WITH APPLICATIONS TO
MATHEMATICAL PHYSICS
RUDOLF SCHMID
Department of Mathematics
EMORY University
Atlanta, GA 30032 , USA
e-mail: rudolf@mathcs.emory.edu
web: www.math.emory.edu/rudolf
Abstract
We give a survey of innite dimensional Lie groups and show
some applications and examples in mathematical physics. This in-
cludes dieomorphism groups and their natural subgroups like vol-
ume preserving- , symplectic- and contact transformations, as well as
gauge groups , quantomorphisms and loop groups. Applications in-
clude uid dynamics, Maxwells equations, plasma physics and BRST
symmetries in quantum eld theory. We discuss the Lie group struc-
tures of pseudodierential- and Fourier integral operators, both on
compact and non-compact manifolds and give applications to the KdV
equation and quantization.
1
INTRODUCTION: Lie Groups play an important role in physical sys-
tems both as phase spaces and as symmetry groups. Innite dimensional
Lie groups occur in the study of dynamical systems with an innite number
of degrees of freedom such as PDEs and in eld theories. For such innite
dimensional dynamical systems dieomorphism groups and various exten-
sions and variations thereof, such as gauge groups, loop groups and groups
of Fourier integral operators occur as symmetry groups and phase spaces.
Symmetries are fundamental for Hamiltonian systems. They provide conser-
vation laws (Noether currents) and reduce the number of degrees of freedom,
i.e the dimension of the phase space. Cohomological aspects of Lie groups
come into the picture when studying anomalies and BRST symmetries in
quantum eld theory.
This paper is based on a series of lectures given at the 4th International
Conference on Geometry, Integrability and Quantization in Sts Constantine
and Elena (Varna) Bulgaria, June 5-16, 2002. We thank the organizers for
their good work and their hospitality.
The topics selected for these lectures aim to illustrate some of the ways
innite dimensional geometry and global analysis can be used in mathemat-
ical problems of physical interest.
1. Innite dimensional Lie groups
2. Dieomorphism groups
3. Subgroups of dieomorphism groups and applications
4. BRST symmetries
5. Lie groups of pseudodierential - and Fourier integral operators
6. Dieomorphism groups and FIO for non compact manifolds
7. Applications to uid dynamics and quantization.
1 INFINITE DIMENSIONAL LIE GROUPS
In physics Lie groups appear as symmetry groups or conguration spaces
of dynamical systems. Some classical, nite dimensional examples are: The
linear and angular momentum are related to the groups of translations and
rotations; a rigid bodys position f(x, t) at time t is determined by f(x, t) =
A(t) x , where A(t) SO(3) is a proper rotation. The group SO(3) is
conguration space and symmetry group. The heavy top has as conguration
space also SO(3) and in addition the circle group S
1
as a symmetry group,
2
which represents rotations about the direction of gravity. Eliminating this
S
1
symmetry leaves the Euclidean group E
3
of rigid motions.
Some innite dimensional examples are: For an incompressible uid the
conguration space is Diff
vol
() , the volume preserving dieomorphisms
of a region . This is an innite dimensional Lie group. For compress-
ible uids the conguration space is Diff() and as symmetry group acts
Diff
: ( ( : (g) = g
1
inversion is C
.
In nite dimensions, the second condition follows from the implicit function
theorem ! The group operations satisfy associativity of the product g(hk) =
(gh)k and ge = eg = e , gg
1
= g
1
g = e. As a manifold, ( is locally
dieomorphic to an innite dimensional vector space V , which can be a
Banach space (with norm |.|), a Hilbert space (with inner product < ., . >
and norm |x|
2
=< x, x >), or a Frechet space (with metric d(., .) but no
norm). Correspondingly we call these Banach Lie groups, Hilbert Lie groups
or Frechet Lie groups. If ( is locally dieomorphic to R
n
,n < , then ( is
a nite dimensional Lie group.
1.1.2 The Dierentiable Structure of (
For g ( consider left- and right translations , L
g
: ( (; L
g
(h) = gh
and R
g
: ( (; R
g
(h) = hg. These are dieomorphisms with inverses given
3
by (L
g
)
1
= L
g
1 and (R
g
)
1
= R
g
1 . Let (U, ) be a chart at the identity
e ( and dene a chart (U
g
,
g
) at g ( by U
g
= L
g
(U) = L
g
h ; h U
and
g
= L
g
1 : U
g
V, h (g
1
h). So a single chart at the identity
denes the dierentiable structure of ( with smooth transition maps (change
of coordinates) given by
g
1
h
= L
g
1
h
.
1.1.3 The Lie Algebra g of a Lie Group (
The Lie algebra g of a Lie group ( is dened as the space of left invariant
vector elds on ( , which is isomorphic to T
e
( the tangent space at the
identity e. Let X(() denote the space of all vector elds on (, which is an
innite dimensional Lie algebra with Lie bracket given by the commutator
bracket [X, Y ] = XY Y X for X, Y X((). A vector eld X X(() is
called left invariant i L
g
X = X. If X, Y are left invariant, then L
g
[X, Y ]
= [L
g
X, L
g
Y ] = [X, Y ], i.e. [X, Y ] is left invariant. Hence the space of left
invariant vector elds X
L
(() is a Lie subalgebra of X(().
The Lie algebra g of ( is dened as g = X
L
((). Let T
e
(, then
X
(g) := T
e
L
g
() T
g
( denes a left invariant vector eld and X
(e) = .
This denes a isomorphism between T
e
( and X
L
((). We dene the Lie
bracket for any , T
e
( by
[, ] := [X
, X
](e). (1.1)
This bracket satises the conditions for a Lie algebra, i.e.[ , ] is
bilinear: [t
1
+s
2
, ] = t[
1
, ] +s[
2
, ] t, s R
skew symmetric: [, ] = [, ]
Jacobi identity [[, ], ] + [[, ], ] + [[, ], ] = 0 is satised.
With this we can identify g T
e
( as Lie algebras.
1.1.4 The Exponential Map
The exponential map exp : T
e
( g ( is dened as follows: For T
e
(
let X
(t) be the ow of
X
through e ( i.e.
(0) = e ,
(t) = X
(1). (1.2)
If ( nite dimensional exp denes a local dieomorphism from a neighbor-
hood of 0 g onto a neighborhood of e ( . So in nite dimensions, the
4
exponential map denes a local chart about e ( (called canonical chart),
hence the dierential structure of (. In innite dimensions this is not the
case in general.
1.2 Classical Lie Groups
We examine these structures and constructions on some examples.
1.2.1 Vector Groups
Let ( = V be a Banach space with (x, y) = x + y, (x) = x, e = 0,
which makes ( into an abelian Lie group i.e. (x, y) = (y, x). For the Lie
algebra we have g T
e
V V . For u T
e
V the corresponding left invariant
vector eld X
u
is given by X
u
(v) = u, v V , i.e X
u
= const.. Hence the
Lie algebra g = V with the trivial Lie bracket [u, v] = 0 is abelian. For the
exponential map we get exp : V V, exp = id
V
.
1.2.2 Matrix Groups
A) The General Linear Group GL(n, R) is dened as
( = GL(n, R) := A L(R
n
, R
n
) ; det A ,= 0
the group of invertible nn matrices. The determinant map det : L(R
n
, R
n
)
R is continuous, hence GL(n, R) = det
1
(R0) L(R
n
, R
n
) is an open
submanifold . It is disconnected and dimGL(n, R) = n
2
. The Group mul-
tiplication (A, B) = AB (matrix multiplication) is C
as the restriction
of the continuous bilinear map (A, B) L(R
n
, R
n
) L(R
n
, R
n
) AB
L(R
n
, R
n
). The group inversion (A) = A
1
(matrix inversion) is C
by the
implicit function theorem, indeed (A, (A)) = e = I the identity.
The Lie algebra: Since GL(n, R) L(R
n
, R
n
) is open T
I
GL(n, R)
L(R
n
, R
n
) with the Lie bracket [A, B] = AB BA. For A L(R
n
, R
n
) the
corresponding vector eld X
A
on on GL(n, R) given by X
A
: GL(n, R)
L(R
n
, R
n
), X
A
(Y ) = Y Ais left invariant (linear), indeed X
A
(L
Z
Y ) = X
A
(ZY ) =
ZY A = (T
Y
L
Z
)X
A
(Y ). Hence the Lie bracket on T
I
GL(n, R) dened by
[A, B] = [X
A
, X
B
](I) = DX
B
(I)X
A
(I) DX
A
(I)X
B
(I). Since X
B
(Z) is
linear we have DX
B
(I)Z = ZB and DX
B
(I)X
A
(I) = AB . So [A, B] =
AB BA is the the usual commutator bracket.
5
The exponential map: For A L(R
n
, R
n
) the curve
A
: R
GL(n, R) :
A
(t) =
i=0
t
i
i!
A
i
is a one-parameter subgroup with
A
(0) = I
and
A
(t) =
i=1
t
i1
(i1)!
A
i
=
A
(t)A. Hence
A
is the (unique) integral curve
of X
A
and the exponential map exp : L(R
n
, R
n
) GL(n, R) : exp(A) =
A
(1) =
i=0
1
i!
A
i
, becomes exp(A) = e
A
.
1.2.3 Lie Subgroups of GL(n, R)
A) The Special Linear Group SL(n, R) is dened as
SL(n, R) := A GL(n, R); det A = 1 = det
1
1.
SL(n, R) is a closed Lie subgroup of GL(n, R). It is non-compact, connected
and dimSL(n, R) = n
2
1.
The Lie algebra: sl(n, R) = A L(R
n
, R
n
) ; trA = 0
with the commutator bracket [A, B] = AB BA.
B) The Orthogonal Group O(n) is dened as
O(n) := A L(R
n
, R
n
) ; < Ax, Ay >=< x, y > AA
T
= I.
A O(n) detA = 1. O(n) is compact , disconnected (2 comp.) and
dimO(n) = n(n 1)/2.
The Lie algebra: o(n, R) = A L(R
n
, R
n
); A skew symmetric.
C) The Special Orthogonal Group SO(n) is dened as
SO(n) = A L(R
n
, R
n
) ; AA
T
= I & detA = +1
= O(n) SL(n, R) = comp
I
O(n).
SO(n) is compact, connected and dimSO(n) =
1
2
n(n 1).
The Lie algebra: so(n, R) = A L(R
n
, R
n
) ; A skew symmetric =
o(n, R).
D) The Symplectic Group Sp(2n) is dened as
Sp(2n, R) = A L(R
2n
, R
2n
) ; A
T
JA = J
where J =
_
O I
I 0
_
. Sp(2n, R) is noncompact and dimSp(2n) = 2n
2
+n.
6
The Lie algebra: sp(2n, R) = A L(R
n
, R
n
) ; A
T
J +JA = 0.
Similar constructions and results hold for the complex matrix groups
GL(n, C) and their corresponding subgroups.
1.3 Classical Results in Finite Dimensions which are
NOT True in Innite Dimensions:
1) There is NO Implicit Function Theorem or Inverse Function Theorem in
innite dimensions ! (except Nash-Moser type theorems.)
2) The exponential map exp : g G dened by exp() =
(1), where
(t)
is the ow of the left invariant vector eld X
(M) C
(M) C
(M). This is a
C
(M) with
respect to the C
k
-norm, k < , and obtain the Banach Lie group C
k
(M)
, or with respect to the H
s
-Sobolev norm, s >
1
2
dimM, and obtain the
Hilbert Lie group H
s
(M). For example, if (M, g) is a Riemannian manifold
with Levi-Civita connection the H
s
(M) is the completion of C
(M) with
respect to the norm |f|
s
= (
_
M
s
i=1
[
i
f[
2
x
dvol
x
(g))
1/2
< .
Application: Maxwells Equations
Let E, B be the electric and magnetic elds respectively. Then the vacuum
Maxwell equations are
E = curl B,
B = curl E, div B = 0, div E = . (1.3)
Let V be the space of vector elds (potentials) on R
3
and the phase space P =
T
V = V V
V are
A =
H
E
,
E =
H
A
. (1.5)
With the Hamiltonian (energy) H(A, E) =
1
2
_
([curl A[
2
+[E[
2
)dx and B :=
curl A we get divB = div curlA = 0 and Hamiltons equations
A =
H
E
=
E
B = curl E and
E =
H
A
= curl curlA = curlB . These are the
rst 3 of Maxwells equations, the 4th equation divE = is obtained from
gauge symmetry and reduction.
