Table of contents : CONTENTS PREFACE CHAPTER I: STATEMENT OF THE THEOREM OUTLINE OF THE PROOF §1. The index theorem §2. The topological index §3. The analytical index §4. Appendix CHAPTER II : REVIEW OF K-THEORY §1. K(X) a finite CW-complex §2. The Chern character §3. The difference construction §4. L-theory §5. Products in L-theory CHAPTER III : THE TOPOLOGICAL INDEX OF AN OPERATOR ASSOCIATED TO A G-STRUCTURE CHAPTER IV: DIFFERENTIAL OPERATORS ON VECTOR BUNDLES §1. Notation §2. Jet bundles §3. Differential operators and their symbols §4. Hermitian bundles and adjoint operators §5. Green's forms §6. Some classical differential operators §7. Whitney sums §8. Tensor products §9. Connections and covariant derivatives §10. Spin structures and Dirac operators CHAPTER V: ANALYTICAL INDICES OF SOME CONCRETE OPERATORS §1. Review of Hodge theory §2. The Euler Characteristic §3. The Hirzebruch signature theorem §4. Odd-dimensional manifolds CHAPTER VI: REVIEW OF FUNCTIONAL ANALYSIS CHAPTER VII: FREDHDIM OPERATORS CHAPTER VIII: CHAINS OF HTLBERTIAN SPACES §1. Chains §2. Quadratic interpolation of pairs of hilbert spaces §3. Quadratic interpolation of chains §4. Scales and the chains (Z^n, V) CHAPTER IX: THE DISCRETE SOBOLEV CHAIN OF A VECTOR BUNDLE §1. The spaces C^k(ξ) §2. The hilbert space H^0(ξ) §3. The spaces H^k(ξ) CHAPTER X: THE CONTINUOUS SOBOLEV CHAIN OF A VECTOR BUNDLE §1. Continuous Sobolev chains §2. The chains {H^k(T^n, V)} §3. An extension theorem §4. The Rellich, Sobolev, and restriction theorems CHAPTER XI: THE SEELEY ALGEBRA CHAPTER XII: HOMOTOPY INVARIANCE OF THE INDEX CHAPTER XIII: WHITNEY SUMS §1. Direct sums of chains of hilbertian spaces §2. The Sobolev chain of a Whitney sum §3. Behaviour of Smblk with respect to Whitney sums §4. Behaviour of Intk and σ^k under Whitney sums §5. Behaviour of the index under Whitney sums CHAPTER XIV: TENSOR PRODUCTS §1. Tensor products of chains of hilbertian spaces §2. The Sobolev chain of a tensor product of bundles §3. The # operation §4. The property (S6) of the Seeley Algebra §5. Multiplicativity of the index CHAPTER XV: DEFINITION OF ia AND it ON K(M) §1. Definition of the analytical index on K(B(M), S(M)) §2. Multiplicative properties of it §3. Proof of Lemma 1 §4. Definition of it and ia on K(M) §5. Summary of the properties of ia and it on K(M) §6. Multiplicative properties of i on K(X ) §7. Direct check that ia = it in some special cases CHAPTER XVI: CONSTRUCTION OF Intk §1. The Fourier Transform §2. Calderón-Zygnund operators §3. Calderón-Zygmund operators for a compact manifold §4. Calderón-Zygmund operators for vector bundles §5. Definition and properties of Intr (ξ, η) §6. An element of Into(S^1) with analytical index -1 §7. The topological index of the operator of §6 §8. Sign conventions CHAPTER XVII: COBORDISM INVARIANCE OF THE ANALYTICAL INDEX CHAPTER XVIII: BORDISM GROUPS OF BUNDLES §1. Introductory remarks §2. Computation of Ωk(X) ⊗ Q §3. The bordism ring of bundles CHAPTER XIX: THE INDEX THEOREM: APPLICATIONS §1. Proof of the index theorem §2. An alternative formulation of the index theorem §3. The non-orientable case of Theorem 2 §4. The Riemann-Roch-Hirzebruch theorem §5. Generalities on integrality theorems §6. The integrality theorems APPENDIX I : THE INDEX THEOREM FOR MANIFOLDS WITH BOUNDARY §1. Ellipticity for manifolds with boundary §2. The difference element [σ(d, b)] §3. Comments on the proof APPENDIX II : NON STABLE CHARACTERISTIC CLASSES AND THE TOPOLOGICAL INDEX OF CLASSICAL ELLIPTIC OPERATORS §1. Characteristic classes §2. τ-homomorphism §3. The character of classical elliptic operators
Citation preview
Annals of Mathematics Studies Number 57
ANNALS OF MATHEMATICS STUDIES E dited by Robert C. Gunning, John C. M oore, and Marston Morse 1.
Algebraic Theory of Numbers, by
H erm ann W
3. Consistency of the Continuum Hypothesis, by
eyl
K u r t Go d el
11.
Introduction to Nonlinear Mechanics, bu N.
20.
Contributions to the Theory of Nonlinear Oscillations, Vol. I, edited by S.
21.
Functional Operators, Vol.
24.
Contributions to the Theory of Games, Vol. I, edited by H. W.
25.
Contributions to Fourier Analysis, edited by A. A. P. C a l d e r o n , and S. B o c h n e r
28.
Contributions to the Theory of Games, Vol. II, edited by H. W.
30.
Contributions to the Theory of Riemann Surfaces, edited by L.
33.
Contributions to the Theory of Partial Differential Equations, edited by L. n e r , and F. J o h n
34.
Automata Studies, edited by C. E.
38. 39. 40.
1,
by
J ohn
and N.
Kr ylo ff
B o g o l iu b o f f L
efsch etz
N eum a n n
von
and J.
Sh annon
Kuh n
W.
Zyg m u n d ,
M
and A. W.
T uck er
M.
M o rse,
T ra n sue,
Kuhn
Ah lfo rs
and A. W.
T uckj
et al. B ers,
S.
Boc
cC a rth y
Linear Ineaualities and Related Systems, edited bu H. W.
Kuhn
and A. W.
T uck er
Contributions to the Theory of Games, Vol. Ill, edited by M. D r e s h e r , A . W. T u c k e r and P. W o l f e Contributions to the Theory of Games, Vol. IV, edited by R. D u n c a n L u c e and A. W. T u ck er
41.
Contributions to the Theory of Nonlinear Oscillations, Vol. IV, edited by S.
42.
Lectures on Fourier Integrals, by S.
43.
Ramification Theoretic Methods in Algebraic Geometry, by S.
44.
Stationary Processes and Prediction Theory, by H.
45.
Contributions to the Theory of Nonlinear Oscillations, Vol. V, edited by L. S a l l e , and S. L e f s c h e t z
F
e fsc h etz
46.
Seminar on Transformation Groups, by A. Theory of Formal Systems, by R.
48.
Lectures on Modular Forms, by R. C.
49.
Composition Methods in Homotopy Groups of Spheres, by H.
50.
Cohomology Operations, lectures by N. E.
B o rel
A bhyankar
u rsten berg
47.
E
L
B o ch n er
C e s a r i,
J.
et al.
G u n n in g
St e en r o d ,
T oda
written and revised by D. B. A.
M il n o r
52. Advances in Game Theory, edited by M. 53. Flows on Homogeneous Spaces, by L.
D resh er,
A u sla n d er,
54. Elementary Differential Topology, by J. R. 55. Degrees of Unsolvability, by G. E. 56. Knot Groups, by L. P.
a
Sm u lly a n
p s t e in
51. Morse Theory, by J. W.
L
L. L.
Sh a p l e y , Green ,
F.
and A. W. H
ahn,
M u n kres
Sacks
N e u w ir t h
57. Seminar on the Atiyah-Singer Index Theorem, by R. S. 58. Continuous Model Theory, by C. C.
C hang
and
H.
J.
P a l a is , K e is l e r
et al.
T u ck er
et al.
SEMINAR ON THE A T IY A H -S IN G E R INDEX THEOREM BY
Richard S. Palais
WITH
CONTRIBUTIONS
M . F . A T IY A H
BY
R. T . SE E L E Y
A . BO REL
W . SH IH
E. E. FLO Y D
R. SO LO VAY
PRINCETON, NEW JERSEY PRINCETON UNIVERSITY PRESS 1965
STATEMENT OF THE THEOREM OUTLINE OF THE PROOF, by A. B o r e l....................................................
§ 1. §2. §3. §k .
The index th e o r e m ...................................................................................... The to p o lo g ic a l index ............................................................................. The a n a ly t ic a l index ............................................................................. Appendix...........................................................................................................
6 9
K(X) a f i n i t e CW -com plex..................................................................... The Chern c h a r a c t e r .................................................................................. The d iffe r e n c e co n stru ctio n ................................................................ L - t h e o r y ....................................................................................................... Products in L -th e o ry ..................................................................................
13 it
CHAPTER IV:
27
DIFFERENTIAL OPERATORS ON VECTOR BUNDLES, by R. S. P ala is 51
N o t a t i o n ....................................................................................................... J e t bundles ................................................................................................... D i f f e r e n t i a l op erato rs and t h e ir sym bols....................................... Hermitian bundles and a d jo in t o p e r a t o r s ....................................... G reen 's f o r m s .............................................................................................. Some c l a s s i c a l d i f f e r e n t i a l op erato rs ........................................... Whitney s u m s ............................................................................................... Tensor products .......................................................................................... Connections and co v a ria n t d e r iv a t iv e s ........................................... Spin str u c tu r e s and D irac op erato rs ...............................................
CHAPTER V:
15 18 20
THE TOPOLOGICAL INDEX OF AN OPERATOR ASSOCIATED TO A G-STRUCTURE, by R. S o lo v a y ...............................................
§ 1. §2. §3. § 14-.
1 k
13
CHAPTER I I I :
§ 1. §2. §3. § 14-. §5. §6. §7. §8. §9. §10.
1
REVIEW OF K-THEORY, by R. S o l o v a y ...........................................
CHAPTER I I : § 1. §2. §3. § 14-. §5.
ix
ANALYTICAL INDICES OF SOME CONCRETE OPERATORS, by M. Solovay ...............................................................................................
51 55 61
69 73 75 79 81 Qk
91 95
Review o f Hodge th e o r y ............................................................................. 95 The E u ler C h a r a c te r is tic ..................................................................... 96 The H irzebruch sig n a tu re t h e o r e m .................................................... 98 Odd-dimensional m a n i f o ld s ................................................................. 103
CHAPTER V I: CHAPTER V II:
REVIEW OF FUNCTIONAL ANALYSIS, by R. S. P a la is . . . . FREDHDIM OPERATORS, by R. S. P a l a i s .............................. v
107 119
CONTENTS
CHAPTER V I II: §1. §2. §3. § 14-.
. . .
125
C hains........................................................................................................... Q uadratic in te r p o la tio n o f p a ir s o f h ilb e r t spaces . . . Q uadratic in te r p o la tio n o f chains ............................................... S c a le s and the chains Z n , V ) } ...........................................
125 131 139 i in
CHAPTER IX: § 1. §2. §3.
THE DISCRETE SOBOLEV CHAIN OF A VECTOR BUNDLE, by R. S. P a l a is .............. ’ .................................................................. ;
The spaces Ck (£) ................................................................................. The h ilb e r t space H°U) ................................................................ The spaces Hk ( | ) ................................................................................. ...
CHAPTER X: § 1. §2. §3. § i.
CHAINS OF HTLBERTIAN SPACES, .by R. S. P a la is .
THE CONTINUOUS SOBOLEV CHAIN OF A VECTOR BUNDLE, by R. S. P a l a i s .........................................................................................
Continuous Sobolev c h a i n s ................................................................ ... The chains {H^T11, V ) } ........................................................................ An ex ten sio n t h e o r e m ............................................................................ The R e llic h , Sobolev, and r e s t r ic t i o n theorem s.........................
CHAPTER XI:
THE SEELEY ALGEBRA, by R. S. P a l a i s .................................. ...
U7 114-7 1 14-8 1 14-9 155 155 156 162 1 6b
175
CHAPTER X II: HOMOTOPY INVARIANCE OF THE INDEX, by R. S. P a la is . . , 185 CHAPTER X III: WHITNEY SUMS, by R. S. P a l a is ................................................... § 1. §2. §3. § t. §5.
D ir e c t sums o f chains o f h ilb e r t ia n sp a ce s.............................. ... The Sobolev chain o f a Whitney s u m ............................................., Behaviour o f Smbl^. w ith re sp e c t to Whitney sums . . . ., Behaviour o f In t^ and o^. under Whitney sums....................... . Behaviour o f the index under Whitney sums ..................................
CHAPTER XIV: § 1. §2. § 3. § t. §5.
Tensor products o f chains o f h ilb e r t ia n sp ace s......................... The Sobolev chain o f a tensor product o f b u n d les..................... The # o p e r a t i o n ................................................................................. ... The p ro p erty (S6) o f the S e ele y A l g e b r a .................................. ... M u l t i p li c a t iv i t y o f the index ............................................................
CHAPTER XV: § 1. §2. §3. §1*. §5. §6. §7.
TENSOR PRODUCTS, by R. S. P a l a i s ...........................................
DEFINITION OF
i & AND
it
191 192
193 193 19t 197 197 201 206 209 210
ON K(M) , by R. M. Solovay, 215
D e fin itio n o f the a n a ly t ic a l index on K(B(M), S(M)). . .. M u lt ip lic a t iv e p r o p e rtie s o f i^ ................................................... Proof o f Lemma 1 ..................................................................................... D e fin it io n o f i^ and i a on K(M).............................................. Summary o f the p r o p e rtie s o f i & and i t on K(M) . . ., M u lt ip lic a t iv e p r o p e rtie s o f i on K ( X ) ................................ .. D ir e c t check th a t i & = i^ in some s p e c ia l cases . . . ., vi
191
215 217 222 223 226
228 232
CONTENTS CHAPTER XVI:
CONSTRUCTION OF
I n t k ; b y R. S. P a la is
and R. T. S e e l e y .......................................................................................235 §1. §2. §3 . § 3+. §5 . §6.
The F o u r ie r T r a n s f o r m ........................................................................................235 C alderon-Zygnund o p e r a t o r s * . 2^3 Calderon-Zygm und o p e r a to r s f o r a compact m a n i f o l d .........................259 Calderon-Zygm und o p e r a to r s f o r v e c t o r b u n d l e s ..................................266 D e f i n i t i o n and p r o p e r t ie s o f I n t r (£ , r\) ............................................... 270 An elem ent o f I n t Q( S 1 ) w ith a n a l y t i c a l in d e x - 1 .........................272
§7. §8.
The t o p o l o g i c a l in d e x o f th e o p e r a to r o f §6 . . . . . . . . . 275 S ig n c o n v e n t i o n s ............................................................................................2 8 1
CHAPTER X V II:
COBORDISM INVARIANCE OF THE ANALYTICAL INDEX, b y R. S . P a l a is and R. T. S e e l e y ....................................................... 285
CHAPTER X V III: §1. §2 .
BORDISM GROUPS OF BUNDLES, b y E . E. F lo y d ............................ 303
In tr o d u c to r y re m a rk s.............................................................................................303 Com putation o f ftk (X) ® Q ............................................................................... 306
§ 3 . The bord ism r i n g o f b u n d l e s ...........................................................................307 CHAPTER XIX:
THE INDEX THEOREM:
APPLICATIONS, b y R. M. S o lo v a y . .
313
§1. §2.
P ro o f o f th e in d e x theorem ............................................................................... 313 An a l t e r n a t i v e fo r m u la tio n o f th e in d e x t h e o r e m ............................. 315
§3. § 3+.
The n o n - o r ie n ta b le ca se o f Theorem 2 318 The R iem ann-R och-H irzebruch t h e o r e m .........................................................32^
§ 5 . G e n e r a l i t i e s on i n t e g r a l i t y th eorem s........................................................ 326 §6. The i n t e g r a l i t y th eo rem s................................................................................... 329 APPENDIX I :
THE INDEX THEOREM FOR MANIFOLDS WITH BOUNDARY, by M. F . A t iy a h .................................................................................................... 337
§1. §2.
E l l i p t i c i t y f o r m a n ifo ld s w ith b o u n d a r y ............................................... 338 The d if f e r e n c e elem ent [ cr(d, b) ] .............................................................3^6
§3.
Comments on th e p r o o f ........................................................................................350
APPENDIX I I :
NON STABLE CHARACTERISTIC CLASSES AND THE TOPOLOGICAL INDEX OF CLASSICAL ELLIPTIC OPERATORS, by W. S h ih
C h a r a c te r is tic c la s s e s
. .
353
§1. §2 .
t - h o m o m o rp h ism .................................................................................. ....
. . . . . .
360
§3.
The c h a r a c te r o f c l a s s i c a l e l l i p t i c o p e r a to r s ............................
362
vii
353
PREFACE
The Index Theorem is a striking and central result in a rapidly developing field of research which may be described as the study of the re lation between analytic and topological invariants of a certain class of linear maps between sections of differentiable vector bundles (the class of integro-differential, or pseudo-differential operators).
The field is not
really new and has several classical results, for example the Hodge theory of harmonic forms and the Hirzebruch formulation of the Riemann-Roch theorem, both of which are in fact closely related to the Index Theorem.
Moreover,
since the Index Theorem there have been other notable results discovered by Atiyah, Bott, Singer and others.
To mention only two there is a mod 2 "in
dex” defined for elliptic operators on real bundles with self adjoint sym bols, which in certain cases seems to be related to the Arf invariant of a manifold, and there has been a remarkable generalization of the Lefschetz fixed point formula, which already has had important applications to the theory of h-cobordism and to the study of the fixed points of periodic trans formations . This book consists mainly of slightly revised notes of a semi nar held at the Institute for Advanced Study in 1 9 6 3 -6k upon the initiative of A. Borel.
Exceptions are Chapters XVT and XVII and the Appendix by M.
Atiyah which were written somewhat later. Aside from going through the details of the proof of the Index Theorem, the major emphasis of the seminar was placed on developing the to pological and analytical machinery associated with integro-differential operators.
On the topological side the agreement was to assume a reasonable
degree of sophistication.
Thus, the basic facts concerning K-theory and
characteristic classes are reviewed rather than proved and the emphasis is on showing how, with these tools, elliptic operators give rise to cohomology
ix
PREFACE classes and on studying the properties of these classes. side, it was decided to start more or less ab i n i t i o .
On the analytical
The reason for this
somewhat unbalanced exposition is in part due to the predelictions of the organizers of the seminar, but also in part it is due to the fact that while most of the algebraic topology involved is covered in complete detail in easily accessible published papers, much of the analysis is quite recent, and the published versions often refer explicitly only to the case of trivial bundles over domains in Euclidean space. I would like to thank the many persons who attended the semi nar lectures and whose suggestions lead to a smoother presentation.
In par
ticular the section on jet bundles in Chapter IV was considerably improved by A. Vasquez and L. Charlap, and a suggestion by M. Kneser led to a smoother version of the section on Fredholm operators (Chapter VII). I would also like to express my great appreciation to F. Browder, E. Nelson and E. Stein for their constant and invaluable advice while I was writing my part of these notes. Finally, a very careful reading of the entire manuscript by W. Shih led to the elimination of a great number of misprints and other errors. For this I am sure he has the readers thanks as well as mine.
Richard S. Palais Brandeis University January, 1 9 6 5
CHAPTER I STATEMENT OF THE THEOREM OUTLINE OF THE PROOF A. Borel As an introduction to the subject matter of this seminar, this lecture gives the statement and a rough description of the proof of the index theorem.
More details on the proofs of the results stated and on
the concepts discussed here will be found in the subsequent lectures. Manifolds are compact, smooth (i.e.,
C°°), orientable and o r i
e n t e d , consist of connected components of the same dimension, and, unless otherwise stated, have no boundary. be denoted at a point
Complex vector bundles will usually
bythe same letter as their total spaces. x of a bundle
E and
C°°(E)
tions of a smooth complex vector bundle
§1.
the space E
Let
on
X.
d:
C°°(E) -► C°°(F)
X
be a manifold,
A differential operator
of smooth cross sec
over a manifold.
d
E, F
from
E
smooth complex vector bundles to
F
is a linear map
which is given locally by a matrix of ordinary (i.e.,
scalar) differential operators.
More precisely, let
definition of a local chart, x 1, ..., xn (n = dim X) C°°(U)
taking the terms of degree partial derivatives vector
P^j
of
d
in the partial derivatives.
at
x
is obtained from
r, substituting indeterminates
d/dx^,
^ d x +-..,+ £n &X-n ,
and multiplying by
(-l)r ^2 .
there is then associated a
with complex coefficients.The system is elliptic if matrix is invertible for all
i ^ 0
and for all
cumstances, it is known that
ker d
and
coker
P
for the To each cotangent
p x q matrix
p =
q and
x e X.
by
Q(x, |)
ifthis
Under those cir
d = C°°(F) /im d
are finite
dimensional; the difference of their dimensions is the index of
d.
In
order to distinguish it from another index, to be defined below, we shall call this the analytical index of (1)
d
and denote it by
ia (&)•
Thus
iQ cl(d) = dim ker d - dim coker d
It is known that
ia (d)
is invariant under deformations of
d, and this led
Gelfand to ask whether it could be expressed in terms of topological data. The index theorem provides a positive answer to that question. formulate it, a mixed rational cohomology class on the symbol of
d
ch d e H (X; Q ) , depending
(see section 2, below), is introduced.
^(X) be the Todd class or
In order to
the complexified tangentbundle of
Let moreover X.
It is
obtained by considering the product of formal power series -xf(l-e_V where the
x^
1 • (l- e V 1
(i = 1, ..., s > dim X/2)
,
are indeterminates, expressing it
as a formal sum of homogeneous polynomials, which are then symmetric in the 2
x^, the
writing these as polynomials in the elementary symmetric functions in 2 x^ 's, and then replacing the j-th symmetric function by the j-th
Pontrjagin class of
X
(j = 1, 2, ...).
The topological index i^Cd)
is
then defined as (2)
it(d) = (ch d *^(X))[X]
(n = dim X)
,
where the right hand side stands for the value of the n-dimensional component of
ch d. 3T(Yf)
on the fundamental cycle of
THEOREM (Atiyah-Singer) . on the manifold
X.
Then
Let
d
X.
We have then the
be an elliptic operator
ia (d) = it(d) .
§1
I: OUTLINE OF PROOF
3
This is not the most general form of the theorem. Singer have extended it:
Atiyah and
(a) to elliptic complexes, I.e., sequences of
operators whose spribols form an exact sequence, but this can easily be re duced to the case of one operator; (b) to a wider class of operators; this is quite important for the proof, and will be dealt with at length in this seminar; (c) to boundary value problems; this will probably not be touched upon here, for lack of time and material. M. Atiyah.]
[See however, Appendix I, by
A fourth generalization; which would include Grothendie.ck> s
version of the Riemann-Roch theorem over
C, is contemplated but, as far as
I know, has not yet been carried out. The proof falls naturally into two parts: which investigates the properties of cerned with
ia (d).
it(d) > and an analytical one, con
The former one also gives the motivation for the latter
one, and here we shall summarize it first. give some more details on 2.
a topological one,
ch d. ch d.
The definition of
The Grothendieck group
Before doing that, however, we
K(X)
of
X
Let
X
be a finite CW-complex.
is the quotient of the free commutative
group generated by the isomorphism classes of complex vector bundles on by the subgroup generated by the elements E*
0
K(X)
is an exact sequence.
If a base point
is the kernel of the homomorphism
bundle
E
the dimension of
relative Grothendieck group means
X
with
Y
E - E ! - EM
E .
If
K(X, Y)
Y
x e X
K(X) -► Z
where
o -► E" -*■ E -►
has been chosen, then
which assigns to each
is a closed subcomplex of
is by definition
K(X/Y),
X
be again a manifold.
