X, ftY ] for all X, Y ∈ V . This means that ψ1 = dft ,
and the cocycle is trivial, i.e., a coboundary. Thus trivial deformations are generated by
2-coboundaries B 2 (V, V ) [21]. It can therefore be expected that the vanishing of H 2 (g, g)
implies some important property concerning the corresponding Lie algebra. Actually, this
is the contents of an important structural theorem due to Nijenhuis and Richardson [19]:
Theorem 1 Let g be a (real) Lie algebra such that H 2 (g, g) = 0. Then the orbit O (g) of
g is an open set in the manifold Ln .
Roughly speaking, this result means that any Lie algebra g0 close to g is isomorphic to
g [23, 10, 21], i.e., that the algebra has no non-trivial deformations. We therefore define
Definition 1 A Lie algebra g = (V, µ) is called stable 6 if its orbit O(µ) is open.
As known, the Whitehead lemmas imply that H 2 (s, s) = 0 for any semisimple Lie
algebra. Therefore these algebras provide very important examples of rigid algebras. This
result was generalized in [32], where the stability of any parabolic subalgebra was shown.
However, the stability theorem of Nijenhuis and Richardson only constitutes a sufficient,
but not necessary condition for a Lie algebra to be stable [25]. An interesting structural
result concerning stable Lie algebras was obtained in [33]:
Theorem 2 Any rigid Lie algebra g satisfies one of the following conditions:
1. The radical r7 is not nilpotent and satisfies dim Der (g) = dim g. If moreover
codimg [g, g] > 1, then g is complete 8.
2. The radical is nilpotent and verifies one of the following constraints:
where te (g) denotes the common dimension of Abelian subalgebras of Der(g) gen-
erated by outer semisimple derivations.
where H j (r, g)g is the space of g-invariant cocycles. These are the multilinear skew-
symmetric maps ϕ ∈ C j (r, s) that satisfy the coboundary operator (4) and such that
j
X
(Xϕ)(Y1, .., Yj ) = [X, ϕ(Y1, .., Yj )] − ϕ [X, Yi ] , Y1, , .Ybi, .., Yj = 0,
i=1
∀X ∈ s, Y1 , .., Yj ∈ r.
This result suggests that Levi subalgebras are stable in some sense, and that deformations
are determined by appropriate modification of the brackets in the radical. This idea actually
constitutes the germ of the important strong stability theorem of Page and Richardson [25],
which will be essential in our analysis. This result makes a precise statement about stability
of Levi subalgebras.
2. µ(X, Y ) = µ0 (X, Y ), ∀X ∈ s, Y ∈ r.
In essence, this stability theorem establishes that if the Lie algebra g has a semisimple
subalgebra s, then its deformations will have some subalgebra isomorphic to s, and that the
action of s on the remaining generators is preserved, that is, the representation describing
the semidirect product is preserved. Taking into account the Hochschild-Serre spectral
sequence, this means that the main information about deformations of semidirect products
is codified in the radical of the algebra. As a consequence of this theorem, some properties
on the cohomology of Lie algebras can be easily derived:
Proposition 2 Let g = s ⊕ r be the direct sum of a semisimple Lie algebra s and an arbi-
trary algebra r. Then H 2 (g, g) ' H 2 (r, r).
It suffices therefore to consider the s-invariance. Now, for any ϕ ∈ H 2 (r, g)s and X ∈
s, Y, Z ∈ r we have
(Xϕ) (Y, Z) = [X, ϕ (Y, Z)] − ϕ ([X, Y ] , Z) − ϕ (Y, [X, Z]) = [X, ϕ (Y, Z)] = 0. (13)
408 R. Campoamor-Stursberg and J. Guerón
because the sum is direct. Now ϕ (Y, Z) ∈ g, and by the decomposition of g we can rewrite
it as
ϕ (Y, Z) = W1 + W2 , W1 ∈ s, W2 ∈ r. (14)
Since s is semisimple, for any X there exists X 0 ∈ s such that [X, X 0] 6= 0. By the
invariance condition (13) we must have W1 = 0, thus ϕ (Y, Z) ∈ r for all Y, Z ∈ r. This
proves that any invariant cochain is actually a 2-cochain of the radical, from which the
assertion follows by imposing the coboundary condition.
Thus for direct sums s ⊕ r of a semisimple and an arbitrary Lie algebra r, the deforma-
tions are exclusively those of r. It should be remarked that if none of the direct summands is
semisimple, this results does not longer hold. Some interesting criteria were found in [34],
in connection with the problem of space-time and internal symmetry of elementary particle
physics.
If the limit
[X, Y ]∞ := lim Φ−1
t [Φt (X), Φt(Y )] (16)
t→∞
exists for any X, Y ∈ g, then equation (16) defines a Lie algebra g0 called the contraction
of g by Φt ). It is called non-trivial if g and g0 are non-isomorphic Lie algebras. With this
definition it is obvious that contractions are transitive, i.e., given two contractions g g0
and g0 g00, we obtain the contraction g g00.10 Taking into account the expressions of
the coboundary operator d used in cohomology, it is not difficult to see that the infinitesimal
version of equation (16) is generated by a coboundary [21]. In fact, if we consider a trivial
cocycle ψ ∈ B 2 (g, g), let σ be the 1-cochain such that dσ = ψ. Using the exponential map
we obtain the linear transformation ft = exp(−tσ), and expressing the brackets over the
transformed basis {ft (Xi)}, we get
Therefore a contraction can be obtained by taking limits in (17). This is the main point to
relate contractions of Lie algebras with deformations. An important result states that for any
contraction of Lie algebras g → g0, there is a deformation of g0 that reverses it [5]. However,
it should be remarked that a formal deformation is not necessarily related to a contraction [9,
40]. In this context, it is worthy to be mentioned that nontrivial cocycles generate nontrivial
deformations, while coboundaries generate trivial deformations. In particular, this implies
that the infinitesimal version of equation (16) is generated by a trivial cocycle. This allows
us to interpret contractions as finite deformations generated by a trivial cocycle [31]. Using
this fact, we can define a contraction by pure geometrical means:
Otherwise stated, contractions correspond to points in the closure O(g) of the orbit
of the Lie algebra g. This definition is completely independent on the particular form of
the endomorphims Φt used, and comprises all different types of contractions considered in
the literature, like simple Inönu-Wigner, Saletan, Lévy-Nahas or generalized Inönü-Wigner
contractions [1, 2, 3, 5, 6, 9]. From this representative free definition, the notion of equiv-
alence of contractions follows also at once. In particular, it implies that the orbit O(g0 ) is
contained in the orbit of g.
The analysis of many important types of Lie algebras, like classical kinematical alge-
bras [17], have shown explicitly the close relation between deformations and contractions
commented above, and may suggest that those contractions and deformations appearing in
physical applications are inverse procedures. Although it is not globally true, since there
are deformations not related to contractions [36], any contraction is actually related to a
deformation [9].
The problem of which deformations are related to a contraction is solved in the follow-
ing result:
As a consequence, non-invertible deformations are not of plateau type, i.e., for differ-
ent values of the parameter the deformed Lie algebras are pairwise non-isomorphic. In
particular, this implies that a stable Lie algebra can never appear as the contraction of a
non-isomorphic algebra. It should however be observed that there exist Lie algebras which
are not contractions, and nevertheless they are not stable. Further, this result gives a hint
for which classes of Lie algebras the invertibility of deformations can fail, namely families
of Lie algebras with some parameter that acts as a scaling factor on some of its generators.
Resuming these observations, it follows that contractions of Lie algebras can be studied
either directly, as has been done in low dimensions (see [15, 14, 9] and references therein),
or applying cohomological tools [5]. More specifically, in the latter case the deformations
410 R. Campoamor-Stursberg and J. Guerón
of Lie algebras are computed, and those being invertible provide the searched contractions
[31, 36]. This procedure seems to be more effective for algebras in high dimensions, or
having a Levi subalgebra, due to the successive simplifications provided by results like the
Hochschild-Serre reduction. This approach also allows us to find some criteria that simplify
the study of contractions.
ential operators:
Xbi = C k xk ∂ , (18)
ij
∂xj
where [Xi , Xj ] = Cij k X (1 ≤ i < j ≤ n) and {x , .., x } is a dual basis of {X , .., X }.
k 1 n 1 n
The invariants of g (in particular, Casimir operators) are the solutions of the system of
partial differential equations:
bi F = 0,
X 1 ≤ i ≤ n. (19)
The number N (g) of functionally independent solutions is obtained from the classical cri-
teria for differential equations, and equals:
k
N (g) := dim g − rank Cij xk , (20)
kx
where A(g) := Cij k is the matrix associated to the commutator table of g over the
given basis. It is known (see e.g. [36]) that for a contraction g g0 of Lie algebras, the
following inequality must be satisfied
N (g) ≤ N g0 . (21)
That is, contractions may generate additional independent invariants for the coadjoint
representation. By Theorem 4, any deformation of plateau type reverses the preceding
inequality. For completeness in the exposition, the generalized Casimir invariants of all
indecomposable non-solvable real Lie algebras up to dimension eight are given in Tables 4
and 5.
This result constitutes a slight variation of the Page-Richardson stability theorem seen
→
−
before. If g = s ⊕ R r is a contraction of s0, then there exists some deformation of g reversing
the contraction [9]. By the stability theorem, this deformation has some subalgebra that is
isomorphic to the Levi part of g, and acts the same way on the generators of the radical.
Therefore the embedding of semisimple Lie algebras s ,→ s0 induces a branching rule for
representations, and the quotient algebra s0/s, seen as an s-module, is isomorphic to the
representation R, that is, ad(s0)|s = ad(s) ⊕ R. This proves (i) and (ii). Finally, the third
condition follows from the properties of contractions of invariants [36].
Case 2. is typical for double inhomogeneous Lie algebras, and has also appeared in
the classification of kinematical Lie algebras [17, 35, 44, 43]. The first possibility has been
used to establish the stability of certain semidirect products of simple Lie algebras with
Abelian algebras such that the describing representation is irreducible [25, 45].
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 413
where εi ∈ R are real parameters. It follows from the stability theorems that
L8,14 (ε1 , ε2, ε3) is a semisimple Lie algebra or isomorphic to one of the Lie algebras hav-
ing the describing representation 2D 1 ⊕ D0 of sl (2, R) in its Levi decomposition. For this
2
Lie algebra it can be shown that
dim H 3 (L8,14, L8,14) = 3,
which implies that the Jacobi condition will be satisfied under constraints of the parameters
εi , i.e., constraints on the integrability condition are expected. In fact, these are given by
ε2 ε3 = 0, ε1 ε3 = 0.
if and only if ε1 6= 0.
11
Since both the brackets of the Levi subalgebra and the action on the radical remain unchanged by the
deformation, we skip its explicit presentation.
414 R. Campoamor-Stursberg and J. Guerón
X60 = X6 − ε−1 0 −1 0 −1
1 X4 , X7 = X7 − ε1 X5 , X8 = ε1 X8 .
Over the new basis {X1 , .., X5, X60 , X70 , X80 } the brackets of L8,14 (ε1 , ε2, 0) are
transformed to
[X4, X80 ] = X4, [X5 , X80 ] = X5, [X60 , X80 ] = 0, [X7, X8] = 0,
proving that
L8,14 + ε1 ϕ1 ' L08,17.
Since the deformation does not depend on the particular value of ε1 , this de-
formation is of plateau type, and therefore reversible. In order to obtain the
corresponding contraction
L08,17 L8,14,
we first consider the change of basis
1 0 1 1 1 1
X400 = X , X500 = 3 X50 , X600 = X60 , X700 = X70 , X800 = 2 X80 ,
t2 4 t t t t
where t ∈ R. The brackets are
h 00 i
X4 , X800 = t12 X400, [X500, X800] = 1
X 00,
t2 5
[X600, X800] = X400,
[X700, X800] = X500,
and it is easily verified that for t → ∞, the resulting algebra is identical with
L8,14.
(b) Let ε2 6= 0 and ε1 = ε3 = 0. As follows from the structure of the deformation,
for any value of ε2 the deformation L (0, ε2, 0) satisfies the condition
therefore L (0, ε2, 0) must be isomorphic to one of the following Lie algebras:
Lε8,13, L8,15, L−1 0
8,17 , L8,18 . Further, dim [L (0, ε2 , 0) , L (0, ε2, 0)] = 7, which
excludes the Lie algebras Lε8,13 and L8,15, since these are perfect, i.e., they
12
Again, the invariant generators are skipped for brevity.
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 415
X40 = γβX4, X50 = γβX5, X60 = βX6, X70 = βX7 , X80 = γX8.
Then
[X40 , X80 ] = γ 2 ε2 X60 , [X50 , X80 ] = γ 2 ε2 X70 , [X60 , X80 ] = X40 , [X70 , X80 ] = X50 .
and thus
γ 2ε2 = ±1
depending whether ε2 is positive or negative.
(a) If ε2 = 1. In this case, consider the change of basis
X40 = αX4 + αX6 , X50 = αX5 + αX7 , X60 = βX4 − βX6 , X70 = βX5 − βX7,
where αβ 6= 0. It follows at once that
[X40 , X80 ] = X40 , [X50 , X80 ] = X50 , [X60 , X80 ] = −X60 , [X70 , X80 ] = −X70 ,
showing that
L8,14 + ε2 ϕ2 ' L−1
8,17 .
>0
L−1
8,17 L8,14,
In this case, the contraction of L08,18 onto L8,14 is determined by the transformations
L8,16, Lp68,17
=−1
, Lp68,18
=0
.
shows that
[X40 , X80 ] = X40 , [X50 , X80 ] = X50 , [X60 , X80 ] = X40 + X60 , [X70 , X80 ] = X50 + X70 ,
thus
L8,14 + 2ϕ1 − ϕ2 ' L8,16.
X40 = αX4 + αε2 X6 , X50 = αX5 − αε2 X7, X60 = βX4 − βX6 , X70 = βX5 − βX7
Define
X40 = t−1 X4 , X50 = t−1 X5, X60 = t−1 X4 + X6
X70 = t−1 X5 + X7 , X80 = t−1 X8.
for the deformed algebra. It is straightforward to verify that this family defines a con-
traction L−ε 2
8,17 L8,14.
3. Finally, if ε1 + ε2 6= 1, the change of basis
X40 = αX4 + βX6, X50 = αX5 + βX7, X60 = γX4 + δX6 , X70 = γX5 + δX7
with
ε1 γ ε21 ε1
α=δ+ , β=− +1 γ− δ
2 4 2
leads to the deformation
1 ε /4
L8,14 + ε1 ϕ2 + ε2 ϕ2 ' L8,18 .
It should be remarked that, in contrast with the previous deformations, this is not re-
versible. In fact, the structure of the 3-dimensional subalgebras of Lp8,18 and L8,14
generated by the elements {X4 , X6, X8} and {X5 , X7, X8} prevent the existence of a
contraction onto L8,14 [13].
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 417
Lp68,18
=0
L08,18 Lq68,17
=−1
6 -
↓
↓
- L8,15
L8,16 L8,14
-
L08,17 L−1
8,17
7.1. Dimension 5
There is only one indecomposable Lie algebra in this dimension having a non-trivial Levi
decomposition, the semidirect product of sl (2, R) and the irreducible representation D 1 .
2
This algebra is obviously stable, and any contraction is necessarily solvable. The remaining
algebras having a Levi subalgebra are the direct sums so (3) ⊕ r and sl (2, R) ⊕ r, with r a
two dimensional Lie algebra. For these, the deformation problem is trivial, since there are
only two Lie algebras in dimension two.
7.2. Dimension 6
Six is the lowest dimension where we find non-solvable Lie algebras appearing as contrac-
tions of semisimple algebras. Any six dimensional real Lie algebra with Levi subalgebra is
either semisimple, indecomposable (isomorphism classes L6,1 − L6,4 in Table 4), isomor-
phic to L5,1 ⊕ R or the direct sum of a simple algebra of rank one and a three dimensional
solvable Lie algebra. For the latter class, the deformation problem again reduces to the
analysis of deformations and contractions in dimension three, which is well known. The
semisimple algebras so (1, 3), so (4), so∗ (4) and so (2, 2), as well as the algebras L6,2 and
L6,3 are stable, and therefore have no deformations. The Lie algebras L6,1 and L6,4 , being
inhomogeneous, appear as contractions of semisimple algebras, while the decomposable al-
gebra L5,1 ⊕ L1 deforms onto L6,2 and L6,3. The situation can be resumed in the following
7.3. Dimension 7
Seven dimensional algebras with non-trivial Levi decomposition were partially studied in
[24] and [40], where some interesting patterns were observed in their behavior with respect
to contractions and deformations. Any non-solvable seven dimensional algebra is either
reductive (i.e, the direct sum of a semisimple and an Abelian algebra), indecomposable
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 419
and isomorphic to L7,1 − L7,7 (see Table 4), isomorphic to L6,1 ⊕ L1, L6,2 ⊕ L1, L6,3 ⊕
L1 , L6,4 ⊕ L1 , L5,1 ⊕ 2L1 or L5,1 ⊕ r2 , or the direct sum of a rank one simple Lie algebra
and a four dimensional solvable Lie algebra. The interesting fact about this dimension is
that almost any indecomposable non-solvable algebra is stable [24].
Proposition 6 The reductive Lie algebras so (1, 3) ⊕ L1 , so (4) ⊕ L1, so∗ (4) ⊕ L1 ,
so (2, 2) ⊕ L1 and the indecomposable algebras L7,1, L7,2, L7,4, L7,5, L7,6 and L7,7 are
stable.
8. L5,1 ⊕ 2L1 appears as contraction of L6,2 ⊕ L1, L6,3 ⊕ L1, Lp7,3, L7,4 and L5,1 ⊕ r2 .
Only the Lie algebras L7,6 and L7,7 have no non-solvable contraction, due to the struc-
ture of their defining representation R.
The corresponding bases of the cohomology space, as well as other important characteristic
like the dimension of the commutator and derivation algebras, number of invariants etc. are
given in Table 2. The structure constants of the Lie algebras are given in Table 1.
420 R. Campoamor-Stursberg and J. Guerón
8.2. R = R4 ⊕ D0
Three types of isomorphism classes are found for this describing representation, one of
them parameterized continuously.
8.2.1. L8,2
→
−
The algebra L8,2 has Levi decomposition L8,2 = so (3) ⊕ R A5,4 and a one dimensional
cohomology space. Let L8,2() = L8,2 + ϕ be a linear deformation of L8,2. For any
value of the deformed commutator satisfies the Jacobi identity, thus defines a Lie algebra.
Computing the Killing metric tensor κ over the basis {X1 , .., X8}, we obtain the matrix
−3 0 0 0 0 0 0 0
0 −2 0 0 0 0 0 3
2
0 0 −3 0 0 0 0 0
0 0 0 −6 0 0 0 0
κ= 0
0 0 0 −6 0 0 0
0 0 0 0 0 −6 0 0
0 0 0 0 0 0 −6 0
3
0 2 0 0 0 0 0 −92
We have det(κ) = 2238 76 6= 0 for 6= 0, and therefore the deformation is a semisimple
Lie algebra. To identify to which real form g is isomorphic, we compute the spectrum of
κ and obtain
( s )
3 9 2 1 3 7
Spec(κ) = −3, 3, (−6) , − − 1 ± 92 − + . (25)
2 2 2 4
q
Since 92 + 2 > 92 − 32 + 74 for any , the two last roots of Spec(κ) are always
negative, and the signature σ of κ is given by
−8, ε > 0
σ (κ) = .
0, ε < 0
For σ = −8 we obtain the compact Lie algebra su(3), while for σ = 0 we get the
pseudo-unitary algebra su(2, 1) [46]. Finally, starting from the deformed bracket, apply-
ing formula (16) to the family of linear maps defined by ft (Xi ) = Xi, (i = 1, 2, 3),
ft (Xi) = t−1 Xi , (i = 4, ..., 7) and ft (X8) = t−2 X8, we obtain the contraction of su(2, 1)
onto L8,2.
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 421
8.2.2. L8,3
→
−
The Lie algebra L8,2 = so (3) ⊕ R A1,1,1
5,7 has no invariants, it can thus never appear as the
contraction of a semisimple Lie algebra, in view of proposition 3. For a formal deformation
L8,3 (ε1 , ε2, ε3) = L8,3 + ε1 ϕ1 + ε2 ϕ2 + ε3 ϕ3
there is no obstruction to the Jacobi identity, therefore for any values of εi it defines a Lie
algebra. Moreover, independently of the values of the deformation parameters, we have
N (L8,3 (ε1 , ε2, ε3)) = 0.
In view of Table 2, this means that there is only one possibility for the deformations, namely
L8,3 (ε1 , ε2, ε3) ' Lp8,4.
It follows from the structure of the three dimensional Lie algebras spanned by {X4, X6, X8}
and {X5, X7, X8} that no contraction of Lp8,4 onto L8,3 can exist.
8.2.3. Lp8,4
→
−
For the family Lp8,4 = so (3) ⊕ R A1,p,p
5,17 two cases have to be distinguished, according to p
being zero or not.
1. If p 6= 0, then there is only one cocycle class ϕ, and it follows at once that
Lp8,4 + εϕ ' Lε+p
8,4 ,
which cannot give rise to a contraction for being a member in the same family.
2. For p = 0, we have dim H 2 L08,4, L08,4 = 2. A deformation L08,4 (ε1 , ε2) = L08,4 +
ε1 ϕ1 + ε2 ϕ2 has to satisfy the integrability condition ε1 ε2 = 0. It is straightforward
to verify that the linear deformation L08,4(0, 2) = L08,4 + 2 ϕ2 has a codimension one
derived ideal, and cannot be semisimple. Indeed,
L08,4 (0, ε2) ' Lε8,4
2
,
and no reversal is possible because both algebras have the same number of deriva-
tions. Considering the remaining deformation L08,4 (1, 0) and computing the spec-
trum of the Killing tensor κ, we obtain that
Spec(κ) = (−3)3 , −4, (−61)4 .
Thus
−8, 1 > 0
σ (κ) = ,
0, 1 < 0
which means that L08,4 + 1 ϕ1 ' su(3) if 1 > 0 and L08,4 + 1 ϕ1 ' su(2, 1) if
1 < 0. Defining on L08,4(1 , 0) the linear maps
8.3. R = R5
→
−
Since L8,5 = so (3) ⊕ R5 5L1 is an inhomogeneous algebra, any non-trivial deformation
must be a semisimple Lie algebra. For L8,5() = L8,5 + ϕ the spectrum of the Killing
form κ is given by
Spec(κ) = (−12)2, −8, −4, −6, (−24)2, −72 ,
thus σ(κ) = −8 if > 0 and σ(κ) = 2 if < 0. This proves that L8,5() ' su(3) if > 0
and L8,5() ' sl(3, R) otherwise. The contraction of su (3) respectively sl (3, R) onto L8,5
are defined by the transformations
8.4. R = D 12 ⊕ 3D0
This representation is the first where a considerable number of parameterized families with
very similar properties appear. This fact is a consequence of the high number of copies of
the trivial representation that R contains. We also remark that this case has no counterpart
for the Levi subalgebra so (3), because the half-spin representation D 1 of sl (2, R) is of
2
second class.
8.4.1. L8,6
→
−
For the Lie algebra L8,6 = sl (2, R) ⊕ R h2 the integrability condition of a formal deforma-
tion L8,6 (ε1 , ε2, ε3 , ε4) is given by the non-linear system
of L6,2 ⊕ r2 . On the other hand, the deformation L8,6 (ε1 , 2ε1, ε3, −2ε3) leads also to a
decomposable Lie algebra, namely
It may wonder why the remaining eight dimensional algebras with the same R do not appear
as deformations of L8,6. This is easily explained observing that L8,6 (ε1 , ε2, ε3, ε4) has a
one dimensional center, which happens only for L08,8. However, this algebra cannot appear
because of the structure of the radical.
8.4.2. L8,7
→ 1,p,q
−
The algebra Lp,q
8,7 = sl (2, R) ⊕ R A5,7 is the first of the two families to depend on two
parameters. As expected, for this class of algebras the cohomology depends heavily on the
values of p and q, and gives rise to a high number of subcases (see Table 3). Here we also
find the highest dimensional cohomology space.
1. p 6= q, p 6= 2, q 6= 2.
There are no integrability conditions for the deformation Lp,q
8,7 (ε1 , ε2 ).
Lp+ε 1 ,q+ε2
, (p + ε1 ) (q + ε2) 6= 0,
Lp,q
8,7 (ε1 , ε2 ) ' 8,7
p
L7,3 ⊕ L1, (p + ε1 ) (q + ε2) = 0.
2. p 6= q, p = 2, q 6= 2 [equivalent
to p 6= q, q = 2, p 6= 2]
For these values the space H L2,q
2 2,q
8,7 , L8,7 has dimension three. The integrability
2,p
condition for a deformation L8,7 (ε1 , ε2, ε3) is
ε1 ε3 = 0.
L2,p k
8,7 (0, ε2, ε3 ) ' L8,10
with k = (q + ε2 ) ε−1
3 . For any value of k we can construct the contraction
Lk8,10 L2,q
8,7
3. p = q
For these values of p and q, a formal deformation depends on four parameters if
p 6= 2, and on six for the equality. For the case p = 2 the integrability condition is
ε1ε5 + ε4 ε6 = 0, ε2 ε6 + ε2 ε5 = 0.
424 R. Campoamor-Stursberg and J. Guerón
Among these possibilities, only two of the deformations can be reversed to give a
contraction onto L2,2
8,7 . The contraction
L8,11 L2,2
8,7
The different criteria for contractions show that none of these deformations is of
plateau type, and cannot thus be reversed.
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 425
8.4.3. Lp8,8
This family presents the particularity that the radical depends on the values of the parameter
p. Specifically, the Levi decomposition equals
(
→
− Ap−1,p−1 , p 6= 0
Lp8,8 = sl (2, R) ⊕ R 5,9
1 .
A5,8, p=0
1. p = 0
The integrability condition for a deformation L08,8 (ε1 , ε2 , ε3) reads
ε2 ε3 = 0.
L6,3 ⊕ r2 L08,8.
2. p = 2
The integrability conditions for L28,8 (ε1 , ε2, ε3) are
ε1 ε3 = ε2 ε3 = 0.
Only the first can be reversed. The linear isomorphisms ft of L8,11 defined by
3. p 6= 0, 2
As follows from the structure of the cohomology space, the deformations of Lp8,8 for
general p are special cases of the special values analyzed before. The deformations
are therefore p,q
2
L8,9, 4ε2 + ε1 < 0
p ε1 /2
L8,8 (ε1 , ε2 , ε3) ' L8,8 , 4ε2 + ε21 = 0 .
Lp,q , 4ε + ε2 > 0
8,7 2 1
8.4.4. Lp,q
8,9
→ 1,p,q
−
Although the Lie algebras Lp,q 8,9 = sl (2, R) ⊕ RA5,13 depend on two parameters p and q 6=
0, its cohomology space is quite simple (see Table 2), and for all values of p, q the cocycle
classes coincide. This absence of special cases makes the analysis of deformations quite
easy. There is no integrability condition to be satisfied. In this case, it can be shown that
any formal deformation Lp,q p,q
8,9 (ε1, ε2 ) is equivalent to a deformation of the type L8,9 (ε1 , 0).
For the latter, the possibilities are
p+ε1 /2, 21 q 0
L8,9 , q 02 = 4q − ε21 > 0
p,q
L8,9 (ε1 , 0) '
Lp±q , q 02 = 4q − ε21 = 0 .
8,8
r,s
L8,7, q 02 = 4q − ε21 < 0
The dimension of the corresponding algebras of derivations show that none of these defor-
mations can be reversed to a contraction onto Lp,q
8,9 .
8.4.5. Lp8,10
→
−
The algebras Lp8,10 = sl (2, R) ⊕ R A2,p
5,19 must also be analyzed separately for some values
of the parameter p.
1. p 6= 2
For these values of p, there is only one cocycle class, which turns out to be integrable.
The corresponding deformation Lp8,10 (ε1 ) is easily seen to belong to the same family
or be decomposable. More precisely,
(
p(1+ε)−1
p
L8,10 (ε1 ) ' L8,10 , p + ε 6= 0
L6,2 ⊕ 2L1 , p + ε = 0
None of these deformations can be reversed. For the first it is obvious from the
dimension of the derivation algebra, while for the second, it follows from the fact
that L6,2 ⊕ 2L1 has non-trivial center, while Lp8,10 has not.
2. p = 2
For L28,10 (ε1 , ε2) there is no integrability condition, thus for any values of ε1 and ε2
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 427
they define Lie algebras. If ε1 6= 0 and ε2 = 0, the deformations are exactly those
previously obtained for p 6= 2. Taking L28,10 (0, ε2), a change of basis proves that
In this case, there is a contraction of L8,11 onto L28,10 determined by the linear iso-
morphisms
For arbitrary non-zero values of ε1 and ε2 the deformations are the following
(
2(1+ε )−1
L28,10 (ε1 , ε2) ' L8,10 1 , 1 + ε1 =6 0 ,
L6,2 ⊕ r2, 1 + ε1 = 0
8.4.6. L8,11
→
−
The only cocycle class of L8,11 = sl (2, R) ⊕ R A25,20 is integrable, and any deformation
L8,11 (ε) satisfies the constraint dim [L8,11 (ε) , L8,11 (ε)] = 7, which excludes L8,6 to be
reached. Further, for any value of ε we obtain that dim Der (L8,11 (ε)) = 9, which implies
that none of the deformations can be reverted to give a contraction. The detailed analysis
gives
(
2(1+ε)−1
L8,11 (ε) ' L 8,10 , 1 + ε 6= 0 .
L6,2 ⊕ r2, 1 + ε = 0
8.5. R = D1 ⊕ 2D0
There is only one algebra having this describing representation, namely Lp8,12 =
→
−
sl (2, R) ⊕ RA1,1,p p
5,7 . For any value of ε, the deformation L8,12 + εϕ defines a Lie algebra. It
is trivial to verify that
Lp8,12 (ε) ' Lε+p
8,12 ,
8.6. R = 2D 1 ⊕ D0
2
This representation provides the second largest group of Lie algebras, although only two
of them depend on a continuous parameter. Some of these Lie algebras turn out to have
the same complexification as the algebras having the compact Levi subalgebra so (3) and
describing representation R4 ⊕ D0 . This is also the group where most of the non-solvable
contractions of simple Lie algebras appear. The algebra L8,14, studied in detail in the pre-
vious section, has been omitted from this list.
428 R. Campoamor-Stursberg and J. Guerón
The signature is σ (κ) = 0 for = 1 and 2 for = −1, proving that L18,13 (µ) is isomorphic
to su (2, 1) and L−1
8,13(µ) is isomorphic to sl(3, R). In both cases, the deformations can
be reversed, and the corresponding contraction is obtained from the changes of basis in
L8,13(µ) defined by
ft (Xi) = Xi, (i = 1, 2, 3); ft (Xi) = t−1 Xi , (i = 4, ..., 7); ft (X8) = t−2 X8.
8.6.2. L8,15
→
−
Let L8,15 (ε) = L8,15 + εϕ be a formal deformation of L8,15 = sl (2, R) ⊕ R A5,3. The
computation of the Killing form gives det (κ) = 21438ε5 6= 0 for nonzero ε, and the
spectrum is given by
n o
Spec (κ) = −6, 6, 12, (−12ε)3 , (12ε)2 ,
thus σ (κ) = 0 for positive ε and σ (κ) = 2 for ε < 0. We obtain the deformations
su (2, 1) , ε > 0
L8,15 + εϕ ' .
sl (3, R) , ε < 0
ft (Xi) = Xi , (i = 1, 2, 3); ft (Xi) = t−3 Xi , (i = 4, 5); ft (Xi ) = t−1 Xi, (i = 6, 7); ft (X8) = t−2 X8.
8.6.3. L8,16
The third of the algebras not depending on continuous parameters is L8,16 =
→
−
sl (2, R) ⊕ RA15,15. The single cohomology class is not subjected to integrability condi-
tions, and in any case the deformation satisfies the constraint
N (L8,16 (ε)) = 0.
In addition, any of the possible target Lie algebras has a nine dimensional derivation alge-
bra, which proves that no algebra can contract onto L8,16. Specifically, we have
p
L8,17, ε 6= −1
L8,16 (ε) ' ,
L−1
8,18 , ε = −1
where q
1
p= 3ε − 1 + (ε − 1) + 4 (1 + ε)−1 .
2
2
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 429
8.6.4. Lp8,17
→
−
The family Lp8,17 = sl (2, R) ⊕ R A1,p,p
5,7 , originally studied in [40], presents some interesting
properties with respect to deformations.
1. p = 1
In this case there is no integrability condition, thus L18,17 (ε1 , ε2, ε3 ) satisfies the Ja-
cobi identity for all values of εi . Ignoring the cases which lead to equivalent defor-
mations, the possibilities are the following
ε1 +1
L8,17 , ε1 6= 0
L8,16, (0, ε2, 0) , (0, 0, ε3)
L18,17 (ε1 , ε2, ε3) ' 1
(1−ε2 ε3 ) .
L2 , ε2 ε3 ± 1 = 0
8,18q
L8,17, ε2 ε3 ± 1 6= 0
2. p 6= ±1
For any value of ε we immediately obtain
3. p = −1
This case is the most interesting, since it corresponds to a singular point of the family
that deforms onto a semisimple Lie algebra [40]. The integrability condition for a generic
deformation L−18,17 (ε1 , ε2 ) is ε1 ε2 = 0.
Let ε1 6= 0. We note that L−1 8,17 + ε1 ϕ1 is semisimple by considering the Killing form κ.
The spectrum of κ equals σ (κ) = 2 for any nonzero values of ε1 , therefore obtain the
deformation L−18,17 + ε1 ϕ1 ' sl (3, R). The corresponding contraction is determined by the
linear maps
ε2 −1
L−1
8,17 (0, ε2) ' L8,17 ,
8.6.5. Lp8,18
→
−
This family, Lp8,18 = sl (2, R) ⊕ R A1,p,p
5,7 , similar to the previous, has the particularity of
having a value for which it appears as the contraction of a simple Lie algebra.
1. p = 0
In this case we find two independent cocycle classes. The integrability condition for
the deformation L08,18 (ε1 , ε2) = L08,18 + ε1 ϕ1 + ε2 ϕ2 is ε1 ε2 = 0. Considering the
first possibility L08,18 + ε1 ϕ1 and computing the Killing tensor, we obtain det (κ) =
21437ε41 6= 0 and the spectrum
n o
Spec (κ) = −6, −4, 6, 12, (−12ε1 )2 , (12ε1 )2 ,
and in any case σ (κ) = 0, showing that L08,18 + ε1 ϕ1 ' su (2, 1). To obtain the
contraction, we consider on the deformations the transformations
where p
ε2 ± ε22 − 4
p= p .
ε2 ∓ ε22 − 4
2. p 6= 0
8.7. R = D2
There is only one algebra with this describing representation. Since the Lie algebra L8,21 =
→
−
sl (2, R) ⊕ R5L1 is inhomogeneous and dim H 2 (L8,21, L8,21) 6= 0, it follows from [25]
that L8,21 is the contraction of a semisimple Lie algebra. Considering L8,21 + εϕ, the
spectrum of the Killing form κ is given by
Spec (κ) = {−24, 24, 48, −48ε, −12ε, 12ε, 48ε, 72ε} .
The signature is σ (κ) = 2 for ε > 0, σ (κ) = 0 for ε < 0. We thus obtain that
su (2, 1) , ε < 0
L8,21 + εϕ ' .
sl (3, R) , ε > 0
In both cases, the contraction follows at once from the linear maps
8.8. R = D 1 ⊕ D1
2
→
−
The only semidirect product having this R, L8,22 = sl (2, R) ⊕ R 5L1 is also inhomoge-
neous, but, in contrast to the previous case, R is not irreducible. Therefore a deformation
is not necessarily semisimple. A formal deformation L8,22 (ε) satisfies the Jacobi condi-
tion for any ε. The Killing form κ of L8,22 (ε) turns out to be degenerate, thus L8,22 (ε)
cannot be semisimple. However, the subalgebra generated by the elements {X1, .., X6} is
semisimple and isomorphic to sl (2, R) ⊕ sl (2, R) for any value of ε. In view of this, it is
not difficult to show that
Lie algebras g = s ⊕ r being a direct sum of a rank one and a five dimensional solvable
algebra are left out, since the deformation problem is reduced exclusively to the solvable
radical. On the other hand, no detailed study has been yet done for the contractions of solv-
able Lie algebras in this dimension. One of the reasons is the high number of parameterized
families, as well as the difficulty of determining the cohomologies for these Lie algebras,
where no reduction criteria exist. We also ignore the Lie algebras L5,1 ⊕ r, r being a three
dimensional algebra. With two exceptions, algebras of this type deform onto algebras with
the same decomposition, and are therefore not of great interest in our analysis. We first of all
determine those algebras that do not have deformations and therefore are not contractions.
Proposition 8 The Lie algebras so (1, 3) ⊕ r2 , so (4) ⊕ r2, so (2, 2) ⊕ r2, so∗ (4) ⊕ r2 and
L6,3 ⊕ r2 are stable.
s ⊕ r2 s ⊕ 2L1,
where s = so (1, 3) , so (4) , so (2, 2) or so∗ (4). There are also no other possibilities for
the deformations of these direct sums with the two dimensional Abelian algebra. For the
remaining, non-stable algebras we study the problem separately, as done for the indecom-
posable case. Up to some special cases, the endomorphisms that give rise to the contractions
are skipped, for being formally very similar to the contractions already studied.
ε1 ε3 = ε2 ε3 = 0, ε2 ε4 + ε3 ε5 = 0.
We discard those solutions to the integrability conditions which determine equivalent de-
formations. For the remaining cases, we get the scheme
so (1, 3) ⊕ r2, ε1 6= 0, ε4 ε5 6= 0
so (1, 3) ⊕ 2L1, ε1 6= 0, εi = 0, i 6= 1
L6,1 ⊕ 2L1 (ε1 , .., ε5) ' .
6,1 ⊕ r2,
L ε4 6= 0 or ε5 6= 0, εi6=4,5 = 0
p
L8,1, ε2 ε5 6= 0
From any of these deformation we can easily construct the corresponding contraction onto
L6,1 ⊕ 2L1, which shows that all of them can be reversed.
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 433
9.2. L6,1 ⊕ r2
A basis of the cohomology space is given by the cocycles
The integrability condition for L6,1 ⊕ r2 (ε1 , ε2) = L6,1 ⊕ r2 + ε1 ϕ1 + ε2ϕ2 is given by
ε1 ε2 = 0.
Only the first deformation is of plateau type and defines a contraction (see the case in
dimension six), while the second cannot be reversed because of the dimension of the algebra
of derivations.
ε1 ε4 + ε2 ε5 = 0.
It is straightforward to verify that all these deformation can be reversed and give rise to
contractions onto L6,2 ⊕ 2L1 .
9.4. L6,2 ⊕ r2
The Lie algebra L6,2 ⊕ r2 has a one dimensional cohomology space
H 2 (L6,2 ⊕ r2 , L6,2 ⊕ r2 ) generated by the cocycle class
There is obviously no integrability condition, and for any nonzero value of ε we obtain the
deformation
−1
L6,2 ⊕ r2 + εϕ ' L−ε 8,10 .
Since dim Der Lp8,10 ≥ 9 and dim Der (L6,2 ⊕ r2) = 9, this deformation does not give
rise to a contraction.
For the decomposable algebra L6,3 ⊕ 2L1 the cohomology space is six dimensional, gener-
ated by the cocycles
ε3 ε5 − ε2 ε6 = 0, ε1 ε6 − ε3 ε4 = 0.
L6,3 ⊕ r2, ε4 = ε5 = ε6 = 0, ε3 6= 0
Lp7,3 ⊕ L1, ε3 = ε4 = ε5 = ε6 = 0, ε1 ε2 6= 0
L0 ,
8,8 ε1 = ε2 = ε4 = ε5 = ε6 = 0
L6,3 ⊕ 2L1 (ε1 , .., ε6) ' p,q .
L 8,7 , ε1 ε5 6= 0, ε3 = ε6 = 0
p
L 8,8 , ε2 6= 0, ε1 = ε5 , ε3 = ε6 = 0
Lp,q ,
8,9 ε1 = ε5 , ε2 = −ε4 , ε3 = ε6 = 0
Among these deformations, it is not difficult to see that the only contractions that can exist
are
the other being forbidden by the properties of contractions or the structure of three dimen-
sional subalgebras [14].
The only integrability conditions that a deformation depending on the ε0i s has to satisfy are
ε1 ε2 = ε1 ε3 = 0. As a consequence of a case by case inspection, the deformation scheme,
avoiding equivalent cases, can be resumed as:
so (1, 3) ⊕ 2L1, ε1 < 0, εi = 0, i 6= 4, 5
so (2, 2) ⊕ 2L1, ε1 > 0, εi = 0, i 6= 4, 5
so (1, 3) ⊕ r2 , ε1 < 0, ε4 ε5 6= 0,
L6,4 ⊕ 2L1 (ε1 , ε2, ε3, ε4, ε5) ' so (2, 2) ⊕ r2 , ε1 > 0, ε4 ε5 6= 0, .
L6,2 ⊕ r2 , ε4 ε5 6= 0, εi = 0, i 6= 4, 5
L7,5 ⊕ L1 ,
ε2 6= 0, εi = 0, i 6= 2
p
L8,12, ε5 = pε2 6= 0, εi = 0, i 6= 5
Basing on the deformations studied in dimensions n ≤ 7 and the structure of this algebra, it
is routinary to verify that all these deformations admit an inverse, and therefore any of the
deformed algebras contracts onto L6,4 ⊕ 2L1.
9.7. L6,4 ⊕ r2
The cohomology satisfies dim H 2 (L6,4 ⊕ 2L1 , L6,4 ⊕ 2L1) = 2, spanned by
As follows from the lower dimensional analysis, the two first deformations can be reversed
and provide a contraction, while the third cannot be reversed because of the dimensions of
the corresponding derivation algebras.
9.8. L7,1 ⊕ L1
→
−
For L7,1 ⊕L1 = so (3) ⊕ adso(3)⊕D0 A1,1 2
4,5 ⊕L1 we find dim H (L7,1 ⊕ L1 , L7,1 ⊕ L1 ) = 1,
generated by
ϕ (X7, X8) = X8.
In this case, the deformation problem is trivial, and for any value of ε we find that
9.9. L7,2 ⊕ L1
This algebra provides the highest dimensional cohomology space among the decomposable
algebras, with dim H 2 (L7,2 ⊕ L1, L7,2 ⊕ L1 ) = 7. A basis of cocycles can be chosen as
ϕ1 (X4 , X5) = X8, ϕ1 (X6 , X7) = −X8, ϕ2 (X4, X7) = X8 , ϕ2 (X5, X6) = −X8 ,
ϕ3 (X4 , X6) = X8, ϕ3 (X5 , X7) = X8, ϕ4 (X4, X8) = X4 , ϕ4 (X5, X8) = X5,
ϕ4 (X6 , X8) = X6, ϕ4 (X7 , X8) = X7, ϕ5 (X4, X8) = −X6 , ϕ5 (X5, X8) = −X7 ,
ϕ5 (X6 , X8) = X4, ϕ5 (X7 , X8) = X5, ϕ6 (X4, X8) = X4 , ϕ6 (X5, X8) = X5,
ϕ7 (X4 , X8) = −X6, ϕ7 (X5 , X8) = −X7.
In this case, the integrability conditions are rather complicated, but fortunately, many of
them lead to equivalent deformations, since there are only three possible target algebras,
namely L8,2, L8,3 and Lp8,4. Ignoring those cases which are equivalent, the possibilities for
a formal deformation of L7,2 ⊕ L1 are:
7
X L8,2, ε1 6= 0, εi = 0, i 6= 1
L7,2 ⊕ L1 + εi ϕi ' L8,3, ε4 =6 0, εi = 0, i 6= 1 .
p
j=1 L8,4, pε6 + ε4 = 0.
A simple computation (see Table 1) shows that any of these deformations can be inverted,
thus give rise to the contractions L8,2 L7,2 ⊕L1, L8,3 L7,2 ⊕L1 and Lp8,4 L7,2 ⊕L1
for any p.
ϕ1 (X4, X7) = X4, ϕ1 (X5, X7) = X5 , ϕ2 (X4 , X8) = X4, ϕ2 (X5, X8) = X5 ,
ϕ3 (X7, X8) = X8, ϕ4 (X4 , X5) = X6 [p = 2].
while the integrability condition is ε2 ε3 = 0 in the general case and the additional con-
straints ε2 ε4 = ε1 ε4 if p = 2. The complete deformation scheme resulting for these Lie
algebras is
−pε−1
L5,1 ⊕ A3,5 3 , ε3 6= 0, 1 + ε1 = 0
p(1+ε
L8,7 1 ) −1
,−ε3 (1+ε1 ) −1
, ε3 6= 0, 1 + ε1 6= 0
p L6,3 ⊕ r2, 2ε2 = −1 − ε1 6= 0
L7,3 ⊕ L1 (ε1 , ε2 , ε3) ' .
L ⊕ r ⊕ L ε =
6 0, ε = ε = 0
5,1 2 1 1 2 3
L−ε 3
8,10 , ε4 6= 0, p = 2
Lq ⊕ L ε3 6= 0, εi = 0, i 6= 3
7,3 1
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 437
Of these possibilities, only two contractions arise. The contraction L6,3 ⊕ r2 Lp7,3 is
determined by the endomorphisms
ft (X7) = t−1 X7, ft (X8 ) = pX7 + X8, ft (Xi ) = Xi , i 6= 7, 8.
The other contraction L08,10 ' L7,4 ⊕ L1 L27,3 ⊕ L1 is the same as in proposition 6,
where the additional generators remains unchanged.
9.11. L7,4 ⊕ L1
The cocycle ϕ (X7, X8) = X8 generates the cohomology for the Lie algebra L7,4 ⊕ L1 .
Since there is no integrability condition, for any ε the deformation L7,4 ⊕ L1 + εϕ de-
fines a Lie algebra, which is easily seen to be isomorphic to Lε8,10. By the dimension of
the corresponding derivation algebras, the deformation cannot be reversed and there is no
contraction.
9.12. L7,5 ⊕ L1
The Lie algebra L7,5 ⊕ L1 has also an adjoint cohomology space generated by a unique co-
cycle, which can be chosen as ϕ (X7 , X8) = X8 , as before. For any value of the parameter
ε we find that the deformation L7,5 ⊕ L1 + εϕ isisomorphic
to the indecomposable algebra
Lε8,12. Since dim Der (L7,5 ⊕ L1) = dim Der Lp8,12 = 9, no contraction of Lp8,12 onto
L7,5 ⊕ L1 can exist.
9.13. L7,6 ⊕ L1
Although separately the Lie algebras L7,6 and L1 are stable14 , their direct sum is not, and in
fact it satisfies the equality dim H 2 (L7,6 ⊕ L1 , L7,6 ⊕ L1) = 2, generated by the cocycles
ϕ1 (X4 , X7) = X8 , ϕ1 (X5, X6) = −3X8, ϕ2 (X4 , X8) = X4 ,
ϕ2 (X5 , X8) = X5 , ϕ2 (X6, X8) = X6, ϕ2 (X7 , X8) = X7 .
For a formal deformation (L7,6 ⊕ L1 ) (ε1 , ε2) the integrability condition is ε1 ε2 = 0. It is
trivial to verify that
L8,19, ε1 6= 0
L7,6 ⊕ L1 + ε1 ϕ1 + ε2 ϕ2 ' ,
L8,20, ε2 6= 0
and that both deformations are of plateau type, thus are reversible to give the contractions
L8,19 L7,6 ⊕ L1 and L8,20 L7,6 ⊕ L1 .
9.14. L7,7 ⊕ L1
This Lie algebra, having the same complexification as L7,2 ⊕ L1 , has also a seven dimen-
sional cohomology space. A basis of cocycles can be taken as
ϕ1 (X4, X8) = X4, ϕ1 (X5 , X8) = X5, ϕ2 (X6 , X8) = X4, ϕ2 (X7, X8) = X5,
ϕ3 (X4, X8) = X6, ϕ3 (X5 , X8) = X7, ϕ4 (X6 , X8) = X6, ϕ4 (X7, X8) = X7,
ϕ5 (X4, X5) = X8, ϕ6 (X4 , X7) = X8, ϕ6 (X5, X6) = −X8 , ϕ7 (X6, X7) = X8.
14
This happens because L7,6 , although stable, is not semisimple.
438 R. Campoamor-Stursberg and J. Guerón
For a generic deformation L7,7 ⊕ L1 (ε1 , .., ε7), the integrability constraints are the most
complicated of the studied algebras
ε1 ε5 + ε3 ε6 = 0, ε4 ε7 + ε2 ε6 = 0, ε1 ε6 − ε2 ε5 = 0,
ε3 ε6 − ε4 ε5 = 0, ε4 ε6 − ε3 ε7 = 0, ε2 ε6 − ε1 ε7 = 0,
ε1 ε6 + ε3 ε7 + ε2 ε5 + ε4ε6 = 0.
After cumbersome computations, and ignoring equivalent cases, the possibilities for defor-
mations are resumed in the following scheme:
±1
L8,13, ε5 ε7 6= 0, εi = 0, i 6= 5, 7
L±1
ε6 6= 0, εi = 0, i 6= 6
8,13
8,14 ε2 6= 0, εi = 0, i 6= 2
L ,
L7,7 ⊕ L1 (ε1 , .., ε7) ' L8,15, ε2 ε7 6= 0, εi = 0, i 6= 2, 7 .
L 8,16, ε 1 ε2 ε4 =
6 0, εi = 0, i =
6 1, 2, 4
Lp8,17, ε1 ε4 6= 0, εi = 0, i 6= 1, 4
Lp , ε = ε 6= 0, ε = −ε 6= 0
8,18 1 4 2 3
As could be expected from the structure of L7,7 ⊕L1 , all these deformations can be reversed,
we thus obtain a contraction of any of the target algebras onto the decomposable Lie algebra
L7,7 ⊕ L1.
10. Conclusion
We have determined all deformations of eight dimensional real Lie algebras with a non-
trivial Levi decomposition. Once these deformations were obtained, those being invertible
were computed, and the corresponding contractions of Lie algebras obtained. Due to the
transitivity properties of contractions, the analysis provides the contractions among this
class of Lie algebras. Decomposable algebras were also considered, which provided addi-
tional possibilities and confirmed some properties of contractions and deformations already
commented in [40]. It should be remarked that the same problem for the nine dimensional
case, as classified in [27], is much more complicated, not only because of the high number
of isomorphism classes, but also to the presence of many continuous parameters in several
families. For a satisfactory approach to this dimension, alternative criteria that allow to
reduce the number of cases to be analyzed should be developed. In contrast to dimension
eight, however, the analysis of contractions of semisimple algebras is muss less interesting,
due to the non-existence of simple algebras in dimension nine. Therefore, determining their
contractions essentially reduces to the lower dimensional cases.
In this context, it is worthy to recall those Lie algebras that have appeared as a contrac-
tion of an eight dimensional semisimple Lie algebra, decomposable or not. The contraction
diagram, following the results already obtained in [24], are reproduced in Figure 2.
Acknowledgment
During the preparation of this work, the first author (RCS) was financially supported by the
research project MTM2006-09152 of the M.E.C. and the project and CCG07-UCM/ESP-
2922 of the U.C.M.-C.A.M.