Gauge invariance: The Lie group ( = C
(R
3
) acts on V by A =
A + ( (, A V ). We lift this action to P = V V
by (A, E) =
(A +, E). The Hamiltonian H is ( invariant and has a momentum map
J : V V
becomes P
= J
1
()/( = (E, B) ; div E = , div B = 0 and the
reduced Hamiltonian becomes H
(E, B) =
1
2
_
([E[
2
+ [B[
2
)dx . Computing
the reduced Poisson bracket on P
we get
F, H
(E, B) =
_
(
F
E
curl
H
B
H
E
curl
F
B
)dx . (1.6)
Now Hamiltons equations on the reduced phase P
F = F, H
_
E = curl B ,
B = curl E
div B = 0 , div E = .
(1.7)
1.4.3 Loop Groups
Let M be a nite dimensional manifold and ( = C
(M, R). If
M is compact then C
k
(M, R0) is a Banach Lie group for k < and
H
s
(M, R0) (dieomorphisms of H
s
Sobolev class) is closed under multi-
plication if s >
1
2
dimM . Then H
s
(M, R0) is a Hilbert Lie group.
As a generalization we replace R 0 by any nite dimensional Lie
group G with Lie algebra g. Then ( = C
k
(M, G) with pointwise dened
multiplication and inversion, (i.e. for f, g (, x M , (f, g)(x) = f(x)g(x)
and (f)(x) = f
1
(x)) is a Banach Lie group with Lie algebra g = C
k
(M, g)
where the bracket is again dened pointwise [, ]
g
(x) = [(x), (x)]
g
for
, g, x M .
The exponential map EXP : g = C
k
(M, g) ( = C
k
(M, G) is given by
EXP() = exp , where exp : g G is the nite dimensional exponential
map. In this case EXP is a local dieomorphism .
Special case: If M = S
1
the circle then ( = C
k
(S
1
, G) = L
k
(G) is
known as the loop group and g = C
k
(S
1
, g) = l
k
(g) its loop algebra, see [29]
for details.
Applications: These innite dimensional Lie groups have wide appli-
cations in gauge theories and quantum eld theory. Their Lie algebras and
their representation theory play an important role to ane Lie algebras,
Kac-Moody Lie algebras (central extensions) , vertex algebras, completely
integrable systems and soliton equations (Toda, KdV, KP equations).
9
2 DIFFEOMORPHISM GROUPS
2.1 Overview of Diff(M)
Dieomorphism groups and their subgroups provide an important and wide
range of examples of innite dimensional Lie groups with very interesting
applications to mathematical physics. Let M be a smooth manifold, and
consider
Diff(M) = f : M M ; f dieomorphism. (2.1)
2.1.1 The Algebraic Structure of Diff(M)
Diff(M) is a group with composition as group operation
: Diff(M) Diff(M) Diff(M), (f, g) = f g (2.2)
and inversion
: Diff(M) Diff(M) , (f) = f
1
. (2.3)
The unit element e is the identity map e = id
M
: M M . We call Diff(M)
the dieomorphism group of M. There are very interesting algebraic
properties like cohomologies, isomorphy types, simplicity ect. studied mainly
by A. Banyaga in the 70s, [5].
We concentrate here on the geometric structures and its applications.
2.1.2 Geometric Structure of Diff(M)
We investigate the topology , the manifold structure and global analysis on
Diff(M). One of the main question is, in what sense is Diff(M) a Lie
group ?
For dierent types of dieomorphisms we will obtain dierent structures:
C
k
, k < dieomorphisms Diff
k
(M) is a Banach manifold (Palais
[28], Omori [26]).
C
dieomorphisms Diff
(M) = lim
s
Diff
s
(M) as inverse limit a Frechet Lie group called
ILH Lie group (Inverse Limit of Hilbert Lie groups) . We give the exact
denition and properties below.
2.2 The Manifold Structure on C(M, M)
Well see that the dieomorphism group Diff(M) is an open set of the
manifold of all maps C(M, M). So we rst need to construct a manifold
structure on the spaces of maps C
(M, M), C
k
(M, M) and C
s
(M, M) .
Let M be a compact and N any nite dimensional manifold and consider
C(M, N) = f : M N ; f dierentiable. (2.4)
We rst consider two dierent manifolds M and N to make the construc-
tion clearer, later well set N = M. Also we dont specify at the moment the
dierentiability class C
, C
k
or H
s
.
2.2.1 The Parameter Spaces of C(M, N)
In order to construct an innite dimensional manifold structure on a set like
C(M, N) one rst has to nd a candidate for the parameter spaces. As for
nite dimensional manifolds, these parameter spaces should be isomorphic to
the tangent spaces at the corresponding points. Heuristically one determines
the tangent space as follows:
11
Let f C(M, N) and : I R C(M, N) be a curve with (0) = f.
Then (0) is a tangent vector to C(M, N) at the point f, i.e.
(0) =
d
dt
t=0
(t) T
f
C(M, N). (2.5)
We make the following identication to interpret (0): For each x M let
x
: I R N be the curve in N dened by
x
(t) = (t)(x) . Then
x
(0) = f(x) and
x
(0) T
f(x)
N , i.e.
x
(0) is a tangent vector to N at the
point f(x). We identify
x
(0) (0)(x); hence we can regard (0) as a map
(0) : M TN such that (0)(x) T
f(x)
N . That means (0) is a vector
eld along f. With this motivation we dene the tangent space of C(M, N)
at the point f C(M, N) by
T
f
C(M, N) =
f
C(M, TN) ;
N
f
= f (2.6)
where
n
: TN N is the canonical projection. We can identify T
f
C(M, N)
with the space (f
N
) of sections of the pull back bundle f
N
. This is
the parameter space at the point f C(M, N), i.e. T
f
C(M, N)
= (f
N
)
which is an innite dimensional vector space.
Note, for N = M and f = id : M M we have T
id
C(M, M) = (
M
) =
X(M) the space of vector elds on M.
For dierent dierentiability types of f : M N we choose the corre-
sponding topologies on the vector spaces (f
M
) as follows:
a) C
(f
M
) of C
i
(x)| < (2.7)
where
i
is the local representative of
(f
M
) in a chart U
i
of M.
Note there is no single norm dening this topology, but it is metrizable and
complete.
b) C
k
(M, N), 0 k < : The space
k
(f
M
) of C
k
-dierentiable sec-
tions with the uniform C
k
-topology is a Banach space with norm
|| = max
0pk
[[
p
, with [[
p
as above (2.8)
12
c) C
s
(M, N) , s >
1
2
dim M: The space
s
(f
M
) of H
s
-Sobolev dieren-
tiable sections is a Hilbert space with inner product
< , >= max
i
_
U
i
(
0|k|s
D
k
i
(x)D
k
i
(x))dx . (2.9)
2.2.2 Local Charts (coordinates) of C(M, M)
Now let N = M in the above construction. We choose a Riemannian metric
on M, which allows us to dene a (Riemannian) exponential map exp :
TM M as follows. For each v
x
T
x
M there is a unique geodesic
x
through x whose tangent vector at x is v
x
, i.e
x
(0) = x and
x
(0) = v
x
.
Then dene
exp
x
(v
x
) :=
x
(1) , v
x
T
x
M. (2.10)
In general exp
x
is a local dieomorphism from a neighborhood of 0 T
x
M
onto a neighborhood of x M. However, since M is compact, exp
x
is dened
on all of T
x
M and can be extended to a map exp : TM M such that the
map
Exp := (
M
, exp) : TM M M , Exp(v
x
) = (x, exp
x
(v
x
))
is a dieomorphism from a neighborhood O(0) of the zero section in TM
onto neighborhood |() of diagonal M M,
Exp : TM O(0) |() M M. (2.11)
We dene f : M M to be close to the identity id : M M i graph(f)
|() . This denes a neighborhood 1(id
M
) in C(M, M) by
1(id
M
) = f C(M, M) ; graph(f) |(). (2.12)
We dene : M TM to be close to the zero section 0 : M TM i
(M) O(0). This denes a neighborhood J(0) in X(M) = T
id
C(M, M)
by
J(0) = C(M, TM) ; (M) O(0). (2.13)
We want to point out that compactness of M is important. We have the
following lemma:
13
Lemma 2.1 If : E M is any vector bundle over M , denote by () =
: M E ; = id the space of sections. Let O E be an open set
of E. Then the set (O) := () ; (M) O is open in () (open
in any of the topologies
,
k
,
s
) if and only if M is compact.
From this lemma follows that the neighborhood J(0) dened above is
open in (
M
) = X(M) = T
id
C(M, M).
Dene : 1(id
M
) C(M, M) J(0) X(M) by
(f) = Exp
1
(id
M
, f) , f 1(id
M
). (2.14)
is a bijection from 1(id
M
) onto J(0) with inverse
1
() = pr
2
Exp , J(0) (2.15)
where pr
2
: M M M is the projection onto the second factor.
Therefore we have a chart about the identity id
M
on C(M, M)
: 1(id
M
) C(M, M) J(0) (
M
). (2.16)
A similar construction yields charts about any f C(M, M), i.e. there
exists open neighborhoods J
f
in (f
M
) and 1
f
in C(M, M) and bijections
f
: 1
f
J
f
where
f
(g) = Exp
1
f
(id
M
, g) with Exp
f
= (
f
M
, exp f).
2.2.3 Change of Coordinates:
The change of charts (change of coordinates) for any f, g C(M, N) is give
by the map
f,g
=
g
1
f
:
f
(1
f
1
g
) (f
M
) (g
M
) (2.17)
which turns out to be
f,g
(
f
) = Exp
1
g
Exp
f
f
(2.18)
where Exp
f
= (
f
M
, exp f) and Exp
g
= (
g
M
, exp g) are the from Exp
induced local dieomorphisms in the bundles (f
M
) and (g
M
) respec-
tively. To show that these transition maps
f,g
are smooth one needs the
-Lemma (Palais [28]).
14
Lemma 2.2 (-Lemma) Let : E M and
: E
M be vector
bundles over M (compact !) and let O E be open. Let : O E E
be
a C
: (O) () (
) ,
() :=
is a C
, i.e D
k
() =
k
, where
k
denotes the ber derivative.
Remarks: For the dierent topologies
,
k
,
s
we have:
a)
() and
) is C
0
in the sense of Keller [19].
b)
k
, 0 k < : If is C
then
:
k
(O)
k
()
k
(
) is C
.
If is C
k+p
, then
:
k
(O)
k
()
k
(
) is C
p
.
c)
s
, s >
1
2
dimM: If is C
then
:
s
(O)
s
()
s
(
) is C
.
If is Sobolev H
s+t
, then
:
s
(O)
s
()
s
(
) is H
t
.
It follows from the -Lemma (2.2) that the change of charts
f,g
is C
,
because
f,g
=
for = Exp
1
g
Exp
f
: O
f
f
TM g
TM and is
C
(M, M), C
k
(M, M) , 0
k < and C
s
(M, M) , s >
1
2
dimM are innite dimensional C
Frechet-,
Banach- and Hilbert manifolds, respectively.
Remark: A similar construction works for the space C(M, N),where M
is a compact manifold and N is a Riemannian (not nec. compact) manifold.
Again C
(M, N), C
k
(M, N), 0 k < and C
s
(M, N), s >
1
2
dimM are
innite dimensional C
,in
order to dene a Lie group structure on Diff
s
(M). The answer in NO! The
situation is more complicated as indicated at the beginning.
Right multiplication: Let g Diff
s
(M). Right multiplication is the
map R
g
: Diff
s
(M) Diff
s
(M) dened by R
g
(f) = f g. Its tangent
map (derivative) at any f Diff
s
(M) is the map T
f
R
g
: T
f
Diff
s
(M)
T
fg
Diff
s
(M) : T
f
R
g
(
f
) =
f
g. So TR
g
= R
g
, therefore R
g
is C
for
any g Diff
s
(M).
Left multiplication: Let g Diff
s
(M). Left multiplication is the map
L
g
: Diff
s
(M) Diff
s
(M) : L
g
(f) = g f . Its tangent map (deriva-
tive) at f Diff
s
(M) is the map T
f
L
g
: T
f
Diff
s
(M) T
gf
Diff
s
(M) :
T
f
L
g
(
f
) = Tg
f
. Therefore TL
g
= L
Tg
. But if g is H
s
then Tg is H
s1
therefore L
g
: Diff
s+k
(M) Diff
s
(M) is C
k
for any g Diff
s+k
(M) ,
(-Lemma).
Together we get: The group multiplication is C
k
as a map
: Diff
s+k
(M) Diff
s
(M) Diff
s
(M) . (2.19)
Note k = 0 the multiplication is only continuous between the same spaces
Diff
s
(M).
Similar, the group inversion is C
k
as a map
: Diff
s+k
(M) Diff
s
(M) , (2.20)
and k = 0 the inversion is only continuous between the same spaces
Diff
s
(M).