Let
B(X)
(resp.
with respect to some Riemannian metric, and It is known that, via cup product,
coefficients) is a free module over n = dim X,
n,
the inverse of
H*(X),
it: B(X) -+ X
where
X/Y
a
be
T*(X),
the natural pro
H*(B(X), S(X)) (any ring of with a canonical generator
whence the existence of an isomorphism
cp*: H (B(X) , S(X)) -► H (X), by
the
S(X))
the unit ball (resp. unit sphere) bundle of the cotangent bundle
of degree
X,
pinched to a point, which is then taken as base point.
Let now
jection.
X
U
cp*: H*(B(X),
the Thom isomorphism, which decreases dimensions a o U(a e H*(X)).
d: C°°(E) -+ C°°(P)
Let on
§1
BOREL
k
X.
be a differential operator of order
It is well-known that the matrix
r
Q(x, t), which was defined above
using local coordinates, has in fact an intrinsic meaning and associates to each
£ e T*(X)x
a linear map of
of course smoothly on the bundles on symbol of
d.
B(X)
x, |,
Ex
Fx «
This linear map depends
whence a homomorphism
lifted from
E
and
Ellipticity means that
isomorphism.
into
F
via
*,
[.trp (d) ] e K(B(X), S(X))
**F
of
to be called the
&r (d), restricted to
In that case, one can associate to
difference element
a^(d) : jt*E
S(X),
is an
(jt*E, tc*F, ap (d)|S(X))
(see [2]).
Then
ch d
a
is de
fined by (3)
ch d = ( - D n(n+1)/2 T#ch[
0, a. € z ) , QJ
-L
QJ
the Fourier transform
=
-I-
1 ... sn n . We put a lso
1 ... 3 ^ n , denote by
7
f -+ f
where
1112 = Z I?,
and
f(|) =
(2*)-n/2 / f(x)ei(x'l)dx. The Riesz operator
Ra
$ “1 o (|/| g1 )a o $ f
is
course, the middle factor stands for the multiplication by A = Za aa (x)Ra
a finite linear combination efficients in the space
B°° = B°°(R n ) of
where, of (l/|l|)a .
To
of Riesz operators, with co
C°°-functions all of whose partial
derivatives arebounded, we associate the symbol
a ( A) = Z aa (x) (|/| 11 )Q;,
which may be viewed as a function on R n x Sn_1 .
The symbol map
A -*> a (A)
is injective, and extends by continuity (in suitable topologies) to a 1-1 map of a certain class of operators of H“ ( R n )
H°°(R n )
to
(where
is the space of functions all of whose partial derivatives are
square integrable), to be called
B°° singular integral operators of order
zero, onto the space of functions on R n x Sn_1, are bounded.
A
B°° operator of order
A
is of order zero, and
of
1 +a ,
a
H°°( R n),
(a
A = iF_1 o (1
being the Laplacian) .
r
all of whose derivatives
is then a product o $
+ 1112) 1 The symbol
B = AAr ,
where
is the square root
cj (B)
is then
(A) •U | r . These operators, modulo operators of lower order, include the
the symbols is shown to have an intrinsic meaning,
and associates to every Ex
into
Fx ,
Intr (E, F) denoted class
A.
The operator
and if
cr^(A)
Ellp (E, F) . ch A
g € T*(X) - 0
whence a homomorphism
the symbol of
§3
A
(x e U)
cr^(A) : jt*E -► jc*F,
r
A e Ell (E, F), then, using
if it is in
a (A), the cohomology
are defined as in the case of
Moreover, it can be shown that
ker A, coker A
finite dimensional, so that
ia (A)
are true.
may now be viewed as a function on
Consequently
i&
S(X) ,
The set of such operators is
and the topological index
differential operators.
defined on
is elliptic of order
is an isomorphism.
If
a linear transformation of
are
is again defined, and that (ii), (iii) K(B(X), S(X)),
and the next steps of the proof consist in showing that it has the same pro perties as
i^, namely: 7.M u l t i p l i a a t i v i t y .
ellipticoperators on in the sense
X, Y
and
One X x Y
wants to know:
if
A, B, C,
such that [a(A)] • [o(B)J
of the pairing of §2, section 3, then ia (A) *
are =
[cr(C) ],
= ^a^ *
This will be proved by making use of a pairing of operators: (A,B) *-► A # B which is suggested by the construction underlying section 3 above. however, meets with some technical difficulties since eral an integro-differential operator.
A # B
This,
is not in gen
It will nevertheless be a limit of
such operators and this will allow one to keep track of symbols and indices. 8.
If
X
is even-dimensional, and
V
a bundle on
X,
we
define V*'
v) ■
•V
»
where the right hand side stands for the analytical index of any operator with symbol belonging to
V •
X
.
In fact, using a connection on
the operator
D Q mentioned in k, we may find a differential
that symbol.
One then has to prove that (a) to (d)
with
i^
replaced by
1&.
V
operator with
in section k are
This is fairly standard, except,
and
true
however, for
(c), which is one of the main parts of the whole proof. The index theorem for even-dimensional manifolds then follows from t, 5 and 8.
In order to extend it to odd-dimensional manifolds, we
need the following statement:
§3
Is OUTLINE OP PROOF 9. There exists on the circle
9
an elliptic operator
from the trivial bundle to the trivial bundle 3 with
i^E^)
EQ,
= i^(EQ ) 7^
0,
which, together with the multiplicativity properties 3, 7, allows one to reduce the odd-dimensional case to the even-dimensional one by multiplying with
Eq .
10. every element of
The analytical index is an integer by definition. K(B(X), S(X))
Since
is the class of a symbol of an elliptic
operator in the Seeley algebra, the main theorem for these operators implie the COROLLARY.
Let
X
be a manifold, and
Then the topological index of
a
a € K(B(X), S(X)).
is an integer.
As we shall see, this yields all the integrality theorems per taining to the Todd genus or the
A-genus.
§h . Appendix The order of exposition in the sequel does not coincide with the one adopted in the previous outline.
In order to orient the reader, we
make here some comments on the contents of the different chapters. Chapters II to X give some background material both for the to pological and the analytical parts.
In conformity with the purpose of this
seminar, the treatment of the latter one is practically self-contained, omitting only proofs of some quite standard facts, while much is taken for granted on the topological side.
Chapter II reviews briefly K-theory,
Chern characters, and shows that
K(X, Y) may be defined as the set of equi
valence classes of suitable sequences of vector bundles.
Chapter III de
scribes a method to compute the topological index when the bundles underly ing the differential operator and the tangent bundle to the base manifold are associated in a suitable way to a given principal bundle. Chapter IV introduces some basic material on differential oper ators: definition, symbols, the jet bundle exact sequence, adjoints, Green operators, some classical differential operators. Chapter V defines the differential operator leading to the in dex of a manifold and checks the index theorem in some special cases.
10
BOREL Chapters VI to X review some notions and results in functional
analysis:
in particular bounded operators On Banach spaces with finite di
mensional kernel and cokernel (to be called Fredholm operators) the Sobolev ir spaces H , the Sobolev inequality, and Rellich's theorem. The Seeley algebra is introduced axiomatically in XI, by means of five conditions, the existence proof being postponed to XVI. XII, XIII, XIV, and part of XV Seeley algebra: invariance of
Chapters
are devoted to the main properties of the
regularity properties of elliptic operators (XI), homotopy ia (XII), (this is (i) of §3; property (ii) is essentially
built in the axioms, so that its validity is really part of the existence proof), behaviour under Whitney sums (XIII) ; and the multiplicativity pro perty 7 above, which is dealt with with the help of the sixth condition im posed on the Seeley algebra (XIV) . Chapter XV proves that both conditions (a), (b), (d) of t, and that that (c) also is true for
i
s
i^ and i i^
satisfy the condition
also verifies (c).
The proof
is much harder and is given in XVII.
Chapter XVI proves the existence of the Seeley algebra of integro-differential operators, with the properties postulated in XI and XIV. The method finally adopted in these Notes is different from that of [5 ], after which the above summary was patterned. convenient for a self-contained exposition.
This turned out to be more Chapter XVI also contains the
construction of an elliptic scalar operator on the circle with both indices equal to -1, comments on the sign conventions made in defining
i& and i^,
and historical remarks on singular integral operators. Chapter XVIII is topological, and gives the proof of the unique ness theorem in Section 5 of 1, using the main results of Thom's cobordism theory. How the results of Chapters XI to XVIII yield a proof of the index theorem is briefly recapitulated in Chapter XIX, which also contains an extension to the non-orientable case, further remarks on the theorem, and some of the main applications (the Riemann-Roch theorem, the Hirzebruch in dex theorem, various integrality theorems of algebraic topology).
I: OUTLINE OP PROOF Appendix I is devoted to the index theorem on manifolds with boundary.
It gives the precise statement, and discusses the notions under
lying it; but the proof is only briefly sketched.
Finally, there is an ap
pendix by Weishu Shih, pertaining to Chapter III, which treats in a more general setting, characteristic classes and the topological index of differ ential operators associated to G-structures.
REFERENCES [1 ] M. F. Atiyah, "The index of elliptic operators on compact manifolds," Sem. Bourbaki, Mai 1963, Exp. 2 5 3 . [2 ] M. F. Atiyah and F. Hirzebruch, "Analytic cycles on complex manifolds," Topology 1 (1 9 6 2 ), pp. 2 5 -^6 . [3 ]
M. F. Atiyah and I. M. Singer, "The index of elliptic operators on compact manifolds," Bull. A. M. S. 69 (1 963), pp. ^22-I4-3 3.
[k]
R. T. Seeley, "Singular integrals on compact manifolds," Amer. J. M. 81 (1959), PP. 658-690.
[5 ] R. T. Seeley, Integro-differential operators on vector bundles, Trans. A.M.S. 117 (1965), pp. 167-20^.
CHAPTER II REVIEW OF K-THEORY Robert Solovay
§1. group, {|},
K(X),
Let
X
be a finite CW-complex.
as follows.
The generators will be the equivalence classes,
of complex vector bundles over
X.
allows the dimensions of the fibers of ness components of
X.) (E)
there is a relation,
One defines an abelian
(If £
X
is not connected, one
to differ on different connected
For each short exact
sequence of vectorbundles,
0
-
i*
-+ | -
{|} = U'} + {£"},
in
o
,
K(X) .
The tensor product of vector bundles makes tative ring:
{(•} • {rj} ={£ ® t]} . A mapping
the pull-back of vector bundles) a map Let dim({|})
x
be a point.
=dim £. If
X
Then
f:
K(X)
into a commu
Y -► X induces
f ': K(X) -* K(Y) : dim: K({x}) = Z
is a space with basepoint
(through
f ‘(Q)) = {f*|}.
is defined by x, we put
K(X)
=
kernel (K(X) — K((x))) . If
(X, Y)
is a finite CW-pair, we let
tained by collapsing for
X/Y.
(If
Y
Y = 4,
The relative group,
to a point X/Y
K(X, Y) ,
(Y)
X/Y be thespace
ob
which we take as the basepoint
is the disjoint union of is by definition
X
with a basepoint.)
K(X/Y) .
In [1 ], Atiyah and Hirzebruch have shown
that one can define
"cohomology groups" K t x , Y)
(i e Z 2)
which satisfy all the Eilenberg-Steenrod axioms (modified for except for the dimension axiom, and such that 13
Z 2-grading)
§1
SOLOVAY
^k
K°(X, Y) = K(X, Y) The exact sequence in cohomology reduces to an exact hexagon: K°(X, Y)
K°(X)
K'(Y)
K°(Y) 5
K (X)
K (X, Y)
(The construction of this cohomology theory uses the Bott periodicity theorem.) Moreover, the ring structure on commutative ring structure on
K(X, Y)
extends to an anti-
*
K (X, Y) :
K t x , Y) • KJ(X, Y) C Ki+J(X, Y)
§2. If
.
The Chern Character. X
is a topological space,
n j> 0 HJ(X, A) . Let
H**(X, A)
is the direct product
| be an n-dimensional complex vector bundle over
with total Chern class (rational coefficients) c( i) = 1 + c1(!) + ..*+ cR (g) Following Borel and Serre, one considers
c^(|)
as the
symmetric function in the indeterminants
x 1, ..., xn :
. i—
n
c( e) = II (1+x1)
.
i=1 Then by definition,
ch(|) e H**(X, Q)
is
i=i Of course, if
X
is finite dimensional,
H (X) = H *(X)
and
elementary
X
§3
CHAPTER II: The map
ch
K (X, Y) - H
B**(X, Y ; Q)
the
the
H* = IIi=o(2) ^
(X, Y; Q)
Z 2-grading: H
(where
15
extends to a ring homomorphism ch:
If we give
REVIEW OF K-THEORY
311(1
=tf®H
= Hj_=i (2)
>
tlien
cl1
compatible w ith
Z 2-gradings and commutes with the coboundary homomorphism, If
(X, Y)
is a finite CW-pair, then
K*(X, Y)
5.
is finitely
generated and ch 0 1q : K*(X, Y) 0 Q s H is an isomorphism. K (X)
* K (X, Y)
admits a natural graded module structure over
which is compatible, via
H*(X, Yj Q)
over
(X, Y; Q)
ch,
with the module structure of
H*(X; Q) .
The Bott periodicity theorem entails the following description of
K*(Sm ): The map
ch:
K* (Sm ) -►lf1(Sm ,Q)
the subgroup of integral elements,
§3.
is a monomorphism with image
if^S111, Z).
The Difference Construction. (After [2 ]).
3.1.
Suppose that
are vector bundles over
X
(X, Y)
is a finite CW-pair, that
E
P
and that cc:
E|Y s F |Y
is an isomorphism of the restrictions of
and
difference element d(E, F, a) € K(X, Y) is defined as follows: Let
A C X x I
be
X x {0 } U Y x I U X x (13
to
Y.
Then the
E
and
16
Thus
SOLOVAY X x I/A s S1(X/Y)
(S1(X)
§3
is the reduced suspension of
X.)
Put
AQ = X x {0} U Y x (o, 1) and
A 1 = X x {1} U Y x (0, 1)
and let
f^:
A^ -+ X
Define a bundle
% is
f*(E);
Let d(E,
|
over
on
F, a)
be the restriction of the projection map A
AQ n A 1
as follows: we identify
be theimage of
U)
on
AQ ,
f*(E)
|
with
is
X x I -* X.
fQ (F);
f*(E)
on
using
A 1,
a.
under the composition
K°(A) —►KtXx I, A) s K1(s’ (X/Y)) sK°(X, Y) 3.2.
The following lemma gives the principal properties of
a priori, an abelian semigroup with respect to Whitney sum.
is,
We put
L 1(X) = L 1(X, Of) . L 1(X, Y)
is a contravariant functor of the pair
(X, Y ) . We
define a natural transformation L 1(X, Y) - K(X, Y)
X:
as follows: then X
X(
E €
If
S 1(X, Y)
[E ]) = d(EQ , E 1, a).
represents the element
[E]
of
L 1(X, Y ) ,
Using Lemma 1of Section 3, one checks
is well-defined and additive.
Moreover, if E 0 - Ei
Y = 0,
x([El)
that
is just
.
The main result of this paragraph is THEOREM 2.
The map
L 1(X, Y) — K(X, Y)
X:
is an
isomorphism. PROOF. s K(X/Y, (Y)) ,
We first show that
we may assume that
is surjective.
X
Y
So let where
dim Ex = dim E^.
Let
a:
CY))
j
s
a € K(X, {x}) . The image of
Y)
is a point:
L,(X, Y)«-------- V X / Y ,
’ K(X, Y)
Since K(X,
K(X/Y, (Y)) a
in
K(X)
may be written as
E - E’
Ex ^ E^. Then
E = (E, E», a) lies in
t*N, a ) . f duced by
f.
Let f:
(Cf. ( h ) . )
:
(B(X), S(X)) — (B(V), S(V)) be the map
in-
Then we put Qf = f* 7
+ In Lecture II, K-theory was discussed only for the category of finite CW-pairs. Therefore, strictly speaking, we should work with N-universal classifying space, Bq n , and then let N tend to infinity. (The spaces Bq ^ can be finite CW-complexes.) However, we shall continue to speak only of
Bq .
III:
§2 2.1+. on
31
OPERATORS ASSOCIATED TO G-STRUCTURES
Let
a e K(B(X), S(X))
be associated to the G-structure
X. We are going to show eventually how to find an explicit formula
chCD^)
for various concrete a.
for
The following theorem is a first step
in
that direction. THEOREM 1 . Suppose that map
f:
Let X
X
2f t .
be a manifold of dimension
has a G-structure with classifying
a e K(B(X), S(X))
X -► Bq. Let
ciated to the G-structure. constructed from M,
N,
be asso
We suppose that
oc
and o' as discussed
is above.
We suppose also that the rational Euler class x(V*) e H*(Bq, Q)
is non-zero.
Then
ch(M) - ch(N) is divisible by (6 )
x(V ),
and
ch(D ) ■ f * * ( ( V “ V REMARK 1 : According to Borel,
sors
of zero.
(Cf. [1 ].)
Thus
ch(M)-ch(N) ) X(V*) '
.
the ring H*(Bq, Q)
has no divi
of (6 ) iswell-defined.
the righthand side
REMARK 2 : The right hand sideof (6 ) depends only and not
on
the isomorphism PROOF of Theorem 1 .
o.
Therefore, the same
Let
y = d(jt*M, **N, a).
on
M
and
is true of Since
chCD^) .
ch and
the Thom isomorphism are natural, ch(Da) = (-1 )^ cp* " 1 ch(a) = (-1)* cp* " 1 ch(f! 7 ) = (-1 )* f**(cp* " 1 ch(r)) Thus it suffices to show that X ( f ) U cp* " 1 ch(7 ) = ch(M) - ch(N)
(7) Let
i:
diagram
(B(V ) , Q) -+ (B(V ), S(V ))
.
be the inclusion map.
N
Consider the
32
SOLOVAY
§2
! K(B(V*), S(V*))--
^ ------------ -
ch .
K(B(V), 0)
ch I ► H**(B(V), 0; Q) I **
**
H~"(B(V"), S ( T ) ; Q)
*
fl Since
ch
v;
----------------------- “
H *(X) = II p > o H^(X).) (1 ) and (2 ) of 1 .0 . (8 ) Since
--------------------
is natural, the upper square is commutative.
taG, Q)
The commutativity of the lower square follows from
Since the diagram is commutative,
q>*~ (ch(r)) U x(V ) = (it
y = d(jt* M,
a
(We recall that
)“ (ch(i*(r))
N, cr) ,
jt*
i, . * ~ ~ i ’(7 ) = it M - jt N (by IILemma 1 (iv))
so the right hand side of (8 ) is just
ch M
- ch N .
The proof is complete.
Review of the Borel-Hirzebruch Formalism (after [2 ]).
§3. 3.1.
Let Tr
be a torus of dimension
Tr = S 1 x •* * x S 1 Then the classifying space,
B^,
(r
r: times)
is the Cartesian product of
infinite-dimensional complex projective space
r copies
of
CP(00) :
Bt = CP(oo) x ••• x CP(00)
.
Therefore we have H*(Bt , Q) sQtx,,
xp ]
and H**(Bt , Q) ^ Q(x 1 , ..., xp ]
(= the ring of
formal power series) where
x^ e H 2 (B,p, Q) , 3 .2 .
mal torus of sifying spaces
G.
Now let
G
The inclusion map of
be a compact Lie group, and let T
in
G
T
determines a map of clas
be a ma
III:
§3
33
OPERATORS ASSOCIATED TO G-STRUCTURES
BT
BG
(well-defined up to homotopy). The Weyl group,
W(G, T) ,
is the group of automorphisms of
which extend to inner automorphisms of homology ring,
G.
The Weyl group acts on the co
H*(Bt , Q ) ,(since it consists of automorphisms of
We can now state Borel*s description of
T
T).
H (Bq , Q ) .
THEOREM 2 . The map P* :
H* (Bq , Q) — H* (Bt , Q)
is injective.
The image consists of those elements
in
which are invariant under the action of
H*(Bt , Q)
the Weyl group
of
G. (I.e.,
g •y = y
y
for all
g € W(G, T ) .) We shall frequently identify H*(5j, Q)
under
with its image in
p.
3.3for any torus
*
H (Bq , Q)
Let T,
S1
be the group of reals
mod 1: S1 =R/Z.
One has
canonical isomorphisms p:
Hom(T, S1) — H1 (T, Z)
v:
Hom(T, S1) — H2^ ,
and Z)
which we shall treat in the future as identifications .
ed as follows: Let * = h a. To describe
They may be describ
o € H 1(S1, Z)
Then
u,
and
be the canonical generator. 1 we identify temporarily the groups S
n(h) U(1)
via the map s - e 2*13 A homomorphism on
B,p.
h:
T -*■ S 1
(s € S 1)
gives rise to a principal
U(1)-bundle,
We put u(h) = c/lh) 3.^.
M
Let
a G-module, and
Eq
G
be a compact Lie group,
T
a maximal torus of
the universal principal G-bundle over
G,
Bq . Recall
3^
SOLOVAY
that
M = Eq Xq M
§3
is the vector bundle over
corresponding to
M.
We
shall now recall the Borel-Hirzebruch description of the characteristic classes of
be an oriented 2 -dimension real vector bundle, Then if
y
is the Euler class of -y.
ch(A (V)) - ch(A_(V)) = e
1
y.
- e 1
V,
§6
III: PROOF.
By naturality, we may assume
R 2-bundle
ented
over
V
e1, e2
2 is the usual basis in R ,
*e l
= e 2>
*02
1- i ( e 1
a
A+
e2)
If y:
^9
is the universal ori
b sq(2) *
If
Thus a basis for
is
OPERATORS ASSOCIATED TO G-STRUCTURES
*1
=
1 + i(e1 a e2)
is
and
= -e i 9
e1
a
and
e 2 ,
* (e 1 a
e1 + ie2 ;
e 2)
= 1
.
a basis for
A_
ie2 .
SO(2) -► S*
is
thestandard weight,
then
ysends the
matrix ^ cos 2*0
-sin 2*0 v
sin 2*0 into 0.
Then the
(Here
0
Let
[X]
€ S 1.) denote the1-dimensional
SO ( 2) -module s
[e^ + ie2]
cos 2*0
have weights
[i
+ i(e1
+ y
a e2)]
subspace spanned have weight
respectively.
0,
by x.
and the modules
The Lemma now results from
(1 0 ) .
REFERENCES [1]
A. Borel, Sur la cohomologie des espaces fibres principaux et des espaces homogbnes de groupes de Lie compacts, Ann. of Math., 57(1953), pp. 1 1 5 -2 0 7 .
[2]
A. Borel and F. Hirzebruch, Characteristic classes and homogeneous spaces
I, Amer. J. Math., 8 0 (1 9 5 8 ), pp. t58-538.
CHAPTER IV DIFFERENTIAL OPERATORS ON VECTOR BUNDLES Richard S. Palais § 1. M ary,
will denote a paracompact
C°°(M,
C)
the subring gory of
Notation C°° manifold, possibly with bound
the ring of complex valued
of real valued functions.
C°°complex vector
ally be denoted by
V(M)
bundles over
g, n, £
C°°
M.
Objects of
F*:
of
g . We denote by
C°° cross-sections of C°°( r\)
C°°(g)
is of course a
g
) . C“ (g)
linearly into
V(M)
and to
The fiber of
|
into
t\
g
F € Hom( g, tj)
F*(f)(x) = F(f(x))
at
x
which for each
CQ (g)
V(M)
modules, that 0“ (ti)
is a
F*
and maps
the induced map
x € M
maps
C°°(g)
the subspace of
C” (g)
(g, i\)
then
into L
C“ (g)
and
L( g
F*
consisting C” (g)
T -*• T
con
dM.
C” (g) maps
L( g, n) tj )
V(M) x V(M)
whose fiber
of linear maps of
T(x) = T|gx e L(g, tj)
We
are subC“ (g)
into
V(M) which
gx
into
at
x
is
tj .