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 439
I2 = x27x−p6
L8,11 I1 = 2x2x3x6 + x25 x2 + x1x4x5 − x3 x24 x−1 1 2
6 + 2 x1 x6
−1
I2 = (2x7 − x6 ln x6 ) x6
p
L8,12 I1 = x25 − 4x4x6 x−2 7
p
I2 = (x1 x5 − 2x3x4 + 2x2 x6 ) x−2 7
Lε8,13 I1 = x8
ε = ±1 I2 = εx28 4x2x3 + x21 + 2x2x8 x27 + εx25 − 2x3x8 εx24 + x26
+2x6 x7 (x1 x8 + x4x5 ) − x24 x27
L8,14 I1 = x4x7 − x5 x6
I2 = x1x4x5 + x5x6x8 − x4 x7x8 + x2x25 − x3 x24
L8,15 I1 = x4x7 − x5 x6 − 12 x28
I2 = x1x4x5 + x5x6x8 − x4 x7x8 + x2x25 − 13 x38 − x3x24
L−1
8,17 I1 = x4x7 − x5 x6
I2 = x1 (x4x7 + x5 x6) + 2x2x5 x7 − 2x3x4x6 + x8 (x4x7 − x5x6 )
L08,18 I1 = x4x7 − x5 x6
I2 = x1 (x4x7 + x5 x6) + x2 x25 + x27 − x3 x24 + x8 (x5 x6 − x4x7 ) − x3 x26
L8,19 I1 = x8
I2 = 12 x2 x3x28 + x2x5 x7x8 − x3x4 x6x8 + 4 x4 x36 + x3 x25x8 + x35x7 − x2x26x8
+18 (x1 x4x7x8 − x4x5x6 x7) + 27x24x27 + 3x21x28 − 2x1x5x6 x8 − x25x26
L8,21 I1 = 3x5 x7 − x26 − 12x4 x8
I2 = −2x36 + 9x5x6x7 + 72x4 x6 x8 − 27x4 x27 − 27x25x8
L8,22 I1 = x25 − 4x4x6
I2 = x4x28 + x6 x27 − x5 x7x8
444 R. Campoamor-Stursberg and J. Guerón
su(3)
? -
L8,5 L7,2 ⊕ L1 L8,2
-
-
L08,4
? - ? ?
-
L−1 L−1 - L8,14 L18,13 L08,18
8,13 8,17
-
---
-
-
L8,21
L7,7 ⊕ L1
References
[1] I. E. Segal, Duke Math. J 18, 221 (1951)
[2] E. Inönü and E. P. Wigner, Proc. Natl. Acad. Sci. U.S. 39, 510 (1953)
[3] E. Saletan, J. Math. Phys. 2, 1 (1961); M. Kupczyński, Comm. Math. Phys. 13, 154
(1969)
[6] Ya. Kh. Lykhmus, Predel’nye gruppy Li, (Institut Fiziki AN Estonskoi’ SSR, Tartu,
1969)
[8] R. Gilmore, Lie Groups, Lie Algebras and Some of Their Applications , (Wiley, N.Y.,
1974)
Deformation and Contraction Schemes for Non-solvable Real Lie Algebras 445
[10] J. F. Cariñena, J. Grabowski and G. Marmo, J. Phys. A: Math. Gen. 34, 3769 (2001)
[16] A. Fialowski and M. Penkava, Int. J. Theor. Phys. 47, 561 (2008)
[19] A. Nijenhuis and R. W. Richardson, Bull. Amer. Math. Soc. 72, 1 (1966)
[20] G. Hochschild and J.-P. Serre, Ann. Math. 57, 591 (1953)
[22] J. A. de Azcárraga, J. M. Izquierdo, M. Picón and O. Varela, Int. J. Theor. Phys. 46,
2738 (2007)
[25] R. W. Richardson and S. Page, Trans. Amer. Math. Soc. 127, 302 (1967)
[30] J. A. de Azcárraga and J. M. Izquierdo, Lie Groups, Lie Algebras, Cohomology and
some Applications to Physics , (Cambridge Univ. Press, Cambridge, 1995)
[31] C. Chryssomalakos and E. Okon, Int. J. Mod. Phys. D 13, 1817 (2004)
[41] L. Abellanas and L. Martinez Alonso, J. of Math Phys 16, 1580 (1975)
[42] J. Patera, R. T. Sharp, P. Winternitz, and H. Zassenhaus, J. of Math. Phys. 17, 986
(1976)
[43] F.J. Herranz, J.C. Perez-Bueno and M. Santander, J. Phys. A: Math. Gen. 31, 5327
(1998)
[45] Y. Folly, Rend. Sem. Fac. Sci. Univ. Cagliari 67, 1 (1997)
[46] A. L. Onishchik, Lectures on real semisimple Lie algebras and their representations ,
(ESI, Zürich, 2004)
In: Lie Groups: New Research ISBN 978-1-60692-389-4
Editor: Altos B. Canterra, pp. 447-483
c 2009 Nova Science Publishers, Inc.
Chapter 16
Abstract
The Ehresmann’s theorem about a Lie structure in the hole automorphism group
of a finite type G-structure on manifold is generalized to orbifolds.
Estimates for dimension of such Lie group are established, depending on strat-
ifications of orbifolds. Particular attention is devoted to affine connected, pseudo-
Riemannian and Riemannian orbifolds. The content is illustrated by examples.
Introduction
Orbifold can be regard as a manifold with singularities. The topological space of n-
dimensional orbifold is locally homeomorphic to a quotient space of Rn by a finite group
Γ of diffeomorphisms of Rn . The group Γ is not fixed and can be changed by passing from
the one chart of an orbifold to an other chart. Orbifolds were introduced by Satake [29].
They were named V -manifolds by him.
Orbifolds appear naturally in many branches of mathematics and mathematical physics.
A theory of deformation quantization on symplectic orbispaces, which include symplectic
orbifolds, is developed in [28]. For example, symplectic reduction often gives rise to orb-
ifolds [22]. Orbifolds are used in string theory [13]. Orbifolds arise in foliation theory as
∗
E-mail address: an bagaev@rambler.ru
†
E-mail address: n.i.zhukova@rambler.ru
448 A.V. Bagaev and N.I. Zhukova
“good“ spaces of leaves [36]. Famous result s of Thurston [33] on classification of closed
3-manifolds use the classification of 2-dimensional orbifolds.
The problem of a finding of conditions guaranteeing existence of Lie structure for trans-
formation group is one of the central problems of differential geometry [19]. Myers and
Steenrod [26] proved that the group of all isometries of a Riemannian manifold is a Lie
group. Kobayashi [18] demonstrated that the group of all conformal transformations of a
Riemannian manifold is a Lie group. The theorem that the automorphism group of a fi-
nite type G-structure on a manifold admits a Lie group structure is due to Ehresmann [14].
Nomizu [27] proved that the group of all affine transformations of a complete affinely con-
nected manifold is a Lie group. Later Hano and Morimoto [17] have received this result
without the assumption of completeness. The isometry groups of pseudo-Riemannian and
Lorentzian manifolds is devoted numerous paper of authors Zimmer [38], D’Ambra and
Gromov [12], Adams and Stuck [1, 2], Zeghib [34, 35].
We generalize the Ehresmann’s theorem, mentioned above, and prove that the automor-
phism group A(N ) of a G-structure of finite type and order m on a smooth n-dimensional
orbifold is a Lie group of dimension at most d(n, g) := n+dim g+dim g1 +. . .+dim gm−1 ,
where gi is the ith prolongation of the Lie algebra g of G; moreover, the group A(N ) ad-
mits a unique topology and a unique smooth structure that makes it into a Lie group (The-
orem 3.1).
The presence of orbifold points is shown to sharply decrease the dimension of the trans-
formation group A(N ) of an orbifold, with the equality dim A(N ) = d(n, g) is possible
only in the case when N is a smooth homogeneous space with transitive action of A(N )
(Theorem 3.1).
We investigate an influence of the existence of k-dimensional stratum ∆k of N on the
dimension of the automorphism group A(N ). It is shown that a G-structure on an orb-
ifold N inducts G-structure on each connected component ∆ck of the stratum ∆k (sub-
section 2.5). We observe that the subgroup A(N , ∆ck ) consisting from automorphisms of
A(N ) which preserve the connected component ∆ck is an open-closed Lie subgroup of the
Lie group A(N ) (Proposition 3.1). Hence, A(N , ∆ck ) has the same dimension as the Lie
group A(N ). Using this observation we get some estimates of the dimension of the Lie
group A(N ) (Theorems 3.3 and 3.4). The specific character of automorphism groups of
G-structures on good orbifolds is indicated (Theorem 3.2).
Some classes of geometric structures on orbifolds are considered: affine connected,
pseudo-Riemannian and Riemannian orbifolds (Section 4).
In particular, we demonstrate that a pseudo-Riemannian metric of signature (p, q) on
a n-dimensional orbifold N induces pseudo-Riemannian metrics on each connected com-
ponent ∆ck and on the closure ∆ck of ∆ck (Proposition 2.9). It is shown that the induced
pseudo-Riemannian metric on ∆cs can have arbitrary signature (k, l) where 0 ≤ k ≤ p,
0 ≤ l ≤ q, s = k + l < n, in general (Examples 4.1 and 4.2). Estimates are established
for the dimension of the isometry group of pseudo-Riemannian orbifolds, depending on the
types of orbifold points (Theorem 4.2).
For each n ≥ 3 we construct n-dimensional compact Lorentzian orbifolds N with
noncompact isometry groups I(N ) and nonproper actions of I(N ) on N (Example 4.3).
In Sections 4 we present some results about the Lie groups of automorphisms of affine
connected orbifolds and the isometry groups of Riemannian orbifolds belong to the au-
The Automorphism Groups of Some Geometric Structures on Orbifolds 449
1. An Introduction to Orbifolds
1.1. The Category of Orbifolds
Throughout this article we understand by smoothness the smoothness of class C ∞ . Given
some smooth mapping of manifolds f : M → N, denote by f∗ and f ∗ the differential and
codifferential of f .
Recall the definition of a smooth orbifold [6, 14]. Let N be a connected Hausdorff
topological space with a countable base, let U be an open subset of N , and let n be a fixed
natural number. An orbifold chart on N is a triple (Ω, Γ, p), consisting of a connected open
subset Ω of the n-dimensional arithmetic space Rn , a finite group Γ of diffeomorphisms of
Ω, and the composition p : Ω → N of the quotient mapping r : Ω → Ω/Γ and a homeomor-
phism q : Ω/Γ → U of the quotient space Ω/Γ onto U. The subset U is called a coordinate
neighborhood of (Ω, Γ, p). Note that, unlike Satake [6], we do not require the dimension of
the fixed-point set FixΓ of Γ to be smaller than n − 1.
Let U and U ′ be coordinate neighborhoods of orbifold charts (Ω, Γ, p) and (Ω′ , Γ′ , p′ ),
with U ⊂ U ′ . An embedding φ : Ω → Ω′ such that p′ ◦ φ = p. is called an embedding of
the orbifold chart (Ω, Γ, p) into the orbifold chart (Ω′ , Γ′ , p′ ) corresponding to the inclusion
U ⊂ U ′ . It is known [15] that each embedding φ induces a (unique) monomorphism of
groups ψ : Γ → Γ′ for which φ ◦ γ = ψ(γ) ◦ φ ∀γ ∈ Γ, and if φ is a diffeomorphism then
ψ is an isomorphism between the groups Γ and Γ′ .
Two orbifold charts (Ω1 , Γ1 , p1 ) and (Ω2 , Γ2 , p2 ) with coordinate neighborhoods U1
and U2 are called compatible if in the case U1 ∩ U2 6= ∅ for each point x ∈ U1 ∩ U2 there
exist: (a) an orbifold chart (Ω, Γ, p) with coordinate neighborhood U such that x ∈ U ⊂
U1 ∩ U2 ; (b) embeddings of orbifold charts φ1 : Ω → Ω1 and φ2 : Ω → Ω2 , corresponding
to inclusions U ⊂ U1 and U ⊂ U2 .
A set A = {(Ωi , Γi , pi ) | i ∈ J} of orbifold charts is called an orbifold atlas if
the family {Ui := pi (Ωi ) | i ∈ J} is an open covering of N and each pair of orbifold
charts in A is compatible. An orbifold atlas A is called maximal if A coincides with every
orbifold atlas that includes it. A maximal orbifold atlas is called the structure of a smooth
n-dimensional orbifold on N . A pair (N , A), where A is a maximal orbifold atlas on N ,
is called a smooth n-dimensional orbifold. Note that each orbifold atlas is included in a
unique maximal orbifold atlas, and thus defines the structure of a smooth orbifold.
Henceforth we assume all orbifolds N to be smooth and denote by A = {(Ωi , Γi , pi ) |
i ∈ J} the maximal atlas of N . The embedding φij of an orbifold chart (Ωi , Γi , pi ) into
an orbfold chart (Ωj , Γj , pj ) corresponding to the inclusion of coordinate neighborhoods
Ui ⊂ Uj , is called an embedding of orbifold charts and is denoted by φij : Ωi → Ωj ,
i, j ∈ J.
Note that the coordinate neighborhood U of an orbifold chart (Ω, Γ, p) which is home-
omorhic to Ω/Γ belongs to the orbifolds. Orbifolds U are called elementary.
For each point x ∈ N there exists an orbifold chart (Ω, Γ, p) ∈ A, such that Ω is a
n-dimensional arithmetic space Rn , p(0) = x, with 0 = (0, . . . , 0) ∈ Rn , and Γ is a finite
450 A.V. Bagaev and N.I. Zhukova
underlying topological space of N is not locally Euclidean. This example shows that, in
contrast to manifolds and 2-dimensional orbifolds, for n ≥ 3 the underlying topological
spaces of n-dimensional orbifolds may be not locally Euclidean.
(a) bi (γ)◦ φ̄ij = φ̄ij ◦bj (ψij (γ)) ∀γ ∈ Γi where ψij : Γi → Γj is a monomorphism
of groups induced by the embedding φij ;
(b) if Ui ⊂ Uj ⊂ Uk with the corresponding embeddings of charts φij and φjk
then φjk ◦ φij = φ̄ij ◦ φ̄jk .
Denote by ξ = {Pi , bi , φ̄ij }i,j∈J the fiber bundle over N described above.
A fiber bundle over an orbifold can be defined starting from an arbitrary atlas; see [30].
For each orbifold N there exists some atlas B = {(Ωβ , Γβ , pβ ) | β ∈ B} with contractible
coordinate neighborhoods of all charts. For such an atlas the fiber bundles Pβ are trivial;
i.e, Pβ = Ωβ × F and πβ : Pβ → Ωβ is the canonical projection onto the first factor.
Let ξ = {Pi , bi , φ̄ij }i,j∈J be a fiber bundle with standard fiber F and structure group H
over some orbifold N . For each chart (Ωi , Γi , pi ) ∈ A the antimonomorphism bi determines
the smooth left action Φi : Γi × Pi → Pi : (γ, z) 7→ bi (γ −1 )(z) of Γi on the manifold Pi .
Since Γi is a finite group, the quotient space P̄i := Γi \Pi is a smooth orbifold of dimension
dim N + dim F, and the diagram
p̄i
Pi −−−−→ P̄i = Γi \Pi
π π̄
y i y i
pi
Ωi −−−−→ Ui
is commutative, where p̄i : Pi → Γi \Pi is the quotient mapping and π̄i : P̄i → Ui F takes the
orbit of z ∈ Pi into pi (πi (z)) ∈ Ui = pi (Ωi ). Denote by P̄ the disjoint union i∈J P̄i .
Define on P̄ the equivalence relation ρ : say that two points z̄i ∈ P̄i and z̄j ∈ P̄j are ρ-
equivalent if: (a) π̄i (z̄i ) = π̄j (z̄j ) = x ∈ Ui ∩Uj ; (b) there exist two points zi ∈ (p̄i )−1 (z̄i ),
zj ∈ (p̄j )−1 (z̄j ) and a chart (Ωk , Γk , pk ) ∈ A with coordinate neighborhood Uk , such that
x ∈ Uk ⊂ Ui ∩ Uj and zj = (φ̄kj )−1 ◦ φ̄ki (zi ). It is easy to check [6] that ρ is indeed an
The Automorphism Groups of Some Geometric Structures on Orbifolds 453
equivalence relation, the quotient space P = P̄ /ρ is naturally equipped with the structure
of a smooth orbifold, and the submersions πi : Pi → Ωi define a submersion π : P → N of
orbifolds.
Thus we have proved
P ROPOSITION 1.1. If a fiber bundle with the standard fiber F and the structure group H
over an orbifold N is given then the smooth orbifold P of dimension dim N + dim F is
naturally defined together with the submersion π : P → N of orbifolds.
D EFINITION 1.2. The orbifold P is called the total space; and the mapping π : P → N is
called the natural projection of the fibre bundle.
Remark that the standard fiber F is diffeomorphic to π −1 (x) for any regular point x ∈
N.
D EFINITION 1.3. A smooth section of a fiber bundle ξ = {Pi , bi , φ̄ij }i,j∈J with the stan-
dard fiber F and the structure group H over an orbifold (N , A) is defined [8] as a family
{si }i∈J of smooth sections si : Ωi → Pi of the fiber bundles Pi if the following conditions
are satisfied:
(a) bi (γ) ◦ si ◦ γ = si ∀γ ∈ Γi , i ∈ J;
2. G-structures on Orbifolds
2.1. Proper Actions of Lie Groups on Orbifolds
is proper;
(ii) if there exist such consequences {gn } ⊂ G and {xn } ⊂ N that xn → x and
xn · gn → y, x, y ∈ N , then {gn } has a convergence subsequence in G;
P ROOF is analogously to the proof of the respectively assertion for manifold ([23, p. 41]).
proposition 5.4 [23] the orbit x·G is an embedded submanifold of ∆k , consequently, x·G is
an embedded submanifold in N . Since (idN , Φ) is a closed mapping as a continuous, proper
mapping between topological spaces N ×G and N ×N , then (idN , Φ)(x, G) = {x}×x·G
is closed in the fiber {x} × N of product N × N . Hence, the orbit x · G is closed in N .
(ii) a locally free proper action of the Lie group H on P is defined, with N = P/H and
the quotient mapping π : P → P/H = N is submersion of orbifolds.
P ROOF. The first statement follows from Proposition 1.1. Define a smooth action of the
Lie group H on the total space P of the principal bundle ξ = {Pi , bi , φ̄ij }i,j∈J . For each
i ∈ J the smooth right action Υi : Pi × H → Pi with (z, h) 7→ z · h where z ∈ Pi ,
h ∈ H, of H on the total space Pi is defined. Since bi (γ), γ ∈ Γi , is an automorphism
of the principal bundle Pi , it follows that bi (γ)(z · h) = (bi (γ)(z)) · h; consequently, the
mapping Ῡi : P̄i × H → P̄i : (z̄, h) 7→ p̄i (z · h), where z̄ ∈ P̄i , z F ∈ p̄−1
i (z̄), h ∈ H,
defines a smooth right action of H on P̄i = Γi \Pi . As above, P̄ = i∈J P̄i . Denote by
q : P̄ → P̄ /ρ = P the natural projection. The composition qi := q ◦ j : P̄i → P of the
inclusion j : P̄i ֒→ P̄ with the projection q is a homeomorphism onto the image. Take
u ∈ P, x = π(u), and some chart (Ωi , Γi , pi ) ∈ A with coordinate neighborhood Ui ∋ x.
The formula Υ(u, h) := qi ◦ p̄i (z · h) where z ∈ (qi ◦ p̄i )−1 (u), h ∈ H, defines a smooth
right action Υ : P × H → P of the Lie group H on orbifold P. The orbit space P/H of
the action Υ is the orbifold N . The following diagram is commutative:
(p̄i , idH ) (qi , idH )
Pi × H −−−−−→ P̄i × H −−−−−→ P × H
Υ
y i yῩi yΥ
p̄i qi
Pi −−−−→ P̄i −−−−→ P
π π̄ π
y i y i y
pi
Ωi −−−−→ Ui ֒→ N.
Since each isotropy group Hu , u ∈ P, is isomorphic to some subgroup of the finite orbifold
group Γ of the point x = π(u), so the action of H on P is locally free.
Show that Υ is a proper action. Denote by [u] the orbit u·H and it will be considered as a
point of orbit space N = P/H. Let {un } and {hn } be sequences in P and H, respectively,
and un → u, un · hn → y, where u, y ∈ P. As the projection π : P → N is continuous
456 A.V. Bagaev and N.I. Zhukova
mapping, so un → u and un · hn → y yield [un ] → [u] and [un ] = [un · hn ] → [y]. The
uniqueness of the limit of a sequence in Hausdorff topological space N implies [u] = [y],
i.e. u · H = y · H, consequently, there exists h0 ∈ H such that y = u · h0 . Since the isotropy
group Hu is isomorphic to same subgroup of the orbifold group Γ of point x = π(u), then
{h ∈ H | u · h = y} = h0 · Hu is a finite subset of H. The continuousness of the ac-
tion Υ of H on P yields the convergence of the sequence u · hn to y. As it is known [3,
Theorem 2], the topology of each group of homeomorphisms of a locally compact Haus-
dorff topological space acting continuously on this space contains all subsets open in the
compact-open topology. Therefore, there exist such n0 ∈ N that the consequence {hn }
belongs to some neighborhood V of h0 · Hu with compact closure V , hence, {hn } has a
convergent subsequence. According to Proposition 2.1, the action Υ is proper.
Using the fact that an action of H keeps the stratification of orbifold N and applying
the theorem about existence of slices for a proper action of a Lie group on a manifold [23,
Theorem 5.7, p. 44–45], we get the following statement.
P ROPOSITION 2.4. Let P be a smooth m-dimensional orbifold and H be a Lie group, with
H effective locally free proper acts on P. Then:
(i) the orbit space N := P/H becames a n-dimensional orbifold by the natural a way,
where n = m − dim H;
(ii) the canonical projection π : P → N = P/H forms a principal bundle over N with
structure group H;
(a) the connected components of the fibers of the bundle π : P → N form a smooth
foliation F of codimension n;
(b) if the Lie group H is connected then the holonomy group of a leaf L = π −1 (x)
∀x ∈ N is isomorphic to the orbifold group Γ of the point x ∈ N .
R EMARK 2.1. Thus, the family ξ = {Pi , bi , φ̄ij }i,j∈J forms a principal bundle over an
orbifold N with structure group H and total space P if and only if an effective locally free
proper action of the Lie group H on P is given, with the orbit space P/H is equal to N .
R EMARK 2.2. Let P be a smooth manifold with effective proper locally free action of the
Lie group H. According to Proposition 2.4, the orbit space N := P/H is a smooth orbifold
and the quotient mapping π : P → P/H = N is a principal bundles over N . Let ∆ck be a
connected component of a stratum ∆k of N . According to Theorem 1.1, ∆ck consists from
orbifold points of the same orbifold type. Denote the orbifold group of points of ∆ck by Γ.
Put Rck := π −1 (∆ck ) and πkc := π|Rck . Then πkc : Rck → ∆ck is a fibre bundle with the
structure group G and the standard fibre G/Γ over manifold ∆ck .
structure group H of the bundle Pi over Ωi is reduced to the subgroup G, Ri is the reduced
bundle, and moreover the conditions
(a) bi (γ)(Ri ) = Ri ∀γ ∈ Γi ;
are satisfied, then the principal bundle ξ ′ = {Ri , bi , φ̄ij }i,j∈J over N with the structure
group G is defined, where φ̄ij means φ̄ij |Rj |φ (Ω ) : Rj |φij (Ωi ) → Ri and bi : Γi → AutRi .
ij i
The principal bundle ξ ′ with structure group G over N is referred to as the reduced bundle.
In this case we say that the structure group H of the bundle ξ is reduced to G.
Let (Ωi , Γi , pi ) be a chart of a n-dimensional orbifold (N , A). We will regard a frame
at a point x ∈ Ωi as a vector space isomorphism z : Rn → Tx Ωi , where Tx Ωi is the tangent
space of Ωi at x. Denote by πi : Pi → Ωi the principal GL(n, R)-bundle of frames over
Ωi . Define an anti-homomorphism bi of Γi into the automorphism group of bundles Pi
as bi (γ)(z) := (γ −1 )∗x ◦ z, where γ ∈ Γi and z is a frame at x ∈ Ωi . For embedding
φij : Ωi → Ωj , i, j ∈ J, define the mapping φ̄ij : Pj |φij (Ωi ) → Pi by the equality φ̄ij (z) :=
(φ−1
ij )∗φij (x) ◦ z where z is a frame at φij (x) ∈ φij (Ωi ) ⊂ Ωj . The so-constructed principal
bundle with structure group GL(n, R) is called the frame bundle over the orbifold N .
If the structure group GL(n, R) of the frame bundle over N is reduced to a Lie subgroup
G ⊂ GL(n, R) then the reduced principal bundle is called a G-structure on the orbifold N .
Let ξ = {Ri , bi , φ̄ij }i,j∈J be a G-structure on an orbifold N , let (Ωi , Γi , pi ) be an
arbitrary chart of N . Fix x ∈ FixΓi , z ∈ πi−1 (x). As bi (γ) = (γ −1 )∗x is an automorphism
of Ri for any γ ∈ Γi , so γ∗x (z) ∈ Ri and, consequently, the linear isomorphism z −1 ◦
γ∗x ◦ z : Rn → Rn belongs to the Lie group G. Straightforward verification shows that the
mapping χz : Γi → G : γ 7→ z −1 ◦ γ∗x ◦ z correctly defines a faithful representation of the
finite group Γi in the Lie group G.
The equality bi (γ)(z) = (γ −1 )∗x ◦ z = z where z ∈ Ri , x = πi (z), γ ∈ Γi , implies
γ∗x = idTx Ωi . As the group Γi is finite, so γ = idΩi . Therefore, the group Γi free acts on
Ri and the total space R of G-structure is a manifold. Thus, applying Proposition 2.3 we
have
P ROPOSITION 2.5. Let R be a G-structure on a smooth n-dimensional orbifold N . Then
R is a smooth manifold of dimension n+dim G and the connected components of the fibers
of the bundle π : R → N constitute a smooth foliation F of codimension n; moreover, if
the Lie group G is connected then the holonomy group of a leaf L = π −1 (x) is isomorphic
to the orbifold group Γ of the point x ∈ N .
Let R be a G-structure on a smooth n-dimensional orbifold N , let g be the Lie algebra
of the Lie group G. Denote by V the smooth distribution on R tangent to the fibres of π.
Given X ∈ g, xt be the global one-parameter subgroup in G generated by X. Then X
d
defines to the vector field X ∗ on R by the formula Xu∗ := dt (u · xt )|t=0 , u ∈ R. The vector
field X is called the fundamental vector field corresponding to X. The vector field X ∗ is
∗
P ROPOSITION 2.6. The connection form ω satisfies the conditions ω(X ∗ ) = X and
(Rg )∗ ω = Ad(g −1 )ω for every X ∈ g, g ∈ G, where Ad is the adjoint representation
of G in g. Conversely, if ω is some g-valued 1-form on R satisfying these conditions then
there is a unique connection H on R whose connection form is ω.
The canonical form θ on R is the Rn -valued 1-form defined as follows: For any Xu ∈
Tu R and every chart (Ωi , Γi , pi ) at π(u) ∈ N , let Yz ∈ Tz Ri be such that (qi ◦ p̄i )∗ (Yz ) =
Xu , where as above p̄i : Ri → Γi \Ri = R̄i is the quotientF mapping and qi := q ◦ j : R̄i →
R is the composition of the inclusion j : R̄i ֒→ R̄ = i∈J R̄i and the quotient mapping
q : R̄ → R̄/ρ = R. Then we put by definition θ(Xu ) := z −1 (πi )∗ (Yz ). It is easy to check
that the value of θ is independent on the choice of the chart (Ωi , Γi , pi ) and z ∈ Ri . The
torsion form Σ is defined to be the exterior covariant differential of the canonical form θ.
(ii) (Rg )∗ θ = g −1 θ, g ∈ G;
(iii) Σ = dθ + ω ∧ θ.
The definition the G-structure on the orbifold N implies that this definition is correct;
i.e., it is independent on the choice of charts at x and f (x) and of the choice of a local lift.
We denote the hole group of automorphisms of a G-structure on an orbifold N by A(N ).
The Automorphism Groups of Some Geometric Structures on Orbifolds 459
According to Theorem 1.1, the connected component ∆ck is formed by points of the
same orbifold type. Therefore, the orbifold groups of all points of ∆ck are isomorphic to the
same finite group Γ. Without loss of generality, we may assume that for each point x ∈ ∆ck
there exists a linearized chart (Ωi = Rk × Rn−k , Γ, pi ), where FixΓ = Rk × {0}. Then
Vi := pi (Rk ×{0}) is an open subset in ∆ck and the pair (Vi , ϕi ), where ϕi = (pi |Rk ×{0} )−1 ,
is a chart of k-dimensional manifold ∆ck at x. We will say that the chart (Vi , ϕi ) is associated
with the chart (Ωi , Γ, pi ). Denote the family of all such charts (Vi , ϕi ) by A(∆ck ).
Let Ri0 = Ωi × G0 be the reduced bundle with structure group G0 . We will regarded
Qi := Vi × H as the principal bundle with structure group H over Vi . The mapping
belongs to gk−1 . The set gk with pointwise addition and multiplication by a number is a
vector space. The Lie bracket in gk is defined by the equality [t, s]k := t ◦ s − s ◦ t. The
Lie algebra gk is called the kth prolongation of the Lie algebra g. The order of g is defined
as the least k for which gk = 0. In this case, gk+l = 0 for every l ∈ N. If gk 6= 0 for every
k = 0, 1, . . ., then g is called an algebra of infinite type.
Let G be an arbitrary Lie subgroup of the group GL(n, R), g be the Lie algebra of G and
gk be the k th prolongation of g. The first prolongation G1 of G is the set of transformations
t̄ of the vector space V + g which are defined by the elements t ∈ g1 in accordance with
the following formulas:
t̄(v) := v + t(·, v), t̄(x) := x
for v ∈ V and x ∈ g.
The kth prolongation Gk of G is defined as the group of all linear transformations t̄ of
the vector space V + g + g1 + . . . + gk−1 , induced by elements t ∈ gk by the following
formulas:
t̄(v) := v + t(·, . . . , ·, v), t̄(x) := x,
for v ∈ V, x ∈ g + . . . + gk−1 , t ∈ gk .
V
Let V ∗ be the dual linear space to V . Then V ⊗ 2 V ∗ can be treated as the space
of all skew-symmetric bilinear mappings from V × V into V and g ⊗ V ∗ can be regarded
as the
V2space of linear mappings from V into g. Define the linear mapping ν : g ⊗ V ∗ →
V⊗ V by the formula (νf )(v1 , v2 ) := f (v1 )v2 −f (v2 )v1 for f ∈ g⊗V ∗ , v1 , v2 ∈ V .
∗
The definition of ν implies that f belongs to the kernel ker ν if and only if the mapping
V × V → V : (v1 , v2 ) 7→ f (v1 )v2 belongs to the first prolongation g1 .
V2 ∗
Fix an arbitrary vector subspace
V2 ∗ C in V ⊗ V , complementary to ν(g ⊗ V ∗ ), i.e.,
satisfying the equality V ⊗ V = ν(g ⊗ V ∗ ) ⊕ C.
Let R be a G-structure on a smooth n-dimensional orbifold N . Suppose that u ∈ R
and Hu is an arbitrary n-dimensional subspace of Tu R, complementary to Vu . We call Hu
a horizontal subspace at u ∈ R. Then the restriction θ|Hu of the canonical form θ on Hu is
a vector space isomorphism Hu → V . Therefore, the restriction of the exterior differential
dθ on Hu × Hu defines some V skew-symmetric bilinear mapping V × V → V , i.e., an
element c(u, Hu ) ∈ V ⊗ 2 V ∗ . If Hu′ is another horizontal subspace at u then, using
relation (iii) of Proposition 2.7, we can easily verify that the difference c(u, Hu′ ) − c(u, Hu )
belongs to ν(g ⊗ V ∗ ).
Each horizontal subspace Hu at Tu R defines some frame at u ∈ R. Indeed, as men-
tioned, the connection form ω gives rise to an isomorphism of g onto the vertical space Vu
and the canonical form θ yields an isomorphism of V on Hu . We thus obtain some iso-
morphism of the vector space V + g onto the tangent space Tu R = Hu ⊕ Vu . We call it
the frame at u corresponding to the subspace Hu . We denote by R1 the set of frames on
R, corresponding to the horizontal subspaces Hu , u ∈ R, such that c(u, Hu ) ∈ C. The
following holds:
P ROPOSITION 3.1. The set R1 is a G1 -structure on R, with G1 the first prolongation of
the group G.
By induction we define the kth prolongation Rk of a G-structure R as the first prolon-
gation of the Gk−1 -structure Rk−1 , i.e., Rk = (Rk−1 )1 .
464 A.V. Bagaev and N.I. Zhukova
homogeneous space.
(iii) the hole group A(N ′ ) of automorphisms of the G-structure R′ is isomorphic to the
quotient group N(Γ)/Γ of the normalizer N(Γ) of Γ in the hole group A(N ) of
automorphisms of the G-structure R.
The Automorphism Groups of Some Geometric Structures on Orbifolds 467
Demonstrate that H(N , ∆ck ) is an open-closed Lie subgroup of the Lie group H. Sup-
pose that f ∈ H(N , ∆ck ) and h : [0, 1] → H is a continuous path in H, with f = h(0). Take
x ∈ ∆ck . Denote the action of the Lie group H on N by Φ. Since the path h is continuous
and the group H smoothly acts on N , then the path h̃ : [0, 1] → N , defined by the equality
h̃(t) := Φ(h(t), x) is continuous. As f preserves ∆ck , i.e., f (∆ck ) = ∆ck , and g(∆k ) = ∆k
for all g ∈ H, so h̃(t) ∈ ∆ck , ∀t ∈ [0, 1]. Therefore, for each t ∈ [0, 1] the automor-
phism h(t) keeps invariant each connected component ∆ck of stratum ∆k . Consequently,
468 A.V. Bagaev and N.I. Zhukova
h(t) ∈ H(N , ∆ck ), ∀t ∈ [0, 1], and the group H(N , ∆ck ) contains whole connected com-
ponent of element f. Remark that the group H(N , ∆ck ) also contains connected component
of identity idN of H. Thus, the group H(N , ∆ck ) consists from connected components of
the Lie group H and, hence, it is an open-closed subgroup Lie of the Lie group H, with
dim H(N , ∆ck ) = dim H. So we obtain
P ROPOSITION 3.1. Let H be an arbitrary Lie group of automorphisms of an orbifold N ,
∆ck be a connected component of a stratum ∆k of N . Then the subgroup H(N , ∆ck ) in H,
consisted from automorphisms of H which preserved ∆ck , is an open-closed Lie subgroup
of the Lie group H.
D EFINITION 3.1. The Lie group
(ii) the images imχ = AN (∆ck ) and imχ̄ = AN (∆ck ) are closed Lie subgroups of the
Lie groups A(∆ck ) and A(∆ck ) respectively;
T HEOREM 3.4. Let R be a G-structure of finite type and order m on a n-dimension orbifold
N . If N has a k-dimensional stratum, then
Since each automorphism of H preserves e-structure on Rm−1 , then the Lie group H freely
acts on Rm−1 and, consequently, on Rcm−1 . It means that each orbit H · u, u ∈ Rcm−1 , is
diffeomorphic to the Lie group H. Then
(b) an embedding φij a chart (Ωi , Γi , pi ) into chart (Ωj , Γj , pj ) with coordinate neighbor-
hood Ui and Uj , Ui ⊂ Uj , satisfies to the equality (φij )∗ (∇iX Y ) = ∇j(φij )∗ X (φij )∗ Y
for all vector fields X, Y on Ωi .
The pair (N, ∇) is called an affinely connected orbifold.
Remark that an affine connection ∇ is given on an orbifold N if and only if a connection
form ω is given on the frame bundle over N .
D EFINITION 4.2. Let (N , ∇) be a affinely connected orbifold. By an automorphism of
(N , ∇) we mean an diffeomorphism f : N → N of the orbifold (N , A) satisfying the
condition: for each point x ∈ N and each pair charts (Ωi , Γi , pi ), (Ωj , Γj , pj ) ∈ A with
coordinate neighborhoods Ui ∋ x and Uj such that f (Ui ) = Uj there exists a local lift
fˆ: Ωi → Ωj fulfilled the equality ∇jˆ fˆ∗ Y = fˆ∗ (∇iX Y ) for all X, Y ∈ X(Ωi ).
f∗ X
(ii) the group A(N , ∇) admits a unique topology and smooth structure which make it
into a Lie group;
The Automorphism Groups of Some Geometric Structures on Orbifolds 471
(iii) the equality dim A(N , ∇) = n2 + n is satisfied if and only if (N , ∇) is the ordinary
affine space with the flat affine connection;
(vi) the equality dim A(N , ∇) = n2 implies ∆k = ∅ for all k ∈ {1, . . . , n − 1}; if
dim A(N , ∇) > n2 , then N is an affinely connected n-dimensional manifold with
zero torsion; if dim A(N , ∇) > n2 and n ≥ 4, then N is the ordinary affine space
with affine connection.
P ROOF. The items (i), (iii), (iv) and (vi) are proved by authors in [6]. For proof of items (ii)
and (v) we repeat some reasoning of proof of Theorem 3 [6]. In [6] some estimates of
dimension of the Lie group A(N , ∇) were obtained. Here we improve these estimates.
Let P be the linear frame bundle over a n-dimensional orbifold N with the natural
projection π : P → N . The affine connection ∇ on N determines a connection form ω on
P. Denote the Lie algebra of the Lie group G = GL(n, R) by g. Fix some Euclidean scalar
products d0 and d1 on the vector spaces Rn and g respectively. Recall that the canonical
form θ and the connection form ω receive the values in Rn and g respectively. Then the
formula d(X, Y ) := d0 (θ(X), θ(Y ))+d1 (ω(X), ω(Y )), where X and Y are smooth vector
fields on the manifold P, defines a Riemannian metric on P.
According to Lemma 3 [6], the diffeomorphism fˆ of P induced by an automorphism
f ∈ A(N , ∇) keeps invariant the connection form ω and the canonical form θ :
The bijection σ forms on the set A(N , ∇) the structure of a smooth manifold. Since
σ is a group isomorphism, with respect to the induced smooth structure A(N , ∇) is a Lie
group.
Receive the estimates (4.1) and (4.2). Let ∆ck 6= ∅, k < n. By Remark 4.1, the affine
connection ∇ induces the affine connections on the connected component ∆ck of the stra-
tum ∆k and on the closure ∆ck of ∆ck . So we can consider the inducted groups AN (∆ck ) and
AN (∆ck ). According to Proposition 3.1, the subgroup A(N , ∆ck ) consisted from automor-
phisms of A(N , ∇) which preserved the connected component ∆ck is an open-closed Lie
subgroup of the Lie group A(N , ∇). Using the topology of the Lie groups A(N , ∆ck ) and
A(∆ck ), it is easy to check that the mapping χ : A(N , ∆ck ) → A(∆ck ) : f 7→ f |∆ck is a ho-
momorphism of the Lie groups, with the image imχ = AN (∆ck ) is a closed Lie subgroup of
the Lie group A(∆ck ). Then dim A(N , ∇) = dim A(N , ∆ck ) ≤ dim AN (∆ck ) + dim ker χ.
Estimate the dimension of the kernel ker χ := {f ∈ A(N , ∆ck ) | f |∆ck = id∆ck }.
Let x be an arbitrary point of ∆ck , let (Ω, Γ, p) be an chart with coordinate neighborhood
U ∋ x such that y = p−1 (x) ∈ FixΓ. Since Γ is finite group of diffeomorphisms of the
manifold Ω, so there exists Riemannian metric g on Ω for which Γ is an isometry group.
In the point y choice a normal coordinate system (y 1 , . . . , y n ) such that (y 1 , . . . , y k ) is
a coordinate system of k-dimensional manifold FixΓ. Then with relation to the selected
coordinate system Jacobi matrix of transformation γ ∈ Γ at the point y has the form
E 0
,
0 C
where E is the unit of the orthogonal group O(k, R) and C ∈ O(n − k, R).
Let f ∈ ker χ. As f is an automorphism of the orbifold N , so the triple (Ω, Γ, p′ :=
f |U ◦ p) is a chart of N with coordinate neighborhood f (U ). By the definition of the
automorphism f of the affinely connected orbifold (N , ∇) a representative f¯: Ω → Ω of f
in the charts (Ω, Γ, p) and (Ω, Γ, p′ ) is automorphism of the affinely connected manifold Ω.
The differential f¯∗y of f¯ we will consider as a linear transformation of Rn , with f¯∗y (0) = 0,
0 ∈ Rn . Since f ∈ ker χ, so f¯|FixΓ = idFixΓ and with relation to the selected normal
coordinates at the point y Jacobi matrix of the mapping f¯ at y has the form
E A
0 B
where B ∈ GL(n − k, R), A is a matrix of size k × (n − k) and E is the unit in the group
GL(k, R).
For any γ ∈ Γ there exists γ ′ ∈ Γ such that f¯ ◦ γ = γ ′ ◦ f¯; then f¯∗y ◦ γ∗y = γ∗y
′ ◦ f¯ .
∗y
Consequently,
E A E 0 E 0 E A
= ,
0 B 0 C 0 C′ 0 B
respectively, where Ek is the unit in the group GL(k, R), En−k is the unit in the
group GL(n − k, R). Put Γ1 = hγ1 | γ12 i ∼ = Z2 , Γ2 = Γ1 × hγ2 | γ22 i ∼ =
Z2 × Z2 . The quotient spaces N1 = R /Γ1 and N2 = Rn /Γ2 are smooth n-dimensional
n
orbifolds with stratifications ∆(N1 ) = {∆n , ∆k } and ∆(N2 ) = {∆n , ∆n−k , ∆k , ∆0 }, re-
spectively. Note that if k > 0 then the stratum ∆k of N2 is not closed in the topology of
N2 and ∆k = ∆k ⊔ ∆0 . Further, we assume that k > 0 for the orbifold N2 .
474 A.V. Bagaev and N.I. Zhukova
The transformations of the groups Γ1 and Γ2 keep the ordinary flat affine connection of
affine space An . Hence, N1 and N2 are affinely connected orbifolds. According to Proposi-
tion 7 [6], the Lie group A(Ni , ∇i ), i = 1, 2, is isomorphic to the factor-group N(Γi )/Γi ,
where N(Γi ) is the normalizer of Γi in the hole group A(An ) of all affine transformations
of the affine space An . The group A(An ) is isomorphic to the semi-direct product of the
linear group GL(n, R) and the translation group Rn , which is a normal subgroup in this
product. Therefore, any transformation of A(An ) can be consider as a pair hA, ai, where
A ∈ GL(n, R), a ∈ Rn , with the multiply in A(An ) is given by the equality
Therefore the group A(N2 , ∇2 ) is isomorphic to the product GL+ (k, R) × GL+ (n − k, R),
where GL+ (k, R) is the group of nondegenerate matrixes with positive determinates. Thus
dim A(N2 , ∇2 ) = k 2 + (n − k)2 = n2 + n − (n − k)(2k + 1) − k.
T HEOREM 4.2. Let I(N ) be the hole isometry group of a n-dimensional pseudo-
Riemannian orbifold N and let ∆(N ) = {∆k }k∈{0,...,n} be the stratification of N . Then:
(i) the group I(N ) is a Lie group of dimension at most n(n + 1)/2, with the action of
the Lie group I(N ) on orbifold N is smooth and nonproper in general;
(ii) the group I(N ) admits a unique topology and smooth structure which make it into
a Lie group;
(iii) the equality dim I(N ) = n(n + 1)/2 implies that pseudo-Riemannian orbifold N
is a homogeneous n-dimensional pseudo-Riemannian manifold;
The Automorphism Groups of Some Geometric Structures on Orbifolds 475
n(n + 1)
dim I(N ) ≤ − (n − k)(k + 1), (4.6)
2
n(n + 1)
dim I(N ) ≤ − (n − k)(k + 1) − k; (4.7)
2
(v) if N is the proper orbifold, then dim I(N ) ≤ n(n − 1)/2; moreover, the equality
dim I(N ) = n(n − 1)/2 yields ∆k = ∅ for all k ∈ {1, . . . , n − 2}.
is an isometry of R(p,q) . Since A2 = E, then the group Γk,l generated by γk,l is isomorphic
to Z2 . The quotient space N = R(p,q) /Γk,l is a pseudo-Euclidean orbifold with the strati-
fication ∆(N ) = {∆n , ∆k+l }. The pseudo-Euclidean metric induced on the stratum ∆k+l
has the signature (k, l).
E XAMPLE 4.2. Let the group Γ of isometries of the n-dimensional pseudo-Euclidean space
R(p,q) , p + q = n, generated by the isometries
∼ (Z2 )n and
where (x1 , . . . , xi−1 , xi , xi+1 , . . . , xn ) ∈ R(p,q) , i = 1, . . . , n. Then Γ =
the quotient space N = R(p,q) /Γ is a pseudo-Euclidean orbifold with the stratification
∆(N ) = {∆n , ∆n−1 , . . . , ∆1 , ∆0 }, and ∆i 6= ∅ ∀i ∈ {0, . . . , n} (see Example 1.7). Let
(k, l) be an arbitrary pair satisfying (4.8). The isometry αk+1 ◦ . . . ◦ αp ◦ αp+l+1 ◦ . . . ◦ αn
coincides with the isometry γk,l ∈ Γk,l given by matrix of form (4.9) (see Example 4.1).
Consequently, the group Γ contains a subgroup Γk,l . Hence the orbifold N has the (k + l)-
dimensional stratum ∆k+l on which the pseudo-Euclidean metric of signature (k, l) is in-
ducted according with Example 4.1.
P ROOF. Let ψ be an arbitrary isometry of I(N , g). Then for any point x ∈ N there exist
charts (Ωi , Γi , pi ), (Ωj , Γj , pj ) ∈ A with coordinate neighborhoods Ui = pi (Ωi ) ∋ x,
Uj = pj (Ωj ) ∋ ψ(x) such that ψ(Ui ) ⊂ Uj and an isometry ψij : Ωi → Ωj of pseudo-
Riemannian manifolds (Ωi , gi ) and (Ωj , gj ) satisfying the equality pj ◦ ψij = ψ ◦ pi . Since
the mapping f depends only on the coordinates of the orbifold L, then the mapping id ×
ψij : Ω′α ×Ωi → Ω′α ×Ωj given by the formula (id×ψ)(y, z) := (y, ψij (z)) ∀(y, z) ∈ Ω′α ×
Ωi is an isometry of the warped product Ω′α ×f¯α Ωi where f¯α : Ω′α → R is a representative
of f in the chart (Ω′α , Γ′α , p′α ) ∈ B. By the definition, it follows that id × ψ is an isometry
of the warped product L ×f N . Thus the mapping ν is correctly defined. It is clear that ν is
an injective group homomorphism.
The warped product L ×f N has the product topology. Using this fact and the
way of introduction of topology in the isometry group of pseudo-Riemannian orbifold
L ×f N , indicated in the proof of Theorem 3.1, we get that the convergence of a sequence
of isometries {ψn } ⊂ I(N , g) to ψ implies the convergence of the sequence {id × ψn } to
{id × ψ}. It means that ν is continuously mapping and consequently ν is a homomorphism
of the Lie groups.
P ROOF. Suppose opposite, let G be the compact Lie group. Since the action of the group
G on N is continuous, so the mapping
α : G → R : h 7→ kh∗x Xk
is continuous. Therefore, the image α(G) is compact in R and hence it is bounded. But ac-
cording to the condition of Lemma the subset {k(f n )∗x Xk | n ∈ Z} of α(G) is unbounded.
The contradiction shows that the Lie group G is a noncompact.
C OROLLARY 4.1. Let G be a Lie group of diffeomorphims of an orbifold product L × N
acting on L × N continuously. If the group G contains a diffeomorphism (id, f ), where f is
an Anosov diffeomorphism of N , then G is noncompact.
P ROOF. Let g be a Riemannian metric on N satisfying to the definition of Anosov dif-
feomorphism f. Consider a Riemannian metric h on L. Let h ⊕ g be a Riemannian met-
ric on L × N . According to definition of f there exist constants c > 0, λ > 0 such
that for any integer m > 0 and x ∈ N , k(f m )∗ Xk ≥ cλm kXk when X ∈ Exu
where Tx N = Exs ⊕ Exu is a corresponding splitting of the tangent space Tx N . Re-
mark that Y = 0 ⊕ X ∈ T(z,x) (L × N ) = Tz L ⊕ Tx N , (z, x) ∈ L × N , with
k(id, f )n∗(z,x) Y k = k(f n )∗x Xk. Consequently, the subset {k(id, f )n∗(z,x) Y k | n ∈ Z} is
unbounded in R. Thus, the all conditions of Lemma 4.1 are satisfied, hence G is noncom-
pact Lie group.
2 0
E XAMPLE
1 −1
4.3. Let (R , g ) be a pseudo-Euclidean plane with the metric defined the matrix
2 → R2 of the plane R2 given by the matrix
−1 −1 . The affine transformation f0 : R
A = ( 52 21 ) is an isometry of (R2 , g 0 ). As f0 ◦Z2 = Z2 ◦f0 , where Z2 is the translation group
of R2 on arbitrary vectors with integer coordinates, so f0 projects to some diffeomorphism √
fA of the torus T 2 = R2 /Z2 . The matrix A has two proper numbers λ1,2 = 3 ± 2 2,
and λ1 > 1, 0 < λ2 < 1. Thus fA is an Anosov diffeomorphism of the torus T 2 . The
invariance of the Lorentzian metric g 0 relatively Z2 admits to define a Lorentzian metric g
on the torus T 2 such that the quotient mapping π : R2 → R2 /Z2 is a local isometry, with
fA is an isometry of (T 2 , g).
Let (L, h) be a compact m-dimensional Riemannian orbifold, m ≥ 2. If m = 1, we
take as (L, h) the segment [0, 1] considered as 1-dimensional Euclidean orbifold with two
orbifold points {0, 1} and with orbifold groups isomorphic to Z2 . Let f : L → R be a
smooth positive function on the manifold L. Then for each m ≥ 1 the warped product
(L × T 2 , h ⊕ f g) is the (m + 2)-dimensional Lorentzian orbifold, with its hole isometry
group I(L × T 2 , h ⊕ f g) contains the isometry {(id, fA )}. According to Corollary 4.1 the
isotropy group of I(L × T 2 , h ⊕ f g) at the point (x, y), where x = π(0, 0), is noncompact,
hence by Corollary 2.1 the action of I(L×T 2 , h⊕f g) on the orbifold L×f T 2 is nonproper.