16
2.3.2 The Lie Algebra of Diff
s
(M)
Recall for a nite dimensional Lie group G, the Lie algebra structure on g
=
T
e
G (e=identity) is dened as follows: for , T
e
G dene the corresponding
left invariant vector elds X
(g) := T
e
L
g
() , X
(g) := T
e
L
g
() , g G .
Then dene the Lie bracket on g by [, ] := [X
, X
](e), where [X
, X
] =
X
and X
g
Y = Y g Diff
s
(M).
Let T
id
M
Diff
s+1
(M) = X
s+1
(M) be an H
s+1
-vector elds on M, and
dene the right invariant vector eld Y
on Diff
s
(M) by
Y
(f) := TR
f
() = f , f Diff
s
(M). (2.21)
Y
is a C
1
vector eld on Diff
s
(M) if X
s+1
(M). The bracket [Y
, Y
] of
two right invariant vector elds is given by
[Y
, Y
](id
M
) = [, ] , , X
s+1
(M) (2.22)
where the bracket [, ] is the commutator bracket of the vector elds and
on M ,i.e. [, ] = . If , X
s+1
(M) the [, ] X
s
(M), hence
[Y
, Y
](id
M
) X
s
(M) and [Y
, Y
] is a right invariant C
0
vector eld on
Diff
s
(M). So the ordinary Lie bracket [, ] = of vector elds on M
is the bracket on the Lie algebra g = T
id
M
Diff
s+1
(M) but g is not closed
under the bracket! we loose derivatives. This type of Lie algebra structure
is called ILH Lie algebra which well dene below.
2.3.3 The Exponential Map exp : g (
Recall for a nite dimensional Lie group G with Lie algebra g = T
e
G the
exponential map exp : g G is dened as follows: For g let X
be
the corresponding left invariant vector eld and
: R G smooth,
(0) = e and
(t) = X
(t)). Now
dene exp() :=
on Diff
s
(M) through e = id
M
. Indeed, (0) =
0
= id
M
, and for x M
(t)(x) =
d
dt
[
t=0
(t)(x) =
d
dt
[
t=0
t
(x) = (
t
(x)) = ( (t))(x) = TR
t
(x) =
Y
((t))(x). Hence
(t) = Y
(
t
) and
(0) = . Now dene
exp() :=
1
. (2.24)
Then exp : X
s
(M) Diff
s
(M) is continuous, but not C
1
. Moreover exp is
not locally surjective, i.e. exp does not map onto any neighborhood of the
identity in Diff
s
(M). This is the reason we cannot use this exponential map
to dene directly charts on Diff
s
(M). In other words, the dieomorphism
group ( = Diff
s
(M) is not generated by exponentiating vector elds. In
every neighborhood of id
M
there are dieomorphisms which are not part of
the ow of any vector eld, in short
Diff
s
(M) ,= exp X
s
(M).
Summarizing we have the following Lie group structure of Diff
s
(M):
(i) multiplication : Diff
s+k
(M) Diff
s
(M) Diff
s
(M) , is C
k
(ii) inversion : Diff
s+k
(M) Diff
s
(M) , is C
k
(iii) the Lie bracket for , X
s+1
(M) is only H
s
i.e. [, ] X
s
(M)
(iv) the exponential map exp : X
s+1
(M) Diff
s
(M) is not locally onto.
This non-classical Lie group structure of Diff
s
(M) leads to the de-
nition of so called ILH - Lie groups (Inverse Limit of Hilbert).
2.4 ILH - Lie Groups and ILH - Lie Algebras
Denition 2.1 A collection of groups G
, G
s
; s s
0
is called an ILH-
Lie group (Inverse Limit of Hilbert) if:
(i) each G
s
is a Hilbert manifold of class C
k(s)
modeled on a Hilbert space
18
E
s
, and k(s) as s
(ii) for each s s
0
there are linear continuous, dense inclusions E
s+1
E
s
and dense inclusions of class C
k(s)
: G
s+1
G
s
(iii) each G
s
is a topological group and G
= lim
s
G
s
is a topological group
with the inverse limit topology
(iv) if (U
s
,
s
, E
s
) is a chart of G
s
then (U
s
G
t
,
s
[
U
s
G
t , E
t
) is a chart for
G
t
for all t s
(v) the multiplication : G
extends to a C
k
map : G
s+k
G
s
G
s
for all s such that k k(s)
(vi) the inversion : G
extends to a C
k
map : G
s+k
G
s
, for all
s such that k k(s)
(vii) the right multiplication R
g
by g G
s
extends to a C
k(s)
map R
g
: G
s
G
s
.
Denition 2.2 A collection of vector spaces g
, g
s
; s s
0
is called an
ILH-Lie algebra if:
(i) each g
s
is a Hilbert space and for each s s
0
there are linear, continuous,
dense inclusions g
s+1
g
s
and g
= lim
s
g
s
is a Frechet space with the
inverse limit topology
(ii) there exist bilinear, continuous, antisymmetric maps [ , ] : g
s+2
g
t+2
g
min(s,t)
, for all s, t s
0
, which satisfy the Jacobi identity on g
min(s,t,r)
for
elements in g
s+4
g
t+4
g
r+4
.
Theorem 2.3 Let G
, G
s
; s s
0
be and ILH-Lie group and for each
s let g
s
:= T
e
G
s
and g
= lim
s
g
s
. Then g
, g
s
; s s
0
is the ILH-Lie
algebra of the ILH Lie group G
, G
s
; s s
0
.
Example : Dieomorphism Groups
Diff
(M), Diff
s
(M) ; s >
1
2
dimM
is an ILH- Lie group with ILH-Lie algebra
X
(M), X
s
(M) ; s >
1
2
dimM
where k(s) = for all s.
19
Remarks:
1) The Nash-Moser implicit function theorem works for ILH !
2) C
k
-dieomorphism groups, 0 k < .
For the group Diff
k
(M) of dieomorphisms of class C
k
we have a similar
situation as in the case Diff
s
(M) of H
s
dieomorphisms discussed above.
Diff
k
(M) is a Banach manifold and one denes the notion of ILB (Inverse
Limit of Banach)- Lie groups and ILB-Lie algebras similar to ILH. Then
Diff
(M), Diff
k
(M) ; k N is an ILB Lie group with ILB-Lie algebra
X
(M), X
k
(M) ; k N.
3) C
-dieomorphisms.
For the group Diff
(P, G) ; (p g) = g
1
(p)g . (2.26)
Gau(P) is a group under pointwise multiplication, hence a subgroup of the
loop group C
(P, g) ; (p g) = Ad
g
1(p). (2.27)
The Lie algebra lie ( (running out of symbols) of the gauge group ( is the
Lie subalgebra of X
(P) ; R
g
X = X, X(p) g, g G, p P (2.28)
with commutator bracket [X
1
, X
2
] = X
1
X
2
X
2
X
1
lie (.
On the other hand, the Lie algebra of C
(Ad P) is C
(AdP). They
are all group isomorphic. There is a natural group isomorphism Gau(P)
( : dened by (p) = p (p), p P which preserves the product
1
2
1
2
. Identifying ( with Gau(P) we can avoid the troubles
with dieomorphism groups and we can extend ( to a Hilbert Lie group (
s
.
So (
s
is actually a Hilbert Lie group in the classical sense, i.e. the group
operations are C
(adP)
are canonically isomorphic. Indeed, for s C
(adP)
by s((p)) := [p, (p)].
On the other hand, for gau(P) dene Z
lie ( by
Z
(p) =
d
dt
[
t=0
R(p, exp t(p)), (= (p)
(p)),
i.e. Z
is
invariant i (p g) = Ad
g
1(p).
To topologize lie ( we complete C
(ad P) in the H
s
-Sobolev norm . If
s >
1
2
dimM then lie (
s
H
s
(ad P) gau
s
(P) are isomorphic Hilbert Lie
algebras.
There is a natural exponential map Exp : gau(P) Gau(P), which is a
local dieomorphism. Let exp : g G be the nite dimensional exponential
map . Then dene
Exp : gau
s
(P) Gau
s
(P) : (Exp )(p) = exp((p)) , gau
s
(P). (2.29)
Or in terms of (, Exp : lie (
s
(
s
: (Exp )(p) = p exp(
p
).
Proposition 2.1 For s >
1
2
dimM
(
s
Gau
s
(P) H
s
(Ad P)
is a smooth Hilbert Lie group with Lie algebra
lie (
s
gau
s
(P) H
s
(ad P)
and smooth exponential map , which is a local dieomorphism
EXP : lie (
s
(
s
: (EXP )(p) = p exp((p)).
3 SUBGROUPS OF DIFFEOMORPHISM
GROUPS AND APPLICATIONS
3.1 Volume Preserving Dieomorphisms and
Fluid Dynamics
In this section we show that the Euler equations of uid dynamics are equiv-
alent to the geodesic equations on the group of volume preserving dieomor-
phisms as well as existence and uniqueness of solutions [13].
22
Let M be a compact manifold, dimM = n and a volume form on M
i.e. is a nondegenerate n-forms with (x) ,= 0, x M. Let
Diff
s
(M) = f Diff
s
(M) ; f
= (3.1)
be the set of volume preserving H
s
dieomorphisms on M . Since the pull
back satises (f g)
= g
(M) Diff
s
(M) is a
subgroup ; is it a LIE SUBGROUP in the sense of ILH-Lie groups ?
Lets rst look for a candidate for the corresponding Lie algebra, i.e.
the tangent space at the identity T
e
Diff
s
(M). Let X
s
(M) with ow
t
Diff
s
(M). Since
t
= it follows that 0 =
d
dt
[
t=0
t
= L
, the Lie
derivative of along . Recall that the divergence div of a vector eld is
dened by L
= (div ), so L
(M) = X
s
(M) = X
s
(M) ; div = 0 (3.2)
the space of divergence free (incompressible) vector elds on M. Since L
[,]
=
[L
, L
] we conclude that X
s
(M) X
s
(M) is a Lie subalgebra.
How do we show that Diff
s
: Diff
s
(M)
n
(M) : F
(f) = f
. (3.3)
Then one shows that for a small enough subspace []
s
of
n
(M) this map is
a C
(M) = F
1
(), i.e.
23
Diff
s
: Diff
s
(M) []
s
, F
(f) := f
(). Then F
is
well dened, i.e. F
(f) []
s
is in the right space. Indeed by the change of
variable theorem we have
_
M
(f
= d for some H
s1
(
n1
(M)).
The second step uses the -lemma to prove that F
is a C
submersion.
The derivative of F
is obtained as
T
f
F
(
f
) = f
(L
) ,
f
T
f
Diff
s
(M) (3.4)
where =
f
f
1
X
s
(M). Hence F
is C
.
For f = e = id
M
: T
e
F
() = L
= di
+ i
(M) i
H
s
(
n1
) (3.5)
is an isomorphism (i
is onto,
hence F
is a submersion and F
1
() = Diff
s
(M), Diff
s
(M) ; s >
1
2
dim M is an ILH Lie group
with X
(M), X
s
(M) ; s >
1
2
dim M its ILH Lie algebra.
Application to Fluid Dynamics: The main application of Diff
s
(M)
is based on the fact that it is the conguration space of incompressible,
homogeneous, ideal uids . Let M be a compact orientable, nite dimensional
Riemannian manifold and let be the Riemannian volume. Then Diff
s
(M)
admits the smooth weak Riemannian metric as follows: For m M let
< , >
m
be the inner product on T
m
M. Let Diff
s
(M) and X, Y
T
Diff
s
Diff
s
(M) is dened as
(X, Y )
=
_
M
< X(m), Y (m) >
(m)
d(m) (3.6)
24
X, Y T
Diff
s
(M) we
can talk about geodesics on Diff
s
(M) and
t
=
M
V
t
i.e.
t
= V
t
then
v
t
:= V
t
1
t
(3.7)
satises the Euler equations:
v
t
+
v
v = p , div v = 0 (3.8)
where
v
is the covariant derivative of the metric on M. The time depen-
dent vector eld v
t
on M represents the Eulerian (or spatial) velocity of
the uid , whereas the solution V
t
of the geodesic spray equation represents
the material velocity of the uid. Incompressibility of the ow implies that
t
Diff
s
(M)
existence of C
geodesics on Diff
s
on the
initial condition v
0
.
The relationship v
t
= V
t
1
t
represents the momentum map of the right
action of the group Diff
s
(M),
the symplectic form being induced naturally by the weak metric from the
canonical symplectic structure on the cotangent bundle T
Diff
s
(M).
We (Eichhorn and Schmid [15]) proved the same theorem for the topo-
logical Euler equation on open (non compact) manifolds (see last chapter).
