Lecture Notes in

Mathematics
Edited by A. Dold and B. Eckmann
Subseries: Mathematisches Institut der Universit~it und
Max-Planck-lnstitut for Mathematik, Bonn - voL 5
Adviser: E Hirzebruch

1111

Arbeitstagung Bonn 1984

Proceedings of the meeting
held by the Max-Planck-lnstitut fur Mathematik, Bonn
June 15-22, 1984

Edited by E Hirzebruch, J. Schwermer and S. Suter

I IIII

Springer-Verlag
Berlin Heidelberg New York Tokyo
Herausgeber
Friedrich H i r z e b r u c h
Joachim Schwermer
Silke Suter
M a x - P l a n c k - l n s t i t u t fLir M a t h e m a t i k
Gottfried-Claren-Str. 26
5 3 0 0 B o n n 3, Federal R e p u b l i c of G e r m a n y

A M S - S u b j e c t C l a s s i f i c a t i o n (1980): 1 0 D 1 5 , 1 0 D 2 1 , 1 0 F 9 9 , 1 2 D 3 0 , 1 4 H 1 0 ,
14H40, 14K22, 17B65, 20G35, 22E47, 22E65, 32G15, 53C20, 5 7 N13,
58F19

I S B N 3 - 5 4 0 4 5 1 9 5 - 8 S p r i n g e r - V e r l a g B e r l i n H e i d e l b e r g N e w York T o k y o
I S B N 0 - 3 8 7 - 1 5 1 9 5 - 8 S p r i n g e r - V e r l a g N e w York H e i d e l b e r g Berlin T o k y o

CIP-Kurztitelaufnahme der Deutschen Bibliothek. Mathematische Arbeitstagung <25. 1984.

Bonn>: Arbeitstagung Bonn: 1984; proceedings of the meeting, held in Bonn, June 15-22,
1984 / [25. Math. Arbeitstagung]. Ed. by E Hirzebruch ... - Berlin; Heidelberg; NewYork;
Tokyo: Springer, 1985. (Lecture notes in mathematics; Vol. 1t11: Subseries: Mathematisches
I nstitut der U niversit~it und Max-Planck-lnstitut for Mathematik Bonn; VoL 5)
ISBN 3-540-t5195-8 (Berlin...)
ISBN 0-387q5195-8 (NewYork ...)
NE: Hirzebruch, Friedrich [Hrsg.]; Lecture notes in mathematics / Subseries: Mathematischee
Institut der UniversitAt und Max-Planck-lnstitut fur Mathematik Bonn; HST
This work ts subject to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specifically those of translation, reprinting, re~use of illustrations, broadcasting,
reproduction by photocopying machine or similar means, and storage in data banks. Under
§ 54 of the German Copyright Law where copies are made for other than private use, a fee is
payable to "Verwertungsgesetlschaft Wort", Munich.
© by Springer-Verlag Berlin Heidelberg 1985
Printed in Germany
Printing and binding: Beltz Offsetdruck, Hemsbach/Bergstn
2146/3140-543210
 VORWORT

Die m a t h e m a t i s c h e Arbeitstagung in B o n n hatte seit ihrem Anfang im

Jahre ]957 bis hin z u m 25. Treffen 1984 stets ihren eigenen unver-
wechselbaren Charakter. Das P r o g r a m m wird nicht im v o r a u s festme-
legt, sondern Vortragende und Themen ergeben sich spontan aus einer
Programmdiskussion aller Teilnehmer unter Leitung v o n F. Hirzebruch.
Dies erm~glicht, auf neueste Entwicklungen in e i n z e l n e n mathemati-
schen Gebieten unmittelbar einzugehen,und sie a u f der Tagung vorzu-
stellen. Dar~ber hinaus ist die Arbeitstagung nicht ausschlieSlich
einem speziellen mathematischen Thema gewidmet, sondern es finden
sich vonder Zahlentheorie [ber die Topologie und Geometrie bis hin
zur A n a l y s i s ~ber die Jahre hinweg Beitr~ge aus fast allen Gebieten
der Mathematik. Gerade diesem informellen und arbeitsmiBigem Charak-
ter d e r Tagung h~tte die V e r ~ f f e n t l i c h u n g von Ergebnisberichten
widersprochen. Da jedoch immer wieder stark danach gefragt wurde,
h a t man, erst seit dem Jahre 1974, einen Kompromi8 gefunden. Die
Sprecher fassen ihre V o r t r ~ g e - meist nur handschriftlich - kurz
zusam~,en, und diese werden gesammelt zum Abschlu8 eines jeden Tref-
fens a n die T e i l n e h m e r verteilt.

Anl~81ich der 25. mathematischen Arbeitstagung im J a h r e 1984

wurde und wird hiervon in m e h r f a c h e r Hinsicht abgewichen: Einige
Mathematiker wurden im v o r a u s gebeten, in U b e r b l i c k s v o r t r ~ q e n Ent-
wicklungen und Probleme in i h r e n eigenen Arbeitsgebieten darzustel-
len. Diese Beitr~ge finden sich im e r s t e n Teil des vorliegenden
Bandes. Den zweiten Teil bilden Ausarbeitungen der meisten der ad-
hoc V o r t r ~ g e , deren Sprecher erst w~hrend der A r b e i t s t a g u n g aus-
gewihlt wurden. Vielleicht spiegelt sich trotz dieser ~nderungen
im P r o c e d e r e ein wenig der Charakter der Arbeitstagung wider. Im
Anhang sind die Programme der Arbeitstagungen 1957 - 1984 w i e d e r -
gegeben.

Die M~he der Schreibarbeit fast aller Beitr~ge haben mit v i e l

Geduld und Sorgfalt Frau K. D e u t l e r und Frau C. Pearce auf sich
genommen, denen wir daf~r sehr dankbar sind.

Bonn, November 1984 Die Herausgeber

 25. M a t h e m a t i s c h e A r b e i t ~ t a ~ u n ~ 1984 (15. - 22. Ju~i)

J. Tits: Groups and g r o u p functors attached to K a c - M o o d y data

M. Atiyah: The e i g e n v a l u e s of the Dirac operator

A. Connes: K-theory, cyclic cohomology and o p e r a t o r algebras
G. Segal: Loop g r o u p s

G. Harder : Special v a l u e s of Hecke L-functions and a b e l i a n

integrals
H. Wente: A counterexample in 3-space to a c o n j e c t u r e of
H. Hopf
G. Faltings: Compactification of Ag/Z
C.T.C. Wall: Geometric structures and a l g e b r a i c surfaces

J. Harris: Recent work on Hodge structures

Y.T. Siu: Some r e c e n t results in c o m p l e x m a n i f o l d theory

r e l a t e d to v a n i s h i n g theorems for the s e m i p o s i t i v e
case
W. Schmid: Recent progress in r e p r e s e n t a t i o n theory
W. Ballmann: Manifolds of n o n - p o s i t i v e curvature
B. M a z u r -
Conjectures of B e i l i n s o n on L - f u n c t i o n s and K - t h e o r y
Ch. Soul@:

H.-O. Peitgen: Morphology of Julia sets

S.S. Chern: Some applications of the m e t h o d of m o v i n g frames
S. Lang: Vojta's conjecture on heights and G r e e n ' s function
S. Donaldson: 4-manifolds with indefinite intersection form

D. Zagier: M o d u l a r points, m o d u l a r curves, modular surfaces

and m o d u l a r forms
G. v a n der Geer: Schottky's problem
R. Bryant: G 2 and S p i n ( 7 ) - h o l o n o m y
S. Wolpert: Homology of T e i c h m ~ l l e r spaces

J.-P. Serre: l-adic representations

M.F. Atiyah: On Manin's manuscript "New d i m e n s i o n s in geometry"
 Inhaltsverzeichnis

~BERBLICKS VORTR~GE

Connes, A. - Introduction to non commutative differen- 3-16

tial geometry
H a r d e r , G., - Special values of Hecke L-functions and 17-49
Schappacher, N. abelian integrals
Harris, J. - An introduction to infinitesimal varia- 51-58
tions of Hodge structures
Manin, Y. - New dimensions in geometry 59-101
Atiyah, M. - Commentary on the article of Manin 103-109
Peitgen, H.O., - The Mandelbrot set in a model for phase 111-134
Richter, P.H. transitions
Schmid, W. - Recent developments in r e p r e s e n t a t i o n 135-153
theory
Segal, G.B. - Loop groups 155-168
Siu, Y.-T. - S o m e r e c e n t r e s u l t s in c o m p l e x m a n i f o l d 169-192
t h e o r y r e l a t e d to v a n i s h i n g t h e o r e m s f o r
the s e m i p o s i t i v e case
Tits, J. - Groups and group functors attached to 193-223
Kac-Moody data
Zagier, D. - Modular points, modular curves, modular 225-248
surfaces and modular forms

Ad-hoc VORTR~GE

Atiyah, M. - Eigenvalues of the Dirac operator 251-260

Ballmann, W. - Manifolds of non positive curvature 261-268
Bryant, R. - Metrics with holonomy G 2 or Spin(7) 269-277
Chern, S.S., - On Riemannian metrics adapted to three- 279-308
H a m i l t o n , R.S. dimensional contact manifolds (Appendix
by Alan Weinstein)
Donaldson, S.K. - 4-manifolds with indefinite intersection 309-320
form
Faltings, G. - Arithmetische Kompaktifizierung des 321 - 383
Modulraums der abelschen Variet~ten
van der G e e r , G. - T h e Schottky problem 385-406
Lang, S. - Vojta's conjecture 407-419
Wente, H.C. - A counterexample in 3 - s p a c e to a 421-429
c o n j e c t u r e o f H. Hopf
Wolpert, S.A. The topology and geometry of the moduli 431-451
s p a c e of R i e m a n n surfaces

ANHANG

Programme der 25 A r b e i t s t a g u n g e n von 453-478

1957-1984
Teilnehmer der Arbeitstagung 1984 479-481
 INTRODUCTION TO N O N COMMUTATIVE
DIFFERENTIAL GEOMETRY

A. Connes
Institut des Hautes
Etudes Scientifiques
35, Route des Chartres
F-91440 Bures-Sur-Yvette
France

This is the introduction to a series of p a p e r s in w h i c h we shall extend

the calculus of d i f f e r e n t i a l forms and the de Rham homology of currents
beyond their customary framework of m a n i f o l d s , in o r d e r to d e a l with
spaces of a more elaborate nature, such as,

a) the space of leaves of a foliation,

b) the dual space of a finitely generated non-abelian discrete group

(or L i e group),

c) the orbit space of the action of a discrete group (or Lie group) on
a manifold.

What such spaces have in c o m m o n is to be, in g e n e r a l , badly behaved as

point sets, so t h a t the usual tools of m e a s u r e theory, topology and
differential geometry lose their pertinence. These spaces are much
better understood by m e a n s of a canonically associated algebra which
is the group convolution algebra in c a s e b). When the space V is an
ordinary manifold, the associated algebra is commutative. It is an
algebra of complex-valued functions on V, endowed with the pointwise
operations of sum and product.