(i) the group I(N ) endowed with compact-open topology is a Lie group of dimension
at most n(n + 1)/2, with the action of the Lie group I(N ) on orbifold N is smooth
and proper;
(ii) the group I(N ) admits a unique topology and smooth structure which make it into
a Lie group;
(iv) the equality dim I(N ) = n(n + 1)/2 holds if and only if Riemannian orbifold N
is isometric to one of the following n-dimensional Riemannian manifolds of constant
curvature: (a) the Euclidean space En ; (b) the sphere S n ; (c) the projective space
RP n ; (d) the simply connected hyperbolic space Hn ;
n(n + 1)
dim I(N ) ≤ − (n − k)(k + 1), (4.10)
2
moreover, if ∆ck 6= ∆ck , k < n, then
n(n + 1)
dim I(N ) ≤ − (n − k)(k + 1) − k; (4.11)
2
(vi) if N is the proper orbifold, then dim I(N ) ≤ n(n − 1)/2; the equality dim I(N ) =
n(n − 1)/2 implies ∆k = ∅ for all k ∈ {1, . . . , n − 2}; moreover, if in this case
∆n−1 6= ∅, then each connected component ∆cn−1 of the stratum ∆n−1 is the one
of the following (n − 1)-dimensional Riemannian manifold of constant curvature:
(a) the Euclidean space En−1 ; (b) the sphere S n−1 ; (c) the projective space RPn−1 ;
(d) the simply connected hyperbolic space Hn−1 .
The following proposition proves the precision of the estimates (4.10) and (4.11).
P ROPOSITION 4.4. 1. For each pair of integer numbers (n, k), where 0 ≤ k < n, there ex-
ists a n-dimensional Riemannian orbifold N1 having the k-dimensional stratum ∆k which
the hole isometry group I(N1 ) has the dimension n(n+1)
2 − (n − k)(k + 1).
2. For each pair of integer numbers (n, k), where 0 < k < n, there exists a n-
dimensional Riemannian orbifold N2 having the nonclosed k-dimensional stratum ∆k , with
dim I(N2 ) = n(n+1)
2 − (n − k)(k + 1) − k.
P ROOF. Let N1 = Rn /Γ1 and N2 = Rn /Γ2 be smooth n-dimensional orbifolds con-
structed in the proof of Proposition 4.1. The orbifolds N1 and N2 have the stratifications
∆(N1 ) = {∆n , ∆k } and ∆(N2 ) = {∆n , ∆n−k , ∆k , ∆0 } respectively. Remark that the
stratum ∆k of the orbifold N2 for k 6= 0 does not closed in the topology of N2 and
∆k = ∆k ⊔ ∆0 . Since the group Γ1 and Γ2 are isometry group of the Euclidean space
En , than N1 and N2 are flat Riemannian orbifolds. Calculate the isometry groups I(N1 )
and I(N2 ).
According to Theorem 3.2, the group I(Ni ), i = 1, 2, is isomorphic to the quotient
group N(Γi )/Γi of the normalizer N(Γi ) of Γi in the group I(En ) of all isometries of
480 A.V. Bagaev and N.I. Zhukova
the Euclidean space En . The group I(En ) is isomorphic to the semi-direct product of the
orthogonal group O(n, R) and the translation group Rn , with En is a normal subgroup
of this product. Each transformation of I(En ) has the form hA, ai where A ∈ O(n, R),
a ∈ Rn . As hA, ai ∈ N(Γ1 ) if and only if
′
A 0
A= , A′ ∈ O(k, R), A′′ ∈ O(n − k, R),
0 A′′
a = (a1 , . . . , ak , 0, . . . , 0) ∈ Rn ,
′
so the group I(N1 ) is isomorphic to the semi-direct product of the groups G1 := { A0 A0′′ |
A′ ∈ O(k, R), A′′ ∈ SO(n − k, R)} and G2 := {a = (ai ) ∈ Rn | ai = 0, i = k +
1, . . . , n}. Therefore, dim I(N1 ) = k(k−1)
2 + (n−k)(n−k−1)
2 + k = n(n+1)
2 − (n − k)(k + 1).
The group N(Γ2 ) consists from the transformations having the form
′
A 0
A= , A′ ∈ O(k, R), A′′ ∈ O(n − k, R).
0 A′′
Then the group I(N2 ) is isomorphic to SO(k, R) × SO(n − k, R) where SO(k, R),
SO(n − k, R) are the special orthogonal groups. Thus we have dim I(N2 ) = k(k−1)
2 +
(n−k)(n−k−1) n(n+1)
2 = 2 − (n − k)(k + 1) − k.
Let (N , g) be a Riemannian orbifold. Let Si be Ricci tensor of the Riemannian manifold
(Ωi , gi ), (Ωi , Γi , pi ) ∈ A. The definition of the Riemannian metric g = {gi }i∈J implies that
the family S = {Si }i∈J of tensors is a tensor of type (0, 2) on the orbifold N . The tensor
S is called the Ricci tensor of (N , g).
We say that a symmetric bilinear form t = {ti }i∈J on an orbifold N is negative (or
nonpositive) definite at x ∈ N if there is some chart (Ωi , Γi , pi ) ∈ A with coordinate
neighborhood Ui ∋ x such that the form ti is negative (respectively nonpositive) definite at
x0 ∈ p−1 i (x). Conditions (a) and (b) in the definition of a section imply that this definition is
independent of a choice of a chart (Ωi , Γi , pi ) with the coordinate neighborhood Ui ∋ x and
a point x0 ∈ p−1 i (x). We say also that a symmetric bilinear form t is negative (respectively
nonpositive) definite on N if t possesses this property at each x ∈ N .
Using the integration on orbifolds introduced by Satake [30] and some obtained integral
formulas, we receive [7] the following theorem which can be regarded as an analogy of the
well-known Bochner’s theorem [11].
T HEOREM 4.4. If N is a compact Riemannian orbifold with nonpositive definite Ricci
tensor and at some point of N the Ricci tensor is negative definite, then the isometry group
of N is finite.
A Riemannian orbifold (N , g) is said to be a Riemannian orbifold of constant curvature
k ∈ R if the Riemannian manifold (Ωi , gi ) ∀(Ωi , Γi , pi ) ∈ A has constant curvature k. A
Riemannian orbifold of constant curvature k is called hyperbolic (flat, elliptic) if k < 0
(respectively k = 0, k > 0).
C OROLLARY 4.2. The isometry group of every compact hyperbolic orbifold is finite.
In [7] we calculated the isometry groups of some hyperbolic orbifolds.
The Automorphism Groups of Some Geometric Structures on Orbifolds 481
References
[1] Adams, S.; Stuck, G. The isometry group of a compact Lorentz manifold. Invent.
Math. 1997, 129, 239-261.
[2] Adams, S.; Stuck, G. The isometry group of a compact Lorentz manifold, II. Invent.
Math. 1997, 129, 263-287.
[3] Arens, R. A topology for spaces of transformations. Ann. of Math. 1946, 47, 3, 480–
495.
[4] Arouche, A.; Deffaf, M.; Zeghib A. On Loretzian dynamics: from group actions to
warped products via homogeneous spaces. Trans. Amer. Math. Soc. 2007, 359, 3,
12531263
[5] Bagaev, A. V.; Zhukova, N. I. The automorphism groups of finite type G-structures on
orbifolds. Siberian Math. J. 2003. 44, 2, 213224.
[6] Bagaev, A. V.; Zhukova, N. I. Affinely connected orbifolds and their automorphisms.
In Non-Euclidean Geometry in Modern Physics and Mathematics: Proceedings of
the International Conference BGL-4 (Bolyai-Gauss-Lobachevsky) (Nizhny Novgorod,
Sept. 7–11, 2004); Ed.: Jenkovszky L., Polotovkiy G.; Kiev: Bogolubov Institute for
heoretical Physics, 2004. 31–48.
[7] Bagaev, A. V.; Zhukova, N. I. The isometry groups of Riemannain orbifolds. Siberian
Math. J. 2007, 48, 4, 579592.
[8] Baily, W. L. Jr. The decomposition theorem for V -manifolds. Amer. J. of Math. 1956,
78, 4, 862–888.
[9] Beem, J. K.; Ehrlich, P. E. Global Lorentzian Geometry, Dekker, New York, 1981.
[10] Blumenthal, R. A.; Hebda, J. J. Ehresmann connections for foliations. Indiana Univ.
Math. J. 1984, 33, 4, 597–611.
482 A.V. Bagaev and N.I. Zhukova
[11] Bochner, S. Vector fields and Ricci curvature. Bull. Amer. Math. Soc. 1946, 52, 776–
797.
[12] D’Ambra, G.; Gromov, M. Lectures on transformation groups: geometry and dynam-
ics. In Surveys in Differential Geometry (supplement to the Journal of Differential
Geometry), 1, 1991, pp. 19-111.
[13] Dixon, L.; Harwey, J. A.; Vafa, C.; Witten, E. Strings on orbifolds (1). Nucl. Phys. B.
1985, 261, 4, 678–686.
[14] Ehresmann, C. Sur les pseudo-groupes de Lie de type fini. C. R. Acad. Sci. Paris.
1958, 246, 360–362.
[15] Ermolitski, A. A. Riemannian manifolds with geometric structures (Monograph),
Minsk, BSPU, 1998.
[16] Haefliger, A.; Salem, E. Action of tori on orbifolds. Ann. Global Anal. and Geom.
1991, 9, 1, 37–59.
[17] Hano, J.; Morimoto, A. Note on the group of affine transformations of an affinely
connected manifold. Nagoya Math. J. 1955, 88, 71–81.
[18] Kobayashi, S. Groupe de transformations qui laissent invariante une connexion in-
finitesimale. C. R. Acad. Sci. Paris A. 1954, 238, 644-645.
[19] Kobayashi, S. Transformation groups in diffrential geometry, Springer-Verlag, 1972.
[20] Kobayashi, S.; Nomizu, K. Foundations of differential geometry, N. Y., John Wiley
and Sons, 1963; Vol. 1.
[21] Kowalsky, N. Actions of non-compact simple groups on Lorentz manifolds and other
geometric manifolds. Ann. of Math. 1996, 144, 2, 611–640.
[22] Lermann, E.; Tolman, S. Torus actions on symplectic orbifolds and toric varieties.
Trans. Amer.Math. Soc. 1997, 349, 4201–4230.
[23] Michor, P.W. Isometric actions of Lie groups and invariants; Lecture course at the
University of Vienna, 1996/97. Vienna: University of Vienna, 1997.
[24] Moerdijk, I.; Pronk, D. Orbifolds, sheaves and groupoids. K-theory. 1997, 12, 3–21.
[25] Mounoud, P. Dynamical properties of the space of Lorentzian metrics. Com-
ment.Math. Helv. 2003, 78, 463–485.
[26] Myers, S.B.; Steenrod, N. The group of isometries of a riemannian manifold. Ann.
Math. 1939, 40, 400–416.
[27] Nomizu, K. On the group of affine transformations of an affinely connected manifold.
Proc. Amer. Math. Soc. 1953, 4, 816–823.
[28] Pflaum, M.J. On deformation quantization of symplectic orbispaces. Diff. geom. and
its Appl. 2003, 19, 3, 343–368.
The Automorphism Groups of Some Geometric Structures on Orbifolds 483
[29] Satake, I. On a generalization of the notion of manifold. Proc. of the Nat. Ac. of Sci-
ences. 1956, 42, 6, 359–363.
[30] Satake, I. The Gauss-Bonnet theorem for V -manifolds. J. Math. Soc. Japan. 1957, 9,
464–492.
[31] Sjamaar, R.; Lerman, E. Stratified symplectic spaces and reduction. Ann. of Math.
1991, 134, 375-422.
[34] Zeghib, A. The identity component of the isometry group of a compact Lorentz man-
ifold. Duke Math. J. 1998, 92, 321–333.
[35] Zeghib, A. Sur les espaces-temps homogènes. In The Epstein birthday schrift, Geom.
Topol. Monogr., Geom. Topol. Publ., Coventry, 1998; Vol. 1, pp. 551576.
[36] Zhukova, N. On the stability of leaves of Riemannian foliations. Ann. Global Anal.
and Geom. 1987, 5, 3, 261–271.
[38] Zimmer, R.J. On the automorphism group of a compact Lorentz manifold and other
geometric manifolds. Invent. Math. 1986, 83, 411-426.
In: Lie Groups: New Research ISBN 978-1-60692-389-4
Editor: Altos B. Canterra, pp. 485-561
c 2009 Nova Science Publishers, Inc.
Chapter 17
Abstract
This article is devoted to the investigation of wrap groups of connected fiber bun-
dles over the fields of real R, complex C numbers, the quaternion skew field H and
the octonion algebra O. These groups are constructed with mild conditions on fibers.
Their examples are given.
It is shown, that these groups exist and for differentiable fibers have the infinite
dimensional Lie groups structure, that is, they are continuous or differentiable mani-
folds and the composition (f, g) 7→ f −1 g is continuous or differentiable depending
on a class of smoothness of groups. Moreover, it is demonstrated that in the cases of
real, complex, quaternion and octonion manifolds these groups have structures of real,
complex, quaternion or octonion manifolds respectively. Nevertheless, it is proved
that these groups does not necessarily satisfy the Campbell-Hausdorff formula even
locally. Iterated wrap groups are studied as well.
Their smashed products are constructed. Cohomologies of wrap groups and their
structure are investigated. Sheaves of wrap groups are constructed and studied. More-
over, twisted cohomologies and sheaves over quaternions and octonions are investi-
gated as well. CW-groups associated with wrap groups are studied.
1. Introduction
Wrap groups of fiber bundles considered in this paper are constructed with the help of
families of mappings from a fiber bundle with a marked point into another fiber bundle with
a marked point over the fields R, C, H and the octonion algebra O. Conditions on fibers
supplied with parallel transport structures are rather mild here. Therefore, they generalize
geometric loop groups of circles, spheres and fibers with parallel transport structures over
them. A loop interpretation is lost in their generalizations, so they are called here wrap
groups. This paper continues previous works of the author on this theme, where generalized
486 S.V. Ludkovsky
loop groups of manifolds over R, C and H were investigated, but neither for fibers nor over
octonions [24, 32, 30, 31].
Loop groups of circles were first introduced by Lefshetz in 1930-th and then their con-
struction was reconsidered by Milnor in 1950-th. Lefshetz has used the C 0 -uniformity on
families of continuous mappings, which led to the necessity of combining his construc-
tion with the structure of a free group with the help of words. Later on Milnor has used
the Sobolev’s H 1 -uniformity, that permitted to introduce group structure more naturally
[37]. Iterations of these constructions produce iterated loop groups of spheres. Then their
constructions were generalized for fibers over circles and spheres with parallel transport
structures over R or C [14].
Wrap groups of quaternion and octonion fibers as well as for wider classes of fibers
over R or C are defined and investigated here for the first time.
Holomorphic functions of quaternion and octonion variables were investigated in
[28, 29, 26]. There specific definition of super-differentiability was considered, because
the quaternion skew field has the graded algebra structure. This definition of super-
differentiability does not impose the condition of right or left super-linearity of a super-
differential, since it leads to narrow class of functions. There are some articles on quaternion
manifolds, but practically they undermine a complex manifold with additional quaternion
structure of its tangent space (see, for example, [39, 52] and references therein). Therefore,
quaternion manifolds as they are defined below were not considered earlier by others au-
thors (see also [26]). Applications of quaternions in mathematics and physics can be found
in [11, 16, 17, 23].
Fiber bundles and sheaves and cohomologies over quaternions and octonions are inter-
esting in such a respect, that they take into account spin and isospin structures on manifolds,
because there is the embedding of the Lie group U (2) into the quaternion skew field H.
In this article wrap groups of different classes of smoothness are considered. Hence-
forth, we consider not only orientable manifolds M and N , but also nonorientable mani-
folds.
In particular, geometric loop groups have important applications in modern physical
theories (see [20, 34] and references therein). Groups of loops are also intensively used in
gauge theory. Wrap groups defined below with the help of families of mappings from a
manifold M into another manifold N with a dimension dim(M ) > 1 can be used in the
membrane theory which is the generalization of the string (superstring) theory.
Section 2 is devoted to the definitions of topological and manifold structures of wrap
groups. The existence of these groups is proved and that they are infinite dimensional Lie
groups not satisfying even locally the Campbell-Hausdorff formula (see Theorems 3, 6,
12, Corollaries 5, 8, 9 and Examples 10). In the cases of complex, quaternion and octo-
nion manifolds it is proved that they have structures of complex, quaternion and octonion
manifolds respectively.
In Section 3 smashed products of wrap groups are constructed. Iterated wrap groups are
studied as well. Their structure is investigated in more details. The main results of Section
3 are Theorems 2, 6, 9, 10, 20, 21, Propositions 3, 7, 8, 12, 13, 17 and Corollary 11.
Section 4 is devoted to constructions and investigations of cohomologies and sheaves of
wrap groups. Moreover, over quaternions and octonions twisted cohomologies and sheaves
are studied. Twisted analogs of bar resolutions of sheaves and smooth Deligne cohomology
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 487
are investigated as well. This is done over twisted multiplicative groups. Previously the
complex case and with loop groups of fiber bundles on spheres was only studied. The main
results of Section 4 are given in Theorems 34, 36, 44, 48.1, 55, 58, 60, Propositions 6, 14,
15, 19, 26, 27, 29, 32, Corollaries 7, 8, 33, 45 and 47.
In Section 5 a structure of wrap groups as CW-groups is studied.
All main results of this paper are obtained for the first time.
1.2.3. Definition and notation. An Ar -manifold M with corners is defined in the usual
way: it is a metric separable space modelled on X = Anr or X = l2 (Ar ) respectively and
is supposed to be of class C ∞ , 0 ≤ r ≤ 3. Charts on M are denoted (Ul , ul , Ql ), that is,
ul : Ul → ul (Ul ) ⊂ Ql is a C ∞ -diffeomorphism for each l, Ul is open in M , ul ◦ uj −1
is biholomorphic for 1 ≤ r ≤ 3 from the domain uj (Ul ∩ Uj ) 6= ∅ onto ul (Ul ∩ Uj ) (that
is, uj ◦ u−1 and ul ◦ u−1 are holomorphic and bijective) and ul ◦ u−1
j satisfy conditions
l Sj
(i − iii) from §1.2.2, j Uj = M .
A point x ∈ M is called a corner of index j if there exists a chart (U, u, Q) of M with
x ∈ U and u(x) is of index indM (x) = j in u(U ) ⊂ Q. A set of all corners of index
j ≥ 1 is called a border ∂M of M , x is called an inner point of M if indM (x) = 0, so
S
∂M = j≥1 ∂ j M , where ∂ j M := {x ∈ M : indM (x) = j}.
For a real manifold with corners on the connecting mappings ul ◦ u−1 j ∈ C ∞ of real
charts only Condition 1.2.2(i) is imposed.
1.2.4. Terminology. In an Ar -manifold N there exists an Hermitian metric, which in
P
each analytic system of coordinates is the following nj,k=1 hj,k dzj dz̄k , where (hj,k ) is a
positive definite Hermitian matrix with coefficients of the class C ∞ , hj,k = hj,k (z) ∈ Ar ,
z are local coordinates in N .
As real manifolds we shall consider Riemann manifolds.
In accordance with the definition above for internal points of N it is supposed that
they can belong only to interiors of charts, but for boundary points ∂N it may happen that
x ∈ ∂N belongs to boundaries of several charts. It is convenient to choose an atlas such
that ind(x) is the same for all charts containing this x.
1.3.1. Remark. If M is a metrizable space and K = KM is a closed subset in M of
codimension codimR N ≥ 2 such that M \ K = M1 is a manifold with corners over Ar ,
then we call M a pseudo-manifold over Ar , where KM is a critical subset.
Two pseudo-manifolds B and C are called diffeomorphic, if B \ KB is diffeomorphic
with C \ KC as for manifolds with corners (see also [14, 36]).
Take on M a Borel σ-additive measure ν such that ν on M \ K coincides with the
Riemann volume element and ν(K) = 0, since the real shadow of M1 has it.
The uniform space Hpt (M1 , N ) of all continuous piecewise H t Sobolev mappings from
M1 into N is introduced in the standard way [30, 31], which induces Hpt (M, N ) the uniform
space of continuous piecewise H t Sobolev mappings on M , since ν(K) = 0, where R ∋
t ≥ [m/2]+1, m denotes the dimension of M over R, [k] denotes the integer part of k ∈ R,
T
[k] ≤ k. Then put Hp∞ (M, N ) = t>m Hpt (M, N ) with the corresponding uniformity.
For manifolds over Ar with 1 ≤ r ≤ 3 take as Hpt (M, N ) the completion of the family
of all continuous piecewise Ar -holomorphic mappings from M into N relative to the Hpt
uniformity, where [m/2] + 1 ≤ t ≤ ∞. Henceforth we consider pseudo-manifolds with
′
connecting mappings of charts continuous in M and Hpt in M \ KM for 0 ≤ r ≤ 3, where
t′ ≥ t.
1.3.2. Note. Since the octonion algebra O is non-associative, we consider a non-
associative subgroup G of the family M atq (O) of all square q × q matrices with entries in
O. More generally G is a group which has a Hpt manifold structure over Ar and group’s
operations are Hpt mappings. The G may be non-associative for r = 3, but G is supposed
to be alternative, that is, (aa)b = a(ab) and a(a−1 b) = b for each a, b ∈ G.
As a generalization of pseudo-manifolds there is used the following (over R and C
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 489
see [14, 45]). Suppose that M is a Hausdorff topological space of covering dimension
dim M = m supplied with a family {h : U → M } of the so called plots h which are
continuous maps satisfying conditions (D1 − D4):
(D1) each plot has as a domain a convex subset U in Anr , n ∈ N;
(D2) if h : U → M is a plot, V is a convex subset in Alr and g : V → U is an Hpt
mapping, then h ◦ g is also a plot, where t ≥ [m/2] + 1;
(D3) every constant map from a convex set U in Anr into M is a plot;
(D4) if U is a convex set in Anr and {Uj : j ∈ J} is a covering of U by convex sets in
n
Ar , each Uj is open in U , h : U → M is such that each its restriction h|Uj is a plot, then h
is a plot. Then M is called an Hpt -differentiable space.
A mapping f : M → N between two Hpt -differentiable spaces is called differentiable
if it continuous and for each plot h : U → M the composition f ◦ h : U → N is a plot of
N . A topological group G is called an Hpt -differentiable group if its group operations are
Hpt -differentiable mappings.
′ ′
Let E, N , F be Hpt -pseudo-manifolds or Hpt -differentiable spaces over Ar , let also
′
G be an Hpt group over Ar , t ≤ t′ ≤ ∞. A fiber bundle E(N, F, G, π, Ψ) with a fiber
space E, a base space N , a typical fiber F and a structural group G over Ar , a projection
π : E → N and an atlas Ψ is defined in the standard way in §II.1 [47] (see also [14, 36])
′
with the condition, that transition functions are of Hpt class such that for r = 3 a structure
group may be non-associative, but alternative.
Local trivializations φj ◦ π ◦ Ψ−1 t′
k : Vk (E) → Vj (N ) induce the Hp -uniformity in the
′
family W of all principal Hpt -fiber bundles E(N, G, π, Ψ), where Vk (E) = Ψk (Uk (E)) ⊂
X 2 (G), Vj (N ) = φj (Uj (N )) ⊂ X(N ), where X(G) and X(N ) are Ar -vector spaces on
which G and N are modelled, (Uk (E), Ψk ) and (Uj (N ), φj ) are charts of atlases of E and
N , Ψk = ΨE N
k , φj = φj .
If G = F and G acts on itself by left shifts, then a fiber bundle is called the principal
fiber bundle and is denoted by E(N, G, π, Ψ). As a particular case there may be G = A∗r ,
where A∗r denotes the multiplicative group Ar \ {0}. If G = F = {e}, then E reduces to
N.
2. Definitions. Let M be a connected Hpt -pseudo-manifold over Ar , 0 ≤ r ≤ 3
satisfying the following conditions:
(i) it is compact;
(ii) M is a union of two closed subsets over Ar A1 and A2 , which are pseudo-manifolds
and which are canonical closed subsets in M with A1 ∩ A2 = ∂A1 ∩ ∂A2 =: A3 and a
codimension over R of A3 in M is codimR A3 = 1, also A3 is a pseudo-manifold;
(iii) a finite set of marked points s0,1 , ..., s0,k is in ∂A1 ∩ ∂A2 , moreover, ∂Aj are
arcwise connected j = 1, 2;
(iv) A1 \ ∂A1 and A2 \ ∂A2 are Hpt -diffeomorphic with M \ [{s0,1 , ..., s0,k } ∪ (A3 \
Int(∂A1 ∩ ∂A2 ))] by mappings Fj (z), where j = 1 or j = 2, ∞ ≥ t ≥ [m/2] + 1,
m = dimR M such that H t ⊂ C 0 due to the Sobolev embedding theorem [35], where the
interior Int(∂A1 ∩ ∂A2 ) is taken in ∂A1 ∪ ∂A2 .
Instead of (iv) we consider also the case
(iv ′ ) M , A1 and A2 are such that (Aj \ ∂Aj ) ∪ {s0,1 , ..., s0,k } are
C 0 ([0, 1], Hpt (Aj , Aj ))-retractable on X0,q ∩ Aj , where X0,q is a closed arcwise connected
490 S.V. Ludkovsky
limn hn = h such that fn (x) = hn (ηn (x)) for each x ∈ M and n ∈ ω, where ω is a
directed set and convergence is considered in Hpt (M, {X0,q : q = 1, ..., k}; W, y0 ). Hence-
forward in the case 2(iv) we get s0,q instead of X0,q in the case 2(iv ′ ).
Thus there exists the quotient uniform space
Hpt (M, {X0,q : q = 1, ..., k}; W, y0 )/Rt,H =: (S M E)t,H . In view of [41, 42] Dif Hpt (M )
is the group of diffeomorphisms for t ≥ [m/2] + 1. The Lebesgue measure λ in the real
shadow of M̂ by the mapping Ξ induces the measure λΞ on M which is equivalent to ν,
since Ξ is the Hpt -mapping from the compact space onto the compact space, λ(∂ M̂ ) = 0
and Ξ : M̂ \ ∂ M̂ → M is bijective.
Due to Conditions (P 1 − P 5) each element f = Pγ̂,u up to a set QM of measure zero,
ν(QM ) = 0, is given as f ◦ Ξ−1 on M \ QM , where π ◦ f = γ̂, γ̂ = γ ◦ Ξ. Denote f ◦ Ξ−1
also by f . Thus, for each (E, f ) ∈ Hpt (M, {s0,q : q = 1, ..., k}; W, y0 ) the image f (M ) is
compact and connected in E.
Therefore, for each partition Z there exists δ > 0 such that for each partition Z ∗ with
supi inf j dist(Mi , M ∗ j ) < δ and (E, f ) ∈ H t (M, W ; Z), f (s0,q ) = uq , there exists
(E, f1 ) ∈ H t (M, W ; Z ∗ ) with f1 (s0,q ) = uq for each q = 1, ..., k such that f Rt,H f1 ,
where Mi and Mj∗ are canonical closed pseudo-submanifolds in M corresponding to parti-
tions Z and Z ∗ , H t (M, W ; Z) denotes the space of all continuous piecewise H t -mappings
from M into W subordinated to the partition Z such that Z and Z ∗ respect Hpt structure of
M.
Hence there exists a countable subfamily {Zj : j ∈ N} in the family of all partitions Υ
such that Zj ⊂ Zj+1 for each j and limj diamZ ˜ j = 0. Then
(i) str − ind{H (M, {s0,q : q = 1, ..., k}; W, y0 ; Zj ); hZ
t i M
Zj ; N}/Rt,H = (S E)t,H
is separable if N and G are separable, since each space Hpt (M, {s0,q : q =
1, ..., k}; W, y0 ; Zj ) is separable.
The space str − ind{H t (M, {s0,q : q = 1, ..., k}; W, y0 ; Zj ); hZ i
Zj ; N} is complete due
to Theorem 12.1.4 [40], when N and G are complete. Each class of Rt,H -equivalent ele-
ments is closed in it. Then to each Cauchy net in (S M E)t,H there corresponds a Cauchy
net in str − ind{H t (M × [0, 1], {s0,q × e × 0; W, y0 ; Zj × Yj ); hZ i ×Yi
Zj ×Yj ; N} due to the-
orems about extensions of functions [35, 44, 50], where Yj are partitions of [0, 1] with
˜
limj diam(Y j ) = 0, Zj × Yj are the corresponding partitions of M × [0, 1]. Hence
M
(S E)t,H is complete, if N and G are complete.
If f, g ∈ H t (M, X) and f (M ) 6= g(M ), then
(ii) inf ψ∈Dif Hpt (M,{s0,q :q=1,...,k}) kf ◦ ψ − gkH t (M,X) > 0. Thus equivalence classes
< f >t,H and < g >t,H are different. The pseudo-manifold M̂ is arcwise connected. Take
η : [0, 1] → M̂ an Hpt -mapping with η(0) = ŝ0,q and η(1) = ŝ0,k+q , where 1 ≤ q ≤ k.
Choose in M̂ Hpt -coordinates one of which is a parameter along η. Therefore, for each
gq , gk+q ∈ G there exists Pγ̂,u with Pγ̂,u (s0,q ) = y0 × gq and Pγ̂,u (s0,k+q ) = y0 × gk+q
for each q = 1, ..., k. Since E and G are arcwise connected, then N is arcwise connected
and (S M E)t,H is locally connected for dimR N > 1. Thus, the uniform space (S M E)t,H
is non-discrete.
The tangent bundle T Hpt (M, E) is isomorphic with Hpt (M, T E), where T E is the
′
Hpt −1 fiber bundle, t′ ≥ t+1. There is an infinite family of fα ∈ Hpt (M, T E) with pairwise
S
distinct images in T E for different α such that fα (M ) is not contained in β<α fβ (M ),
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 493
manifolds over the same Ar such that φ(s1,0,q ) = s2,0,a(q) for each q = 1, ..., k1 , where
{sj,0,q : q = 1, ..., kj } are marked points in Mj , j = 1, 2, 1 ≤ a ≤ k2 , l1 ≤ k2 ,
l1 := card φ({s1,0,q : q = 1, ..., k1 }). Then there exists an induced homomorphism of
monoids φ∗ : (S M2 E)t,H → (S M1 E)t,H . If l1 = k2 , then φ∗ is the embedding.
Proof. Take Ξ1 : M̂1 → M1 with marked points {ŝ1,0,q : q = 1, ..., 2k1 } as in §2, then
take M̂2 the same M̂1 with additional 2(k2 −l1 ) marked points {ŝ2,0,q : q = 1, ..., 2k3 } such
that ŝ1,0,q = ŝ2,0,q for each q = 1, .., k1 , k3 = k1 + k2 − l1 , then φ ◦ Ξ1 := Ξ2 : M̂2 → M2
is the desired mapping inducing the parallel transport structure from that of M1 . Therefore,
each γ̂2 : M̂2 → N induces γ̂1 : M̂1 → N and to Pγ̂2 ,u2 there corresponds Pγ̂1 ,u1 with
additional conditions in extra marked points, where u1 ⊂ u2 . The equivalence class <
(E2 , Pγ̂2 ,u2 ) >t,H ∈ (S M2 E)t,H gives the corresponding elements < (E1 , Pγ̂1 ,u1 ) >t,H ∈
(S M1 E)t,H , since Dif Hpt (M̂1 , {ŝ0,q : q = 1, ..., 2k2 }) ⊂ Dif Hpt (M̂1 , {ŝ0,q : q =
1, ..., 2k3 }). Then φ∗ (< (E2 , Pγ̂2 ,u2 ) ∨ (E1 , Pη̂2 ,v2 ) >t,H ) = φ∗ (< (E2 , Pγ̂2 ,u2 ) >t,H
)φ∗ (< (E1 , Pη̂2 ,v2 ) >t,H ), since f2 ◦ φ(x) for each x ∈ Ξ1 (M̂1 \ ∂ M̂1 ) coincides with
f1 (x), where fj corresponds to Pγj ,y0 ×e (see also the beginning of §3).
If l1 = k2 , then M̂1 = M̂2 and the group of diffeomorphisms Dif Hpt (M̂1 , {ŝ0,q :
q = 1, ..., 2k1 }) is the same for two cases, hence φ∗ is bijective and inevitably φ∗ is the
embedding.
6. Theorems. 1. There exists an alternative topological group (W M E)t,H containing
the monoid (S M E)t,H and the group operation of Hpl class with l = t′ − t (l = ∞ for
t′ = ∞). If N and G are separable, then (W M E)t,H is separable. If N and G are
complete, then (W M E)t,H is complete.
2. If G is associative, then (W M E)t,H is associative. If G is commutative, then
(W E)t,H is commutative. If G is a Lie group, then (W M E)t,H is a Lie group.
M
group (W M N )t,H . Algebraically it is the quotient group F/B, where F is the free com-
mutative group generated by (S M N )t,H , while B is the minimal closed subgroup in F
generated by all elements of the form [f + g] − [f ] − [g], f and g ∈ (S M N )t,H , [f ]
denotes the element in F corresponding to f (see also about such abstract Grothendieck
construction in [?, 48]).
By the construction each point in (S M N )t,H is the closed subset, hence (S M N )t,H is
the topological T1 -space. In view of Theorem 2.3.11 [12] the product of T1 -spaces is the
T1 -space. On the other hand, for the topological group G from the separation axiom T1 it
follows, that G is the Tychonoff space [12, 43]. The natural mapping η : (S M N )t,H →
(W M N )t,H is injective. We supply F with the topology inherited from the topology of the
P
Tychonoff product (S M N )Z t,H , where each element z in F has the form z = f nf,z [f ],
P
nf,z ∈ Z for each f ∈ (S M N )t,H , f |nf,z | < ∞. By the construction F and F/B
are T1 -spaces, consequently, F/B is the Tychonoff space. In particular, [nf ] − n[f ] ∈ B,
hence (W M N )t,H is the complete topological group, if N and G are complete, while η
is the topological embedding, since η(f + g) = η(f ) + η(g) for each f, g ∈ (S M N )t,H ,
η(e) = e, since (z + B) ∈ η(S M N )t,H , when nf,z ≥ 0 for each f , and inevitably in the
general case z = z + − z − , where (z + + B) and (z − + B) ∈ η(S M N )t,H .
′
Using plots and Hpt transition mappings of charts of N and E(N, G, π, Ψ) and equiv-
alence classes relative to Dif Hpt (M, {s0,q : q = 1, ..., k}) we get, that (W M E)t,H has the
structure of the Hpt -differentiable manifold, since t′ ≥ t.
The rest of the proof and the statements of Theorems 6(1-4) follows from this and
Theorems 3(1-3) and [30, 31]. Since (S M E)t,H is infinite dimensional due to Theorem
3.3, then (W M E)t,H is infinite dimensional.
7. Definition. The (W M E)t,H = (W M,{s0,q :q=1,...,k} E; N, G, P)t,H from Theorem
6.1 we call the wrap group.
8. Corollary. There exists the group homomorphism h : (W M E)t,H →
C (Hpt (M, {s0,q : q = 1, ..., k}; N, y0 ), Gk ).
0
−1
Proof follows from §6 and putting hf (γ) = (hf (γ))−1 .
9. Corollary. If M1 and M2 and φ satisfy conditions of Corollary 5, then there exists a
homomorphism φ∗ : (W M2 E)t,H → (W M1 E)t,H . If l1 = k2 , then φ∗ is the embedding.
10. Remarks and examples. Consider examples of M which satisfy sufficient condi-
tions for the existence of wrap groups (W M E)t,H . Take M , for example, DR n , S n \ V with
R
n n n n
s0 ∈ ∂V , DR \Int(Db ) with s0 ∈ ∂Db and 0 < b < R < ∞, where SR denotes the sphere
of the dimension n > 1 over R and radius R, V is Hpt -diffeomorphic with the interior
n ) of the n-dimensional ball D n := {x ∈ Rn : Pn 2
Int(DR R k=1 xk ≤ R} or in
P
n dimensional
over R subspace in Ar and is the proper subset in SR := {x ∈ Rn+1 : n+1
l n 2
k=1 xk = R}.
Instead of sphere it is possible to take an Hpt pseudo-manifold Qn homeomorphic with a
sphere or a disk, particularly, Milnor’s sphere. Indeed, divide M by the equator {x1 = 0}
into two parts A1 and A2 and take A3 = {x ∈ M : x1 = 0} ∪ P , where s0 ∈ ∂A1 ∩ ∂A2 ,
while P = ∅, P = ∂V , P = ∂Dbn correspondingly. Then take also V and Dbn such that
their equators would be generated by the equator {x1 = 0} in SR n or D n respectively or
R
more generally Q . n
S
Take then M = Qn \ lk=1 Vk , where Vk are Hpt -diffeomorphic to interiors of bounded
quadrants in Rn or in n dimensional subspace in Aar , where l > 1, l ∈ N, ∂Vk ∩ ∂Vj =
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 497
A1 and A2 are retractable into a point. For L take M̂ = I 2 , then take a connected curve
η̂ consisting of the left side {0} × [0, 1] joined by a straight line segment joining points
{0, 1} and {1, 0} and then joined by the right side {1} × [0, 1]. This gives the proper cut-
ting of M̂ which induces the proper cutting of L and of M with A3 ⊃ η ∪ ∂L up to an
Hpt diffeomorphism, where η := Ξ(η̂), hence the Möbius band L and M satisfy Conditions
2(i − iii, iv ′ , v).
Take a quotient mapping φ : I 2 → S 1 such that φ({s0,1 , s0,2 }) = s0 ∈ S 1 ,
s0,1 = (0, 0), s0,2 = (0, 1) ∈ I 2 , where I = [0, 1], hence there exists the embedding
1 2
φ∗ : (W S ,s0 E)t,H ֒→ (W I ,{s0,1 ,s0,2 } E)t,H .
The Klein bottle K has M̂ = I 2 with twisting equivalence relation on ∂I 2 so it satis-
fies sufficient conditions. Moreover, K is the quotient φ : Z → K of the cylinder Z with
twisted equivalence relation of its ends S 1 using reflection relative to a horizontal diam-
eter. Thus A3 ⊃ φ(S 1 ). Therefore, there exists the embedding φ∗ : (W K,{s0 } E)t,H →
(W Z,{s0,1 ,s0,2 } E)t,H , where s0,1 , s0,2 ∈ ∂Z, φ({s0,1 , s0,2 }) = s0 .
Take a pseudo-manifold Qn Hpt -diffeomorphic with S n for n ≥ 2, cut from it β non-
intersecting open domains V1 , ..., Vβ Hpt -diffeomorphic with interiors of bounded quadrants
Ta1 +...+aq
in Rn , s0,q ∈ j=a 1 +...+aq−1 +1
∂Vj , a0 := 0, a1 + ... + ak = β, q = 1, ..., k. Then glue
for V1 , ..., Vl , 1 ≤ l ≤ β, by boundaries of slits Hpt -diffeomorphic with S m−1 the reduced
498 S.V. Ludkovsky
Therefore, the slit along C of T2 is the non-orientable band which inevitably is the Möbius
band with twice larger number of marked points {sL 0,j : j = 1, ..., 2k} ⊂ ∂L.
Therefore, for M = T2 as M̂ take a quadrant in R2 with 2k pairwise opposite
marked points ŝ0,q and ŝ0,q+k on the boundary of M̂ , q = 1, ..., k, k ∈ N. Suit-
able gluing of boundary points in ∂ M̂ gives the mapping Ξ : M̂ → T2 , Ξ(ŝ0,q ) =
Ξ(ŝ0,q+k ) = s0,q , q = 1, ..., k. Proper cutting of M̂ into Âj , j = 1, 2, or of L in-
duces that of T2 . Thus we get a pseudo-submanifold A3 (T2 ) =: A3 ⊃ C, while
A1 and A2 are retractable into a marked point s0,q ∈ C for each q, hence T2 sat-
isfies Conditions 2(i − iii, iv ′ , v). In view of Corollary 9 there exists the embedding
L
φ∗ : (W T2 ,{s0,q :q=1,...,k} E)t,H → (W L,{s0,q :q=1,...,2k} E)t,H , where φ : L → T2 is the
quotient mapping with φ({sL L
0,q , s0,q+k }) = {s0,q }, q = 1, ..., k.
For the n-dimensional torus Tn in Aar with n > 2 take a n − 1-dimensional surface
B such that each its projection into T2 is Hpt -diffeomorphic with C for a loop C as above.
Therefore, the slit along B up to a Hpt -diffeomorphism gives M0 := L × I n−2 for even
n or M0 := S 1 × I n−1 for odd n, where I = [0, 1]. Since I m is retractable into a point,
where m ≥ 1. Thus we lightly get for Tn a pseudo-submanifold A3 ⊃ B and two A1
and A2 retractable into points and satisfying sufficient Conditions 2(i − iii, iv ′ , v), where
M̂ = I n up to a Hpt -diffeomorphism, s0,q ∈ B ⊂ A3 := A3 (Tn ), {sM M0
0,q , s0,q+k } ⊂ ∂M0 ,
0
M0
embedding φ∗ : (W Tn ,{s0,q :q=1,...,k} E)t,H ֒→ (W M0 ,{s0,q :q=1,...,2k} E)t,H due to Corollary
9.
More generally cut from Tn open subsets Vj which are Hpt diffeomorphic with in-
teriors of bounded quadrants in Rn embedded into Alr , j = 1, ..., β, such that s0,q ∈
Ta1 +...+aq
B ∩ ( j=a 1 +...+aq−1 +1
∂Vj ), Vj ∩ Vi = ∅ for each j 6= i, Vj ∩ B = ∅ for each j, where
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 499
M1
This gives the embedding φ∗ : (W M2 ,{s0,q :q=1,...,k} E)t,H ֒→ (W M1 ,{s0,q :q=1,...,2k} E)t,H .
Another example is M3 obtained from the previous M2 with 2k marked points and 2β
cut out domains Vj , when s0,q is identified with s0,q+k and each ∂Vj is glued with ∂Vj+β
for each j ∈ λq ⊂ {d : a1 + ... + aq−1 + 1 ≤ d ≤ a1 + ... + aq }, q = 1, ..., k,
k ∈ N, by an equvalence relation υ. Such M3 is obtained from the torus Tn,m with
m holes instead of one hole in the standard torus Tn,1 = Tn cutting from it Vj with
S
j ∈ {1, ..., 2β} \ ( q=1,...,k λq ), where m = m1 + ... + mk , mq := card(λq ). For Tn
and M2 the surface B is Hpt diffeomorphic with (∂L) × I n−2 for even n or S 1 × I n−1
S
for odd n. Take A3 ⊃ B ∪ ( j∈λq υ(∂Vj )), it is arcwise connected and contains all
marked points. Therefore, M3 satisfies conditions of §2 and there exists the embedding
M3 M2
υ ∗ : (W M3 ,{s0,q :q=1,...,k} E)t,H ֒→ (W M2 ,{s0,q :q=1,...,2k} E)t,H . This also induces the em-
Tn,m Tn
bedding (W Tn,m ,{s0,q :q=1,...,k}
E)t,H ֒→ (W Tn ,{s0,q :q=1,...,2k−1} E)t,H such that each el-
Tn,m
ement g ∈ (W Tn,m ,{s0,q :q=1,...,k}
E)t,H can be presented as a product g = (..(g1 g2 )...gm )
Tn ,{sT n :q=1,...,2k−1}
of m elements gj ∈ (W 0,q E)t,H , gj =< fj >t,H , supp(π ◦ fj ) ⊂ Bj ,
B1 ∪ ... ∪ Bm = Tn , Bi ∩ Bj = ∂Bi ∩ ∂Bj for each i 6= j, each Bj is a canonical closed
subset in Tn , s0,1 ∈ B1 , s0,2q , s2q+1 ∈ Bd for m1 + ... + m0 + 1 ≤ d ≤ m1 + ... + mq ,
q = 1, ..., k − 1, where m0 := 0.
Evidently, in the general case for different manifolds M and N wrap groups may be
non isomorphic. For example, as M1 take a sphere S n of the dimension n > 1, as M2 take
M1 \ K, where K is up to an Hpt -diffeomorphism the union of non intersecting interiors
Bj of quadrants of diameters d1 , ..., ds much less, than 1, K = B1 ∪ ... ∪ Bl , l ∈ N. Let
N be a δ-enlargement for M2 in Rn+1 relative to the metric of the latter Euclidean space,
where 0 < δ < min(d1 , ..., dl )/2. Then the groups (W M1 N )t,H and (W M2 N )t,H are not
isomorphic. This lightly follows from the consideration of the element b :=< f >t,H ∈
(W M2 N )t,H , where f : M2 → N is the identity embedding induced by the structure of the
δ-enlargement.
Recall, that for orientable closed manifolds A and B of the same dimension m the
degree of the continuous mapping f : A → B is defined as an integer number deg(f ) ∈ Z
such that f∗ [A] = deg(f )[B], where [A] ∈ Hm (A) or [B] ∈ Hm (B) denotes a generator,
defined by the orientation of A or B respectively [5]. Consider mappings fj : S n → N
such that Vj ⊃ ∂Bj ∩ N , where Vj is a domain in Rn+1 bounded by the hyper-surface
fj (Bj ), fj is w0 on each Bi with i 6= j, while the degree of the mapping fj from S n onto
fj (S n ) is equal to one. If there would be an isomorphism θ : (W M2 N )t,H → (W M1 N )t,H ,
then θ(b) would have a non trivial decomposition into the sum of non canceling non zero
additives, which is induced by mappings fj : S n → N . Nevertheless, an element b in
(W M2 N )t,H has not such decomposition.
If two groups G1 and G2 are not isomorphic, then certainly (W M E; N, G1 , P)t,H and
500 S.V. Ludkovsky
f , g and w0 in the considered wedge products, where < f ∨ w0 >=< w0 ∨ f >=< f >
is the same class of equvalent elements.
Since G = {e}, (ab)−1 = b−1 a−1 , then g(1 − x, y) ∨ f (1 − x, y) is in the same
class of equivalent elements as g(x, 1 − y) ∨ f (x, 1 − y). But due to inclusions (1, 2)
< g(1 − x, y) ∨ f (1 − x, y) >=< f (x, y) ∨ g(x, y) >−1 and < f (x, y) ∨ g(x, y) >=<
g(x, 1 − y) ∨ f (x, 1 − y) >−1 and < h(x, y) >−1 =< h(x, 1 − y) >=< h(1 − x, y) > for
h ∈ ab, consequently, < h(x, y) >=< h(1 − x, 1 − y) > and < (f ∨ g)(x, 1 − y) >=<
f (x, 1 − y) ∨ g(x, 1 − y) >∈ (ab)−1 , since (x, y) 7→ (1 − x, 1 − y) interchange two spheres
in the wedge product S 2 ∨ S 2 . Hence a−1 b−1 = b−1 a−1 and inevitably ab = ba.
12. Theorem. Let M and N be connected both either C ∞ Riemann or Ar holomorphic
manifolds with corners, where M is compact and dimM ≥ 1 and dimN > 1. Then
(W M N )t,H has no any nontrivial continuous local one parameter subgroup g b for b ∈
(−ǫ, ǫ) with ǫ > 0.
Proof. Suppose the contrary, that {g b : b ∈ (−ǫ, ǫ)} with ǫ > 0 is a local non-
trivial one parameter subgroup, that is, g b 6= e for b 6= 0. Then to g δ for a marked
0 < δ < ǫ there corresponds f = fδ ∈ Hp∞ such that < f >t,H = g δ , where f ∈ Hpt . If
f (U ) = {y0 × e} for a sufficiently small connected open neighborhood U of s0,q in M ,
then there exists a sequence f ◦ ψn in the equivalence class < f >t,H with a family of dif-
feomorphisms ψn ∈ Dif Hpt (M ; {s0,q : q = 1, ..., k}) such that limn→∞ diamψn (U ) = 0
T
and ∞ n=1 ψn (U ) = {s0,q }. If h(x) 6= y0 , then in view of the continuity of h there exists
an open neighborhood P of x in M such that y0 ∈ / h(P ). Consider the covariant differen-
tiation ∇ on the manifold M (see [22]). The set Sh of points, where ∇k h is discontinuous
is a submanifold of codimension not less than one, hence of measure zero relative to the
Riemann volume element in M . For others points x in M , x ∈ M \ Sh , all ∇k h are
continuous.
Take then open V = V (f ) in M such that V ⊃ U and ∇kν f |∂V 6= 0 for some k ∈ N,
where ∇ν f (x) := limz→x,z∈M \V ∇ν f (z), ν is a normal (perpendicular) to ∂V in M at a
point x in the boundary ∂V of V in M . Practically take a minimal k = k(x) with such
property. Since M is compact and ∂V := cl(V ) ∩ cl(M \ V ) is closed in M , then ∂V
is compact. The function x 7→ k(x) ∈ N is continuous, since f and ∇l f for each l are
continuous. But N is discrete, hence each ∂q V := {x ∈ ∂V : k(x) = q} is open in V .
Therefore, ∂V is a finite union of ∂q V , 1 ≤ q ≤ qm , where qm := maxx∈∂V k(x) < ∞
for f = fδ , since ∂V is compact. Thus, there exists a subset λ ⊂ {1, ..., qm } such that
S
∂V = q∈λ ∂q V and ∂q V 6= ∅ for each q ∈ λ. If ∇l f (x) = 0 for l = 1, ..., k(x) − 1 and
∇k(x) f (x) 6= 0, then ∇k(x) f (ψ(y)) = ∇k(x) (ψ(y)).(∇ψ(y))⊗k(x) 6= 0 for y ∈ M such
that ψ(y) = x, since ∇ψ(y) 6= 0, where ψ ∈ Dif Hp∞ (M ; {s0,q : q = 1, ..., k}).
We can take ǫ > 0 such that {g b : b ∈ (−ǫ, ǫ)} ⊂ U , where U = −U is a connected
symmetric open neighborhood of e in (W M N )t,H . Since g b1 + g b2 = g b1 +b2 for each
b1 , b2 , b1 + b2 ∈ (−ǫ, ǫ), then limt→0 g b = e for the local one parameter subgroup and in
particular limm→∞ g 1/m = e, where m ∈ N. Take δ = δm = 1/m and f = fm ∈ Hp∞
such that < fm >t,H = g 1/m . On the other hand, jg 1/m = g j/m for each j < mǫ, j ∈ N,
hence fj/m (M ) = f1/m (M ) for each j < mǫ, since f ∨ h(M ∨ M ) = f (M ) ∨ h(M ) and
using embedding η of (S M N )t,H into (W M N )t,H .
k(x)
The function |∇ν fδ (x)| for x ∈ ∂V is continuous by δ due to the Sobolev embedding
502 S.V. Ludkovsky
k(x)
theorem [35], 0 < δ < ǫ, consequently, inf x∈∂V |∇ν fδ (x)| > 0, since ∂V is compact.
We can choose a family fδ such that z (l) (δ, x) := ∇l fδ (x) is continuous for each 0 ≤ l ≤
k0 by (δ, x) ∈ (−ǫ, ǫ) × M , since {g b : b ∈ (−ǫ, ǫ)} is the continuous by b one parameter
subgroup, where k0 := qm (δ0 ). Therefore, for this family there exists a neighborhood
[−ǫ+c, ǫ−c] such that δ0 ∈ [−ǫ+c, ǫ−c] ⊂ (−ǫ, ǫ) with 0 < c < ǫ/3 such that qm (δ) ≤ k0
for each δ ∈ [−ǫ+c, ǫ−c] with a suitable choice of V (fδ ), since N is discrete. On the other
k(x) k(x)
hand, supx∈∂V (fδ ),0<δ≤ǫ−c |∇ν fδ (x)| ≤ supx∈M,0<δ≤ǫ−c |∇ν fδ (x)| =: B < ∞,
since M and [−ǫ + c, ǫ − c] are compact.