If
and the bundle structure of
is a natural equivalence of the functors
which we regard as an identification, i.e., 51
to
L(g, r\)
T e C°°(L(g, r\)) .
L(g, t|) “is uniquely characterized by the property that Then
tx
C~( n) .
the functor from
the bundle
the complex vector space T e Hom(g, t\)
module, that
is a module homomorphism, and that
We denote by assigns to
C°°(M, C)
to
the vector space
sisting of sections whose supports are compact and disjoint from C°°(g)
will
(an element of Hom( g, r\)
will denote the subspace of
of sections with compact support and
note that
R)
will usu
C°° the additive functor from
g
defined by
C°° map of
C°°(M,
we will generally use the same symbol to
the category of complex vector spaces which assigns to C°°(g)
M,
will denote the additive cate
denote a vector bundle and its total space. be denoted by
functions on
Horn
Horn = C°°L .
and
C°°L
52
PALAIS If
¥^ e V(M) bold
and
C°°(M, ¥)
If
and we will use the sym-
If
g e V(M)
is a C°°
its dual bun
real vector bundle then
denote the dual real bundle. IfM = M 1 x •• • x
a bundle
7
vector space then
M x¥
interchangeably.
= L(-g, C M) . Similarly if
g*
will
is afinite dimensional complex
will denote the product bundle
C°°(¥m )
dle is 7*
¥
§1
over
M
f1 €
g1 e V(M^)
and
whose fiber at
x = (x-,..., xv)
K is
I
f 1 0 ••• 0 fk ,
we define
g1 0 ••• 0 gk
then
0
g.L
X1
a function on
is
••*0 g^
Xk
M, by
(f1 0 ••• 0 fk) (x) = f 1 (x1) 0 ••• 0 fk (xk) . The bundle structure of v i v ••• 0 g is characterized by the property that f 0 •••0 f e C& (|1 0 ••• 0 gk)
wherever
f^ € C°°(gi).
Ifg1 € V(M) i = 1 , .. ., k g1 0 •••
0 gk = A* (g1 0 *•* 0 gk),
agonal map,
x -*■ (x, ..., x) .
g1 0 ••• 0 gk = then
.
if
ments of
where
0 kg
lift fixed by all such
0 kg
into itself
. The set of ele
ir is denoted by
Sk (g),
where the sum is over all
(1 , ..., k) . ¥e will also use of 0 kg 0 t] onto
•x M
•
the
k—
( k) There is a canonical projection S^ ’ e
Z *,
namely
•
by
{1 , ..., k}
it of
jt(v1 0 ••• 0 v^) = v ^ . ^ 0 • • • 0
Hom(0 kg, Sk (g)),
0 id
M x
M -►
a:
jr is a permutation of the set
g.
tions of
••♦ 0 gk e V(M)
g1 = g (i = 1 , ..., k) we write
If
symmetric tensor product of
S ^
define g1 0
we
ir induces a bundle morphism, also denoted by
characterized by
g1 0
S ^
to denote
k! permuta
theprojection
Sk (g) 0 tj.
¥e note that the above tensor and symmetric tensor products are of course
also defined for Let 7
be
a
non-negative integer k bundle over
M
C°° real bundles.
C°° real bundle over
M.
For each
we defineLk (7 ,
to
be aC°°
is the space L k (7
whose fiber at x
(over R) maps of
(rx )k
into
gx
having the obvious bundle structure. bundle whose fiber at maps of iL
(7 v)k into
X
x g .
X
is If
7 x -► gx , defined byf(v)
g)
(we define ¥e define
the space SX
X
= f(v,..., v)
, g) of
Lk (7 , l)
and
complex vector k
L°(rx , lx ) = ix ),
Lg(7 x , !x )
f € Lk (7 „, g„)
g e V(M)
linear and
to be the sub
of symmetric k-linear
there is an associated map i.e.,
f = f°A
where
is
§1 a
:
IV: k rx -► (?x)
DIFFERENTIAL OPERATORS ON VECTOR BUNDLES
is
diagonal map.
identified, it is the space of degree
k
then
7X
of
, ix )
into
The image of the map
P ^ ( r x , Sx ) | .
[If
e^
f f
1=1
= k,
xiei)
-*■ ( I
..., en
map is called polarization.
a = (c^ , .
P ^ ( 7 , |),
are the elements of
L^( 7 „, £„) S X X
P^ ( 7 X
with
It follows that
L g(?,
the bundle whose fiber at
characterized by the property that corresponds to the map
&x ) •
(w1 , ..., w^) -* v 1 (w1) ••• v^-(w^-)e
identified with the sub-bundle
of
£ g
then
corresponding element of
L^ ( 7 X, lx ),
of
Lk (7 , |) .
(v 0 ••• 0 v) 0 e
the inverse
Lk (7 , |),
0 e in ®^7X ® l x
Sk (7*) §> t
e
iY),
X
0 ^7 * 0 | with
Under this identification the sub-bundle
and
£x ] . The map
is
(v1 0 **• 0 v^)
Lg(7, |)
a )
i)is naturally equiva
x
There is a canonical identification of
v € 7*
7x
is a basis for
X 1 1 ••• xn n)^a
a
ja
and the
is an isomorphism of
lent to
is easily
of homogeneous polynomial maps
where the sum is over all n-tuple of non-negative integers
f -*• f
v
consists of all maps of the form
(I
such that
53
is in
? x > lx ) •
in
0k7* §> I
is
In particular if
^ ( 7 *) 0 |
and the
namely
(w1, ..., wk) — v(w1) •**v(wk)e will be denoted by
S^(v) 0 e.
Note that we how havethree
naturally equivalent bundles
Sk (r*) ® I *» Lg(r, I) «p (k)(r, I) In general
we will find it convenient to "prefer"
Lg(r, I)
(some choice
has to be made) however, it will beconvenient to pass back and forth be tween these various ways of looking at the same thing and we shall often do so
tacitly. We denote the tangent bundle of
T(M)*,
the cotangent bundle of
bundle
T*(M)
pairs
(v, x)
by
T(M),its dual bundle
T*(M) andwe denoteby
T'(M)
the
with the zero section removed; so elements of
T'(M)
are
where
x e M
and
M, by
M
v
is a non-zero linear functional on the
^
PAIjAIS
tangent space to jt(v, x) = x. (v, x) T ’(M) number
M
Then
is
x, and the projection
**(g)
so
such that k
at
§1 «:
T'(M) -► M
is a vector bundle over
Hom(ir*|,
cr(v, x)
tj)
T'(M)
whose fiber at
consists of functions
is a linear map of
g„ X
a
For each real *)£ * Hom(ir g, * t]) by X
Smblk (g, n) = (a e Hom(ir*g, *%)|a(pv, x) = pkcr(v, x) If
k
with domain
t\'.
into
Smblk (g, i\) of
we define a linear sub space
is given by
p > 0}
if
is a non-negative integer then we define a subspace
It will be obvious that the trivializations so obtained are smoothly related Namely each into
W.
d111?^ = 0
f e C°°(|)
gives rise to a
By Taylor's Theorem if m = 0, 1, ... k.
p e N
C°° map then
f = t ° f ° cp-1 j, (f)p = 0 k
(See the proof of Lemma 1 .)
P =
if and
satisfies
vanishes in a neighborhood of
g(p) = 0 f(p) = e.
p
then clearly
j, (f) = 0 k p
Hence
Ji_(f*| O) K
0
T(p) e L( Jk (|
PROOF. ^ k ^ fi^p^
base
for Jk (|)
5ij
for x
near
M then
, tip).
is a base for
P* Let
such that
T € C°°(L(Jk (|), n)) = Hom(Jk (|), r\)
Necessity is clear.
p.Let
M
A necessary and sufficient
x »-► T(x) jk (f) be inC°°(Tj)
near
jk (f)p M>
Jk (|) | 0 ** Jk (t \ 6 ) .
T be a function on
condition that is that
is open in
induces an isomorphism
P
THEOREM 3. Let
and let
i(Sk (v) 0 e) =
satisfies
f € C°°(|)
is
Proved above.
THEOREM 2. If
so that
Jk(f)p ^ Jk_-, (f)p
Given
jk( ^ p
T (x )Jk (fj_)x = £ aij(x)g^(x)
p e M
5131(1 5161:106
hp € C°°(Jk U)*) §1 > •••> Ss
for each
so that
f e C°°(g).
choose ^ k ^ fi ^ hp (x)
be a local basis for
near
p.
Then
{f^}
in
C°°(g)
is a local (jk(fj)x ) =
i\
near
p
§3
61
IV: DIFFERENTIAL OPERATORS ON VECTOR BUNDLES
T(x) = 2 a1j(x)(h1 (x) ® 8j(x) then
is
L(Jk (|), n)
C“
near
and hence
near
T
p.
If
x*-» T(x)Jk (f1)x
is a smooth section of
is in
C ° ° ( t|)
Jk (t)* ® n
p. q.e.d.
§3.
Differential operators and their symbols DEFINITION. g to
from
A k-th order differential operator
t) is a linear map
such that, for each Df(p) = 0 C°°U)
into
g
p € M,
= 0
tip vanishes on
implies
f -+ Df(p)
(i.e., the linear map
the set of from
D: C°°(|) -*• C°°(t])
Zp(|)).
of
We denote
k-th order differential operators
to
t]
by
Diffk (£, ti) .
We now record some trivial consequences of the definition. First we note that a k-th order differential operator is local, i.e., if D e Diff^U, ti) and if and
Dfg
Diffk (|, C°°(M, C)
0.
agree on ti)
f 1, fg e C°°(|) agree on an open set Secondly
is a subspace of
then
Hom^,(C°°( |),C°°(ti))
it is a sub
D e Diffk (£, ti)and
will be called a universal k-th
order differential operator for
£
k-th order differential operator there is a unique
|
and in fact
A k-th order differential operator
U e Diffk (|, w)
of
Df1
T*D e Diffk (£, £) •
DEFINITION.
If
then
Diffk-1(|, ti) C_Diffk (£, tj) . Thirdly,
module. Finally we note that if
T e Hom(Ti, (;)
0
if for each
D e Diffk (|, ti)
T e Hom(w, ti) such that
U f€ Diffk (g, w ’)
D = T*U.
is a second universal k-th order operator
then by the usual uniqueness argument for universal objects there is
a unique isomorphism
T:
THEOREM 1.
w ~ w*
such that
U* = TU .
jk : C°°(|) - C ” (Jk (|))
k-th order differential operator for PROOF.
That
jk
is a universal g.
is a k-th order differential operator is just
the tautologous statement that if
J^^p = 0
then
Jf*)
= 0.
Given
62
PALAIS
D € Diff*k U , n) ZpU) into
the map
f -► Df(p)
§3
of
C°°(0
hence there is a unique linear map
into
vanishes on
of
C°°(|)/Z^(l) = Jk (l)p
T(p)
n
such that Df(p) = T(p)jlr(f)_. By Theorem 3 of Section 2 P K P T € Hom(Jk (|), ti) and by definition D = T*L . K q.e.d. COROLLARY 1 . The map ofHom(Jk (|), ti)
T -*■ T*jk
is an isomorphism
Diffk (S, n)
with
(as C°°(M, C)
modules). COROLLARY 2.
T -*■ T*
Hom( i , t})
is an isomorphism of
with DiffQ (|, n)• PROOF.
Since
Z°(|) = Cf € C°°(|) |f(p) = 0} there is a
identification of
J°(t)
with
|
through which jbecomes
natural
the identity
map. Since
differential operators are local it follows that if
D € Diff^d, r\) ential
@ is open in
and
operator denoted
= (Df) | O for each
by
f € C°°(i) .
M
there is a unique k-th order differ
%| 0
D|a n
existsit is e € ix#
i(Sk (v) ® e),
where
i:
and
f
D(^y(g-g(x))kf)(x)
on
If
of Section 2
T'(M),
D e Diffk (| ,
we define a function
called the symbol of
it is
is the ’’inclusion" map
It follows that if
D e Diffk (£, ^)
such
^y(g-g(x))kf
is well-defined.
DEFINITION. a^(D)
f e C°°(|)
the k-jet of
Lk (T(M), |) -► Jk (|)
of the Jet Bundle Exact Sequence.
unique.
and infinitely many
f(x) = e, however given any suchg
Smbl£(S, n) - 0
Recall that since
M
.
is paracompact the additive cate
is semi-simple (i.e., every exact sequence of vector bundles
splits) hence any additive functor from
category is exact.
V(M) to any additive
In particular applying the contravariant functor
L( , t|) : V(M) -► V(M)
to the Jet Bundle Exact Sequence (Theorem 1 of Sec
tion 2) 0 -*■ Lk (T(M) , |)-i* Jk (|) - Jk_1 (I) - 0 we get an exact sequence in
V(M); namely .
*
0 - L(Jk_1 (|), ti) — L( Jk (|), T,)-n. L(Lk (T(M), |), T,) - 0 If now we apply the functor
C°°
.
to this sequence and recall that C°°L = Horn,
then using Corollary 1 above we get the exact sequence o where of
rk
—
(|, n ) - D i f f k U ,
= C°°(i ). Let
Hom(Jk (|), t|)such that
choose
g
e C°°(M, R) with
by Theorem l of Section 2
n) -L-. Hom(Lk (T(M), i ) , n) -
D e Diffk (£, ti)
and let
D = T*jk - Given dgx = v
and
v e
f € C°°(|)
T
0
bethe unique
T(M)X
and e e
with
f(x) = e.
i(Sk (v) 0 e) = Jk (^r( g-g(x))kf)
hence
element | Then
PALAIS
§3
rk (D)(Sk (v) ® e) = (C“ (i*)T)(Sk(v) ® e) = (i*T(x))(Sk (v) ® e) = T*(i(Sk (v) ® e)) = D(-jjr(g-g(x))kf) (x) = ok(D)(v, x)e If we now compose
.
with the isomorphism *k : Hom(Lk (T(M), |), r,)) ~ Smbl£( I, t,)
of Section 1 we get the exact sequence 0 - Diffk_1 (I, n) - Diffk(I, n) — —
^ — ► Snibl£( s, T)) - 0
.
Moreover, by the above calculation ck (D)(v, x)e = rk (D)(Sk (v) ® e) = xkrk(D)(v, x)e
Let us now paraphrase the information contained in the exactness of the Symbol Exact Sequence. symbol of some
First every element of
k-th order differential operator from
Smbl^U, n) £
to
is the
r\ and second
ly two such operators have the same symbol if and only if they differ by a differential operator of order An element
a
of
is a linear isomorphism of £x D € Diff^fl, t\)
k-1. is called e l l i p t i c
Smbl^( |, n) tj
a(v, x)
for all
(v, x) €
T»(M), and
is called an elliptic k-th order
operator
if and only if
its symbol is elliptic.
with
if
We note that we can now assert the existence of an
elliptic operator corresponding to any elliptic symbol. Caution: a
If
D
is an elliptic
k-th order operator then
D
is
(k+l)st order operator but n o t an elliptic (k+l)st order operator, in
fact
*k+1(D) = 0 We now investigate what differential operators and their symbols
look like in coordinates. R n
and let
linear map
V D:
and
W
N
be either R n
or a closed half space in
be finite dimensional complex vector spaces.
C°°(N, V)
ential operator with
Let
C°°
A
C°°(N, W) is called a k-th order partialdiffer coefficients if it is of the form
§3
IV:
DIFFERENTIAL OPERATORS ON VECTOR BUNDLES
Df ‘
where
for each
into
n-multi-index
L(V, W) .
Daf(p)
belongs to where
X 1 9 '**> xn
a with
which implies
Diff^U, ti)
y € N,
Ac P ° f
I
I or | < k
|a| v ) be
Then
°k P)(v,
y) (e) = jjr D(gkf)(y)
= jjr
]>
Aa (y) (D°gk) (y)e
I« I < k Clearly
Dagk (y) = 0
if
|a| < k
and
(Dagk)(y) = k!va
if
|a| = k,
hence : 0 ~ R n
such that if
positive
C°° function
p
on R n
is a char1: such that
any continuous, complex valued function
0.
compact subset of
Jf
is a Radon measure
M
there is a strictly
/ fdii = /(f°q>_1 (x))p(x)dx on
M
having as support a M
gives
M
and also one on
for the corresponding integrals.
dM
and we shall write
We note that if
f
M
with compact support then
We will denote by whose fiber at
x
Met (|)
|f| = 0 = > f = 0.
the complex fiber
bundle over
the obvious bundle structure. C°°(Met(£))
( , )^
are hermitian bundles over (which
namely if
An hermitian structure for
£
£ ,
with
is an element
and by an hermitian bundle we mean a complex vector bundle
with a specific choice
L( tj,|)
M
is the space of positive definite, hermitian symmetric,
conjugate bilinear forms (i.e., Hilbert space inner products) on
I
is a
m
continuous function on
of
for
p(x) = (det g ^ ( x ) ) 2 . We shall assume as
given a fixed such measure on and
f
M
For example, a Riemannian structure on
rise to such a measure, with
Jf M
69
M
of hermitian structure.
there is a natural map
*
If
T eL( £x , tix )
then
T*€ L( tjx , ix )
isdefined
r\
and
L(|, t\) into
of
is a real bundle isomorphism but is conjugate
£
linear)
by
(Te, e 1)
*
(e, T*e1)^.
This gives rise to a conjugate linear map, also denoted by * of Smb 1^.(1, r\) into Smbl^(r], |) defined by a (v, x) =■ a(v, x) . If a
|
is an hermitian bundle then for
C°° complex valued function
Clearly,
(f, g) -► (f, g) ^
and the support of of
f
and
g.
< f, g >
and
(f, g)g(x) = (f(x), g(x))^.
f
and
g
belong to the space
= JM (f, g)
is well defined. 00
C_(|) c
(f, g) Clearly
and the associated norm
C~(|) of
has compact < , >
If
I
and
D e Dif fk (|, r\)
tj are hermitian bundles over then
D * € D i f f k (ii, |)
is a -
< f, f >?
||f||^ -
DEFINITION. M
by
having compact support, then
prehilbert space structure for will be denoted by
M
is included in the intersection of the supports
In particular if |
on
we define
is hermitian symmetric and conjugate bilinear,
(f, g)
C°° cross-section of support and
(f, g) ^
f, g € C°°(£)
*
will be
5
70
PALAIS called a formal adjoint for < f, D*g >£
whenever
g € Cq (t])
PROOF.
Let
then
if
< Df, g >^ =
f e C” (g) and
S € Differ], i)
LEMMA 1 . If all
D
and if
g € C” (t]). Sg =
0 for
S = 0.
h € C°°(g) . We must show
Sh = 0,
and since e
is continuous it will suffice to prove that
Sh(p) = 0 if p
cp be a
subset ofM - dM
C°° function with support a compact
identically one in a neighborhood of and
since g
agrees
with
h
p
and let
g = cph.
in a neighborhood of
p,
dM.
Sh Let
which is
Then
Sg = 0
Sh(p) =Sg(p)
= 0. q.e.d.
LEMMA 2.
If
D € Diff^(|, t}) has a formal
adjoint
then it has a unique formal adjoint. PROOF T
Suppose
T
In particular taking
Sg
D
such that for each
n 0^
gives
||Sg|L =0hence 1
Diffk (g, tj)
D* e Differ)! O a , I I &a ) • D|
= 0
> g - < f, T'g > g = g = < f, Tg
for
T ’ are adjointsfor
- T T.By Lemma 1 it will suffice to show that
Now if
of
and
- < Df,
q.e.d.
{0
and that
is an open cover
hasa formal adjoint
Then clearly
D*|0 a n 0
D* | 0 a n 0
= 0 .
Sg = 0.
a Da = D| 0 a
so by Lemma 2
g
is a formal adjoint
= D* | 6 a n0 ^ •By
Corollary 3
of Theorem 1 of Section 3 there is a uniqueD € Differ], g) such that * # *#■ D |0 a = Da for all a and clearly D is a formal adjoint for D (by an obvious partition of unity argument) hence LEMMA 3.
If
D e Diffk (g, t\)
cp: 0 ~ ® C_Hn adjoint then LEMMA Ij-. space
and if for
of an atlas for M, D|
0
each chart hasa formal
D has a formal adjoint.
Let
M
be either
{x e R n |xn < 0} . Then
formal adjoint
D*
and
Rn
or theclosedhalf
D e Diffk (g, ti) = (-1)kcjk (D)* .
has a
IV: PROOF. space, the proof M = Rn.
We shall consider only the case where being essentially the same but
Let the measure on
that on dM = R n_1 a
are positive
M
be given by
t| = M x V 2
inner products by
T f:
where V^ ( , )^.
V 2 -*■ V 1 ,so
are If
alittle simpler when
L
= J^f(x)p(x)dx and let
We can assume
| = M x V1
finite dimensional Hilbert spaces with
T:
V1
V 2 is linear we denote its adjoint
(Tv, w) 2 = (v, T'w)^ The hermitian structure for
of positive operators on las
is a half
/^M f = / n _1 f(x)a(x)dx where p and R ^ C°° functions on M and R n_ respectively and dx are
r\are uniquely given by
and
M
be given by
the respective elements of Lebesgue measure. and
71
DIFFERENTIAL OPERATORS ON VECTOR BUNDLES
C°° maps
V1
(e, e')g = (A(x)e, e ,)1
and
A
V2
if
and
e, e ’ € nX .
Then if
by
T* = A(x) _1T'B(x) . Let
D:
of
e, e 1 € lx
and
^
into the spaces
(e, e ’)
its adjoint
T*:
C~(M, V 1) — C°°(M, V 2)
Df =
M
respectively according to the formu
T e L(e X, Xr\ )
if
B
|
= (B(x)e, e ’)2
X -►X|
is given
be given by
Ca (D“f)
a, where the sum is over all n-multi-indices is a
C°° map of If
M
into
s ^m b M a or \ i i
D g =
^
where
(-l)Wi A _1D >'(pC^Bg)
x i =
G o k_,f, jk-ig)
dM Then if so
f
and
g
have supports disjoint from
< Df, g >^ = < f, D*g >£
hence
D*
is a formal adjoint for
Moreover the characteristic polynomial for
(-Ok
G(jk_.jf, ^k-1^ = 0
dM,
D
D.
is
£ A(x)‘1C^(x)B(x)va = (-1)k £ C*(x)va |a|=k la I=k
- 0 - l)k
*
I
a |=k which since that
Z|a |_k Ca (x)va
is the characteristic polynomial of
D
proves
*) = (-l)kak (D)*. q.e.d. We note the following Corollary of Proof.
Grk (|, t])
be the bundle over
of conjugate bilinear maps of
In the case dM = R n”1 x
M = {x e R n |xn < o},
whose fiber at into
x C.
let
is the space Then for each
§5
IV:
D e Diffjj.Cl, n)
DIFFERENTIAL OPERATORS ON VECTOR BUNDLES
there is a cross
< Df, g > n - < f,
whenever
f € C^(|)
and
section G €C (Gr^l, 11)
D*g > 5 =
73
such
that
G(jk. 1 f, J ^ g )
J dM
g € C^(rj) .
The main result of this section, which now follows trivially from Lemmas 2 , 3 and k and Theorem 3 of Section 3 THEOREM. M
If
then each
adjoint
|
r\ are hermitian bundles over
and
D e Diff^d, r\)
D*e Diff^d,
We note that
D
DEFINITION.
If 5
is
= D
has a unique formal
|) . Moreover and if D €
ak (D*) = (-1 )kak (D)*.
Differ), £)
then
(DD)* = d V .
D € Diff2k(|, |) 2k
isan hermitian bundle over
is called strongly elliptic of order (-1 )^ ^ ( D ) (v, x)
if and only if
operator on for all
COROLLARY.