A smooth manifold V can be considered from different points of view

such as

~) Measure theory (i.e. V appears as a measure space with a fixed

measure class),
8) T o p o l o g ~ (i.e. V appears as a locally compact space),

T) D i f f e r e n t i a l geometry (i.e. V appears as a smooth manifold).

Each of these structures on V is fully specified by the c o r r e s p o n d i n g

algebra of functions, namely:

~) The c o m m u t a t i v e von Neumann algebra L~(V] of classes of e s s e n t i a l l y

bounded measurable functions on V,

8) The C * - algebra C0(V) of c o n t i n u o u s functions on V which vanish

at infinity,

y) The algebra C~(V) of smooth functions with compact support.

C

It has long been known to o p e r a t o r algebraists that measure theory and

topol o g y extend far b e y o n d their usual framework to:

A) Th__e theory of w e i g h t s and yon N e u m a n n algebras.

B) C*- algebras, K theory an d index theory.

Let us b r i e f l y discuss these two fields,

A) The theory of weights and yon N e u m a n n algebras.

To an o r d i n a r y measure space (X,~) correspond the von N e u m a n n algebra

L (X,~) and the weight ~ :

~(f) : SX fd~ Vf 6 L~(x,~) +

Any pair (M,~) of a c o m m u t a t i v e yon N e u m a n n algebra M and weight

is o b t a i n e d in this way from a m e a s u r e space (X,~). Thus the place
of o r d i n a r y measure theory in the theory of weights on von N e u m a n n
algebras is similar to that of c o m m u t a t i v e algebras among a r b i t r a r y ones.
This is why A) is o f t e n called non-commutative measure theory.

Non-cO~autative measure theory has many features w h i c h are trivial in

the c o m m u t a t i v e case. For instance to each w e i g h t ~ on a v o n Neumann
algebra M corresponds canonically a one-parameter group ~C
ot Aut M
of a u t o m o r p h i s m s of M, its m o d u l a r automorphism group. When M is
commutative, one has a ~(x)
t = x, V x 6 M, and for any w e i g h t ~ on M.
We r e f e r to [13] for a s u r v e y of n o n - c o m m u t a t i v e measure theory.

B) C * - a l g e b r a s , K theory and index theory_u.

Gel'fand's theorem implies that the c a t e g o r y of c o m m u t a t i v e C*-algebras

and *-homomorphisms is d u a l to the category of locally compact spaces
and p r o p e r continuous maps.

Non-commutative C~algebra~ have first been used as a tool to construct

von Neumann algebras and weights, exactly as in o r d i n a r y measure theory,
where the Riesz representation theorem [38], Theorem 2.14, enables to
construct a measure from a positive linear form on continuous functions.
In this use of C * - a l g e b r a s the m a i n tool is p o s i t i v i t y . The fine topo-
logical features of the "space" under consideration do not show up.
These fine features came intoplay thanks to A t i y a h ' s topological K-theory
[2]. First the p r o o f of the p e r i o d i c i t y theorem of R. B o t t shows that
its n a t u r a l set up is n o n - c o m m u t a t i v e Banach algebras (cf. [46]). Two
functors K 0 , K I (with v a l u e s in the c a t e g o r y of a b e l i a n groups) are de-
fined and any short exact sequence of B a n a c h algebras gives rise to an
hexagonal exact sequence of K groups. For A = C0(X) , the c o m m u t a t i v e
C*-algebra associated to a l o c a l l y compact space X, Kj(A) is (in a
natural manner) isomorphic to KJ(x) , the K theory with compact sup-
ports of X. Since (cf. [41]) for a c o m m u t a t i v e Banach algebra B, Kj(B)
depends only upon the G e l ' f a n d spectrum of B, it is r e a l l y the C*-alge-
bra case which is m o s t relevant.

Secondly, Brown, Douglas and Fillmore have classified (cf. [8]) short
exact sequences of C * - a l g e b r a s of the form:

0÷ K+A÷ C(X) ÷ 0

where K is the C * - a l g e b r a of c o m p a c t operators in H i l b e r t space, and

X is a c o m p a c t space. They have shown how to c o n s t r u c t a group from
such extensions. When X is a f i n i t e dimensional compact metric space,
this group is n a t u r a l l y isomorphic to KI (X) , the Steenrod K homology
of X , cf. [19],[24].
Since the o r i g i n a l classification problem of e x t e n s i o n s did arise as
an i n t e r n a l question in o p e r a t o r and C*-algebra theory, the w o r k of
Brown, Douglas and Fillmore made it c l e a r that K theory is an in-
dispensable tool even for studying C*-algebras per se. This fact was
further emphasized by the role of K theory in the classification of
C*-algebras which are inductive limits of finite dimensional ones
(cf. [7] [22] [21]),and in the w o r k of C u n t z and K r i e g e r on C*-
algebras associated to t o p o l o g i c a l Markov chains [18].

Finally the w o r k of the Russian school, of M i s c e n k o and K a s p a r o v in

particular, ([30] [26] [27] [28]), on the N o v i k o v conjecture, has shown
that the K theory of n o n - c o m m u t a t i v e C*-algebras plays a crucial role
in the solution of c l a s s i c a l problems in the t h e o r y of n o n - s i m p l y - c o n -
nected manifolds. For such a space X , a basic homotopy invariant is
the F-equivariant signature q of its u n i v e r s a l covering X, w h e r e
r = z1(X) is the fundamental group of X. This invariant ~ lies in
the K group, K0(C*(F)), of the g r o u p C*- algebra C*(F).

The K theory of C * - a l g e b r a s , the e x t e n s i o n theory of Brown, Douglas

and F i l l m o r e a n d the Ell t h e o r y of A t i y a h ([I]) are all special cases
of K a s p a r o v ' s bivariant functor KK(A,B) . G i v e n two 2/2 graded
C*-algebras A and B, KK(A,B) is an a b e l i a n group whose elements are
homotopy classes of K a s p a r o v A-B bimodules (cf. [26] [27]).

After this quick overview of m e a s u r e theory and topology in the non-

commutative framework, let us be m o r e specific about the algebras
associated to the "spaces" occuring in a) b) c) above.

a) Let V be a s m o o t h manifold, F a smooth foliation of V. The

measure theory of the leaf space "V/F" is d e s c r i b e d by the y o n N e u m a n n
algebra of the foliation (cf.[I0][11][12]). The topology of the leaf
space is d e s c r i b e d by the C * - a l g e b r a C*(V,F) of the foliation
(cf. [11] [12] [43]).

b) Let F be a d i s c r e t e group. The m e a s u r e t h e o r y of the (reduced)

A
dual space F is d e s c r i b e d by the yon N e u m a n n a l g e b r a I(F) of o p e r a -
tors in the H i l b e r t space i2(r) which are invariant under right trans-
lations. This von Neumann algebra is the w e a k closure of the g r o u p ring
•F acting in £2(r) by left translations.
 A
The topology of the (reduced) dual space r is d e s c r i b e d by the C*-
algebra c*(r), the n o r m closure of {F in the algebra of b o u n d e d
r ~2
o p e r a t o r s in (F).

b') For a Lie group G the d i s c u s s i o n is the same, with C ~c(G) instead
of ~F.

c) Let F be a d i s c r e t e group acting on a m a n i f o l d W. The m e a s u r e

theory of the "orbit space" W/F is d e s c r i b e d by the yon N e u m a n n
algebra crossed product L~(W) ~ F (cf. [33]).

The situation is s u m m a r i z e d in the following table:

A A
Space V V/F F G W/F

Measure v.N.algebra
theory L~(V) of (V,F) }< (F) ~ (G) L(W) ~ F

Topology C0(V) C*(V,F) Cr(F) C*(G)r C0(W) ~ F

It is a g e n e r a l principle (cf. [3] [14] [4]) that for families of

elliptic operators (Dy)y6y parametrized by a "space" Y s u c h as those
occuring above, the index of the family is an e l e m e n t of K0(A) , the
K group of the C*-algebra associated to Y. For i n s t a n c e the F-equivari-
ant signature of the universal covering X of a c o m p a c t oriented mani-
fold is the F-equivariant index of the elliptic signature operator on
X. We are in case b) and 0 6 K0(C~(F)) . The o b v i o u s problem £hen is to
compute K(A) for the C * - a l g e b r a s of the above spaces, and then the
index of f a m i l i e s of e l l i p t i c operators.

After the breakthrough of P i m s n e r and V o i c u l e s c u ([34]) in the c o m p u t a -

tion of K groups of c r o s s e d products, and under the influence of the
Kasparov bivariant theory, the general program of c o m p u t a t i o n of the
K groups of the a b o v e spaces (i.e. of the associated C*-algebras) has
undergone rapid progress in the last y e a r s ([12] [43] [31] [32] [45] [44]).

So far, each new r e s u l t confirms the v a l i d i t y of the general conjecture

formulated in [4]. In o r d e r to state it b r i e f l y , we shall deal only with
case c) above. We also assume that F is d i s c r e t e and torsion free ,
cf. [4] for the g e n e r a l case. By a f a m i l a r construction of a l g e b r a i c
topology a space such as W/F, the o r b i t space of a d i s c r e t e g r o u p action,
can be r e a l i z e d asa simplicialcomplex, up to h o m o t o p y . One lets F act f r e e l y
and p r o p e r l y on a c o n t r a c t i b l e space EF, and forms the homotopy
quotient W AF EF which is a m e a n i n g f u l space even when the q u o t i e n t
topological space W/F is p a t h o l o g i c a l . In case b) (F acting on
W = {pt)) this yields the classifying space BF. In case a), see [12]
for the analoguous construction. In [4] (using [12] and [14]) a map
is d e f i n e d from the t w i s t e d K homology K*,T(W ×F EF) to the K
group of the C*-algebra C0(W)~F. The conjecture is that this map
is a l w a y s an ismorphfsm.

~:K,, ~ (W ~.EF) ÷ K, (C O (W) ~ F)

At this point it w o u l d be t e m p t i n g to a d v o c a t e that the space w ×FEF

gives a sufficiently good description of the t o p o l o g y of W/F and
that we can dispense with C* algebras. However, it is a l r e a d y clear
in the simplest examples that the C*-algebra A = C0(W) ~ F is a
finer description of the "topological space" of o r b i t s . For instance,
with W = SI and F = Z , the actions given by two irrational rotations
R01,R 9 yield isomorphic C*-algebras if and o n l y if 01 = ±0 ([34]
[35]) 2 and M o r i t a equivalent C*-algebras iff 8I and 82 belong to
the same orbit of the action of PSL(2,~) on PI (~)[36]. On the c o n t r a r y ,
the h o m o t o p y quotient is i n d e p e n d e n t of e (and is h o m o t o p i c to the
two torus) .