Therefore, for this family there exists a neighborhood [−ǫ + c, ǫ − c] such that δ0 ∈
[−ǫ + c, ǫ − c] ⊂ (−ǫ, ǫ) with 0 < c < ǫ/3 such that qm (δ) ≤ k0 for each δ ∈ [−ǫ + c, ǫ − c]
with a suitable choice of V (fδ ), since N is discrete.
k(x)
Then limδ→0,δ>0 |∇ν fδ (x)| =: b > 0 for x ∈ ∂V with a suitable choice of V =
V (fδ ), since M is connected, dimM ≥ 1 and inf m∈N diamfj/m (M ) > 0 for a marked
δ0 = j/m0 < ǫ with j, m > m0 ∈ N mutually prime, (j, m) = 1, (j, m0 ) = 1. To <
fl/m >t,H there corresponds < f1/m >t,H ∨...∨ < f1/m >t,H =:< f1/m >∨l t,H which is the
k(x)
l-fold wedge product. Thus there exists C = const > 0 for M such that |∇ν fl/m (x)| ≥
k(y)
Cl inf y∈∂V (f1/m ) |∇ν f1/m (y)| ≥ Clb, where C > 0 is fixed for a chosen atlas At(M )
with given transition mappings φi ◦ φ−1j of charts.
Consider δ0 ≤ l/m < ǫ−c and m and l tending to the infinity. Then this gives B ≥ Clb
for each l ∈ N, that is the contradictory inequality, hence (W M N )t,H does not contain any
non trivial local one parameter subgroup.
1, ..., k}; W, y0 ) := Hpt (M, {s0,q : q = 1, ..., k}; Hpt,l (M, {s0,q : q = 1, ..., k}; W, y0 )),
Hpt,1 (M, {s0,q : q = 1, ..., k}; W, y0 ) := Hpt (M, {s0,q : q = 1, ..., k}; W, y0 ). Therefore,
there exist iterated wrap semigroups and groups (S M E)l+1;t,H := (S M (S M E)l;t,H )t,H
and (W M E)l+1;t,H := (W M (W M E)l;t,H )t,H , where (S M E)1;t,H := (S M E)t,H and
(W M E)1;t,H := (W M E)t,H .
′
Evidently, if there are Hpt and Hpt diffeomorphisms ρ : M → M1 and η : N →
N1 mapping marked points into respective marked points, then Hpt (M, W ) is isomorphic
with Hpt (M1 , W1 ) and hence (W M E)b;t,H is Hpt isomorphic as the Hpt -manifold and Hpl -
isomorphic as the Hpl -Lie group with (W M1 E1 )b;t,H for each b ∈ N, where l = t′ − t,
l = ∞ for t′ = ∞, t′ ≥ t ≥ [dim(M )/2] + 1. If f : N → N1 is a surjective map and N is
an Hpt -differentiable space, then N inherits a structure of an Hpt -differentiable space with
plots having the local form f ◦ ρ : U → N1 , where ρ : U → N is a plot of N .
′
5. Lemma. Let E be an Hpt principal bundle and let D be an everywhere dense subset
in N such that for each y ∈ D there exists an open neighborhood V of y in N and a
differentiable map p : V → Hpt (M, {s0,q : q = 1, ..., k}; V, y) := {f ∈ Hpt (M ; V ) :
f (s0,q ) = y, q = 1, ..., k} such that êv ŝ0,q (p̂(y)) = y for each q = 1, ..., 2k and each
y ∈ N , where p ◦ Ξ = π ◦ p̂. Then êv : Hpt (M ; W ) → N k is an Hpt differentiable principal
(S M E)t,H bundle.
Proof. Let {(Vj , yj ) : j ∈ J} be a family such that yj ∈ Vj ∩ D for each j and
there exists pj : Vj → Hpt (M, {s0,q : q = 1, ..., k}; Vj , yj ) so that p̂j (ŝ0,q )(y) = y × e
for each q = 1, ..., 2k and every j, where {Vj : j ∈ J} is an open covering of N , y is a
constant mapping from M̂ into Vj with y(M̂ ) = {y}, where p̂j (ŝ0,q ) is the restriction to
Vj of the projection p̂(ŝ0,q ) : (P M E)t,H → E, while pj (Ξ(x̂))(y) = π ◦ p̂j (x̂)(y × e) for
each y ∈ N and x = Ξ(x̂) in M , where x̂ ∈ M̂ , Ξ : M̂ → M . Then (W M E)t,H and
(P M E)t,H are supplied with the Hpt -differentiable spaces structure (see Remark 4 above
and Theorem 6 in Section 2), where the embedding (S M E)t,H ֒→ (P M E)t,H and the
projection êv ŝ0,q : (P M E)t,H → N are Hpt -maps.
Let ψj ∈ Dif Hpt (N ) such that ψj (y) = yj . Specify a trivialization φj :
p̂−1 M
j (ŝ0,q )(Vj ) → Vj × (S E)t,H of the restriction p̂j (ŝ0,q )|Vj of the projection p̂j (ŝ0,q ) :
(P M E)t,H → E by the formula φj (f ) = (f (ŝ0,q ), ψj ◦ p̂j (ŝ0,q )(f )) for each f ∈
(P M E)t,H with π ◦ f (ŝ0,q ) = y, where ψj ◦ p̂j (f ) = ψj (p̂j (f )). Then φ−1 j (y, g) =
−1 M
g (ψj ◦ p̂j (y)) =: η, η ∈ (P E)t,H with π ◦ ψj ◦ f (ŝ0,q ) = yj , since G is a group,
where g = ψj ◦ p̂j (f ). Finally the combination of the family {êv ŝ0,q : q = k + 1, ..., 2k}
induce the mapping êv : Hpt (M ; W ) → N k . By the construction a fiber of this bundle is
the monoid (S M E)t,H .
6. Theorem. If N is a smooth manifold over Ar (holomorphic for 1 ≤ r ≤ 3
respectively), then there exists an Hpt -differentiable principal (S M E)t,H bundle êv :
(P M E)t,H → N k .
Proof. In view of Lemma 5 it is sufficient to prove that for each y ∈ N there exists a
neighborhood U of y in N and an Hpt -map pq : U → Hpt (M, W ) such that evs0,q (pq (z)) =
z for each q = 1, ..., k, z ∈ U , where evx (f ) = f (x).
In M̂ consider a rectifiable curve ζq : [0, 1] → M̂ joining ŝ0,q with ŝ0,q+k , where
1 ≤ q ≤ k. Then consider a coordinate system (x1 , ..., xm ) in M̂ such that x1 corresponds
to a natural coordinate along ζq . This coordinate system is defined locally for each chart of
506 S.V. Ludkovsky
g(s, < f >t,H ) ⊂< g(s, f ) >t,H for each s ∈ [0, 1], since f ∈ gs−1 (< g(s, f ) >t,H )
and g is continuous while < g(s, f ) >t,H by its definition is closed in Hpt (M, W ),
where gs (z) := g(s, z). Therefore, id = g(0, ∗) : (P M E)t,H → (P M E)t,H and
g(1, (P M E)t,H ) =< w0 >t,H .
8.1. Notation. Denote by Homtp ((W M E)t,H , G) or Homtp ((S M E)t,H , G) the group
or the monoid of Hpt differentiable homomorphisms from (W M E)t,H or (S M E)t,H re-
spectively into G. By A∗r is denoted the multiplicative group of Ar \ {0}, where 0 ≤ r ≤ 3.
′
9. Theorem. Let Dif Hpt (N ) acts transitively on N , t ≤ t′ . For each H ∞ manifold
N and an Hpt differentiable group G such that A∗r ⊂ G with 1 ≤ r ≤ 3 there exists
a homomorphism of the Hpt differentiable space of all equivalence classes of (P M E)t,H
′
relative to Dif Hpt (N ) (see §§1.3.2 and 3 in Section 2) and Homtp ((S M E)t,H , Gk ). They
are isomorphic, when G is commutative.
Proof. Mention that due to Theorem 6 the Hpt -differentiable principal (S M E)t,H bun-
dle êv : (P M E)t,H → N k has a parallel transport structure P̂γ̂,uz (x) = P̂γ̂,u (x)z for each
x ∈ M̂ and all γ ∈ Hpt (M, N ) and u ∈ êv −1 (γ(s0,k )) and every z ∈ G and the corre-
sponding γ̂ : M̂ → N such that γ ◦ Ξ = γ̂. If x = ŝ0,q with 1 ≤ q ≤ k, then P̂ gives the
identity homomorphism from (S M E)t,H into (S M E)t,H . If θ : (S M E)t,H → Gk is an Hpt
differentiable homomorphism, then the holonomy of the associated parallel transport P̂θ on
the bundle (P M E)t,H ×θ G → N k is the homomorphism θ : (S M E)t,H → Gk (see §2.3
in Section 2). At the same time the group G contains continuous one-parameter subgroups
from A∗r , where 1 ≤ r ≤ 3. If g ∈ (W M N )t,H and g 6= e, then g is of infinite order, since
w0 does not belong to g n for each n 6= 0 non-zero integer n, where w0 (M ) = {y0 }.
This holonomy induces a map h : (P M E)t,H /Q → Homtp ((S M E)t,H , Gk ), where
′
Q is an equivalence relation caused by the transitive action of Dif Hpt (N ) such that
(S M E)t,H with distinct marked points either {s0,q : q = 1, ..., k} in M and y0 or ỹ0 in
′
N are isomorphic, since there exists ψ ∈ Dif Hpt (N ) such that ψ(y0 ) = ỹ0 .
If G is commutative, then this map is the homomorphism, since (S M E)t,H is the
commutative monoid for a commutative group G (see Theorem 3.2 in Section 2) and
uPγ̂1 ,v1 (x1 )Pγ̂2 ,v2 (x2 ) = uPγ̂2 ,v2 (x2 )Pγ̂1 ,v1 (x1 ) for each x1 , x2 ∈ M̂ and u, v1 , v2 ∈
Ey0 . There is the embedding (S M E)t,H ֒→ (W M E)t,H , hence a homomorphism θ :
(W M E)t,H → Gk has the restriction on (S M E)t,H which is also the homomorphism.
For G ⊃ A∗r there exists a family of f ∈ Homtp ((S M E)t,H , Gk ) separating ele-
ments of the wrap monoid (S M E)t,H , hence there exists the embedding of (S M E)t,H into
Homtp ((S M E)t,H , Gk ). The bundle (P M E)t,H ×θ G → N k has the induced parallel trans-
port structure Pθ . The holonomy of the parallel transport structure on (P M N )t,H ×θ G →
N k is θ. Therefore, the map Hpt ((S M E)t,H , Gk ) ∋ θ 7→ Pθ is inverse to h.
10. Theorems. Suppose that M2 ֒→ M1 and M = M1 \ (M2 \ ∂M2 ) and M̂2 ֒→ M̂1
and M̂ = M̂1 \(M̂2 \∂ M̂2 ) and N2 ֒→ N1 are Hpt -pseudo-manifolds with the same marked
points {s0,q : q = 1, ..., k} for M1 and M2 and M and y0 ∈ N2 satisfying conditions of
§2 in Section 2 and G2 is a closed subgroup in G1 with a topologically complete principal
fiber bundle E with a structure group G1 .
1. Then (W M2 ,{s0,q :q=1,...,k} E; N2 , G2 , P)t,H has an embedding as a closed subgroup
into (W M1 ,{s0,q :q=1,...,k} E; N1 , G1 , P)t,H .
508 S.V. Ludkovsky
Proof. The manifold M has marked points {s0,q : q = 1, ..., k} such that s0,q cor-
responds to s0,q,1 glued with s0,q,2 in the bunch M1 ∨ M2 for each q = 1, ..., k, where
s0,q,j ∈ Mj are marked points j = 1, 2. Since each Mj satisfies Conditions 2.2(i − v) in
Section 2, then M satisfies them also.
In view of Theorem 10.1 (W Mj ,{s0,q :q=1,...,k} N )t,H has an embedding as a closed
subgroup into (W M,{s0,q :q=1,...,k} N )t,H for j = 1, 2. If γj ∈ Hpt (Mj , {s0,q : q =
1, ..., k}; N, y0 ) for j = 1, 2, then γ1 ∨ γ2 ∈ Hpt (M, {s0,q : q = 1, ..., k}; N, y0 ). On
the other hand, each γ ∈ Hpt (M, {s0,q : q = 1, ..., k}; N, y0 ) has the decomposition
γ = γ1 ∨ γ2 , where γj = γ|Mj for j = 1, 2. Therefore, < γ >t,H =< γ1 ∨ w0,2 >t,H ∨ <
w0,1 ∨ γ2 >t,H , where w0 (M ) = {y0 }, w0,j = w0 |Mj for j = 1, 2, hence (W M N )t,H is
isomorphic with (W M1 N )t,H ⊗ (W M2 N )t,H .
13. Propositions. 1. Let θ : N1 → N be an embedding with θ(y1 ) = y0 , or F : E1 →
E be an embedding of principal fiber bundles over Ar such that π ◦ F |N1 ×e = θ ◦ π1 ,
then there exist embeddings θ∗ : (W M N1 )t,H → (W M N )t,H and F∗ : (W M E1 )t,H →
(W M E)t,H .
2. If θ : N1 → N and F : E1 → E are a quotient mapping and a quotient ho-
momorphism such that N1 is a covering pseudo-manifold of a pseudo-manifold N , then
(W M N )t,H is the quotient group of some closed subgroup in (W M N1 )t,H and (W M E)t,H
is the quotient group of some closed subgroup in (W M E1 )t,H .
′
3. If there are an Hpt diffeomorphism f1 : M → M1 and an Hpt -isomorphism f2 :
E → E1 , then wrap groups (W M1 E1 )t,H and (W M E)t,H are isomorphic.
Proof. 1. If γ1 ∈ Hpt (M, {s0,q : q = 1, ..., k}; N1 , y1 ), then θ◦γ1 = γ ∈ Hpt (M, {s0,q :
q = 1, ..., k}; N, y0 ), < γ >t,H = θ∗ < γ1 >t,H , where θ∗ < γ1 >t,H := {θ◦f : f Rt,H γ1 }.
In addition F |E1,v gives an embedding F : G1 → G, where G1 and G are structural groups
of E1 and E. Therefore, for the parallel transport structures we get
(1) F ◦ P1γ̂1 ,v (x) = Pγ̂,u (x)
for each x ∈ M̂ , where F (v) = u, π ◦ F = θ ◦ π1 , where P1 is for E1 and P for E.
Define F∗ < P1γ̂1 ,v >t,H := {F ◦ g : gRt,H P1γ̂1 ,v }. Since θ and F are Hpt differentiable
mappings, then θ∗ and F∗ are embeddings of Hpt manifolds and group homomorphisms of
Hpl differentiable groups (see also Theorems 6 in Section 2).
2. If γ ∈ Hpt (M, {s0,q : q = 1, ..., k}; N, y0 ), then there exists γ1 ∈ Hpt (M, {s0,q : q =
1, ..., k}; N1 , y1 ) such that θ ◦ γ1 = γ, since N1 is a covering of N , that is each y ∈ N
has a neighborhood Vy for which θ−1 (Vy ) is a disjoint union of open subsets in N1 for
each y ∈ N . This γ1 exists due to connectedness of M and γ(M ), where γ(M ) ⊂ N . To
each parallel transport in E1 there corresponds a parallel transport in E so that Equation (1)
above is satisfied. Put θ∗−1 < γ >t,H = {< γ1 >t,H : θ ◦ γ1 = γ} and F∗−1 < Pγ̂,u >t,H :=
{< P1γ̂1 ,v >t,H : F ◦ P1γ̂1 ,v = Pγ̂,u }, where F (v) = u.
This gives quotient mappings θ∗ and F∗ from closed subgroups θ∗−1 (W M N )t,H
and F∗−1 (W M E)t,H in (W M N1 )t,H and (W M E1 )t,H respectively onto (W M N )t,H and
(W M E)t,H by closed subgroups θ∗−1 (e) and F∗−1 (e) correspondingly.
3. We have that g ∈ Hpt (M, {s0,q : q = 1, ..., k}; W, y0 ) if and only if f2 ◦ g ◦ f1−1 ∈
t
Hp (M1 , {s0,q,1 : q = 1, ..., k; W1 , y1 ), where f1 (s0,q ) = s0,q,1 for each q = 1, ..., k,
f2 (y0 × e) = y1 × e. At the same time ψ ∈ Dif Hpt (M ) if and only if f1 ◦ ψ ◦ f1−1 ∈
Dif Hpt (M1 ). Hence (S M E)t,H is isomorphic with (S M1 E1 )t,H and inevitably wrap
510 S.V. Ludkovsky
groups (W M E)t,H and (W M1 E1 )t,H are Hpt diffeomorphic as manifolds and isomorphic
as Hpl groups.
14. Note. If N is a manifold not necessarily orientable, then it contains up to equiv-
alence of atlases a connected chart V open in N such that y ∈ V and V is orientable.
Since (W M E|V )t,H is the infinite dimensional group, then (W M E)t,H is also infinite di-
mensional even if N is not orientable due to Proposition 13.1. If N is not orientable, then
there exists an orientable covering manifold N1 and a quotient mapping θ : N1 → N as
in Proposition 13(2) (see also about coverings and orientable coverings in §§50, 51 [43],
§§II.4.18,19 [7]).
It is necessary to mention that some circumstances of wrap groups are related also with
their infinite dimensionality.
15. Note. Let G be a topological group not necessarily associative, but alternative:
(A1) g(gf ) = (gg)f and (f g)g = f (gg) and g −1 (gf ) = f and (f g)g −1 = f for each
f, g ∈ G
and having a conjugation operation which is a continuous automorphism of G such that
(C1) conj(gf ) = conj(f )conj(g) for each g, f ∈ G,
(C2) conj(e) = e for the unit element e in G.
If G is of definite class of smoothness, for example, Hpt differentiable, then conj is
supposed to be of the same class. For commutative group in particular it can be taken the
identity mapping as the conjugation. For G = A∗r it can be taken conj(z) = z̃ the usual
conjugation for each z ∈ A∗r , where 1 ≤ r ≤ 3.
Suppose that
(A2) Ĝ = Ĝ0 i0 ⊕ Ĝ1 i1 ⊕ ... ⊕ Ĝ2r −1 i2r −1 such that G is a multiplicative group of
a ring Ĝ with the multiplicative group structure, where Ĝ0 , ..., Ĝ2r −1 are pairwise isomor-
phic commutative associative rings and {i0 , ..., i2r −1 } are generators of the Cayley-Dickson
algebra Ar , 1 ≤ r ≤ 3 and (yl il )(ys is ) = (yl ys )(il is ) is the natural multiplication of any
pure states in G for yl ∈ Gl . For example, G = (A∗r )n and Ĝ = Anr .
16. Lemma. If G and K are two topological or differentiable groups twisted over
{i0 , ..., i2r −1 } satisfying conditions 15(A1, A2, C1, C2) and K is a closed normal sub-
group in G, where 2 ≤ r ≤ 3, then the quotient group is topological or differentiable and
twisted over {i0 , ..., i2r −1 }.
Proof. Since Ĝ = Ĝ0 i0 ⊕ Ĝ1 i1 ⊕ ... ⊕ Ĝ2r −1 i2r −1 , where Ĝ0 , ..., Ĝ2r −1 are
pairwise isomorphic, then Ĝ/K̂ = (Ĝ0 /K̂0 )i0 ⊕ ... ⊕ (Ĝ2r −1 /K̂2r −1 )i2r −1 is also
twisted. Each Ĝj is associative, hence G/K is alternative, since 2 ≤ r ≤ 3 and us-
ing multiplicative properties of generators of the Cayley-Dickson algebra Ar . On the
other hand, conj(K) = K, hence conj(gK) = Kconj(g) = conj(g)K ∈ G/K
and conj(ghK) = conj(gh)K = (conj(h)conj(g))K = (conj(h)K)(conj(g)K) =
conj(hK)conj(gK) = conj(gKhK).
The subgroup K is closed in G, hence by the definition of the quotient differentiable
structure G/K is the differentiable group (see also §§1.11, 1.12, 1.15 in [45]).
′ ′
17. Proposition. Let η : N1 → N2 be an Hpt -retraction of Hpt manifolds, N2 ⊂ N1 ,
η|N2 = id, y0 ∈ N2 , where t′ ≥ t, M is an Hpt manifold, E(N1 , G, π, Ψ) and
′
E(N2 , G, π, Ψ) are principal Hpt bundles with a structure group G satisfying conditions of
§2 in Section 2. Then η induces the group homomorphism η∗ from (W M E; N1 , G, P)t,H
onto (W M E; N2 , G, P)t,H .
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 511
Proof. In view of Proposition 7(1) the wrap group (W M E; N1 , G, P)t,H is the princi-
pal Gk bundle over (W M N1 )t,H . Extend η to ϑ : E(N1 , G, π, Ψ) → E(N2 , G, π, Ψ) such
that π2 ◦ ϑ = η ◦ π1 and pr2 ◦ ϑ = id : G → G, where pr2 : Ey → G is the projection,
y ∈ N1 . If f ∈ Hpt (M, N1 ), then η ◦ f := η(f (∗)) ∈ Hpt (M, N2 ). If f (s0,q ) = y0 , then
η(f (s0,q )) = y0 , since y0 ∈ N2 . Since N2 ⊂ N1 , then Hpt (M, N2 ) ⊂ Hpt (M, N1 ). The
parallel transport structure P is over the same manifold M .
Put η∗ (< Pγ̂,u >t,H ) =< Pη◦γ̂,u >t,H , where γ̂ : M̂ → N1 . In view of Theorems
2.3 and 2.6 in Section 2 η∗ (< Pγ̂1 ,u ∨ Pγ̂2 ,u >t,H = η ∗ (< Pγ̂1 ,u >t,H )η∗ (< Pγ̂2 ,u >t,H
), and we can put η∗ (q −1 ) = [η∗ (q)]−1 , consequently, η∗ is the group homomorphism.
Moreover, for each g ∈ (W M E; N2 , G, P)t,H there exists q ∈ (W M E; N1 , G, P)t,H such
that η∗ (q) = g, since γ : M → N2 and N2 ⊂ N1 imply γ : M → N1 , while the structure
group G is the same, hence η∗ is the epimorphism.
18. Definition. Let G be a topological group satisfying Conditions 15(A1, A2, C1, C2)
such that G is a multiplicative group of the ring Ĝ, where 1 ≤ r ≤ 2. Then define the
smashed product Gs such that it is a multiplicative group of the ring Ĝs := Ĝ ⊗l Ĝ, where
l = i2r denotes the doubling generator, the multiplication in Ĝ ⊗l Ĝ is
(1) (a + bl)(c + vl) = (ac − v ∗ b) + (va + bc∗ )l for each a, b, c, v ∈ Ĝ, where v ∗ =
conj(v).
A smashed product M1 ⊗l M2 of manifolds M1 , M2 over Ar with dim(M1 ) =
dim(M2 ) is defined to be an Ar+1 manifold with local coordinates z = (x, yl), where
x in M1 and y in M2 are local coordinates.
Its existence and detailed description are demonstrated below.
19. Proposition. The ring Ĝs has a multiplicative group Gs containing all a + bl 6= 0
with a, b ∈ Ĝ. If Ĝ is a topological or Hpt differentiable ring over Ar for t ≥ dim(G) + 1,
then Ĝs is a topological or Hpt differentiable over Ar+1 ring.
Proof. For each 1 ≤ r ≤ 2 the group G is associative, since the generators
{i0 , ..., i2r −1 } form the associative group, when r ≤ 2. An element a + bl ∈ Ĝs is non-
zero if and only if (a + bl)(a + bl)∗ = aa∗ + bb∗ 6= 0 due to 15(A1, A2, C1, C2) and
18(1). For a + bl 6= 0 put u = (a∗ − lb∗ )/(aa∗ + bb∗ ), where aa∗ + bb∗ ∈ G0 , hence
u(a + bl) = (a + bl)u = 1 ∈ G0 , since Gj is commutative for each j = 0, ..., 2r − 1,
where Gj denotes the multiplicative group of the ring Ĝj . For r ≤ 2 the family of gen-
erators {i0 , ..., i2r+1 −1 } forms the alternative group, hence Ĝs = Ĝ0 i0 ⊕ ... ⊕ Ĝ2r+1 −1 is
alternative, where Ĝj are isomorphic with Ĝ0 for each j.
If an addition in Ĝ is continuous, then evidently (a + bl) + (c + ql) = (a + c) + (b + q)l
is continuous. If the multiplication in Ĝ is continuous, then Formula 18(1) shows that the
multiplication in Ĝs is continuous as well.
We have the decomposition Ar+1 = Ar ⊕ Ar l. If Ĝ is Hpt differentiable, then from
the definition of plots it follows, that Ĝs is Hpt differentiable over Ar+1 (see also in details
20(1 − 5)).
20. Theorem. Let M1 , M2 and N1 , N2 be Hpt manifolds over Ar with
1 ≤ r ≤ 2, and let G be a group satisfying Conditions 15(A1, A2, C1, C2),
let also M1 ⊗l M2 , N1 ⊗l N2 be smashed products of manifolds and Gs be
a smashed product group (see Proposition 19), where dim(M1 ) = dim(M2 ),
dim(N1 ) = dim(N2 ), t ≥ max(dim(M1 ), dim(N1 ), dim(G)) + 1. Then the wrap
group (W M1 ⊗l M2 ;{s0,j,1 ⊗l s0,v,2 :j=1,...,k1 ;v=1,...,k2 } E; N1 ⊗l N2 , Gs , Ps )t,H is twisted over
512 S.V. Ludkovsky
=<< Pγ̂a ,ua ;a >∞,H;a ⊗ < Pγ̂b ,ub ;b >∞,H;b >∞,H;a+b << Pη̂a ,ua ;a >∞,H;a ⊗ <
Pη̂b ,ub ;b >∞,H;b >∞,H;a+b
= θ(< Pγ̂a ,ua ;a >∞,H;a , < Pγ̂b ,ub ;b >∞,H;b )θ(< Pη̂a ,ua ;a >∞,H;a , < Pη̂b ,ub ;b >∞,H;b ).
Thus θ is the group homomorphism.
The mapping Hp∞ (M a , N ) × Hp∞ (M b , N ) ∋ (γa × γb ) 7→ (γa , γb ) ∈ Hp∞ (M a+b , N 2 )
is of Hp∞ class. The multiplication in Gv is Hp∞ for each v ∈ N, since it is such in G, since
the multiplication in Gv is (a1 , ..., av ) × (b1 , ..., bv ) = (a1 b1 , ..., av bv ), where Gv is the v
times direct product of G, a1 , ..., av , b1 , ..., bv ∈ G.
The iterated wrap group (W M E)l;t,H for the bundle E is the principal Gkl bundle over
the iterated commutative wrap group (W M N )l;t,H for the manifold N , since the number
of marked points in M l is kl, where E is the principal G bundle on the manifold N , l ∈ N.
Thus the iterated wrap group is associative or alternative if such is G. In view of Proposition
7 and Remark 4 the homomorphism θ is of Hp∞ class. From the wrap monoids it has the
natural Hp∞ extension on wrap groups.
If G is associative, then
< Pγ̂,u;a+b+v >∞,H;a+b+v =<< (< Pγ̂a ,ua ;a >∞,H;a ⊗ < Pγ̂b ,ub ;b >∞,H;b
) >∞,H;a+b ⊗ < Pγ̂v ,uv ;v >∞,H;v >∞,H;a+b+v
=<< Pγ̂a ,ua ;a >∞,H;a ⊗(< Pγ̂b ,ub ;b >∞,H;b ⊗ < Pγ̂v ,uv ;v >∞,H;v ) >∞,H;a+b+v =
θ(θ(< Pγ̂a ,ua ;a >t,H;a , < Pγ̂b ,ub ;b >t,H;b ), < Pγ̂v ,uv ;v >t,H;v )
θ(< Pγ̂a ,ua ;a >t,H;a , θ(< Pγ̂b ,ub ;b >t,H;b ), < Pγ̂v ,uv ;v >t,H;v )),
consequently, θ is the associative homomorphism.
If G is alternative, then
< Pγ̂,u;a+a+b >∞,H;a+a+b =<< (< Pγ̂a ,ua ;a >∞,H;a ⊗ < Pγ̂a ,ua ;a >∞,H;a
) >∞,H;a+a ⊗ < Pγ̂b ,ub ;b >∞,H;v >∞,H;a+a+b
=<< Pγ̂a ,ua ;a >∞,H;a ⊗(< Pγ̂a ,ua ;a >∞,H;a ⊗ < Pγ̂b ,ub ;b >∞,H;b ) >∞,H;a+a+b =
θ(θ(< Pγ̂a ,ua ;a >t,H;a , < Pγ̂a ,ua ;a >t,H;a ), < Pγ̂b ,ub ;b >t,H;b )
θ(< Pγ̂a ,ua ;a >t,H;a , θ(< Pγ̂a ,ua ;a >t,H;a ), < Pγ̂b ,ub ;b >t,H;b )),
consequently, the homomorphism θ is alternative from the left, analogously it is alternative
from the right.
alternative or associative) such that F (U ) is a group or a ring for each U open in X and for
each U ⊂ V with U and V open in X there exists a homomorphism sU,V : F (V ) → F (U )
such that sU,U = 1 and sU,V sV,Y = sU,Y for each U ⊂ V ⊂ Y with U, V, Y open in X.
Let Fx denotes the family of all elements f ∈ F (U ) for all U open in X with x ∈ U .
Elements f ∈ F (U ) and g ∈ F (V ) are called equivalent if there exists an open neighbor-
hood Y of x such that sY,U (f ) = sY,V (g). This generates an equivalence relation and a
class of all equivalent elements with f is called a germ fx of f at x. A set Fx of all germs
of the pre-sheaf F at a point x ∈ X is the inductive limit Fx = ind − lim F (U ) taken by
all open neighborhoods U of x in X.
In the set F of all germs Fx take a base of topology consisting of all sets {fx ∈ Fx :
x ∈ U }, where f ∈ F (U ). This induces a sheaf S generated by a pre-sheaf F .
A sheaf of groups or rings (all either alternative or associative) on X is a pair (S, h)
satisfying Conditions (S1 − S4):
(S1) S is a topological space;
(S2) h : S → X is a local homeomorphism;
(S3) for each x ∈ X the set Fx = h−1 (x) is a group called a fiber of the sheaf S at a
point x;
(S4) the group or the ring operations are continuous, that is, S∆S ∋ (a, b) 7→ ab−1 ∈
S or S∆S ∋ (a, b) 7→ ab ∈ S and S∆S ∋ (a, b) 7→ a + b ∈ S are continuous respectively,
where S∆S := {(a, b) : a, b ∈ S, h(a) = h(b)}.
We can consider pre-sheafs and sheafs of different classes of smoothness, for example,
H or Hpt , when the corresponding defining sheaf and pre-sheaf mappings sU,V , h and
t
group operations are such and S and F are H t or Hpt differentiable spaces respectively (see
also §1.3.2 in Section 2).
Consider a sheaf SN,G generated by a pre-sheaf U 7→ {f ∈ Homtp ((W M E)t,H , G) :
supp(f ) ⊂ U }, where U is open in N and supp(f ) ⊂ U means that there exists y ∈ N
and η̂ ∈ Hpt (M̂ , N ) with η̂(ŝ0,q ) = x for each q = 1, ..., k and η̂(ŝ0,q ) = y for each
q = k + 1, ..., 2k and γ̂ ∈ Hpt (M̂ , {ŝ0,q : q = 1, ..., 2k}; N, y) such that γ̂ = γ ◦ Ξ and
f =< η̂ ∨ γ̂ >t,H , where the wrap group (W M E)t,H is taken for a marked point y ∈ N ,
Ξ : M̂ → M is the quotient mapping as in Section 2.
In particular, we can take G = A∗r , and call SN,A∗r the sheaf of infinitesimal holonomies,
where 1 ≤ r ≤ 3.
In view of Property (P 4) in Section 2 for each non-singular points y ∈ N and u ∈ Ey
in the fiber Ey of E over y there exists an Ar vector subspace Hu of the tangent bundle Tu E
at u called a horizontal subspace of Tu E such that π∗ |Hu : Hu → Ty N is an isomorphism,
where π(u) = y, t′ ≥ [dim(E)/2] + 2 or t′ = ∞, since there exist generalized derivatives
′
in the Sobolev space H t (see §III.3 [35]). This is the case for all y ∈ N and u ∈ Ey when
′ ′
N and E are of class H t instead of Hpt .
Due to (P 1) the family {Hu } of horizontal subspaces of T E depends smoothly on u.
Suppose that Y is a vector field in T E corresponding to a vector field X in T N such that
π∗ (Y ) = X, then
(CD1) Tu E = Hu ⊕ Vu , where Vu = π∗−1 (0) ⊂ Tu E is the space of vectors tangent
to Eu at u. In accordance with (P 3) the horizontal spaces are G-equivariant, that is,
(CD2) Huz = (Rz )∗ Hu , where Rz is the diffeomorphism of E given by the multi-
plication on z from the right and (Rz )∗ corresponds to the tangent mapping T Rz for the
516 S.V. Ludkovsky
13 and Note 14 Section 3. If γ̂ ∈ Hpt (M̂ , {ŝ0,q : q = 1, ..., k}; N, y0 ), then consider a
path lq joining the point ŝ0,q with ŝ0,q+k , where 1 ≤ q ≤ k, ˆlq : [0, 1] → M̂ . Therefore,
p̂q := γ̂ ◦ ˆlq : [0, 1] → N and p̂∗q w := (p̂q , id)∗ w is a differential form on [0, 1], where w is
an Ar holomorphic connection one-form on E. We get that γ̂ ∗ w is a differential one-form
on M̂ and there exists its restriction νγ,q := γ ∗ w|l̂q [0,1] .
Then we have also γ ∈ Hpt (M, {s0,q : q = 1, ..., k}; N, y0 ) and lq : S 1 → M and
pq : S 1 → N respectively, where γ̂ = γ ◦ Ξ (see Section 2), S 1 is the unit circle in C
with the center at zero, while C = CM is embedded into Ar as R ⊕ M R with M ∈ Ar ,
Re(M ) = 0, |M | = 1, when 2 ≤ r ≤ 3.
R
Since w is Ar holomorphic, then φ w does not depend on a rectifiable curve φ but only
on the initial and final points φ(0) and φ(1), φ : [0, 1] → E (see Theorems 2.15 and 3.10
in [28, 29] and [31]).
Consider now the principal fiber bundle E with the structure group A∗r , where 1 ≤ r ≤
3. Then the pull-back p∗q E of the bundle E is a trivial A∗r -bundle over S 1 . The latter bundle
carries a pull-back connection differential one-form p∗q w. Take the pull-back ρ∗ (p∗q w) one
form, where ρ : S 1 → p∗q E is a trivialization of the fiber bundle p∗q E → S1 .
The parallel transport structure Pγ̂,u (x) for (M, E) with x ∈ M̂ induces the parallel
transport structures Pp̂q ,u∗ (s) for (S 1 , p∗q E) with s ∈ [0, 1] for each q = 1, ..., k, where
pq (u∗ ) = u. Then the holonomy along γ is given by
R
(H) h(γ) = (h1 , ..., hk ) ∈ Gk with hq = hq (γ) = exp[− S 1 ρ∗ (p∗q w)] for each
q = 1, ..., k.
If ζ : S 1 → p∗q E is another trivialization and f : S 1 → C∗M satisfies ζ = fR ρ, so
that f (v) = exp(M 2πθ(v)), where θ(v) ∈ R, M 2πdθ(v) = dLn(f (v)), v ∈ S 1 , S 1 dθ
is an integer number, since R is the center of the algebra Ar , where Ln is the natural
logarithmic function over Ar (see §3.7 and Theorem 3.8.3 [29] and [28, 33]). Therefore,
Formula
R
(H) is independent of a trivialization ρ, since ζ ∗ (p∗q w) = ρ∗ (p∗q w) + dLn(f ), but
exp[ S 1 dLn(f )] = 1.
4. Non-associative bar construction. Let G be a topological group not necessarily
associative, but alternative:
(A1) g(gf ) = (gg)f and (f g)g = f (gg) and g −1 (gf ) = f and (f g)g −1 = f for each
f, g ∈ G
and having a conjugation operation which is a continuous automorphism of G such that
(C1) conj(gf ) = conj(f )conj(g) for each g, f ∈ G,
(C2) conj(e) = e for the unit element e in G.
If G is of definite class of smoothness, for example, Hpt differentiable, then conj is
supposed to be of the same class. For commutative group in particular it can be taken the
identity mapping as the conjugation. For G = A∗r it can be taken conj(z) = z̃ the usual
conjugation for each z ∈ A∗r , where 1 ≤ r ≤ 3.
Denote by ∆n := {(x0 , ..., xn ) ∈ Rn+1 : xj ≥ 0, x0 + x1 + ... + xn = 1} the standard
S
simplex in Rn+1 . Consider (AG)n as the quotient of the disjoint union nk=0 (∆k × Gk+1 )
by the equivalence relations
(1) (x0 , ..., xk , g0 , ..., gk ) ∼ (x0 , ..., xj + xj+1 , ..., xk , g0 , ..., ĝj , ..., gk ) for gj = gj+1
or xj = 0 with 0 ≤ j < k; (x0 , ..., xk , g0 , ..., gk ) ∼ (x0 , ..., xk−1 + xk , g0 , ..., gk−1 ) for
gk−1 = gk or xk = 0.
518 S.V. Ludkovsky
tσ(1) ≤ tσ(2) ≤ ... ≤ tσ(n+k+1) , σ ∈ Sn+k+1 , Sm denotes the symmetric group of all
permutations of the set {1, ..., m}. Define the multiplication for pure states in AG:
(1) |t1 , ..., tn , h0 [h1 |...|hn ]| ∗ |tn+1 , ..., tn+k+1 , hn+k+2 [hn+1 |...|hn+k+1 ]| :=
|tσ(1) , ..., tσ(n+k+1) , (−1)q(σ) (h0 hn+k+2 )[hσ(1) |...|hσ(n+k+1) ]|,
where hl = yl ij(l) , yl ∈ Gj(l) for each l = 0, ..., 2r − 1, q(σ) ∈ Z is such that
(−1)q(σ) ij(0) (ij(1) ...(ij(n+k+1) ij(n+k+2) )...) = (ij(σ(0)) ij(σ(n+k+2)) )(ij(σ(1)) ...
(ij(σ(n+k))) ij(σ(n+k+1)) )...) in Ar ; while in BG:
(2) |t1 , ..., tn , [h1 |...|hn ]| ∗ |tn+1 , ..., tn+k+1 , [hn+1 |...|hn+k+1 ]| :=
520 S.V. Ludkovsky
B4 in [15]), where G0 denotes the multiplicative group of the ring Ĝ0 . Therefore, AG and
BG are the multiplicative groups of the rings AĜ and B Ĝ.
If a ∈ AG0 or a ∈ BG0 , then ab = ba for each b ∈ AG or BG respectively. From
Definition 6 it follows, that they are C ∞ or Hpt groups, when such is G.
The inverse element is
(10) {|t1 , ..., tk , h0 [h1 |...|hk ]| : k}−1 = {|t1 , ..., tk , h−1 −1 −1
0 [h1 |...|hk ]| : k}
due to (2, 6), since
(h0 (h1 ...(hk−1 hk )...))((...(h−1 −1 −1 −1
k hk−1 )...h1 )h0 ) =
−1 −1 −1 −1
(...(((h0 h0 )(h1 h1 ))(h2 h2 )...)(hk hk ) = e
for pure states for each k in view of the Moufang identities (4 − 6). In general it follows
from (3, 8, 9), since v ∗conj(v) ∈ AG0 or BG0 for v ∈ AG or v ∈ BG respectively, hence
v −1 = conj(v)(v ∗ conj(v))−1 = {|t1 , ..., tk , h0 [h1 |...|hk ]| : k}−1 .
In view of (8, 9) and the existence of an inverse element we get, that AG is alternative,
since 1 ≤ r ≤ 3. Putting h0 = 1 and applying the equivalence relation Y we get, that BG
is also an alternative group, since the multiplicative group {i0 , ..., i7 } is alternative. If G
is associative, for example, when 1 ≤ r ≤ 2, then AG and BG are associative, since the
multiplicative group {i0 , i1 , i2 , i3 } is associative.
Thus, groups AG and BG are Z2 graded, hence they are twisted over {i0 , ..., i2r −1 }.
Consider for Ĝ multiplication and addition operations, then they induce them for AG and
BG as above. It follows, that E Ĝ and B Ĝ are twisted rings.
7. Corollary. Let suppositions of Proposition 6 be satisfied, then AB m G and B m G are
topological or C ∞ or Hpt differentiable groups respectively for each m ≥ 1. Moreover, all
maps in the short exact sequence e → B a G → AB a G → B a+1 G → e are continuous or
C ∞ or Hpt correspondingly.
Proof. Define differentiable space structure by induction. Suppose that it is defined on
B G and ∆k ×(B a G)m for k, m ≥ 0, where a ≥ 1. Then f : AB a G → Ar is C ∞ or Hpt if
a
` q f ′
the composition n≥0 (∆n − ∂∆n ) × (B a G)n+1 −→ A
AB a G −→ Ar is either C ∞ or Hpt ,
` qB
while f : B a+1 G → Ar is C ∞ or Hpt if the composition n≥0 (∆n − ∂∆n ) × (B a G)n −→
f ′
B a+1 G −→ Ar is either C ∞ or Hpt , where 0 ≤ r ≤ 3.
A function f : ∆k × (B a+1 G)m → Ar is C ∞ or Hpt if the composition ∆k ×
` id×(qB )m f
( n≥0 (∆n − ∂∆n ) × (B a G)n )m −→ ∆k × (B a+1 G)m −→ Ar is either C ∞ or
′
Hpt . From this it follows that all maps in the short exact sequences are of the same class of
smoothness.
Then the mappings B a G × B a G → B a G and AB a G × AB a G → AB a G of the form
(f, g) 7→ f g −1 are C 0 or C ∞ or Hpt in respective cases due to Formulas 6(1 − 3, 8 − 10)
(see also §1.3.2 in Section 2 and §1 and Appendix B in [15]).
8. Corollary. If a group G satisfies Conditions 4(A1, A2, C1, C2), then there exist Hpt
groups AB a (W M,{s0,q :q=1,...,k} E)t,H and B a (W M,{s0,q :q=1,...,k} E)t,H for each a ∈ N.
Proof. The wrap group (W M,{s0,q :q=1,...,k} E; N, G, P)t,H is a principal Gk bundle over
(W M,{s 0,q :q=1,..,k} N ) M,{s0,q :q=1,...,k} N )
t,H , where (W t,H is commutative and associative
(see Proposition 7(1) in Section 3).
If g ∈ Ĝ, then g has the decomposition g = g0 i0 + ... + g2r −1 i2r −1 with gj ∈ Ĝj for
each j = 0, 1, ..., 2r − 1 and
522 S.V. Ludkovsky
P2r −1
(1) g0 = (g + (2r − 2)−1 {−g + s=1 is (gi∗s )})/2 and
P r −1
(2) gj = (ij (2r − 2)−1 {−g + 2s=1 is (gi∗s )} − gij )/2
r
for each j = 1, ..., 2 − 1. Therefore, each g0 , ..., g2r −1 has analytic expressions through
g due to Formulas (1, 2). Fix this representations. Then the Hpt differentiable parallel
transport structure P with the groups G induces the Hpt differentiable parallel transport
structures j P with groups Gj .
L
Since Ĝk is isomorphic with 0≤j(1),...,j(k)≤2r −1 (Ĝj(1) ij(1) , ..., Ĝj(k) ij(k) ) which is
L
isomorphic with 0≤j(1),...,j(k)≤2r −1 Ĝk0 (ij(1) , ..., ij(k) ), then
(W M,{s0,q :q=1,...,k} E; N, G, P)t,H is isomorphic with a group
L
{f = (f1 , ..., fk ) ∈ 0≤j(1),...,j(k)≤2r −1 [(W M,{s0,q :q=1,...,k} E; N, G0 , P)t,H ∪ {0}]
(ij(1) , ..., ij(k) ) : f1 6= 0, ..., fk 6= 0}, where (ij(1) , ..., ij(k) ) ∈ (A∗r )k and (A∗r )k has the
embedding into the family of all k × k matrices with entries in Ar as diagonal matrices,
(W M,{s0,q :q=1,...,k} E; N, Gj , P)t,H is commutative for each j = 0, ..., 2r − 1 due to Theo-
rem 2.6.
The construction of Proposition 5 above has the natural generalization for Gk instead
of G such that
L
AĜk = 0≤j(1),...,j(k)≤2r −1 (AĜj(1) ij(1) , ..., AĜj(k) ij(k) )
L
which is isomorphic with 0≤j(1),...,j(k)≤2r −1 AĜk0 (ij(1) , ..., ij(k) ), also
L
B Ĝk is isomorphic with 0≤j(1),...,j(k)≤2r −1 B Ĝk0 (ij(1) , ..., ij(k) ), consequently,
A(W M,{s0,q :q=1,...,k} E; N, G, P)t,H is isomorphic with a group
L M,{s0,q :q=1,...,k} E; N, G , P)k
{v ∈ 0≤j(1),...,j(k)≤2r −1 [A(W 0 t,H ∪ {0}](ij(1) , ..., ij(k) ) :
vn = |t1 , ..., tn , h0 [h1 |...|hn ]|, hj 6= 0∀j, ∀n} and
B(W M,{s0,q :q=1,...,k} E; N, G, P)t,H is isomorphic with
L M,{s0,q :q=1,...,k} E; N, G , P)k
{v ∈ 0≤j(1),...,j(k)≤2r −1 [B(W 0 t,H ∪ {0}](ij(1) , ..., ij(k) ) :
vn = |t1 , ..., tn , [h1 |...|hn ]|, hj 6= 0∀j, ∀n}. Continuing this by induction on a and using
Corollary 7 we get the statement of this corollary for each a ∈ N.
9. Lemma. Let N be a C ∞ or Hpt manifold over Ar with 0 ≤ r ≤ 3 and G a C ∞ or Hpt
differentiable group. If f : N → BG is a mapping such that for each y ∈ N there exists an
open neighborhood V of y in N such that f |V = |f0 , f1 , ..., fn , [g1 |...|gn ]| with f0 , ..., fn
being C ∞ or Hpt differentiable mappings, then f is either C ∞ or Hpt differentiable mapping
correspondingly.
Proof. If h : BG → Ar is a C ∞ or Hpt mapping, then for each n ≥ 1 the composition
q h
∆n × Gn −→ B
BG −→ Ar is of the corresponding class. For the commutative diagram
f h f¯ qB
consisting of N −→ BG −→ Ar and N −→ ∆n × Gn −→ BG and f = qb ◦ f¯,
¯ ¯
where f := (f0 , ..., fn , h1 , ..., hn ) both f and h ◦ qB are continuous C ∞ or Hpt . Then the
composition h ◦ f = h ◦ qB ◦ f¯ is continuous and either C ∞ or H t , where as usually
p
h ◦ f (y) := h(f (y)). Thus f : N → BG is continuous either C ∞ or Hpt respectively.
10. Twisted bar resolution and hypercohomologies. For a twisted group G satisfying
Conditions 4(A1, A2, C1, C2) the composition of the short exact sequences
(1) e → B a G → AB a G → B a+1 G → e induces the long exact sequence
σ σ σ σ σ
(2) e → G → AG −→ ABG −→ AB 2 G −→ ... −→ AB a G −→ ...,
where for each a ≥ 0 the homomoprhism σ : AB a G → AB a+1 G is the composition
AB a G → B a+1 G → AB a+1 G of the surjection AB a G → B a+1 G and the monomor-
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 523
is C ∞ or Hpt differentiable and arcwise connected. Therefore, for each b > 0 the cohomol-
ogy group Hb (N, AGN ) is trivial (see also §2.2 [3] and §54 below).
12. A sheaf S on X is called soft if for each closed subset Y of X the restriction map
S(X) → S(Y ) is a surjection.
12.1. Lemma. For a C ∞ or Hpt differentiable group G satisfying conditions
4(A1, A2, C1, C2) the sheaf AGN is soft.
Proof. Consider a closed subset Y of N and a section σY of AGN over Y . In accor-
dance with the definition of a section over a closed subset there exists an open set U and
an extension σU of σY from Y onto U such that Y ⊂ U ⊂ N . From the paracompactness
of N there exists a neighborhood V of Y such that cl(V ) ⊂ U , where cl(V ) denotes the
closure of V in N . Therefore, the extension σ of σY to a global section of AGN is provided
by the formula σ(x) = fˆ(σU (x), ψ(x)), where fˆ : AG × [0, 1] → AG is a deformation
retraction (see §11) and ψ : N → [0, 1] is either a C ∞ or Hpt differentiable function equal
to 1 on V and equal to 0 on M \ U .
13. Remark. Let now N be a C ∞ or H ∞ manifold over Ar and T (N, G, π, Ψ) be a
tangent bundle with T = T N and the projection π : T → N , where connecting mappings
φj ◦ φ−1 k for Vj ∩ Vk 6= ∅ of the atlas At(N ) = {(Vj , φj ) : j} of the manifold N are Ar
holomorphic for 1 ≤ r ≤ 3, φj ◦ φ−1 ∞
k ∈ H . Denote by T the sheaf of germs of smooth
sections of T . Then BT denotes the sheaf associated with a the pre-sheaf assigning to each
`
open subset V of N the group of sections of the natural projection y∈V B(π −1 (y)) → V ,
which are locally of the form
(1) y 7→ |t1 (y), ..., tn (y), [σ1 (y)|...|σn (y)]|, where t1 , ..., tn are C ∞ for r = 1 or H ∞
for 1 ≤ r ≤ 3 functions and σ1 , ..., σn are C ∞ or H ∞ sections of the vector bundle
T (N, G, π, Ψ). Using constructions above we define B a+1 T for each a ∈ N by induction.
Then B a+1 T is the sheaf associated with the pre-sheaf assigning to an open subset V
of N the group of sections of the natural projection
` a+1 (π −1 (y)) → V having the local form (1), where σ , ..., σ are sections of
y∈V B 1 n
B a T over V . Similarly we define AB a T .
If now T = Λb T ∗ N is the b-th exterior power of the cotangent bundle of N , then
the above construction produces the sheaves AB a SN,A b
r
and B a+1 SN,A
b
r
of AB a Ar and
B a+1 Ar valued respectively differential C ∞ forms on N , where the index Ar may be
omitted, when the Cayley-Dickson algebra Ar is specified. In the equation
P
(2) w = J fJ (z)dxb1 ,j1 ∧ dxb2 ,j2 ∧ ... ∧ dxbk ,jk ,
where fJ : N → AB a Ar or fJ : N → B a+1 Ar , z = (z1 , z2 , ...) are local coordinates in
N , zb = xb,0 i0 + xb,1 i1 + ... + xb,2r −1 i2r −1 , where zb ∈ Ar , xb,j ∈ R for each b and every
j = 0, 1, ..., 2r − 1, J = (b1 , j1 ; b2 , j2 ; ...; bk , jk ).
Since each topological vector space Z over Ar with 2 ≤ r ≤ 3 has the natural twisted
structure Z = Z0 i0 ⊕ Z1 i1 ⊕ ... ⊕ Z2r −1 i2r −1 with pairwise isomorphic topological vector
spaces Z0 , ..., Z2r −1 over R, then T N and T ∗ N and Λb T ∗ N have twisted structures, where
X ∗ denotes the space of all continuous Ar additive and R homogeneous functionals on X
with values in Ar , when 2 ≤ r ≤ 3, while X ∗ over C is the usual topologically dual space
of continuous C-linear functionals on X. Therefore, due to Proposition 6 B a T and AB a T
have the induced twisted structure for each a ∈ N.