25
3.2 Canonical Transformations (Symplectomorphisms)
and Plasma Physics
In this section we show that the Maxwell-Vlasov equations of plasma physics
are an innite dimensional Hamiltons system on a space involving the dieo-
morphism subgroup of canonical transformations.
Let M be a compact manifold, dimM = 2n, and be a symplectic struc-
ture M, i.e. is a nondegenerate closed (d = 0) 2-form on M. Consider
Diff
s
(M) = f Diff
s
(M) ; f
= (3.9)
the group of symplectomorphisms , or canonical transformations, on M.
Using similar arguments as for volume preserving dieomorphisms it is shown
that
Theorem 3.4 Diff
(M), Diff
s
(M), Diff
s
(M) ; s > n, with ILH-Lie algebra
X
(M), X
s
(M) = X
s
(M) ; L
= 0 is the
Lie algebra of locally Hamiltonian H
s
- vector elds.
See Ebin-Marsden [13] and Omori [26] for details. Again the charts on
Diff
s
(M) are obtain only implicitly. However, in this case, there is a direct
method due to Weinstein [40] to construct explicit charts using Poincares
generating functions. In Schmid [31], [33] it is shown directly using the so
called -dierentiability for Frechet spaces, that Diff
(M) is a Frechet
manifold and Frechet Lie group with -dierentiable group operations.
3.2.1 Application to Plasma Physics
A plasma is a collection of charged particles of various species (electrons,
protons, ect.) moving in R
3
as a charged uid. For simplicity one assumes
that there is only one species of particles of charge q and mass m, and it is
useful to approximate their positions x R
3
and velocities v by a density
function on phase space which may be a smooth function f(x, v, t). Denote
by E(x, t) and B(x, t) the electric and magnetic elds respectively generated
by the motion of the charged particles. Then the plasma can be described
by the
26
Maxwell-Vlasov equations
(MV )
_
_
f
t
+v
f
x
+
q
m
(E +v B)
f
v
= 0 (Boltzman equation)
B
t
= curlE
E
t
= curlB J
f
, where the current density J
f
= q
_
vf(x, v, t)dv
divE =
f
, where the charge density
f
= q
_
f(x, v, t)dv
divB = 0 .
We think of this system of coupled, non-linear system of evolution equations
as an initial value problem for the variables f, E and B. The following was
shown by Marsden and Weinstein in 1982, see [23] for a summary.
Theorem 3.5 The Maxwell-Vlasov equations are an innite dimensional
Hamiltonian system, i.e. they can be written in the form
F = F, H
for a certain non-canonical Poisson bracket , and some Hamiltonian H.
Let us outline the ideas of the construction of this Hamiltonian structure
, see [23], [33] for details.
3.2.2 The Poisson-Vlasov System
We rst consider the limit case where B = 0 . Then the Maxwell-Vlasov
system reduces to a single equation of the eld variable f(x, v, t) the Poisson-
Vlasov equation
f
t
+v
f
v
q
m
f
x
f
x
= 0 (3.10)
where the scalar potential
f
is given by
f
=
f
. One can show
that f(x, v, t) evolves in time by a canonical transformation
t
of R
6
, i.e.
27
f(x, v, t) =
t
f(x, v, t
0
) where
t
Diff
(R
6
) and is the canonical sym-
plectic form on R
6
. If we identify any Hamiltonian vector eld X
h
(x, v) on
R
6
with its Hamiltonian function h(x, v) : R
6
R, we get a Lie algebra iso-
morphism g = X
(R
6
)
= C
(R
6
) with , [X
h
, X
g
] = X
{h,g}
, where h, g is
the canonical Poisson bracket of functions on R
6
. Moreover, if we identify the
dual of this Lie algebra g
via the L
2
-pairing < h, f >=
_
h(x, v)f(x, v)dxdv
with itself g , we can regard the plasma density as f(x, v) g
= C
(R
6
).
On the dual g
,
G
]) , g
(3.11)
where
F
,
G
and g.
Now a direct computation shows that the Poisson-Vlasov equation (3.10)
is in Lie-Poisson form on g
= C
(R
6
) i.e.
F = F, H with energy
H(f) =
1
2
_
mv
2
f(x, v, t)dxdv +
1
2
_
f
f
dx (3.12)
and , the Lie-Poisson bracket on g
= C
(R
6
) given by
F, G(f) =
_
f
_
F
f
,
G
f
_
dxdv. (3.13)
3.2.3 The Maxwell Equations
As we discussed in section 1.4.2, Maxwells equations
E = curlB ,
B = curlE (3.14)
are canonical Hamiltons equations on T
(R
3
) leads via the
reduction procedure to the two remaining Maxwell equations
divE = , divB = 0. (3.15)
28
3.2.4 The Coupled System: The Maxwell-Vlasov Equations
Combining the Poisson-Vlasov system (3.10) and the Maxwell equations
(3.14),(3.15) we get the Maxwell-Vlasov system. The same symmetry group
C
(R
6
) that leaves Maxwells equations invariant, acts on the coupled phase
space
T
id
Diff
(R
6
) T
/ C
(R
6
) T
F = F, H (3.16)
with the Hamiltonian
H(f, E, B) =
1
2
_
mv
2
f(x, v)dxdv+
1
2
_
[B(x)[
2
dx+
1
2
_
[E(x)[
2
dx (3.17)
which is the total energy of the plasma. The noncanonical Poisson bracket on
the reduced phase space turns out to be the following: For any two functions
F and G of the eld variables (f, E, B) we have
F, G(f, E, B) =
_
f
_
F
f
,
G
f
_
dxdv
+
_ _
F
E
curl
G
B
G
E
curl
F
B
_
dx
+
_ _
F
E
f
v
G
f
G
E
f
v
F
f
_
dxdv
+
_
fB
_
v
F
f
v
G
f
_
dxdv. (3.18)
With this Poisson bracket and Hamiltonian (3.17) the Maxwell-Vlasov equa-
tions are an innite dimensional Hamiltonian system of the form
F = F, H.
Notice that the coupling of the plasma and electromagnetic elds appears in
the Poisson structure rather then in the Hamiltonian, and it is produced by
the action of the innite dimensional gauge group C
(R
6
) on the uncoupled
phase space of matter and elds.
29
In this example where a dynamical system of innitely many degrees
of freedom (the Maxwell-Vlasov equations) is described as a Hamiltonian
system, innite dimensional Lie groups appear as a conguration space as
well as symmetry groups. Similar structures were found for multiuid plasma,
see [23] and references therein.
3.3 Contact Transformations on
T
M
This example of a dieomorphism group plays an important role in our next
chapter, where well discuss the group of invertible Fourier integral operators.
Let M be a compact manifold, dim M = n and T
dp
i
dq
i
(locally) ,
where =
p
i
dq
i
is the canonical 1-form on T
M. We want to study
dieomorphisms : T
M T
= . Such
a dieomorphism is necessarily a lift, i.e. of the form = T
for some
Diff
s
(M) i.e. is just an extended point transformation . To avoid
this trivial situation, we delete the zero section in T
M and we consider
M := T
M0. Then
= and (
x
) = (
x
) for all
> 0,
x
T
x
M i.e.
(
T
M) = Diff
s
(
T
M) ; f
= (3.19)
the group of H
s
- contact transformations on
T
M. Note that
T
M is
not compact , so we cannot apply our previous constructions to establish
Diff
s
(
T
(
T
M), Diff
s
(
T
M).
Therefore, the subgroup Diff
s
(
T
M) of Diff
s
(
T
(
T
M) is algebraically isomor-
phic to the group of contact H
s
- transformations (quantomorphisms) on the
cosphere bundle.
30
3.3.1 The Cosphere Bundle S(T
M)
The multiplicative group R
+
acts smoothly on
T
M by
x
x
,
R
+
,
x
T
x
M . This action is free and proper, hence the orbit space
S(T
M) := (
T
M)/R
+
is a smooth manifold, called the cosphere bundle of
M. Note that S(T
M) = 2n1.
Moreover, the canonical projection :
T
M S(T
M) denes a smooth
principal ber bundle with structure group R
+
. There is no canonical con-
tact structure on S(T
= , but one
can construct a whole family of contact structures by use of global sections
: S(T
M)
T
M , = id.
Given such a global section ,dene
:=
. Then
is an exact
contact 1-form on S(T
M) , i.e.
)
n1
is a volume form on S(T
M).
Note that
:
T
M
R dened by ((
x
)) = f
(
x
)
x
; i.e. f
= f
. (3.20)
3.3.2 Contact Transformations on S(T
M)
Let Diff
s
(
T
M) and : S(T
M)
T
M) S(T
M) dened by = . Write =
1
f
, indeed
for
x
T
M, ((
x
)) = (
1
f
(((
x
))) = (
1
f
(f
(a
x
)
x
) = ((
x
)).
Compute
=
1
f
=
1
f
(f
) =
(f
f
)
(f
f
)
= (f
f
)
= h
where h
:= f
f
.
This means that the pair (, h
i.e.
= h
, where h
: S(T
M) R
+
.
So to each with
= h
= h
we
dene by = ( )/(h )f
. Then
(
T
M) < C
s
(S(T
M),
R),
where C
s
(S(T
M),
R) is regarded as a multiplicative group.
31
Theorem 3.7 (Ratiu-Schmid[30])
1) The group of contact transformations
Con
s
(S(T
M)) := (, h) Diff
s
S(T
M) < C
s
(S(T
M),
R) ;
= h
is a closed H
s
Lie subgroup of the semidirect product Diff
s
S(T
M) <
C
s
(S(T
M),
R).
2) Diff
s
(
T
(S(T
M)).
Remarks: 1) Since the cosphere bundle S(T
M)< C
s
(S(T
M).
2) For two dierent sections and we have an isomorphismCon
s
(S(T
M))
Con
s
(S(T
(
T
M) is indepen-
dent of the choice of the section .
3.3.3 The Corresponding Lie Algebras
The Lie algebra of Diff
s
S(T
M)< C
s
(S(T
M),
R) is the semidirect prod-
uct X
s
(S(T
M))< C
s
(S(T
(S(T
M)) is
con
s
(S(T
M)) = (Y, g) X
s
(S(T
M)) < C
s
(S(T
M), R) ; L
Y
= g
(3.22)
the space of innitesimal contact transformations.
On the other hand, the Lie algebra of Diff
s
(
T
M) is X
s
(S(T
M)) =
Y X
s
(S(T
M)) ; L
Y
= 0. Now L
Y
= 0 Y is a globally Hamiltonian
vector eld, homogeneous of degree 0, i.e. Y = Y
H
and H is homogeneous of
degree 1. H is determined uniquely by H = (Y ) .
Let C
s
hom
(
T
M) = H C
s
(T
M, R) ; H homogeneous degree 1.
C
s
hom
(
T
F
q
i
H
p
i
H
q
i
F
p
i
, and is isomorphic to the Lie algebra X
s
(
T
(
T
M) and Con
s
(S(T
M))
are isomorphic as ILH-Lie groups with isomorphic ILH-Lie algebras X
s
(
T
M)
=
con
s
(S(T
M))
= C
s
hom
(
T
M).
32
Remarks: 1) We will see in the next section that elements of Diff
s
(
T
M)
represent phase functions of Fourier integral operators and elements of
C
s
hom
(
T
(M)
whose Lie algebra consists of all globally Hamiltonian vector elds.
Let (M, ) be a compact symplectic manifold and denote by H
(M)
the space of globally Hamitonian vector elds on M, i.e. X H
(M)
i X = X
H
with H : M R and (X
H
, Y ) = dH Y or equivalently
i
X
H
= dH. We have the relation
X
{F,H}
= [X
H
, X
F
] (3.23)
hence H
(M).
The question is : Is there a corresponding Lie group ?
We have seen that Diff
s
(M), X
(M)] =
H
(M)
0
denote the identity component of Diff
(M).
Theorem 3.9 (Ratiu - Schmid [30])
The commutator subgroup [Diff
(M)
0
, Diff
(M)
0
] is a simple, closed
ILH-Lie subgroup of Diff
(M)
0
with ILH Lie algebra
H
(M) = [X
(M), X
(M)].
Proof: (idea, Banyaga [5], Calabi [9]) Consider the map
S : Diff
s+1
(M)
0
H
1
(M, R) , S(h) = [A(h
t
)]
where h
t
is a symplectic homotopy fromh to id
M
with the locally Hamiltonian
vector eld X
t
=
dht
dt
h
1
t
. Dene A(h
t
) =
_
1
0
i(X
t
)dt which is a closed
H
s+1
-one form dening a cocycle in H
1
(M, R).
33
One shows the following:
S is a group homomorphism and KerS is perfect. H
1
(M, R) is abelian,
therefore KerS = [Diff
(M)
0
, Diff
(M)
0
].