Moreover, as we a l r e a d y mentioned, an i m p o r t a n t role of a "space" such

as Y = W/F is to p a r a m e t r i z e a family of e l l i p t i c operators, (Dy) y6y.
Such a family has b o t h a topological index Indt(D) , w h i c h belongs to
the twisted K homology group K,(WI F EF), and an a n a l y t i c index
Inda(m) = ~(Indt(m)), which belongs to K,(C0(W) ~ F) (cf. [4] [16]).
But it is a p r i o r i only through Inda(D) that the analytic properties
of the family (Dy)y6y are r e f l e c t e d . For instance, if e a c h Dy is the
Dirac operator on a Spin Riemannian manifold M of s t r i c t l y p o s i t i v e
Y
scalar curvature, one has Ind a (D) = 0 (cf. [ 3 7 ] [ 1 6 ] ) , b u t the e q u a l i t y
Indt(D) = 0 follows only if one knows that the m a p ~ is i n j e c t i v e
(cf. [4][37][16]). The p r o b l e m of i n j e c t i v i t y of b is an i m p o r t a n t
reason for d e v e l o p i n g the analogue of de R h a m homology for the above
"spaces". Any closed de R h a m c u r r e n t C on a m a n i f o l d V yields a
map @C from K*(V) to

ec(e) = <C,che> Ve £ K*(V)

where ch:K*(V) + H*(V,]R ) is the usual Chern character.

Now , any "closed de R h a m current" C on the o r b i t space W/F should

yield a map ~C from K.(C0(W) ~ F) to {. The rational injectivity
of ~ would then follow from the e x i s t e n c e , for e a c h ~ 6 H*(W×E EF) ,
of a " c l o s e d current" C(~) making the following diagram commutative,

K.,T(W ×I,EI') ........~.. K.((C0(W) n V)

]ch. I <0C(~)

H . ( W x F El', ~) ~

Here we a s s u m e that W is ~ - e q u i v a r i a n t l y oriented so that the dual

Chern character ch.:K., T ÷ H. is w e l l defined (See [16]). Also, we
view ~6 H*(W×F EF,~) as a l i n e a r map from H.(W ×F E F , ~ ) to ~.

This leads us to the subject to our series of p a p e r s which is;

I. The construction of de Rham homology for the aboye__spaces I

2. Its a p p l i c a t i o n s to K theory and index theory__t.

The construction of the t h e o r y of c u r r e n t s , closed currents, a n d of the

maps ~C for the above "spaces", requires two q u i t e different steps.
The first is p u r e l ~ algebraic:

One starts with an a l g e b r a A over C, w h i c h plays the role of C~(V),

and one d e v e l o p s the analogue of de R h a m h o m o l o g y , the p a i r i n g with the
algebraic K theory groups K0(A), KI (A), and algebraic tools to p e r -
form the c o m p u t a t i o n s . This step y i e l d s a contravariant functor H~
f r o m non c o m m u t a t i v e algebras to g r a d e d modules over the p o l y n o m i a l ring
{~(u) with a generator ~ of d e g r e e 2. In the d e f i n i t i o n of this functor
the finite cyclic groups play a crucial role, and this is why H* is
called cyclic cohomology_m Note that it is a c o n t r a v a r i a n t functor for al-
gebras and h e n c e a covariant one for"spaces". It is the s u b j e c t of p a r t II
under the title,
 10

De R h a m h o m o l o g y and non-commutative algebra

The second step involves analysis:

The non-commutative algebra A is n o w a d e n s e subalgebra of a C*-

algebra A and the problem is, given a closed current C on A as
above satisfying a suitable continuity condition relative to A, to
extend ~c:K0(A) ÷ { to a m a p from K0(A) to ~. In the simplest
situation, which will be t h e o n l y o n e t r e a t e d in p a r t s I and II, the
algebra A cA is s t a b l e under holomorphio functional calculus (cf.
Appendix 3 of part I) and the above problem is t r i v i a l to h a n d l e since
the inclusion A cA induces an i s o m o r p h i s m K0(A)~ K0(A) . However,
even to treat the fundamental class of W/F, where r is a d i s c r e t e
group acting by orientation preserving diffeomorphisms on W, a more
elaborate method is r e q u i r e d and will be d i s c u s s e d in p a r t V (cf. [16]).
In the context of a c t i o n s of d i s c r e t e groups we shall construct C(~)
and ~C(~) for a n y cohomology class ~ 6 H*(W ×F EF,~) in the subring
R generated by the following classes:

a) C h e r n classes of r-equivariant (non u n i t a r y ) bundles on W.

b) r-invariant differential forms on W.

c) G e l ' f a n d Fuchs classes.

As applications of o u r construction we get (in the above context):

~) If x 6K,,~(W x r EF) and <ch,x,~> ~ 0 for s o m e ~ in the a b o v e

ring R then Z(x) ~ 0.

In fact we shall further improve this result by varying W; it w i l l

then apply also to the case W = {pt}, i.e. to t h e usual Novikov con-
jecture. All this will be d i s c u s s e d in p a r t V, b u t see [16] for a p r e -
view.

8) For any ~ 6 R and any family (Dy)y£y of e l l i p t i c operators para-

metrized by Y = w/r , one has the index theorem.

~c(Inda(D)) = <ch,Indt(D),~>
 11

When Y is an o r d i n a r y manifold, this is the cohomological form of

the A t i y a h Singer index theorem for families ([3]).

It is i m p o r t a n t to n o t e that, in a l l cases, the r i g h t hand side is

computable by a standard recipe of a l g e b r a i c topology f r o m the symbol
of D. The left hand side carries the a n a l y t i c information s u c h as
vanishing, homotopy invariance,...

All these results w i l l be e x t e n d e d to t h e c a s e of foliations (i.e.

when Y is t h e leaf space of a foliation) in p a r t VI.

As a t h i r d application of o u r analogue of de R h a m h o m o l o g y for the

above spaces we shall obtain index formulae for t r a n s v e r s a l l y elliptic
operators; that is e l l i p t i c operators o n the above "spaces" Y. In p a r t
IV w e shall work out the p s e u d o - d i f f e r e n t i a l calculus for c r o s s e d pro-
ducts of a C*-algebra by a Lie group, (cf. [15]), thus yielding many
non-trivial examples of e l l i p t i c operators on spaces o f the above type.
Let A be the C * - a l g e b r a associated to Y, any such elliptic operator
on Y yields a finitely summable Fredholm module over the dense sub-
algebra A of s m o o t h elements of A.

In p a r t I we s h o w h o w to c o n s t r u c t canonically from such a Fredholm

module a closed current on the dense subalgebra A. T h e title of part I,
the C h e r n character in K homology is m o t i v a t e d b y the specialization
of the a b o v e construction to the case when Y is an o r d i n a r y manifold.
Then the K homology K,(V) is e n t i r e l y described by elliptic opera-
tors on V ([6] [14]) a n d the association of a c l o s e d current provides
us w i t h a map,

K, (V) ÷ H, (V,~)

which is e x a c t l y the dual Chern character ch,.

The explicit computation of t h i s m a p ch, will be t r e a t e d in p a r t III

as a n i n t r o d u c t i o n to the asymptotic methods of computations of c y c l i c
cocycles which will be used again in p a r t IV. As a c o r o l l a r y we shall,
in p a r t IV g i v e completely explicit formulae for i n d i c e s of finite
difference, differential operators on t h e real line.
 12

If D is a n e l l i p t i c operator on a " s p a c e " Y and C is t h e closed

current C = ch,D (constructed in p a r t I), the m a p ~c:K,(A) ÷
makes sense and one has,

~c(X) = <x,[D]> : Index Dx V x 6 K,(A)

where the right hand side means the i n d e x of D with coefficients in

x, o r e q u i v a l e n t l y the value of the pairing between K homology and
K cohomology. The integrality of this value, Index Dx6~, is a b a s i c
result which will be a l r e a d y used in a v e r y e f f i c i e n t way in p a r t I,
to c o n t r o l K.(A) .

The a i m of p a r t I is to s h o w t h a t the c o n s t r u c t i o n of the C h e r n character

ch, in K homology dictates the b a s i c definitions and operations -
s u c h as the suspension map S - in c y c l i c cohomology. It is m o t i v a t e d
b y the p r e v i o u s work of Helton and Howe [23], Carey and Pincus [9] a n d
Douglas and V o i c u l e s c u [20].

There is a n o t h e r , equally important, natural route to c y c l i c cohomology.

It was taken by L o d a y and Quillen ([29]) and by Tsigan ([42]). Since
the latter work is independent from ours, cyclic cohomology was dis-
covered from two quite different points of view.

There is a l s o a strong relation with the w o r k o f I. S e g a l [39] on

quantized differential forms, which will be d i s c u s s e d in p a r t IV and
with the w o r k of M. Karoubi on secondary characteristic classes [25],
which is d i s c u s s e d in p a r t II, Theorem 33.

Our results and in p a r t i c u l a r the spectral sequence of part II w e r e

announced in the conference on operator algebras held in O b e r w o l f a c h
in S e p t e m b e r 1981 ([17]).

Besides parts I and II, which will soon appear in the IHES Publications,
our set of p a p e r s will contain:

I. The Chern character in K homology.

II. De Rham homology and non commutative algebra.
III. Smooth manifolds, Alexander Spanier cohomology and index theory.
IV. Pseudodifferential calculus for C* d y n a m i c a l systems, index
 13

t h e o r e m for c r o s s e d p r o d u c t s and the p s e u d o torus.

V° D i s c r e t e groups and a c t i o n s on s m o o t h manifolds.
VI. F o l i a t i o n s and t r a n s v e r s a l l y e l l i p t i c o p e r a t o r s .
VII. Lie groups.
 14

References

[I] M.F. Atiyah, Global theory of e l l i p t i c operators, Proc. Internat.

Conf. on functional analysis and related topics, Univ. of Tokyo
Press, Tokyo (1970).

[2] M.F. Atiyah, K theory, Benjamin (1967).

[3] M.F. Atiyah and I. Singer, The index of e l l i p t i c operators IV,

Ann. of Math. 93 (1971) p. 119-138.

[4] P. Baum and A. Connes, Geometric K theory for Lie groups and
Foliations, Preprint IHES, 1982.

[5] P. Baum and A. Connes, Leafwise homotopy e q u i v a l e n c e and rational

Pontrjagin classes, Preprint IHES, 1983.

[6] P. Baum and R. Douglas, K homotopy and index theory, O p e r a t o r

Algebras and applications, Proc. Symposia Pure Math. 38 (1982),
part I, p. 117-173.

[7] O. Bratteli, Inductive limits of finite d i m e n s i o n a l C*-algebras,

Trans. AMS 171 (1972), p. 195-234.

[8] L.G. Brown, R. Douglas and P.A. Fillmore, Extensions of C*-algebras

and K homology, Ann. of Math. (2) 105 (1977) p. 265-324.