Each section σ of the sheaf AB a SN k can be written in the form: σ =
a
|h0 , .., hn , σ0 , ..., σn |, where σ0 , ..., σn are smooth differential B Ar valued differential k-
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 525
where the restriction of φ to Vj is φ(x)|Vj = |fj0 , fj1 (x), ..., fjn (x), [gj0 (σ(x)) :
gj1 (σ(x)) : ... : gjn (σ(x))]|, σ : Vj → π −1 (Vj ) is a smooth section of the restric-
tion π −1 (Vj ) → Vj for π : E → N . Consider equivalence classes qj ∼ gj if
and only if there exist s1 , ..., sm ∈ G such that (sm (sm−1 ...(s1 (qj )...) = gj , hence
qj h ∼ gj h, since qj h ∼ gj h if and only if h−1 qj ∼ h−1 gj , which is equivalent
with h(sm (sm−1 ...(s1 (h−1 qj )...) = gj . Due to the alternativity of the group G we get
[(...(gj−1 s−1 −1 −1 −1
1 )...sm−1 )sm ][(sm (sm−1 ...(s1 gl )...)] = gj gl .
Therefore, in the non-homogeneous coordinates the mapping φ takes the form
φ(x) = |fj0 (x), fj1 (x), ..., fjn (x), [gj0 ,j1 (x)|gj1 ,j2 (x)|...|gjn−1 ,jn (x)]|, where gj,l (x) =
[gj (σ(x))]−1 gl (σ(x)) are transition functions associated with the open covering of N by
the open sets {x ∈ N : fj (x) > 0}. Then the mapping φ(x) is independent from the
choice of σ, since gj are G-equivariant. All functions gj are either C ∞ or Hpt , hence φ is
either C ∞ or Hpt correspondingly.
16. Corollary. For each smooth C ∞ or Hpt principal B b A∗r bundle with 1 ≤ r ≤ 3
there exists a C ∞ or Hpt differentiable mapping φ : N → B b+1 A∗r such that E(N, G, π, Ψ)
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 527
smooth maps from N to BG for which there exist lift mappings from N to AG.
On the other hand, f ∈ C ∞ (N, BG) has a lift F : N → AG if and only if f is
smooth (or continuous) homotopic to a constant mapping, since [g0 : ... : gn ] in (BG)n is
the equivalence class {(g0 , ..., gn ) ∼ (sm (...(s1 g0 )...), ..., (sm ...(s1 gn )...)) : s1 , ..., sm ∈
G, m ∈ N}, consequently, C ∞ (N, BG)/π∗ C ∞ (N, AG) ∼ = [N, BG]∞ . In the class of
continuous mappings we get analogously C (N, BG)/π∗ C (N, AG) ∼
0 0
= [N, BG]0 .
18. Notes. In view of §§4-6 there exists a short exact sequence
e → G → AG → BG → e
′
of Hpt homomorphisms due to the twisted structures of G, AG and BG (see Equations
4(A2) and 6(8, 9)).
To groups AG and BG are assigned simplicial topological groups AG. and BG. with
face homomorphisms ∂j : AGn → AGn−1 given by:
(1) ∂j (h0 [h1 |...|hn ]) = h0 h1 [h2 |...|hn ] for j = 0,
∂j (h0 [h1 |...|hn ]) = h0 [h1 |...|hj hj+1 |...|hn ] for 0 < j < n,
∂j (h0 [h1 |...|hn ]) = h0 [h1 |...|hn−1 ] for j = n. While ∂j : BGn → BGn−1 has the
form:
(2) ∂j ([h1 |...|hn ]) = [h2 |...|hn ] for j = 0,
∂j ([h1 |...|hn ]) = [h1 |...|hj hj+1 |...|hn ] for 0 < j < n,
∂j ([h1 |...|hn ]) = [h1 |...|hn−1 ] for j = n.
The degeneracy homomorphisms sj : AGn → AGn+1 are prescribed by the formula:
(3) sj (h0 [h1 |...|hn ]) = h0 [e|h1 |...|hn ] for j = 0,
sj (h0 [h1 |...|hn ]) = h0 [h1 |...|hj |e|hj+1 |...|hn ] for 0 < j < n,
sj (h0 [h1 |...|hn ]) = h0 [h1 |...|hn |e] for j = n. While sj : BGn → BGn+1 is given by:
(4) sj ([h1 |...|hn ]) = [e|h1 |...|hn ] for j = 0,
sj ([h1 |...|hn ]) = [h1 |...|hj |e|hj+1 |...|hn ] for 0 < j < n,
sj ([h1 |...|hn ]) = [h1 |...|hn |e] for j = n.
Analogous mappings are for simplices:
(5) ∂ j (t0 , ..., tn+1 ) = (t0 , ..., tj , tj , tj+1 , ..., tn+1 ) and
(6) sj (t0 , ...., tn+1 ) = (t0 , ..., tj , t̂j+1 , tj+2 , ..., tn+1 ), where t̂j+1 means that tj+1 is
absent.
528 S.V. Ludkovsky
The geometric realization |AG. | of the simplicial space AG. is defined to be the quotient
space of the disjoint union ⊔∞ n
n=0 ∆ × G
n+1 by the equivalence relations
(7) (∂ x, ḡ) ∼ (x, ∂j ḡ) for each (x, ḡ) ∈ ∆n−1 × Gn+1 , while (sj x, ḡ) ∼ (x, sj ḡ)
j
for each (x, ḡ) ∈ ∆n+1 × Gn+1 . At the same time the geometric realization |BG. | of the
simplicial space BG. is the quotient of the disjoint union ⊔∞ n n
n=0 ∆ × G by the equivalence
relations
(8) (∂ j x, ḡ) ∼ (x, ∂j ḡ) for each (x, ḡ) ∈ ∆n−1 × Gn , while (sj x, ḡ) ∼ (x, sj ḡ) for
each (x, ḡ) ∈ ∆n+1 × Gn .
Consider a non-commutative sphere Cr := {z ∈ Ir : |z| = 1} for r = 2, 3, where
Ir := {z ∈ Ar : Re(z) = 0}. For r = 1 put Cr = {i, −i}, where i = (−1)1/2 . Let Z(Cr )
Q
denotes the additive group ZCr /Z, where ZCr := b∈Cr Tb , Tb = Zb for each b ∈ Cr , Z
is the additive group of integers, Z is the equivalence relation such that Tb × T−b /Z = Tb
for each b ∈ Cr . For 2 ≤ r ≤ 3 the group Z(Cr ) is isomorphic with Zα , where card(α) =
card(R) =: c. Particularly, Z(C1 ) = Zi for r = 1.
Henceforth, we consider twisted sheaves and cohomologies over {i0 , ..., i2r −1 }, where
2 ≤ r ≤ 3. In particular, the complex case will also be included for r = 1, but the latter
case is commutative over C. So we can consider simultaneously 1 ≤ r ≤ 3 and generally
speak about twisting undermining that for r = 1 it is degenerate.
19. Proposition. Let G be the group either A∗r or Z(Cr ), where 1 ≤ r ≤ 3. Then
for each H ∞ smooth manifold N over Ar and each b ≥ 2 the group Hb (N, Z(Cr )) is
isomorphic with:
(1) the group E(N, B b−2 G) of isomorphism classes of smooth principal B b−2 G-
bundles over N ;
(2) the group [N, B b−1 G]∞ of smooth homotopy classes of smooth mappings from N
to B b−1 G.
Proof. In view of Corollary 3.4 [28, 29] there exists the short exact sequence
η
(1) 0 → Z(Cr ) −→ Ar → A∗r → 1,
since exp(M + 2πkM/|M |) = exp(M ) for each non-zero purely imaginary M ∈ Ir (with
Re(M ) = 0) and every k ∈ Z, 1 ≤ r ≤ 3, where η(z) = 2πz for each z ∈ Ar . If
f : Ar → A∗r is a differentiable function, then (dLnf ).h = w(h) is the differential one-
form considering d as the external differentiation over R, where h ∈ Ar . In the particular
case of G = A∗r with 1 ≤ r ≤ 3 there exist further short exact sequences
(2) 1 → A∗r → AA∗r → BA∗r → 1
(3) 1 → BA∗r → ABA∗r → B 2 A∗r → 1
(4) 1 → B m A∗r → AB m A∗r → B m+1 A∗r → 1.
Therefore, identifying the ends of these short exact sequences we get the long exact
sequence
(5) 0 → Z(Cr ) → Ar → AA∗r → ABA∗r → ... → AB m A∗r → ...,
where σ : Ar → AA∗r ,..., σ : AB m−1 A∗r → AB m A∗r are homomorphisms, all terms Ar ,
AA∗r ,...,AB m A∗r ,... are contractible spaces.
Suppose now that N and E are of class H ∞ . Let C∞ (N, AB m A∗r ) denotes the sheaf
of germs of C ∞ functions from N into AB m A∗r . Thus, we get the functor C ∞ . Then the
application of C ∞ functor to the long exact sequence (5) gives:
(6) 0 → Z(Cr )N → C∞ (N, Ar ) → C∞ (N, AA∗r ) → C∞ (N, ABA∗r ) → ... →
C∞ (N, AB m A∗r ) → ...,
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 529
where σ∗ : C∞ (N, Ar ) → C∞ (N, AA∗r ),...,σ∗ : C∞ (N, AB m−1 A∗r ) → C∞ (N, AB m A∗r )
are induced homomorphisms.
The latter exact sequence is called the bar resolution of Z(Cr )N . Sheaves C∞ (N, Ar )
and C∞ (N, AB m A∗r ) are contractible, since Ar and AB m A∗r are contractible. Therefore,
the cohomology of the sheaf Z(Cr )N can be computed using the complex
(7) C ∞ (N, Ar ) → C ∞ (N, AA∗r ) → ... → C ∞ (N, AB m A∗r ) → ...
with homomorphisms
σ∗ : C ∞ (N, Ar ) → C ∞ (N, AA∗r ),...,σ∗ : C ∞ (N, AB m−1 A∗r ) → C ∞ (N, AB m A∗r ).
The long exact sequence (7) we call a bar cochain complex of Z(Cr )N . The cohomology of
Z(Cr )N computed with the help of the bar complex is denoted by H∗b (N, Z(Ar )N ) and it is
called the bar cohomology of Z(Cr )N . Then π0 C ∞ (N, BA∗r ) is the first bar cohomology
H1b (N, Z(Ar )N ) of Z(Cr )N .
For the generalized exponential sequence
0 → Z(Cr )N → AB <b−2 GN → B b−2 GN [2 − b] → e there exists the cohomology
long exact sequence
... → h Hb−1 (N, AB <b−2 GN ) → H1 (N, B b−2 GN )
→ Hb (N ; Z(Cr )) → h Hb (N, AB <b−2 GN ) → ...,
where AB <b GN is the complex
σ σ σ σ σ
KN −→ AGN −→ ABGN −→ AB 2 GN −→ ... −→ AB b−1 GN ,
K is equal to either Ar or AZ(Cr ) for G = A∗r or G = BZ(Cr ) respectively, where
b
h H (N, B) denotes a hypercohomology on N with coefficients in a complex of sheaves B
(see its definition in [3, 14, 15] and §10 above).
We have that for each b ≥ 0 the sheaf AB b GN is acyclic. Consider the groups of global
sections AB b GN (N ) of the sheaves AB b GN . Therefore, the cohomology of the complex
AB <b−2 GN is equal to the cohomology of the cochain complex
σ σ σ σ σ
K −→ AGN (N ) −→ ABGN (N ) −→ AB 2 GN (N ) −→ ... −→ AB b−1 GN (N ).
Thus, h Hm (N, AB <b−2 GN ) ∼ = Hm (N, AB <b−2 GN (N )) ∼ = e for each m > b − 2,
consequently, the coboundary homomorphism H (N, B GN ) → Hb (N ; Z(Cr )) is an
1 b−2
alence relations, since X = X0 i0 ⊕ ... ⊕ X2r −1 i2r −1 , where X0 , ..., X2r −1 are pairwise
isomorphic topological vector spaces over R such that we put vx = xv for each v ∈ Xj
and x ∈ R.
Since R is the center of the algebra O, then the projection from AX to BX is R-linear.
Evidently, it is additive as the additive group homomorphism.
21. Remark. Let N. be a simplicial smooth manifold over Ar , where 0 ≤ r ≤ 3.
A smooth m-form w on the geometric realization |N. | of N. is defined to be as a family
{wk : k} of smooth differential m-forms wk on ∆k × Nk with values in Ar being applied
to vectors, satisfying for each 0 ≤ j ≤ n the compatibility conditions:
(1) (∂ j × id)∗ wn = (id × ∂j )∗ wn−1
(2) (sj × id)∗ wn = (id × sj )∗ wn+1 ,
where ∂ j × id, id × ∂j , sj × id and id × sj are the maps as follows:
(3) id × ∂j : ∆n−1 × Nn → ∆n−1 × Nn−1 , ∂ j × id : ∆n−1 × Nn → ∆n × Nn ,
id × sj : ∆n+1 × Nn → ∆n+1 × Nn+1 , sj × id : ∆n+1 × Nn → ∆n × Nn such that ∂ j and
sj are coface and the codegeneracy maps on ∆n and ∂j , sj are the face and the degeneracy
maps on Nn . We consider w taking values in a vector space or an algebra over Ar as is
specified below.
For a Lie group G either over R or may be twisted over Ar and its Lie alge-
bra g put g −1 dg as the canonical g-valued connection 1-form on G (see also Lemma
20). Under the mapping g 7→ hg and dg 7→ hdg we have (g −1 h−1 )(hdg) = g −1 dg
due to the alternativity of G and the Maufang identity (xy)(zx) = x(yz)x for each
x, y, z ∈ O and de = d(g −1 g) = 0 = (dg −1 )g + g −1 dg = [(dg −1 )h−1 ](hg) +
(g −1 h−1 )(hdg). Iterating this relation due to the alternativity of O and the Moufang iden-
tities in it we get the equivariance condition in homogeneous coordinates over O as well:
[(...(g −1 s−1 −1 −1
1 )...sm−1 )sm ][sm (sm−1 ...(s1 dg1 )...)] = g1 dg1 .
−1
|t1 , ..., tj , tj , tj+1 , ..., tn−1 , [z1−1 dz1 |...|zn−1 dzn ]| and
(id × ∂j )∗ A(z −1 dz)n−1 ||t1 ,...,tn−1 ;z0 [z1 |...|zn ]| =
|t1 , ..., tn−1 ; (z0−1 dz0 ) + (z1−1 dz1 )[z2−1 dz2 |...|zn−1 dzn ]| for j = 0,
(id × ∂j )∗ A(z −1 dz)n−1 ||t1 ,...,tn−1 ;z0 [z1 |...|zn ]| =
|t1 , ..., tn−1 ; z0−1 dz0 [z1−1 dz1 |...|zj−1 −1
dzj−1 |(zj−1 dzj )+(zj+1 −1 −1
dzj+1 )|zj+2 dzj+2 |...|zn−1 dzn ]|
for 0 < j < n,
(id × ∂j )∗ A(z −1 dz)n−1 ||t1 ,...,tn−1 ;z0 [z1 |...|zn ]| =
|t1 , ..., tn−1 ; z0−1 dz0 [z1−1 dz1 |...|zn−1 −1
dzn−1 ]| for j = n, while
∗
(id × ∂j ) B(z dz) −1 n−1 ||t1 ,...,tn−1 ;[z1 |...|zn ]| =
|t1 , ..., tn−1 ; [z2−1 dz2 |...|zn−1 dzn ]| for j = 0,
and (id × ∂j )∗ B(z −1 dz)n−1 ||t1 ,...,tn−1 ;[z1 |...|zn ]| =
|t1 , ..., tn−1 ; [z1−1 dz1 |...|zj−1 −1
dzj−1 |(zj−1 dzj ) + (zj+1 −1 −1
dzj+1 )|zj+2 dzj+2 |...|zn−1 dzn ]| for
0<j<n
of Proposition 26 for each α < β ∈ Λ and every U open in X1 there exists the group
α : S α
homomorphism πβ,∗ W,X1 ,X2 ,Gα (U ) → SW,X1 ,X2 ,Gβ (U ) generated by πβ . Thus, there
exists SW,X1 ,X2 ,G := ind − limα∈Λ SW,X1 ,X2 ,Gα .
33. Corollary. Let X1 = ind − limα∈Λ X1,α and G = ind − limα∈Λ Gα satisfy
conditions of Proposition 26, where Gα = GX1,α and X1,α is an Hpt pseudo-manifold for
each α in a directed set Λ. Then SW,X1 ,X2 ,G = ind − limα∈Λ SW,X1,α ,X2 ,Gα .
Proof. For an Hpt pseudo-manifold X1 its base of topology consists of all those subsets
T
U in X1 such that U = m v=1 Uα(v) for some m ∈ N and α(1), ..., α(m) ∈ Λ, where
−1
Vα = πα (Uα(v) ) is open in Xα , πα : X1,α → X1 is an embedding for each α in Λ.
In view of Proposition 26 for each U open in X1 and α < β ∈ Λ there exists a group
α : S
homomorphism πβ,∗ W,X1,α ,X2 ,Gα (U ) → SW,X1,β ,X2 ,Gβ (U ). Due to Proposition 32 this
generates SW,X1 ,X2 ,G as the inductive limit of sheaves SW,X1,α ,X2 ,Gα .
′
34. Theorem. Let X2 = X2,1 × X2,2 , where X1 , X1,2 and X2,2 are Hpt and Hpt
pseudo-manifolds respectively over Ar as in §25. Then the restriction of the complete
tensor product of wrap sheaves SW,X1 ,X2,1 ,G1 ⊗S ˆ W,X1 ,X2,2 ,G2 on ∆1 × X2 is isomorphic
with SW,X1 ,X2 ,G , where G is the tensor product G := G1 ⊗G2 of sheaves G1 and G2 on X1 ,
∆1 := {(x, x) : x ∈ X1 } is the diagonal in X12 .
Proof. If B1 and B2 are sheaves on a topological space X, then B1 ⊗B2 denotes the sheaf
on X generated by the presheaf U 7→ B1 (U ) ⊗ B2 (U ), where (B1 ⊗ B2 )x ∼ = B1,x ⊗ B2,x is
the natural isomorphism of fibers. The sheaf B1 ⊗B2 is called the tensor product of sheaves.
Consider the natural projections φ1 : X2 → X2,1 and φ2 : X2 → X2,2 having exten-
sions id × φ1 : X1 × X2 → X1 × X2,1 and id × φ2 : X1 × X2 → X1 × X2,2 . Therefore,
there exists the sheaf S := SW,X1 ,X2,1 ,G1 ⊗S ˆ W,X1 ,X2,2 ,G2 := [(id × φ1 )∗ SW,X1 ,X2,1 ,G1 ] ⊗
∗
[(id × φ2 ) SW,X1 ,X2,2 ,G2 ] which is the complete tensor product of sheaves (see in general
Chapter 1 §5 [3]).
If γ : M → X2 is an Hpt mapping preserving marked points, then γ = (γ1 , γ2 ), where
γj : M → X2,j for j = 1, 2, γ(s0,q ) = y0 , γj (s0,q ) = yj,0 for each q = 1, ..., k and
j = 1, 2, y0 = y1,0 × y2,0 . Then we get a lifting γ̂ : M̂ → X2 such that γ ◦ Ξ = γ̂ (see §§2,
3 and 6 in Section 2). Therefore, Pγ̂,u (ŝ0,k+q ) = Pγ̂1 ,u1 (ŝ0,k+q ) ⊗ Pγ̂2 ,u2 (ŝ0,k+q ) ∈ G for
each q = 1, ..., k, with G = G1 ⊗ G2 being the direct product of groups for G1 = G1 (U1 )
and G2 = G2 (U2 ) for every Uj open in X1 , j = 1, 2, where u ∈ Ey0 , uj ∈ Ej,yj,0 ,
N = N1 × N2 , Nj ⊂ Vj ⊂ X2,j , E = E(N, G, π, Ψ), Ej = E(Nj , Gj , πj , Ψj ) are
principal bundles, y0 = y0,1 × y0,2 , yj,0 ∈ Nj are marked points, Vj is open in X2,j for
j = 1, 2 (see also §25).
For classes of equivalent parallel transport structures we get < Pγ̂,u >t,H =<
Pγ̂1 ,u1 >t,H ⊗ < Pγ̂2 ,u2 >t,H , hence F (U × (V1 × V2 )) is isomorphic with (φ1 )∗ F (U ×
V1 ) ⊗ (φ2 )∗ F (U × V2 ) for each U open in X1 and all Vj open in X2,j , j = 1, 2, since open
sets of the form V = V1 × V2 form a base of topology in X2 , where F (U × Vj ) is given for
the group Gj (U ). Here U = U1 = U2 and (φ1 )∗ F (U ×V1 )⊗(φ2 )∗ F (U ×V2 ) is isomorphic
with the restriction of (id × φ1 )∗ F (U × V1 ) ⊗ (id × φ2 )∗ F (U × V2 ) from U 2 × V1 × V2 onto
∆(U ) × V1 × V2 , where ∆(U ) denotes the diagonal in U 2 . Thus, SW,X1 ,X2 ,G is isomorphic
with the restriction of the complete tensor product of sheaves SW,X1 ,X2,1 ,G1 ⊗S ˆ W,X1 ,X2,2 ,G2
on ∆1 × X2 .
35. Twisted Alexander-Spanier cohomologies.
Let G be a group satisfying Conditions 4(A1, A2), which may be in particular a wrap
538 S.V. Ludkovsky
For j ∈ J(x) we get θ(sj (x)) = s(x). Take W (x) sufficiently small such that
sj |W (x) =: sx does not depend on j ∈ J(x), since J(x) is finite, consequently, sx ∈
A(W (x)).
Let x, y ∈ U , z ∈ W (x) ∩ W (y) and j ∈ J(z), where J(z) ⊂ J(x) ∪ J(y). Then
sx |W (x)∩W (y) = sy |W (x)∩W (y) . Due to Condition (S2) there exists β ∈ A(U ) such that
β|W (x) = sx for each x ∈ U , clearly, θ(β) = s|U .
Take now C ∈ φ such that |s| ⊂ Int(C) and C ⊂ U , where Int(C) denotes the interior
of C. If x ∈ C \ Int(C), then θ(β)(x) = s(x) = 0. Therefore, there exists a covering
{Qj } of C \ Int(C) with open in X sets Qj such that Qj ⊂ U and t|Qj = 0 for every j.
Choose the open covering {Qj } ∪ {Int(C), X \ C} of X and elements e ∈ A(Qj ),
β|Int(C) ∈ A(Int(C)) and e ∈ A(X \ C). Restrictions of each two elements on a common
part of their domains of definition coincide. Hence due to Condition (S2) such elements
have a common extension q ∈ A(X) and inevitably θ(q) = s and |q| = |θ(q)| = |s| ∈ φ.
θ
The sequence e → A0 (X) → Aφ (X) −→ Γφ (S) → e is exact, since each subsequence
θ P r
e → A0,k (X) → Aφ,k (X) −→ Γφ,k (S) → e is exact, where Âφ = 2k=0 −1
Âφ,k ik and
P2r −1
Γ̂φ = k=0 Γ̂φ,k ik , where each Âφ,k is commutative and they are pairwise isomorphic for
different k, as well as Γ̂φ,k are commutative and pairwise isomorphic for different values of
k, since the sheaf S is twisted over the group of standard generators {i0 , ..., i2r −1 } of the
Cayley-Dickson algebra Ar .
Mention, that for a pre-sheaf A satisfying Condition (S1) we have A0 (X) = e.
37. Corollary. Let conditions of Theorem 36 be satisfied. Then for a paracompactifying
family φ of supports there exists the natural isomorphism:
Hm ∼ m ∗
φ (X; G) = H (Γφ (S (X; G)).
Proof. This follows immediately from Theorem 36 and §35.
38. Twisted singular cohomologies.
Let B be a locally finite twisted sheaf on X, that is a group B(U ) satisfies Conditions
4(A1, A2, C1, C2) for each U open in X. For U ⊂ X denote by S m (U ; B) the group
of singular m-dimensional cochains of the space U with coefficients in B. Each element
f ∈ S m (U ; B) is a function posing for each m-dimensional simplex σ : ∆m → U a section
f (σ) ∈ Γ(σ ∗ (B)), where ∆m is a standard m-dimensional simplex.
The pre-sheaf S m (.; B) satisfies Condition (S2). The sheaf B is locally constant, then
the sheaf σ ∗ (B) is constant on ∆m , since the simplex ∆m is simply connected, where
m ≥ 1. Therefore, there exists a usual coboundary operator d : S m (U ; B) → S m+1 (U ; B).
Consider the sheaf S m (U ; B) generated by a pre-sheaf U 7→ S m (U ; B). Then the
differential d in the pre-sheaf induces the differential in the sheaf. For a locally constant
sheaf B singular cohomologies with coefficients in B and supports in the family φ are
defined as ∆ Hm m ∗
φ (X; B) = H (Sφ (X; B)). Since B is the twisted sheaf over {i0 , ..., i2r −1 },
then Sφ (X; B) and inevitably ∆ Hm
∗
φ (X; B) are twisted over {i0 , ..., i2r −1 }.
Let U := {Uj : j} be an open covering of X and let S ∗ (U; B) be a group of singular
cochains defined on singular simplices subordinated to the covering U. With the help of
the subdivision we get, that the homomorphism bU : S ∗ (X; B) → S ∗ (U; B) induces the
isomorphism of cohomologies, consequently, the complex KU = kerbU is acyclic. On
S
the other hand, S0∗ (X; B) = KU∗ = ind − lim KU , hence H∗ (S0∗ (X; B)) = H∗ (ind −
lim KU∗ ) = ind − lim H∗ (KU∗ ) = e.
540 S.V. Ludkovsky
Thus for a paracompatifying family of supports from the exactness of the sequence
e → S0∗ → Sφ∗ → Γφ (S ∗ ) → e
and Theorem 36 it follows the isomorphism ∆ Hm ∼ m ∗
φ (X; B) = H (Γφ (S (X; B))).
39. Twisted differential sheaves.
A graded sheaf is a sequence {S m : m ∈ Z} of sheaves, which is called a differential
sheaf if there are homomorphisms
(1) d : S m → S m+1 such that d2 = 0 for each m. This sheaf may be twisted over
{i0 , ..., i2r −1 }, where 2 ≤ r ≤ 3. In this case we suppose that
(2) up to an automorphisms θm : S m → S m we have θm+1 ◦ d(Skm ) ⊂ Skm+1 for each
k = 0, ..., 2r − 1.
A differential sheaf S ∗ having S m = 0 for each m < 0 and supplied with the augmen-
tation homomorphism ε : B → S 0 is called the resolvent of the sheaf, if the sequence
ε d d
e → B −→ S 0 −→ S 1 −→ S 2 → ...
is exact.
The notion of differential graded pre-sheaves is formulated analogously. If S m is
twisted, that is Ŝ m = Ŝ0m i0 + ... + Ŝ2mr −1 i2r −1 , where Ŝkm and Ŝjm are pairwise isomorphic
and commutative for each k 6= j, then Ker(d : S m → S m+1 ) and Im(d : S m−1 → S m )
are twisted as well, since up to isomorphisms θm : S m → S m we have θm+1 ◦ d(Skm ) ⊂
Skm+1 for each k = 0, ..., 2r − 1.
A sheaf of cohomologies (in another words a derivative sheaf) is defined as Hm (S ∗ ) =
Ker(d : S m → S m+1 )/Im(d : S m−1 → S m ). If S ∗ is generated by a differential
pre-sheaf S ∗ , then Hm (S ∗ ) is generated by the pre-sheaf U 7→ Hm (S ∗ (U )).
For a sheaf B on topological space X and an open subset U ⊂ X denote by Y0 (U ; B) a
set of all mappings (may be discontinuous) f : U → B such that π ◦ f = id is the identity
Q
mapping on U , where π : B → X is the canonical projection. Thus Y0 (U ; B) = x∈U Bx
and it is the group with the pointwise group operation. Therefore, U 7→ Y0 (U ; B) is the pre-
sheaf satisfying Conditions (S1, S2), hence it is the sheaf which we denote by Y 0 (X; B).
If B is twisted, then Y 0 (U ; B) is twisted as well.
The inclusion of all continuous sections of B into the family of all sections not neces-
sarily continuous induces the augmentation homomorphism ε : B → Y 0 (X; B).
For a family φ of supports put Yφ0 (X; B) = Γφ (Y 0 (X; B). If e → B1 → B2 → B3 → e
is a short exact sequence of sheaves (may be twisted), then the sequence of pre-sheaves
e → Y0 (X; B1 ) → Y0 (X; B2 ) → Y0 (X; B3 ) → e is exact. If f ∈ Yφ0 (X; B), then its
support is |f | := cl{x : f (x) 6= e}. Therefore, f is an image of a section g of the sheaf
B such that g is not necessarily continuous and g(x) = e if f (x) = e for x ∈ X, hence
|g| = |f | ∈ φ.
Denote by Z 1 (X; B) the cokernel of the homomorphism ε such that the sequence e →
ε ∂
B −→ Y 0 (X; B) −→ Z 1 (X; B) is exact. Define by induction the sheaves
Y m (X; B) = Y 0 (X; Z m (X; B)), Z m+1 (X; B) = Z 1 (X; Z m (X; B)).
If B is twisted over {i0 , .., i2r −1 }, then Z 1 (X; B) is twisted as well and by induction
Z m (X; B) and Y m (X; B) are twisted for each m ∈ N. Therefore, the sequence e →
ε ∂
Z m (X; B) −→ Y m (X; B) −→ Z m+1 (X; B) → e is exact. Consider the composition
∂ ε
d = ε ◦ ∂ for Y m (X; B) −→ Z m+1 (X; B) −→ Y m+1 (X; B), then the sequence
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 541
ε d d
e → B −→ Y 0 (X; B) −→ Y 1 (X; B) −→ Y 2 (X; B) → ... is exact. Thus, Y ∗ (X; B)
is the resolvent of the sheaf B, which is called the canonical resolvent.
40. Proposition. The canonical resolvent of the twisted sheaf B is fiberwise homotopi-
cally trivial.
Proof. Consider the homomorphism Y0 (U ; B) → Bx such that U ∋ x 7→ f (x) ∈ Bx
for each f ∈ Y0 (U ; B) and x ∈ U . The direct limit by neighborhoods of a point x induces
the homomorphism ηx : Y 0 (X; B)x → Bx , consequently, ηx ◦ ε : Bx → Bx is the identity
isomorphism, where ηx ◦ ε(z) = ηx (ε(z)). Define the homomorphism νx : Z 1 (X; B) →
Y 0 (X; B) by the formula νx ◦ ∂ = 1 − ε ◦ ηx which defines νx in a unique way. Therefore,
there exists a fiber splitting
ε ∂
Z m (X; B)x −→ Y m (X; B) −→ Z m+1 (X; B)x and
ηx νx
Z m (X; B)x ←− Y m (X; B) ←− Z m+1 (X; B)x . Put Dx := νx ◦ ηx : Y m (X; B)x →
Y m−1 (X; B)x for m > 0. Therefore, d ◦ Dx + Dx ◦ d = ε ◦ ∂ ◦ νx ◦ ηx + νx ◦ ηx ◦ ε ◦ ∂ =
ε ◦ ηx + νx ◦ ∂ = 1 on Y m (X; B) for m > 0. At the same time on Y 0 (X; B)x we have
Dx ◦ d = νx ◦ ηx ◦ ε ◦ ∂ = νx ◦ ∂ = 1 − ε ◦ ηx . This means, that Y ∗ (X; B)x is homotopically
fiberwise trivial resolvent.
41. Remark. The functor Y 0 (X; B) is exact by B, hence Z 1 (X; B) is also the exact
functor by B. Using induction we get, that all functors Y m (X; B) and Z m (X; B) are exact
by B. For an arbitrary family φ of supports on X put Yφm (X; B) := Γφ (Y m (X; B)) =
Yφ0 (X; Z m (X; B)). Since the functor Y0 (X; ∗) is exact, then the functor Yφm (X; B) is
exact.
42. Definition. Cohomologies in X with supports in φ with coefficients in B are defined
as Hm m ∗
φ (X; B) := H (Yφ (X; B)).
42.1. Note. The sequence e → Γφ (B) → Γφ (Y 0 (X; B)) → Γφ (Y 1 (X; B)) is exact,
consequently, Γφ (B) ∼ = H0φ (X; B). If there is a short exact sequence of twisted sheaves
e → B1 → B2 → B3 → e on X, then it implies the exact sequence of cochain complexes
e → Yφ∗ (X; B1 ) → Yφ∗ (X; B2 ) → Yφ∗ (X; B3 ) → e, that in its turn induces the long
exact sequence
δ m+1
... → Hm m m
φ (X; B1 ) → Hφ (X; B2 ) → Hφ (X; B3 ) −→ Hφ (X; B1 ) → ....
43. Definition. Let G be a topological group satisfying conditions 4(A1, A2, C1, C2)
such that G is a multiplicative group of the ring Ĝ, where 1 ≤ r ≤ 2. Then define the
smashed product Gs such that it is a multiplicative group of the ring Ĝs := Ĝ ⊗l Ĝ, where
l = i2r denotes the doubling generator, the multiplication in Ĝ ⊗l Ĝ is
(1) (a + bl)(c + vl) = (ac − v ∗ b) + (va + bc∗ )l for each a, b, c, v ∈ Ĝ, where v ∗ =
conj(v).
A smashed product M1 ⊗l M2 of manifolds M1 , M2 over Ar with dim(M1 ) =
dim(M2 ) is defined to be an Ar+1 manifold with local coordinates z = (x, yl), where
x in M1 and y in M2 are local coordinates.
44. Theorem. There exists smashed products S s := S1 ⊗l S2 on X = X1 = X2 and
s
Ŝ := S1 ⊗ ˆ l S2 on X = X1 × X2 over {i0 , ..., i2r+1 −1 } of isomorphic twisted sheaves S1 on
X1 and S2 on X2 over {i0 , ..., i2r −1 } with X1 = X2 , in particular of wrap sheaves, where
1 ≤ r ≤ 2, l = i2r .
Proof. If Sj is a sheaf on a topological space Xj twisted over {i0 , ..., i2r −1 }, then
Ŝj = Ŝ0,j i0 ⊕ ... ⊕ Ŝ2r −1,j i2r −1 , where Ŝk,j (U ) = Sk,j (U ) ∪ {0} are commutative rings
542 S.V. Ludkovsky
for each U open in Xj , Ŝk,j are sheaves on Xj pairwise isomorphic for different values of
k. Then for X = X1 = X2 take Sxs := (S1 )x ⊗l (S2 )x for each x ∈ X in accordance with
Definition 43, that defines the twisted sheaf S on X over {i0 , ..., i2r+1 −1 } due to Proposition
3.19. This sheaf S is the smashed tensor product of sheaves.
If X = X1 × X2 , then take Ŝ s := S1 ⊗ ˆ l S2 = (π1∗ S1 ) ⊗l (π2∗ S2 ), which is the smashed
complete tensor product of sheaves, where π1 : X → X1 and π2 : X → X2 are projections.
45. Corollary. Let X2 = X2,1 ⊗l X2,2 be the smashed product, where X1 and X2 are
′
Hpt and Hpt pseudo-manifolds respectively over Ar+1 , 1 ≤ r ≤ 2. Then the restric-
tion the smashed complete tensor product of wrap sheaves SW,X1 ,X2,1 ,G ⊗ ˆ l SW,X1 ,X2,2 ,G
s
on ∆1 × X2 is isomorphic with SW,X1 ,X2 ,G s , where G is the smashed tensor product
G s := G⊗l G twisted over {i0 , ..., i2r+1 −1 } of a sheaf G twisted over {i0 , ..., i2r −1 } on
X1 , ∆1 := {(x, x) : x ∈ X1 } is the diagonal in X12 .
Proof. The smashed product of manifolds was described in details in the proof of
Theorem 3.20. Consider an Ar shadow of X1 that exists, since Ar+1 = Ar ⊕ Ar l, where
l = i2r . For each U open in X1 there exists a group G(U ), hence G(U ) ⊗l G(U ) is defined
due to Proposition 3.19, that gives the sheaf G s on X1 . Then wrap sheaves SW,X1 ,X2,b ,G
over Ar are defined, where b = 1, 2. Thus the statement of this corollary follows from
Proposition 3.19 and Theorem 44, modifying the proof of §34 for the smashed complete
tensor product instead of complete tensor product so that Pγ̂,u (ŝ0,k+q ) = Pγ̂1 ,u1 (ŝ0,k+q )⊗l
Pγ̂2 ,u2 (ŝ0,k+q ) ∈ Gs with E = E(N, Gs , π, Ψ), where G = G(U ), U = U1 = U2 ,
consequently, < Pγ̂,u >t,h =< Pγ̂1 ,u1 >t,H ⊗l < Pγ̂2 ,u2 >t,H .
46. Corollary. Let X1 = X1,1 ⊗l X1,2 and X2 = X2,1 ⊗l X2,2 are smashed products,
′
where X1 , and X2 are Hpt and Hpt pseudo-manifolds respectively over Ar+1 , 1 ≤ r ≤ 2.
Then the wrap sheaf SW,X1 ,X2 ,G s is twisted over {i0 , ..., i2r+1 −1 } and is isomorphic with
the smashed complete tensor product of twice iterated wrap sheaves
ˆ l SW,X1,2 ,X2,2 ,SW,X ,X ,G ,
SW,X1,2 ,X2,1 ,SW,X1,1 ,X2,1 ,G ⊗ 1,1 2,2
where G is the smashed tensor product G s := G⊗l G of a twisted sheaf G over
s
G(U ), pseudo-manifolds X1 and X2 and the sheaf G are of class Hp∞ (see also §25).
Corollary. There exists a homomorphism of of iterated wrap sheaves θ : SW,X1 ,X2 ,G;a ⊗
SW,X1 ,X2 ,G;b → SW,X1 ,X2 ,G;a+b for each a, b ∈ N. Moreover, if G is either associative or
alternative, then θ is either associative or alternative.
Proof. For pre-sheaves the mapping
(2) θ : Fa (U × V ) ⊗ Fb (U × V ) → Fa+b (U × V )
is induced by Formula 47(1) and due to Theorem 3.21. Then θ has the extension on the
sheaf of iterated wrap groups, since (SW,X1 ,X2 ,G;a )z = ind − lim Fa (U × V ), where the
direct limit is taken by open subsets U × V for a point z = x × y ∈ X1k × X2 , x ∈ X1k ,
y ∈ X2 , such that x ⊂ U , y ∈ V , U is open in X1 , V is open in X2 .
The inductive limit topology in (SW,X1 ,X2 ,G;a )z is the finest topology relative to which
each embedding Fa (U × V ) ֒→ (SW,X1 ,X2 ,G;a )z is continuous. If f ∈ (SW,X1 ,X2 ,G;a )z and
g ∈ (SW,X1 ,X2 ,G;b )z , then there exist open U1 × V1 and U2 × V2 such that f ∈ Fa (U1 × V1 )
and g ∈ Fb (U2 ×V2 ), consequently, f ∈ Fa (U ×V ) and g ∈ Fb (U ×V ), where U = U1 ∪U2
and V = V1 ∪ V2 , hence θ(f, g) ∈ Fa+b (U × V ). From (2) and the definition of the
inductive limit topology it follows, that θ is continuous, since on iterated wrap groups θ is
Hp∞ differentiable.
Moreover, in accordance with Theorem 3.21 θ is either associative or alternative if G is
associative or alternative.
48. Note. Let φ be a family of supports in X and B be a sheaf on X, where B may be
twisted. A sheaf B is called φ-acyclic, if Hbφ (X; B) = 0 for each b > 0.
Let L∗ be a resolvent of B. Put Z b := Ker(Lb → Lb+1 ) = Im(Lb−1 → Lb ), where
0
Z = B. An exact sequence
(1) e → Z b−1 → Lb−1 → Z b → e
induces an exact sequence
(2) e → Γφ (Z b−1 ) → Γφ (Lb−1 ) → Γφ (Z b ) → H1φ (X; Z b−1 ).
Therefore, there exists the monomorphism
(3) Hb (Γφ (L∗ )) = Γφ (Z b )/Im(Γφ (Lb−1 → Γφ (Z b )) → H1φ (X; Z b−1 ).
Moreover, the sequence e → Z b−v → Lb−v → Z b−v+1 → e induces the homomor-
phism:
(4) Hb−1
φ (X; Z
b−v+1 ) → Hv (X; Z b−v ).
φ
Define κ as the composition
(5) Hb (Γφ (L∗ )) → H1φ (X; Z b−1 ) → H2φ (X; Z b−2 ) → ... → Hbφ (X; Z 0 ).
If all sheaves Lb are φ-acyclic, then (3, 4) are isomorphisms. We call κ natural, if from
the commutativity of the diagram:
B −→ L∗
↓f ↓g
E −→ M∗
so that wk and νk,j and ηk,j,l are H ∞ differential 1-forms. Then from (6 − 8) it follows,
that
(9) wk − wl + νk,l = wk − wj + νk,j + wj − wl + νj,l + ηk,j,k and hence
(10) ηk,j,l = νk,l − νk,j − νj,l .
Generally ηk,j,l may be non-zero because of non-commutativity or non-associativity.
In view of the alternativity of the octonion algebra O the identities eM eN = eK , eM =
e e−N , eN = e−M eK and e−K = e−N e−M are equivalent, that leads to the identities:
K
a group E(N, B b−2 A∗r , ∇b−1 ) of equivalence classes of differentiable principal B b−2 A∗r -
bundles with (b − 1)-connections.
An assignment (g, w1 , w2 , ..., wb−1 ) 7→ (−1)b−1 dwb−1 induces a homomorphism
K : E(N, B b−2 A∗r , ∇b−1 ) → SA b (N ) called the curvature of the b-connection
r
(g, w1 , w2 , ..., wb−1 ). The kernel ker(K) is isomorphic to the group E(N, B b−2 A∗r , ∇f lat )
of isomorphism classes of differentiable principal B b−2 A∗r -bundles with flat connections.
51. Curvature of holonomy. If v, w ∈ T0 Rn , put
(1) γv,w (u) = 4uv for 0 ≤ u ≤ 1/4, γv,w (u) = v + 4(u − 1/4)w for 1/4 ≤ u ≤ 1/2,
γv,w (u) = w −4(u−3/4)v for 1/2 ≤ u ≤ 3/4, γv,w (u) = 4(1−u)w for 3/4 ≤ u ≤ 1 and
s (u) := γ
γv,w sv,sw (u), where 0 ≤ u, s ≤ 1. For a sequence of vectors w = (w0 , w1 , ..., wq )
in T0 Rn with q ∈ N define a (q + 1)-dimensional parallelepiped p[w0 , ..., wq ] in the
Euclidean space Rn with q < n if w0 , ..., wq are linearly independent. Then define
γw0 ,w1 ,w2 (u1 , u2 ) := γγw0 ,w1 (u1 ),w2 (u2 ) and by induction
(2) γw0 ,...,wq (u1 , ..., uq ) = γγw0 ,...,wq−1 (u1 ,...,uq−1 ),wq (uq ) and γws (u , ..., u ) :=
1 q
γsw (u1 , ..., uq ), where 0 ≤ u1 , ..., uq , s ≤ 1. This gives the natural parametrization of
the parallelepiped p[w0 , ..., wq ] and the mapping γw : ∂I q+1 → Rn which is continuous
and piecewise C ∞ . Denote by ej = (0, ..., 0, 1, 0, ..., 0) the standard orthonormal basis in
Rn with 1 in the j-th place. Put Ln(diag(a1 , ..., ak )) := diag(Ln(a1 ), ..., Ln(ak )), where
Ln is the principal branch of the logarithmic function with Ln(1) = 0 and diag(a1 , ..., an )
is the diagonal matrix with entries a1 , ..., ak ∈ A∗r .
If h is an (A∗r )k -valued C n holonomy or an homomorphism for a wrap group
(W M E)∞,H with M̂ being Hp∞ diffeomorphic with ∂I m+1 and ψ = (y1 , ..., yn ) is a
coordinate system centered at y, ψ : V → Rn , V is an open neighborhood of a point y in
N , then a curvature of h at y is a q-form
P
(3) Ky := 1≤j1 <...<jq ≤n Kj1 ,j2 ,...,jq (y)dyj1 ∧ dyj2 ∧ ... ∧ dyjq ∈ Λq Ty∗ N , where
(4) Kj1 ,....,jq (y) = (−1)q lims→0 Ln[h(ψ −1 (γesj ,...,ejq ))]s−q−1 ,
1
where m ≥ q.
Consider the inversion (wj , wj+1 ) 7→ (wj+1 , wj ). In view of Theorem 3.2 for M̂ being
∞
Hp diffeomorphic with ∂I m+1 using the iterated loops and the mapping uj 7→ (1 − uj )
we get, that
(5) Ky (wg(1) , ..., wg(q+1) ) = (−1)|g| Ky (w1 , ..., wq+1 ),
where g ∈ Sq+1 , Sq denotes the symmetric group of the set {1, ..., q}, |g| = 1 for odd g,
while |g| = 2 for an even transposition g.
52. Remark. Consider an H ∞ manifold N and a pseudo-manifold X. A mapping
γ : X → N is called piecewise C ∞ or H ∞ smooth if it is continuous and the restriction
of γ to each top dimensional simplex of X is a C ∞ or H ∞ mapping. A piecewise smooth
mapping γ : X → N is called an oriented singular pseudo-manifold q-cycle, if X is an
oriented pseudo-manifold q-cycle. Denote by Zψ ψ
q (N ) := Zq (X, N ) the group of oriented
singular pseudo-manifold q-cycles in N .
If there exists an oriented pseudo-manifold with boundary (Y, ∂Y ) with a pseudo-
diffeomorphism η : ∂Y → X and a piecewise smooth mapping ζ : Y → N such that
γ = ζ|∂Y ◦ η −1 , where γ is an oriented singular pseudo-manifold q-cycle, then γ is called
an oriented singular pseudo-manifold q-boundary in N . Denote by Bψ ψ
q (N ) := Bq (X, N )
the group of oriented singular pseudo-manifold q-boundaries in N .
548 S.V. Ludkovsky
(j −R 1)-form on X. Therefore, R R
∗ j−1 dŵ ∗ j−1 dw̄ j−1
R X (γ wj + (−1) j−1 ) = X (γ wj + (−1) j−1 ) + (−1) X dvj−1 =
(γ ∗ w + (−1)j−1 dw̄ ) in the considered here case of X with ∂X = 0.
X j j−1
This holonomy can be generalized in an abstract way for an equivalence class η
of a q-connection (g, w1 , ..., wq ) along a singular oriented pseudo-manifold X of di-
mension q with an H ∞ mapping γ : X → N such that hη (γ) ∈ Ar . Define
Hq+1 (X, Z(Cr )(q + 1)∞ D ) := H
q+1 (X \ S , Z(C )(q + 1)∞ ), where S
X r D X is a singular-
ity of X. If the dimension of X is q, then Hq+1 (X, Z(Cr )(q + 1)∞ ) ∼
= H q (X, A∗ ). Since
D r
codim(SX ) ≥ 2, then Hq (X, A∗r ) has a fundamental class that induces an integration along
the fundamental class isomorphism and Hq (X, A∗r ) ∼ = A∗r . Thus we get the isomorphism
TX : H (X, Z(Cr )(q + 1)D ) → Ar . Therefore, hη (γ) = TXq (γ ∗ (η)) is the holon-
q q+1 ∞ ∗
omy of a q-connection corresponding to an element η ∈ Hq+1 (N, Z(Cr )(q + 1)∞ D ) along
γ : X → N for a singular oriented Ar pseudo-manifold φ : X → N of a real dimension q,
where 2 ≤ r ≤ 3.
54. Twisted cohomology. Consider a twisted sheaf B over {i0 , ..., i2r −1 }. Then a
twisted analog of an Alexander-Spanier (or of isomorphic Čech) cohomology with coeffi-
cients in B and supports in the family φ is AS H∗φ (X; B) = H∗ (Γφ (S ∗ ⊗ B)).
<q+1 <q+1
Take an element η ∈ AS Hq (SN,A r
(dLn)) := AS Hq (X, SN,A r
(dLn)), where N is
of dimension q. Write it as η = (gjq , wjq−1 , ..., wj0 ), where jb := (j0 , ..., jb ) is a multi-
index. The A∗r -valued q-cocycle gjq is cohomologous to zero, since N is of dimension q.
Therefore, Hq (C∞ ∗ ∼ q+1 (N, Z(Cr )) ∼ = 0. If ḡjq−1 is a (q − 1)-cochain such that
N (Ar )) = H
−1
δ(ḡjq−1 ) = gjq , then
(gjq δ(ḡjq−1 ), dLn(ḡjq−1 ) + wjq−1 , ..., wj0 ) = (1, dLn(ḡjq−1 ) + wjq−1 , ..., wj0 ) =: η ′ .
<q+1
Denote by D the differential in the twisted complex of SN,A r
(dLn), then the cocycle
′
condition D(η ) = 0 leads to δ(dLn(ḡjq−1 ) + wjq−1 ) = 0. Since SN,A 1 is an acyclic
r
sheaf, then its twisted complex is exact and inevitably there exists a (q − 2)-cocycle w̄jq−2
for which δ(w̄jq−2 ) = dLn(ḡjq−1 ) + wjq−1 . Then D(1, −w̄jq−2 , 0, ..., 0) + η ′ is coho-
molous to a cocycle having the form (1, 0, w′ jq−1 , ..., wj0 ). Continuing this procedure gives
<q+1
a (q − 1)-cochain η̄ = (ḡjq−1 , w̄jq−2 , ..., w̄j0 ) of the twisted complex SN,A r
(dLn) such that
η + D(η̄) = (1, 0, ..., 0, ŵj0 ). From the cocycle condition D(η + D(η̄)) = 0 it follows that
ŵj0 is a global q-form on N which R
we will denote by ŵ. Put
(1) TNq (η) := exp((−1)q N ŵ).
The mapping TNq of Formula (1) depends only on the cohomology class of η, since
if η = D(η̄), then ŵ = 0. Moreover, TNq is independent from a choice of the chain η̄.
Indeed, if η̃ is another (q − 1)-cochain with ηR+ D(η̃) = (1, 0, ..., 0, νj0 ), then ν − ŵ is an
2πZ(Cr )-integral q-form and inevitably exp( N (ν − ŵ)) = 0.
Define the isomorphism tqN : Hq (N, A∗r ) → A∗r as the restriction of TNq to Hq (N, A∗r )
or it can be written as tqN = TNq ◦ iqN , where iqN : AS Hq (N, A∗r ) → AS Hm (YN,A <q+1
r
(dLn)),
iqN (gjq ) = (gjq , 0, ..., 0) is the monomorphism induced by the morphism of complexes of
<q+1
sheaves (A∗r )N → SN,A r
(dLn).
Now construct an isomorphism eqN : Hq (SN,A <q+1
r
(dLn)) → Hq (N, A∗r ). Consider
the q-cocycle η as above, then dwj0 = 0, since N is of dimension q. This implies
<q+1
an existence of a (q − 1)-cochain η̄ of the twisted complex of SN,A r
(dLn) such that
η + D(η̄) = (ḡjq , 0, ..., 0). The cocycle condition D(η + D(η̄)) = 0 implies that ḡjq is
550 S.V. Ludkovsky
a locally constant A∗r -valued cochain. Then the mapping η 7→ ḡjq induces the isomorphism
eqN : AS Hq (SN,A <q+1
r
(dLn)) → AS Hq (N, A∗r ). This construction implies, that eqN is the
inverse of iqN , hence TNq = tqN ◦ eqN .