If
in
is injective
| )
for all
D € Diff^d, t\)
elliptic of order
2k
for all
is a positive
( ( - 1 )ka 2 k (D) (v, x)e, e) | > 0
!x (i.e.,
e ^ 0
M then
(v, x) € T !(M) . then
D*D
is strongly
if and only if ^(D) (v,
x) :Sx -► r\x
(v, x) e T ’(M). Hence if
Tj have the same fiber dimension then elliptic if and only if
DD
is
D
£ and
is k-th order
2 k-th order strongly
elliptic.
§ 5.
Green* s forms In this section we again assume given strictly positive smooth
measures on M.
If
M
and on
D € Diff^d, t))
f € C“ (g)
and
dM
and that
|
and
then the formula
r\ arehermitian bundles over
< Df, g
= < f, D*g
g € C^d) have supports disjoint from
generally we should be able to express
< Df, g
£M
suggests that more
- < f, D*g
as an in
tegral
of
some expression involving the jetsof f
theory
of
boundary value problems requires that thisbemade precise and we
shall now do so.
andg
where
over
dM.
The
PALAIS DEFINITION.
§5
We define a bundle
whose fiber at
x
I, ti)
over
SM
is the space of conjugate bilinear
maps of
jk~1(£)x x
into
section
G e C°°(Grk (i, t}))
order Green's form of
g
A C°° cross
will be called a t\
and
and will be called
D e Diffk (g, i\)
a Green's form for
k-th
if
< Df, g > n - < f, D*g > t = J G(Jk_,(f), j ^ g ) ) 9M for all f € C” (|) REMARK.
Let
is a Green's form for
and
g £ C“ (n) .
M = Cx £ R n |xn < o},
D
Diff2(g,
e
also
so
t\)
| = ti = M x C . Then if isG !,
G't^f, j,g) = G(J,f, J,g) + This
G
defined by
(fg)
.
shows thatin general Green's forms for operators are not
uniquely
determined. REMARK.
If
valued functions on
G e C°°(Grk (g, ti))
M
and
is a bilinear form in
similarly for let
{cp^}
{ ^ a },
j^.-j^g)*
be a locally finite
Ga e C°°(Grk (g| O that
G(jk_,(f), jk_i(g))
D e Diffk(|, t]) and
tj| ,
for
has a Green's form. D € Diffk (t, i\)
THEOREM.
Every
has a Green's form.
PROOF.
Immediate from the Lemma and the Corollary ofProof of
Lemma k of the preceding section.
§6.
Some classical differential operators a.
The Lie derivative Let
g be a complex vector bundle over
consists of all complex valued tensors at metry type.
Given a vector field
be the Lie derivative operator.
X
x
on
M
whose fiber at x
of a fixed variance and sym M
define
C°°(|) -► C°°(|)
Then bya basic property of
^x
to
we have
a>x (gf) = (xg)f + gdxf if
g € C°°(M, R)
that = v
f € C°°(|) .
and
^x e Diff^S, g) . Also if
It follows from Theorem 5 of Section 3 g € C°°(M, R)
vanishes at
x
and
dg^.
then tf-l^xMv, x)f(x) = dx (gf)(x) = v(X)f(x)
i.e.,
a1(^x )(v, x)
is scalar multiplication by
is elliptic if and only if b.
M
v(X).
is one dimensional and
X
It follows that dx never vanishes.
The exterior derivative Let
| = a (T*(M) ) ® C = ® L q
complex valued differential forms and let
A1 (T*(M)) ® C be the bundle of d:
C°°(g) -► C°°(|)
as usual de
note exterior derivation. If
g
is a C°° real
valued function on M and
d(gca) = dg a cd + g By Theorem5 of Section
3 again
g(x)
then
= o and
dg^^. = v
A
cde
then
da)
we see that d e Diff1(g, |).
d(go))x = v a cdx
C°°(|)
Also if
76
§6
PALAIS a1(d)(v, x) = va 9
which proves that c.
left exterior multiplication by v.
The codifferential and Laplacian Maintaining the notation of (b) assume now in addition that
is an n-dimensional Riemannian manifold.
The Riemannian structure for
defines strictly positive smooth measures on hermitian structure on to
Af = 0.
hence
df = 0 ; Clearly if
and on
6 = d
*
dM, and also a
t
= (d+5 )2 .
a
Recall that
a
2
Note that 5
and a = d5 + 5d.
f € C°°(i) is called 0
it is called co-closed if 5f = f
M
5 , a first order operator and the Laplacian
a second order operator, by
closed if
M
| . We define two differential operators from
the codifferential
(d*)2 = (d2)* = 0
M
and harmonic if
is closed and co-closed it is harmonic.
If
f,
g ^ cjd) < Af, g >£ = < d5f, g >£ + < 5df, g >£ =< 5f, Sg >| Af = 0
In particular if
then taking
g = f gives
5f =
+ < df, dg > s . df = 0 ,i.e.,
conversely a harmonic form is closed and co-closed if its support is com pact and disjoint from
dM.
In particular on a compact manifold without
boundary every harmonic form is closed and co-closed. To give a more precise description of M
is oriented and we let
sure (i.e.,
forT (M) ).
Var i£t £s D i f f e r e n t i a b l e s .
and
we assume that
a
a> be the n-form describing the Riemannian mea
cox= e1 a... aeR
normal base
s
where
(e^
..., en )
is any oriented ortho-
For details of what we sketch below Recall that the inner product in
pletely described by the properties that the
see de Rham'e |
A^(T*(M)x ) ® C
is com
are mutually
orthogonal and (v,a...a
vp ,
V / ...A
Wp) 5 =
£
eCnMv,,
W)[(1))
•••
( v p> w f l ( p ) )
Jt where the sum is over all permutations of (1 ,...,
p ) . Equivalently, if
(e., ..., e )
is an orthonormal basis for T
then
(1 < p < n,
i1< i2 e = y tfA * g in M Let (-1)p
on
w be the automorphism of
Ap (T*(M)) ® C
|
which is multiplication by
so d(f a g) = df a g + w(f) a dg
We note that
[w(f) a d*g]R = -[f a d*w(g)ln hence if f, g [d(f a* g)]n =
[df a*
f, g € C*(s)
Stokes'
y [f ** g]n _.,
= y [d(f A* g)]n = < df, g > 5
dMdefining the
= G(e, e !)vx for x €
is the
(n-l)-form on
dM and
e, e 1 e J°(l)x = £x
then
by G
5
has degree -1, hence since
d
has degree
+1,
0.
Given =
7
d.
We note that has degree
and moreoverthat if
Riemannian measure and if we define G € C°°(Gr.j(£, |))
reducible modules, elements of which are called positive and negative spinors
92
§1 0
PALAIS
respectively.
Moreover
have
vS* C S”
gard
Sn
and
vS~ C S*
ble.
If
u € Spin(n)
and S^
A^S^
C S* for
so in particular since R n C A “ v e R n . Since Spin(n) C A*
we
we can re
as Spin(n) modules, and as such they are still irreduci v € Rn
and
then we note that for
s € sn
u(vs) =
(uvu_1)(us) = (p(u)v)(us). Now let structure for from
S0(n)
M
M
be an oriented Riemannian manifold.
is meant a reduction of the structural group of
We are given an indexed set
{ 6>x) is an open covering of
fields defined on
(9
and
M
and
..., X^)}
X^, ..., X^
are
C°° vector
forming an oriented crthonormal basis at each point.
And we are also given maps forx e Ox
T(M)
to Spin(n). Expressed in the language of coordinate bundles
this means the following. where
By a spin
n (9^ -► Spin(n) satisfying
g^:
gX|i(x) g^v (x) = g^v (x)
p(gXtJ(x))1j = (Xj(x),
x € &x n
for
X^(x)) or equivalently
such a spin structure we let
Sn (M)
y , and finally
Xj =
. Given
be the associated bundle with fiber Sn .
Thus the total space of
Sn (M) is the disjoint union of the sets x n O .
(x) Since
S*
Sn (M)
splits canonically as the Whitney sum of sub-bundles
S”(M) . if
and
S~
are Spin(n) invariant submodules of
There is a bilinear
w € T(M)X ,
x e 0^
pairing
and w =
Sn
T(M) x Sn (M) Sn (M) v^X^(x)
it follows that S*(M)
and
given as
follows:
then
w(x, s, \i) = (x, vs, |i) . The identity
u(vs) = (p(u)v)(us),
noted above, with
u = g^(x)
this definition is consistent with the identifications. it follows that we have induced bilinear maps T(M)
x S“(M) — S*(M). We now define
o(v,
x)s = vs
in maps
Smbl^(S+(M), S'(M)) Sn (M)x
C S”
R nS*
T(M) x S*(M) -+ S”(M)
a e Smbl^(Sn (M), Sn (M))
(where we have identified
mannian structure). By
Since
shows
T*(M)
with T(M)
and
by via the Rie
the above we can consider a by restriction to be or Smbl^(S‘(M), S*(M)). If
isomorphically onto itself, in fact
the defining relations of the Clifford algebras.
v^O
a(v, x) Thus-
a
2
then
±- e -> dSx(w ) = with
v
xi it follows that Now let
f(x) = s
®x
1 V^ for
we can define
T(M) . x
D maps
C°°(S*)
into
C°°(S”)
and let
g e C°°(M, R) with g(x) = 0
v) •
Then if
s € Sn (M)x (w^, v) s
and
f e C°°(Sn (M))
hence
a1(d)(v, x)s = D(gf)(x) = ^ ( w 1 , v)wis = vs = a ( v , x)s i D hasthe required symbol
S*(M)
and vice versa.
T(M)
and
That
we have
restricts to covariant derivatives for
then v^(gf)(x) = dg(w^)s =
which proves that
and
associated with the Riemannian
w • • •> ¥n
v 1, ••*> vn
so that
S*(M)
but there seems to be no particular advantage involved
in taking it) . Let T(M)X
V
S (M)
and hence is elliptic.
CHAPTER V ANALYTICAL INDICES OF SOME CONCRETE OPERATORS Robert M. Solovay In § 1 , we review Hodge theory.
This will be used in §§2 , 3 to
check the index theorem in a number of concrete cases. a role in the proof of the index theorem in general.
This check will play In §^, we show that
the topological index of a differential operator on an odd dimensional mani fold is zero. § 1 . Review of Hodge theory 1 .0 .
mension 5 , and
n. *
Let X
be a closed oriented Riemannian manifold of di
We shall recall the principalproperties of the operators introduced 1.1.
in Chapter IV, § 6 . = Ak (T*(X)) 0 C
Let
differential k-forms over
X.
a Hermitian structure on X.
(a,
p) j 1 .2 .
(1 ) If of
be the bundle of complex valued
The Riemannian structure on
X
gives rise to
| , and to a strictly positive smooth measure on
In this way, we get a Hermitian inner product on
note by
d,
C^U^)
which we de
(a, 0 e C°°(|k)). There is a bundle isomorphism *: Ak (T*(X)) ® C a An~k (T*(X)) ® C
V
is the fibre of T*(X) over
x,
at the point
x
of
X, then, on the fibre
(1 )is the complexification of theisomorphism *!
discussed in III, § 6 . will also be denoted by
Ak (V) S An_k(V)
The induced isomorphism of *
C“ (S^)
with
C°°(|n_P)
Recall the following properties of the map 95
*:
96
SOLOVAY
(2)
(a,
(3) If (10
p) = / a a * p f
( a> p e C°°( |k)) ;
*2 - (-l)k(n-k)
on
(5)
= *P, 1.3.
Then
d
=0.
C°°( |k) .
n is even, (3) simplifies to *2 = (-l)k
p
§1
Let
d:
on
C°°(?k),
i.e.,
*
(dim X
even).
is real.
C°°( £k) -► C°°( |k+1)
be the exterior derivative.
DeRham's theorem asserts that the cohomology groups of H*(X; C),
with complex coefficients,
X
are canonically isomorphic to the
cohomology groups of the complex C°°(|0) - i * Let- 6:
... c“ (ik) - L * c°°(!k+1) — C“ (|k+1) — C"(5k)
...
be the adjoint of
d.
. When
X
is
even-dimensional, we have (6)
5 = -*d* Let
is formally
| = A(T# (X)) ® C = E£=q
self-adjoint, (i.e.,
A = (d+5)2 = ds + 5d. Then C°°(|lc).
We have,
(7) If
dcp = 0;
0, we say that
this:
e very
C"(|) - C“ (|) We put
is homogeneous of degree zero: a : C^fS^) “**
0,
X K.
has a
112
PAIAIS PROOF.
B y Asooli's theorem it will suffice to prove that the
V
implies |4(x)| < SupC ||k|| | k e K}, and since X U(x) - H ( j ) | < ||x - y|| the elements of B * are even equicontinuous on X X. q.e.d. clearly
H€ B *
THEOREM 12.
If
T e K(X, Y)
then
PROOF. {J^n )
X
and
Y
are Banach spaces and
T* € K(Y*, X*).
By the lemma if
)
is a sequence in _
has a subsequence which converges uniformly on
or equivalently (since
TB^-
B *
then
y
hence on
TB^,
T*4n (x) = ^n (Tx))T*£n
has a subsequence which con* verges uniformly on B^., i.e., which converges in X . But this means pre * * cisely that T B * is relatively compact in X . Y q.e.d. THEOREM 13.
(F. Riesz)
k e K(X, X)
and
Let
T = I - k.
1)
ker T = T”1(o)
2)
T(X)
X
be a Banach space,
Then
is finitedimensional;
is closed in
X
and coker T = X/T(X)
is finite
dimensional. PROOF. k(b)
Let
= b - T(b) = b
V = ker T.
for
relatively compact, hence
b e By,
Then
if
i.e.,
By
is the unit ball of
k(By) = By.
But,
k(By)
is locally compact and by Theorem k
V
V is
dim V o ||Tw|| > e ||w|| for
will be a Banach space, (wn ) x €
and we can assume wn
But But
dim (X/T(X)) < then
T € K(X, Y) .
is a bounded subset of
T(X)
and hence by
Theorem 3 it is relatively compact. THEOREM 15. G
If
H
is a Hilbert space then the group
of invertible elements of
subset of PROOF.
L(H, H)
is a connected
L(H, H) . T € G
If
then
T*T
is a strictlypositive operator
on
H and hence has a strictly positive square root
so
U*U = A"1T*TA“1 = A ' V a '1 = I
and
U
A. Let
is unitary.
U = TA_1
Since the set of
strictly positive operators is clearly a convex subset of
G
arc in Gfrom
U. Let 1/i log
A
to
I, hence an arc from
be the inverse of the function Then
T
= UA
t -► e ^ of
[0, 2*)
to onto
there is an
{z e C|
1 /i log is a bounded Borel function on the spectrum of
U
|z| = 1 }.
so by the
B = 1 /i log U is a bounded self 2it it U = J e dP^ is the spectraldecompo-
functional calculus for normal operators adjoint operator (equivalently if sitlon of
U
then
is an arc in
G
B = J2n tdPt) .
from
I
to
Then
U = elB
and
t — eltB 0 < t < 1
U. q.e.d.
LEMMA.
If
H
is a Hilbert space and
is self adjoint then either eigenvalue of PROOF. ||Txn || -► IITII^. and then
Since
||y|| = IMI^.
IMI*,
or “M o ,
is an
T.
Choose a sequence (xn ) T
T € K(H, H)
on the unit sphere of
is completelycontinuous we can assume
H
with
Txn -► y
By Schwartz's inequality
||Ty|| = 11m ||T2xn || > llmfT2^ ,
xn ) = 11m ||Txn ||2 = ||T||2
and HT2y|| • ||y|| > (T2y, y) = ||Ty||2 > ||T|£ = ||T||2 ||y||2 > ||T2y|| • ||y|| so
2
2
(T y, y) = ||T y|| • ||y||
the scalar being and we define
and therefore
2
T y
is a scalar multiple of
(T2y, y) /(y, y) = ||T||^/||T||2 = ||T||2 . We can assume
x = y
+ ||T||^1Ty.
If
x = 0
then
y,
T/0
y is an eigenvector of
T
11 k
PALAIS
belonging to
the eigenvalue
vector of
belonging to the eigenvalue
T
— | | T w h i l e if
x / o
then
x
is an eigen
||TH^. q.e.d.
THEOREM 1 6 .
Let
T e K(H, H)
be self adjoint.
H
be a Hilbert space and let
E X(T) = {x € X|Tx = Xx) .
Then the E X(T)
mutually orthogonal subspaces of their Hilbert space direct sum.
®|x|>s set
H
is
Moreover, the
In fact, if
is
let
are
H, and
E^(T), with the possible exception of finite dimensional.
characterized by the properties that the identity map of
sp(A)
PROOF.
corresponds'to
A and that
f(A) = f(A)*.
A short, elegant, and completely elementary proof, due
to John von Neumann, can be found in Appendix I of S. Lang's "introduction to Differentiable Manifolds" (Interscience, 1 9 6 2 ). structs the isomorphism
f -► f(A)
f(A)
If f = g + ih
are self adjoint.
the definition fying
where
f
g,
h
xn ~*"°*
are real valued then
is clearly the unique extension satis
then by the corollary of Theorem 1 6 ,
A € K(H, H)
(fn)
of eigenvectors of
sp(A) = (0 } U (\n )
Clearly,
f(A)(fn ) = f(xn )fn
A
is completely continuous then
Recall that if 6 £ C
is open and
into a complex Banach space X, then
Given
z e C
pz :
[0 , 0
then
of
z > 0into
Re
PROOF. x
equalities
log t -► -oo
is given
f
is
If
with
6 -► X
is a map of
called holomorphic in exists in
Re z > 0
p z(t) = ez log t p_z
f(A)
f(o) = 0 .
f:
l l m 2^ z (z-zQ)~1 (f(z) - f(zQ))
zQ € O
has
Afn =xn ? n >
A,
and the mapf -► f(A)
is completely continuous if and only if
for each
H
and it follows that
COROLLARY.If
since
and the resulting
f (A) = f (A) *.
a completeorthonormal basis
by
for real valued
f(A) = g(A) + ih(A)
If
and
Lang in fact only con
V
6
if
X.
define t> 0
if
is continuous and sat-
= PzP z t P- = Pz -Moreover, if z -► p z |[0 , L]
map
C([o, L]).
z = x + iy,
as t -► 0
is a holomorphic
x > 0
then
it follows that
p_,_. z+ z = p„p„. z z andp- = zp_
are obvious. z
|pz(t) | = |ex log t | and pz
is continuous.
If for
Re z
The
> 0
we
11 6
PAIAIS
define
qz(t) = (log t)pz(t)
that by L ’Hospital's rule n
for
t > o
limt_^0
qz(o) = 0
and
(log t)n ex
^ = 0
a positive integer) just as above it follows that
moreover, an elementary calculation gives (ez lo®
h”1 (e*1 lo^ ^ - 1 - h log t)
h(-l-(log t)2 e^z+p^
log
then (recalling
qz
k-1
for
“ pz ^ ^
h_1(p2+h(t) - pz(t))
” qz ^
=
which by Taylor's theorem equals
0 < p < 1,
'*'og ^
and
is continuous,
so putting
x = Re z > 0,
lh_1(Pz+h(t) - p z(t)) - qz(t)| < |h|(l(log t)2 e ^x'lh ^ log *0 (log t)2
x > 0
is bounded on
[o, L]
tends uniformly on
and since
it follows that as h — 0,
[0, L]
to
qz(t).
q.e.d. THEOREM 18.
Let
A € L(H, H)
be a positive operator.
sp(A) £ [o, oo) and
pz
H
be a Hilbert space and let
and hence if
is as in the lemma then
a well-defined element of 1)
z € C
z —*■A z
(i.e.,
with Re z > 0
A z = P Z(A)
Re z > 0
A z+z’ = A ZA Z 1);
A z = (Az)*;
3)
If A
is completely continuous and
then
Az
If A
is strictly positive then so is
for
into
and is a semi-group homomorphism
2)
k)
is
L(H, H) . Moreover,
is a holomorphic map of
L(H, H)
Then
Re z > 0
is completely continuous;
x > 0.
Moreover
z -► A z
Ax
extends to a
(not necessarily continuous) function of Re z > 0 into
L(H, H)
whose value at a point iy of the
imaginary axis is a unitary operator such a way that
z -► A zv
Re z > 0 into H
for each
Ax+iy = A lyAx PROOF. if
f
If
is non-negative
if
sp(A)
in
is a continuous map of v e H.
Moreover,
x > 0.
f e C(sp(A)) on
Aiy,
then
f(A)
is positive if and only
(cf. S. Lang, l o o . o i t .
page 118) hence
V I : REVIEW OF FUNCTIONAL ANALYSIS in particular if we take
11 7
f = identity map of sp(A) we get
if and only ifsp(A) C [o,
«>) .
The existence of
A
is positive
A z as well
as properties
1), 2), and 3) now follow from the lemma together with Theorem 17 and its corollary.
If
t > 0
0 < t < 1
then
A = A^A1
More generally if again
A^
t > 0
V = im (A)
A
hence if then
v -► H
by
= ker(A1“iy).
H
(since
ker(A1“iy) = 0
im(A^)
If
is strictly positive so is
A^ = (At/n)n
A ^ v = A 1+^oo
But
A
is clearly positive.
where
n = [t] + 1
.
and so
A is.
is strictly positive, and hence injective,
is dense in
hence the closure of
so
A^ = (A*^2)* ( A ^ 2)
is strictly positive if Now if
A^-Y;
then
V = (ker A*)1 = (ker A)1).
if v =
Acd.
then
Define
Then im(A^) = im(A1+^ ) ,
is the orthogonal complement of ker(A1+^ ) *
A, and hence
A 2,
and
is dense in
im(Aiy)
is injective and H.
A 2 = (A1+iy) (A1”iy)
Now if
v =
Acd
e V
HA^vll2 = (A1+i^co, A 1+i^co)
hence
A^
=
(A1"1*
=
( A c d , Acd)
v =
with
Acd
cd)
e V.
In general given
||u-v || < e.
u €H
=
= ||v||2
A 1 + z cd
and
( A 2cd,
cd)
,
extends uniquely to a unitary map of lim A zv = lim z iy z iy
if
A 1+ 1 ^ c d ,
H onto itself.
= A 1+ e >
i y cD
Note that
= Aiyv
0we can find
v € V
Then
||Azu - A^ull < ||Az(u-v) || + ||Azv - A^vll + A ^ v - u ) || < (1 + ||Az||Je + ||Azv - A ^ V H Since
I|Az ||m 0
Re z > 0 it continues
Re z ! > 0, and in particular
Ax+iy =
Ai3rAx .
q.e.d.
CHAPTER VII FREDHDIM OPERATORS Richard S. Palais DEFINITION. element atorfrom
T
If
X
of
X
and
Y
L(X, Y)
to
Y
are Banach spaces, an
is called a Fredholm oper
( o r simply an F-operator) if
1)
ker T = T -1(o)
is finite dimensional;
2)
coker T = Y/T(X)
is finite dimensional.
We denote the set of F-operators from F(X, Y) ind:
S - dim coker S = dim ker T - dim coker T = ind T. q.e.d.
COROLLARY.
If
T € F(X, Y)
ind(T + k) = ind(T) .
and
k € K(X, Y) then
VII s FREDHDLM OPERATORS PROOF. and since
123
By Corollary 1 of Theorem 2,
K(X, Y)
T + K(X, Y) C F(X, Y)
is connected (in fact a subspace of
L(X, Y)) the corol
lary follows. LEMMA.
Let X
and
with
ind(T) > 0
dim
V = ind(T) .
component of
find
asub space
and extend Then tP
P
W
as
such that
L(X,
as S.
T1:
P
of
by letting
in
K(X, Y)
W
from
Theorem
zero on
T
there is a unitary map Let T 1 = SU.