[9] R. Carey and J.D. Pincus, Almost c o m m u t i n g algebras, K theory and

o p e r a t o r algebras, Lecture Notes in Math. N°575, Springer Berlin-
New York (1977).

[10] A.Connes, The von Neumann algebra of a foliation, Lecture Notes in

Physics N°80 (1978) p. 145-151, Springer B e r l i n - N e w York.

[11] A. Connes, Sur la th&orie non commutative de l'int&gration, Alg~bres

d'op&rateurs, Lecture Notes in Math. N°725, Springer B e r l i n - N e w York
(1979).

[i2] A. Connes, A Survey of foliations and operator algebras, Operator

algebras and applications, Proc. Symposia Pure Math. 38 (1982)
Part I, p. 521-628.

[13] A. Connes, C l a s s i f i c a t i o n des facteurs, Operator algebras and

applications, Proc. Symposia Pure Math. 38 (1982) Part II, p. 4 3 - I 0 ~

[14] A. Connes and G. Skandalis, The l o n g i t u d i n a l index t h e o r e m for

foliations, to appear in Publ. R.I.M.S. Kyoto.

[15] A. Connes, C * - a l g d b r e s - e t g6om6trie diff@rentielle, C.R.Acad.

Sci. Paris, tome 290, S@rie I (1980).

[16] A. Connes, Cyclic c o h o m o l o g y and the transverse fundamental class of

a foliation, Preprint IHES M/84/7.

[17] A. Connes, Spectral sequence and homology of currents for operator

algebras. Math. F o r s c h u n g s i n s t i t u t O b e r w o l f a c h T a g u n g s b e r i c h t 42/81,
F u n k t i o n a l a n a l y s i s und C * - A l g e b r e ~ , 27.9. - 3.10.1981.
 15

[18] J. Cuntz and W. Kr~eger, A class of C * - a l g e b r a s and t o p o l o g i c a l

M a r k o v chains, Invent. Math. 56 (1980) p. 2 5 1 - 268.

[19] R. Douglas, C * - a l g e b r a e x t e n s i o n s and K homology, Annals of Math.

Studies, N°95, P r i n c e t o n U n i v e r s i t y Press 1980.

[20] R. Douglas and D. V o i c u l e s c u , On the s m o o t h n e s s of sphere extensions,~

J. O p e r a t o r T h e o r y 6(I) (1981) p. 103.

[21] E.G. Effros, D.E. H a n d e l m a n and C.L. Shen, D i m e n s i o n groups and

their affine r e p r e s e n t a t i o n s , Amer, J. Math. 102 (1980) p. 385-407.

[22] G. Elliott, On the c l a s s i f i c a t i o n of inductive limits of sequences

of semi-simple finite d i m e n s i o n a l algebras, J. Alg. 38 (1976)
p. 29-44.

[23] J. Helton and R. Howe, Integral operators, commutators, traces,

index and homology, Proc. of Conf. on o p e r a t o r theory, Lecture Notes
in Math. N°345, Springer B e r l i n - N e w York (1973).

[24] D.S. Kahn, J. Kaminker and C. Schochet, Generalized homology theo~

ries on compact m e t r i c spaces, M i c h i g a n Math. J. 24 (1977)
p. 203-224.

[25] M. Karoubi, Connexions, c o u r b u r e s et classes c a r a c t 6 r i s t i q u e s en

K theorie alg6brique, C a n a d i a n Math. Soc. Proc. Vol.2, part I (1982)
p. 19-27.

[26] G. Kasparov, K functor and e x t e n s i o n s of C * - a l g e b r a s , Izv. Akad.

Nauk SSSR, Ser. Mat. 44(1980) p. 571-636.

[27] G. Kasparov, K theory, group C * - a l g e b r a s and higher signature,

Conspectus, C h e r n o g o l o v k a (1983).

[28] G. Kasparov, Lorentz groups: K theory of unitary r e p r e s e n t a t i o n s

and crossed products, preprint, C h e r n o g o l o v k a , 1983.

[29] J.L. Loday and D. Quillen, Cyclic h o m o l o g y and the Lie a~gebra of
matrices, C.R. Acad. Sci. Paris, S6rie I, 296 (1983) p. 295-297.

[30] A.S. Miscenko, Infinite d i m e n s i o n a l r e p r e s e n t a t i o n s of d i s c r e t e

groups and higher signature, Math. USSR Izv. 8 (1974) p. 85-112.

[31] M. P e n i n g t o n , K theory and C * - a l g e b r a s of Lie groups and Foliations,

D. Phil. thesis, M i c h a e l m a s , Term. 1983, Oxford,

[32] M. P e n i n g t o n and R. Plymen, The Dirac o p e r a t o r and the p r i n c i p a l

series for c o m p l e x s e m i - s i m p l e Lie groups, J. Funct. A n a l y s i s 53
(1983) p. 269-286.

[33] G. Pedersen, C * - a l g e b r a s and their automorphism groups. Academic

Press, New York (1979).

[34] M. Pimsner and D. Voiculescu, Exact sequences for K groups and

Ext groups of certain c r o s s - p r o d u c t C*-algebras. J. of o p e r a t o r
theory 4 (1980), 93-118.

[35] M. Pimsner and D. Voiculescu, I m b e d d i n g the irrational r o t a t i o n C*

algebra into an AF algebra, J. of o p e r a t o r theory 4 (1980) 201-211.
 18

[36] M. Rieffel, C*-algebras associated with irrational rotations, Pac.

J. of Math. 93, N°2 (1981).

[37] J. Rosenberg, C* algebras, p o s i t i v e scalar curvature and the

N o v i k o ~ conjecture, Publ. Math. IHES, Vol. 58 (1984) p. 409-424.

[38] W. Rudin, Real and complex analysis, Mc. Graw Hill, New York (1966).

[39] I. Segal, Q u a n t i z e d differential forms, Topology 7 (7968)

p. 147-172.

[40] I.M. Singer, Some remarks on o p e r a t o r theory and index theory,

Lecture notes in Math. 575 (1977) p. 128-138, Springer New York.

[41] J.L. Taylor, T o p o l o g i c a l invariants of the maximal ideal space of

a Banach algebra, A d v a n c e s in Math. 19 (1976) N°2, p. 149-206.

[42] B.L. Tsigan, Homology of m a t r i x Lie algebras over rings and Hoch~
schild homology, Uspekhi Math. Nauk. Vol. 38 (1983) p. 217-218.

[43] A.M. Torpe, K theory for the leaf space of foliations by Reeb
components, Mat. Institut, Odense Univ., P r e p r i n t (1982).

[44] A. Valette, K. T h e o r y for the reduced C * - a l g e b r a of semisimple Lie

groups with real rank one, Q u a r t e r l y J. of Math., O x f o r d S~rie 2,
Vol. 35 (1984) p. 334-359.

[45] A. Wasserman, Une d 6 m o n s t r a t i o n de la conjecture de C o n n e s - K a s p a r o v ,

To appear in C.R. Acad. Sci. Paris.

[46] R. Wood, Banach algebras and Bott p e r i o d i c i t y , Topology 4 (1965-66)

p. 371-389.
 SPECIAL VALUES OF HECKE L-FUNCTIONS AND

ABELIAN INTEGRALS

G ,. Harder and N. S c h a p p a c h e r
Max-Planck-Institut
f~r M a t h e m a t i k
Gottfried-Claren-Str. 26
5300 Bonn 3

In this article we attempt to e x p l a i n the f o r m a l i s m of D e l i g n e ' s ratio-

nality conjecture for special values of motivic L-functions (see [DI])
in the p a r t i c u l a r case of L - f u n c t i o n s attached to algebraic Hecke charac-
ters ("Gr~Bencharaktere of type A0"). In this case the c o n j e c t u r e is
now a theorem by virtue of two c o m p l e m e n t a r y results, due to D. Blasius
and G. Harder, respectively: see §5 below.

For any "motive" over an a l g e b r a i c number field, Deligne's conjecture

relates certain special values of its L - f u n c t i o n to certain p e r i o d s of
the motive. Most of the time when motives come up in a g e o m e t r i c situa-
tion, we tend to know very little about their L-functions. In the special
case e n v i s a g e d here, however, the situation is quite different: The L-
functions of a l g e b r a i c Hecke characters are among those for w h i c h Hecke
proved analytic continuation to the whole complex plane and f u n c t i o n a l
equation. But the "geometry" of the c o r r e s p o n d i n g motives has e m e r g e d
only fairly recently - see §3 below.

The r e l a t i v e l y good command we n o w have of the motives attached to alge-

braic Hecke characters reveals that many n o n - t r i v i a l period r e l a t i o n s
are in fact but r e f l e c t i o n s of c h a r a c t e r - i d e n t i t i e s . This point of v i e w
is sy s t e m a t i c a l l y perused in [Seh], and we shall illustrate it here by
the s o - c a l l e d formula of C h o w l a and Selberg: see § 6.

This formula, in fact, goes back to the year 1897, as does the instance
of D e l i g n e ' s conjecture with w h i c h we start in § I. Tying up these two
relations in the m o t i v i c formalism, we hope to make it apparent that
both r e s u l t s really should be v i e w e d "comme les deux volets d'un m~me
d i p t y q u e '~, as A. Weil has p o i n t e d but in [WIII], p. 463.
 18

Contents

§ I A formula of Hurwitz
§ 2 Algebraic Hecke characters
§ 3 Motives
§ 4 Periods
§ 5 The rationality conjecture for Hecke L-functions
§ 6 A formula of Lerch

§ 1o A formula of Hurwitz

In 1897, Hurwitz [Hu] proved that

1 4v
(I) E' 4v ~ × (rational number),
a,b6~ (a+bi)

for all ~ = 1,2,3,... , where

rl dx £(
(2) ~ = 21_ - 2.62205755,.. =
u

Notice the analogy of these identities with the well-known formula for
the Riemann zeta-function at p o s i t i v e even integers:

1 2v
(3) E' 2~ - (2~i) x (rational number).
a£ ~ a

Both formulas are special cases of D e l i g n e ' s conjecture. To understand

this in H u r w i t z ' case, we look at t h e elliptic curve A given by the
equation

A : y2 = 4x 3 _ 4x .

A is d e f i n e d over ~ , but we o f t e n prefer to look at it as d e f i n e d over

the field k = ~(i)c ~ . Over this field of definition, we can see that
A admits complex multiplication by the same field k :
 19

k > End (A) ®

--X
i I >
I---> iy

D e l i g n e ' s a c c o u n t of Hurwitz' formula would start from the o b s e r v a t i o n

that both sides of (I) express information about the h o m o l o g y

H I (A) ® 4 9 c H4~ (A 4~)

The left hand side of (I) carries data c o l l e c t e d at the finite places of
k , as does the right hand side for the infinite places.