55. Theorem. For an H ∞ manifold N over Ar the mapping η 7→ hη (see §54) induces
<q+1
an isomorphism h : Hq (SN,A r
(dLn)) → Ĥq (N, Z(Cr )).
Proof. It is sufficient to show, that the following diagram with the upper row
ip <q+1 d
0 → Hq (N, A∗r ) −→
N
Hq (SN,A r
(dLn)) −→ Sq+1
Ar (N )0 → 0
and the lower row
îp q π
q+1
0 → Hom(Hψ ∗ N 2
q (N ), Ar ) −→ Ĥψ (N, Z(Cr )) −→ SAr (N )0 → 0
and with the vertical rows
u
Hq (N, A∗r ) −→ Hom(Hψ ∗
q (N ), Ar ) and
<q+1 h
Hq (SN,A r
(dLn)) −→ Ĥqψ (N, Z(Cr )) commutes. For each η ∈ Hq (SN,A <q+1
r
(dLn)) put
η η η η
h(η) := (h , K ), where h is the holonomy of η and K denotes the curvature of η. If
η = (gjq , wjq−1 , ..., wj0 ), then K η = K(gjq , wjq−1 , ..., wj0 ) = dwj0 , consequently, the
right hand side of the above diagram commutes.
The universal coefficient theorem isomorphism u : Hq (N, A∗r ) → Hom(Hψ ∗
q (N ), Ar )
∞
is induced by a pairing assigning to a Hp mapping γ : M → N and a cohomology
class η ∈ Hq (N, A∗r ) an octonion or quaternion number tqM ◦ γ ∗ (η), where tqM denotes the
q P
restriction of TM to Hq (N, A∗r ). Therefore, for η ∈ Hq (N, A∗r ) and a q-cycle j nj γj ,
P Q
where γj : M → N we get u(η)( j nj γj ) = tqM ( j γj∗ (η)nj ).
q
From the equalities hη (γ) = u(η)(γ) and TM = tqM ◦ eqM and eqM ◦ iqM = id for an arbi-
q q ∗ q q q ∗
trary Hp∞ mapping γ : M → N it follows that hiN (η) (γ) = TM γ iN (η) = TM iM γ (η) =
q q q ∗ q ∗ q
tM eM iM γ (η) = tM γ (η) = îN u(η)(γ). Since h is the homomorphism, then the left
hand side square of the diagram is commutative as well.
56. Remark. In view of Theorem 55 every element of Ĥqψ (N, Z(Cr )) is a holon-
omy homomorphism. The operator TXq in the definition of the holonomy uses the in-
tegration which is invariant under the equivalence relation 52(E1). Then the quotient
q q
mapping Zψ ψ̃
q (N ) → Zq (N ) induces an isomorphism Ĥψ̃ (N, Z(Cr )) → Ĥψ (N, Z(Cr )),
ξ : π −1 (U ) → G.
Since the group G is twisted, then due to Proposition 19 and Theorem 3.20 it is sufficient
to prove this Lemma for the commutative group G0 .
Consider a deformation retraction η : [0, 1] × V → A of V onto A, where V is an
open neighborhood of A, put U = θ∗ [(W M E; V, G0 , P)t,H ]. A (W M E; A, G0 , P)t,H -
equivariant mapping ξ : (θ∗ )−1 (U ) = (W M E; W, G0 , P)t,H → (W M E; A, G0 , P)t,H is
given by the formula ξ(< Pγ̂,v >t,H ) =< η(1, Pγ̂,v >t,H due to Propositions 7.1 and
13(2) in Section 3.
58. Theorem. For each connected smooth manifold N , the homomorphism κ induces
an isomorphism
κ∗ : π0 ((W M E; N, A∗r , P)b;∞,H ) → Hψ̃
qb (N, Z(Cr )), where 1 ≤ b ∈ N, q is a dimen-
sion of M .
Proof. The uniform space Hp∞ (M, E) is everywhere dense in the uniform space
C (M, E) of all continuous mappings from M into E, where M is an Hp∞ -pseudo-
0
this proposition.
60. Theorem. The restriction homomorphism κ∗ : Hom(Zψ̃ ∗
qb (M, N ), Ar ) →
M ∗
Hom((W N )b;∞,H , Ar ) induces an isomorphism
κ̂ : Ĥqb
ψ̃
(N, Z(Cr )) → Hom∞ ((W M N )b;∞,H , Ar ), where 1 ≤ b ∈ N, q is a covering
dimension of M , M and N are over Ar .
Proof. Each homomorphism h : Zψ̃ ∗ qb
qb (M, N ) → Ar is a holonomy of Ĥψ̃ (N, Z(Cr )),
<qb+1
since the holonomy induces an isomorphism Hqb (SN,A r
(dLn)) → Ĥqb
ψ̃
(N, Z(Cr )). Then
the restriction of h on (W M E)b;∞,H is of H ∞ class, where E = E(N, G, π, Ψ) with G =
CM (see §56.1). Therefore, κ∗ (h) ∈ Hom∞ ((W M N )b;∞,H , A∗r ), since (W M E)b;∞,H is
the principal Gbk -bundle over (W M N )b;∞,H .
Consider an extension ĥ : Zψ̃ ∗ qb
qb (M, N ) → Ar of h, hence ĥ ∈ Ĥψ̃ (N, Z(Cr )).
We have the locally analytic mapping Ln from A∗r onto Ar . The group (W M N )b;∞,H
is commutative, therefore instead of Hom∞ ((W M N )b;∞,H , A∗r ) we can consider the com-
mutative additive group Hom∞ ((W M N )b;∞,H , Ar ), where Ar is considered as the addi-
tive group (Ar , +). At the same time the group Z(Cr ) is commutative. Then Ln(κ∗ (h)) ∈
Hom∞ ((W M N )b;∞,H , Ar ).
For each ξ ∈ Zψ̃ M ψ̃
qb (M, N ) there exists ζ ∈ (W N )b;∞,H and ∂η ∈ Bqb (M, N ) such
that ξ = κ(ζ) + ∂η, since κ∗ : π0 ((W M E; N, A∗r , P)b;∞,H ) → Ĥqb
ψ̃
(N, Z(Cr )) is an
isomorphism due to Theorem 58. Then for each extension ĥ : Zψ̃ ∗
qb (M, N ) → Ar of h there
is the identity:
R
ĥ(ξ) = h(ζ) + exp((−1)qb η K h )
due to Section 53 and Lemma 59. Therefore, h has a unique extension ĥ. This implies that
κ̂ is an isomorphism.
61. Remark. Mention that Theorems 55, 58 and 60 can be proved in another way
using the corresponding statements over C and the twisted structure of sheaves over
{i0 , ..., i2r −1 }.
R : Ty N ×Ty N → R, which is the trace of the linear transformation w → R̂(v1 , w)u2 from
Ty N into Ty N , where R̂ denotes the Riemann curvature tensor and R is its contraction.
Therefore, there exists min{R(v, v) : v ∈ Ty N, y ∈ N, kvk = 1} =: (n − 1)ρ−2 , where n
denotes the dimension of N .
The manifold M̂ is compact, consequently, there exists a finite partition T of M̂ con-
sisting of Uj such that each Uj is homeomorphic with a cube [0, 1]m , while Uj \ ∂Uj is
S
C ∞ diffeomorphic with [0, 1]m \ ∂[0, 1]m , j Uj = M̂ , m denotes the dimension of M̂ ,
Uj ∩ Ul = ∂Uj ∩ ∂Ul , j = 1, ..., a0 , a0 ∈ N.
Consider a path γ̂ : M̂ → N such that γ̂(ŝ0,q ) = y0 and γ̂(ŝ0,q+k ) = y1 for each
q = 1, ..., k, where M̂ is the corresponding C ∞ Riemannian manifold satisfying Conditions
§2 in Section 2 and Ξ : M̂ → M is the quotient mapping as in §2.2, Ξ(ŝ0,q ) = Ξ(ŝ0,q+k ) =
s0,q for each q = 1, ..., k, s0,q and ŝ0,q , ŝ0,q+k are marked points in M and M̂ respectively
for every q = 1, ..., k, k ∈ N. Therefore, the path γ̂ can be presented as the combination of
its restrictions γ̂|Uj .
Without loss of generality we can take a partition T such that each marked point ŝ0,q in
S 0
M̂ belongs to aj=1 ∂Uj . If Uj has less, than two distinct marked points s0,q , then introduce
in Uj additional marked points x0,a,j such that to have not less than two distinct marked
points in Uj . The manifold N has the homotopy type of a CW-complex, hence N b has the
homotopy type of a CW-complex for each b ∈ N (see also [1, 38] and below).
In view of the Sard theorem II.2.10.2 [7] and §III.6 [35] the set of all Hpt dif-
feomorphisms of M̂ is everywhere dense in the uniform space Hpt (M̂ , M̂ ). Then
S
(P M̂ N ; y0 , y1 )t,H has the homotopy type of ( aj=1 0
(P Uj N ; y0,j , y1,j )t,H ) × N 2a0 −2 , where
y0,j , y1,j are 2a0 distinct marked points in N containing y0 , y1 with the corresponding
marked points in Uj .
In accordance with Proposition (H) [51] if L is a locally finite complex and K is a
CW-complex, then K × L is a CW-complex.
The sum of CW-complexes is a CW-complex, the product of CW-complexes is a CW-
complex in accordance with Section 5 and Proposition (H) of [51]. The manifolds M̂ and
N are connected, consequently, it is sufficient to prove this theorem in the special case of
M̂ = [0, 1]m .
Therefore, consider γ̂ : [0, 1]m → N , γ̂(x) ∈ N , x = (x1 , ..., xm ), xj ∈
[0, 1] for each j = 1, ..., m. Suppose that ηs (xs ) is a geodesic between points as
and bs ∈ N , where ηs (xs ) := η(z1 , ..., zs−1 , xs , zs+1 , ..., zm ) with marked values of
z1 , ..., zs−1 , zs+1 , ..., zm ∈ [0, 1] and η : [0, 1]m → N , as = ηs (0), bs = ηs (1). If ηs (xs )
has a length greater than πρ, then it has an index λ ≥ 1 (see also §§16, 17, 19 in [37]).
Let E(ζ) denotes the energy functional of a geodesic in the Riemannian manifold and
E∗∗ be its Hessian (see §12 in [37]).
Generally consider a geodesic ζ of length greater than gπρ, consequently, ζ has an
index λ ≥ g, where g ∈ N. For each j = 1, ..., g there exists a vector field Yj in N such
that Yj along ζ vanishes outside the interval ((j − 1)/k, j/k), and so that E∗∗ (Yj , Yl ) < 0.
Since E∗∗ (Yj , Yl ) = 0 for each j 6= l, then Y1 , ...Yg span a g-dimensional subspace of
S
y∈ζ([0,1]) Ty N on which E∗∗ is negative definite (see §19 in [37]).
Suppose that points y0,j and y1,j are not conjugate along any geodesic from y0,j to y1,j ,
hence there exists only a finite number of geodesics like ηs from y0,j to y1,j in N by the
556 S.V. Ludkovsky
variable xs of length not greater than gπρ. Hence there exists only finitely many geodesics
with index less than g.
In accordance with Theorem 17.3 [37] if N is a complete Riemannian manifold
and y0 , y1 ∈ N are two points, which are not conjugate along any geodesic, then
(P [0,1] N ; y0 , y1 )t,H has the homotopy type of a countable CW-complex containing one
cell of dimension λ for each geodesic from y0 to y1 of index λ.
Together with Theorem 17.3 [37] this completes the proof for dim(M ) = 1. For m > 1
proceed by induction:
m m−1
(P [0,1] N ; y0 , y1 )∞,H = (P [0,1] (P [0,1] N ; y0 , y1 )∞,H ; y0 , y1 )∞,H , where yb in
[0,1] l
(P N ; y0 , y1 )∞,H denotes the constant mapping yb : [0, 1]l → N , yb ([0, 1]l ) = {yb },
{yb } denotes the singleton in N , b = 1, 2, l ∈ N, here the notation yb corresponds to yb,j
for some j.
This procedure lowers a number of variables on each step by one. In view of Theorem
19.6 [37] (P [0,1] N )t,H has the homotopy type of a CW complex B, which is σ-compact,
that is a countable union of compact sets.
Consider now (P [0,1] B)t,H , where B is a countable union of compact Riemannian man-
ifolds may be with corners, since each polyhedron in Rn with n ∈ N is a manifold with
S S
corners. Put B = j∈Λ Bj , B k := kj=1 Bj , where Bj is a compact Riemannian manifold
with corners being a j-skeleton of a CW complex, Λ ⊂ N. Up to a homotopy type or bend-
ing Bj a little in the corresponding Euclidean space Rn of dimension n ≥ 2 dim (Bj ),
Bj ֒→ Rj ֒→ Rn , we can consider, that each Bj is homotopy equivalent to a compact Rie-
mannian manifold Xj with positive definite Ricci tensor. Therefore, we have to consider
S S
now (P [0,1] X)t,H , where X = j Xj . Put X j = k≤j Xk , then X j ⊂ X j+1 for each
j ∈ Λ, dim (Xj ) = j.
Each path from the compact manifold M into a CW-complex B has a compact image,
consequently, it has a finite covering by cells. Hence a continuous path from M into X up
to a homotopy equivalence has a finite covering by X j .
If N1 and N2 are homotopy equivalent Riemannian manifolds, then
(P [0,1] N1 ; y0,1 , y1,1 )t,H and (P [0,1] N2 ; y0,2 , y1,2 )t,H are homotopy equivalent, when y0,1 6=
y0,2 and y0,2 6= y1,2 simultaneously. On the other hand, (P [0,1] X; y0 , y1 )t,H is homotopy
S
equivalent with a CW-complex K = j∈Λ Kj , where each Kj is a CW-complex homotopy
equivalent with (P [0,1] X j ; y0 , y1 )t,H , where y0 , y1 ∈ X1 , so that Kj ⊂ Kj+1 for each j,
since X j ⊂ X j+1 .
Denote by W the class of all spaces having the homotopy type of a CW-complex. By
a CW-n-ad K = (K; K1 , ..., Kn−1 ) is undermined a CW-complex together with (n − 1)
numbered sub-complexes K1 , ..., Kn−1 . Then W n denotes the class of all n-ads which
have the homotopy type of a CW-n-ad. As usually AC denotes the subspace of the space
AC of all continuous functions f from A into C such that f : C → A is a mapping of n-ads,
that is the induced mappings are fj : Cj → Aj from the j-skeleton to the j-skeleton for
each 1 ≤ j ≤ n.
In accordance with Theorem 3 [38] if A belongs to the class W n and C is
a compact n-ad, then the function space AC belongs to W. In fact the n-ad
(AC ; (A, A1 )(C,C1 ) , ..., (A, An−1 )(C,Cn−1 ) ) belongs to the class W n .
Thus, (P M̂ N ; y0 , y1 )t,H has the homotopy type of the CW-complex.
Wrap Groups of Connected Fiber Bundles, Their Structure and Cohomologies 557
′
2. Corollary. If M and M̂ and N are manifolds Hpt and Hpt diffeomorphic with C ∞
Riemannian manifolds M1 and M̂1 and N correspondingly, t′ ≥ t, where M1 , M̂1 and
N1 satisfy conditions of the preceding theorem, then the path space (P M̂ N ; y0 , y1 )t,H and
the quotient path space (P M N ; y0 , y1 )t,H for marked points y0 and y1 in N are of the
homotopy types of CW-complexes having only finitely many cells in each dimension.
′
Proof. Let φ : M̂1 → M̂ and θ : N1 → N be homeomorphisms, which are Hpt and Hpt
diffeomorphisms. Then the uniform spaces (P M̂ N ; y0 , y1 )t,H and (P M̂1 N1 ; y0,1 , y1,1 )t,H
are isomorphic, where the mapping f 7→ θ−1 ◦ f ◦ φ establishes the isomorphism,
f ∈ (P M̂ N ; y0 , y1 )t,H , θ(yb,1 ) = yb for b = 1, 2. Using this isomorphism and
applying the preceding theorem to (P M̂1 N1 ; y0,1 , y1,1 )t,H and the quotient path space
(P M1 N1 ; y0,1 , y1,1 )t,H we get the statement of this corollary.
3. Corollary. Let M and N be satisfying conditions of the preceding Corollary. Then
the wrap monoid (S M N )t,H and the wrap group (W M N )t,H have homotopy types of CW-
complexes having only finitely many cells in each dimension.
Proof. The wrap monoid has the homotopy type of (P M N ; y0 , y0 )t,H . On the other
hand, the wrap group is the quotient of the free commutative group F generated by
(S M N )t,H by the closed equivalence relation, which is obtained factorizing by the min-
imal closed normal subgroup B containing all elements of the form [a + b] − [a] − [b],
where a, b ∈ (S M N )t,H , [a] and [b] are the corresponding elements of F . Topologically
F is isomorphic with [(S M N )t,H ]Z supplied with the weak (Tychonoff) product topology.
Applying Corollary on page 153 from Theorem A [37] and the preceding theorem we get
the statement of this corollary.
4. Corollary. Let M and N be satisfying conditions of Corollary 2, while E be a
principal fibre bundle with the structure group G, which is up to the homotopy a CW-group.
Then a wrap monoid (S M E)t,H and a wrap group (W M E)t,H have homotopy types of a
CW-monoid and a CW-group correspondingly.
Proof. By Proposition (N ) any covering complex of a CW-complex is a CW-complex
[51]. Therefore, if prove that (S M E)t,H is a CW-complex, then it would mean that
(W M E)t,H is a CW-complex. This follows immediately from the preceding corollary and
Proposition 7.1 in Section 3 and Proposition (H) [51], since (S M E)t,H and (W M E)t,H
have structures of principal Gk -bundles over (S M N )t,H and (W M N )t,H .
On the other, hand the mapping (S M N )t,H ∋ (f, g) → f g ∈ (S M N )t,H is cellular,
since if a, b ∈ K n , then a ∨ b ∈ K n ∨ K n , where the bunch K n ∨ K n of K n by a
finite number of marked points consists of cells of dimension at most n. Therefore, in
(W M N )t,H the group multiplication is cellular as well (see also §3). In (W M N )t,H the
mapping f 7→ f −1 is cellular due to the definition of the wrap group. Since G is the
CW-group, then Gk is the CW-group, consequently, (S M E)t,H and (W M E)t,H are the
CW-monoid and the CW-group respectively.
5. Remark. A topological space P is said to be dominating a topological space X if
and only if there are continuous mappings f : X → P and g : P → X such that gf ≈ 1X .
In accordance with Theorem 1 [38] A belongs to the class W0 if and only if A is dominated
by a countable CW-complex.
If G is a compact simply connected Lie group, then in accordance with Theorem 21.7
[37] (P [0,1] G; y0 , y1 )t,H has the homotopy type of a CW-complex with no odd-dimensional
558 S.V. Ludkovsky
cells and with only finite number of n-cells for each even number n. These two theorems
imply that G also is a CW-group, since (P [0,1] G)t,H dominates G and applying the homo-
topy equivalence.
If G is not associative, but alternative, then the corresponding CW-group is alternative
as well, since if a1 ≈ a2 , b1 ≈ b2 are homotopy equivalent elements of G, then (a1 a1 )b1 =
a1 (a1 b1 ) ≈ a1 (a2 b2 ) ≈ a2 (a2 b2 ) = (a2 a2 )b2 and b1 = a−1 −1
1 (a1 b1 ) ≈ a1 (a2 b2 ) ≈
a−1 −1
2 (a2 b2 ) = (a2 a2 )b2 = b2 and analogously for identities with aj on the right from bj .
In accordance with Corollary 1 [38] every separable finite dimensional manifold be-
longs to the class W0 , where W0 denotes the class of topological spaces having the homo-
topy type of countable CW-complexes. Due to Corollary 2 [38] if A belongs to W0 and C is
a compact metric space, then the function space AC in the compact open topology belongs
to W0 . Therefore, modifying Theorem 2 and Corollary 4 we get.
6. Proposition. If N is a finite dimensional separable manifold, G is a CW-group, then
(P M E; y0 , y1 )t,H has the homotopy type of a CW-complex, (S M E)t,H and (W M E)t,H
have homotopy types of a CW-monoid and a CW-group respectively.
References
[1] E.J. Beggs. ”The de Rham complex of infinite dimensional manifolds”. Quart. J.
Math. Oxford (2) 38 (1987), 131-154.
[2] R. Bott, L.W Tu. ”Differential forms in algebraic topology” (New York: Springer-
Verlag, 1982).
[4] J.-L. Brylinski. ”Loop spaces, charateristic classes, and geometric quantization”, PM
107 (Boston: Birkhäuser, 1992).
[5] Y.H. Ding, J.Z. Pang. ”Computing degree of maps between manifolds”. Acta Mathem.
Sinica. English Series. 21: 6 (2005), 1277-1284.
[7] B.A. Dubrovin, S.P. Novikov, A.T. Fomenko. ”Modern geometry” (Moscow: Nauka,
1979).
[8] D.G. Ebin, J. Marsden. ”Groups of diffeomorphisms and the motion of incompressible
fluid”. Ann. of Math. 92 (1970), 102-163.
[9] J. Eichhorn. ”The manifold structure of maps between open manifolds”. Ann. Glob.
Anal. Geom. 3 (1993), 253-300.
[10] H.I. Eliasson. ”Geometry of manifolds of maps”. J. Differ. Geom. 1 (1967), 169-194.
[14] P. Gajer. ”Higher holonomies, geometric loop groups and smooth Deligne cohomol-
ogy”. In: ”Advances in geometry”. Progr. in Math. 172, 195-235 (Boston: Birkhäuser,
1999).
[15] P. Gajer. ”Geometry of Deligne cohomology”. Invent. Mathem. 127 (1997), 155-207.
[16] F. Gürsey, C.-H. Tze. ”On the role of division, Jordan and related algebras in particle
physics” (Singapore: World Scientific Publ. Co., 1996).
[17] W.R. Hamilton. ”Selected papers. Optics. Dynamics. Quaternions” (Moscow: Nauka,
1994).
[18] F.R. Harvey. ”Spinors and calibrations”. Perspectives in Mathem. 9 (Boston: Aca-
demic Press, 1990).
[19] P. Iglesias. ”Difféologie d’espace singulier et petits diviseurs”. C.R. Acad. Sci. Paris.
Ser. I. 302: 14 (1986), 519-522.
[20] C.J. Isham. ”Topological and global aspects of quantum theory”. In: ”Relativity,
groups and topology.II” 1059-1290, (Les Hauches, 1983). Editors: R. Stora, B.S. De
Witt (Amsterdam: Elsevier Sci. Publ., 1984).
[23] H.B. Lawson, M.-L. Michelson. ”Spin geometry” (Princeton: Princ. Univ. Press,
1989).
[25] S.V. Ludkovsky. ”Poisson measures for topological groups and their representations”.
Southeast Asian Bulletin of Mathematics. 25 (2002), 653-680. (shortly in Russ. Math.
Surv. 56: 1 (2001), 169-170; previous versions: IHES/M/98/88, 38 pages, also Los
Alamos Nat. Lab. math.RT/9910110).
[26] S.V. Ludkovsky. ”Functions of several Cayley-Dickson variables and manifolds over
them”. J. Mathem. Sci. 141: 3 (2007), 1299-1330 (previous variant: Los Alamos Nat.
Lab. math.CV/0302011).
[29] S.V. Ludkovsky. ”Differentiable functions of Cayley-Dickson numbers and line inte-
gration”. J. Mathem. Sci. 141: 3 (2007), 1231-1298 (previous version: Los Alam. Nat.
Lab. math.NT/0406048; math.CV/0406306; math.CV/0405471).
[31] S.V. Ludkovsky. ”Geometric loop groups and diffeomorphism groups of manifolds,
stochastic processes on them, associated unitary representations”. In the book: ”Focus
on Groups Theory Research” (Nova Science Publishers, Inc.: New York) 2006, pages
59-136.
[32] S.V. Ludkovsky. ”Generalized geometric loop groups of complex manifolds, Gaussian
quasi-invariant measures on them and their representations”. J. Mathem. Sci. 122: 1
(2004), 2984-3011 (earlier version: Los Alam. Nat. Lab. math.RT/9910086, October
1999).
[34] M.B. Mensky. ”The paths group. Measurement. Fields. Particles” (Moscow: Nauka,
1983).
[35] V.P. Mihailov. ”Differential equations in partial derivatives” (Moscow: Nauka, 1976).
[37] J. Milnor. ”Morse theory” (Princeton, New Jersey: Princeton Univ. Press, 1963).
[38] J. Milnor. ”On spaces having the homotopy type of a CW-complex”. Transactions of
the A.M.S. 90 (1959), 272-280.
[41] H. Omori. ”Groups of diffeomorphisms and their subgroups”. Trans. Amer. Math. Soc.
179 (1973), 85-122.
[42] H. Omori. ”Local structures of groups of diffeomorphisms”. J. Math. Soc. Japan 24:
1 (1972), 60-88.
[44] R.T. Seeley. ”Extensions of C ∞ functions defined in a half space”. Proceed. Amer.
Math. Soc. 15 (1964), 625-626.
[46] N. Steenrod. ”The topology of fibre budles” (Princeton, New Jersey: Princeton Univ.
Press, 1951).
[48] R.C. Swan. ”The Grothendieck ring of a finite group”. Topology 2 (1963), 85-110.
[49] R.M. Switzer. ”Algebraic topology - homotopy and homology” (Berlin: Springer-
Verlag, 1975).
[51] J.H.C. Whitehead. ”Combinatorial homotopy.I”. Bull. Amer. Mathem. Soc. 55 (1949),
213-245.
[52] K. Yano, M. Ako. ”An affine connection in almost quaternion manifolds”. J. Differ.
Geom. 3 (1973), 341-347.
Chapter 18
S.V. Ludkovsky
Dept. of Applied Mathematics, Moscow State Technical Univ.,
Moscow, Russia
Abstract
The article is devoted to the investigation of groups of diffeomorphisms and wraps
of manifolds over non-archimedean fields of zero and positive characteristics. Dif-
ferent types of topologies are considered on groups of wraps and diffeomorphisms
relative to which they are generalized Lie groups or topological groups. Among such
topologies pairwise incomparable are found as well. Topological perfectness of the
diffeomorphism group relative to certain topologies is studied. There are proved theo-
rems about projective limit decompositions of these groups and their compactifications
for compact manifolds. Moreover, an existence of one-parameter local subgroups of
diffeomorphism groups is investigated.
1. Introduction
Non-archimedean analysis has rather long history, but it is much less developed in compar-
ison with the classical analysis over the fields R and C. Therefore, the theory of groups
on manifolds over non-archimedean fields is not so well investigated as for Riemann or
complex manifolds [35, 16, 12, 5].
As it is known fields with multiplicative ultra-norms such as the field of p-adic num-
bers were first introduced by K. Hensel [14]. Several years later on it was proved by A.
Ostrowski [33] that on the field of rational numbers each multiplicative norm is either
the usual norm as in R or is equivalent to a non archimedean norm |x| = p−k , where
x = npk /m ∈ Q, n, m, k ∈ Z, p ≥ 2 is a prime number, n and m and p are mutually pair-
wise prime numbers. Each locally compact infinite field with a non trivial non archimedean
valuation is either a finite algebraic extension of the field of p-adic numbers or is isomor-
phic to the field Fpk (θ) of power series of the variable θ with expansion coefficients in the
finite field Fpk of pk elements, where p ≥ 2 is a prime number, k ∈ N is a natural number
[38, 43]. The valuation group ΓK := {|x| : x ∈ K, x 6= 0} ⊂ (0, ∞) of a locally compact
564 S.V. Ludkovsky
field K is discrete. Non locally compact fields are wide spread as well and among them
there are fields with the valuation group ΓK = (0, ∞) [9, 38, 40]. The non-archimedean
analog of the field of complex numbers is Cp which is complete algebraically and as the
uniform space relative to its multiplicative norm and ΓCp = {x ∈ Q : x > 0}.
The importance of transformation groups of manifolds in the non-Archimedean func-
tional analysis, representation theory and mathematical physics is clear and also can be
found in the references given below [2, 21, 22, 24, 25, 38, 39]. Here we use the termi-
nology a wrap group instead of a (geometric) loop group, because in the non-archimedean
case there are not any circles like as in classical case, moreover, manifolds are already to-
tally disconnected and of arbitrary dimension over a non-archimedean field. It was shown
earlier in preceding works of the author that such groups are smooth manifolds with differ-
entiable group operations depending on a considered class of smoothness, that is they are
Lie groups.
This article is devoted to several aspect of such groups and to further investigation
of their structure. One of them is on their structure from the point of view of the non-
archimedean compactficiation (see also about Banaschewski compactification in [38]).
Though a new topology used for compactification may be different from the initial topol-
ogy of a group or even may be non-comparable, because on the same group it is possible
an existence of several different topologies making it a topological group. This is useful
also for studying their representations as restrictions of representations of non-archimedean
compactifications, which are constructed below such that they also are groups. Apart from
previous works [21, 22, 23, 24, 25], where the characteristic char(K) = 0 was zero, in
this paper groups on manifolds over fields with non-zero characteristics also are defined
and investigated. Different types of topologies are considered on groups of wraps and
diffeomorphisms relative to which they are generalized Lie groups or topological groups.
Among such topologies pairwise incomparable are found as well. It is caused by the fact
that repeated application of projective and inductive limits of topological spaces generate
topologies and spaces in general dependent from an order of taking limits and their types,
so that such topologies may appear incomparable on a subset contained in these topological
spaces. It is proved, that relative to the C ∞ bounded-open topology groups of geometric
wraps and groups of diffeomorphisms of manifolds over non-archimedean ultra-normed
fields are the generalized Lie groups.
Previously one-parameter subgroups over fields of non-zero characteristics were not
studied. This article contains as well results on one-parameter subgroups over K with
char(K) = p > 1 using its multiplicative subgroup K∗ := K \ {0}. It is proved below that
the diffeomorphism group of a compact manifold is topologically simple relative to the C ∞
compact-open topology, that develops previous results [24], where topological simplicity
and perfectness was proved over fields of zero characteristic.
At first in Section 2 we remind basic facts and notations, which are given in detail in
references [38, 40, 21, 22, 23]. A wrap group Lt (M, N ) is defined as a quotient space of
a family of mappings f : M → N of class C t of one Banach manifold M into another
N over the same local field K such that limz→s (Φ̄m f )(z; h1, ..., hm; ζ1, ..., ζm) = 0 or
limz→s (Υm f )(z [m]) = 0 for each 0 ≤ m ≤ t, where M and N are embedded into the
corresponding Banach spaces X and Y , cl(M ) = M ∪ {s}, cl(M ) and N are clopen
in X and Y respectively, 0 ∈ N , (Φ̄m f )(z; h1, ..., hm; ζ1, .., ζm) and (Υm f )(z [m] ) are
Groups of Diffeomorphisms and Wraps of Manifolds... 565
2. Groups of Diffeomorphisms
1. Definitions. Let K be an infinite field with a non trivial non archimedean valuation, let
also X and Y be topological vector spaces over K and U be an open subset in X. For a
function f : U → Y consider the associated function
f [1](x, v, t) := [f (x + tv) − f (x)]/t
on a set U [1] at first for t 6= 0 such that U [1] := {(x, v, t) ∈ X 2 × K, x ∈ U, x + tv ∈ U }.
If f is continuous on U and f [1] has a continuous extension on U [1] , then we say, that
566 S.V. Ludkovsky
f is continuously differentiable or belongs to the class C 1 . The K-linear space of all such
continuously differentiable functions f on U is denoted C [1] (U, Y ). By induction we define
functions f [n+1] := (f [n] )[1] and spaces C [n+1] (U, Y ) for n = 1, 2, 3, ..., where f [0] := f ,
f [n+1] ∈ C [n+1] (U, Y ) has as the domain U [n+1] := (U [n] )[1].
The differential df (x) : X → Y is defined as df (x)v := f [1](x, v, 0).
Define also partial difference quotient operators Φn by variables corresponding to x
only such that
Φ1f (x; v; t) = f [1](x, v, t)
at first for t 6= 0 and if Φ1f is continuous for t 6= 0 and has a continuous extension on
U [1] =: U (1), then we denote it by Φ̄1 f (x; v; t). Define by induction
Φn+1 f (x; v1, ..., vn+1; t1, ..., tn+1) := Φ1 (Φn f (x; v1, ..., vn; t1 , ..., tn))(x; vn+1; tn+1 )
at first for t1 6= 0, ..., tn+1 6= 0 on U (n+1) := {(x; v1, ..., vn+1; t1, ..., tn+1) : x ∈
U ; v1, ..., vn+1 ∈ X; t1, ..., tn+1 ∈ K; x + v1 t1 ∈ U, ..., x + v1t1 + ... + vn+1 tn+1 ∈ U }.
If f is continuous on U and partial difference quotients Φ1 f ,...,Φn+1 f has continuous
extensions denoted by Φ̄1f ,..., Φ̄n+1 f on U (1),...,U (n+1) respectively, then we say that f
is of class of smoothness C n+1 . The K linear space of all C n+1 functions on U is denoted
by C n+1 (U, Y ), where Φ0 f := f , C 0 (U, Y ) is the space of all continuous functions
f :U →Y.
Then the differential is given by the equation dn f (x).(v1, ..., vn) :=
n!Φ̄n f (x; v1, ..., vn; 0, ..., 0), where n ≥ 1, also denote Dn f = dn f . Shortly we
shall write the argument of f [n] as x[n] ∈ U [n] and of Φ̄n f as x(n) ∈ U (n) , where
x[0] = x(0) = x, x[1] = x(1) = (x, v, t), v [0] = v (0) = v, t1 = t, x[k] = (x[k−1], v [k−1], tk )
for each k ≥ 1, x(k) := (x; v1, ..., vk; t1 , ..., tk).
Subspaces of uniformly C n or C [n] bounded continuous functions together with Φ̄k f
or Υk f on bounded open subsets of U and U (k) or U [k] for k = 1, ..., n we denote by
[n]
Cbn (U, Y ) or Cb (U, Y ) respectively.
Consider partial difference quotients of products and compositions of functions
and relations between partial difference quotients and differentiability of both types.
Denote by L(X, Y ) the space of all continuous K-linear mappings A : X →
Y. By Ln (X ⊗n , Y ) denote the space of all continuous K n-linear mappings
A : X ⊗n → Y , particularly, L(X, Y ) = L1 (X ⊗1, Y ). If X and Y are
normed spaces, then Ln (X ⊗n , Y ) is supplied with the operator norm: kAk :=
suph1 6=0,...,hn 6=0;h1 ,...,hn ∈X kA.(h1, ..., hn)kY /(kh1kX ...khnkX ).
2. Lemma. The spaces C [1] (U, Y ) and C 1 (U, Y ) are linearly topologically iso-
morphic. If f ∈ C n (U, Y ), then Φ̄n f (x; ∗; 0, ..., 0) : X ⊗n → Y is a K n-linear
C 0 (U, Ln (X ⊗n , Y )) symmetric map.
Proof. From Definition 1 it follows, that f [1](x, v, t) = Φ̄1 f (x; v; t) on U [1] = U (1) ,
so both K-linear spaces are linearly topologically isomorphic. On the other hand, due to
its definition Φ̄n f (x; ∗, 0, ..., 0) is the K n-linear symmetric mapping for each x ∈ U and
it belongs to C 0 (U, Ln(X ⊗n , Y )), since Φ̄n f (x; v1, ..., vn; t1 , ..., tn) is continuous on U (n)
and for each x ∈ U and v1 , ..., vn ∈ X there exist neighborhoods Vi of vi in X and W of
zero in K such that x + W V1 + ... + W Vn ⊂ U .
3. Lemma. Operators Υn (f ) := f [n] from C [n] (U, Y ) into C 0 (U [n] , Y ) and Φ̄n :
C (U, Y ) → C 0 (U (n) , Y ) are K-linear and continuous.
n
Proof. Since [(af + bg)(x + vt) − (af + bg)(x)]/t = a(f (x + vt) − f (x))/t + b(g(x +
Groups of Diffeomorphisms and Wraps of Manifolds... 567
vt) − g(x))/t for each f, g ∈ C 1 (U, Y ) and each a, b ∈ K, then applying this formula by
induction and using definitions of operators Υn and Φ̄n we get their K-linearity. Indeed,
Υn (af + bg)(x[n]) = Υ1 (Υn−1 (af + bg)(x[n−1]))(x[n]) = Υ1 (af [n−1] +
bg [n−1])(x[n]) = af [n] (x[n] ) + bg [n](x[n] ) and
Φ̄n (af + bg)(x(n)) = Φ̄1 (Φ̄n−1 (af + bg)(x(n−1)))(x(n)) = Φ̄1(af (n−1) +
bg (n−1))(x(n)) = af (n) (x(n) ) + bg (n)(x(n) ).
The continuity of Υn and Φ̄n follows from definitions of spaces C [n] (U, Y ) and C n (U, Y )
respectively.
4. Definitions. Let M be a manifold modelled on a topological vector space X over K
0
such that its atlas At(M ) := {(Uj , M φj ) : j ∈ ΛM } is of class Cβα0 , that is the following
four conditions are satisfied:
(M 1) {Uj : j ∈ ΛM } is an open covering of M , Uj = M Uj ,
S
(M 2) j∈ΛM Uj = M ,
(M 3) M φj := φj : Uj → φj (Uj ) is a homeomorphism for each j ∈ ΛM , φj (Uj ) ⊂ X,
(M 4) φj ◦ φ−1 α0
i ∈ Cβ 0 on its domain for each Ui ∩ Uj 6= ∅,
T [∞] T
where ΛM is a set, C ∞ := ∞ l
l=1 Cβ , Cβ := ∞ [l] 0
l=1 C , α ∈ {n, [n] : 1 ≤ n ≤ ∞},
0 0
β ∈ {∅, b}, C∅α := C α .
Supply Cβα (U, Y ) with the bounded-open Cβα topology (denoted by τα,β generally or
τα for β = ∅ or for compact U ) with the base W (P, V ) = {f ∈ Cβα (X, Y ) : S k f |P ∈
V, k = 0, ..., n} of neighborhoods of zero, where P is bounded and open in U ⊂ X,
P ⊂ U , V is open in Y , 0 ∈ V , S k = Φ̄k or S k = Υk for α = n or α = [n] respectively,
[k]
v1 , ..., vn ∈ (P − y0 ), vl ∈ (P − y0 ) for each k, l for some marked y0 ∈ P and |tj | ≤ 1
for every j.
0
If M and N are Cβα manifolds on topological vector spaces X and Y over K, then
consider the uniform space Cβα (M, N ) of all mappings f : M → N such that fj,i ∈ Cβα on
its domain for each j ∈ ΛN , i ∈ ΛM , where fj,i := N φj ◦ f ◦ M φ−1 i is with values in Y ,
α ≤ α0 . The uniformity in Cβα (M, N ) is inherited from the uniformity in Cβα (X, Y ) with
the help of charts of atlases of M and N . If M is compact, then Cbα (M, N ) and C α (M, N )
coincide.
The family of all homeomorphisms f : M → M of class Cβα denote by Dif fβα(M ).
Let γ be a set, then denote by c0 (γ, K) the normed space consisting of all vectors
x = {xj ∈ K : j ∈ γ, for each > 0 the set {j : |xj | > } is finite }, where kxk :=
supj∈γ |xj |. In view of the Kuratowski-Zorn lemma it is convenient to consider γ as an
ordinal. Henceforth, suppose that X = c0(γX , K) and Y = c0 (γY , K).
5. Theorem. The uniform space Dif fbα (M ) (see §4) is the topological group relative
to compositions of mappings.
Proof. The group operation in Dif fβα(M ) is (f, g) 7→ f ◦g, where f ◦g(x) := f (g(x))
for each x ∈ M . Then f = id is the unit element in Dif fβα(M ), where id(x) = x for each
x ∈ M . Since the composition of mappings is associative, then f ◦ (g ◦ h) = (f ◦ g) ◦ h
is associative as the group operation. For each f ∈ Dif fβα there exists its inverse mapping
f −1 such that f −1 (y) = x for each y = f (x), x ∈ M , since f : M → M is the
homeomorphism. It remains to verify that f −1 ∈ Dif fβα (M ) for each f ∈ Dif fβα (M )
and the composition (Dif fβα)2 3 (f, g) 7→ f ◦ g ∈ Dif fβα(M ) and inversion f 7→ f −1
are continuous operations.
568 S.V. Ludkovsky
then Υk φi ◦ f ◦ φ−1j (V
[k] ) is bounded and kf k k
i,j n,U,X < ∞, analogously for Φ̄ f . Thus
f ∈ Cbα (M, M ), consequently, f ∈ Dif fbα (M ).
9. Theorem. If a manifold M has a finite atlas having clopen bounded φj (Uj ) ⊂ X,
0 ≤ α < ∞, then Dif fbα (M ) is metrizable by a left invariant metric.
Proof. The metrization Theorem 8.3 [16] states that if G is a T0 topological group,
then G is metrizable if and only if there is a countable open basis at e. In this case,
the metric can be taken left-invariant. If At(M ) is finite such that each φj (Uj ) is a
clopen bounded subset in X, then the base of neighborhoods of id in Dif fbα (M ) is
countable and Dif fbα(M ) is metrizable by a left invariant metric in accordance with the
general metrization theorem. Practically take as the metric ρ(f, g) := ρ(id, f −1g) :=
maxi,j∈ΛM kidi,j − (f −1 g)i,j kCbα (Vj ,X), where Vj := φj (Uj ).
10. Remark. As it is known left and right uniformities in a topological group may be
different. Here a left uniformity was considered above. On the same group there may exist
several topologies supplying it with structures of a topological group and these topologies
need not be comparable.
k ∈ N.
Decompositions of continuous and differentiable functions on compact subsets of lo-
cally compact fields K of zero and non-zero characteristics with values in K into series of
polynomials were studied in [1, 40, 7] and references therein. Each function f ∈ C t (M, N )
has a C t (B(Km , 0, 1), Kn)-extension by zero on B(Km , 0, 1). Therefore, it has the decom-
position
P l
(1) f = l,m fm Q̄m el
in the Amice’s basis Q̄m , where el is the standard orthonormal in the non-archimedean
sense [38] basis in Kn such that el = (0, ..., 0, 1, 0, ...) with 1 in the l-th place,
m ∈ Zn , ml ≥ 0, m = (m1, ..., mn), fm l ∈ K are expansion coefficients such that
Proof. In accordance with appendix in [4] and §2.5 [11] and Formulas 2(1−6) above the
spaces C 0 (M, N ) and Cw (M, N ) coincide algebraically, since the connecting mappings
πnm are uniformly continuous for each m ≥ n. The space C 0 (M, Kn) is K-linear and its
uniformity is completely determined by a neighborhood base of zero. The space Cw (M, N )
is uniformly homeomorphic with pr − limk (Spk )Mk , which is compact by the Tychonoff
Theorem 3.2.4 [11]. Since C 0 (M, N ) 6= C α (M, N ) for α > 0, then Cw (M, N ) and
C α (M, N ) are different algebraically see Formulas 2(1 − 3)).
Remark. In general in two consequtive projective limits of topological spaces two
limits may be non commuting.
6. Suppose that Dif fw (M ) := pr − limk Hom(Mk ) is supplied with the uniformity
inherited from Cw (M, M ). The group Dif fw (M ) is called the non-archimedean compact-
ification of Dif f α(M ).
Theorem. The group Dif fw (M ) is the compact topological group and it is the com-
pactification of Dif f α (M ) in the projective weak topology τw . If α > 0, then Dif f α (M )
does not coincide with Dif fw (M ).
Proof. From Dif f α (M ) ⊂ C α (M, M ) it follows that Dif f α(M ) has the corre-
sponding algebraic embedding into Cw (M, M ) as the set. Since Cw (M, M ) is compact
and Hom(M ) is a closed subset in Cw (M, M ), then due to Corollary 5 Hom(M ) ∩
Cw (M, M ) = Dif fw (M ) is compact. The topological space C α (M, M ) is dense
in C 0 (M, M ), consequently, Dif f α(M ) is dense in Dif fw (M ). If α > 0, then
Dif f α (M ) 6= Hom(M ), hence Dif f α(M ) and Dif fw (M ) do not coincide alge-
braically. It remains to verify, that Dif fw (M ) is the topological group in its projective
weak topology τw . If f, g ∈ C α (M, N ), then πk∗(Q̄m (g(x))) = Q̄m,k (gk (x(k)) due to
P
Formula 2(4), consequently, πk∗(f ◦ g) = l,m πk (fm l )Q̄
m,k (gk (x(k))el , hence
(1) (f ◦ g)k = fk ◦ gk .
Since πk (x) = x(k), then πk∗(id(x)) = idk (x(k)), where id(x) = x for each x ∈ M .
Therefore, for f = g −1 we have (f ◦ g)k = fk ◦ gk = idk , hence
(2) πk∗(g −1) = gk−1.
The associativity of the composition (fk ◦ gk ) ◦ hk = fk ◦ (gk ◦ hk ) of all func-
tions fk , gk , hk ∈ Hom(Mk ) together with others properties given above means, that
Dif fw (M ) is the algebraic group.
Indeed, inverse limits of mappings f = pr − limk fk , g = pr − limk gk and h =
pr − limk hk satisfy the associativity axiom as well, each f has the inverse element f −1 =
pr − limk fk−1 such that f −1 (f (x)) = id and e = id is the unit element. By the definition
of the weak topology in Dif fw (M ) for each neighborhood of e = id in Dif fw (M ) there
exists k ∈ N and a subset Wk ⊂ Hom(Mk ) such that ek ∈ πk−1 (Wk ) ⊂ W . On the
other hand, Hom(Mk ) is discrete, hence there are ek ∈ Vk ⊂ Hom(Mk ) and ek ∈ Uk ⊂
Hom(Mk ) such that Vk Uk ⊂ Wk , hence there are neighborhoods e ∈ V ⊂ Dif fw (M )
and e ∈ U ⊂ Dif fw (M ) such that V U ⊂ W , where V = πk−1 (Vk ), U = πk−1(Uk ) and
V U = {h : h = f ◦ g, f ∈ V, g ∈ U }. Consider a neighborhood W 0 of f −1 , then V :=
W 0 f −1 is the neighborhood of e and there exists k ∈ N such that πk−1(ek ) =: U ⊂ V −1 ,
since e−1
k = ek and πk is the homomorphism. Thus, f U := W is the neighborhood of f
such that W −1 ⊂ W 0 , hence the inversion operation f 7→ f −1 is continuous.
7. Theorem. The initial C α topology τα and the weak projective limit topology τw in
Dif f α (M ) are incomparable, where M is compact.
Groups of Diffeomorphisms and Wraps of Manifolds... 575
∞
! ∞
!
X k − l1 X k − l1 − l2
[ fm − δl1 ,0 (n − 1)][ fm − δl2 ,0 (n − 2)]...
m=0
m − l1 m=0
m − l2
∞
!
X k − l1 − ... − ln−1
[ fm − δln−1 ,0 ]fln ,
m=0
m − ln−1
so that it is necessary to evaluate coefficients
! ! ! !
X k k − l1 k − l1 − ... − ln−2 k − l1
(14) Ωk,n
m1 ,...,mn := ...
l1 +...+ln−1 =k−mn
l1 l2 ln−1 m1 − l1
! !
k − l1 − l2 k − l1 − ... − ln−1
.. .
m2 − l2 mn−1 − ln−1
There are identities:
(15) ml = (−1)l l−m−1 l ,
P b ac+e d
(16) (1 + x) (1 + (1 + x)a)b =
e
c d x , such that for e = b and a = −1 this
gives ! ! ! !
X k − l1 k − l1 − l2 k k − l1
(17) =
l ,l
m1 − l1 m2 − l2 l1 l2
1 2
! ! !
X k − l1 k k − l1 m 2
2 =
l1
m1 − l1 l1 m2
! ! !
k X m1 k − l1 m2
2 , since
m1 l l1 m2
1
! !
X b−l b X
= b!/[(m − l)!(b − m + l)!l!].
l
m−l l l
General expressions are too complicated, but it is suffucient to construct a generating func-
tion for these coefficients:
m m +mn
(18) y1m1 ...yn−2
n−2 n−1
yn−1 (x1 y2 ...yn + (y1 + z2 )(x2 y3 ...yn−1 + (y2 + z3 )(x3 y4 ...yn−1 +
Groups of Diffeomorphisms and Wraps of Manifolds... 579
X
m m +mn k k − l1 k − l1 k − l1 − l2
y3m3 ...yn−2
n−2 n−1
yn−1 (x1 y3 ...yn−1)l1 y1k+m1 −q1
l1 q1 − l 1 l2 q2 − l 2
l1 ,q1 ,l2 ,q2
z2q1−l1 (x2 y3 ...yn−1 )l2 y2k+m2 −q2 z3q2 −l2 (x3 y4 ...yn−1 + (y3 + z4 (x4 y5 ...yn−1 + ...)...)k−l1−l2 = ....
m
In this series coefficients in front of xm1 m2 n−1 k k,n
1 x2 ...xn−1 (y1...yn−1 ) are equal to Ωm1 ,...,mn ,
where xj = zj+1 for each j = 1, ..., n−1, xj and yj are variables. Therefore, the generating
function has the form:
m m +mn
(19) y1m1 ...yn−2
n−2 n−1
yn−1 (x1y2 ...yn + (x1 + y1 )(x2y3 ...yn−1 + (x2 + y2 )(x3y4 ...yn−1 +
!
Y k − l1 − ... − ls−1
(js − n) .
(js∈(1,...,n)\{i1,...,in−p};j1 <j2 <...;lj1 =0,...,ljp−1 =0)
ms
lk−1 −1 ! ! !
X lX
n−1 1 −1 X n l1 lk−1 lk
... ... x , consequently,
l1 =0 l2 =0 lk =0
l1 l2 lk
580 S.V. Ludkovsky
lk−2−1 ! ! !
X lX
n−1 1 −1 X
k n n l1 lk−2
(24) (∆ x )x=0 = ... ... =: Tn,k .
l1 =0 l2 =0 lk−1 =0
l1 l2 lk−1
On the other hand,
x P
(25) m = (m!)−1 m l=0 x
m−l (−1)l α (1, ..., m − 1) =
l
P l
l Sm,l x , where
P
(26) αl (z1, ..., zm) := i1 <i2 <...<il zi1 ...zil , consequently,
P
(27) l Sm,l Tl,j = δm,j and
P
(28) l Tm,l Sl,j = δm,j . Then for analytic g and f :
P
(29) g(x) = m am (g)xm and
X
(30) (g ◦ f )(x) = am (g)[m!/(l1!...lm!)]alk11 (f )...alkmm (f )]xn,
m,lj ,kj ,n
1 ∈ πsv (B(K, 1, |π|s) and πsv (x1x2 ) = πsv (x1 )πsv (x2 ) ∈ πsv (B(K, 1, |π|s)) for each
x1 , x2 ∈ B(K, 1, |π|s), since |x1x2 −1| ≤ max(|x1 −1||x2|, |x2 −1|) ≤ |π|s and |x2 | = 1.