U
to
W 1. hence
by Corollary 1 of
ind(S) = dim ker(S)
T1
we can
onto T(X)1
P be
and since U
of
X
Then ker T 1 = U ”1S“1(o)
is in the same component of
But by [VI, Theorem 15] there is a path
from Corollary 3 of
is
for* 0 < t < 1 ,
L(X, Y) as
by Theorem
X
X -► Y
dim T(X)1 = ind(T) > 0
Y)
U(V) = ker S.
of automorphisms of
to
T such that
is surjective so
so itwill suffice to prove that
F(X, Y)
with
T 1in the same
and an isomorphism
to an element of
S
X
ker(T.j) = V.
of ker T
ind(S) = ind(T) = dim V onto itself
exists
Since dim ker (T) -
Clearly,
T £ F(X, Y)
be a subspace of
Then there
is in the same component of
Theorem 2.
G
V
has finite rank and hence is
S = T+ P
= V
Hilbert spaces.,
and let
F(X, Y)
surjective and PROOF.
Y be
I-^. Since
2 that SU(t)
U(t)
in the group
G C_ F(X, X)
is a path in
it follows
F(X, Y) from
T1
S. q.e.d. THEOREM 5.
If
two elements
X
and
S and
Y
T
index, then they are in PROOF. *
and
S
have non negative index.
T
ind(T ) = -ind(T)
the same components of and S1
S1 and
group from
G A
F(X, Y)
are surjective and T 1 map
V1
of
have the same
F(X, Y)
F(X, Y) .
topologically onto
it suffices to consider the cage where
as
By the lemma choose T
and S
Y,
T1
respectively,
ker T 1 = ker S1 = V.
isomorphic ally onto
of automorphism of
F(X, Y)
the same component of
Since T -► T* maps *
F(Y , X ) and
are Hilbert spaces and if
and S1 such that
V1
lying in
T^
Then by [VI, Theorem 1 ]
hence
A = T-1S1
is in the
V 1. By [VI, Theorem 15] there is a path
to the identity map of
in
G C F(VL, V 1) .
A^
The orthogonal
121*-
PAIAIS
projection
P
of
of Theorem 2
,
T-jA^P
clearly
X
T 1AP = S1
onto
V1
is clearly in
is a path in
and
component of F(X, Y) ,
T.,P = T1,
F(X, Y) hence
hence so also are
F(X, V 1), from
S1 S
T1AP
and
T1
and
so by Corollary 3 to
^P.
But
are in the same
T. q.e.d.
Let
X
be a Hilbert space of infinite dimension.
Corollary of Theorem 11 ] A = L(X, X)/K(X, X)
K(X, X)
is a closed ideal in
is a Banach algebra and we let
the canonical homomorphism.
Let
the identity component
of
U.
corollary of Theorem ^
there is awell-defined map iffd:
ind
° it = ind,
-1
so
X
and let
ind(it(T)) = -1
T
ind
A
U
Z
is a group homomorphism.
ker(iffd) = UQ . Let be an isomorphism of
V X
and since -1 generates
hence
L(X, X) -► A
F(X, X) = jt_1(U).
ind „ o - UQ - U ----- ► Z -► o is exact.
L(X, X) ,
be the group of units of
By Theorem 2,
and by Theorem 3
Theorem 5 it follows that space of
U
it:
By [VI,
be
and
U0
By the such that From
be a one-dimensional sub onto V 1.
Then
Z the sequence
ind(T) =
CHAPTER VIII CHAINS OP HILBERTIAN SPACES Richard S. Palais
§1 .
Chains If
X
is a Banach space and we change the norm in
equivalent norm, then the norm on valence.
In other words, if
of a Banach space then
X
X
X
is the underlying topological vector space
is a well defined topological vector space. X
then
x -► < x,
>
H.
X).
If H
write
X
H
is the with
X
*
*
is a hilbert space
we shall always regard H
*
= H.
this
It follows
is an hilbertian space then there will be no confusion if we
X as the anti-dualrather
canonical isomorphism as an identification, so that that if
If
is an isomorphism of
(this is precisely the reason we chose to define than the dual of
We
hilbertian if it is the underly
ing topological vector space of a hilbert space, H
to an
is also changed only to within equi
shall call a topological vector space
inner product in
X
>
for4(x)
when
x € X
and
ofhilbertian spaces wecan identify =
if x e X
DEFINITION.
and
Let o f
real numbers and
H € X . Moreover, by the reX**
=
X
and then
ft e X*.
denote either the
o f +
with
{k e o/ |k > 0 } .
hilbertian spaces is a set
{H^}
integers or A chain of
of hilbertian
spaces indexed by
o f+
1)
If
then the underlying vector space
of
H
k > £ > 0
space of
is a linear H^
such that:
subspace ofthe underlying vector
and the inclusion map
-► H^
tinuous; 2)
H°° =
3)
H°
Hk
is dense in each
is a hilbert
space 125
(so
Hk ; (H0)* = *H°) .
is con
126
PAIAIS If whenever
k >
> o
I
§1
the inclusion
Hk -*■
is not only
continuous but even completely continuous then we shall say that the chain {H^}
satisfies the Rellich Condition or is a Rellich chain. If
=Z
spaces, while if o/ = R
we call {Hi we call
(Hk)
a discrete chain of hilbertian a continuous chain of hilbertian
spaces.
H"k = (Hk)* ity,
If
{H }
if
k > 0 . If k = 0
while if If
define
is a chain of hilbertian spaces then we define
k >
> 0
H
3 (j^ k) : ^
-i
“*■H
H”k = (Hk)*
then by 2 )
Hk
_4 )
is dense in 30
J*(£,m) = J*(4, m) = ^(4,k)^(k,m)
the inclusion map being
of
H
H 00( 0 0 > k
ous inclusion
ft
j(_k _ji y
then
Hk
f€
and if
- 00)
k): H -*• H
the pairing of k >
>
k
H
If
< g, f > 0 defined if
B°
e
f € H
and
H“°° and
then
00
>
k
>
S.
*
j^_^ ^
k) • Denote, temporarily,
g
g €
c
k < ,
We topologize
H‘"k and
H“^ then
f >k
< g, f >k =
< g, f >Q
is then well
H°° as the inverse limit
H“°° as the inductive limit
with the finest topology such that each inclusion Then it follows that H~°° with
< , >Q
•
extends the pairing
3jjn Hk, i.e., with the least fine topology such that each inclusion is continuous and we topologize
=
we have a continu-
> -°o
and
g € H~ . We note that
f e H00.
8X1(1 i1:
onto a dense linear subspace k we can regard each H as a vec
>a = < 3(.k, . a ) e, f >k = 0 > -JI
< , >k . Iff e Hk
It follows in particular that for each of
m > k
as a dense linear subspace of
. If it should so happen that also
< ^ i >1 = 0
is injective it follows
and
and H_k by
= ,
j^
H
d-eIInes “the inclusion of —°° k . Thus if we define H = Ik H
tor subspace of
j(_k
^(-m,-£) = ^ (-m, -k) ^ (-k, -S .)
k)
o )j(o
n , hence if we
H_k. Moreover, if
follows that we may consistently regard
—ft
is dense in
is injective, and since
by reflexiv-
by 3) of the definition.
to be the inclusion map then
-k
that the image of
H“k,
H"k = (Hk)*
we then have
we still have
k) : Hk
*
k < 0
H°° -► Hk
lim H”k, i.e.,
H -► H~°° is continuous.
is a continuous conjugate bilinear pairing of
(in fact it is easily seen that
and
H“°° are locally
§1
VIII: CHAINS OF HILBERTIAN SPACES
convex topological vector spaces and that
H -00
127
is the antidual of
H00,
i.e., all continuous conjugate linear functionals on H°°) .
DEFINITION.
If
(H^)
and
(Hg)
are chains of hilbertian
spaces a morphism ( - * • (Hg) is a collection {T^) k k that T^: H 1 -► H2 is for each k e o9 a continuous linear map, and
Tk = T^|H^
set of such morphisms by
k > I . We denote the
if
Hom(CH^), {Hg}).
We note that the operation Hom({H^), (H^))
a vector space.
such
+ PT^}
Also if
H({H^})
makes
and clearly with
this composition law the class of discrete (continuous) chains of hilbertian spaces forms an additive category (not abelian; there are kernels but not cokernels). If
{Tk) € Hom({H^}, (H^))
T^|H^
if
k > i it follows that
since
Toof
= L J£ f
of
H^
namely
into Tk
H^.
d
e for all
Since
H^
and
Tkf = k, 7
f e H^° then since
T^f= T^f
T oof € H^, d 7
is the unique continuous extension
DEFINITION.
If
(H^)
and
is well defined.
i.e.,7
is dense in each
H^, of
T00
we define
each map
k € Z( R)
Tk : H^ - K*~r
Remark. that and
Tk
T
T:
the corresponding
T^
H^°
k > ft,
it follows that
is clearly
THEOREM 1 .
If
CH^j
H^ -► H^
such that for
is dense in
T, and
is continuous if r = 0
to be the vector
.
We note that since
In particular, if
r e
extends to a continuous linear
is uniquely determined by
H2”rH^_r
is a linear map
(Hg} are twodiscrete
OPr ({H^}, {H^f})
space of all linear maps
Also
Tto determines each Tk , v to a map of H 1 into
(continuous) chains of hilbertian spaces and Z( R)
Tk =
T and
since
H^
it follows
H^ -*• H^ is continuous
it follows that Tk = T^|H^. (Tk) €Hom({H^), (H^)).
we have: (h |)
are two discrete
(continuous) chains of hilbertian spaces then the
Since
128
PALAIS map
(T^) — T^
of
defined above is a linear isomorphism
Hom( {H^}, {Hg))
DEFINITION.
§1
If
with
(H^J
0PQ ({H^), {Hg}) .
and
(H^l
are two discrete r e
(continuous) chains of hilbertian spaces and Z ( R)L(H^, H2 r)
is a Banach space (in the topology
of uniform convergence on bounded sets) and we take the weakest topology on maps T -►
of
(i.e. a net
[Ta] in
T
0Pp
0Pr
L(H^, H2”r)
the net
formly to Remark.
T^
is continuous
OPr ({H^}, (H^))
if and only if for each
B C
such that each of the
k
converges to
and each bounded set
{T^} in L(H^, H2”r) on
converges uni
B) .
Note in particular that by Theorem 1 this topologizes
Hom({Hb, (Hg}) . THEOREM 2.
Let
{H^}
and
bertian spaces and let OPa((H^}, {Hg})
{Hg}
be chains of hil-
r, s € o f with
is a subspace of
tion
{H^} and
or
[H2)
Then
OPr ({H^), (H^))
and the inclusion map is continuous. either
r > s.
Moreover, if
satisfies the Rellich condi
T e OPg({H^}, (H^))
then for each
k € of
the unique continuous extension of T to a linear map k k—r of H 1 into H2 is in fact completely continuous. PROOF. and
j 1: H2“s
continuous and
H2~s_t j
condition while
j'T^
proves
t = r - s = H2~r
j*
sion is immediate. B
(H^)
is completely continuous if T € O P s ({H^}, {h |})
let
T
j: j
-► and
j1
are
satisfies the Rellich
(H2)
satisfies the Rel
Tk_t :
be continuous extensions of
give continuous extensions of
T € 0Pp .
let
be the inclusion map, so
- H2"t_s
= H|"r
T.
Then either T^-t*^ k k—r to linear maps H 1 -+ H2 which
Since the composition of a continuous and a completely
continuous map is completely continuous
and let
k e of
and given
is completely continuous if
lich condition. If k k—s and T^: H 1 -*• H2 or
Let
Finally, suppose
be a bounded set in
H^.
[VI, Theorem 11] the final conclu(ex')
T v ’ -► T Then
in
0Pg
-*■ T^
and let
k e of
uniformly on
B,
§1
129
VIII: CHAINS OF HILBERTIAN SPACES
hence,
j
_► j'Tk
uniformly on
B
w h i c h proves continuity.
q.e.d. THE O R E M 3.
Let
{Hk }, (Hk ), and {Hk }
be discrete
(continuous) chains of hilbertian spaces and let r, s € o f . map of
Then the map
PROOF.
Obvious. Let
{Hk }
bertian spaces and
T:
linear map
-►
if
into
(Hk }).
DEFINITION.
T
is a continuous
(Hk ) ) x OPr ({Hk ), (Hk ))
O P s ({Hg},
O P r + s ((H^},
(S, T) -► ST
S:
and
CHk }
be chains of hil-
a linear map.
A
is called a transpose of
< Tf, g >Q = < f, Sg
for all f t
>Q
,
g €
Remark. < f, (S-S')g >0 = ° dense in
H°
If
ST
is a second transpose for
for all
it follows that
f €
and
S = S ’,
it is unique and will be denoted by THEOREM it-.
Let
(Hk )
T^. and
(Tt)k :
- H^_r
of
k.
hence
(T_k+r)*:
Hk - H g k
for some
By assumption
T_k + p : then
f f
< Tf, g > Q= < T 0f, g > 0 - < f, so
f
has
of.
k €
for each
k € Q = < f,
on
-
(Tt )k =
k
f f
€ H^,
g e H^
then
hence
and
(Tt )_r g > 0
T. Since for each
h£
independent
k,
f e
r estriction of the continuous linear ma p
the final conclusion is immediate.
T
(H^)) and
for each
is i n fact the transpose of
n ition the
Then
the adjoint map of
H" k+r -*• H ‘k .
then
exists
belongs
T^ € OP
T_ i+r = T _ k + r |H~£+r
is a linear m ap of
T
Ttt: = T.
-►
(t ) k |H^1 is a well-defined function
If f e f
is
is
be two chains of
r € of .
k > £ so
(Hk )
to
uous linear m a p
(TtjjlHg
We note that
T:
PROOF.
If
hence since
hilbertian spaces and suppose OPr ({H^), (H^))
then
i.e., if a transpose for
a transpose and moreover
Hk_r.
g € H^j
T
= < f, f g f
(T_ k+r) * '■
>Q
is b y d efi H k -*• H k_r
130
PALAIS THEOREM 5.
Let
(H^)
and
§1 {H2)
be discrete (con
tinuous) chains of hilbertian spaces and suppose T:
H^
-► H^
is a linear map with a transpose
r, s € o f
Let with T^:
admits a continuous linear extension
-► H2”r
and that for each
k > r - s,
T^
(Tt)k : h £
- H^‘r .
if
Tk
k < s,
PROOF.
If
we
and, since
have H“
with
Then T € 0Pr ((H*}, {H^})
is the adjoint of then
H^
-k + r > r - s
into
so
>hence
H2”r .
and
(Tt)_k+r*
(Tt)_k+r>:H2k+P
is a continuous linear map of g e H^
k e of
admits a continuous linear extension
k < s
tinuous linear extension
k e of
and suppose that for each
k > s, T
T^.
If
T"k
admits a con
Tk = (Tt)*^+r
f €
then
forall
< T^f, g >Q = < f , (Tt)_k+pg >Q = < f, ^ g >0 = Q
is dense in
H2,
Tkf = Tf
so
Tk
extends
T. q.e.d.
COROLLARY 1 . If map
Tk :
-*■
T
extends to a continuous linear
H2
for all
to a continuous linear map k> 0 T
then
T e OP0 ({H^), {Hg}),
defines amorphism
PROOF.
k > 0 and T1" extends t k k (T )k : H2 -► Ej for all
of {H^1}into
hence (Theorem 1) {H^} .
Take r = s = 0.
COROLLARY 2.
If
{H^} and
of hilbertian spaces and if T
(H2)
are discrete chains
extends to a
continuous
linear map
Tk :
and if
extends to a continuous linear map
T^
(T^) k :
H^ -*• H2_1 for each integer
for ea°h integer
k > 1
k > 1
then
T € OP 1 ({H^}‘, {H|}) . PROOF. equivalent to
Take r = s = 1
k > 1since o¥
andnotethat
k > r - s = 0
is
= Z.
q.e.d.
§2
VIII:
§2.
CHAINS OF HTLBERTIM SPACES
Quadratic interpolation of pairs of hilbert
131
spaces.
Given a continuous chain of hilbertian spaces, by restricting the indexing set to the integers we get a discrete chain, and this corre spondence is clearly functorial from the category to the category
of continuous chains K In this and the next section we
of discrete chains.
C
* Cr
will construct a sectioning (i.e. right inverse) functor
called-
quadratic interpolation. Let (Hq , H 1) -► H 1
H^q ^
denote the category whose objects are pairs
of hilbert spaces such that
is a continuous linear map of
whose morphisms TQ : Hq -► note by
T:
HQ C_ H 1 HQ
and the inclusion map
onto a dense subspace of
H 1,
and
(HQ , H 1) -► (H^, H.J) are all continuous linear maps
which extend to continuous linear maps
H^q ^
HQ
:
H1
Hj . We de
the analogous category formed with hilbertian rather than
hilbert spaces and note that there is an obvious
"weakening of structure"
or "forgetful" functor
Similarly, we define
F:
H^Q ^
— ► H^Q
to be the category whose objects are indexed sets indexed by inclusion of
[0,1],
such that if
Hg
map
Tq :
HQ -►
T^_: H^
then
is a continuous linear map of
H^, and whose morphisms
maps
0 < s < t < 1
{H^}
T:
{H^}
which for all
H|.
{H£)
of hilbert spaces, H g C_
onto a dense subspace
Hj-0 1 ]
forgetful functorF: If
i^: * *
i^:
tive.
H q -*■ Ht, *
-► H q
{H^}
is
to be the analogous category
object of
anti-dual of
an objectof
H*
which
{H^} .
If
*
i^
with its image under we also denote by TQ=
T:
T*: H.J* -► H*.
{H^} -► {H|.} TQ ,
Thus if we defineD(T) =T*
is a contravariant functor. we define
or
E(HQ , E^) =
Similarly, if for
, HQ)
inclusion
itsadjoint, the restriction map
has dense range because
-*■ H.J. is the continuous extension of to
Hj-Q 1 j then since the
]has denserange,
is injective, and
Hj-Q 1 j
have an obvi
Hj-0 -j] — * H j-q 1j.
t e [0,1
If we identify
linear
extend to a continuous linear
formed with hilbertian rather than hilbert spaces and again we ous
and the
are allcontinuous
t e [o, 1]
Again we define
Hg
H^Q ^
i*
i^
then
is another
ECH^}
and
is a morphism and then T * : then
Hj *
call the T^:
H* extends
D:Hj-Q 1 j ^ — ► H^Q 1 j
(HQ ,H 1)
(where we identify
is injec
H*
an object of H^Q with itsimage under
^
PALAIS
132 the adjoint of the inclusion -+ (H^, H.j)
we define
i:
HQ -+
D(T) = T*
then
§2 ) and for a morphism D: H^Q
^
of H^Q
^ let
T:
— ► H^Q ^
(HQ ,
)
is a contra-
variant functor, the anti-dual functor. object (HQ , E^)
Given an inner product of norm
C)
if
H^.
Since theinclusionHQ -► ^
x eHQ
then yH- (x, y) 1
is a unique elementAx y € Hq .
of
Clearly,A:
HQ
HQ -*■ HQ
< ||x]l1 ||y||1 < C2 ||x||0 ||y||0 Since
DEFINITION.
If
is continuous
is an element
such that
(x, y) 1
is linear and since
if
x ^ 0,
(HQ , H , )
A
(HQ , H 1)
If
(x, y)1 = (Ax, y)Q
0 < t < 1
then
| (Ax, y)Q | = |(x, y) 1 | and
product on
HQ
= ||At/2x||0
by
A is continuous.
H {0
is the unique HQ -+ HQ
A:
for all
x, y € HQ .
[VI, Theorem 18, h ) ] is
A^
strictly positive continuous operator on (x, y)^ = (A^x, y)Q
for all
is a strictly positive operator.
bounded, strictly positive operator such that
hence there
= (Ax, y)Q
is a n object of
the defining operator of
(say with
of HQ ,
HAH^ < C2
it follows that
(Ax, x)Q = (x, x)1 > 0
( ,denote the
HQ
hence
is a positive definite inner
and we denote the completion of HQ t 1/2 relative to the corresponding norm l|x||t = (A x, x)0'
the indexed set (H^*, E[)
Q t (H0 , H,)
and write
(HQ , H 1 )} .
is a morphism of
If
H^Q ^
T:
Q(Hq , H , )
for
(HQ , H.j) -*■
we define
Q(T) = T. Remark. functor
^
The fact that
Q
is, as the notation suggests, a
-- H^Q 1 j will be proved in Theorem 1 .
The proof of
functoriality was discovered independently by several persons^ including Stein whose previously unpublished proof is used here.
E.
Q, is called quad
ratic interpolation. It is clear that identification)
If
< ||A(t_s) /2 \\^ ||x|| H
and that (with an obvious
Q 1 (HQ , H 1) = Ej . For simplicity we shall often put
Qt(H0 , H,) = Ht .
linear map
Q^Hq, H 1) = HQ
“♦ H^.
0< s < t < 1 then
||x||t = ||At/2x||0 = ||A(t_s)/2A s/2x|l0
so that the identity map of Since the composition
HQ
extends to a continuous
HQ -► H g -►
H1
is just the
§2
VIII: CHAINS OF HILBERTIAN SPACES
inclusion of
HQ
in H 1
we may regard each
Hg -► Ht
it follows that
it follows that a fortiori
Hg
with the above convention H q -*■ H1
is injective and that
E^(the image of
H^ as included in
become inclusion maps.
suppose
H s -+ H^
133
Since
H^ —►
so that
)
HQis by definition dense
is dense in
Q,(Hq , ^ ) =
H^
if
0 < s < t < 1,
{H^} is an object
is completely continuous
in H^,
and let
of Hj-q
Now
(xn ) be a bounded se
quence in
Hq .
Then passing to a subsequence we can suppose
(xn )
Cauchy in
^ ,
i.e. that
hence
A 1/2 : Hq
-*• H q
Ar/2o < r < 1 . Hq
such that
so that
Hx^ - xn ||1 = ||A1/2 (xn - x ^ ||Q — 0,
is
is completelycontinuous and by [VI, Theorem 18, 3)1 so is Then if
0 < s Q =
given by
Clearly
2
lows that
H^q ^
fn : X -*■ V
and the
2
k > 0,
2
so
f -*■ cck /2f
Given
we have
L^(X, V, cc) C l|(X, V, cc)
||f||| < m ^ ”k^ ||fH^. which proves that
integer
k > £
m = Inf{cc(x) |x € X) . Then if
L^(X,
V, cc)
with respect
i.e.,
^ ( L ^ X , V, a), l |(X, V, a)) - l£+t(i_k)(X, V, a) 2
which proves that
{L^(X, V, cc))
is a scale. q.e.d.
The following class of scales will be of particular interest in the sequel. DEFINITION. for each
If
V
k € R
is an hermitian vector space then
we define
bert space of functions l|f|lk
=
£
fL^( Zn , V)
f:
Zn
llf(v) ||2(1
+
to be the hil
V such that ||v||2)k
< 00
v€ Z 11 THEOREM 3.
If
V
is an hermitian vector space then
p
(^k( X , ^ ^ k e R is' foP each non-negative integer
n,
a continuous Rellich chain and a scale. PROOF.
If we make
zn
into a measure space by declaring each
set measurable and each point to have mass one, and define then clearly
Zn , V) = L^( Zn , V, cc) ,
remains is to verify the Rellich condition. F
be the finite set of
v e zU
such that
cc(v) = (i + ||v|| )
so by Theorem 2 above all that Let
k > &.