In fact, look at the d i f f e r e n t c o h o m o l o g y theories:

- Etale cohomology: Fix a rational prime number ~ , and denote, for

n
n ~ 1 , by A[£ n] the group of ~ - t o r s l o n points in A(~), ~ being the
a l g e b r a i c closure of ~ in ~ . Then

V£(A) = I<lim A[£n])®ZZ£ ~£

n

is the dual of the first £-adic c o h o m o l o g y of A×k~ with coefficients

in ~£

By functoriality, the i s o m o r p h i s m k ~ ~ ® ~ End A makes Vi(A) into

a k ® ~£-module, free of rank I. The natural continuous action
of Gal (W/k) on V£(A) is k ® ~ -linear, and therefore given by a
c o n t i n u o u s character

~£: Gal (~/k)ab - - > GL k ® ~ £ ( V Z(A)) = (k®~z)* .

This character was e s s e n t i a l l y d e t e r m i n e d - if from a rather d i f f e r e n t

p o i n t of v i e w - on July 6, 1814 by Gauss, [Ga] . The e x p l i c i t analysis of
the G a l o i s - a c t i o n on t o r s i o n - p o i n t s of A was carried out (in a stunning-
ly "modern" fashion) in 1850 by Eisenstein, [El]. - In any case, if
is a prime element of ~ [i] not dividing 2Z , n o r m a l i z e d so that
~ I (mod (I+i) 3) , and if F 6 Gal (~/k)ab is a g e o m e t r i c Frobenius
 20

element at (~) (i.e., F-1(x)~ ~ x ~ ~ (mod P ) for any p r i m e P of kab

dividing (~), any algebraic integer x 6 kab), then one finds

~ (F) = -I 6 k* c > ( k ® Q£)*

The characters ~£ all fit t o g e t h e r to g i v e an " a l g e b r a i c Hecke charac-

ter" ~ defined on the group 12 of ideals of k that are p r i m e to 2:

I ~ > k*
2

F ~
I2~ > Gal (~/k)ab > (k®~g)*

Then for all ~ Z I, the c h a r a c t e r ~4~ can be d e f i n e d on all ideals of

k by (~) ~__> - 4 v . Remember that k is e m b e d d e d into ~ , so t h a t it
makes sense to c o n s i d e r the L-functions

L(~ 4~ ,s) = ~ I (Re(s) > I-2~) ,

~ (~) 4~
p I s

where p ranges over all p r i m e ideals of ~ [i] . Then the left hand side
of H u r w i t z ' f o r m u l a (I) is s i m p l y 4 .L(~ 4~ ,0). We h a v e shown how this is
a special value, of the L - f u n c t i o n afforded b y the l - a d i c cohomologies
V~(A)®k®~ 4~

- Betti and de R h a m c o h o m o l o g y . Here we shall use the fact that the

curve A (if not its c o m p l e x multiplication) is a l r e a d y defined over ~.
B
Denote by HI(A) = H I ( A ( ~ ) , ~ ) the first rational singular homology of the
Riemann surface A(~) , with the H o d g e decomposition

B H -I'0 @ H 0'-I
H I (A) ®~ • =

Complex conjugation on A×~ induces an e n d o m o r p h i s m

F : HI(A) ~ (the "Frobenius at ~ " ) .

 21

B
Call HB the f i x e d p a r t of HI(A) under F , and let be a b a s i s of
this onedimensional G-vector space.

Let H R(A) = H D1R (A)V be the dual of the f i r s t a l g e b r a i c de R h a m c o h o m o -

logy of A over ~ , g i v e n w i t h the H o d g e filtration

DR +
H I (A) ~ F D {0}

where F + ®~ ~ H 0'-1 u n d e r the G A G A isomorphism over ¢:

I : H~(A) ®~ (1 > H I R (A) ®(~(E

I induces an i s o m o r p h i s m of onedimensional ~-vector spaces

i+ : +
HB(A) ®~ .... > (HIR(A)/F +) ®~

Then, ~I . i +(n) 6 H I R ( A ) / F + , for ~ defined by (2) . In fact,

= f 1 ~dx is a r e a l f u n d a m e n t a l p e r i o d of our curve, and so, up to

~* , Q is the d e t e r m i n a n t of the i n t e g r a t i o n - p a i r i n g

(HB(A) ®(~(~) x (H0(A,~ I) ®(~(Z) S > (~

calculated in t e r m s of R - r a t i o n a l b a s e s of b o t h spaces. This d e t e r m i n a n t

equals t h a t of the m a p I+ since H 0 ( A , ~ I) c H I (A) is the dual of
DR
H ~ R (A) /F +

Passing to tenser p o w e r s of the o n e d i m e n s i o n a l vector s p a c e s a b o v e we find

the p e r i o d s ~4~ occuring in (I).

In a sense, we h a v e c h e a t e d a little in d e r i v i n g the p e r i o d ~ f r o m the

cohomological setup: In the ~ t a l e c a s e we h a v e u s e d the a c t i o n of k via
complex multiplication to o b t a i n a o n e d i m e n s i o n a l situation (i.e., the
k-valued character ~ ). In the c a l c u l a t i o n of the p e r i o d , too, we s h o u l d
have considered H~R(A/k). = H~R(A)j ®~ k , e n d o w e d w i t h the f u r t h e r action
B
of k via c o m p l e x m u l t i p l i c a t i o n , and two c o p i e s of HI(A), i n d e x e d by
the two p o s s i b l e embeddings of the b a s e field k into • .... B u t in the
 22

presence of an e l l i p t i c curve over ~ , this would have seemed too a r t i -

ficial, and the g e n e r a l procedure will be t r e a t e d in § 4.

As a final remark about formula (I), it s h o u l d be n o t e d that it is p r o v e d

fairly easily. Any lattice F = I . (ZZ + ~ i ) gives a WeierstraB ~func-
tion such that

3
~' (z,F) = 4~(z,r) - g2lF) ~(z,F),

and for I = ~ we g e t g2(F) = 4. T h e rational numbers left unspecified in

(1) are then essentially the c o e f f i c i e n t s of the z-expansion of ~(z,F) .
It is t h e s e numbers that H u r w i t z studied in his papers.

§ 2. A l g e b r a i c Hecke Characters

Let k and E be t o t a l l y imaginary number fields (of f i n i t e degree over

~) , and w r i t e

Z = Hom (k,~) and T = Hom (E,~)

the sets of c o m p l e x embeddings of k and E. The group Gal(~/~) acts

on Z×T , transitively on e a c h individual factor. An algebraic homomor-
phism

8 : k* > E*

is a h o m o m o r p h i s m induced by a r a t i o n a l character

: Rk/~ (~m) > RE/~ (~m)-

This means that, for all • 6T, the c o m p o s i t e

• oS : k* > ~*

is g i v e n by

(4) ~oB (x) = ~E-~ O(x) n(~,~) ,

 23

for certain integers n(o,T) , such that n(~o,~) = n(a,T) for a l l

p6 G a l (~/~) .

Let k~ ,f ~,, > k ~ be the topological group of f i n i t e id~les of k -

i.e., those id~les whose components at the infinite places are I. F o r
x 6k*, let x also denote the corresponding principal idele in k~ ,and
xf the finite id~le obtained by changing the infinite components of x
to 1.

An algebraic Hecke character ~ of k with values in E , o f (infinity-)

type ~ , is a c o n t i n u o u s homomorphism

: k* > E*
~,f

s u c h that, f o r all x £ k*,

(xf) = B (x)

If ~ is t h e infinity-type o f an a l g e b r a i c Hecke character Y , then, by

continuity, 8 has to k i l l a subgroup of f i n i t e index of the units of k.
It f o l l o w s that the integer

(5) w = n(o,T) + n(co,T) = n(o,T) + n(o,cT)

(where c : complex conjugation on ~) is i n d e p e n d e n t of ~,T. It is

called the weight of

For any T6 T , w e g e t a c o m p l e x valued Gr~Sencharakter Toy w h i c h extends

to a q u a s i c h a r a c t e r of the id~le-class-group:

* ToY ~.
k]A,f .......... >

klA
* ToY ~.
klA/k. >

Consider the array of L - f u n c t i o n s , indexed by T:

 24

L (~,s) = (L(To~,S))~6 T ,

W
where, for Re(s)> ~ + I ,

I
#

the product being over all prime ideals p of k for which the
value ~(~p) does not depend on the choice of uniformizing parameter
wp of kp

The point s = 0 is c a l l e d critical for ~ , if for any T , no F-

factor on either side of the functional equation of L(To~,s) has a
pole at s = 0 . This is r e a l l y a property of the infinity-type B of
, for it turns out that s = 0 is c r i t i c a l for ~ if and only if
there is a disjoint decomposition

Z ×T : { (o,T) n(o,T) <0} 0 { (d,T) I n(co,7) < 0}

In o t h e r words, for every TET , there is a "CM-type" ~(TOB)C Z such

that

• (~ToB) = ~(ToB) ~ , for ~6 G a l ( ~ / ~ )

(6)

• O6~(~8) ¢¢ n ( d , T ) <0 ~ n(co,<) ->0

For ~ such that s =0 is c r i t i c a l Deligne defined an array of periods

~(~) : (Q(~}''T))T6T 6 ({*)T = (E®~{)* , and conjectured that

(7) L(~,0) 6 E c > E®{

~(~)

In other words, he conjectured that there is x 6 E such that for all

Y : E r~ >{ ,

L(To~,0) = T(X) . ~ ( ~ , T )
 25

The definition of ~(~) is d i s c u s s e d in § 4. It r e q u i r e s attaching a

motive to an a l g e b r a i c Hecke character.

§ 3. M o t i v e s

3.1 In t h e e x a m p l e of § 1, we c o n s t r u c t e d a "motive" for our Hecke

characters ~4m by taking tensor powers of HI (A) , i.e., a certain di-
rect factor of H4~(A4m), in the v a r i o u s cohomology theories. This
illustrates fairly well the g e n e r a l idea of w h a t a motive should be:
Starting f r o m an a l g e b r a i c variety over a number field, we have the
right to c o n s i s t e n t l y choose certain parts of its c o h o m o l o g y . Just
what "consistenly" means constitutes the d i f f e r e n c e between various
notions of m o t i v e . Here we shall be c o n c e r n e d with a fairly weak and
therefore half way manageable version: motives defined using "absolute
Hodge cycles" - see [DMOS], I and II. In this theory motives
can often be shown to be isomorphic when their L - func-
tions and periods coincide. A little more precisely, giving a homo-
morphism between two such motives M and N amounts to g i v i n g a fa-
mily of homomorphisms

HO (M) - - > H° (N) (Betti cohomology depends on the choice of

o : k - - > ~ yielding M}--> Mxo~ )
HDR(M) --> HDR(N)

H£ (M)--> H£ (N) (for all £ )

compatible w i t h all the n a t u r a l structures on these cohomology groups:

Hodge decomposition, Hodge filtration, Gal(k/k)-action, as w e l l as
w i t h the comparison isomorphisms between HB and HDR , HB a n d the
His •

3.2 L e t us state more precisely what a motive attached to a n a l g e -

braic Hecke character ~ should be[ - In the e x a m p l e o f § I, t h e c u r v e
A/~ defines the m o t i v e HI (A) over ~ whose L-function is L(~,s).
(This is r e a l l y what Gauss observed in 1814; nowadays this follows from
a result of D e u r i n g , which has been further generalized by S h i m u r a
[Sh I]...) But this is n o t w h a t we are looking for. The complex multi-
plication of A and therefore the Hecke character ~ are n o t v i s i b l e
 28

over ~ . That is w h y we considered A over k in o u r treatment of

the @tale cohomology, and used the field of v a l u e s of ~ (which again
happened to be k ) to obtain onedimensional Galois-representations,
and thus

Given a general algebraic Hecke character ~ like in § 2, a motive M

for ~ has to be a motive defined over the base field k such that
the field E acts on all the realizations of M in the various eoho-
mology theories, and such that for all Z, Hz(M) is an E®~i-module
of rank I with Gal (T/k) acting via ~ . The action of E on the
various realizations of M should of course be compatible with their
extra structures and with the various comparison isomorphisms. In o t h e r
words (see 3.1), E should embed into End M . Thus the rank-condition
on HI(M ) can also be stated by saying that Betti cohomology H (M)
should form a onedimensional E-vector space.