Thus, B(Fpu , 1, |π|s) for a natural number s ≥ 1 is the multiplicative commutative group
and B(Fpu , 1, |π|s1 )/B(Fpu , 1, |π|s2) for each natural numbers 1 ≤ s1 < s2 is the finite
multiplicative commutative quotient group.
If g ∈ W and k0 ≥ 1, then |g(y)−y| ≤ kg −idkC 1 |y| ≤ |π|k0 |y| and |g(y1)−g(y2 )| ≤
kgkC 1 |y1 − y2 | and |g −1 (y1) − g −1 (y2)| ≤ kg −1kC 1 |y1 − y2 |, since t ≥ 1, consequently,
g is the isometry, that is, |g(y)| = |y| for each y ∈ M .
For given s and s1 choose k0 and q1 ≥ s1 such that for each g ∈ W we have
q1 g = id on Mq1 . Then for a given g ∈ W there exists the algebraic homomorphism
η1 from πs1 (B(K, 1, |π|s)) into πq1 (W ) satisfying Conditions (1 − 3) with πs1 (x0) in-
stead of x0 . Each g ∈ W and each x ∈ K is the projective limit of the inverse sequences
g = pr − lim{ qv g, πqqlv , Λq1 } and x = pr − lim{ sv x, πsslv , N}, since sequences {qv } and
{sv } are cofinal with N, where v ≥ l ∈ N (see also §2.5 [11]). For example, we can
q
take sv+1 = sv + 1 for each v ∈ N. Consider qv+1 g, then πqvv+1 ( qv+1 g) = qv g. Each
qv g ∈ Hom(Mqv ) is the element of the symmetric group Sbv , where bv is the cardinality of
the finite set πqv (M ) = Mqv (see also Theorem 3.6 above). Thus, qv g has a decomposition
Groups of Diffeomorphisms and Wraps of Manifolds... 583
into a product of nonintersecting finite cycles. If yv ∈ Mqv , then the cardinality of each
q
(πqvv+1 )−1 (yv ) is the same for each yv ∈ Mqv . The homomorphism qv+1 g on each subset
q
(πqvv+1 )−1 (yv ) acts as the isometry. Then each cycle of qv g splits into the product of cycles
or becomes a cycle of greater length for qv+1 g. Moreover, qv+1 g is the product of two ho-
q q
momorphisms h and f , where f ((πqvv+1 )−1 (yv )) = (πqvv+1 )−1 (yv ) for each yv ∈ Mqv and
qv+1 ∗
πqv (f ) = πqv (id) is the identity on πqv (M ), while h(zv+1 ) = h(yv+1,0 )+(zv+1 −yv+1,0 )
q q
with a marked yv+1,0 ∈ (πqvv+1 )−1 (yv ) and for each zv+1 ∈ (πqvv+1 )−1 (yv ).
If σ ∈ Σl and σ is a cycle of length u, then its algebraic order is u, that is, the cyclic
group {σ a : a ∈ N} is of order u. On the step v there is the commutative local subgroup
xv = η (x ) for each x ∈ π (B(K, 1, |π|s)). Therefore, on the v + 1-th step choose
qv g v v v sv
qv+1 sufficiently large such that a number of nonintersecting cycles σv,j or their length
are sufficiently large in the sense that a commutative subgroup of πqv+1 (W ) generated by
bj
finite products of σv,j , bj ∈ N, is of sufficiently large order that to provide the algebraic
homomorphism ηv+1 from πsv+1 (B(K, 1, |π|s)) into πqv+1 (W ) satisfying Conditions (1 −
s
3) with πqv+1 (x0) instead of x0 and πsvv+1 (ηv+1) = ηv .
In accordance with Lemma 2.5.10 [11] if {φ, fj 0 } is a mapping of an inverse system S =
0
{Xj , πij , Ψ} into an inverse system S 0 = {Yj 0 , πij0 , Ψ0} and all fj 0 are homeomorphisms,
then the limit mapping f = pr−lim{φ, fj 0 } is also the homeomorphism of X = pr−lim S
onto Y = pr−lim S 0, where φ : Ψ0 → Ψ is a nondecreasing function, Ψ0 and Ψ are directed
sets.
Using the same notation which can not cause a confusion we mention the fol-
lowing. Each this induction step gives the corresponding solution of (4 − 6)
with ηv (πsv (x)) ∈ πqv (B(K, 1, |π|s)) and πqv (y) and πqv (cj,k,m̄), πq∗v (Qm̄(y)) and
πsv (χj,k (x)) ∈ πqv (B(K, 1, |π|s)) instead of x, y, cj,k,m̄, Qm̄ (y) and χj,k (x) respectively.
If f ∈ Cbα (M, Kn ) and kf − idkCbα (M,Kn) < 1, then f ∈ Dif f α (M ). We say that a subset
A in M is an -net, if for each y ∈ M there exists z ∈ A such that |z − y| < , where
0 < < ∞. We have limv→∞ |π|v = 0 and representatives of Mqv in M form a ρv net
for suitable subsequence qv , which is denoted by the same notation, where 0 < ρ < 1 is a
constant and ρ is a parameter characterizing a net Aqv corresponding to {Qm̄ : m̄ ∈ N0 n }
for mj ≤ qv for each j = 1, ..., n, where each z ∈ Aqv is a zero of Qm̄ as soon as
mj ≥ mj,0 for some j = 1, ..., n, where m̄0 := (m1,0, ..., mn,0) corresponds to z. On
the other hand, |χj,k,m (x)| ≤ 1 on B(K, 1, |π|s) for each j, k, m, moreover, x0 = 1 + θs ,
|x0| = 1. Then the corresponding to it series converges, since qv g ηv (xv ) ∈ πq∗v (W ) for
each xv ∈ πsv (B(K, 1, |π|s)) and W is complete relative to the Cbα uniformity. Thus,
the inverse sequence of qv g ηv(πsv (x)) converges to a local multiplicative continuous one-
parameter subgroup g x(y) ∈ W relative to the Cbα × Cb0 uniformity which is Cbα by y ∈ M
and Cb0 by x ∈ B(K, 1, |π|s).
2. Theorem. If M is a manifold on the Banach space X := c0(γX , K) with a finite at-
las and charts with bounded φj (Uj ) in X for each j ∈ ΛM , where char(K) = p > 1, then
in Dif fbα (M ) in each neighborhood of id there are g 6= id which does not belong to any
non-trivial one-parameter subgroup g y relative to the additive group (K, +). Moreover, if
M is embedded into X and
(i) (h − h ◦ g −1 ), (h − h ◦ g −1 ) ◦ g 2,...,(h− h ◦ g −1 ) ◦ g p−1 are K-linearly independent
and g 6= id, where h := g − id, then g does not belong to any one-parameter subgroup
584 S.V. Ludkovsky
{g y : y ∈ (K, +)}.
Proof. From the conditions of this theorem it follows, that M has a C t embedding as
an open bounded subset in X. In view of Theorem 2.9 the group Dif fbα (M ) is metrizable.
Since Dif fbβ (M ) is everywhere dense in Dif fbα (M ) for each β > α, then it is sufficient
to consider the case t ≥ 1, α ∈ {t, [t]}. If g ∈ Dif fbα(M ), then up to a diffeomorphism of
manifolds g|U ∈ Dif fbα (U ) for any U open in M , since g : U → g(U ) ⊂ M . Let U ⊂ M
be a clopen bounded subset such that U ⊂ Uj for some j ∈ ΛM , g ∈ Dif fbα (M ) and
supp(g) ⊂ U , then g|U ∈ Dif fbα(U ). Take in particular U := φ−1 j (B(X, 0, |π|)), where
π ∈ K, 0 < |π| < 1. Thus, if prove theorem for U = B(X, 0, |π|), then it will be also true
for Dif fbα (M ).
Consider in Dif fbα(U ) a left-invariant metric ρα (see Theorem 2.9). Take id 6= g ∈
W := {f ∈ Dif fbα(U ) : ρα (f, e) ≤ |π|}, then g = id + h, where 0 < khkCbα (U,X) ≤ |π|.
If we consider pn as the element of K, where n ∈ N := {1, 2, ...}, then pn = 0 ∈
n
K, since char(K) = p. So we need to have g p = g 0 = id on U , where g 1 = g,
n n P
g k+1 = g k g 1,...,gp = g p −1 g 1. Consider (1 + a)k = ls=0 ks as , where a ∈ K, then
n n n n
(1 + a)p = 1 + ap and inevitably |(1 + a)p | = |1 + ap |, since each binomial coefficient
pn
n − 1 and hence pn is equal to zero in K. Then
l is divisible on p for each 1 ≤ l ≤ p l
Pl l l+(l−1)s s
(xi + axli ) ◦ (xi + axli ) = xi + axli + a s=0 s xi a and so on. In particular,
p2 p+1
(xi + axpi ) ◦ (xi +axpi )= xi + 2axpi + xi a for l = p and so
on. This shows, that
elements of the form g(x) = id(x) + axli ei with a 6= 0 and 1 < l ∈ N can not lie on any
one-parameter subgroups, since g p 6= id.
Demonstrate in general for M embedded into X, that each id 6= g ∈ W satisfying
Condition (i) does not belong to any one-parameter subgroup. On the other hand, g 2 (x) =
g◦g(x) = id◦(id+h)(x)+h◦(id+h)(x) = g(x)+h◦g(x), g n (x) = g n−1(x)+h◦g n−1(x),
consequently, ρα(g n, g n−1) = ρα (g 2, g) ≤ |π| for each n ∈ N and kh ◦ g n−1 kCbα ≤
|π|. Then by induction g n (x) = g(x) + h ◦ g(x) + ... + h ◦ g n−1 (x) for each n ≥ 2,
hence ρα(g n , g) ≤ |π|. Then |h(y) − h(x)| ≤ khkCbα (U,X)|x − y| for each x, y ∈ U , in
particular, for y = g(x), where y − x = g(x) − x = h(x), hence |h ◦ g(x) − h(x)| ≤
khkCbα (U,X)|h(x)| ≤ |π|2. Therefore, |g n(x) − g(x) − (n − 1)h(x)| ≤ khkCbα (U,X)|h(x)|
for each x ∈ U , where |h(x)| ≤ khkCbα (U,X)|x| ≤ |π||x|, consequently, |g n(x) − g(x) −
k
(n − 1)h(x)| ≤ (khkCbα (U,X))2 |x| ≤ |π|2|x|. Thus |g p (x) − id(x)| ≤ (khkCbα (U,X))2|x|,
k
since |pk h(x)| = 0. Suppose that g p = id for each k ∈ N, then 0 = g p(x) − id(x) =
g p(x) − g(x) − (p − 1)h(x) = h(x) + h ◦ g(x) + ... + h ◦ g p−1(x) = (h ◦ g(x) − h(x)) +
... + (h ◦ g p−1(x) − h(x)) = h(x) + p(h ◦ g 1(x) − h(x)) + (p − 1)(h ◦ g 2(x) − h ◦
g(x)) + ... + 3(h ◦ g p−2(x) − h ◦ g p−3(x)) + 2(h ◦ g p−1(x) − h ◦ g p−2(x)) − h ◦ g p−1(x),
since ph(x) = 0 and p(h ◦ g − h)(x) = 0 identically and g p−1 (x) = g −1 (x). Therefore,
it would be 0 = |h(x) − h ◦ g −1(x)| on U , since Condition (i) is supposed to be satisfied,
that leads to the contradiction, since h 6= 0, g 6= id, h ◦ g 6= h. Thus g p(x) 6= id(x)
for each id 6= g ∈ W satisfying Condition (i), consequently, g does not belong to any
one-parameter subgroup {g y : y ∈ (K, +)}.
2.1. Remark. On the other hand, if M ⊃ B(Kn , x0, r), then g y (x) := x + yz for
x ∈ B(Kn , x0, r) and g y (x) := x for x ∈ M \ B(Kn , x0, r) is the non-trivial one-
parameter subgroup for a marked z ∈ B(Kn , 0, r), where y ∈ B(K, 0, 1), 0 < r, x0 ∈ M .
Therefore, for each neighborhood W of id there exists 0 6= z ∈ B(Kn , 0, r) such that
Groups of Diffeomorphisms and Wraps of Manifolds... 585
id 6= g ∈ W .
M is totally disconnected and has a base of topology consisting of clopen compact subsets
in M . As U take a system of all clopen proper subsets U in M for which there exists g ∈ G
such that θ(gU ) is a ball of finite positive radius in Kn . Since M is totally disconnected
and modelled on Kn as the manifold, then U forms the base of the topology of M .
In particular, for the diffeomorphism group G = Dif fcα(M ) and the family U of
M = X we have gU ∈ U for each g ∈ G, since g −1(gU ) = U and g : M → M is
the continuous bijective epimorphism together with g −1. If U = M , then g(M ) = M . A
disjoint clopen covering C of θ(M ) by balls form a C ∞ atlas (moreover, it is the analytic
atlas) such that if f |U is of C α class for each U ∈ C, then f is C α on θ(M ). If U ∈ U
and U 6= M , then M \ U is a clopen nonvoid subset in M . Therefore, if U1 and U2
are two clopen nonvoid subsets in U different from M , then there exists g ∈ G such that
g(U1) = U2 , since each two clopen balls in Kn are analytically diffeomorphic. Thus
axioms (E1, E2) are satisfied.
If g ∈ G, then supp(g) := cl{x ∈ M : g(x) 6= x} is compact and for an open
covering B of M there are V1, ..., Vm in B such that supp(g) ⊂ V1 ∪ ... ∪ Vm . Suppose that
U ∈ U and cl(U ) 6= M , where cl(U ) denotes the closure of U in M . Since U is clopen
in M , then cl(U ) = U . The topological Tychonoff space M is totally disconnected and
locally compact paracompact, since it is modelled on Kn . In view of Theorem 6.2.9 [11] it
is strongly zero dimensional. Therefore, V1, ..., Vm has a refinement P1 , ..., Pm consisting
of clopen compact subsets in M such that supp(g) ⊂ P1 ∪ ... ∪ Pm . Take P0 := ∅ and
Wi := Pi \j<i Pj , then W1 , ..., Wm is the disjoint clopen covering of supp(g), where
Wi ⊂ Pi ⊂ Vi for each i = 1, ..., m. Therefore, g(W1),...,g(Wm ) is the clopen disjoint
covering of g(supp(g)). Each θ(Wi ) is a finite union of clopen balls Bi,j in Kn . Then put
gm |Wm = g|Wm and gm |M \Wm = id, gm−1 |Wm−1 = gm −1 g|
Wm−1 and gm−1 |M \Wm−1 = id,
−1 −1
gi |Wi = gi+1 ...gm g|Wi and gi|M \Wi = id for each i = 1, ..., m − 2, hence Conditions
(i, ii) of (E3) are satisfied. For each 1 ≤ k < j < m there is the identity gj ...gk |Wk =
−1 −1g| −1 −1
gj ...gk+1gk+1 ...gm Wk = gj+1 ...gm g|Wk and gm ...gk |Wk = g|Wk for each 1 ≤ k < m.
Then gi−1 ...g1(U ) ⊂ (U \ (W1 ∪ ... ∪ Wi−1 )) ∪ (gi−1...g1(U ∩ W1) ∪ gi−1 ...g2(W2 ∩
U )... ∪ gi−1 (U ∩ Wi−1 )) and Wi ∪ gi−1 ...g1(U ) 6= M for each U ∈ U , since U 6= M and
g(U ) 6= M , consequently, Condition (iii) of (E3) is also satisfied.
For each clopen compact V in M the group GV := {g ∈ G : supp(g) ⊂ V } is topo-
S
logically perfect by Theorem 3. Moreover, G = V GV , where the union is by all clopen
compact subsets V in M . The group G is supplied with the C α compact-open topology.
Therefore, G is topologically perfect. In view of the Epstein Theorem 2.2.1 and Corol-
lary 2.2.2 [5] the commutator group [G, G] is topologically simple, since each [GV , GV ]
is topologically simple. On the other hand, [G, G] is dense in G, hence G is topologically
simple. If M is locally compact and noncompact, then Dif fcα (M ) is everywhere dense in
Dif f α (M ), hence the latter group is topologically simple relative to its C α compact-open
topology.
5. Remark. If M is a compact manifold, t ≥ 0, α ∈ {t, [t]}, then Dif f α (M ) is not
algebraically simple being the projective limit of groups due to Theorem 3.6.
6. Theorem. Let M be a C α compact manifold, t ≥ 0, α ∈ {t, [t]}. Then
each continuous automorphism of Dif f α (M ) belongs to the group of homeomorphisms
Hom(M, M ) =: Hom(M ).
Proof. Let ψ be a continuous automorphism of Dif f α (M ). In view of Theorem
Groups of Diffeomorphisms and Wraps of Manifolds... 587
3.6 above or Theorem 2.5 [24] for each v ∈ N, v ≥ s, there exists the quotient map-
ping πv∗ : Dif f α (M ) → Σbv , where Σm is the symmetric group of the set {1, ..., m},
bv is the cardinality of the finite set Mv = πv (M ), πv∗ is induced by the quotient map-
ping πv : K → K/B(K, 0, |π|v ) with the help of polynomial expansions by Formulas
3.2(4, 5). Consider M embedded into Kn . The set of diffeomorphisms f such that f − id
is a piecewise affine on balls of the covering of M is contained in Dif f α (M ), conse-
quently, πv∗ (Dif f α (M )) = Sbv is the epimorphism. This also follows from Formulas
3.2(4−7). The automorphism ψ induces the automorphism ψv of πv∗ (Dif f α (M )) such that
πv∗ (ψ(g)) =: ψv (πv∗ (g)), since then πv∗ (ψ(gh)) = πv∗ (ψ(g)ψ(h)) = πv∗ (ψ(g))πv∗ (ψ(h)) =
ψv (πv∗ (g))ψv (πv∗ (h)) for each g, h ∈ Dif f α (M ). Consider v sufficiently large such that
bv > 6. In view of the Hölder Theorem 5.3.1 [19] the automorphism ψv is internal:
ψv (a) = hv ah−1 ∗ α ∗ α
v for each a ∈ πv (Dif f (M )), where hv ∈ πv (Dif f (M )) is a marked
v v
element. This defines the inverse sequence {hv , πl , Λs } such that πl (hv ) = hl for each
v ≥ l ∈ Λs , where πlv ◦ πv∗ = πl∗ and πlv : Gv → Gl are algebraic epimorphisms of discrete
groups Gv := πv∗ (Dif f α (M )) for each v ≥ l ∈ Λs , πlv are written without star for sim-
plicity of notation. Each hv : Mv → Mv is the homeomorphism and each ψv : Gv → Gv
is the homeomorphism. A limit of an inverse mapping system of homeomorphic mappings
is a homeomorphism by Proposition 2.5.10 [11]. Therefore, h = pr − lim{hv , πlv , Λs } is
the element of Dif fw (M ) (see Theorems 3.6, 3.7 above and Theorem 2.5 [24]), moreover,
ψ(g) = lim{hv πv∗ (g)h−1 v
v , πl , Λs } = hgh
−1 for each g ∈ Dif f α (M ). On the other hand,
pings for each l ≥ k, Ω(Mk , Nk ) = {fk : fk ∈ NkMk , fk (s0,k ) = y0,k }/Kα,k , Kα,k is
an equivalence relation induced by an equivalence relation Kα . Moreover, Ωw (M, N ) is a
compactification of Ωα (M, N ) relative to the projective weak topology τw .
Proof. In view of Corollary 3.3 πk (C0α (M, N )) is isomorphic with {fk : fk ∈
Mk
Nk , fk (s0,k ) = y0,k }, where the quotient mapping is denoted by πk both for M and
N , since it is induced by the same ring homomorphism πk : K → K/B(K, 0, 1), s0,k :=
πk (s0 ) and y0,k := πk (y0 ), where k ≥ s = max(s(M ), s(N )). Then πk∗ (Dif f0α (M )) is
isomorphic with Hom0 (Mk ) := {ψk : ψk ∈ Hom(Mk ), ψk (s0,k ) = s0,k }. All of this is
also applicable with the corresponding changes to classes of smoothness C α (or C(α) in the
notation of [23]), where α = t or α = [t] with substitution of Φ̄k on Υk in the latter case. If
f and g are two Kα -equivalent elements in C0α (M, N ), that is, there are sequences fn and
gn in C0α (M, N ) converging to f and g respectively and also a sequence ψn ∈ Dif f0α (M )
such that fn (x) = gn (ψn (x)) for each x ∈ M , then πk∗ (fn ) =: fn,k and gn,k := πk∗ (gn )
converge to πk∗ (f ) and πk∗ (g) respectively and also ψn,k := πk∗ (ψn ) ∈ Hom0 (Mk ). From
the equality fn,k (x(k)) = gn,k (ψn,k (x(k))) for each n ∈ N and x(k) ∈ Mk it follows,
that the equivalence relation Kα induces the corresponding equivalence relation Kα,k in
πk∗ (C0α (M, N )) such that classes < πk∗ (f ) >K,α,k of Kα,k -equivalent elements are closed.
Each element fk ∈ πk∗ (C0α (M, N )) is characterized by the equality fk (s0,k ) = y0,k . This
induces the quotient mapping πk∗ : Ωα (M, N ) → Ω(Mk , Nk ) and surjective mappings
πkl : Ω(Ml , Nl ) → Ω(Mk , Nk ) for each l ≥ k. Each Ω(Mk , Nk ) is the finite discrete set,
since each NkMk is the finite discrete set. This produces the inverse sequence of finite dis-
crete spaces, hence the limit Ωw (M, N ) := pr − lim{Ω(Mk , Nk ), πlk , Λs } of the inverse
sequence is compact and totally disconnected. It remains to verify that Ωw (M, N ) is the
commutative topological monoid with the unit element and the cancelation property.
From the equality M = M̄ \ {s0 }, it follows that Mk = M̄k , since for each k ∈ N
there exists x ∈ M such that x + B(Km , 0, p−k ) ∋ s0 . Moreover, Mk and Nk are finite
discrete spaces. Then πk (M ∨ M ) = Mk ∨ Mk , where A ∨ B := A × {b0 } ∪ {a0 } × B ⊂
A × B is the wedge product of pointed spaces (A, a0 ) and (B, b0 ), A and B are sets with
marked points a0 ∈ A and b0 ∈ B. The composition operation is defined on threads
{< fk >K,α,k : k ∈ N} of the inverse sequence in the following way. There was fixed
a C [∞] -diffeomorphism χ : M ∨ M → M [23]. Let x ∈ M , then πk (x) ∈ Mk and
χ−1 (U ) ∈ M ∨ M , where U := πk−1 (x + B(K, 0, p−k )) ∩ M. On the other hand χ−1 (U ) is
a disjoint union of balls of radius p−2k in B(K2m , 0, 1), hence there is defined a surjective
mapping χk : M2k ∨M2k → Mk induced by χ, πk and π2k such that χk (χ−1 (U )) = πk (x).
If f and g ∈ C α (M, N ), then f ∨ g ∈ C α ((M ∨ M ), N ) and χ(f ∨ g) ∈ C α (M, N ) as in
§2.6 [23]. Hence χk (f2k ∨ g2k ) ∈ C α (Mk , Nk ) and inevitably χk (< f2k ∨ g2k >K,α,2k ) =
χk (< f2k >K,α,2k ∨ < g2k >K,α,2k ) ∈ Ω(Mk , Nk ).
There exists a one to one correspondence between elements f ∈ Cw (M̄ , N ) and {fk :
k} ∈ {NkMk : k ∈ Λs }. Therefore, pr − limk Ω(Mk , Nk ) algebraically this is the commu-
tative monoid with the cancelation property. Let U be a neighborhood of e in Ωw (M, N ),
then there exists Uk = πk−1 (Vk ) such that Vk is open in Ω(Mk , Nk ), e ∈ Uk and Uk ⊂ U .
−1
On the other hand there exists U2k = π2k (V2k ) such that V2k is open in Ω(M2k , N2k ),
e ∈ U2k and U2k + U2k ⊂ Uk . Therefore, (f + U2k ) + (g + U2k ) ⊂ f + g + Uk ⊂ f + g + U
for each f, g ∈ Ωw (M, N ), consequently, the composition in Ωw (M, N ) is continuous.
Since C0α (M, N ) is dense in C0,w (M̄ , N ), then Ωα (M, N ) is dense in Ωw (M, N ) relative
Groups of Diffeomorphisms and Wraps of Manifolds... 589
wrap groups SL := {L(M̂k , Nk ); π̂kl ; Λs }. Its projective limit Li,w (M, N ) := pr − lim SL
is a commutative topological totally disconnected group and Lα (M, N ) has an embedding
in it as a dense subgroup.
Proof. Due to the Grothendieck construction the inversion operation fk 7→ fk−1 is
continuous in L(M̂k , Nk ) and homomorphisms π̂kl and π̂k have continuous extensions
from wrap submonoids onto wrap groups L(M̂k , Nk ). Each monoid Ω(M̂k , Nk ) is to-
tally disconnected, since NkM̂k is totally disconnected and Ω(M̂k , Nk ) is supplied with
the quotient ultrametric, hence the free Abelian group Fk generated by Ω(M̂k , Nk ) is
also totally disconnected and ultramertizable, consequently, L(M̂k , Nk ) is ultrametriz-
able. Evidently their inverse limit is also ultrametrizable and the equivalent ultrametric
can be chosen with values in Γ̃K := {|z| : z ∈ K}, where Γ̃K ∩ (0, ∞) is discrete in
(0, ∞) := {x : 0 < x < ∞, x ∈ R}. Then the projective limit (that is, weak) topology of
Li,w (M, N ) is induced by the projective weak topology of Cw (M, K).
5. Theorem. For each prime number p the wrap group Lα (M, N ) in its weak topol-
ogy inherited from Li,w (M, N ) has the non-archimedean compactification isomorphic with
Zp ℵ0 , moreover, Li,w (M, N ) has the compactification (νZ)ℵ0 , where νZ is the one-point
Alexandroff compactification of Z.
Proof. The projective ring homomorphism πk : K → Kk induces
π̂k∗ (Φ̄m (f (x; h1 , ..., hm ; ζ1 , ..., ζm )) = Φ̄m fk (x(k); h1 (k), ..., hm (k); ζ1 (k), ..., ζm (k))
and π̂k∗ (Υm (f (x[m] )) = Υ[m] fk (x(k)[m] ),
where m ∈ N, Φ̄m fk and Υm fk are defined for the field of fractions generated by Kk ,
since K is the commutative field (see also [4] and §§2.1-2.6 [23]). Then the condition
lim Υm f (x[m] ) = 0
|x|→∞
lim Υm fk (x(k)[m] ) = 0
|x(k)|→∞
respectively, where x[1] = (x, v [0] , ζ1 ), x[m+1] := (x[m] , v [m] , ζm+1 ). Therefore,
supp(fk ) := M̂kf := {x(k) : fk (x(k)) 6= 0} is a finite subset of the discrete space M̂k
for each k ∈ N. Then evidently, π̂k∗ (< g >K,α ) is a closed subset in NkM̂k for each
g ∈ C0α ((M̂ , ∞), (N, 0)), since for each limit point fk of π̂k∗ (< g >K,α ) its support is the
finite subset in M̂k . Let k0 be such that Nk0 6= {0}, then this is also true for each k ≥ k0 .
If fk ∈/ π̂k∗ (< w0 >K,α ) and k ≥ k0 , then fk∨n ∈/ π̂k∗ (< w0 >K,α ) for each n ∈ N, where
fk := fk ∨ ... ∨ fk denotes the n-times wedge product, since kf ∨n kC α ≥ kf kC α > 0
∨n
to π̂k∗ (< gi >K,α ) and the embedding of Ω(M̂k , Nk ) into L(M̂k , Nk ), zi ∈ {−1, 0, 1},
l = card(M̂kf ), M̂kgi are singletons for each i = 1, ..., l.
Each M̂k is the finite discrete set as well as Nk . For each x 6= y ∈ M̂k there exists ψ ∈
Hom0 (Mk ) such that ψ(x) = y, where 0 corresponds to s0 for convenience of the notation.
Using the group Hom0 (Nk ) we get that L(M̂k , Nk ) is isomorphic with Znk , where nk =
card(Nk ) > 1. For each prime number p > 1 there exists the p-adic completion of Z which
is Zp . In view of Corollary 4 Lα (M, N ) has the non-archimedean completion isomorphic
with Zℵp 0 , since Z is dense in Zp and pr − limk Znk = Zℵ0 .
On the other hand, we can take the multiplicative subgroup {θl : l ∈ Z} of the lo-
cally compact field Fpu (θ) which gives the embedding φ of Z into Fpu (θ), where θ0 = 1.
The completion of φ(Z) in Fpu (θ) is φ(Z) ∪ {0} which is the one-point Alexandroff com-
pactification νZ of Z. This gives the non-archimedean completion (νZ)ℵ0 of Lα (M, N ).
Moreover, Zℵp 0 and (νZ)ℵ0 are compact as products of compact spaces and Li,w (M, N )
has the aforementioned embeddings into them.
6. Note. Using quotient mappings ηp,s : Z → Z/ps Z we get that Lα (M, N )ℵ0 has
Q
the compactification equal to { p∈P Zp ℵ0 } × (νZ)ℵ0 relative to the product Tychonoff
topology, where P denotes the set of all prime numbers p > 1, s ∈ N. These compact-
ifications produce characters of Li,w (M, N ), since each compact Abelian group has only
one-dimensional irreducible unitary representations [16]. On the other hand, there are irre-
ducible continuous representations of compact groups in non-archimedean Banach spaces
[39]. Among them there are infinite-dimensional [10, 37]. Moreover, in their initial C α
topologies diffeomorphism and wrap groups also have infinite-dimensional irreducible uni-
tary representations [22, 23]. At the same time topologies of Lα (M, N ) and Lw (M, N )
or Li,w (M, N ) are incomparable, since the topologies of C α (M, N ) and Cw (M, N ) are
incomparable (see Theorem 3.7 above).
Projective limits of groups obtained above have the non-archimedean origin related with
non-archimedean families of semi-norms on spaces of continuous or more narrow classes of
functions between manifolds over ultra-normed fields. Generally, if a topological space X
has a projective limit decomposition X = pr − lim{Xα , πβα , Λ}, then if fβ : Xβ → Y is a
continuous function into a topological space Y , then f := fβ ◦ πβ : X → Y is a continuous
function, where the mapping πβα : Xα → Xβ is continuous for each α ≥ β ∈ Λ, Λ is a
directed set, πβα ◦ πα = πβ , πα : X → Xα is continuous and epimorphic. Therefore, fα =
π̃αβ (fβ ) := fβ ◦πβα for each α ≥ β generate the inductive limit ind−lim{C(Xβ , Y ); π̃αβ ; Λ},
where C(X, Y ) denotes the family of all continuous mappings from X into Y . On the other
hand, if Y = pr − lim{Yγ , pγδ , Ψ}, then one gets pr − lim{C(X, Yγ ); pγδ , Ψ}. Then these
two constructions can be combined with repeated application of projective and inductive
limits, which may be dependent on the order of taking limits. If card(Ψ) ≥ ℵ0 , then a
Q
suitable box topology in γ∈Ψ C(X, Yγ ) is strictly stronger than a weak topology in it and
in its projective limit subspace (see also [32]).
7. Theorem. If M̄ and N are compact manifolds, α ∈ {∞, [∞]}, then Lα (M, N ) is
the C α Lie group.
Proof. The uniform space C α ((M, s0 ), (N, y0 )) has the structure of the C α manifold,
since M and N are C α manifolds, where α ∈ {∞, [∞]}. Therefore, Ωα (M, N ) and
Lα (M, N ) are C α manifolds. The wedge product (f, g) 7→ f ∨ g in C α ((M, s0 ), (N, y0 ))
Groups of Diffeomorphisms and Wraps of Manifolds... 593
8. Appendix
1. Lemma. Let either f, g ∈ C [n] (U, Y ), where U is an open subset in X, Y is an algebra
over K, or f ∈ C [n] (U, K) and g ∈ C [n] (U, Y ), where Y is a topological vector space
over K, then
(1) (f g)[n] (x[n] ) = (Υ ⊗ P̂ + π̂ ⊗ Υ)n .(f ⊗ g)(x[n] )
and (f g)[n] ∈ C 0 (U [n] , Y ), where (π̂ k g)(x[k] ) := g ◦ π10 ◦ π21 ◦ ... ◦ πkk−1 (x[k] ),
P̂ n g := Pn Pn−1 ...P1 g, πkk−1 (x[k] ) := x[k−1] , (A ⊗ B).(f ⊗ g) := (Af )(Bg) for
A, B ∈ L(C n (U, Y ), C m (U, Y )), m ≤ n, (A1 ⊗ B1 )...(Ak ⊗ Bk ).(f ⊗ g) := (A1 ...Ak ⊗
B1 ...Bk ).(f ⊗ g) := (A1 ...Ak f )(B1 ...Bk g) for corresponding operators, Υn f := f [n] ,
(Pk g)(x[k] ) := g(x[k−1] + v [k−1] tk ),
P̂ k π̂ a1 Υb1 ...π̂ al Υbl g = Pk+s ...Ps+1 π̂ a1 Υb1 ...π̂ al Υbl g with s = b1 +...+bl −a1 −...−al ≥
0, a1 , ..., al , b1 , ..., bl ∈ {0, 1, 2, 3, ...}.
Proof. Let at first n = 1, then
(2) (f g)[1] (x[1] ) = [(f g)(x + vt) − (f g)(x)]/t = [(f (x + vt) − f (x))g(x + vt) +
f (x)(g(x + vt) − g(x))]/t = (Υ1 f )(x[1] )(P1 g)(x[1] ) + (π̂10 f )(x[1] )Υ1 g(x[1] ),
since π̂10 (x[1] ) = x and P1 is the composition of the projection π̂10 and the shift operator on
vt. Let now n = 2, then applying Formula (2) we get:
(3) (f g)[2] (x[2] ) = ((f g)[1] (x[1] ))[1] (x[2] ) = (Υ1 (f [1] (x[1] )(x[2] ))g(x + (v [0] +
[1] [1] [1] [1] [1] [1]
v2 t2 )(t1 + v3 t2 ) + v1 t2 ) + f [1] (x[1] )g [1] (x + v [0] t1 , v1 + v2 (t1 + v3 t2 ), t2 ) +
[1] [1]
f [1] (x, v1 , t2 )g [1] (x[1] + v1 t2 ) + f (x)g [2] (x[2] ),
[k] [k] [k] [0]
where v [k] = (v1 , v2 , v3 ) for each k ≥ 1 and v [0] = v1 such that x[k] + v [k] tk+1 =
[k] [k] [k]
(x[k] + v1 tk+1 , v [k−1] + v2 tk+1 , tk + v3 tk+1 ) for each 1 ≤ k ∈ Z. For n = 3 we get
(4) (f g)[3] (x[3] ) = [(Υ3 f )(P̂ 3 g) + (π̂ 1 Υ2 f )(Υ1 P̂ 2 g) + (Υ1 (π̂ 1 Υ1 f ))(P̂ 1 Υ1 P̂ 1 g)
+(π̂ 2 Υ1 f )(Υ2 P̂ 1 g) + (Υ2 π̂ 1 f )(P̂ 2 Υ1 g) + (π̂ 1 Υ1 π̂ 1 f )(Υ1 P̂ 1 Υ1 g)
+(Υ1 (π̂ 2 f ))(P̂ 1 Υ2 g) + (π̂ 3 f )(Υ3 g)](x[3] ),
since by our definition P̂ k π̂ a1 Υb1 ...π̂ al Υbl g = Pk+s ...Ps+1 π̂ a1 Υb1 ...π̂ al Υbl g with s =
b1 + ... + bl − a1 − ... − al ≥ 0, a1 , ..., al , b1 , ..., bl ∈ {0, 1, 2, 3, ...}.
Therefore, Formula (1) for n = 1 and n = 2 and n = 3 is demonstrated by Formulas
(2 − 4). If f, g ∈ C 0 (U [k] , Y ), a, b ∈ K, then (Pk (af + bg))(x[k] ) := (af + bg)(x[k−1] +
v [k−1] tk ) = af (x[k−1] + v [k−1] tk ) + bg(x[k−1] + v [k−1] tk ), moreover, π̂ k (af + bg)(x[k] ) =
(af + bg) ◦ π10 ◦ π21 ◦ ... ◦ πkk−1 (x[k] ) = (af + bg)(x) = af (x) + bg(x) = aπ̂ k f (x[k] ) +
bπ̂ k g(x[k] ) for each x[k] ∈ U [k] , hence π̂ k and Pk and P̂ k are K-linear operators for each
k ∈ N. Suppose that Formula (1) is proved for n = 1, ..., m, then for n = m + 1 it follows
by application of Formula (2) to both sides of Formula (1) for n = m:
(f g)m+1 (x[m+1] ) = ((f g)[m] (x[m] ))[1] (x[m+1] ) = ((Υ ⊗ P̂ + π̂ ⊗ Υ)m .(f ⊗
g)(x[m] ))[1] (x[m+1] ) = (Υ ⊗ P̂ + π̂ ⊗ Υ)m+1 .(f ⊗ g)(x[m+1] ),
594 S.V. Ludkovsky
since x[m+1] = (x[m] )[1] and more generally x[m+k] = (x[m] )[k] for each nonnegative inte-
gers m and k such that πkk−1 (x[m+k] ) = x[m+k−1] for k ≥ 1; Υk , P̂ k and π̂ are K-linear
operators on corresponding spaces of functions (see above and Lemma 2.3) and
(Υ ⊗ P̂ + π̂ ⊗ Υ)m+1 .(f ⊗ g)(x[m+1] ) =
P a1 a1
a1 +...+am+1 +b1 +...+bm+1 =m+1 (Υ ⊗ P̂ )
(π̂ b1 ⊗ Υb1 )...(Υam+1 ⊗ P̂ am+1 )(π̂ bm+1 ⊗ Υbm+1 ).(f ⊗ g)(x[m+1] ),
where aj and bj are nonnegative integers for each j = 1, ..., m + 1, (A1 ⊗ B1 )...(Ak ⊗
Bk ).(f ⊗ g) := (A1 ...Ak ⊗ B1 ...Bk ).(f ⊗ g) := (A1 ...Ak f )(B1 ...Bk g).
2. Note. Consider the projection
(1) ψn : X m(n) × Ks(n) → X l(n) × Kn ,
where m(n) = 2m(n − 1), s(n) = 2s(n − 1) + 1, l(n) = n + 1 for each n ∈ N such
that m(0) = 1, s(0) = 0, m(n) = 2n , s(n) = 1 + 2 + 22 + ... + 2n−1 = 2n − 1. Then
m(n), s(n), l(n) and n correspond to number of variables in X, K for Υn , in X and K
for Φ̄n respectively. Therefore, ψ(x[n] ) = x(n) and ψn (U [n] ) = U (n) for each n ∈ N for
suitable ordering of variables. Thus Φ̄n f (x(n) ) = ψ̂n Υn f (x[n] ) = f [n] (x[n] )|W (n) , where
ψ̂n g(y) := g(ψn (y)) for a function g on a subset V in X l(n) × Kn for each y ∈ ψn−1 (V ) ⊂
X m(n) × Ks(n) , W (n) = U (n) × 0, 0 ∈ X m(n)−l(n) × Ks(n)−n for the corresponding
ordering of variables.
3. Corollary. Let either f, g ∈ C n (U, Y ), where U is an open subset in X, Y is an
algebra over K, or f ∈ C n (U, K) and g ∈ C n (U, Y ), where Y is a topological vector
space over K, then
(1) Φ̄n (f g)(x(n) ) = (Φ̄ ⊗ P̂ + π̂ ⊗ Φ̄)n .(f ⊗ g)(x(n) )
and Φ̄n (f g) ∈ C 0 (U (n) , Y ). In more details:
P P
(2) Φ̄n (f g)(x(n) ) = 0≤a,0≤b,a+b=n j1 <...<ja ;s1 <...<sb ;{j1 ,...,ja }∪{s1 ,...,sb }={1,...,n}
Φ̄a f (x; vj1 , ..., vja ; tj1 , ..., tja )Φ̄b g(x + vj1 tj1 + ... + vja tja ; vs1 , ..., vsb ; ts1 , ..., tsb ).
Proof. The operator ψ̂n is K-linear, since ψ̂n (af + bg)(y) = (af + bg)(ψn (y)) =
af (ψn (y)) + bg(ψn (y)) for each a, b ∈ K and functions f, g on a subset V in X l(n) × Kn
and each y ∈ ψn−1 (V ) ⊂ X m(n) × Ks(n) . Mention that the restrictions of π̂kk−1 and Pk
on W (k) gives πkk−1 (x(k) ) := x(k−1) and (Pk g)(x(k) ) := g(x(k−1) + vk tk ) in the notation
of §1.1. The application of the operator ψ̂n to both sides of Equation 1(1) gives Equation
(1) of this corollary, since ψ̂n Υn = Φ̄n for each nonnegative integer n, where Υ0 = I and
Φ̄0 = I and ψ̂0 = I are the unit operators.
4. Lemma. Let f1 , ..., fk ∈ C [n] (U, Y ), where U is an open subset in X, either Y
is an algebra over K, or f1 , ..., fk−1 ∈ C [n] (U, K) and fk ∈ C [n] (U, Y ), where Y is a
topological vector space over K, then
P
(1) (f1 ...fk )[n] (x[n] ) = [ k−1 α=0 π̂
⊗α ⊗ Υ ⊗ P̂ ⊗(k−α−1) ]n .(f ⊗ ... ⊗ f )(x[n] )
1 k
[n] 0
and (f1 ...fk ) ∈ C (U , Y ), where[n]
π̂ ⊗α ⊗ Υ ⊗ P̂ ⊗(k−α−1) .(f1 ⊗ ... ⊗ fk ) := (π̂(f1 ...fα ))(Υfα+1 )(P̂ (fα+2 ...fk )), where
π̂ := I, P̂ 0 = I is the unit operator, π̂f0 := 1, P̂ fk+1 := 1 (see Lemma 1).
0
(f ◦ u)[1] (t0 , v, t) = [f (u(t0 + vt)) − f (u1 (t0 ), u2 (t0 + vt), ..., um (t0 +
vt))]/t + [f (u1 (t0 ), u2 (t0 + vt), u3 (t0 + vt), ..., um (t0 + vt)) − f (u1 (t0 ), u2 (t0 ), u3 (t0 +
vt), ..., um (t0 + vt))]/t+
... + [f (u1 (t0 ), ..., um−1 (t0 ), um (t0 + vt)) − f (u(t0 ))]/t,
where u = (u1 , ..., um ), uj ∈ K for each j = 1, ..., m. Since uj (t0 + vt) − uj (t0 ) =
[1]
tuj (t0 , v, t), hence
(f ◦ u)[1] (t0 , v, t) = Υ1 f ((u1 (t0 ), u2 (t0 + vt), ..., um (t0 + vt)), e1 , tΥ1 u1 (t0 , v, t))
Υ1 u 1 1
1 (t0 , v, t) + Υ f ((u1 (t0 ), u2 (t0 ), u3 (t0 + vt), ..., um (t0 + vt)), e2 , tΥ u2 (t0 , v, t))
Υ1 u2 (t0 , v, t) + ... + Υ1 f (u(t0 ), em , tΥ1 um (t0 , v, t))Υ1 um (t0 , v, t),
since uj ∈ K for each j = 1, ..., m and K is the field, where ej = (0, ..., 0, 1, 0, ..., 0) ∈
Km with 1 on j-th place for each j = 1, ..., m. With the help of shift operators it is possible
to write the latter formula shorter:
Pm
(2) Υ1 (f ◦ u)(y, v, t) = 1 1
j=1 Ŝj+1,vt Υ f (u(y), ej , tΥ ◦ pj u(y, v, t))(Υ ◦
1
pj u(y, v, t)),
where pj (x) := xj , x = (x1 , ..., xm ), xj ∈ K for each j = 1, ..., m, Ŝj+1,τ g(u(y), β) :=
g(Sj+1,τ u(y), β), y ∈ Ks , τ ∈ Kk , k ≥ s, β is some parameter. Introduce op-
erators Aj,v,t := (Ŝj+1,vt ⊗ tΥ1 ◦ pj )∗ Υ1j , where Υ1 is taken for variables (y, v, t)
or corresponding to them after actions of preceding operators as Υk remembering that
[k] [k] [k] [k] [k] [k]
y [k] , v [k] ∈ (Ks )[k] , t ∈ K, v [k] = (v1 , v2 , v3 ) with v1 , v2 ∈ (Ks )[k−1] , v3 ∈ Kk for
[0]
each k ≥ 1, in particular, v [0] = v1 for k = 0, Υ1j f (x, v, t) := [f (x + ej vj t) − f (x)]/t,
(B ⊗ A)∗ Υ1 fi ◦ ui (y, v, t) := Υ1j fi (Bui , v, Aui ), B : Km(i) → Km(i) , A : Km(i) → K.
For example, in the particular case of s = 1 we have v [k] ∈ (K)[k] . Therefore, in the general
case Formula (2) takes the form:
P
(3) Υ1 f ◦ u(y, v, t) = m 1
j=1 (Aj,v,t f ◦ u)(Υ ◦ pj u)(y, v, t).
Take now n = 2, then
P
Υ2 f ◦ u(y [2] ) = Υ1 m 1
j=1 [(Aj,v,t f ◦ u)(Υ ◦ pj u)(y, v, t)](y ).
[2]
In the square brackets there is the product, hence from Formula 1(1) and Lemma 2.3 we
get:
P
(4) Υ2 f ◦ u(y [2] ) = m 1 1 1
j=1 [(Υ Aj,v [0] ,t f ◦ u)(P2 Υ ◦ pj u) + (π̂ Aj,v [0] ,t f ◦ u)(Υ ◦
2
pj u)](y [2] ).
Then from Formula (3) applied to terms Aj,v,t f ◦u it follows, that Υ1 Aj1 ,v[0] ,t1 f ◦u(y [2] ) =
Pm(2)
j2 =1 (Aj2 ,v [1] ,t2 Aj1 ,v [0] ,t1 f ◦ u)(Υ1 ◦ pj2 Sj1 +1,v[0] t1 u)(y [2] ), where v [0] = v, t1 = t (see
also Lemma 1). Therefore,
P Pm(2)
(5) Υ2 f ◦ u(y [2] ) = [ mj1 =1 j2 =1 (Aj2 ,v [1] ,t2 Aj1 ,v [0] ,t1 f ◦ u)(Υ
1 ◦
P m
pj2 Ŝj1 +1,v[0] t1 u)(P2 Υ1 ◦ pj1 u) + j1 =1 (π̂ 1 Aj1 ,v[0] ,t1 f ◦ u)(Υ2 ◦ pj1 u)](y [2] ).
Then for n = 3 applying Formulas (3) and 4(1) to (5) we get:
P Pm(2) Pm(3)
(6) Υ3 f ◦ u(y [3] ) = [ m j1 =1 j2 =1 j3 =1 (Aj3 ,v [2] ,t3 Aj2 ,v [1] ,t2 Aj1 ,v [0] ,t1 f ◦ u)
(Υ ◦ pj3 Ŝj2 +1,v[1] t2 Ŝj1 +1,v[0] t1 u )(P2 Υ ◦ pj2 Ŝj1 +1,v[0] t1 u)(P3 P2 Υ1 ◦ pj1 u)+
1 2 1
Pm Pm(2) 1 2 1
j1 =1 j2 =1 [(π̂ (Aj2 ,v [1] ,t2 Aj1 ,v [0] ,t1 f ◦ u))(Υ ◦ pj2 Ŝj1 +1,v [0] t1 u)(P3 P2 Υ ◦ pj1 u)+
(π̂ 1 {(Aj2 ,v[1] ,t2 Aj1 ,v[0] ,t1 f ◦ u)(Υ1 ◦ pj2 Ŝj1 +1,v[0] t1 u)}(Υ1 P2 Υ1 ◦ pj1 u)]+
Pm Pm(3) 1 1 2
j1 =1 j3 =1 (Aj3 ,v [2] ,t3 π̂ Aj1 ,v [0] ,t1 f ◦ u)(Υ ◦ pj3 Ŝj1 +1,v [0] t1 u)(P3 Υ ◦ pj1 u)
P m
+ j1 =1 (π̂ 2 Aj1 ,v[0] ,t1 f ◦ u)(Υ3 ◦ pj1 u)](y [3] ).
Groups of Diffeomorphisms and Wraps of Manifolds... 597
Thus Formula (1) is proved for n = 1, 2, 3. Suppose that it is true for k = 1, ..., n and
prove it for k = n + 1. Applying Formula 4(1) to both sides of (1) we get:
Pm Pm(n+1)
(7) Υn+1 f ◦ u(y [n+1] ) = [ j1 =1 ... jn+1 =1 (Ajn+1 ,v [n] ,tn+1 ...Aj1 ,v [0] ,t1 f ◦ u)
(Υ ◦pjn+1 Ŝjn +1,v[n−1] tn ...Ŝj1 +1,v[0] t1 u )(Pn+1 Υ1 ◦pjn Ŝjn−1 +1,v[n−2] tn−1 ...Ŝj1 +1,v[0] t1 un−1 )...
1 n
P Pm(n) 1
(Pn+1 ...P2 Υ1 ◦ pj1 u) + m j1 =1 ... jn =1 (π̂ (Ajn ,v [n−1] ,tn ...Aj1 ,v [0] ,t1 f ◦ u)
Υ ((Υ ◦ pjn Ŝjn−1 +1,v[n−2] tn−1 ...Ŝj1 +1,v[0] t1 un−1 )...(Pn ...P2 Υ1 ◦ pj1 u))+
1 1
P Pm(n−1) 1
Υ1 ( mj1 =1 ... jn−1 =1 (π̂ (Ajn−1 ,v [n−2] ,tn−1 ...Aj1 ,v [0] ,t1 f ◦ u))Υ1 ((Υ1 ◦
pjn−1 Ŝjn−2 +1,v[n−3] tn−2
P Pm(2)
...Ŝj1 +1,v[0] t1 un−2 )...(Pn−1 ...P2 Υ1 ◦ pj1 u)) + ... + Υn−2 { m j1 =1 j2 =1
(π̂ 1 Aj2 ,v[1] ,t2 Aj1 ,v[0] ,t1 f ◦ u)Υ1 ((Υ1 ◦ pj2 Ŝj1 +1,v[0] t1 u)(P2 Υ1 ◦ pj1 u))}+
P
Υn−1 { m 1 2
j1 =1 π̂ Aj1 ,v [0] ,t1 f ◦ u)(Υ ◦ pj1 u)}](y
[n+1] ) =
Pm Pm(n+1) 1
[ j1 =1 ... jn+1 =1 (Ajn+1 ,v [n] ,tn+1 ...Aj1 ,v [0] ,t1 f ◦ u)(Υ ◦ pjn+1 Ŝjn +1,v [n−1] tn
...Ŝj1 +1,v[0] t1 un )(Pn+1 Υ1 ◦ pjn Ŝjn−1 +1,v[n−2] tn−1 ...Ŝj1 +1,v[0] t1 un−1 )...(Pn+1 ...P2 Υ1 ◦
pj1 u)
P Pm(n) 1 Pn−1 ⊗α
+ m j1 =1 ... jn =1 (π̂ (Ajn ,v [n−1] ,tn ...Aj1 ,v [0] ,t1 f ◦ u)[ α=0 π̂ ⊗ Υ ⊗ P̂ ⊗(n−α−1) ]
((Υ1 ◦ pjn Ŝjn−1 +1,v[n−2] tn−1 ...Ŝj1 +1,v[0] t1 un−1 ) ⊗ ... ⊗ (Pn ...P2 Υ1 ◦ pj1 u))
Pn−1 ⊗α ⊗ Υ ⊗ P̂ ⊗(n−α−1) ](Pm Pm(n−1) 1
+[ α=0 π̂ j1 =1 ... jn−1 =1 (π̂ (Ajn−1 ,v [n−2] ,tn−1 ...Aj1 ,v [0] ,t1 f ◦
Pn−2 ⊗α ⊗(n−α−2)
u)) ⊗ [ α=0 π̂ ⊗ Υ ⊗ P̂ ]
((Υ ◦ pjn−1 Ŝjn−2 +1,v[n−3] tn−2 ...Ŝj1 +1,v[0] t1 un−2 ) ⊗ ... ⊗ (Pn−1 ...P2 Υ1 ◦ pj1 u))
1
P P Pm(2) 1
+[ 2α=0 π̂ ⊗α ⊗ Υ ⊗ P̂ ⊗(2−α) ]n−2 { m j1 =1
1
j2 =1 (π̂ Aj2 ,v [1] ,t2 Aj1 ,v [0] ,t1 f ◦ u)(Υ ⊗ P̂ +
1
Mention that in general (Υn+1 f ◦ u)(y [n+1] ) may depend nontrivially on all components
of the vector y [n+1] through several terms in Formula (7). Thus Formula (1) of this Lemma
is proved by induction.