(1 + ||v||2)
Given > e2/2.
e > o Then the
let
VIII: CHAINS OF HILBERTIAN SPACES subspace if
v
4
V)
such that 2 is finite dimensional, so the unit ball of ^ ( F , V)
F
of
and we can choose unit ball of
l|( Z n , V)
f 1, ..., fm ,
£k ( Z n , V)
consisting of
143
e/\/2 -dense in it.
||f(v)||2(U||v||2)k v€ Zn
< M k f^
If
f(v) = o
then
£ ||f(v) ||2(1 + IM|2)* veF
hence we can choose
f
< 1
so that
||f(v) - f± (v) ||2(l + ||v||2)4 < ^
v€F On the other hand since if
v
4
(1 + ||v||2)^_k 1/2
bk = Jj” (l+y2)”kdy.
bk 1/2
-► V
defined by
p(f)(X) =
/2( Zn_1, V) . More-
p: «2( Zn , V) - f2., /£( Zn_1 , V)
is a con-
tinuous linear map and has a continuous linear sec tion (i.e. right inverse)
X: J^2_1j ^ ( Z n-1 , V) -►
*£( Z n, V) . PROOF.
Taking
a^ = ||f(x, p) || and
c = (1 + I|i||2) in Lemma 2
gives
(Z
I l f U , ^ ) l l ) 2 (1 + I|x||2) k -1 / 2 < Bk
pez
Z
t1) II^ ( l +ll^ll2 +^2) k
Since the sum on the right is clearly no greater than the absolute convergence of gives
•
^Z
Z^ f(x, p)
to an element
ll^ll^-
this proves
p(f) (X.) e V
and
§!)•
VIII: CHAINS OP HIIBERTIAN SPACES
1^
in Lemma 1 l|x (g )(x ,
gives
= g(x)
r(A.) < 1 /b^.
H )||2 (1 + | | X ||2 +
^ 2) k
aH
A, e Zn 1 .
Taking
c = (1 + ||x||2)1
so
=
| | g ( x ) | | 2 ( i + | | x ||2)k - 1 r ( x ) 2 f
l + M
8
l k
1+ ||X||2+|i2J
H > 0,
be integers, k > s
f e Jt£( Zn , V)
and
the series
f(x, n)
f Z ,n-i converges absolutely, and the function
p(f) : z^
defined by
belongs to
*£_s(
p(f)(x) =
zf, V).
Moreover,
f(^, 1-0
p: i£( Zn , V) - Je^._s(
V
zf
V)
is a continuous linear map and has a continuous linear section X : *£_s( Z£, V) PROOF.
Z,n , V)
.
Immediate from Theorem k by induction on
COROLLARY.
If
(n-£).
n
is a positive integer and k > n/2 2 n there is a constant B such that f e , V) then
z
Zv€ j n f(v)
converges absolutely and
I
f(v) || < B||f||k
.
r7n vc€ L PROOF.
This is just the case
It = 0
of the theorem.
.CHAPTER IX THE DISCRETE SOBOLEV CHAIN OF A VECTOR BUNDLE Richard S. Palais In this section bly with boundary, and M.
M
will denote a compact
|, t], £
We shall also assume that
C°° manifold, possi
will denote hermitian vector bdndles over
M
is equipped with a fixed strictly positive
smooth measure.
§1.
The spaces
Ck (g)
We let vector space of
Ck (|),
Ck
which it is complete.
in
M
for
and if
\|r:
f € C°°(|)
Thus C°(g)
a map
We give
C°(g) the compact-open is a norm for
| 10 ^ 6 x Y
R n —► ®|a |-< k V
by the map
I
k-jet extension map
C0 (Jk (g))
the requirement that Ck (|)
Jk?f) (y) = (Daf (y)) (a j < k .
jk :
C°°(g) -► C°°(Hk (0)
It follows
extends naturally to
Jk .
is clearly injective and we topologize Jk
shall be a homeomorphism into.
Rn
then
Dafn
from which we see that
jk
maps
Ck (|)
Then
(so
Ck (|)
Then
Ck (|)
by
fR -► f
|a| < k
converges uniformly to
D^f
and on
in K K,
isomorphically onto a closed sub
is the underlying topological vector space
of a Banach space) and that the topology of "Ck topology."
then
so that the same coordinate representation still
is a compact subset of
C°(Jk d))
and if
f° cp-1 : R n -► V,
f =
means that, in terms of the above "coordinates,” if
space of
is a chart
0f
over
holds for this extended map, which we continue to denote by jk : Ck (|)
in
jk (f) I® is then represented by the map
defined by
Ck (|) -*■ C°(Jk d))
cp: 0 ^ R n
is a trivialization of f| 0
we represent
C°(|)
is the underlying topological vector space
We recall from Chapter IV that if
and
that the
g.
|||f|||0 = Sup{ ||f(x) |||x e M)
Jk (|) jk~f):
a non-negative integer, denote the complex
cross sections of
topology and note that
of a Banach space.
k
k7
Ck (|)
is just the usual
1^8
PAIAIS
§2. The hilbert space
§2
H°(g)
We recall that by Section b of Chapter IV /M (f(x), g(x))^ define
H°(|)
defines a prehilbert space inner product on
to be the completion of this prehilbert space.
can (and will) regard I
< f, g >^ =
such that
H°(|)
C°°(i).
We
Clearly we
as the set of measurable cross-sections
||f||g = /M (f(x), f(x))^ 1,
D:
If
D € Diff^g, ti)
C°°(g) -► C°°( r|) By Theorem 5,
Diff^ri, g),
then by Theorem 6 for each integer
extends to a continuous linear map
D
has a transpose
D^
and in fact
so again by Theorem 6 for each integer
to a continuous linear map
Hk (T]) -*Hk_1(|).
k > 1,
D^: Hk (|) -»■ D 1" = D* e D^
extends
Then [VIII, Corollary 2
of Theorem 5 ] completes the proof. q.e.d. LEMMA.
If
k and
JI
are non-negative integers then
Diffk+Jf(|, T,) = D i f f 4(Jk U ), n) • Diffk(S, Jk (I)). PROOF. and
homogeneous of degree Smbl^-U^, t^) , which are such a
C°°
a
r
k € Z,
if
a(pv, x) = pra(v, x)
is precisely the set of
the zero map of
|
less
o(v, x)
T(M)
is possible.
into
tix ,
is independent of
v)
o € C°(L(jt*i1,
v 1 7^ 0
and to be the zero map of
then
to be
r = 0
(and also for
un
a
r > 0
over
we
x.,) 8 aQ (I^ )(v2, x2) into
if
0 ^ 2 ^xg
^i^x
) € C°(L(n*(i1 8 |g), **(1^ 8 i2)) r.
will not necessarily be in
such that
More generally if and
t
[Note that even if
a 8 aQ (I^ )
a e Smblr (£ 1,
Smblr (l1 8 £2, ti1 8 tj2)
since it
C°°, at elements ((v^ v2) , (x1, x 2))
v 1 = 0 . As we saw in [IV, § 8 ] if
o 8 a0 (I^ ) € Smbl^U-j 8 £g,
r
a(o, x)
then
Then from what has just been said it follows that
cept in this case
degree
r > 0,
homogeneous of degree
will in general be only continuous, not T 1 (M)
r < 0
U-,)x ® ^ 2 ^x2
and in fact is homogeneous of degree
of
(so that
a e C°(L(jt*!^,
if we define
)) ((v 1,v 2) , ( x^ x 2)) =
a 8 aQ (I
o < 8> a Q( 1 ^ )
p > 0
no continuous extension of
))
( a 8 aQ( I
v1 = 0 .
T(M^)
while if
define
if
for
and homogeneous of degree k) . We note that if
extends to be continuous on
Given
20 7
(l1 8 £g,
8 £g).]
r, s > 0 and a € C°(L(jt*l1,)) is homogeneous of 0 * € C (L(n £g, it r\2) ) is homogeneous of degree s then
o 8 t = (cr0 (I) 8 t) (a 8 aQ (1 ^ ))
is in
C°(L(n*l1 8 £g), n*(ri1
8 Tjg)) and
is homogeneous of degree r+s. In case r = s we define an element o # t o * * of C (L(jt |, jt T) ) ) f which again is homogeneous of degree r; namely the matrix of components of
where
a # t
[XIII, § 3 ]
is
/ f f ® a 0 (IS2)
-«0 (Ini) ® ( - l ) V \
\ a 0 (I|i) ® T
(-I)V ® ^(1^)
(-l)r = e1Plt. Recall that the matrix of
reflecting the matrix of joint of each entry.
a # t
Since
(o # t)*
J
is obtained by
in the main diagonal and taking the ad
(-1 )p = (-1)-P
this gives
),
208
PALAIS
§3
°* ® °0(Il2) -«0 (I
hence
(a #
t )*(cj
)
#
°0(I|1) ® T*
® (-1)'rT
t)
'a*a 0
(-l)"ra ® x 2 ^ :^i2 ^x2 “*‘ ^ 2 ^x2
t e C°(I(jr*t2, * % 2))
is non-sln® llar>
It follows that if (which implies
dim ^
=
dim | = dim t])then
# t)((Vi, v2), (x,, X 2 ) ) =
l(Xi;X2) - n(x1 ,x2)
is a monomorphism and hence an isomorphism for all € T ’(M).
at each
at least one of the maps
is non-singular at such a point.
a e C°(I(it*^1, *%.,)) dim
T'(M) where
v
((v^ v2), (x1, x 2))
This proves: THEOREM 1 . of
degree
If
ck e C°(I(jt*ii ,
r > 0 , then
a 1 # a2
and is homogeneous of degree Recall from Chapter XII that
is homogeneous is in
C°(I(jt*|, *%))
r. is defined as the restriction of
£-
jr
to the unit sphere bundle of
T(M^) . Recall also that for any
r e R
the restriction mapping is a bisection of the set of elements of C°(L(jt*|^, the inverse map
which are homogeneous of degree r ( r')' being of course defined by a -► av
with
C0 (L('Tj_, ?j_)),
a^rt v , x) = M | r cr(v / 1|v||, x) Finally, recall that
a
(|^, tj^)
is the set of arc components of
c°(i(?'1 , t i p ) . THEOREM 2 . -► 8 , # 8 g
There is a uniquely determined map Of
A d , , n,) X A d 2 ,n2) -*• A d , n)
(&,, 8 2) such that
209
XIV: TENSOR PRODUCTS if
ai e &i €
Tij_)
r > 0
and
then
(o|r)# a^r))|S(M) € 5, # S 2 PROOF.
If
then a e C°(I(k*I1 , «*’i1))
o± e C°(I(|'1 , n±))
ishomogeneous of degree
r
for any
r > o,
C°(I(«*4 , «%)) and hence
(°jr^ #
a^tlS^)
map
a 2) - (0 jr) # a^r))|S(M) of
(a,,
C°(I(T,
.
e
C0 (I(?1, n ,)) x C°(I(t2, n2))
+ (1 -t)r
A,(t) = ts
continuous arc in
s > 0 then as
If
varies from
r
to
t
and (0 ^ ^
s
into
a(I.,, 11 ) x
varies from zero to unity
(0 ^ #
C°(I(T, *0 ) from
hence the above map
e-
C°(I(T, tD ) • Moreover the
is clearly continuous and hence induces a map
a(£2, n2) -*• a( I, t\) .
and
1,
so by Theorem
# a2 ^ ) | S ( M )
a 2 r^)|S(M)
a(I1# t\^) x a(I2, ti2) -*• a(£, t\)
is a
1:0 ( a \3^ # a^s^)|S(M)
is independent of
r
and it is clearly the unique map having the property stated in the theorem. q.e.d. Recall that in Chapter XII we defined an index map ind: Ad^ , -►
Z.
ity,"
t^)
Our goal in the rest of this chapter is to prove its "multiplicativi.e., that ind (&1 # 52) = (ind(51 ))(ind(6 2))
The property (S6 ) of the Seeley Algebra If D 0 I^ e
k
is a non-negative integer,
Diff^( 1 1 0 i2, T|1 0 |2)
and
ak (D 0 1 ^ )
might hope naively that analogously if
Intk d.| 0 l2,
would be in
0 ^))
is in
0 l2)
must be disappointed.
If
k
unless
k
0 |g))
ak (T) ® a0 ^ | ^
ak (T) ® ao ^ | ^
no^
so this hope
31101 is homogeneous of
® ^2^
so we can clearly uniformly approximate it on any compact set by Smblk ( | 1 0 £2,
an element of
0 l2) .
by Seeley that we can approximate elements
T 0 1^
OPk d 1 0 l2,
T e Smbl^d-,, t^),
0ne
is positive, then at least we know that
ak (T) 0 aQ (I^ ) € C° (L( it*d -J ® S2) , degree
then
and its symbol would be
However, as pointed out in the preceding section, even in Smbl^d-, ® l2,
= 0 k (f>) ® ao^I|2^*
T € Int^f^, t^)
(which by Corollary 1 of the Theorem of §2
then [IV, § 8 ]
D € Diff^d.,, t^)
A
in
Intk (1 1 0 l2, ^
T 0 1^ ® l2)
Not so obvious is the fact shown in
0Pk (I 1 ® i2>
so that
0 k (A)
® ^
approximates
210
PAIAIS
a , (T) 0 a (I ) a-^(T) ao(I| K U |p '
in the compact open topology.
This is a basic fact on a
par with the properties (Si)-(S5) listed in Chapter XI.
We state it formal
ly as (S6) and will prove it along with (Sl)-(S5) when we give the explicit construction of the Seeley algebra. (56)
Let
M1
and
M2
be compact C°° manifolds without boundary and with
strictly positive smooth measures. vector bundles over M 2,
and let
{An ) (An in in
in
and
k > 0.
l2, t]1 0 |2) Int^C^ 0 S2, i2)
0Pn_(l, ® 0 Z2’ 1^. 111 ti„ ® 0 ^
^
and
behermitian
l2l2 ananhermitian hermitianvector vectorbundle bundle over
T € Int^( 11, t^),
a^.(T) 0 cr0 (I^ )
§5.
M1
Let
Then there is a'‘sequence
such that
and such such that that
An
converges to T 0 1^
on (A._) aiJAn)
converses converges to to
in the compact-open topology.
Multiplicativity of the index Given
element of
= 1 , 2)
S^ € IntU^, ^ ( i
0P(|, n)
THEOREM 1.
k> 0
then
S^ e OP^^i* ^
S1 § s1 # S2 b2 € 0Pk(|, ti ]) ;.. ir If
k > o 0
then there is a sequence
CAn) CAn ) in
that t*
(
An \
converges to r» n m r c s v i r f Q e i
+• /^>
ak(An) converges to
and
S S1 1 # S2
.-r
S 1 # S2 to be the
of components [XIII, § 2 ]is
whose matrix
If
we define
/ Q ^
/-r
implies
S± € Int^^, t t^) ]±;
Int^d, ttjj)
such
in 0Pk( 0Pk (|, tt] j) and / C!
& ak ^ 2 \^
-? v~i
+- V i o
n r n n o r * +• comPact-
open topology. PROOF.
Since
1 ^ e 0PQ
1^,
1 ^ e OP^i^, n1)
and
e
_L
OPkC^i, thefirst firststatement statementisisa a consequence of [XIII, [XIII, Theorem Theorem 2]2]and OPjJ’l-p ljl_j)_)the consequence of Corollary 1 of of the the theorem theorem of of §2 §2 above. above.
€ € Int^d^, ii\/),
If If
then by (S6) (S6) of of the the preceding preceding section we we can can find find a sequence I nt^£ ^Int^( |0 ® i2, i2 ^ t)1 ^1 00 l2) ^ with
converging 1 0® 1^ converging to to SS1
a^(Bn) a^(Bn ) converging converging to to 0k 0k( (S. S|) .|) 00 aQ aQ(I^ (I )
Similarly, we can find sequences
in
k > 0,
(Bn) (Bn ) in
0Pk (S1 0 tj j1 ® 0 S2) 0Pk( i2, t
in the compact-open topology.
, and
[En) in
§5
X IV : TENSOR PRODUCTS
Intk(T)' 0 n2» ’11 ® 82)»
Intjc(11 ® S2, h ®
respectively, converging in the respective and
211 8X1(1 Intk^nl 0 n2 ' 6l 0 ^
0 Pk*s
1^
to
/V
t S2,
0
1^
^
0 S2
sf J I
and with the corresponding sequences of symbols converging I T)n k * in the compact-open topology respectively to 0 o^tj ^ ® °k^2^ > aQ (I^ ) 0 crk (S2),
(-1 )kak (S1)* 0
and
B
)•
n
-C n
Dn
En
Then by [XIII, §^, Theorem]
An =l
is in
Intk (|, rj)
while
a^(An )
and clearly,
converges to
0 Pk (|, n)
An converges in
CTk (S.,) #
to
S 1 # S2
in the compact-open topology. q.e.d.
THEOREM 2 . PROOF. S 1 # S2
(S, # S2 ) 1 = S^ # (-S2) (S1 # Sg) ^
The matrix of
.
is obtained from that of
by reflecting in the main diagonal and replacing each entry by its
transpose, i.e., S* 0 1 .
-I
n-,
I,
2
61
0 S0
® Sp
S J l
2
1
which is also clearly the matrix of
t2
# (-S2) . q.e.d.
COROLLARY 1 .
Q I
°)
n-|
where
C = S^S-, 0 1 ^ + 1 ^ 0 S2 S 2
0 S0S^ + S.S^ §> I 2 2
(S, # Sg)t(S1 # Sg) is
The matrix of
1 1
COROLLARY 2 . If
PROOF.
s j_ e E1 ( g±, t)±)
(b) (c)
then for all
ker T C ker T^
forany T e OP.j(£, ;
ker T^T = ker T ; (TtT)k - T^_lTk
k e Z
and has dimension
S-jMdim ker S2) + (dimker S^) (dimker
We first note that
(a)
D =
n2
ker(S1 # S2)k = ker(S1 # S2) (dim ker
and
.
S2) . 0
212
PALAIS
Applying(a) and
T = S1
(c) with
§5
# S 2 gives ker (S1 # S2) C ker(S1 # S2)k
C ker((S1 # S2 )t(S1 # S2))k . And applying (b) gives
ker(S1 # Sg) =
ker(S 1 # S 2 )t(S1 # S2) . Hence it will suffice to prove that ker((S1 # S2 )t(S1 # S2))k = ker(S, # S 2 )t(S1 # S2) Now by Corollary 1 it follows that
T: X -*■ Y a continuouslinear map. closed range, then so does PROOF. jectively vn hence
Let
(vn )
T
maps
isomorphic to
has
T. Then
T*T
so by [VI, Theorem 1 ]
T*T
is a sequence in
V.
V isomorphically onto V ,
T*T
V = (ker T ) 1 = (ker T^T)1 .
with the range of
0if
If
Now
TvR -*• 0
T*Tvn -► 0
T(V) = T(X) . Hence
V
bi-
implies
T*Tvn -♦ 0 ,
implies
is a hilbert space, hence closed in
maps
T(X),
being
Y. q.e.d.
LEMMA 2 . Let spaces,
Hj
be chains of hilbertian and suppose that
has closed range.
Then
A^: H^ -*•
has closed range.
PROOF. to an isomorphism
JA*: H° - H*
H2
A e 0Pk (H1, H2)
(A^A)k : H^ -*• H “k H^
and
A choice of admissible inner product for j: H~k s H^.
Now
(A^A)^ = A^A^
is the adjoint of
Ak ,
so
and, by Lemma 1 , so does
lr H 1 gives rise
and by [VIII, Theorem k]
j(AtA)k = A^Ak
has closed range
A^.
q.e.d.
213
XIV: TENSOR PRODUCTS
§5 THEOREM 3.
Let
S^ €
r^)
so that (by Theorem 1 )
(S, # S2), : H 1 (I) - H°(n)
S, # s2 € OP, (I, n) •
Then
is an F-operator and
ind(S1 # S2) 1 = (ind S1)(ind S2) .
PROOF.
Writing
v = dim ker
v((S1 # S2) 1) =
we have
v (S 1 )v (S2) + v(S^)v(S2) by Corollary 2 of Theorem 2 . Since [VIII, Theorem k] •X* +■ (S1 # S2)1 = (S1 # S2)Q , from Theorem 2 and its second corollary we have v((S, # Sg)*) = v (S^ # -S2) = v(S^)v(S2) + v(S 1 )v(Sg)
.
Then
v ((S, # Sg),) - v((S, # Sg)*) = (v(S,) - v(S^))v(S2) - v (Sg)) = (ind( S-j)) (ind( S2)) Hence (cf. [VI, Corollary 2 of Theorem (S1
# S 2 ) 1 has
closed range, and by Lemma 2 above it will suffice to prove
R ^ x , y) = 0( ||x-y||2) . Note that gi( t(x) -i|r(y),J(x) _11)
of
x.
z = i(R*(x, y), l) f
to
m
=
ei(x-y,S)
in the above expansion of
terms and of
ez
to
k
ez . Using the expansion
terms, and neglecting all remainders
we get (5)
(Am>kf)(x) = k
m
.j
J (i \ Qa (x)(y-X)a , | ) 7 J ! j=0 |a|=2
(2 , ) A l(x)v(x)-1 | J
a'(x, I) e(t) (^(y)f(y)dydt
|p| > i.
We shall show the follow
ing two facts. LEMMA k.
Given
s > 0,
y
k
and
m
can be chosen
arbitrarily large so that (6) and (7) have the form (8)
K(x, y)f(y)dy
where the support of
R n x Rn order
and
K
K
,
is a compact subset of
has continuous derivatives of
< ts.
LEMMA 5.
Let
T:
S( Rn) —
S( R n)
be a linear
map and suppose there exists an even integer and a
K
in
C^s( Rn x
j
2s > 0
Rn), with compact support
so that (Tf)(x) = y K(x, y)f(y)dy
§2
PALAIS AND SEELEY
258 Then if
j > -2 s
HJ ( Rn)
into
and
t < 2s,
T
extends to a continuous linear map of
H t R n) . A*-A^ J^ _ 1
Granted this, Theorem 7 is concluded by showing that has order
< r -£,
and hence that the expansion for A^ J^ _ 1 .
obtained from that of that for each f
in
j
||A'f - A^
R n ) . Let
s
To check the order of f*ll^ _r+J£ < cj||f||j,
be an integer with
HJ*
continuously into
H^_r+^.
2&
teger
Take
k
is
we show
||f||k = the norm of
2s > max(|j|,
||A1 f
r-£+i
A'- A^ J^_1,
|j-r+4|).
so that
Since Am k - A^ J^ _ 1
we obtain by the triangle inequality that PROOF of Lemma k.
to order
where
m > H and k > £ - 1
by Lemmas ^ and 5, we may choose maps
A’
Then
A 1- A^ k
has order
< r-A,
- A^ J^ _ 1 f||j_r+Jp 2 4 > r
+ k+
i + i 4-s + n 2
Since (6)
a
. e ^ z'^
= - ||z||2 e ^ z,T1^
(where
At = Z? , —
)
we can write
as a sum of terms of the form
-2s and
is continuous from
HJ*(Rn )
to
H ^ R 11)
t ± (p)q>j(p)a(v, p) • Smblr (M)
into
o e Smblp (M)
into
0Pp (M) ,
Xp (a)
such that support qK p
with image in
Definitions 6 and 7 of §2.
and
ffr(Aij(°)) (v > It follows that
into
= Z^. A^.(a), support cp^ ^ 0 ,
Then by
Ps a continuous lin € CZp (M)
and
= cPi (p)cpj(p)cy(v, p) where the sum is over all pairs i, j satisfies the requirements ofthe
theorem. q.e.d. COROLLARY.
The sequence
0 - CZr _1 (M) - CZr (M) is exact.
Smblr (M) - 0
PALAIS AND SEELEY
266
THEOREM 6.