3.3 The typical example is HI (A) , for an a b e l i a n variety A/k with

E ~ ~®~ End/kA and 2 dim A = [E : Q]. The fact that these motives
always give rise to an a l g e b r a i c Hecke character was one of the main
results of the theory of complex multiplication by Shimura and Taniyama.
The Hecke characters occuring with abelian varieties of CM-type are
precisely those of weight -I such that n(o,7) 6 {-1,0} , for all
(o,T) 6 Z × T .

In fact, given such an algebraic Hecke character ~ of k with

values in E , we can assume without loss of g e n e r a l i t y that E
is t h e field generated by the values of ~ on the finite id~les of
k . Then E is a C M - f i e l d (i.e., quadratic over a totally real subfield),
and a theorem of C a s s e l m a n , [Sh 1], can be applied to get an a b e l i a n
variety A defined over k such that: . 2 dim A = [E : ~]

• there is an isomorphism
N

E > ~®~ End/kA

• HI (A) is a motive for ~ .

 27

3.4. When ~ has arbitray weight (%0) the homogeneity condition (5)
above still forces the infinity-type 8 to be of the form 6 = lilBi ,
with lweight (Bi) I : I , n i ( d , 7 ) 6 {±1,0} . Since twisting with finite
order characters is e a s y to c o n t r o l motivically one w o u l d naively expect
to be a b l e to assemble a motive for any given algebraic Hecke character
essentially as t e n s o r product of c o n s t i t u e n t s of the form HI(A) or
HI(A) like in 3.3.

There is h o w e v e r the n a s t y problem of c o n t r o l l i n g the f i e l d s of v a l u e s

E . For example, if k is i m a g i n a r y quadratic with class number h > I ,
then a Hecke character of k with {n(d,7) } = {-1,0} or
{n(J,T) } = {1,0} can never take all v a l u e s in E = k , but its h-th
p o w e r may.

Constructing a motive for the h-th power as an E - l i n e a r tensor power

of a m o t i v e for the c h a r a c t e r of w e i g h t ±I , one still has to s h o w t h a t
the field of coefficients E c a n be "descended" to k in w e i g h t ±h .

3.5 This "descent" o f the field of c o e f f i c i e n t s c a n be d e a l t w i t h

directly. B u t we g a i n m u c h more insight if we use a v e r y elegant
formalism due to L a n g l a n d s , [La] § 5, a n d D e l i g n e , [DMOS] IV. Langlands
defined a group scheme over ~ , the "Taniyama group" T , of w h i c h
Deligne was subsequently able to s h o w t h a t the category of its ~ - r a t i o n a l
representations is e q u i v a l e n t to the c a t e g o r y of t h o s e m o t i v e s as c a n
be o b t a i n e d (eventually after twisting by a character of finite order)
from abelian varieties over ~ which admit complex multiplication over
. Since the Taniyama group - along with many other beautiful proper-
t i e s - has, for every k , a certain subquotient Sk (isomorphic to a
group scheme constructed by Serre in [S£]) whose irreducible represen-
tations are g i v e n precisely b y the algebraic Hecke characters ~ of k,
we "find" the m o t i v e attached to a g i v e n ~ by lifting the c o r r e s p o n -
ding representation of Sk back to the subgroup of T whose repre-
sentations give the m o t i v e s defined over k .

3.6 SO, for e v e r y algebraic Hecke character ~ of k with values in

E , a motive over k equipped with an E - a c t i o n c a n be c o n s t r u c t e d from
CM-abelian varieties over k, w h o s e £-adic Galois representations are
onedimensional E ®Q£-modules given by ~ . Furthermore, the T a t e - c o n j e c -
ture would imply that the £-adic realizations determine a motive up to
 28

isomorphism - even in the strictest sense of "motives" (algebraic cycles).

As we are dealing with motives for absolute Hodge cycles, it is p e r h a p s
not too surprising that one can actually prove: in the category of m o -
tives that can be obtained from all abelian varieties over k (not
necessarily CM) , any two motives attached to t h e same Hecke character
are actually isomorphic - see [Sch], I. Still, this does not seem to be
known in a n y larger category of motives. In fact, it h i n g e s on Deligne's
theorem that "every Hodge cycle on an a b e l i a n variety over an algebra-
ically closed field is a b s o l u t e l y Hodge" - see [DMOS], I. A n y w a y ,
whenever we find two motives constructed from the cohomology of abelian
varieties that belong to the same Hecke character they will have the
same periods...

§ 4. Periods

As in t h e example of § I, p e r i o d s are going to a r i s e from a comparison

of the Betti and de Rham cohomology groups of o u r motive. So, let us
first look at these cohomologies more closely in the case of a motive
for an algebraic Hecke character. We are going to use some facts which
are well-known for the cohomology of a l g e b r a i c varieties, and which
carry over to m o t i v e s .

4.] As in § 2, let k and E be totally imaginary number fields,

and ~ an a l g e b r a i c Hecke character of k with values in E . Let
M be a motive over k attached to ~ (in the sense of 3.2 above) .
Then for any embedding o6Z , the singular rational cohomology H (M)
is an E - v e c t o r space of dimension 1. T h e E-action respects the Hodge-
decomposition

Ho(M) ® ~ : HP,q
P,q

H (M) ® ~{ is an E ®~{ = {T - m o d u l e of rank I. (~ and T were de-

fined at the beginning of § 2.)

Starting from the special case where M = HI (A) with an a b e l i a n variety

A/k of CM-type, and using the uniqueness of the motive attached to a
Hecke character (see 3.6) , o n e finds that, for any embedding T 6 T ,
 29

the direct factor of Ho(M) ®~ on which E acts via T lies in

Hn(o,x) , w-n(c,T)

(The n(o,T) are given by the infinity-type of ~ : see § 2, formula

(5) .)

4.2 Let us note in p a s s i n g that, if M(~) and M(~') are motives

for Hecke characters ~ and T' of k with values in E , then the
following are equivalent:

• M(~) ~ M(~') over ~ .

• For some 06~ , H @ ( M ( ~ ) ) ~ H (M(T')) , as r a t i o n a l Hodge-structures.

• ~ and ~' have the same infinity-type B .

4.3 Coming back to o u r motive M for ~ , suppose now that s = 0

is c r i t i c a l for ~ (see § 2, formula (6)), and consider the comparison
isomorphism

I : @ H u(M) ® ~ { > HDR(M) ®~

Note that HDR(M) is b y definition a k-vector space, and that

k®~ ~ {Z . So, I is a n isomorphism of k ® E ®{ - modules of rank 1.

For ~EZ , let e be an E - b a s i s of H (M) , and put e : ( ~ ® I)~6 ~

On the right hand side, choose a basis ~ of HDR(M) over k ®~E ,
and decompose

L0 : ~ £0 r
(@,T)6ZxT 0,7

with ~o,T 6 T-eigenspace of H D R ( M ) ®kl { . Writing I(e ) = Z I(e~) T

T6T
for the corresponding decomposition of I(e) , we find for all
(0,T)EZxT that
 30

~ ,~ = p(O,T) . I(eo) T , for some p(O,7) 6 {*

The unit

C(~*) ZxT = (k®E®~)*

(p(~,T))(~,T)E E×T

gives the "matrix" of I and, up to m u l t i p l i c a t i o n by (k®E)* , de-

pends only on ~ .

4.4 Modulo such a factor one has the relation

(8) p(o,T) • p(co,T) N (2 z i ) w

This amounts essentially to L e g e n d ~ e ' s period relation, and can be

proved in o u r context (using uniqueness of motives for Hecke characters)
from the identity ~ = ~w . _ The motive ~(-I) attached to the
norm character is discussed in more detail, e.g., in [DI], § 3. For (8),
it is e n o u g h to know that ~(-1) is a motive defined over ~ , with
coefficients in ~ such that

I
HB(~(-I)) = ~ ~ and HDR(~(-I)) = ~ ,

with trivial comparison isomorphism. Incidentally, ~(-I) has no cri-

tical s , if c o n s i d e r e d over a totally imaginary field k .

With (8) a n d 3.4, c a l c u l a t i n g the p ( o , T ) 's (or their inverses) usually

reduces to i n t e g r a t i n g holomorphic differentials on which E acts via
7 or cT .

4.5 In terms of these p(O,T) , Deligne's period ~(%~)C(E®~)*/E*

(see (7) above) can be defined componentwise by

-I
(9) ~(~{,T) = D(~') T - o6~(~ToB) p(o,'~)

For the definition of the "CM-types" ¢('~oB) , see § 2, formula (6).

 31

Note that the product in (9) is, in fact, well-defined up to a factor

in (E ® 1)* - One definition of the "discriminant factor"
D(~) = (D(~)T)T 6 T can be found in [DI], 8.15. This factor arises when
one computes the cohomology of Rk/Q M by the K~nneth formula: among
other things, one has to choose an ordering of Z . A definition of
D(~) which was born out by these cohomological computations - cf. [Ha],
esp. 2.4.1 and Cor. 5.7.2B - is as follows. Start with one TET , and
let K cE ~ be the fixed field of

{p 6 G a l ( ~ / ~ ) I P @ (T°O) : ~(~o~) }

G a I ( ~ / K T) permutes the set @('loB) . Let L T mK be the fixed field

of the kernel of the character

G a I ( ~ / K T) > ~(~(ToB)) s@n> {-+I}

Then [L : K ] < 2 , and L = K[(D(~) ) for some D(~) T with

2 ~ ~[ T
D (~) 6 K*
T T

NOW, any p 6 Gal(~/~) induces a permutation of the set of infinite

places of k : both ~(To~) and O(pTo6) are in b i j e c t i o n with this
set. Call 8(p) the sign of this permutation. Then we set

D(~) = a(p) (D(~))P

pT T

The array (D(T) T ) < 6 T is independent, up to a factor in (EOI)* ,9f

the choices made in d e f i n i n g its components.