S is a C
(M)
0
.
3.5 The Group of Quantomorphisms
Let (M, ) be a compact exact contact manifold, i.e. M is smooth dimM =
(2n + 1) and is a 1-form such that (d)
n
is a volume element on M.
Consider
Diff
s
(M) = Diff
s
(M) ;
= (3.24)
called the quantomorphism group of M.
The characteristic bundle of d is given by R
d
= v TM ; i
v
d = 0.
It is integrable and one-dimensional. The Reeb vector eld E of R
d
is the
unique section of the line bundle R
d
dened by i
E
= 1 and i
E
d = 0. In
local coordinates (x
1
, . . . , x
n
, y
1
, . . . y
n
, t), we have =
n
i=1
y
i
dx
i
+ dt and
E =
t
. Let R
= 2n and TM = R
d
R
, hence X
s
(M) = X
s
(R
d
) X
s
(R
).
The leaves of the foliation T dened by the line bundle R
d
(circle ac-
tion) are the integral curves of E . Then the quotient manifold N = M/T
is a smooth manifold which carries a symplectic structure such that the
projection : M N satises
and
is its curvature 2-form. Thus M is the quantizing manifold of N whose
automorphism group Diff
s
(M) = Diff
s
(M) ;
= is called the
quantomorphism group of (M, ).
Let /
s
(N) = Diff
s
(N) ; H
= H
(M)
P
/
s
(N) 0
and the following sequence of Lie algebras is a exact
0 R
j
X
s
(M)
p
H
s
(N) 0 .
34
We showed that this is an exact sequence of ILH-Lie groups with corre-
sponding ILH-Lie algebras i.e.
Theorem 3.11 (Ratiu- Schmid [30]) The quantomorphism group Diff
(M)
is an ILH principal circle bundle over the ILH-Lie group /
(M) = X X
(M) ; L
X
= 0 the innitesimal quanto-
morphisms.
3.6 The Group of Gauge Transformations and
Quantum Field Theory
The dieomorphism subgroups that arise in gauge theories as gauge groups
behave nicely because they are isomorphic to subgroups of loop groups as
discussed in section 2.5.
Consider a principal G-bundle : P M and the group ( of gauge
transformations
( = Diff
= C
(P, G) ; (p g) = g
1
(p)g = Gau(P) (3.25)
which is a smooth Hilbert Lie group with smooth group operations (see sec-
tion 2.5).
We only sketch here what role this innite dimensional gauge group (
plays in these quantum eld theories. A good reference for this topics is P.
Deligne et al. [12].
Let / denote the space of connection 1-forms on P (vector potentials) .
Each A / denes a covariant dierential D
A
and a curvature 2-form (eld
strength) F
A
= D
A
A = dA +
1
2
[A, A]. They are locally given by A = A
dx
and F =
1
2
F
dx
dx
where F
+ [A
, A
].
The gauge group ( acts on / via pull-back (, A /, A =
(
1
)
1
and the action on the
eld is F
A
:= F
A
= F
A
1
.
The action functional (Yang-Mills functional) is S(A) = |F
A
|
2
, locally
given by |F
A
|
2
=
1
2
_
M
tr(F
|F
A
|
2
, M compact.
The Feynman path integral quantizes the action and we get the probability
amplitude
W(f) =
_
A/G
e
S(A)
f(A)T(A)
for any gauge invariant functional f(A).
4 BRST SYMMETRY
In 1976 C.Bechi, A.Rouet and R.Stora [6] (and independently I.V. Tyutin
in 1975) discovered that in gauge eld theories the eective Lagrangian,
which is no longer gauge invariant, is still invariant under a new class of
transformations, now called BRST transformations s , given by
s A = d + [A, ] , s =
1
2
[, ] (4.1)
where A is the potential eld (connection one form) and is the ghost eld.
An important property of the BRST charge s is its nilpotency s
2
= 0. We give
an interpretation of these BRST transformation in terms of the Lie algebra
cohomology of the Lie group of gauge transformations.
36
4.1 Quantum Chromo Dynamics (QCD) and
Quantum Electro Dynamics (QED)
In classical eld theory one considers a Lagrangian L(
i
,
i
) of the elds
i
: R
n
R, i = 1, ...k,
=
x
and the corresponding action functional
S =
_
L(
i
,
i
)d
n
x. The variational principle S = 0 then leads to the
Euler-Lagrange equations of motion
L
L
(
i
)
= 0 . (4.2)
In QED and QCE the Lagrangian is more complicated, of the form
L(A, , ) =
1
4g
2
Tr F
+ieA
) +m] +(D
A
)
(D
A
) m
2
(4.3)
Where A
+ [A
, A
i
0
_
where
i
are the Pauli matrices (canonical
basis of su(2) and
=
o
is the Pauli adjoint with
o
=
_
0 1
1 0
_
,
m is the electron mass, e the electron charge and g a coupling constant.
4.1.1 The Equations of Motion
The variational principle of the Lagrangian (4.3) with respect to the elds
A, and gives the corresponding Euler Lagrange equations of motion.
They describe for instance the motion of an electron (x) (fermion, spinor)
in an electromagnetic eld F, interacting with a bosonic eld . We get from
the variational principle
S
A
= 0
= e
= 0 i(,
A
m) = 0, which are Diracs equations, where ,
A
=
+
ieA
) =
A
. In the free case i.e. when A = 0 we get i(, m) = 0, the
classical Dirac equation.
4.1.2 Global Formulation
Consider a principal G-bundle : P M , with M a compact, orientable
Riemannian manifold (e.g. M = S
4
, T
4
) and G a compact nonabelian gauge
group with Lie algebra g. Let / be the innite dimensional ane space of
connection 1-forms on P. So each A / is a g-valued, equivariant 1-form
on P (also called vector potential) and denes the covariant derivative of any
eld by D
A
= d+
1
2
[A, ] . The curvature 2-form F
A
(or eld strength)
is a g valued 2-form and is dened as F
A
= D
A
A = dA +
1
2
[A, A] .
In pure Yang-Mills theory the action functional is given by
S(A) =
1
2
|F
A
|
2
=
1
2
_
M
Tr(F
), (4.4)
and the Yang-Mills equations become globally
d F
A
= 0. (4.5)
With added fermionic eld interaction the action becomes
S(A, ) =
1
2
|F
A
|
2
+ <,
A
, > (4.6)
where is a section of the spin bundle opin
(M) and ,
A
: opin
(M)
opin
(M) by
A = (
1
)
A , and = (
1
)
(M), (. (4.8)
4.2.2 Chiral Symmetry
The chiral symmetry is the symmetry that leads to anomalies and the BRST
invariance. In QCD the chiral symmetry of the Fermi eld is given by
e
i
5
(4.9)
where is a constant and
5
= i
o
3
. The classical Noether current of
this symmetry is given by
J
5
(4.10)
which is conserved, i.e.
= 0 .
This conservation law breaks down after quantization; one gets
= 2im
5
g
2
8
2
TrF
,= 0. (4.11)
This value is called the chiral anomaly.
4.3 Quanization
The quantization is given by the Feynman path integral:
_
A/G
_
Spin
e
iS(A,)
T(A, )TAT =< T(A, ) > (4.12)
which computes the expectation value < T(A, ) > of the function T(A, ).
This in an integral over two innite dimensional spaces , the gauge orbit
space //( and the fermionic Berezin integral over the spin-space opin
(M).
These integrals are mathematically not dened but physicist compute them
by gauge xing, i.e. xing a section : //( /, (e.g. (A) =
= 0,
the Lorentz gauge) and then integrate over the section . Such a section
does not exist globally , but only locally (Gribov ambiguity!). The eect of
39
such a gauge xing is that one gets extra terms in the Lagrangian (gauge
xing terms) and one has to introduce new elds, so called ghost elds via
the Faddeev-Popov procedure. The such obtained eective Lagrangian is no
longer gauge invariant. This eective Lagrangian has the form in QCD:
L
eff
(A, , ) =
1
2
Tr(F
) kinetic energy
+
1
2
Tr(
)
2
gauge xing term
g
A
ghost term
+. . . . . . interaction terms
(4.13)
We can write this globally as
L
eff
=
1
2
|F
A
|
2
+
1
2
|(A)|
2
+ / +... (4.14)
where /=
loc
: (
q,p
(lie (,
loc
) (
q+1,p
(lie (,
loc
),
2
loc
= 0.
Then with s :=
(1)
p+1
q+1
loc
we have s
2
= 0 and the following:
1) for q = 0, p = 1, A / (
0,1
, then sA = d + [A, ]
2) for q = 1, p = 0, (
1,0
, then s =
1
2
[, ] the Maurer Cartan form
3) the chiral anomaly (given by equ. (4.11) ) is represented as coho-
mology class of this complex [] H
1,0
BRST
(lie (,
loc
).
4.4.1 The Chevalley-Eilenberg Cohomology
We are now going to explain the previous theorem, in particular the general
denition of the Chevalley-Eilenberg [11] complex and the corresponding
cohomology.
Let G be a Lie group with Lie algebra g and let be a representation of g
on the vector space W. Denote by (
q
(g, W) the space of W-valued q-cochains
on g and dene the coboundary operator : (
q
(g, W) (
q+1
(g, W) by
(
o
, ,
q
) =
q
i=0
(1)
i
(
i
)(
o
, ,
i
, ,
q
)
+
i<j
(1)
i+j
((
i
)
j
, ,
i
, ,
j
, ,
q
)
(4.16)
We have
2
= 0 and dene the Lie algebra cohomology of g with respect
to (, W) as H
() of the
associated vector bundle . Consider the space of exterior forms on M
,
i.e (M
()) =
p,q
p,q
(M
(). A form
p,q
(M
i,j
i
1
ipj
1
jq
dx
i
1
dx
ip
u
j
1
u
jq
(4.18)
where the
i
1
ipj
1
jq
are local (0, 0) forms, the dx
i
p
p
loc
(P, g) the space of Lie algebra valued ad-equivariant local p-forms on
P with the representation of g on W induced by the adjoint representation.
We have the double complex (
q,p
(
q
(lie (,
p
loc
(P, g)) with coboundary
operator
loc
= + d. The induced cohomology is the BRST-cohomology
H
BRST
(lie (,
loc
).
4.4.4 Proof of the Theorem
We sketch the proof of the above theorem:
1) For q = 0, p = 1 we have the following identications:
(
q,p
= (
0
(lie (,
1
(P, g))
=
1
(P, g) Lie algebra valued 1-forms and /
1
(P, g). Then s = and for A /, X lie ( we get:
s(A)(X) = A(X) = (X) A = ad(X) A = D
A
X = dX +
1
2
[A, X].
Also d(X) = d((X)) = dX, and [X, ](A) = [A, (X)] = [A, X] hence
sA(X) = (d)(X) +
1
2
[A, ](X).
42
2) For q = 1, p = 0 we have the following identications:
(
q,p
= (
1
(lie (,
0
(P, g)) = Hom(lie (, lie (). For = id : lie (
lie (; (A) : T g. Then s =
1
2
and for X
0
, X
1
lie ( we get:
s(X
0
, X
1
) =
1
2
(ad(X
0
)(X
1
)ad(X
1
)(X
0
))[X
0
, X
1
]
=
1
2
(L
X
0
X
1
L
X
1
X
0
[X
0
, X
1
])
=
1
2
[X
0
, X
1
] =
1
2
[(X
0
), (X
1
)]
=
1
2
[, ](X
0
, X
1
).
4.5 Anomalies
The Noether current induced by the chiral symmetry (after quantization) for
the free case ( = 0) i.e. for pure Yang-Mills becomes
=
g
2
8
2
TrF
=
1
4
Tr
(A
+
2
3
A
)
= ,= 0 anomaly , see (4.11)
(4.19)
Note the similarity with the Chern-Simon Lagrangian
L(A) =
_
M
Tr(AdA +
2
3
A
3
). (4.20)
We are going to derive a representation of the chiral anomaly in the
BRST cohomology, i.e. [] H
1,0
BRST
(lie (,
loc
).
The question is, if s = 0, does there exist a local functional F(A), such
that = s(F(A)) ? i.e. is BRST s-exact ? The answer in general is NO,
i.e. represents a nontrivial cohomology class. This class is given by the
Chern-Weil homotopy.
43
4.5.1 The Chern-Weil Homotopy
Let
A = A + (
0,1
(
1,0
and
F s
A +
A
2
= F
A
. For t [0, 1] let
F
t
= t
F + (t
2
t)
A
2
and dene the Chern-Simons form
2q1
q
_
1
0
Tr(
A
F
q1
t
) dt (4.21)
we get
s
2q1
= Tr
F
q
(4.22)
We write
2q1
as sum of homogeneous terms in ghost number (upper index)
and degree (lower index)
2q1
=
0
2q1
+
1
2q2
+
2
2q3
+ +
2q1
0
. Let
(X, A) =
_
M
1
2q2
(X).