Let
M 1 at
M2
§3
be compact Riemannian
manifolds without boundary and let There is a sequence
(By] in
K € 0PQ (M1 x M 2)
such that
in
and
0Pp (M1 x Mg)
PROOF. is a chart for
If
M1
A
r > 0.
and a
converges to A
o (A) (v1, x 1)
0 I- K
( x 1, x 2) )
on the unit
M 1 x Mg.
is of the form T € CZr ( R n)
and
By
or (By)((v1, v 2),
converges uniformly to sphere bundle of
A e CZr (M1)
CZr (M1 x M 2)
A(T, tjT1)
where \|r!
n
=
.
and
X
Let
e C°°(ti*)
g
T = and
with respect to the Her
a trivial computation shows that
h € C°°(M) = C°°( C^)
then
y (f(x), h(x)g(x))T)dn(x)
I
(x*)^ =
C°°(|*).
X e
|. Then
f e C°°( t})
namely if
and similarly
§5
(f(x), g(x))^h(x)d^(x)
=
J
g(f(x) )h(x)d|i(x)
=
< g*(f) , h > r
>
M
Then
Tt =
= M xStg#
and the theorem is now an easy consequence of Theorem 3 above and Theo rem t of § 3 . q.e.d. THEOREM 7.
Let
M1
and
M2
manifolds without boundary, over
M 1,
CZr (£,
£a vector bundle
and
where
£
i)
and
over
vector bundles
M2
and let
r]), r > 0.Then there is a sequence
CZr (| 0 5, t] 0 O that
be compact Riemannian
B^
K e 0PQ (| 0 5, rj 0 £)
T 0 I - K
converges to
ar (^v)
in
in such
0Pp (| 0 5, n ® 5)
converges uniformly on the unit sphere
bundle
of
PROOF.
ByTheorem
A e CZr (M),
and a
{B^}
Te
M1x M2
g € C ° ° ( ti)
to
crp (A) 0 aQ (I^). M °A ° X* § X e C°°(| ), and the theorem then fol
3 above we can assume and
T=
lows easily from Theorem 6 of §3. q.e.d.
§5- Definition and properties of Again
M
Int (£, ti) .
is a compact manifold without boundary and
are vector bundles over
M.
rj, 5
§5
271
XVI: CONSTRUCTION OF Intk THEOREM 1.
For each integer
czr.i(5’ n) = czr u , PROOF.
n) n OPr_,(s, n)
•
As noted in the remark following Definition 6 of §2,
CZr ( R n) =CZC R n) n 0Pp ( R n) CZr ( R n) n0Pr _1 ( R n) . CZp (M)
r
and Theorem
CZr>_1 ( R n) =
from which it follows that
It is then immediate from the definition of
1 of §3 that
CZr _1 (M) = CZp (M) n 0Pp _1 (M) ,
theorem follows directly from this and the definition of DEFINITION 1.
= CZp (|, t f
T e Int (|, ti)
A e CZp (|, r\)
and
0Pr _l(£> n)
If
then A
T = A + S
by
T = A ’ + S'
- A ’ = S' - S.
where
and define
ap (T) = ap (A) . where A ’e CZp (£, q) Now
and
A - A f e CZp (£, t f
S’ e
and
S' - S
^ > hence
€ 0Pr_i
A - A ’ € czr ( i , By
write
+ 0Pr_1(|, t])
S e OPp _1(l, ti)
cjp (T) e Smblr (|, t f Remark.
CZp (|, ti) .
We define
Intr (|, t f Given
and the
Theorem b
of §t
tf
n OPr _1(i, t f
= CJZ^d,
o (A) = a (Af), proving that
ap
tf
is well defined.
It is clearly linear. Since
CZp (£, ti) C_OPp (£, t])
0Pr _l(£> n) £ opr ( ^ perties of
n)
it is clear that
(Theorem b of §t) and Intp (£, ti) £ 0 P p (|, t f . Pro
(S1), (S2) and (S3) of Chapter XI are also immediate consequences
Theorem b
Chapter XI
of §t and Definition 1 above. Properties (St) and (S5) of
follow from Theorems 6 and 5 of §t respectively.
Finally,
Property (S6) of Chapter XIV, §t follows directly from Theorem 7 of §t. This completes the construction and verification of the properties of the Seeley Algebra.
2 72
§6
PALAIS AND SEELEY
§6.
An element of IntQ (S1)
with analytical index -1
Thissection considers the situation in which analytical indices for
singularintegral operators were first computed, namely on the mani
fold
M = S1 ,
plane. ing
realized as the unit circle
Here we have standard coordinate systems
eix
x,
with
in'the complex
X : z -► -i log z
and thus identify the cotangent bundle
infinite cylinder vdx^,
{|z| = h)
S1 x R 1
by letting
T*(S1)
send
with the
(z, v) e S1 x R 1 correspond to
a standard coordinate system.
X
Let
oo”= (z11}.a,00
the usual orthogonal basis of
L2(S1). For M
f e C^CS1)
and
o < r < 1,
set
P^ffz) = -— y- / &£ p 1 5-rz
= 1.
THEOREM 1. Ppf
Let
f € C°°(S1).
P e C Z ^ S 1)
when
=0
v > 0,
n > o,
PROOF.
when
Pcpn = o
if
^^ we
and for
joint supports. K(z,
O
^
when
= cpn
n 1 -
Choose
( |x| > Jt/2),
r -► 1-.
P
\|r in
cp = 0
^
on
i|r with dis
= / K(z,Of(£)d£
-oo . in standard coordi C°°(R1)
with
{ |x| > it] .
dy
00
J-»
For
= 0.
cp and
f(Odt
M^PM^ has
pPf (z) =
z) = rn zn -► zn =
M^PM^fCz) = ^U- /
€ C°°(S1 x S 1),
30
converges uniformly as
Finally, consider the representation of
on
Pcpn
lr z l 0
and
v < 0. Further
For the first,
Pcpn with
r — 1-,
converges uniformly to a limit, denoted
The operator
if
Thenas
+(x)g(y)^(y)
Tfcg). i-re '
dy
,
\|r = 1
Then
§6
XVI: CONSTRUCTION OF Intfc
since
cp(x-y)
= 1
be defined b y
when
Tg(|)
+(x)\|r(y) £ 0.
Let
273
K^x)
= g7
^^"X |x >
811(1
T
We can conclude the proof of
= i/2( — I— + i)g(g).
III the theorem b y showing that the map
g *-►
lim
K_*g
differs from
r- > 1-
g *-► Tg
b y an operator of order
v < >
- w
-oo .
I -n< :1I1X
I
—00 00
I
■ h
Now
dX
0 v d - n T r 11
0 converges boundedly to 00 K(g)
= U.
by h(cz,
C) and we c) =
(z,c) .
then we have e(cz, c) = e(cz, ^
) = (z, cz)
and id^) (z)h(cz, c) = i d ^ ( z ) ( z , c) = (z, cz) i.e.,
we have commutativity in the diagram (H|D0)|S1
h S
hi!
► C D lo1 0
1
identity
C D Is1 -------- . C IS1 o id^j o
It follows (formally, by [II, §3, (v) and (ix) of Lemma 1]) that d(H|Dn , C n , elS1) = d( C n , u o
o
, idM x) 1 '
811(1 hence
d( C D^, C D , id(l)) = -g = 5 (-M-)
de Now
280
PALAIS AM) SEELEY
Recalling our identifications
U(1) = S1
§7
and
= DQ
we get the follow
ing result by comparing with Definition 1. THEOREM 3.
Identify
U( 1)
with
\i
in the standard fashion and let of
H 1 (S t Q)
S1 = (z e C|
M
= 1)
be the generator
corresponding to its standard orienta
tion (i.e., as the boundary of
DQ = {z e C|
|z| < 1},
the latter having its orientation as a complex mani fold) . Then °h( i) = -[I Remark. vay.
The idea of the above proof was suggested by R. Solo-
Note that we have nowhere used the explicit definition of the dif
ference construction, just its functorial properties given in [II, §3, Lemma 1 ]. THEOREM k.
is a coordinate in a tubular neighborhood suppose that
eQ = 1,
the unit interval Rg •
and that
(-1, 1).
£|{|t| < 1)
Then for
C°°(U( |t| < 1}) - C°°(UM)
of
M.
on
Then
t
We may
t, |M
is the product of
-1 < e < 1
given by
x e X_.
if
{|t| < eQ}
t
and
there is a map
(R£f)(m) = f(m, e)
for
m
in
M.
By [X, H , Theorem 7], R£ is continuous from Hs(f;) to H s _1/2((;|M) / * for s > 1/2. For 6 = 0 we need also the adjoint RQ defined by
< f, RQ*g >Q = JM (f |M, g)^ a continuous map of Define
\
Since
(L)
R£L
HS(^|M)
and
L
R q LQg
P+
then as
LQ
differential operator
P+g,
with the symbol de
-P_.
If
then also
converges uniformly as
is bijective.
e - o+
The same holds as
isreplaced by
extends to
(LQf)(x) = \ ( v , x)f(x).
by
converges uniformly to a limit
scribed in Theorem 2.
R Q*
s < 0.
is elliptic, it follows that
P+ e CZq ((;|M, £ |M)
when
By duality,
H s-1^2(^) for
If g e C°°(£ |M) ,
*
where
into
g.
LQ : C°°(UM) -►C00(£|M)
LEMMA B. -1
for smooth
e
o+
D
e -► o-
is any
R£DL
-1
*
RQ LQg
or e -*• o-.
The proof of Lemma B is postponed for
a
few paragraphs. It is
the essential part of Theorem 2. LEMMA C.
If
P
and
P_
constructed in Lemma B, PROOF. any
q>
e C°°(£),
Let
are theoperators
then
f € C°°(?|M),
P+ + P_ = identity.
and set
/M (LQf, 0
< Lu, cp >Q = < u, L*cp >Q = lim
/|t | ^ £
u| ( 111 > e)
Lu = R0*LQf
in
is
{ 111 > e} .
u=
C00,
and since
L'1Ro*L0f.
= < R0*L0f, ^ >o =
(u, L*cp) . From Lemma B, has support in
Thus from Theorem 1
y (L0f, 0+
M -e
Then
(L_g (u|M_e ), cp |M_e )
M,
Lu = 0
for
289
XVII: COBORDISM INVARIANCE OP THE ANALYTICAL INDEX where
M£ = {t = e}
and
Lg
is the appropriate analog of
everything
in sight is
to obtain
/M (LQf, q>|M) = /M (LQ (P+f + P_f) ,
LQ .
Since
C°° inthe right places, we may pass to thelimit cp|M) ,
since
P+ =
-1 * 00 I I + lim R L Rn . But every C section of £ |M has the form cp|M e- > 0+ 6 u cp in t;, so LQf = LQ (P+f + P_f) . Finally,LQ is bijective, so
for a f =
P +f + P f. q.e.d. PROOF of Theorem 2.
as in Lemma B. In view of + Lemmas A and C, it suffices to show that the range of P+ lies in H+, for then ment of
f = P+f + P_f H+
Define
is the unique decomposition of
plus an element of
C“ (5|M),
then
supported
in
Similarly
P_f € H_.
P
H_.
But
f
Lemma B shows that if f €
u|X+ = (L_1R0*L0f)|X+ e C” (C|X+). Since
M,
into an ele
Lu = 0 in the interior of
X +, and
Lu= R0*LQf
is
P+f = u|M e H+ .
q.e.d. The proof of Lemma B uses LEMMA D.
L~1 e CZ_1(5, £). T e CZ_1(5,
PROOF. Choose Then
oQ (TL) =
s°
TL= I - K
(I + K + ... + K ^ T L = I - K®4"1
5) with
with
a_1(T) = X-1 =o ^ L ) -1.
Ke C
on the right by
Z L-1
£)•
Multiplying
yields
L“1 = T + KT + ... + K®T + K ^ ’l -1 Now
K^T e CZ_J_1(^,
Thus
L -1
CZ_1 ,
O,
and
since L'1 e
0P_, .
can be approximated within arbitrarily low order by members of
and the lemma follows easily. PROOF of LemmaB.
when
kP+1L _1 e 0P_m _2U , O
cp1
and
cp2
It
suffices to consider
M
L~1M R *L„g cp1 cp2 0 Oto have disjoint support, or are supported in an appro-
continuous from
for all s < 0, and a fortiori for -1 * L R Q LQg converge uniformly to a
all limit
s
H S(M)
whatsoever, Hg, and
H
is of order cp1
M
R^
is
q>2 0
H°°(X)
Thus
Suppose now with coordinates
to
T-,
L
*
In the first case,
9
M
-1
priate coordinate neighborhood.
and
-oo. cp2
have support in a neighborhood
U -► Rn+1 , and such that
£ |U
is trivial.
If
U C X U
290
PALAIS AND SEELEY
does
*
not intersect M,
RQ LQg = 0 and there is nothing
\|r= (Y-,,
we suppose that the support of
..., Yn ,^n+-j)with
|cp1 | + |cp2 |,
support
tn+1 =
(cp) C U.
to prove; so
t. Let
cp = 1 on
Then according to Lemma
D , the operator
e CZ -I.
B = y»McpL"1M cpf It is crucial to know Let
ff(B)(y, t; r\,
= a1 ti
+ aQ ,
certain details about
t )
=
where a^
Z J^
|r(support( |cp1 | + |cp2 |))
the
7
b^
(y, t ;
^ = (t^,
is "dual to"
t j,t )
...,
t. Because
as matrices of functions.
fwith support (f)C support ( |cp1 |
M L-1M M LM g = g, it cp cp cp cp
\|r*)
+ |cp2 l)
we have
7 Theoreirf 3l that on
satisfy
b _ 1a 1 = identity matrix (1)
b _ 1aQ + b _ 2a,
- ±(Sb_.,/St ) (Sa1/St)
- i E (Sb_1/Sijj) (Sa 1/Syj)
6
Let
A( 0 , b^)
< k,
^[Mp
as in [XVI, §2, Definition 3l.
L”1
Xj£ ^ £
-
JfcQ < -1 - k - (n+l)/2 tinuity of
B^l has
etc.
^n
R D[M L e 9-|
—
1 M
0
92
order < i Q + k,
/ a = 1. weakly in
Set
D
has order
so that for
B c ]Rn*Ln
remarked above show that
is continuous from
-K* - E* N B B 0 ]Rn L g 2* o
a € C°°(R1) ,
a(s)
am (s) = ma(ms) .
HS(X) for each
I (g(y), q>(y, 0))dy for
cp
H “1^2(^|M)
converges uniformly as
and it remains to prove a like result for Let
Then if
Soboleff's theorem [X, H , Theorem ij-] and the con
R0*: H_1/2(UM) — H"1( O
L"1M Cp^ Y2
Hence
0
be a patch function, and set
with
D[M
=
s < in H“s .
** = ^M>2g and
C°(0-
s -* o,
* R£DB^R0 LQg.
vanish except for
Then as
to
-1 < s < -i,
m -► + 0 }.
finite path in
equations ( 1 ) for the
(ti, t )
for
We obtain an integral in which
lim may be taken m- > 00 with an integral over a
oo / dT
under the integral sign by replacing
of
0 = 0
To see that this is possible, consider the
b^.
There
a ^ y , tj n, x)
11 12 + |x |2 > 0,
which is non-singular for
because of the ellipticity of
L.
is a linear function ^
Thus there is a compact set
support(\|r( |cp1 | + | 0 } Extend Since
0
to
Rn x C / 1^
a(x) =
{Im(x) > 0 }
,
and
so that e"itT e
0(t],
x
a(t)dt
)
is bounded for
J , >. dx.
still for
h | 2 + |x|2 >_ 1 .
decays exponentially as
integrand in (2 ) is holomorphic between may be replaced by
= 1
Then
t > 0 r(^)
lim m- > 00
and
and
|x|-* «> in
Im(x) > 0,
(Im(x) = 0 },
and the /“^ dx
may pass inside the integrals,
yielding for (2 ) the expression (3)
(2n)_n'1tp1#
y
y
Rn since
a(x/m)
e ^ £eitTdT ei < y ’ n> dr)
converges boundedly to
immediate that the expression (3) tives in
y
and
t
a(o) = f a = 1 as
isC°° for
extend continuously to
the uniform convergence of
,
r( ti)
RgDB^R0*g.
t > 0, {t
Further
m
00.
and all its
It is deriva
> 0 }. This establishes
292
PALAIS AND SEELEY lim 6-
for if
|n | > 1
and
>
€ CZ,
H B R 0+
U
R |n I > 1
(UM)
;
+'
then the function
b
T)bj^(y, 0; n, t) dx r(n)
is homogeneous of degree
£+1
r\ .
in
(Note that the homogeneity of
is preserved in its holomorphic extension, and for the given = 1
and
is the boundary of
r(r\)
r\
b^
6(n, t)
{|t| < R |t]|} n (Im(T) > o}.)
It remains only to compute
o (lim
*
R B.R
u 6 — > 0+ 6
a (P ) = a (lim R B -R *Ln) . We have for u + u e - > o+ u u
),
and hence
u
x in M
(y, 0) = y ( x )
and
that
( k)
ReB -iRo*s)(x)
o+
t)-n J
( 'l'*.(v, x)- 1 \(v, x)
generalized eigenspaces of
,
with positive imaginary
part along the direct sum of the remaining generalized eigenspaces. Since
k(v, x ) = -l(v, x), by what was remarked above
a (P )(v, x) = i(I + ik(v, x)X(v, x)). B±(x)
=
i(I + ix(v, x)).
We also have
Keeping in mind that -1
anticommute and have square
wehave + aQ (B”)(v, x) =
X(v, x)
and
x(v, x)
a straightforward computation gives
aQ (B+P+)(v, x) = i(I -
ix.(v, x) - iX(v, x)\(v, x)
+ \ ( v , x))
crQ (B~P_) (v, x) = v(I +
il(v, x) + ix(v, x)X(v, x)
+ X(v, x)) .
So if we define
C e CZQ (5|M, U M )
by
cr0 (C)(v, x) = 2
Sincex(v, x)
= -I
and
x(v, x)
C = B+P+ + B ”P_ then
i ( I +\ ( v , x)) *
= -X(v, x)
aQ (C)*(v, x) = i d
- ^(v, x))
and a0 (C)“1 (v, x) = (I - \ ( v , x))
so in particular skew-adjoint so by
C
is elliptic.
Since index(C ) = 2 index (C) the other hand ker(C) = 0 . B+ and B ”
1
2
a0 (C )(v, x) = 2 l(v, x)
XI, Corollary of Theorem 9 ,
2
Since
Now
it follows that
Indeed suppose
C C
0 = Cf
2
has index zero. has index zero. =B+P+f
+ B -P_f.
are orthogonal projections itfollows that B+P+f = B “P_f
=
0.
is
On
296
PALAIS AND SEELEY
Let
P+f = u|M,
u e
Then by the G-reen-Stokes formula (1) with
u =
we have (since Lu = o)
=
so
B P +f shows that
= 0,
(B P+f, P+f)
=
| ||B"P+f||2
hence
P_f = 0
P f+ = B P f+ + B P f+ = 0. A similar argument hence f = P f + P_f = 0. Since C has kernel
and index zero it is bijective and, by an argument similar to that of Lemma D,
C“1 e CZ (£|M, {;|M),
and of
course
a (C"1)(v, x) = a (C)"1(v, x) = (I - X(v, x)) Now define
T e CZ (|+ , g )
by T = B P C"1i
where
i+ e Diff0 (|+, 5 |M)
is the inclusion of
C°°( |+)
into
C°°(t;|M).
Then aQ (B_P+)(v,
x) = i(I
+ i d v,x))(I - ix(v, x)x(v,x))
= i(I
+ idv, x) - idv , x)dv, x)
= i(I
+ i d v ,x))(I - d v , x))
- d v , x))
so a0 (B"P+C_1) (v, x) = i(I + idv, x))(I - X(v, x))2 = -i(I + i d v , x))d v , x) = -idv, x) (I - i d v , x)) Now on
g*,
d v , x)
is multiplication by
i,
hence
I - idv, x)
is
multiplication by two, so a0 (T)(v, x) = a0 (B“P+C"1i+) (v, x) Thus
a
(T) agrees
with
-a
T.
-dv, x) |g*
analytical index of
M, so
T
is
a is equal
to
Thus to complete the proof that
alytical index zero it will suffice to show that bijective.
= -a(v, x)
on the unit sphere bundle of
elliptic and by [VII, Theorem 3 ] the the analytical index of
=
T:
a
has an
C°°(|+) -► C°°( g~) is
297
XVII: COBORDISM INVARIANCE OF THE ANALYTICAL INDEX Define a linear map B+g
into B ”g.
B+g
= 0 then
we take
To seethat B “g = 0.
u = v
0
in
=
U
U
as follows:
with
J (B+g - B _g, g)
is injective.
H°
=
norms.
=
B “g
&
T
B+ ;
from [XI, Theorem 73 that
B"P+ h
h € C°°(i“).
= 0
;
(ii)
Q
= 0
;
(iii)
Q
The first and second becauseB+h = 0
P_h
=
is surjective.
B"P+h
Then
Then
Q = < P+h, B+h >Q is in the range of T,
Then
=
< (B+ + B')(P+ + P")h, h >Q
=
< B “P_h - B+P_h, h >Q
€ H_, say P_h = v|M
with
v € H__.
=
< B"P_h, h >Q
Since
Lv = 0 if
we take
in the Green-Stokes Formula (1) we get 0 = < B"P_h - B+P_h, h >Q
=
||h||2
so
h = 0
which completes the proof that the analytical index of
cr is zero.
298
PALAIS AM) SEELEY To prove the final statement of the Theorem we note that by
[X, H , Theorem 5 ]
Z = £ §> n
Let
t)
is the restriction to
and define
M
A e Smbl^CT, T)
of a bundle by
^
A = A 0 1^.
over
X.
Clearly
X is skew-adjoint and satisfies x ( v , x)2 = -1 if ||v|| = 1. Moreover, ^jb ^ ^ ^jb if we put Tx = (e £ r x l^( v, x)e = +ie) for x e M then clearly T” = 0
£(v, x) IT*
Moreover it isclear that
a
that the analytical index of analytical index of
= (v,x) . The
proof
is zero now implies the same for the
o’.
q.e.d.
We now show how Theorem 3 leads to the main goal of this chap ter; namely the proof of part c) of Theorem k of [XV, § 5 ] for the analy tical index. mension
We assume that
2m+1,
and we let
M
X
has dimension
be the double of
n = 2m, Y.
so
Y
has di
As usual we put
2m+1 A*(T*(X) ®
C)
=y
Ap (T*(X) ® C)
,
p^O m v e n (T * (X) ® C)
A 2 p (T*(X) ® C ) ,
= y P=°
and
m A°dd
(T*(X) ® C)
A 2 p + 1 (T*(X) ® C )
= ^
P=° We recall certain linear maps. terior multiplication nap
e -► v A e
(recall the identification of into
aP+1
First for each
T (X)x
and hence interchanges
is the interior multiplication by
V for
odd..
that
Aeven
(vi> •••> vP -i}
Since iv
and
=
y
e
maps
hP into
A°dd. If we define uy :
uve = v A e A
with
v,
- iye, then
a ]_ s 0
ct1(s )(v
uv
iv ,
T*(X)x ), and
into itself
which carries
Aodd.
Secondly, there
identified by
A^3"1
iv
is the adjoint of
it likewise interchanges
A*(T(X) 0 C)x — A*(T(X) 0 C)x
is skew adjoint and interchanges
and so
Ap
e(v> v 1' **•' VP-1}
recall from [IV, §6c] that
, x)e = -ive
we have the ex
A*(T*(X) 0 C)x
0 e A^(T*(X) 0 C)x . We recall [IV, §6c] that
e »-* v A e. Aeven
of
v e T(X)
a1(d)(v, x)e
a1(d+5)(v, x) = u^..
=
Aeven
v A e,
Also
by and
hence
o 2(a )(v
, x)
299
XVII: COBORDISM INVARIANCE OF THE ANALYTICAL INDEX = cr2((d+e>) 2) (v, x) = u2 Since
v -► uy
which we computed to be multiplication by
— 1|v||2 .
is linear the argument at the beginning of the proof of
Theorem 3 shows that uv, % and in particular if
v-
+ UV2 uv,
vp
and
■
- 2 ( V 1 ’ V2>
are orthogonal then u
and u
V1
v2
antieommute .