Let us list some properties of D(%') - ef. also [Sch].

4.6 a) D(~) depends only on k,E , and the collection of "CM-types"

{¢(~o~) I • CT}
2
b) m(~) 6(E® I)* c ( E ~ ) *
c) If k is a CM-field, with maximal totally real subfield ko,
then D(~) N 6 d i s c r ( k o T , up to a factor in (E® I)*
d) Let F/k be a finite extension of degree n . Then
 32

c ('[o 8) TIT

up t o a f a c t o r in (E ® I)* . Here, the r i g h t h a n d side m e a n s the

following:

Let d(k*) 2 6 k * / ( k * ) 2 be the relative discriminant of F/k . For any

infinite place v of k , choose a square root 6 = /d-6k* . For
V V
6Z , let Icl be t h e infinite place of k determined by ~ and c~ ,
w
and denote by o(6toi) £ 6" the w e l l - d e f i n e d image of 61~16 klo I under
the continuous isomorphism k j o I --~--~
> { given by ~ . - Note that
changing the r e p r e s e n t a t i v e of d or the signs of 6 , a t some p l a c e s
V
v , multiplies the r i g h t h a n d side of o u r formula only bya factor in
(E ® I)*

Assume the situation of 4.6,d) . F r o m the v e r y definition of the p(c,~) ,

a n d the p r o p e r t i e s 4.6,a) a n d d), one finds the following formula for
the b e h a v i o u r of the p e r i o d s under extension of the b a s e field:

(~ ONF/k) D (~ONF/k)
A(F/k,B) • -
(~n) D (~n)

(10)
D(~°NF/k) D(~) n n-1
= 77 ~(6 • D(~)
D(p) n D ( ~ n) Ic I T 6 T

The array A(F/k,B) 6 (E~6)* will reappear in the second theorem of

§ 5 below. Note that, if k is a C M - f i e l d , the second factor of
A(F/k,B) can be evaluated by 4.6,c). Both factors of A are already
present in [Ha], in the case n = 2 , although the formalism there is
still somewhat clumsier than the one employed here.

4.7 Let us c l o s e this section with a few words o n the b e h a v i o u r of

our p e r i o d s under twisting. For the T a t e twist, one finds

(11) 2 ( ~ . ~ m) ~ (2 ~i) m ~(~)

 33

If ~ is a c h a r a c t e r of finite order K* on
/k* with values in E*
~,f
one passes from ~(T) to Q(~T) by leaving D(T) unchanged, and
multiplying the p(0,T) by c e r t a i n algebraic numbers with eigen-
properties under ~ . The details c a n be f o u n d in [Sch]. All we n e e d
to k n o w is the following invariance lemma:

If F is a f i n i t e extension of k, X a character of f i n i t e order on

F~,f* /F* with values in E* , and ~ the r e s t r i c t i o n of X to
k~,f* (in o t h e r words, considering X and ~ on Gal(k/F) , Gal(k/k) ,
resp., via class field theory, ~ : Xo V e r , where
Vet : Gal(kab/k) --> Gal(Fab/F) is the transfer map), then

~(X" (To NF/k) ) ~q(TONF/k)

(12) : £ (F/k,B)
2 ( ~ - ~ n) ~q(Tn)

Let us m e n t i o n in p a s s i n g that the p r o o f of (12) also shows that the

quotients

~(T)

may always be e x p r e s s e d by G a u s s sums.

§ 5. The rationality conjecture for Hecke L-functions

The p r o o f of D e l i g n e ' s conjecture (see e n d of § 2) for the c r i t i c a l

values of L - f u n c t i o n s of a l g e b r a i c Hecke characters falls into two parts:
The case where the b a s e field k is a C M - f i e l d is t r e a t e d first. From
there one passes to the g e n e r a l case by a t h e o r e m about the b e h a v i o u r
of s p e c i a l values under extension of the base field.

(I) Let us b r i e f l y describe the C M - c a s e :

Historically, the m a i n idea for the C M - c a s e goes back to Eisenstein.

But it w a s Damerell who, in his thesis [Da], published the f i r s t
 34

comprehensive account of a l g e b r a i c i t y results for c r i t i c a l values of

Hecke L-functions of imaginary ~uadratic fields. He a l s o a n n o u n c e d
finer rationality theorems in t h a t c a s e , but never published them.
(The c a s e of imaginary quadratic k was later settled completely in
[GS] a n d [GS'].) I n the F a l l of 1974, Andr@ Well gave an e x p o s i t i o n of
w o r k of E i s e n s t e i n and Kronecker including, among other things,
Damerell's theorem as an a p p l i c a t i o n . This course at t h e IAS - w h i c h w a s
later on developed into the b o o k [WEK] - inspired G. S h i m u r a to g e n e -
ralize Damerell's algebraicity results to c r i t i c a l values of H e c k e L-
functions of a r b i t r a r y CM-fields: [Sh 3] (At t h a t p o i n t , he still
needed a technical assumption o n the infinity-type of the H e c k e cha-
racter.)

To e x p l a i n the starting point of t h i s method of p r o o f , recall our

I
example in § I: t h e L-value there appeared (up t o a f a c t o r of ~ ) as
an E i s e n s t e i n series :

4~
a,b6Z (a+bi)

relative to the lattice Z + Zi . Now, sometimes the relation between

L-value and Eisenstein series is not q u i t e as straightforward - e.g.,
if, in § I, we w e r e to study the v a l u e s L(~a,0) for integers a # 0
such that s = 0 is c r i t i c a l for ~a , t h e n we w o u l d h a v e to t r a n s -
f o r m the E i s e n s t e i n series by c e r t a i n (non h o l o m o r p h i c ) differential
operators. But except for s u c h o p e r a t o r s it r e m a i n s true that, in a n y
pair of c r i t i c a l values o f an H e c k e L-function of a C M - f i e l d k which
are symmetric with respect to the f u n c t i o n a l equation, there is a v a l u e
which c a n be w r i t t e n as a l i n e a r combination of E i s e n s t e i n series (viz.,
Hilbert modular forms with respect to the m a x i m a l totally real subfield
of k ), r e l a t i v e to lattices in k .

When k is i m a g i n a r y quadratic, the a l g e b r a i c i t y properties of the

Eisenstein series can be d e r i v e d directly from explicit polynomial re-
lations among them (see, e.g., Well's treatment of D a m e r e l l ' s theorem
in [WEK]). But in g e n e r a l the p r o o f of t h e i r algebraicity depends on a
theory of c a n o n i c a l models for the Hilbert modular group (as in [Sh 3])
or, equivalently, o n an a l g e b r a i c theory of H i l b e r t modular forms.
 35

This latter approach was used by Katz in [KI], [K2]. Just like Shimura,
Katz did not stop to look at m o r e precise rationality theorems about
the special values he had determined up to a n a l g e b r a i c number. In fact,
Katz' main concern was with integrality properties and p-adic inter-
polation.

When Deligne formulated his conjecture in 1977 he felt the need to c h e c k

that, up to a f a c t o r in ~* , it p r e d i c t e d Shimura's theorem. This
turned o u t to be a confusing problem, for the following reason. Shimura
expresses the L-values in t e r m s of periods of abelian varieties con-
structed from lattices in k , which therefore have complex multiplica-
tion by k , and are defined over some number field E' . On the other
hand, the L-function in q u e s t i o n is t h a t of a Hecke character of the
field k , with values in some number field E . The motive of such a
character arises from abelian varieties defined over k , with complex
multiplication b_~ E (or s o m e field closely related to E ). T h i s
double role of k as field of definition and of coefficients was dealt
with by Deligne - up to factors in ~* - by an ad hoc dualization,
see [DI], 8.19. (Its refinement for more precise rationality state-
ments remained the most serious obstacle in the attempt to p r o v e
Deligne's conjecture made in [Sch I].)

Don Blasius managed to solve this problem by writing down an analogue

of Deligne's dualization on the level of m o t i v e s over k , resp. E :
his "reflex motive". Thus he was able to p r o v e

Theorem I: Let k be a CM-field, and ~ a Hecke character of k ,

with values in some CM-field E . If s = 0 is c r i t i c a l
for ~ , then

L('~,o) 6E ~ > E®¢ .

~(~)

(Note that any algebraic Hecke character of any number field takes
values in a C M - f i e l d . )

As B l a s i u s ' paper [B] is a b o u t to be available we shall not enter into

 36

describing the technique of his proof in d e t a i l . Suffice it to say

that, apart f r o m the "reflex motive" mentioned above, he n e e d s , of
course, a very careful analysis of the b e h a v i o u r o f the E i s e n s t e i n
series under Gal(~/~) (i.e., Shimura's reciprocity law in C M - p o i n t s ) ,
and also the e x p l i c i t description - due t o T a t e and Deligne - of the
action of Gal(~/~) on abelian varieties of C M - t y p e : see [LCM],
chapter 7.

(II) We shall now describe a little bit more in d e t a i l the second

part of the p r o o f of D e l i g n e ' s conjecture for H e c k e L-functions. It
relies on a g e n e r a l i z a t i o n of [Ha], § 3, from GL 2 to GL n , and
might n o t be p u b l i s h e d completely before some time.

Consider the following situation: Let k be a t o t a l l y imaginary num-

ber field, and F/k a finite extension of d e g r e e n ~ 2 . Let ~ be an
algebraic Hecke character with values in a n u m b e r field E , of in-
finity-type 8 . Assume s = 0 is c r i t i c a l for Y . Let
X : F~* /F* > E* be a c h a r a c t e r of finite order, and put ~ =Xlk~* '
like in § 4.7 above. Recall the array

A(F/k,8) : (A(F/k, Yo8))76T

defined in § 4.6, formula (10).

Theorem 2:

L F(X- (~ONF/k) ,0)

A (F/k, B) 6E m > E®~
Lk(~-~n,0)

R e m a r k s : (i) A s the E u l e r product for L(T,s) converges for

w
Re(s) > ~ + I , a n d s = 0 is c r i t i c a l for Y , it is w e l l - k n o w n that
the d e n o m i n a t o r in the theorem is n o t zero.