Theorem 4.2 (Schmid [36]) The form (X, A) =
_
M
_
1
0
F
q1
t
(X)dt satis-
es the Wess-Zumino consistency condition (s)(X
0
, X
1
, A) = 0 and repre-
sents the chiral anomaly [] H
1,0
BRST
(lie (,
loc
).
We have an explicit form of the anomaly in (2q 2) dimensions:
1
2q2
= q(q 1)
_
1
0
(1 t)Tr(
loc
(
A
F
q2
t
)) dt. (4.23)
So for q = 2 the non-Abelian anomaly in 2-dim. becomes
1
2
= Tr(
loc
A),
and for q = 3 the non-Abelian anomaly in 4-dim. becomes
1
4
= Tr(
loc
(
A
loc
A +
1
2
A
3
)). (4.24)
4.6 The Wess-Zumino Consistency Condition
This is a problem in local cohomology. We consider the bicomplex (
loc
=
(
q,p
,
q,pN
with total dierential =
loc
+ (1)
p
d where
loc
: (
q,p
(lie (,
loc
) (
q+1,p
(lie (,
loc
) and d : (
q,p
(lie (,
loc
)
(
q,p+1
(lie (,
loc
). We have
2
=
loc
d + d
loc
=
2
loc
= d
2
= 0.
The Wess-Zumino consistency condition for (
loc
means that there
exists an (
loc
such that
loc
+ d = 0 . (WZ)
44
Any solution of (WZ) of the form =
loc
+ d, , (
loc
is trivial, i.e.
loc
= 0. The consistency condition (WZ) produces the so called descent
equations. If
loc
+ d = 0 taking
loc
of (WZ) we get
loc
2
+
loc
d = 0
hence
loc
d = 0 . The Poincare lemma implies there exists a local form
such that
loc
+ d = 0. By denition
loc
[] = []. If is trivial, i.e.
=
loc
+ d then
loc
d = d, hence =
loc
+ d, i.e. [] = 0.
We get the descent equations
loc
+ d
1
= 0
loc
1
+ d
2
= 0
.
.
.
loc
k1
+ d
k
= 0
(4.25)
where k is the smallest integer such that [] H
k
loc
(lie () with
loc
= 0.
4.6.1 g-Symplectic Structures
Denition 4.1 A g-symplectic structure on P is a g - form
2
(P, g)
which is closed and nondegenerate, i.e. d = 0 and for each p P the map
(p) : T
p
P T
p
P g is bilinear and nondegenerate. We have the induced
linear injective map (p)
#
: T
p
P L(T
p
P, g) ; (p)
#
(v) w = (p)(v, w).
A vector eld X on P is called g-Hamiltonian if there exists a g-function
f : P g such that df = i
X
.
A g - vector eld X is locally g -Hamiltonian i its ow
t
is g - sym-
plectic, i.e.
t
= .
We have the
Lemma 4.1 Poincare lemma: For any
p
(R
n
, g) with d = 0 there
exists locally a
p1
(R
n
, g) such that = d and
(x) =
_
1
0
i
x
(tx)dt.
Theorem 4.3 (Schmid [37]) If G is semi simple, then every G - orbit O
p
of the right action of G on P is a g - symplectic manifold; induced by the
Maurer Cartan form on G.
45
4.6.2 The Canonical Momentum Map on O
p
Proposition 4.1 (Schmid [37]) For every g the fundamental vector eld
P
on O
p
dened by
P
(q) =
d
dt|
t=0
R
exp t
(q)
is locally g -Hamiltonian.
Corollary 4.1 For every g there exists a g- function H : O
p
g such
that
P
= X
H
i.e dH = i
p
. Explicitly
H(x) =
1
2
[x, x ].
Proposition 4.2 (Schmid[37]) The g -momentum map J : O
p
L(g, g) of
the right action of G on O
p
dened by J(q), ) = H(q), q O
p
, g. Is
given by
J(q) = ad
TR
q
where = R
p
X
t
(g), q = p g .
4.6.3 A Solution of Consistency Condition
The innite dimensional analogue of the previous construction leads to the
following: Consider a principal ( bundle (T, , /), where T =
(P, g) with
the adjoint ( action and / = T/(. For A
A
on the orbit O
A
induced from the Maurer - Cartan form on ( becomes
a map
A
: O
A
1
(P, g) (
0,1
loc
and the momentum map J : O
A
L(lie (, lie () = (
1,0
loc
.
Theorem 4.4 (Schmid [37]) The momentum map J satises the consistency
condition for the canonical 1-form (Maurer-Cartan)
A
of (
loc
A
+ dJ = 0.
5 LIE GROUPS OF PSEUDODIFFEREN-
TIAL AND FOURIER INTEGRAL
OPERATORS
Pseudodierential operators and Fourier integral operators are used in physics
to construct solutions of partial dierential equations, e.g. the fundamental
46
solution of the heat equation is a pseudodierential operator. They are also
used to prove local formulas for index theorems. Geometric properties are
used to describe Fourier integral operators as a symmetry group for the KdV
equation as Hamiltonian system. These will be discussed in section 6. These
operators have very nice geometric properties which we will discuss here.
Fourier integral operators (FIO for short) generalize pseudodierential
operators (DO for short) which themselves generalize dierential operators
(DO for short). So as sets we have the inclusions
FIO DO DO .
5.1 Pseudodierential Operators DO
Consider a dierential operator P on R
n
of order m with smooth coef-
cients a
Pu(x) =
||m
a
(x)D
x
u(x) , u C
(). (5.1)
We associate to the operator P the polynomial
p(x, )
||m
a
(x)
(5.2)
called the symbol of P . Using the Fourier transform u() of u(x) we have
x
u() =
u() and D
x
u(x) = (2)
n
_
e
ix
x
D
p(x, )[ C
,
(K)(1 +[[)
m||+||
. (5.4)
47
We restrict ourselves to classical symbols, i.e. those who have an asymp-
totic expansion of the form
p(x, )
j=0
p
mj
(x, ) , m = order of p (5.5)
where each p
mj
(x, ) C
(
R
n
) is homogeneous of degree m j, in
, i.e. p
mj
(x, ) =
mj
p
mj
(x, ), > 0. Then P is a classical pseudod-
ierential operatorof order m if P is of the form
Pu(x) = (2)
n
_ _
e
i(xy)
p(x, )u(y) dyd (5.6)
with p(x, ) a classical symbol of the form (5.5). The principal symbol of
P is the leading term p
m
(x, ).
Denote by DO
m
the space of classical pseudodierential operators of
order m and let DO =
m
DO
m
be the space of all pseudodierential
operators of all order. DO is an innite dimensional graded Lie algebra with
commutator bracket satisfying the following: if P DO
m
and Q DO
n
then [P, Q] = PQQP DO
m+n1
. Note that DO
1
the space of pseu-
dodierential operators of order 1 is an innite dimensional Lie subalgebra
of DO .
It is a natural question to ask whether there exist corresponding Lie
groups having DO
1
and DO as their Lie algebras respectively ? The an-
swer is YES ! The Lie group that has DO
1
as its Lie algebra is the group of
invertible Fourier integral operators of order zero, denoted by (FIO
0
)
; and
the group for DO is the group of all invertible Fourier integral operators ,
denoted by FIO
R
n
(x
0
,
0
)
T
= (5.7)
where =
n
i=1
dp
i
dq
i
is the canonical symplectic form on T
R
n
. The
function S is called the generating function of . Vice versa, every canon-
ical transformation has a locally generating function S. Note that S is
homogeneous of degree one in if and only if is homogeneous in .
Example: Let S(x, ) = x . Then =
S
x
= , y =
S
= x hence
= id.
5.2.2 Fourier Integral Operators
Let S(x, ) be a generating function and a(x, ) a classical symbol order m.
Dene a classical Fourier integral operator A of order m by
Au(x) :=
_
e
iS(x,)
a(x, ) u() d = (2)
n
_ _
e
i(S(x,)y)
a(x, )u(y) dyd
(5.8)
More generally , a Fourier integral operator A of order m is dened by
Au(x) = (2)
n
_ _
e
i(x,y,)
a(x, )u(y) dyd (5.9)
where (x, y, ) is nondegenerate phase function (homogeneous +1) and
the symbol a(x, ) S
m
,
.
Notice if S(x, ) = x or general (x, y, ) = (x y) then the
operator A is a pseudodierential operator dened by (5.6) and we have
FIO DO DO.
These Fourier integral operators are singular operators but make sense as
oscillatory integrals and they have nice properties in the following sense:
Property 1) Fourier integral operators are invariant under dieomor-
phisms . So they can be dened on manifolds M as bounded linear op-
erators A : C
(M) C
(M) T
(M). Any P DO
m
extends as bounded linear operator
P : H
s
c
(M) H
sm
c
(M).
Property 2) Pseudodierential operators have properties which are close
to dierential operators (DO):
P DO P is local, i.e. supp Pu supp u.
P DO P is pseudolocal, i.e. sing supp Pu sing supp u.
Property 3) Pseudodierential operators preserves the wave front sets
WF i.e. WF(Pu) WF(u) where WF(u)
T
M,
M
WF(u) = sing supp u.
Fourier integral operators generalize pseudodierential operators in the
sense that they move the wave front sets WF by a canonical relation , i.e.
if A FIO then WF(Au) WF(u) where the canonical relation
T
M
T
= 0. (5.10)
Remarks:
a) If = = ((x, ), (x, )) the diagonal then A DO.
b) Any A FIO is determined by its symbol a(x, ) and canonical relation
.
c) The principal symbol is globally dened a
m
(x, ) :
T
M R.
Property 4) Fourier integral operators are closed under multiplication:
Assume :
T
M
T
M
T
M then
FIO
m
(id) = DO
m
, i.e. Fourier integral operators associated with the iden-
tity are pseudodierential operators.
Example: Let f : M M be a dieomorphism. Then
f
u(x) = (2)
n
_ _
e
i(f(x)y)
u(y)dy d
50
denes a FIO A = f
: C
(M) C
f :
T
M
T
M.
Notation: Denote by FIO
, DO
, (FIO
m
)
, (DO
m
)
the invert-
ible elements in the corresponding spaces, and Diff
(
T
(M) where =
p
i
dq
i
is the canonical
1-form. Note that FIO
, DO
, (FIO
0
)
, (DO
0
)
= we get
= hence Diff
s
(
T
Diff
s
(
T
M) , p(A) =
where graph() = the canonical relation of A . The kernel of p is p
1
(e) =
DO
, e = id
M
. Both DO
and FIO
j
FIO
p
Diff
s
(
T
M) e (5.11)
We want to make this into an exact sequence of LIE GROUPS.
Notice that the zero order operators (DO
0
)
and (FIO
0
)
are groups
and form exact sequence
I (DO
0
)
j
(FIO
0
)
p
Diff
s
(
T
M) e (5.12)
First we are going to give ILH-Lie group structures to this sequence (5.12)
of zero order operators , then well move these structures by a xed elliptic
operator T to any order m, e.g. T = (1 + )
m/2
: FIO
0
FIO
m
.
For the parameter spaces we look at the corresponding Lie algebras of
I (DO
0
)
j
(FIO
0
)
p
Diff
s
(
T
M) e (5.13)
51
The Lie algebras are
I DO
0
j
DO
1
+1
(
T
M) e (5.14)
where (P) is the principal symbol (homog. +1) of P DO
1
. The space
of homogeneous functions C
+1
(
T
+1
(
T
M)
= X
(
T
M) = X ; L
X
= 0, as
discussed in section 3.4.
Remark: Clearly FIO ,= exp(DO) since Diff
s
(
T
M) ,= exp(X
s
(
T
M),
i.e. we cannot obtain a chart at the identity in FIO
by exponentiating the
Lie algebra DO.
The idea to construct a manifold- and a Lie group structure on FIO
is
to construct an innite dimensional principal ber bundle with
base space = Diff
s
(
T
M)
total space = (FIO
0
)
ber = p
1
() = (FIO
0
())
= (DO
0
)
.
We outline this construction in 7 steps (for details see Adams, Ratiu and
Schmid [1],[2],[3]:
Step 1: We show that Diff
s
(
T
M) = lim
Diff
s
(
T
M) is an ILH-Lie group.
Step 2: We show that (DO
0
)
= lim
(DO
s
0
)
is an ILH-Lie group.