We also have a bundle automorphism mapping
AP
into
A2m+1~p
basis element (the volume element) for ®x ,
and defined by
cdx = v., a
any oriented orthonormal basis the uniqueelement of f e AP ,
where
that in
T*(X)x ®
A2m+1'"p
(f, e) C
for
T(X)x
then
and
e
*x .
where
for
Then if
T (X)x .
and
Ae v e n
There is a canonical
... a v 2m+1,
denoted by
v 1 ,..., v2m+1 e€ Ap ,
is
*^-e
is
? o r aH
f A *-^e = (f,
hermitian structure for Ap
induced by
is the complex conjugate
of
v 1, ..., V2m+1
*x (v1 a ... a v_^) =
since the elements of
*
A (T (X) ® C)
A2m+1 (T (X) ® C)x ,
such that
is the
cular it follows that if
*
of
and hence again interchanging
reca3_^ the explicit definition of
Aodd ^
*x
e.
Inparti
as an orien1:ed- orthonormal base ... a v 2m+1 .
a
In particular
of the form v. A ... A v with * * Jp j-1 < ... < jp form an orthonormal basis for A (T (X) ® (D x it follows •0 2 that is a unitary map. It also follows that on A-^ we have *x = (_i )P(2m+1 “P) _ i
A*(T*(X) C d
i.e., that
*x
is the identity.
cause interior and exterior multiplication by have
forany
f
v
If
e € Ap
then be
are adjoint maps we
e AP “1
f a va *^e =
(f A v,e)cox
=(-1) p _ 1 (v
A f, e)^x
= (-i)p_1(f, ive)»x = (-1 )p_1f A *x d ve) and hence
v A *^e = (-i)p_1 *x (ive)Define a bundle automorphism
^(p-lj-m+l ^
on
Ap^
Then clearly
and is a unitary map satisfying i.e.,
a^. i s
v a *xe
oP
A*(T*(X) ® C)
interchanges
A0^
2
a^. = -1 . It follows that
s k e w - a d j o i n t . Also if
e e Ap
= (-T)p-1 *x (iye) = -(i)"2p *x U ve)
by and
*
= Aeven _*| = -c^,
then multiplying the equation by
iP(P“1)“m+1
we get
300
PALAIS AND SEELEY
v A a^e
=
a^i^e.
Replacing
e
by
a^e
a^.(v a e) = L^a^e.
sides of the latter equation gives
aXv Ae ” ^ v 6 = ive = ^v^X6 ” v A aXe = ^ v ^ X 6 * and
oc^
and applying
to both
Hence
axuve =
other> words
anti-oommute .
We now define the bundle Recalling that
both
and uy
(;
interchange
and in fact
considered as an element of
\
X
by
Aeven
f; = Aev0n(T*(X) 0 C ) . anaa °
\ e Smbl^((;, £)
lows that we can define an element x(v, x)e = U-Va x e ,
over
j_t
fol
by the formula (d+5 )Q^
is the symbol of the operator
Diff1(5, 5)•
We note that
*(v, x)2 =v w x Also since
u^
and
are skew adjoint we have
\ ( v , x)* = a* u* Thus the hypotheses of Theorem
To recapitulate we have the following theorem whose proof
was the main goal of this chapter. THEOREM k.
Let
and define
a
i.e.,
e
§* = {e e A*(T*(M) 0 C) |o^e = + e] Smbld (£+ ,
a(v, x) = uy .
| ”)
by
a = a1(d+&)|| + ,
302
PALAIS AND SEELEY If
bundle over
t] is any bundle over
Y, then
M
which is the restriction of a
a ® 1^ e Smbl^(| + ® ti,
symbol with analytical index zero.
®> ti)
is an elliptic
In other words c) of Theorem b of
[XV, §5 ] holds for the analytical index.
CHAPTER XVIII BORDISM GROUPS OF BUNDLES E. E. Floyd
§1 .
Introductory remarks The purpose of this chapter is to prove the uniqueness theorem
for index functions stated in Chapter I.
The proof can he based directly
upon the results of Conner and myself in our Ergebnisse tract [2],
In view
of the fact that torsion does not enter into the proof, it is also possible to give the following complete proof based directly upon Thom's work [^]. We first collect some definitions and elementary remarks, be ginning with Thom's L-equivalence classes. X
without boundary and an integer
k
entiable k-dimensional submanifolds L- e q u i v a l e n t X x I
such
with
M , M1
Fix a differentiable n-manifold 0 < k < n.
Two closed differ
with oriented normal bundle are
if there exists a compact differentiable submanifold that
W intersects
X x 0, X x 1 transversally in
of
MQ x 0,
M 1 x 1 respectively
and such that
normal bundle of
W
can be oriented so as to extend the orientation of the
normal bundle of
MQ x 0, M 1 x 1
Lk (X)
W
W
in
has no other boundary, and if the
X x 0, X x 1
respectively.
the set of L-equivalence classes of closed k-manifolds in
oriented normal bundle. (1)
If
X
X
with
There are the following remarks. is oriented, there is a one-to-onecorrespondence
between orientations for the normal bundle of the tangent bundle of
Denote by
M.
MC X
andorientations
In this case, in the definition of
for
L-equivalence
we suppose the tangent bundle oriented instead of the normal bundle. (2) isotopy of
Mq
orientation and
Given an with
MQ
with oriented normal bundle and a smooth
M 1 C X, then the normal bundle of
MQ and
M1
are L-equivalent. 303
M1
receives an
§1
FLOYD (3)
If
smoothly isotopic to
M1
with
k < dim X/ 2 ,
MQ
and
we can define an abelian operation in Since
L^-(X)
Mj
given
MQ
disjoint.
L^(x )
and
M1
there is an
Hence in this case
^y disjoint union.
is defined only for manifolds, does not always
have a natural operation, etc., it is natural to consider also more function al constructs (see Atiyah [1], Conner-Floyd [2]). Let pairs
(M, f)
f: M - * X
X
be a space, and
where
is a map.
M
n
a non-negative integer.
is a closed oriented differentiable n-manifold and
Two pairs
(MQ, ^q) > ^1 >
are b o r d a n t if there
exists a compact oriented differentiable manifold joint union
MQ U -M1
and a map
F: W -► X
There results an equivalence relation. sented by
(M, f),
and by
nn (X)
the set of all For
¥
with boundary the dis
F|Mq = f , EM, f]
the class repre
[M, f ].
X = point,
F|M1 = f1 .
Then
ftn (X)
fln (X)
is
is the Thom
consisting of all bordism classes of closed oriented differentiable
n-manifolds.
There are the following straightforward remarks. (t)
defined by
A map
cp: X -► Y
cp*[M, f] = [M, cpf].
cpQ , cp1 : X -► Y
gives a homomorphism Moreover,
are homotopic then
homotopy equivalence, then (5)
If
boundary and if ^^(X)
with
Denote by
an abelian group under disjoint union. group
Consider
X
cp*: ^(X) -► ^^(Y)
)* = T2*cp1*;
cpQ* = cp.,*.
Hence if
if
cp: X -*■ Y
is a
cp*: nn (X) « ^n (Y) .
is an oriented differentiable manifold without
2k + 2 < dim X,
there is the natural homomorphism
L-^(X) -♦
which takes the L-equivalence class represented by the closed oriented
submanifold
M
[M, i] C n^(X),
into
where
i
is the inclusion map.
It
follows from the Whitney embedding theorem that this is an isomorphism Lk (X) ~ nk (X) • (6) Sn . Denote by boundary. plies that
Let N
X
be a finite complex embedded as a subcomplex of
the closed regular neighborhood of
Then ( b )
implies
n^(X) « ft^N-dN),
X
and by
for
dN
its
2k+2 < n, (5) im
flk (X) « L^(N-^N) . (7)
Given
f: M -*■ X, where
M
is a closed oriented differen
tiable k-manifold, there are the Pontrjagin numbers of given a cup product p. ... p. e H-^(M) 1r
f
[2],
Namely,
of Pontrjagin classes of the tangent
Mj
§1
XVIII: BORDISM GROUPS OF BUNDLES
bundle of
M
x e H^-(X)
and
with
305
r(i1 + ...+ i ) + q = k,
there is the
integer P1
••• Pn- f*(x) [M] r
1 the value of
,
H^(M) .
... p. f*(x) e H^M) on the orientation class of M 1 r These numbers are functions only of the bordism class [M, f ] e
ft^(X),
and are the Pontrjagin numbers of
x €
p.
Q),
[M, f ].
,
where
G
is a compact Lie group.
can reinterpret this group in the following way. M
We may also let
in which case the Pontrjagin number is a rational number.
Consider now
where
in
is a
Consider pairs
over
M.
Two pairs
(MQ , a ) and
(M1 , a )
there exists a compact oriented differentiable manifold disjoint union
MQ U -M1,
stricts to
on
M^,
and a principal G-bundle
i = 0, 1.
p
(M, a)
a is a prin
closed oriented differentiable k-manifold and
cipal G-bundle
We
are bordant if
W
with
over
dW
W
the
which re
This is an equivalence relation; we see
that the set of equivalence classes is in one-to-one correspondence with (M, a)
ft^BG) .
In fact, given
for
Letting the bordism class represented by
of.
there is a classifying map
ft^BG),
we get the one-to-one correspondence.
for
any classifying space for
BG
consider
BG
G
(M, a)
f: M
-*• BG
map into [M, f] €
It is seen that we may use
in dimensions
< k + 1.
Hence we may
a finite complex. In a similar fashion, we can consider bordism of pairs
where
M
is a closed oriented differentiable manifold, and
pairs
(Mq , aQ)
and
(M1, a 1)
differentiable
manifold
b € K(W)
the restriction
i = 0, 1.
with
W
(M, a)
a € K(M). Two
are bordant if there exists a compact oriented
with boundary the K(W)
disjoint unionMQ
K(M^) sending
b
into
a^
U -M1 ,and for
It can be seen that this is an equivalence relation, using, in
proving transitivity, exactness of K(W1 U W 2)
K(W.,) + K(W2) - K(W1 n W 2)
We are thus lead to an abelian group of bordism classes. one-to-one correspondence with homotopy classes of maps points not considered). correspondence with
But
K(M)
is in
M -+ Z x BU
(base
Hence, as above, the bordism group is in one-to-one x BU).
306 §2.
§2
FLOYD The computation of
fl^(X) ® Q
We first recall somewell-known facts abouthomotopy theory. Given
spaces X, Y
classes of maps X
with base point, denote
X
Y
by
[X, Y]
which preserve base point.
is a finite CW-complex and that
Y
the set ofhomotopy
We suppose always that
is a CW-complex whose skeletons are
finite. If
Y
is
stable range), then [3].
(n-1)-connected and
[X, Y]
(i.e., in the
is an abelian group, also finitely generated
In the stable range there is a homomorphism [X, Y]
where
dim X < 2n - 2
Hom[H*(X), H*(Y) ]
H*(X) -►H*(Y),
- Hom[H*(X), H*(Y) ]
,
is the group of degree preserving homomorphisms
which sends the homotopy class of
f
into
f*:
This is an isomorphism modulo the class of torsion groups [3 ].
H*(X) -*■ H*(Y) . It is equally
true that [X, Y] is an
Hom[H*(Y) , H*(X) ]
isomorphism modulo torsion.
Here homology and cohomology are
taken to
be reduced. Let
A
-► BSO(n)
closed n-ball, and let Then
MSO(n) = A/dA
dA C A
be a universal
S0(n) -bundle with fiber the
denote the union of the boundary spheres.
is the universal Thom space; it is (n-1)-connected.
The cohomology is given by the Thom isomorphism i J k;
Also
consisting of all
flk (1 x BU) = «k (BU) ~
choosing a classifying space
BU(n)
in dimensions
be a finite complex, we may apply (9 ) to compute rank the number
of pairs
(j.,, j2, ...) For
( cd,
cd
’ )
fi*(Z x BU)
of partitions
cd
=
(i
[M,
ftk (BU(n)) < N
flk (1 x BU) . i2,
. . . ) ,
a]
cd
1
to It is
=
with t(i1 + i2 + ...) + 2 (j 1 + j2 + ...) = k. a £ K(X),
dimensional component.
let
ch a = Z ch^ a
Writing
symmetric functions we obtain,
Z
where
ch^ a
is the 2q-
as a polynomial in elementary
308
FLOYD
chqa
Denote by bundle. by
ch^ It
chq,
If ch(a 0 b) =
(-I)11'1 _ ! a _
Q)
€ H2c^(BU;
is seen that
q > 0.
integers with
=
H*(BTJ;
is
where
a
V1
is the universal
the polynomialalgebragenerated
t = (v1, v2, ...) i,
.
of non-negative
let
vp
= (ctLj)
• (ch2)
and
b e K(Y)
are virtual dimension one, then
ch a 0 ch b
implies that ^ chpa ® chqb p+q=n
ch0 = 1 . Consider now
generator sion
cq_^)
chT
c h ^ a b) =
where
Q)
for almost all
a e K(X)
+
the class °hqa
Given a sequence
v^ = 0
§3
g
S
n\r ,
of H2k(S2k;
one. InK((S2k)v),
Z)
C H 2k(S2k;Q),
let
ch(Vk)v =
(Vk)v = V k ®
(chk)v (Vk)v = g ® ... ® g
by
with
Vk
withch^V^.
a
ofvirtual dimen
... V k . Then
(1 + g) ® ... ® (1 + g)
ohk(Vk)v = g ®
Denote
OI/-
e K(S )
and the element
1
v!
® ... ® 1 + . . . + l ® l ® . . . ® l ® g , g ® ... ® g .
gk v - Note also that ch^(Vk)v = 0
if
i < k.
Consider again the sequences t as above. For each t , let o Voh- v-i S = (S ) x ... x (S )where v. = 0 for i > k, and let V € Vlr V1 * K(S_) (V1) 0 ... 0 (V.) . Write elements of H (Sd T be given by V =T I K T * Pi i as linear combinations of elements x 1 0 ... 0 where x^ € H ((S ) ). Then
V1
(ch )
v p
(ch2)
VT = v 1 ! g.,
0 1 0 ... 0 1
V T = v 21 1 0 g2 v
8) 1 0 ... 1 + Z x^ 0 y^ 0 1 0 ... 0 1,
deg x 1 > 0 ,
§3
309
XVIII: BORDISM GROUPS OP BUNDLES Order the partitions
...+ 2kvk .
If
there exists
t
k
x
’ = (v.J, v£, ...) with
v£ = v.^ for
as follows. has
deg t
Let
deg x* = deg x,
i > k
while
x' < t
let
< v^.
+
= 2v1 + if
Computation as
above shows (11)
chT V TT = 0
if
Consider now elements
deg x 1 = deg x [M, a] e
and
fl^(BU),
x’ < x
and consider the x e H (BU; Q) .
characteristic numbers of such elements as in (7) except that
given finite sequences co = (s^ s2, ...), t = (v1 , v 2, ...), let P^ = S1 S2 ^ (P-|) (P2) ... be the appropriate cup product of Pontrjagin classes of M and let POJ°hT [M' “ ] = PtB ' ChT (aH M ] These rational numbers are functions only of the bordism class
[M, oc ].
They are zero unless (is1 + 8s 2 + 12s, + ...) + (2v1 + W 2 + 6v^ + . . . ) = k (3.1)
THEOREM. [M, oc]
classes
Consider
fl*(BU)
where
is a closed oriented differ
entiable manifold and one.
PROOF. (v1# v 2,...) x ...,
Pontrjagin classes of Pa>' • cV Hence, p^, deg x and
where
H2j'(S2f
V. e K(S2j)
Consider finite sequences Let
ch.V.
J J
a gen-
cd
= (s1 , s2, ...), t =
P^ =
(P2(C )) 1x (P^( C)) 2
[P^ x S , 1 V ] € n^(BU). Note that since the ST
are all zero, we have
[Pm x ST , 1 ® V T ] - P ^ . E P J
• ch^jP^ x ST , 1 V ] = 0 unless
• (chT ,VT)[ST ]< . degcd! = deg oj,
deg x f
x 1< x . Supposethe characteristicnumbers [M, a] = Z n ^ t P ^ x
are allzero.
has
[S2j*, Vj ],
Q) .
Z) C H2;>(S21
of non-negative integers.
and consider
is of virtual dimension
[P2i(C), 1], i = 1, 2 , ..., and
j = 1, 2, . . . 9 erator of
a € K(M)
n*(BU) Q is the polynomial algebra gen
Then
erated by
M
as the ring of bordism
We show inductively that
of alinear
combination
ST ,1 ® T = 0 f°p aH
a>,T.
Suppose
=
310
FLOYD
ncD t = 9
fop a11
03' T
§3
deS T =
0 “ Pm' • C\
[M> a] =
T0
^
=
Y n ^ p ,[P ] ■ ch T, [S, ] L »,T ) = < -v, u >.
316
§2
SOLOVAY
By (3),
T(T*(X))
inherits a complex structure (in which the real part
is "vertical" and the imaginary part is "horizontal"); in this way, T*(X)
becomes an almost-complex manifold.
By construction, we have an
isomorphism of complex vector bundles: (10
T(T*(X)) s n*(T(X) ® C) Remark.
*
The almost-complex structure on
T (X)
depends on
the choice of a Riemannian metric; however, it is well-defined up to a homotopy since any two Riemannian structures on Let
B(X)
be the unit ball bundle of
some fixed Riemannian structure on structure from
-ft
T (X) ; we give
almost-complex structure. H 2 n (B(X), S(X); Q) tion of
X
Let
X.
B(X)
B(X)
are homotopic.
T*(X)
inherits analmost-complex
the orientatidn determined by its
n = dim X,
and let
[B(X), S(X) ]
e
be the fundamental class determined by the orienta
B(X) . Finally, let ^(B(X)) = ^(T(B(X)))
the almost-complex manifold THEOREM
determined by
2 . Let
be the Todd class of
B(X). X
be a compact differentiable mani
fold, not necessarily oriented.', let S:
C°°(t) - C°°( r\)
be an elliptic operator on X. (5)
Then
ia (S) = ch(7 (S)) • ^(B(X))[B(X), S(X)] (Note the absence of signs!) PROOF.
[X] e H ^X; Q) tion of
X.
We consider first the case when
X
is oriented.
Let
be the fundamental class determined by some fixed orienta LetU e Hn (B(X), S(X); Q)
be the Thom class associated to
this orientation.(From now on, rational coefficients are Let cp*:
PROOF of Theorem 6 .2 . We may assume, by Proposition 6.7, that X
has an
Give
W
Sc(m)-structure, with associated principal Sc(m)-bundle a hermitian structure.
U(n)-bundle
P2 .
U(n)-bundle
P,
The principal bundle of
The ordered pair and
X
(P1, P2)
W
P1.
is then a
corresponds to an Sc(m) x
may thus be viewed as a manifold with an
Sc(m) x U(n)-structure. For brevity, let projection
G
be
G -► Sc(m) , on the
the unit sphere of R m G that leaves Let N = N1 ® C n.
em M’
and
Sc(m) -module R m
since
fixed
Sc(m)
G
tion 6 .8 , it follows that
M. M
Sc(m)
N M, N
on
H,
of
Sc(m-1 ) x U(n).
M 1 and of
on
Similarly,
and
can apply the results of §5.3 to
acts, via the
and G is transitive
N r be as in Proposition 6 .8 .
on
G
is.Moreover the subgroup,
is precisely
From the actions of
we get an action of
Sc(m) x U(n).Then
G
acts on
are isomorphic as and
G.
Put
M = M 1 ® Cn,
U(n)
on C n ,
N. From Proposi H-modules.
The result is that
We
33*4-
§6
SOLOVAY
y± / 2 -y± /2 v /„ (e - e ) • y. oh(W)e ' J] i [X]
■i 1 - e 1) (1 - e is integral.
1)
(As usual, the Pontrjagin classes of
elementary symmetric functions in
y^,
y^
X
where
are viewed as the r = m/2.)
The result now follows from the identity y± /2 -y± /2 e 1 - e 1
1
PROOF of Theorem 6.3.
-1
The proof is, for the most part, quite
similar to that of Theorem 6.2; we omit many details. 0(m).
Then, using the hypothesis
w 2(X) = 0 ,
we give
Let X
G = Spin(n) x a suitable
G-structure (cf. first paragraph of the proof of Theorem 6.2).
Let M = Sm % Rn N
arequaternionic
by means of
N = Sm
G-modules.
Let V = R
m
Rn ( c f ' §6-9) - Then M .
the natural inner product of a 1*.)
Proposition 1^.
We are going to
with
V
*
apply
Thus we need a map :
Since
(We identify V
S(V*)
Rm C A^,
- S".
If
and v
— HomH (M, N) . A^ S* C_ S^,
€ S(V*)> s e S*,
there is a bilinearpairing w € R n ,we put
ff1 (v) (s ® w) = \|r(v ® s) ® w Then
a1
satisfies the hypotheses of Proposition 5.U. The remainder of the proof follows the last twoparagraphs
the proof of Theorem 6.2. struct the details.
of
The interested reader will be able to recon
§6
XIX: THE INDEX THEOREM: APPLICATIONS
335
REFERENCES
[1]
M. F. Atiyah, R. Bott and Arnold Shapiro, Clifford modules , Topology, vol. 3, supp. 1 (196*0 pp. 3-38.
[2]
F. Hirzebruch, Reue Topologisahe Methoden in der Algebraisohen Geometrie > Springer-Verlag, 1956.
[3]
F. Hirzebruch, A Riemann-Rooh theorem for differentiable manifolds , Seminaire Bourbaki, Fevrier, 1959 -
[*4-]
John Milnor,
The Representation Rings of Some Classical Groups ,
mimeographed notes, Princeton, 1 9 6 3 .
APPENDIX I THE INDEX THEOREM FOR MANIFOLDS WITH BOUNDARY M. F. Atiyah Introduction In this appendix we shall indicate briefly how the index theo rem may be extended to manifolds with boundary. manifold with boundary X
Y
and
d
If
X
is a smooth compact
is an elliptic differential operator on
then there is a definition of an "elliptic boundary condition” (or bound
ary operator) b.
This definition, due originally to Lopatinski, imposes an
algebraic restriction on the symbol of
b
relative to the symbol of
d.
In §1 we shall give this definition, taking care to state everything intrin sically.
Such a pair
(d, b)
then has an index and the problem is to ex
press this index in terms of the topological data provided by the symbols ff(d)
and
o(b) . In the case when the boundary is empty then
difference element ch d e H*(X; Q)
[a(d)] € K(B(X), S(X))
defines a
and hence a cohomology class
by the formula ch d = ( _ D n (n+D/2 ^
cp*
a(d)
ch[cJ(d)]
being the Thom isomorphism. In the general case we shall show in §2 that the pair
a(b)
defines a difference element [o(d, b)] e K(B(X), B(X)|Y u S(X))
and hence a relative cohomology class ch(d, b) € H*(X, Y; Q) by the formula
337
,
a(d),
338
ATIYAH
§1
ch(d, b) = (-i)n (n+1>/2 cp*
oh[o(d, b)]
being now the relative Thom isomorphism H*(B(X), B(X) |Y U S(X); Q ) — H*(X, Y; Q) The topological index
it(d, b)
.
is then defined, in analogy
with the closed case, by it(d, b) = (ch(d, b) 0,
x^,) ^ 0
f = °
n
' p j(ax7 , 1
343
=
0
(J=1> •••'
r)
which is of the form
= exp(ix11, +...+ lxn_1ln _1)h(xn )
(and real) and
h ^ 0.
First let us observe that (ii) and (iii) may be reformulated as (ii) 1
for each
..., in _1) 7^0
(g-,
and real, m. gj e C J ,
and for every set of vectors the boundary problem (*)