7. Corollary. Let u ∈ C n (Ks , Km ), u(Ks ) ⊂ U and f ∈ C n (U, Y ), where U is an
open subset in Km , s, m ∈ N, Y is a K-linear space, then
P Pm(n)
(1) Φ̄n (f ◦ u)(x(n) ) = [ m
j1 =1 ... jn =1 (Bjn ,v (n−1) ,tn ...Bj1 ,v (0) ,t1 f ◦ u)
(Φ̄1 ◦ pjn Ŝjn−1 +1,v(n−2) tn−1 ...Ŝj1 +1,v(0) t1 un−1 )(Pn Φ̄1 ◦
pjn−1 Ŝj +1,v(n−3) ,t ...Ŝj1 +1,v(0) t1 un−2 )
n−2 0 n−2
Pm Pm(n−1)
...(Pn ...P2 Φ̄1 ◦ pj1 u) + j1 =1 ... jn−1 =1 (π̂
1 (B
jn−1 ,v (n−2) ,tn−1 ...Bj1 ,v (0) ,t1 f ◦
Pn−2
u)[ α=0 π̂ ⊗α
⊗ Φ̄ ⊗ P̂ ⊗(n−α−2) ]
((Φ̄1 ◦ pjn−1 Ŝjn−2 +1,v(n−3) tn−2 ...Ŝj1 +1,v(0) t1 un−2 ) ⊗ ... ⊗ (Pn−1 ...P2 Φ̄1 ◦ pj1 u))
Pn−2 ⊗α ⊗ Φ̄ ⊗ P̂ ⊗(n−α−2) ](Pm Pm(n−2) 1
+[ α=0 π̂ j1 =1 ... jn−2 =1 (π̂ (Bjn−2 ,v (n−3) ,tn−2 ...Bj1 ,v (0) ,t1 f ◦
Pn−3 ⊗α ⊗(n−α−3)
u)) ⊗ [ α=0 π̂ ⊗ Φ̄ ⊗ P̂ ]((Φ̄ ◦ pjn−2 Ŝjn−3 +1,v(n−4) tn−3 ...Ŝj1 +1,v(0) t1 un−3 ) ⊗
1
References
[1] Y. Amice. ”Interpolation p-Adique”. Bull. Soc. Math. France 92(1964), 117-180.
[2] Aref’eva I.Ya., Dragovich B., Frampton P.H., Volovich I.V. ”Wave functions of the
universe and p-adic gravity”. Int. J. Modern Phys. 6 (1991), 4341-4358.
[3] J. Araujo, W.H. Schikhof. ”The Weierstrass-Stone approximation theorem for p-adic
C n -functions”. Ann. Math. Blaise Pascal. 1 (1994), 61-74.
[4] Yu.A. Bachturin. ”Basic structures of modern algebra” (Moscow: Nauka, 1990).
[5] A. Banyaga. ”The structure of classical diffeomorphism groups”. Mathem. and its
Applic. 400 (Dordrecht: Kluwer, 1997
[7] P.-J. Cahen, J.-L. Chabert. ”On the ultrametric Stone-Weierstrass theorem on
Mahler’s expansion”. J. de Théorie des Nombres de Bordeaux 14 (2002), 43-57.
[9] B. Diarra. ”Ultraproduits ultrametriques de corps values”. Ann. Sci. Univ. Clermont
II, Sér. Math. 22 (1984), 1-37.
[11] R. Engelking. ”General topology”. 2-nd ed., Sigma Series in Pure Mathematics, V. 6
(Berlin: Heldermann Verlag, 1989).
[12] J.M.G. Fell, R.S. Doran. ”Representations of ∗-algebras, locally compact groups,
and Banach ∗-algebraic bundles”. V. 1 and V. 2 (Boston: Acad. Press, 1988).
[15] I.N. Herstein. ”Topics in algebra” (New-York: Blaisdell Publ. Co., 1964).
[16] E. Hewitt, K.A. Ross. ”Abstract harmonic analysis” (Berlin: Springer, 1979).
[18] I. Kaplansky. ”The Weiertsrass theorem in fields with valuations”. Proc. Amer. Math.
Soc. 1 (1950), 356-357.
[19] M.I. Kargapolov, Yu.I. Merzljakov. ”Principles of group theory” (Moscow: Nauka,
1977).
[25] S.V. Ludkovsky, B. Diarra. ”Profinite and finite groups associated with loop and
diffeomorphism groups of non-archimedean manifolds”. Int. J. of Math. and Mathe-
matical Sciences 2003: 42 (2003), 2673-2688.
[30] S.V. Ludkovsky. ”Approximate and global differentiability of functions over non-
archimedean fields”. Sovrem. Mathem. i ee Pril. 52 (2008)
(previous variant: Los Alamos Nat. Lab. math.CA/0608724).
[32] L. Narici, E. Beckenstein. ”Topological vector spaces” (New York: Marcel Dekker
Inc., 1985).
[33] A. Ostrowski. ”Untersuchungen zur arithmetischen Theorie der Körper”. Math. Zeit.
39 (1935), 269-404.
[34] ”p-Adic Mathematical Physics.” 2nd International Conference (Belgrad, 2005). Ed-
itors: A.Yu. Khrennikov, Z. Rakić, I.V. Volovich. AIP Conference Proceedings 826,
New York, 2006.
[36] J. Riordan. ”Combinatorial identities” (New York: John Wiley and Sons Inc., 1968).
[38] A.C.M. van Rooij. ”Non-Archimedean functional analysis” (New York: Marcel
Dekker Inc., 1978).
[39] A.C.M. van Rooij, W.H. Schikhof. ”Groups representations in non-Archimedean Ba-
nach spaces”. Bull. Soc. Math. France. Memoire. 39-40 (1974), 329-340.
[40] W.H. Schikhof. ”Ultrametric calculus” (Cambridge: Cambr. Univ. Press, 1984).
[41] W.H. Schikhof. ”Non-Archimedean calculus”. Nijmegen: Math. Inst., Cath. Univ.,
Report 7812, 130 pages, 1978.
[42] V.S. Vladimirov, I.V. Volovich, E.I. Zelenov. ”p-adic analysis and mathematical
physics” (Moscow: Nauka, 1994).
application, x, xi, xiii, 44, 62, 86, 126, 131, 134, 140,
142, 175, 203, 242, 283, 285, 294, 305, 326, 337,
A 376, 402, 528, 564, 568, 575, 592, 593, 594
Argentina, 325
ABC, 248 argument, 35, 43, 46, 114, 211, 213, 227, 239, 275,
Abelian, 17, 18, 99, 291, 402, 403, 406, 412, 419, 566
432, 591, 592 arithmetic, 122, 449
absorption, 2 Arizona, 56
absorption spectra, 2 Arizona State University, 56
academic, 27 Asian, 559, 599
access, 2 assignment, 352, 533, 547
accounting, 91 assumptions, 79, 377, 392
accuracy, ix, 169, 180, 184 astrophysics, 36
ad hoc, 201 asymmetry, 37, 53
Adams, 448, 481 asymptotic, viii, 55, 89, 90, 91, 92, 99, 101, 103,
adaptation, 92 104, 106, 107, 108, 109, 110, 112, 114, 116, 268,
additives, 499 269, 271, 275, 281, 290, 360, 371, 383
affine group, 201, 210 asymptotically, 110, 114, 251, 360, 365
age, 49 asymptotics, 91
aid, 6, 9, 14, 62, 72, 76, 77, 212, 220, 222, 240, 243, Australia, 37, 186
261 Austria, 1
AIP, 600 authority, 28
Albert Einstein, 1
algorithm, 84, 124, 170, 172, 180
alternative, 54, 120, 209, 215, 216, 217, 284, 439, B
488, 489, 490, 493, 495, 502, 503, 510, 511, 513,
514, 515, 517, 519, 521, 527, 543, 545, 554, 558 Banach spaces, 564, 592, 599
alternatives, 27 baryon, vii, 3, 13, 23, 27, 28, 32, 40, 42, 43
ambivalence, 5 baryonic matter, 36
amplitude, 46 baryons, 28, 44, 55
AMS, 58, 187, 305, 348, 447 beams, 87
Amsterdam, 87, 559 behavior, 104, 116, 118, 121, 214, 216, 267, 269,
analog, 262, 362, 545, 548, 549, 564, 575 371, 383, 402, 419
angular momentum, vii, 3, 4, 5, 6, 7, 20, 21, 231, behaviours, viii, 89, 90, 92, 98
246, 256, 294, 295 Beijing, 139, 169
angular velocity, 62, 63, 74, 76, 77, 78, 82, 83, 294, Belarus, 483
303, 340 Bessel, 118, 376, 377
annihilation, vii, 3, 17, 32, 42, 43 Bianchi identity, 240, 241, 253
anomalous, 2, 4 birth, 390
antagonistic, 51 black hole, 247
anthropology, 46 blocks, 41
antimatter, 35, 247 Bogolubov, 481
appendix, 574 boson, 30, 34, 53
bosons, 35, 44, 52, 53, 201
602 Index
evolution, viii, 36, 47, 123, 128, 254, 258, 263, 325,
327, 408 G
excitation, 35
exclusion, 30 gas, 124, 138, 298
exercise, 394 gauge, x, 5, 27, 35, 53, 54, 199, 200, 201, 202, 203,
expansions, 587, 589, 599 204, 205, 208, 209, 212, 213, 214, 215, 216, 217,
eyes, 3 219, 221, 222, 223, 224, 225, 227, 232, 233, 235,
239, 241, 243, 246, 247, 251, 255, 256, 257, 264,
486
F gauge fields, x, 199, 200, 201, 202, 203, 219, 221,
232, 233, 256
failure, 170 gauge group, 5, 35, 200, 201, 202, 208, 209, 239
family, viii, 16, 33, 91, 104, 123, 125, 127, 128, 129, gauge theory, x, 53, 54, 199, 200, 201, 202, 203,
132, 133, 136, 137, 144, 145, 148, 189, 191, 195, 224, 227, 256, 486
196, 214, 216, 254, 280, 288, 326, 353, 369, 370, Gauss-Bonnet, 483
371, 393, 408, 414, 416, 420, 421, 425, 427, 429, Gaussian, 91, 103, 104, 108, 372, 373, 560
430, 449, 451, 453, 456, 459, 461, 466, 467, 470, GBA, 260
476, 480, 488, 489, 492, 500, 501, 502, 505, 507, gender, 140
511, 513, 515, 516, 519, 522, 523, 526, 530, 531, gene, 287, 397, 419, 485
538, 539, 540, 541, 542, 543, 545, 546, 548, 549, General Relativity, x, 55, 199, 200, 201, 202, 219,
551, 552, 564, 567, 568, 569, 570, 571, 577, 581, 242, 247, 249, 251, 253, 254, 255, 256, 257, 262
586, 592 generalization, 210, 227, 256, 280, 290, 296, 303,
Far East, 560 327, 339, 342, 483, 486, 488, 522, 535
Fermi, 28, 199, 385 generalizations, 287, 397, 419, 485
fermions, 10, 28, 30, 34, 36, 38, 42, 44 generation, 247, 250
Feynman, 28 generators, ix, xii, 13, 14, 22, 53, 99, 100, 125, 130,
fiber, xiii, 202, 204, 205, 206, 208, 209, 211, 212, 139, 140, 141, 142, 148, 150, 152, 157, 160, 161,
452, 453, 455, 485, 487, 489, 490, 491, 492, 493, 206, 210, 211, 230, 231, 232, 385, 386, 387, 389,
494, 505, 506, 507, 508, 509, 515, 516, 517, 525, 394, 407, 409, 412, 414, 437, 498, 502, 504, 510,
541, 554, 598 511, 512, 519, 525, 534, 535, 538, 539
fiber bundles, xiii, 204, 208, 452, 453, 485, 487, 490, genes, 46
491, 509, 516 goals, 308
fibers, xiii, 205, 208, 209, 212, 213, 214, 259, 456, grades, 5, 18, 29, 30, 32
457, 485, 486, 537 grading, 25, 32, 37, 47
field theory, 55 gravitation, x, 56, 58, 199, 200, 201, 202, 208, 209,
finite volume, 90, 121 230, 241, 256, 264
Finland, 87 gravitational field, x, 1, 200, 226, 242, 246, 256, 306
fixation, 2 gravitational force, 200
flavor, 26, 30, 42, 45, 47 gravity, 200, 201, 202, 203, 224, 227, 242, 246, 247,
flow, x, xi, 89, 90, 91, 106, 107, 110, 119, 121, 177, 252, 253, 256, 263, 264, 598
186, 274, 275, 279, 280, 281, 283, 284, 285, 286, groups, vii, ix, x, xi, xii, xiii, 2, 3, 5, 10, 11, 28, 35,
289, 290, 291, 293, 295, 296, 297, 298, 301, 302, 36, 40, 45, 48, 50, 51, 53, 54, 55, 61, 63, 70, 98,
303, 332, 333, 341, 344, 368, 369, 370, 371, 382, 121, 122, 124, 130, 138, 139, 140, 141, 148, 170,
388 172, 185, 187, 188, 200, 201, 202, 203, 208, 209,
fluid, x, xi, 283, 284, 285, 286, 290, 291, 293, 297, 210, 214, 260, 264, 283, 284, 285, 290, 291, 293,
299, 301, 302, 303, 305, 558 296, 302, 305, 306, 323, 325, 337, 347, 348, 349,
Fourier, 358, 365, 382, 383, 446 351, 352, 353, 358, 366, 367, 381, 382, 383, 398,
Fourier analysis, 382, 383 401, 404, 405, 408, 411, 448, 449, 450, 452, 461,
fractals, 32 468, 469, 472, 474, 476, 477, 478, 479, 480, 481,
framing, 209 482, 485, 486, 487, 490, 495, 496, 498, 499, 500,
France, 89, 383, 397, 598, 600 503, 504, 505, 506, 508, 509, 510, 512, 513, 514,
freedom, x, 4, 24, 25, 28, 33, 41, 47, 50, 199, 202, 515, 520, 521, 522, 523, 525, 527, 529, 532, 534,
256, 304, 342 535, 536, 537, 542, 543, 551, 552, 559, 560, 563,
friction, 285, 292, 294 564, 565, 576, 585, 586, 587, 591, 592, 598, 599,
frog, 175 600
fulfillment, 131
functional analysis, 10, 600
Index 605
iteration, 82, 83, 172 416, 418, 419, 420, 421, 422, 423, 426, 427, 428,
429, 430, 431, 432, 433, 434, 435, 437, 438, 439,
441, 446, 448, 457, 458, 462, 463, 464, 471, 530
J Lie group, vii, viii, ix, x, xi, xii, xiii, 2, 3, 5, 10, 11,
13, 18, 26, 36, 39, 40, 44, 50, 52, 53, 61, 63, 67,
Jacobian, 124, 178, 228 70, 71, 121, 124, 125, 137, 138, 139, 140, 141,
January, 57, 59 144, 145, 146, 147, 148, 149, 157, 163, 164, 165,
Japan, 122, 483, 560 169, 170, 171, 172, 175, 176, 184, 185, 186, 187,
joints, 188 188, 191, 198, 201, 205, 209, 210, 274, 283, 284,
Jordan, 559 285, 286, 287, 289, 290, 291, 292, 293, 294, 296,
justification, 52, 77 297, 299, 300, 301, 302, 303, 304, 305, 306, 326,
327, 328, 333, 334, 335, 337, 340, 343, 344, 345,
347, 348, 349, 351, 352, 353, 355, 357, 358, 359,
K 360, 361, 363, 364, 365, 366, 367, 369, 371, 372,
373, 376, 377, 381, 382, 383, 398, 408, 447, 448,
K+, 28 452, 454, 455, 456, 457, 459, 460, 461, 464, 465,
kernel, 100, 117, 189, 267, 268, 269, 276, 354, 378, 467, 468, 469, 470, 471, 472, 473, 474, 475, 476,
379, 389, 463, 472, 547 477, 478, 479, 482, 485, 486, 490, 495, 502, 530,
kinematics, 188, 198 557, 563, 564, 565, 568, 592, 593
Klein-Gordon, 162, 165, 249, 253 limitation, 10, 39
Kleinian, 121, 122 linear, vii, xi, xii, 3, 8, 10, 11, 15, 21, 43, 46, 52, 65,
Kolmogorov, 302 67, 69, 70, 71, 72, 73, 80, 84, 86, 100, 124, 127,
Korteweg-de Vries, 166, 167 128, 130, 131, 140, 142, 143, 148, 149, 151, 153,
170, 172, 173, 174, 185, 186, 191, 201, 202, 205,
209, 210, 215, 217, 219, 222, 251, 255, 256, 291,
L 301, 305, 325, 326, 327, 329, 331, 332, 333, 336,
340, 342, 343, 344, 346, 347, 352, 357, 358, 364,
Lagrangian, x, 27, 62, 80, 87, 104, 169, 174, 199,
366, 367, 385, 386, 387, 388, 389, 390, 391, 392,
201, 223, 224, 227, 229, 230, 232, 233, 235, 242,
394, 402, 403, 404, 405, 408, 413, 414, 415, 417,
243, 244, 248, 256, 283, 284, 285, 292, 302
420, 421, 422, 423, 425, 426, 427, 428, 430, 431,
Lagrangian density, 224, 227, 229, 230, 232, 233,
432, 457, 459, 460, 463, 471, 472, 474, 555, 566,
242, 244, 248
598
Lagrangian formulation, 62, 80
linear dependence, 301, 388
lambda, 17
linear function, 8, 10, 11, 86, 366
language, xi, 204, 283, 387, 397
linear systems, 305
lattice, 53, 54, 59, 326, 327, 338, 348
location, 11
lattices, 53
locus, 46, 193, 195, 198
law, viii, 89, 90, 91, 92, 96, 103, 104, 108, 109, 110,
London, 2, 52, 55, 59, 87, 121, 267, 306, 323, 347,
111, 112, 113, 114, 115, 116, 117, 118, 119, 120,
348
142, 203, 208, 212, 215, 216, 221, 237, 238, 257,
luminal, 250
258
Lyapunov, 369, 370
laws, 89, 119, 120, 137, 138, 170, 203, 204, 227,
Lyapunov exponent, 369, 370
229, 230, 231, 246, 247, 248, 249, 253, 255, 256,
257, 403, 487
lead, 2, 12, 36, 250, 269, 275, 396, 426, 429, 430, M
436
Lebesgue measure, 492 machinery, 23
Lie algebra, vii, ix, x, xii, 3, 4, 9, 10, 11, 13, 15, 16, magnetic, 50, 52, 53, 58, 181
17, 18, 22, 23, 25, 26, 36, 41, 43, 44, 50, 67, 68, magnetic field, 52, 53, 58, 181
93, 125, 126, 130, 137, 139, 140, 141, 142, 144, magnetization, 181
151, 152, 166, 170, 171, 185, 187, 188, 191, 195, manifold, vii, viii, ix, x, xi, xii, xiii, 3, 8, 9, 20, 21,
205, 206, 207, 216, 217, 218, 219, 222, 226, 230, 22, 25, 36, 40, 41, 42, 44, 48, 52, 53, 61, 62, 63,
232, 255, 259, 261, 268, 272, 277, 280, 283, 284, 64, 65, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77,
285, 286, 287, 288, 289, 290, 292, 294, 297, 299, 78, 79, 80, 83, 84, 86, 89, 90, 96, 98, 121, 122,
300, 303, 304, 325, 326, 327, 328, 331, 332, 333, 139, 140, 142, 143, 144, 165, 169, 170, 171, 175,
334, 335, 336, 337, 338, 339, 340, 341, 342, 343, 176, 186, 189, 190, 198, 199, 201, 202, 204, 205,
344, 345, 347, 348, 352, 353, 354, 357, 358, 360, 209, 224, 238, 239, 241, 246, 247, 248, 249, 251,
362, 364, 366, 367, 369, 372, 401, 402, 403, 404, 256, 257, 259, 267, 272, 274, 279, 280, 281, 283,
405, 406, 407, 408, 409, 410, 411, 412, 413, 414, 284, 285, 286, 290, 291, 293, 296, 297, 299, 300,
Index 607
302, 303, 306, 327, 328, 329, 330, 331, 332, 333, measures, 11, 13, 32, 42, 89, 90, 91, 92, 107, 108,
334, 335, 336, 338, 349, 351, 352, 353, 361, 368, 116, 120, 281, 353, 358, 360, 361, 362, 363, 364,
369, 370, 371, 376, 377, 381, 402, 403, 404, 405, 367, 373, 374, 382, 559, 560, 599
406, 447, 448, 449, 450, 451, 452, 454, 456, 457, mechanical properties, 188
458, 459, 460, 461, 462, 464, 465, 466, 469, 470, media, 124, 138
471, 472, 473, 474, 475, 476, 477, 478, 479, 480, memory, 54, 58
481, 482, 483, 485, 486, 488, 490, 493, 495, 496, mesons, 28, 55
497, 498, 499, 500, 501, 503, 504, 505, 506, 507, metric, viii, x, xi, 10, 36, 49, 50, 51, 54, 89, 92, 93,
508, 509, 510, 511, 512, 513, 514, 516, 519, 522, 94, 96, 102, 104, 199, 200, 203, 210, 211, 221,
524, 526, 528, 530, 532, 534, 535, 536, 541, 542, 222, 238, 241, 242, 243, 244, 245, 247, 249, 251,
545, 547, 548, 550, 552, 554, 555, 556, 557, 558, 252, 253, 255, 257, 259, 260, 267, 268, 269, 270,
559, 560, 561, 563, 564, 565, 567, 568, 570, 571, 271, 272, 273, 274, 275, 276, 277, 278, 279, 280,
572, 576, 577, 580, 583, 584, 585, 586, 587, 589, 281, 283, 284, 285, 286, 287, 288, 290, 291, 293,
592, 593, 599 297, 298, 302, 307, 309, 326, 335, 337, 338, 340,
manifolds, ix, x, xiii, 3, 20, 21, 22, 25, 40, 41, 48, 52, 342, 343, 344, 345, 346, 360, 361, 369, 371, 380,
53, 61, 62, 63, 121, 122, 169, 170, 171, 175, 176, 420, 448, 459, 462, 465, 471, 472, 475, 476, 477,
186, 189, 190, 204, 209, 280, 281, 283, 284, 285, 478, 480, 488, 499, 506, 508, 558, 565, 571, 575,
291, 293, 296, 302, 306, 328, 334, 335, 349, 352, 580, 584, 590
369, 370, 371, 381, 448, 449, 450, 451, 458, 459, metric spaces, xi
461, 464, 476, 477, 479, 482, 483, 485, 486, 488, Mexico, 2, 123
490, 495, 497, 499, 500, 501, 508, 509, 510, 511, military, 1
512, 519, 534, 536, 541, 542, 554, 555, 556, 557, Minkowski spacetime, 3, 10, 36, 44, 202, 210
558, 559, 560, 561, 563, 564, 565, 567, 571, 572, MIT, 166
584, 587, 589, 592, 599 mixing, 92, 110
manners, 321 models, 12, 28, 37, 175
mapping, ix, xi, 63, 65, 67, 72, 75, 77, 86, 94, 95, 98, modules, ix, 37, 139, 141, 151, 404
100, 170, 172, 187, 188, 189, 190, 191, 195, 198, modulus, ix, 169, 170, 175, 176, 177, 179, 180, 182,
205, 208, 298, 307, 312, 316, 323, 449, 450, 452, 183, 184, 201, 254
453, 454, 455, 456, 457, 458, 460, 461, 462, 463, momentum, vii, 3, 4, 5, 6, 7, 20, 21, 45, 231, 246,
464, 466, 467, 468, 470, 471, 472, 473, 475, 476, 251, 256, 274, 285, 286, 287, 288, 289, 291, 292,
477, 478, 481, 487, 489, 490, 491, 492, 493, 494, 293, 294, 295, 296, 297, 298, 303, 304, 327, 462
495, 496, 497, 498, 499, 504, 505, 506, 509, 510, monoids, 495, 514, 587, 590
512, 513, 514, 515, 516, 517, 519, 522, 523, 525, monotone, 582
526, 527, 530, 533, 534, 535, 536, 537, 538, 540, morning, 1
542, 543, 544, 546, 547, 548, 549, 550, 551, 552, morphogenesis, 46
553, 554, 555, 556, 557, 566, 567, 568, 569, 571, Moscow, 323, 485, 558, 559, 560, 561, 563, 598,
573, 575, 576, 581, 582, 583, 587, 588, 590, 592, 599, 600
593 motion, x, xi, 2, 5, 7, 12, 39, 45, 46, 47, 48, 50, 52,
marches, 383 53, 54, 55, 58, 59, 63, 65, 71, 79, 80, 86, 89, 90,
Markov, xii, 113, 114, 116, 351, 354, 361, 376, 377, 91, 92, 102, 103, 104, 113, 114, 116, 117, 118,
379, 380, 382 120, 121, 122, 124, 178, 200, 246, 249, 250, 256,
Markov chain, 114, 116 303, 325, 326, 327, 329, 330, 332, 337, 338, 339,
Markov process, xii, 351, 354, 361, 376, 377, 380, 344, 347, 348, 355, 356, 357, 365, 366, 376, 377,
382 380, 381, 382, 383, 558
martingale, 112, 113, 114, 118, 372, 373, 375, 381 movement, 47
Massachusetts, 166, 348 multiples, 62
massive particles, 36, 250 multiplication, 7, 13, 21, 29, 170, 181, 259, 327, 337,
mathematics, vii, 1, 9, 10, 46, 53, 140, 187, 447, 486 463, 490, 491, 494, 503, 510, 511, 514, 515, 519,
Matrices, 44 521, 525, 529, 538, 541, 557, 571
matrix, 2, 3, 4, 24, 34, 35, 37, 38, 44, 93, 94, 103, multiplicity, 369
149, 155, 156, 170, 171, 172, 173, 176, 178, 179, multiplier, 251, 294, 301
182, 183, 185, 206, 234, 252, 259, 260, 285, 294,
295, 296, 303, 308, 311, 312, 328, 338, 352, 353,
355, 356, 357, 358, 359, 360, 366, 367, 372, 374, N
377, 411, 420, 451, 459, 460, 472, 473, 475, 476,
478, 488, 547 National Academy of Sciences, 55, 483
matrix algebra, 3, 35 natural, vii, 3, 23, 30, 46, 48, 54, 90, 91, 93, 97, 106,
Maxwell equations, 51, 227 176, 247, 251, 253, 257, 258, 268, 270, 272, 273,
measurement, 58 287, 326, 331, 340, 346, 353, 354, 357, 361, 370,
608 Index
373, 374, 375, 377, 378, 380, 449, 451, 453, 455, oscillations, 329, 348
456, 466, 467, 471, 493, 496, 505, 506, 510, 512, oscillator, 30, 32, 33, 35, 254, 327, 339, 391, 394
514, 517, 519, 520, 522, 524, 532, 533, 534, 537,
538, 539, 543, 544, 545, 547, 552, 563, 569, 571,
573, 576, 580, 581, 582 P
Netherlands, 59
neutrinos, 36, 39 PACS, 56
New Jersey, 560, 561 pairing, 281, 300, 519, 550, 554
New York, 55, 58, 87, 137, 138, 166, 184, 262, 264, paper, 10, 28, 37, 50, 52, 54, 55, 61, 62, 73, 81, 86,
347, 348, 349, 397, 399, 481, 558, 560, 600 170, 188, 202, 203, 210, 255, 267, 280, 309, 397,
Newton, 252 399, 448, 485, 487, 564, 565
Newtonian, 201 parabolic, 94, 95, 96, 98, 100, 406
non-Abelian, 200, 402 paradox, 30
nonlinear, ix, x, xii, 7, 21, 82, 86, 127, 138, 139, 140, parallelism, 464, 465
142, 156, 160, 166, 167, 169, 170, 175, 176, 177, parameter, viii, 63, 65, 67, 68, 69, 72, 75, 85, 91, 92,
184, 186, 199, 200, 201, 202, 203, 209, 210, 213, 93, 104, 108, 113, 114, 123, 125, 127, 128, 133,
215, 218, 219, 221, 222, 223, 246, 255, 256, 263, 134, 137, 146, 158, 159, 160, 175, 210, 212, 215,
385, 386, 398, 422 230, 231, 254, 279, 288, 289, 330, 333, 405, 409,
nonlinear optics, 175 425, 427, 428, 437, 492, 501, 502, 513, 565, 581,
nonlinear wave equations, ix, 142 583, 584, 595, 596
non-random, 368, 370, 371 Paris, 55, 121, 482, 558, 559, 560, 598, 600
normal, ix, 23, 91, 95, 99, 178, 187, 195, 278, 312, partial differential equations, viii, ix, xii, 123, 124,
333, 367, 374, 472, 474, 480, 501, 508, 510, 536, 125, 126, 127, 128, 137, 138, 385, 386, 394, 411
557, 585, 587 particle creation, 53
normalization, 271 particle mass, 254
norms, 39, 568 particle physics, 200
Norway, 58 particle-like, 253
numerical analysis, 169 particles, 17, 35, 36, 37, 42, 57, 202, 203, 247, 249,
250, 253, 254, 255, 257, 298, 329, 338
partition, 41, 209, 210, 214, 372, 379, 492, 493, 525,
O 526, 546, 551, 555
passenger, 258
oat, 301 passive, 53, 205
objectivity, 44, 61, 62, 83 pedestrians, 5, 50
observations, x, 200, 409 periodic, 46, 47, 49, 106, 598
obstruction, xii, 281, 401, 421 permit, 49
omentum, 256 personal, 12, 58
one dimension, 420, 423, 434 perturbation, 258
operator, xi, 2, 5, 6, 10, 19, 39, 40, 42, 43, 51, 66, 67, perturbations, 258, 404
68, 69, 71, 73, 74, 75, 76, 77, 78, 80, 83, 84, 86, pH, 211, 212
114, 125, 127, 130, 131, 142, 149, 154, 176, 178, phase space, 28, 145, 289, 293, 297, 326, 327, 338,
220, 223, 225, 249, 252, 253, 259, 260, 268, 274, 342, 344, 347
276, 277, 278, 286, 287, 288, 292, 295, 296, 297, phenomenology, 2, 46
298, 300, 303, 304, 307, 309, 310, 315, 316, 318, Philadelphia, 263
320, 322, 327, 342, 354, 355, 359, 360, 362, 363, photon, 39
365, 374, 375, 377, 380, 392, 393, 394, 404, 407, photons, 37, 39, 53
408, 410, 539, 550, 566, 577, 593, 594, 595 physical fields, 200
optics, 175 physical interaction, x, 199
orbit, ix, xi, 9, 126, 157, 187, 209, 212, 213, 284, physical sciences, 138
285, 286, 289, 291, 295, 296, 301, 303, 327, 328, physics, vii, xii, 1, 3, 12, 13, 17, 28, 50, 53, 55, 137,
334, 335, 336, 338, 340, 343, 344, 346, 348, 353, 138, 175, 200, 247, 249, 251, 257, 347, 387, 393,
376, 377, 380, 406, 409, 450, 451, 452, 454, 455, 398, 399, 401, 408, 447, 486, 559, 564, 600
456, 460, 465, 466, 467, 469 planar, 91, 113, 196
ordinary differential equations, 80, 126, 140, 141, plane waves, 45, 46
144, 145, 147, 172, 185, 289, 398 plasma, 175, 387
organization, 45 plasma physics, 175, 387
orientation, 35, 49, 56, 188, 450, 499, 500, 504, 516 plastic, 129, 134
orthogonality, 3, 11, 25, 29, 54, 103, 219, 230, 388 plasticity, viii, 123, 124, 128, 129, 135, 137, 138
oscillation, 34, 46, 47, 49 play, 188, 223, 247, 290, 296
Index 609
Poincare group, 50 quasilinear, 124, 127, 128, 129, 137, 138, 386
Poisson, 117, 169, 287, 295, 296, 299, 304, 306, 326, quasi-periodic, 296
331, 332, 333, 337, 339, 345, 346, 355, 356, 357,
372, 373, 381, 559
Poland, 1 R
polar coordinates, 129, 135, 136, 376, 383
polarization, 45, 267, 272 radiation, 2, 59
polynomial, 16, 21, 174, 223, 275, 365, 393, 572, radical, 2, 402, 403, 406, 407, 408, 410, 412, 413,
575, 581, 587 423, 425, 432
polynomials, 326, 332, 394, 572, 575, 589 radius, 92, 106, 110, 111, 129, 135, 136, 255, 269,
positive linear functionals, 11 496, 506, 586, 588
positron, 2, 45 random, 92, 113, 114, 355, 356, 357, 358, 366, 369,
power, 524, 563 371, 372, 373, 374, 375, 378, 379, 381, 382, 383
Prandtl, 129, 131, 132 random matrices, 381, 383
predicate, 25 random walk, 358, 366, 383
pressure, x, 129, 135, 283, 290, 291 real forms, 19, 403
private, 28, 55, 57 real numbers, 402
probability, 11, 104, 107, 108, 110, 111, 112, 113, real time, 46
114, 120, 353, 354, 358, 362, 363, 364, 368, 369, reality, 81
372, 374, 378, 379, 381, 382 reasoning, 471
probe, 55 recall, 5, 6, 20, 47, 98, 99, 109, 112, 113, 114, 188,
production, viii, 123 189, 208, 388, 392, 393, 394, 411, 439
prognosis, 2 recalling, 33, 99
program, 40, 57, 211, 279 reciprocity, 50, 53
projector, 54 recombination, 258
propagation, 127 recursion, xii, 16, 385, 386, 392, 393, 394
property, vii, viii, ix, 26, 85, 113, 114, 123, 140, 169, reduction, x, xi, xii, 185, 202, 210, 247, 250, 251,
175, 176, 184, 187, 190, 216, 222, 233, 272, 326, 253, 254, 255, 257, 281, 283, 284, 285, 286, 287,
354, 358, 368, 371, 379, 388, 392, 393, 394, 402, 294, 297, 302, 303, 385, 386, 390, 394, 397, 406,
406, 411, 480, 490, 493, 495, 501, 533, 538, 551, 410, 412, 432, 447, 483
570, 588, 589, 590 reference system, 50, 58
proposition, 101, 198, 421, 437, 455, 473, 479, 529, reflection, 293, 497
553 regular, 97, 98, 109, 110, 127, 144, 287, 367, 378,
prototype, x, 199, 203 379, 450, 451, 453, 466, 481
pseudo, xiii, 5, 39, 335, 447, 462, 474, 476, 477, relatives, 36
548, 550 relativity, 2, 50, 52, 58
relevance, 27, 263
reproduction, viii, x, 123, 127, 138, 200, 203
Q research, vii, xii, 2, 49, 55, 87, 140, 157, 166, 401,
439
QCD, 57 residues, 90, 91, 97, 98
QED, 196 resolution, 124, 137, 188, 522, 523, 525, 529, 544,
quadrupole, 57 545
qualitative research, 157 rings, 510, 514, 515, 519, 520, 521, 523, 541, 571,
quanta, 32, 33, 42, 52, 53 572, 573, 582
quantization, 5, 12, 24, 30, 52, 55, 200, 274, 349, robotics, ix, 187
447, 482, 558 rolling, 303
quantum, vii, viii, 2, 3, 4, 5, 10, 11, 12, 13, 20, 24, rotation axis, 65, 76
25, 30, 34, 40, 42, 44, 49, 55, 59, 200, 201, 202, rotation transformation, 134
247, 255, 326, 327, 348, 391, 394, 398, 559, 599 rotations, viii, 5, 26, 33, 61, 62, 63, 67, 70, 71, 81,
quantum chromodynamics, 200, 201 83, 86, 87, 88, 239, 295, 365, 383, 466
quantum field theory, 200 Royal Society, 55
quantum gravity, 202 Russia, 447, 485, 563
quantum groups, 10 Russian, 137, 138, 306, 307, 323, 483
quantum mechanics, 5, 20, 55, 59, 327, 398, 599
quantum state, 10, 247
quantum theory, 2, 4, 10, 559 S
quark, 3, 28, 35, 38
quarks, 27, 32, 33, 34, 55 SAE, 58
610 Index
sample, 369 50, 51, 56, 199, 200, 201, 202, 203, 210, 217, 221,
sampling, 118 222, 226, 227, 230, 231, 233, 235, 236, 241, 242,
scalar, x, 4, 5, 9, 12, 18, 29, 34, 47, 50, 51, 52, 53, 244, 246, 247, 249, 250, 253, 255, 256, 257, 258,
59, 201, 221, 229, 232, 233, 242, 248, 251, 252, 408
253, 254, 256, 257, 267, 264, 280, 281, 286, 297, spatial, 13, 46, 48, 63, 68, 70, 71, 73, 74, 76, 77, 78,
387, 471, 487 79, 80, 81, 86
scalar field, 50, 52, 248, 251, 252, 253, 254, 257 spatial representations, 63, 79
scaling, 103, 113, 250, 269, 386, 387, 409, 417 spectrum, 49, 57, 255, 420, 422, 428, 430, 431, 573
scattering, 55 speed, 50, 58, 91, 105, 106, 250
school, 28 speed of light, 50, 58
Schrodinger equation, ix, xii, 169 spheres, viii, 89, 90, 92, 485, 486, 487, 494, 501
scientists, 55 spin, vii, 2, 3, 4, 5, 7, 8, 19, 22, 28, 31, 39, 45, 54,
s-compact, 556 78, 181, 186, 201, 231, 237, 239, 242, 246, 251,
S-curves, 133 256, 262, 486
search, 27, 337 spin-1, 35
searching, 1, 6 spinning particle, 202
self-interactions, 249 spinor fields, 202, 243
semigroup, 353, 354, 361, 362, 363, 364, 367, 368, SRT, 59
369, 371, 372, 374, 376, 377, 379, 380 stability, ix, 169, 175, 180, 182, 184, 187, 190, 193,
separation, 258, 496 195, 196, 198, 281, 284, 301, 304, 382, 402, 406,
series, 17, 50, 98, 143, 185, 202, 255, 284, 308, 309, 407, 412, 413, 418, 419, 483
358, 359, 360, 563, 572, 573, 575, 577, 579, 580, stabilizers, 274, 277, 279
582, 583 standard model, 4, 5, 11, 27, 35, 37, 39, 44, 48, 50,
Shanghai, 167 200
shape, 210, 464 statistics, 28, 381
shear, x, 129, 199, 210, 215, 219, 225, 255 stochastic, 102, 104, 121, 354, 355, 356, 357, 366,
Shell, 32, 58 368, 369, 370, 371, 372, 373, 375, 376, 381, 382,
shock, 127 560
shock waves, 127 stochastic processes, 560
shy, 1 stratification, 451, 456, 470, 474, 476, 478
SIGMA, 397, 399 streams, 50
sign, 34, 107, 250 strength, x, 39, 52, 200, 222, 223, 240, 256
similarity, 243 stress, viii, 89, 92, 128, 129, 207, 215, 241, 248, 249,
simulation, 186 253, 256, 261, 396, 460
sine, 34 string theory, 447
Singapore, 559 stroke, 44
singular, 90, 92, 97, 98, 109, 110, 112, 121, 128, strong force, 47, 51, 200
157, 197, 279, 280, 333, 341, 342, 390, 430, 450, strong interaction, 1, 27, 48, 49, 55, 56
469, 539, 547, 548, 549 STRUCTURE, 139, 199, 459, 460, 485
singularities, ix, 53, 127, 187, 188, 189, 193, 195, subgroups, xiii, 17, 35, 43, 93, 188, 209, 279, 291,
198, 247, 447, 500 348, 366, 367, 450, 469, 507, 509, 560, 563, 564,
smoothness, xiii, 171, 352, 449, 465, 485, 486, 510, 565, 580, 581, 584
515, 517, 521, 526, 564, 566, 568, 577, 588 substitution, 117, 213, 216, 227, 233, 588, 590
Sobolev space, 515 summer, 28
social competence, 1 Sun, 170, 172, 174, 176, 178, 180, 182, 184, 186
sociologists, 1 superconductivity, 305
sociology, 1 superposition, 254
software, 2 supersymmetry, 36, 39
solitons, 175 supply, 496
solutions, viii, ix, 2, 6, 19, 21, 45, 55, 123, 124, 125, suppression, 27, 28
126, 127, 128, 129, 130, 135, 137, 138, 139, 140, switching, xii, 401
142, 156, 158, 161, 162, 164, 167, 169, 170, 175, symbols, 13, 242, 249, 309, 312
180, 203, 252, 253, 254, 255, 257, 268, 305, 308, symmetry, vii, viii, x, xi, xii, 1, 21, 27, 32, 33, 34,
312, 330, 337, 386, 393, 394, 396, 397, 411, 424, 36, 37, 38, 39, 42, 50, 51, 55, 58, 123, 124, 125,
433 127, 128, 140, 166, 172, 176, 178, 182, 184, 194,
sorting, 73 199, 200, 201, 202, 208, 210, 224, 226, 231, 239,
Southeast Asia, 559, 599 248, 256, 283, 284, 285, 288, 289, 290, 291, 298,
spacetime, x, 3, 10, 11, 12, 18, 20, 23, 25, 27, 28, 32, 299, 300, 302, 303, 306, 385, 386, 387, 388, 389,
33, 34, 35, 36, 39, 42, 43, 44, 45, 46, 47, 48, 49,
Index 611
390, 391, 392, 394, 395, 396, 397, 398, 401, 404, torus, ix, 277, 294, 296, 302, 332, 333, 478, 481,
408 497, 498, 499
symplectic, 169, 185, 200, 221, 222, 247, 257, 268, total energy, 256
272, 273, 274, 275, 279, 281, 296, 306, 325, 326, traction, 524
327, 328, 329, 330, 331, 333, 334, 335, 336, 338, tradition, 15
339, 340, 342, 343, 345, 347, 447, 460, 462, 464, trajectory, 12, 185, 286, 287, 293, 294, 304
482, 483 trans, 507
synchronous, 46, 48 transformation, vii, viii, 36, 50, 51, 52, 53, 61, 62,
systems, xi, 19, 123, 125, 166, 186, 238, 284, 285, 63, 68, 71, 72, 73, 74, 78, 82, 83, 84, 85, 86, 123,
286, 288, 292, 294, 302, 304, 305, 306, 312, 325, 124, 125, 126, 129, 131, 137, 140, 145, 147, 156,
326, 327, 335, 341, 347, 348, 369, 402 157, 158, 159, 160, 162, 164, 171, 203, 204, 208,
209, 210, 212, 213, 214, 215, 216, 221, 222, 227,
230, 231, 232, 234, 235, 236, 237, 238, 242, 243,
T 255, 258, 294, 371, 408, 448, 450, 460, 466, 472,
474, 478, 480, 482, 532, 533, 552, 555, 564
Taiwan, 351 transformation matrix, 234
teachers, 1 transformations, viii, ix, x, 17, 24, 25, 50, 51, 52, 65,
temporal, 46, 49 78, 123, 124, 125, 126, 127, 128, 130, 131, 134,
tensor field, 50, 453, 460, 461, 462 145, 199, 202, 203, 205, 208, 209, 210, 214, 215,
tetrad, x, 199, 200, 201, 202, 203, 220, 221, 222, 230, 231, 232, 233, 234, 235, 236, 239, 243, 250,
226, 233, 235, 236, 241, 255 251, 256, 259, 269, 305, 333, 371, 390, 391, 415,
Texas, 263 416, 417, 422, 423, 424, 425, 429, 430, 447, 448,
textbooks, 63, 65 450, 459, 460, 463, 465, 466, 467, 471, 472, 473,
theory, vii, viii, ix, x, xi, xii, 2, 5, 8, 9, 28, 35, 36, 37, 474, 476, 480, 481, 482, 560
50, 53, 58, 121, 122, 123, 124, 129, 137, 139, 140, transition, xii, 33, 35, 114, 204, 205, 208, 236, 354,
141, 142, 144, 157, 171, 175, 187, 188, 195, 199, 361, 362, 376, 377, 379, 401, 489, 496, 502, 512,
200, 201, 202, 203, 221, 247, 249, 253, 254, 255, 526, 546, 569
256, 257, 264, 284, 306, 308, 313, 325, 326, 327, transition rate, 35
337, 339, 347, 348, 351, 366, 369, 394, 396, 401, transitions, 34
402, 408, 447, 486, 552, 558, 559, 560, 563, 564, translation, 68, 83, 131, 138, 201, 230, 259, 352,
577, 599, 600 354, 391, 457, 474, 478, 480
third order, viii, 61, 62, 81, 182, 184 translational, 201, 210, 214, 215, 218, 219, 221, 222,
three-dimensional, 65, 263 223, 231, 255, 256
time, x, 2, 3, 8, 10, 11, 27, 28, 30, 34, 36, 45, 46, 47, transparent, 376
48, 49, 51, 52, 62, 73, 74, 76, 77, 78, 80, 81, 82, transport, 33, 485, 486, 490, 491, 493, 494, 495, 506,
83, 86, 91, 107, 108, 113, 114, 117, 170, 176, 178, 507, 508, 509, 511, 513, 516, 517, 522, 535, 537,
180, 182, 195, 200, 205, 209, 247, 250, 252, 257, 542, 554
258, 285, 289, 294, 303, 327, 353, 354, 365, 366, transpose, 3, 47, 48, 353, 358
368, 369, 371, 372, 373, 374, 376, 378, 380, 391, travel, 258
486, 487, 507, 509, 512, 523, 528, 541, 553, 565, trees, 402
571, 575, 577, 592 trend, 50
title, vii, 4, 10 triangulation, 548
Toda model, 55 triggers, 46
Tokyo, 382 two-dimensional, 57, 122, 160, 161, 178
topological, x, xiii, 10, 11, 199, 203, 224, 256, 267,
291, 351, 352, 353, 354, 447, 449, 451, 454, 455,
456, 486, 489, 490, 491, 495, 496, 497, 500, 510, U
511, 514, 515, 517, 519, 520, 521, 524, 525, 527,
529, 530, 532, 534, 537, 540, 541, 551, 554, 557, uncertainty, 5, 48, 257, 258
558, 559, 563, 564, 565, 567, 568, 569, 570, 571, unconditioned, 117, 118
573, 574, 575, 576, 577, 585, 586, 587, 588, 589, unification, 2, 35, 202, 203, 227, 247
590, 591, 592, 593, 594, 595 uniform, 92, 111, 112, 120, 129, 277, 315, 317, 319,
topological invariants, x, 199, 203, 224, 256 320, 322, 359, 366, 488, 490, 491, 492, 493, 508,
topology, 188, 190, 257, 258, 271, 279, 285, 286, 552, 554, 555, 557, 564, 567, 568, 569, 570, 573,
302, 303, 353, 354, 378, 448, 451, 456, 464, 465, 576, 577, 587, 592
469, 470, 471, 472, 473, 474, 477, 479, 481, 483, unions, 568
496, 503, 515, 534, 537, 542, 543, 554, 557, 558, universe, 10, 49, 258, 598
559, 561, 564, 565, 567, 568, 570, 573, 574, 575, USSR, 347
576, 585, 586, 588, 589, 590, 591, 592, 598
612 Index
315, 316, 317, 318, 319, 320, 322, 323, 326, 329,
V 330, 331, 334, 335, 336, 337, 339, 340, 342, 343,
352, 354, 355, 359, 371, 374, 375, 380, 392, 393,
vacuum, 49, 255 394, 395, 396, 397, 403, 453, 457, 458, 462, 463,
validation, 309 464, 470, 471, 473, 477, 487, 490, 515, 524, 525,
validity, 257 529, 530, 532, 555, 560, 565, 567, 593, 594, 595,
values, viii, 61, 62, 82, 84, 85, 86, 134, 170, 208, 597, 600
251, 254, 278, 284, 287, 289, 304, 375, 404, 409, velocity, x, 49, 50, 52, 62, 63, 64, 74, 76, 77, 78, 82,
413, 421, 423, 424, 425, 426, 427, 429, 430, 458, 83, 157, 158, 159, 178, 250, 283, 284, 285, 287,
471, 506, 512, 523, 524, 530, 539, 542, 555, 567, 290, 291, 292, 293, 294, 301, 302, 303, 330, 340
572, 575, 591, 598, 599 Victoria, 186
variable, 114, 133, 134, 200, 248, 296, 306, 332, visible, 365
356, 375, 378, 386, 388, 394, 395, 512, 556, 563 voice, 39
variables, 62, 91, 92, 104, 108, 114, 124, 125, 128, vortex, x, xi, 283, 284, 285, 291, 292, 293, 295, 296,
129, 131, 132, 133, 134, 135, 136, 145, 155, 163, 302, 306
188, 198, 200, 228, 229, 231, 233, 253, 270, 297, vortices, 285
310, 316, 325, 333, 337, 356, 366, 373, 374, 388,
389, 395, 396, 486, 556, 559, 560, 565, 566, 579,
594, 595, 596, 598 W
variance, 91, 102, 104, 108
variation, 80, 112, 114, 137, 215, 225, 226, 227, 228, war, 104
229, 230, 231, 232, 243, 244, 245, 246, 358, 359, wave equations, 2, 55
412 wave number, 46
vector, viii, ix, x, xi, 4, 7, 12, 13, 17, 24, 27, 28, 37, wave propagation, 185
46, 47, 51, 52, 53, 61, 62, 63, 64, 65, 66, 67, 68, weakness, 39
70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, Weinberg, 37, 264
83, 84, 85, 86, 93, 94, 96, 97, 106, 139, 140, 141, workspace, 188
142, 143, 144, 145, 148, 149, 150, 152, 156, 157, writing, 1, 29, 231, 395, 396
165, 166, 171, 175, 176, 178, 179, 181, 183, 190,
192, 196, 199, 201, 205, 207, 215, 220, 221, 227,
234, 236, 244, 250, 259, 260, 261, 268, 271, 277, Y
278, 281, 283, 284, 285, 286, 287, 288, 289, 290,
291, 292, 293, 295, 296, 297, 298, 299, 300, 301, Yang-Mills, x, 200, 201, 202, 256, 402
302, 303, 304, 307, 308, 309, 310, 311, 312, 313, yield, 90, 124, 128, 456, 471