Step 3: We piece 1 & 2 together via a local section : U Diff
s
(
T
M)
(FIO
0
)
. Then (FIO
0
)
. This
gives a chart at the identity I (FIO
0
)
.
Step 4: We move this chart around by the group the structure of Diff
s
(
T
M)
(FIO
0
)
is a topological group.
Step 5: We check that the chart transitions are smooth (FIO
0
)
is a
smooth manifold.
Step 6: We check that multiplication is smooth (FIO
0
)
is an ILH- Lie
group.
Step 7: We identify (1 )
m/2
: (FIO
0
)
(FIO
m
)
FIO
is an ILH-
Lie group.
Lets go through some more details.
52
5.2.4 Step 1: Diff
s
(
T
(
T
M) = lim
s
Diff
s
(
T
(
T
(ST
M) = (, h) Diff
s
(ST
M) < C
s
(ST
M) ;
S
= h
S
(S(T
M)) = Y X
s
(S(T
M)) ; L
Y
= 0 isomor-
phic to C
s
+1
(
T
M) = H C
s
(T
M, R) ; H homog.deg 1.
5.2.5 Step 2: (DO
0
)
j=0
p
j
(x, ), so we
would have to control an innite number of functions and their derivatives,
and an innite product of Hilbert spaces is no longer a Hilbert space. So what
we do is we cut the symbol at the term p
k
for some xed k < . In terms of
the operators, we look at the quotient spaces DO
m,k
:= DO
m
/DO
k1
and similarly FIO
m,k
() := FIO
m
()/FIO
k1
() , FIO
m,k
=
FIO
m,k
(),
where FIO
m
() = A FIO
m
; p(A) = .
Composition is still well dened in DO
0,k
and FIO
0,k
and we denote by
(DO
0,k
)
and (FIO
0,k
)
j
(FIO
0,k
)
p
Diff
s
(
T
M) id (5.15)
For P DO
m,k
with symbol p(x, ) = p
m
(x, ) + + p
k
(x, ) we dene
the norm by
|P|
2
m+k,s
= | p
m
|
2
s+k+m
+| p
m1
|
2
s+k+m1
+ +| p
k
|
2
s
(5.16)
where p
mj
is the restriction of p
mj
to the cosphere bundle S(T
M) and
| p
mj
|
2
s+k+mj
is the H
s+k+mj
-Sobolev norm on S(T
M). Let DO
s
m,k
de-
note the completion of DO
m,k
with respect to this norm and for m = 0
denote by (DO
s
0,k
)
is a
smooth (C
= lim
s
(DO
s
0,k
)
(
T
M) and DO
s
0
by a local section
: | Diff
s
(
T
M) (FIO
0
)
(5.17)
where | is a neighborhood of id Diff
s
(
T
a local
product structure
(|)
= | (DO
0
)
.
Problem: Fourier integral operators are only locally dened but we need
a global writing of Fourier integral operators , i.e. a global phase function
for Fourier integral operators close to I. This is done by constructing an
explicit chart about id of Diff
s
(
T
M M R
H
(
x
, y) :=
x
(exp
1
x
(y)) +H(
x
) (5.18)
(where exp is dened by a Riemannian metric on M). Then there exists an
Diff
s
(
T
M)
onto a neighborhood |(e) Diff
s
(
T
M).
For the inverse let Diff
s
(
T
M) be close to e (
: T
M M the
projection) and dene H :
T
M R by
H(
x
) =
x
exp
1
x
(
1
(
x
)). (5.19)
Then
H
dened by (5.18) is a global phase function for graph().
54
Now we dene a local section of the sequence (5.12)
: | Diff
s
(
T
M) (FIO
0
)
(5.20)
as follows: let | Diff
s
(
T
x
M
_
B
(x)
(x, y)e
i
H
(x,y)
u(y)[ det exp
x
[dyd (5.21)
where B
. In particular () (FIO
0,k
)
the local
product structure
p
1
(|) | (DO
0
)
. (5.22)
Now we dene the topology around the identity in (FIO
0,k
)
by the bijec-
tion : p
1
(U
2t
) U
2t
(DO
2(tk)
0,k
)
1
(, P) = P (), where U
2t
= | Diff
2t
.
This denes a local chart at the identity I (FIO
0,k
)
.
5.2.7 Step 4: (FIO
0
)
as Topological Group
We move this chart (constructed in step 3) around by the group structure
of Diff
s
(
T
a topological group.
To dene the topology on (FIO
0,k
)
t
(FIO
t
0,k
)
with
the inverse limit topology is a topological group as well.
55
To prove this, we have to show that the map (A, B) AB
1
is continuous
for any A, B (FIO
t
0,k
)
as Smooth Manifold
Overlap conditions in local charts give conditions on to make (FIO
0
)
into
a smooth manifold. To prove that the transition maps between local charts
are smooth it amounts to show that the following map is dierentiable
(U
2t
U
2t
) (DO
2(tk)
0,k
)
(DO
2(tk)
0,k
)
(, P) P(
1
)AB
1
(
1
)
1
(5.24)
for any A, B (FIO
t
0,k
)
is a smooth manifold of
class C
t
.
5.2.9 Step 6: (FIO
0
)
(FIO
0
)
(FIO
0
)
, (A, B) = A B (5.25)
: (FIO
0
)
(FIO
0
)
, (A) = A
1
. (5.26)
To show that the group multiplication in (FIO
t
0,k
)
is smooth it amounts to
show that the following map is dierentiable
(U
2(t+s)
) (
2(t+s+k)
0,k
)
(U
2(t+s)
) (
2(t+s+k)
0,k
)
(U
2(t+s)
(r ) (
2(t+s+k)
0,k
)
((
1
, P
1
), (
2
, P
2
))
(
1
2
, P
1
(
1
1
)AP
2
(
2
1
)A
1
(
1
1
)
1
) (5.27)
for any A (FIO
t+s
0,k
)
, B (FIO
t
0,k
)
= lim
s
(FIO
s
0
)
as Lie Group
To obtain a Lie group structure on FIO
(FIO
m
)
. Multi-
plication has the same smoothness properties as before between the appro-
priate spaces, which makes FIO
(M) is and innite dimensional principal ber bundle over the base
manifold Diff
s
(
T
M) of contact transformations of
T
= lim
t
(FIO
t
0,k
)
(FIO
t
0,k
)
(FIO
t
0,k
)
: (A, B) =
AB is C
p
dierentiable,
the inversion : (FIO
t+p
0,k
)
(FIO
t
0,k
)
: (A) = A
1
is C
p
dier-
entiable
right multiplications R
A
: (FIO
t
0,k
)
(FIO
t
0,k
)
: R
A
(B) = BA are
C
t
dierentiable , for any A FIO
t
0,k
the left multiplications L
A
: (FIO
t
0,k
)
(FIO
t
0,k
)
: L
A
(B) = AB are
C
0
(continuous) , for any A FIO
t
0,k
.
57
6 Diff(M) AND FIO FOR NON-COMPACT
MANIFOLDS. APPLICATION TO FLUID
DYNAMICS AND QUANTIZATION
The classical theory of dieomorphism groups and the groups of Fourier
integral operators was developed in the 1970s and 80s. This was all done
for compact manifolds, mainly for technical reasons. Many attempts to deal
with noncompact manifolds failed, until in the 1990s after J. Eichhorn pub-
lished his papers on the manifold structures of maps between open manifolds
[14] and we joined forces to overcome the technical diculties one encoun-
ters dealing with noncompact manifolds. Two typical quotations about the
non-compact case:
1) Topology & Analysis: Atiyah-Singer Index formula & Gauge theoretic
Physics (Boos-Bleecker [7], pg182):
In what follows, the manifolds M is closed i.e. compact, without boundary.
We make this convention in part for convenience ( in order to make some
proofs go easier) but also because otherwise some of the following theorems
would be meaningless or false.
2) J. Eichhorn: There is exactly one thing that work in the non-compact
case: NOTHING
Example of whats going wrong: Let M , N be compact manifolds,
then f : M N is of Sobolev class H
s
the local representatives f
i
j
: U
i
R
m
V
j
R
n
are of class H
s
, where M =
(U
i
,
i
), N =
(V
j
,
j
), f
i
j
:=
j
f
1
i
. These covers are nite if M, N are compact. This denition is
invariant s >
n
2
+ 1. In the compact case we can dene the distance by
d
s
(f, g) := (
i,j
|f
i
j
g
i
j
|
2
s
)
1
2
.
These denitions are meaningless if M and N are open !
6.1 Bounded Geometry
The idea of overcoming compactness is to use bounded geometry, which
means that well have control over the metric and its derivatives and the
58
maps are adapted to the bounded geometry, i.e. we have control over the
mappings and their derivatives by the metric as well.
Denition 6.1 A Riemannian manifold (M
n
, g) has bounded geometry
of order k, 0 k , if it has a positive injectivity radius and the
curvature and all is derivatives up to order k are uniformly bounded; i.e the
following two conditions are satised:
(I) : r
inj
(M) = inf
xM
r
inj
(x) > 0 (6.1)
(B
k
) : [
i
R[ C
i
, 0 i k. (6.2)
These conditions can be expressed as follows:
(I) The exponential map exp
x
: T
x
M M is a dieomorphism from an open
ball B
x
(0, r) T
x
M of radius r around 0 in T
x
M onto an open neighborhood
U
x,r
M of x in M. Let r
x
:= sup(r) be the biggest radius in T
x
M such
that exp
x
is a dieomorphism. Then the injectivity radius is dened by
r
inj
:= inf
xM
r
x
, i.e r
inj
is the smallest distance from x where geodesics
intersect. Hence
(I) there exists a ball around 0 in R
n
which is domain of normal (geodesic)
coordinates for all x M.
(B
k
) there exists a constant d
k
(independent of x M) such that
|g
ij
|
C
k d
k
in any normal coordinate system
[D
g
ij
[ c
k
in any normal coordinate system.
Examples of manifolds with bounded geometry are compact manifolds,
Lie groups, homogeneous spaces, covering spaces of Riemannian manifolds,
leaves of foliations of compact manifolds.
Fact: There is no topological obstruction for the existence of a complete
Riemannian metric with bounded geometry of any order.
6.2 Bounded Maps C
,r
(M, N)
We consider maps which are adapted to the bounded geometry of the mani-
folds. Let (M, g) and (N, h) be open, complete Riemannian manifolds satis-
fying (I) and (B
k
) and f C
= Tf
59
is a section of T
M f
TN. We endow f
h
. Then
g
and f
h
induce connections in all tensor bun-
dles T
q
s
(M) f
T
u
v
(N). Therefore the m
th
derivative
m
df is well dened.
Assume r k. We denote by C
,r
(M, N) the set of all f C
(M, N)
satisfying
[df[
r
:=
r1
i=0
sup
xM
[
i
df[
x
< . (6.3)
Equivalently: f C
,r
(M, N)
(f
TN). Dene g
: M N by g
(x) :=
exp
f(x)
((f(x)), ( exp ). Assume r k and dene
[[
r
:=
r
i=0
sup
xM
[
i
[
x
<
N
< r
inj
(N). (6.4)
Then g
= exp C
,r
(M, N). We dene f to be close to g in C
,r
(M, N)
in the L
p
-category as follows: Let 0 < <
1
2
r
inj
(N), 1 < p < .
V
:= f, g C
,r
(M, N) ; C
(f
TN) s.t. g = g
(6.5)
and [[
p,r
:= (
_
M
r
i=0
[
i
[
p
x
d vol
x
(g))
1/p
< . (6.6)
Then 1 := V
0<<r
inj
(M)/2
is a basis for a metrizable uniform structure on
C
,r
(M, N). Let C
p,r
(M, N) be the completion of C
,r
(M, N) in this uni-
form structure. Then C
p,r
(M, N) is a C
k+1r
-Banach manifold and for p = 2
a Hilbert manifold.
Remarks:
1) A neighborhood |
(f) of f C
p,r
(M, N) is given by: Let 0 < < r
inj
(N)
|
(f) = g C
p,r
(M, N) ; g = exp , C
(f
TN), [[
p,m
< . (6.7)
2) The tangent space at f is given by T
f
C
p,r
(M, N) = C
p,r
(f
TN)
3) The change of coordinates is given by exp
g
exp
1
f
, it is C
kr+1
.
If M is compact then k = .
4) If g comp(f) then T
f
C
p,r
(M, N) T
g
C
p,r
(M, N) , otherwise not.
60
6.3 The Bounded Dieomorphism Group Diff
p,r
(M)
Problem: f bounded , f
1
bounded, i.e. C
,m
(M)
Diff(M) is not a
group, hence f C
,m
(M)
Diff(M) , f
1
C
,m
(M). We need an
additional assumption to obtain a group. Let [[
min
(df) denote the absolute
value of the eigenvalues of the Jacobian of f and set
Diff
p,r
(M) := f C
p,r
(M, M) ; f bijective and [[
min
(df) > 0, (6.8)
then Diff
p,r
(M) is open in C
p,r
(M, M) , hence a C
k+1r
-Banach manifold
and we have the following
Theorem 6.1 ( Eichhorn - Schmid [15]) Let (M
n
, g) be an open, oriented,
complete Riemannian manifold satisfying (I), (B
Diff
p,r
(M) is an ILB - Lie group; and for p = 2 it is
an ILH - Lie group.
6.3.1 Dierentiability of the Group Operations:
The dierentiability of the group operations (composition and inversion) fol-
low from the - and - lemma.
Lemma 6.1 (-lemma) Assume r k, r >
n
p
+1, f Diff
p,r
(M). Then
the right multiplication
f
: Diff
p,r
0
(M) Diff
p,r
(M),
f
(g) = g f is of
class C
k+1r
.
Lemma 6.2 (-lemma) Let k + 1 (r + s) > s, f Diff
p,r+s
0
(M)
Diff
p,r
0
(M), r >
n
p
+ 1. Then the left multiplication
f
: Diff
p,r
(M)
Diff
p,r
(M),
f
(g) = f g is of class C
s
.
Theorem 6.2 ( Eichhorn - Schmid [15]) Let (M
n
, g) be an open, oriented,
complete Riemannian manifold satisfying (I), (B
) . Let Diff
p,
0
(M) :=
lim
Diff
p,r
0
(M) with the inverse limit topology. Then
Diff
p,
0
(M), Diff
p,r
0
(M) ; r >
n
p
+ 1
is an ILB - Lie group; and for p = 2 it is an ILH - Lie group.
61
6.4 Volume Preserving and Symplectic Dieomorphisms.
We have analogous results as in the compact case for the subgroups of volume
preserving and symplectic dieomorphism. The ideas of the proofs are similar
but technically much more complicated, so we just state the results: (for
proofs see Eichhorn and Schmid [15])
Let be a C
= f Diff
r
; f
= .
Theorem 6.3 (Eichhorn and Schmid [15])
a) Diff
p,
= lim
r
Diff
p,r
M
If (M
n
, g) is an open, oriented, complete Riemannian manifold satisfying
(I), (B
k
) then the Sasaki metric on the co-sphere bundle in
T
M satises
(I), (B
k1
). Let be the canonical 1-form on T
M and consider
Diff
p,r
(
T
M) = f Diff
p,r
(
T
M) ; f
=
Theorem 6.4 (Eichhorn, Schmid [15])
Diff
p,
(
T
M) = lim
r
Diff
p,r
(
T
M)
is an ILH-Lie group.
This is the space of phase functions for the Fourier integral operators !
6.6 Pseudodierential Operators and Fourier Integral
Operators on Open Manifolds
If (M
n
, g) is open the previous denitions of pseudodierential operators
and Fourier integral operators does not make sense. We need to adapt the
62
class of symbols and phase functions to the bounded geometry of M in order
to obtain globally dened Fourier integral operators A : C
c
(M) T
(M).
Then the corresponding spaces DO and FIO have similar properties as in
the compact case and we can use the same ideas as before to construct Lie
group structures. We need the following adaptations for a Fourier integral
operator
Au(x) = (2)
n
_ _
e
i(x,y,)
a(x, )u(y)dy d (6.9)
Symbols: The family of local symbols a(x, ) together with their deriva-
tives should be uniformly bounded.
Phase functions: The phase functions (x, y, ) should locally generate
canonical transformations in the space Diff
p,r
(
T
M).
These symbols and phase functions dene the class of so called uniform
pseudo dierential - and Fourier integral operators denoted by |DO
m
and
|FIO
m
respectively. The details are quite technical, so we present here only
the basic ideas and nal result and refer to Eichhorn and Schmid [16] for
details.
As in the compact case (section 5) we get an exact sequence of groups
I (|DO
0
)
j
(|FIO
0
)
p
Diff
p,r
(
T
M) id (6.10)
Now we follow the same ideas as in the compact case: step 1,2...7 to construct
ILH Lie groups structures on these spaces.
Theorem 6.5 (Eichhorn - Schmid [16])
|DO = lim
s
|DO
s
is an ILH Lie group
|FIO = lim
t
|FIO
t
is an ILH Lie group
the multiplication : |FIO
t+r
|FIO
t
|FIO
t
(A, B) = AB is
C
k
dierentiable, k = min(r, t)
the inversion : |FIO
t+r
|FIO
t
; (A) = A
1
is C
k
dierentiable ,
k = min(r, t)
the right multiplications R
A
: |FIO
t
|FIO
t
: R
A
(B) = BA are C
t
dierentiable , for any A |FIO
t
the left multiplications L
A
: |FIO
t
|FIO
t
: L
A
(B) = AB are C
0
(continuous) , for any A |FIO
t
.
63
7 Applications to Fluid Dynamics and
Quantization
We briey discuss some applications to uid dynamics, the periodic and non-
periodic KdV equations, the classical -, topological - and non-homogeneous
Euler equations, as well as quantization. Some of this work is still in progress.
7.0.1 The KdV Equation and the Group of Fourier Integral
Operators
The Korteweg deVries (KdV) equation u
t
= 6uu
x
u
xxx
is an innite di-
mensional Hamiltonian system with respect to the Poisson bracket
F, G(u) =
_
F
u
x
G
u
dx (7.1)
and Hamiltonian
H(u) =
_
(u
3
+
1
2
u
2
x
)dx (7.2)
which means that a function u satises Hamiltons equations u
t
= u, H if
and only if u is a solution of the KdV equation. This was shown by Gardner
and Kruskal in 1971.
The question we asked is: Where does this Poisson bracket (7.1) and
this Hamiltonian (7.2) come from ? Is there a natural space on which this
Hamiltonian system lives ? And the answer is the following:
Theorem 7.1 ( Adams - Ratiu - Schmid [1])
A) The Poisson bracket (7.1) is the natural Lie-Poisson bracket on the coad-
joint orbit of the Lie group of invertible Fourier integral operators ( = FIO
= k
= k
and
k
= h
R
be two functions that are constant on coadjoint orbits of G in g
. Then for
A h
, F
A
, H
A
= 0, where F
A
and H
A
are the restriction of F and H to
the coadjoint orbit of A in h
.
For the KdV equation the Lie group is G = FIO
(S
1
) with Lie alge-
bra g = DO(S
1
), where M = S
1
is the unit circle. Then each pseu-
dodierential operator P DO
m
(S
1
) has a total symbol of the form
p(x, ) =
<jm
p
j
(x)
j
.
The Lie algebra g = DO decomposes into the two subalgebras h =
DO
=
m<0
DO
m
and k = DO
+
=
m0
DO
m
, i.e. g = h k
becomes DO = DO
DO
+
.
We have an inner product < P, Q >:= Trace(P Q) where the trace is
dened by Trace(P) :=
_
p
1
(x)dx. With this we identies DO
DO
and (DO
DO
+
. So for g = h k we get g
= k
, i.e
g
= DO
DO
+
DO
DO
DO
DO
+
DO
.
The Lie-Poisson bracket on h
= DO
DO
+
at A DO
+
becomes
F, H(A) =< A, [
F
A
,
H
A
] >=
_
(A [
F
A
,
H
A
])
1
dx (7.3)
where ...)
1
means taking the order(1) part of the symbol.
The Lie-Poisson evolution equations
F = F, H for any function F on
DO
are equivalent to
A = X
H
(A) = ad
H
A
(A) = [
H
A
, A]
+
(7.4)
on DO
DO
+
, where ...]
+
means taking only the part in DO
+
For the Schrodinger operator A DO
+
with total symbol given by
a(x, ) = a(x)+
2
the Lie-Poisson bracket of two functions F, G : DO
R
at A becomes
F, G =
_
F
a
x
G
a
dx , which is the Gardner bracket (7.1).
65
For the functionals H
k
(A) = Trace(A
k
) =
_
(A
k
)
1
dx, k N we have
H
k
A
= kA
k1
, hence [A,
H
k
A
] = [A, kA
k1
] = 0.
Thus H
k
are constant on coadjoint orbits. By the Kostant-Symes theo-
rem, restricting the H
k
to DO
DO
+
gives the Gelfand-Dikii family of
commuting integrals for KdV.
We get the following integrals, for example : H
0
=
_
a dx , H
1
=
_
1
2
a
2
dx,
H
2
=
_
(a
3
+
1
2
a
2
x
)dx H , which is the Gardner Hamiltonian (7.2)!
H
3
=
_
(
5
8
a
4
+
5
4
aa
2
x
+
1
8
a
2
xx
)dx etc.
7.0.2 The KdV Equation and the Lie Group |FIO
We showed above that the periodic KdV equation (which describes shallow
water waves) is a Hamiltonian system on a coadjoint orbit of the Lie group
of Fourier integral operators on the compact manifold M = S
1
the circle.
Having done all the work for open manifolds we proved the similar result for
the non periodic KdV equation on the real line M = R.
Theorem 7.2 (Eichhorn - Schmid [17]) The non-periodic KdV equation on
the real line is a Hamiltonian system with respect to the Lie-Poisson bracket
on the coadjoint orbit through the Schrodinger operator of the innite dimen-
sional Lie group of invertible Fourier integral operators ( = |FIO
(R).
7.0.3 Hydrodynamics and the Dieomorphism Group Diff
,r
(M)
1. The Euler Equations and Diff
,r
(M)
The topological Euler equations are given by
E
top
_
u
t
+
u(t)
u(t) = grad p
div
u(t) = 0 .
(7.5)
We call these equations the topological Euler equations because is a
xed volume form on (M
n
, g), not necessarily the phase space volume given
by the Riemannian metric g . So we have two dierent volume forms on M.
66
The covariant derivative =
g
is taken with respect to the Riemannian
metric g, whereas the divergence div = div
, dened by L
X
= (div
X),
is the divergence with respect to . The vector eld u = u(x, t) is a time
dependent C
1
vector eld on (M
n
, g).
Then u(t) denes a 1-parameter family of dieomorphisms f
t
dened by
df
s
ds
[
s=t
= u(t) f
t
. (7.6)
We have the analogues result as discussed in section 3.1, Theorem 3.3 ( Ebin
- Marsden) for non-compact manifolds.
Theorem 7.3 (Eichhorn - Schmid [15]) Assume (M
n
, g) is an open Rie-
mannian manifold satisfying (I) and (B
k
). Then u(t) satises the topological
Euler equations E
top
i f
t
t
is a geodesic in Diff
,r
(M).
The Classical Euler equations for an incompressible, homogeneous uid
without viscosity are a special case with = dvol
x
(g) the volume of the Rie-
mannian metric g and =
g
, div = div
dvolx(g)
.
2. The Non-Homogeneous Euler Equations
The non-homogeneous Euler equations with a mass density (x, t) > 0
are given by
E
NH
_
_
_
u
t
+
u(t)
u(t) =
1
grad p
t
+ grad u = 0
div
u(t) = 0 .
(7.7)
If = constant these are the classical homogeneous Euler equations. For
= (g) the corresponding equations on Diff
s
(
T
M) e (7.8)
0 DO
0
j
DO
1
+1
(
T
M) 0 . (7.9)
For P DO
1
its principal symbol (P) :
T
M R is a smooth
function , homogeneous of degree +1. Hence (P) C
+1
(
T
M). Moreover
is a surjective Lie algebra homomorphism
([P, Q]) =
1
i
(P), (Q). (7.10)
That means the commutator bracket of the operators corresponds to the
Poisson brackets of their principal symbols , which means quantization of
C
+1
(
T
M) via DO
1
!
Consider the Lie subalgebra DO
1
DO
1
of all dierential operators of
order 1 which is isomorphic to the space of all smooth vector led on M, i.e.
DO
1
= X
X
j
(x)
x
j
, hence its principal
symbol is given by (X) =
X
j
j
and we write X as pseudodierential
operator
Xu(x) = (2)
n
_ _
e
i(xy)
(X)u(y)dyd . (7.11)
Consider the Lie subalgebra L(T
M) C
+1
(
T
x
M. Locally f L(T
M) is of the form
f(x, ) =
f
j
(x)
j
.
Theorem 7.5 (Schmid [35]) The symbol map induces a Lie algebra
isomorphism : X
(M) L(T
M) : (X)
x
= i
x
X(x), x M,
x
T
x
M
([X, Y ]) =
1
i
(X), (Y ).
68
This means quantization of L(T
M) via X
(M) .
In general, by Egorovs theorem we have the following: For any f, g
C
(T