(ii) Here is h o w t h e o r e m s I and 2 imply Deligne's conjecture for all

critical values of all H e c k e L-functions: Given any totally imaginary
 37

number field F , and any Hecke character ~ of F , with values in

a number field Eo , of infinity-type 8o , the homogeneity condition
(5) of § 2 forces Bo to factor through the maximal CM-field k
contained in F :

8o = ~o NF/k ,

for some algebraic homomorphism

8 : k* > E'o*

Choose a Hecke character ~{ of k with infinity-type B , write

= X -( ~ o N F / k ) , for some finite order character X of F , and
choose E mE big enough to contain the values of T as well as those
o

of X - Define e : Xlk* . Put n = [F : k] . By theorem I,

L(~0"T n,0) 6 E
~ ( w . y n)

But we know the behaviour of the periods Q under twisting and base
extension: see end of § 4. Theorem 2 therefore implies that

L(~,0) L(X-( ~ o NF/k),0)

6E > E@{ •
[2(X- ( T o N F / k ) )

Finally, E may now be replaced by Eo because Deligne's conjecture

is invariant under finite extension of the field of coefficients:
[D1], 2.t0.

This gives Deligne's conjecture for Hecke L-functions of totally ima-

ginary number fields. These are the only fields with honest regard
Hecke L-functions. But it should be said, for the sake of completeness,
that Deligne's conjecture for Hecke (=Dirichlet) L-functions of totally
real fields follows from results of Siegel's (cf. [DI], 6.7) and, in
the case of number fields which are neither totally real nor totally
 38

imaginary, no H e c k e (=Dirichlet) L-function has any c r i t i c a l value.

The r e m a i n d e r of this section is d e v o t e d to s k e t c h i n g the p r o o f of

theorem 2. Let us set up some n o t a t i o n .

We c o n s i d e r the f o l l o w i n g algebraic groups over k :

Go/k : GLn/k -

To/k = standard maximal torus

Bo/k = standard Borel s u b g r o u p of u p p e r t r i a n g u l a r matrices,

and the two m a x i m a l parabolic subgroups

.~
<>ik : (Op) p 6 G L n _ I , t 6 GL I }

q 6 GLn_ 1 , t 6 GL I }

Dropping the s u b s c r i p t zero will m e a n taking the r e s t r i c t i o n of s c a l a r s

to ~ . So,

G/(~ : Rk/,~ (G O/k)

and so on.

We i n t r o d u c e the two c h a r a c t e r s

7P : g =
<0 } >
tn
det(g)

and

tn
 39

which we v i e w as c h a r a c t e r s on the t o r u s extending to Po (resp. Qo )•

The representations of Go/k with highest weight ~Tp (resp. ~TQ )
are the ~-th (resp. b-th) symmetric power of the standard repre-
sentation of Go/k on kn (resp. its dual (kn) ~ )
Coming back to the situation of t h e o r e m 2, d e f i n e a homomorphism

: P(~ih,f) = Po(klA,f) > E*

by

> ~I (tf) ~ (det(gf))

We r e q u i r e that the c e n t r a l character of ~ be our ~ . This means

that ~I is d e t e r m i n e d by

"" "t = w(tf)_ = ~I (tf)_ ~ (tf)

n_

We may view • as an " a l g e b r a i c Hecke character" on P/~ , and it

has an i n f i n i t y - t y p e

type (~) = 7 6 H o m ( P / ~ , RE/~({m))

Hence we get an a r r a y of types, indexed by T 6 T , with components

type(To~) = 'toT 6 HOrn(P,{ m)

Recall that

H o m ( P , G m} = @ H o m ( P o ,~m } ,
gCZ

and that the type 8 of is g i v e n by the integers n(a,<) - see § 2.

 40

It is then easy to c h e c k that

TOy : (n(o,T) .yp)o6E ,

for e v e r y • 6T .

Given T , define an array of d o m i n a n t weights

i(~) = (l(o,<))oE Z

by the rule

{ (-n(@,~)-1)yQ

n(o,~)yp
if

if
n(o,T) < 0

n(o,T) > 0

This affords a representation

p : Gx~ = ~ (GLn/k) > GL(M(A(T))) ,

oGZ

where M(A (T)) =® M(I(O,T)) , M(I(O,T)) b e i n g the r e p r e s e n t a t i o n

o6Z
with highest weight k(O,T) . The s y s t e m {M(A(T))~6 T is a ~ - r a t i o -
nal system of r e p r e s e n t a t i o n s in the sense of [Ha], 2.4 - i.e., the
representations are conjugate under Gal(~/~)

As in [Ha], we study the cohomology of c o n g r u e n c e subgroups of GLn(0)

with coefficients in these modules: Form the q u o t i e n t s

SK = G (~) ~ G ((~IA) / K o'Kf

where K ~ u(n)Z~ is a s t a n d a r d maximal compact subgroup, times

the centre o~bG(~) = G , and where Kf is o p e n compact in G(~ ,f) •
The m o d u l e s M(i(7)) provide coefficient systems M(I(T)) on SK ,
and we c o n s i d e r the ~ - G(~,f)- module

H'(~,M(A(T))) : = lira H'(SK, M(A(T)))

Kf
 41

The embedding of SK into its B o r e l - S e r r e compactification SK is a

homotopy equivalence. The boundary ~S K of this compactification has
a stratification, with strata corresponding to the conjugacy classes
of p a r a b o l i c subgroups of G/~ . The stratum of l o w e s t dimension,
SB SK , corresponds to the c o n j u g a c y class of B o r e l - s u b g r o u p s . The
coefficient system can be e x t e n d e d to the b o u n d a r y , and the limit

H" ( ~ B ~ , M(A(T))) : lim H" ( ~ B S K , M(A(T)))

- - >

Kf

is a g a i n a G(~,f) -module. The diagram

SK ~ i > SK < ~B S K

induces a G(~lh,f)-module homomorphism

rB : H" (~, M(A(T))) > H" ( $ B ~ , M(A(~)))

Just as in [Ha] ,II, the right hand side turns out to be a d i r e c t sum
of modules, induced from an a l g e b r a i c Hecke character

n : B(QIA,f) > ~*

T(~IA, f)

on B(~,f) , up to G(~,f } . The types of these characters are de-

termined by K o s t a n t ' s theorem, [Ko] ; cf. [Ha], II, for n = 2 . In
particular, it is e a s i l y checked that the following induced module
(for • as above, and T 6 T ) is c o n t a i n e d in the cohomology of ~B ~ :
 42

G (QIA, f)
V = Ind
B (QIA, f)

\
h is C , and )

h(bf_gf) = (TOO) (bf). h(if) ,

=lh:G(~,f) --> for all bf 6 B(~ih,f) and

_gf £ G(~IA, f)

(Here, "C " means right invariance under a suitably small open com-
pact subgroup in G(~,f) .)

More precisely, we have

T > H(n-3}do (~B ~ , M(A(T)))

VTo 0

1
where d o : ~ [k : ~] , and the system of maps {iT]<6 T is Q - r a t i o n a l

with r e s p e c t to the two o b v i o u s Q-structures on the systems on both

sides.

Consider the n o n - t r i v i a l submodule

J~o~ = I n d G ( Q ~ ' f ) To~ c V oo

P(~,f)

Obviously, {JToO}T£T is a Q - r a t i o n a l system of G(Q~,f) -submodules

of H ( n - 1 ) d ° ( $ B ~ , M(A(7))) . The first essential step of the p r o o f is

to c o n s t r u c t a Q-rational "section" of rB ,

Eis7 : JTo~ > H(n-1)d° (S, ~{A(T))} ,

for all TCT . Thus, rB o Eis = Id on J . This section is con-

7 T°O
structed first over ~ by m e a n s of r e s i d u a l E i s e n s t e i n series or, in
other words, non c u s p i d a l Eisenstein series attached to P/Q . To prove
that {EisT}TC T is d e f i n e d over ~ one has to use a m u l t i p l i c i t y one
argument, like in [Ha], III. But here this is more complicated. One
 43

has to use the spectral sequence which computes the cohomology of the
boundary in t e r m s of the c o h o m o l o g y of the strata. Then the c o h o m o l o g y
has to be r e l a t e d to a u t o m o r p h i c forms, and one has to a p p e a l to re-
sults of J a c q u e t - S h a l i k a on m u l t i p l i c i t y one, and of J a c q u e t on the
discrete non cuspidal spectrum.

Once we have the m o d u l e s

Eis z (J o~) c H(n-1)do (S, M(A(T) )

we can p r o c e e d more or less in the same way as in [Ha], V: We c o n s t r u c t

an e m b e d d i n g

iH : F* > GLn(k)

H being the torus with H(Q) = ill(F*) Using this torus we can con-
struct homology classes (compact modular symbols)

ZIi H, 7 o X, g) C H (~, M(A(~)))

(n-1)d o

depending on a p o i n t _g6 G(~]A) and on a finite order character

* /F*
X : F]A > E*

whose restriction to k ~* should be ~ .

As in [Ha], V, we get an intertwining operator

G ( ~ m , f)
Int(Z(iH,X)) : JTo ~ > Ind
H(~m, f )

by evaluating E i s T ( J T o Q) on Z(iH,X,~) . There is a n o t h e r intertwi-

ning operator

G ((~]A,f)
int l°c : J > Ind T oX ,
To~
H (~IA, f)
 44

constructed as a p r o d u c t of l o c a l intertwining operators. Both opera-

tors are U-rational, and for some x C E* we f i n d that, for all ~6T ,

L F ( T O (X" (~ONF/k) ) ,0) intlOc

Int(Z(iH,TOX)) = T(X) A (F/k,ToS)
L k ( T ° (~.~n) ,0)

This implies theorem 2. - T h e factor x 6 E* can actually be g i v e n more

explicitly.

§ 6. A formula of L e r c h

The fact that a Hecke character determines its m o t i v e up to isomorphism

produces a period relation whenever two different geometric construc-
tions of a motive f o r the same character c a n be g i v e n . We h a v e seen a
first example of t h i s p r i n c i p l e in f o r m u l a (8) o f § 4. T h e p e r i o d s
p(u,T) occuring in t h i s formula comprise those for w h i c h Shimura [Sh2]
has proved various monomial period relations (up to an a l g e b r a i c number).
These monomial relations were reproven, by m e a n s of m o t i v e s over ~ ,
by D e l i g n e , [D2]. They c a n be r e f i n e d using the above principle. But
we leave aside here this application,as well as some others, referring
the r e a d e r to [Sch]. Instead, let us c o n c e n t r a t e on a typical case in-
volving G . Anderson's motives for J a c o b i - s u m Hecke characters.

Let K = ~(/i-~-) be an i m a g i n a r y quadratic field of d i s c r i m i n a n t -D

Assume for simplicity that D > 4 . Recall the construction of the
simplest Jacobi-sum Hecke character of K , in the sense of [WIII],
[1974 d]: K is c o n t a i n e d in ~ ( ~ D ) , the f i e l d of D-th roots of ] .
Write n : [~(~D ) : K] - ~(D)2 " F o r a prime ideal P of ~(~D ) not
dividing D , put

G(P) : -- [ XD, P (x) "A~ (x) ,

x 6Z [~D]/P

with "ZD, ~ (x) ---x (I~P-I)/D (cod P )" the D th- power residue symbol

cod P , and l(x) : exp(2~i • tr( ~ [~D]/p)/]F p (x))

 45

Then extend the function of prime ideals p of K with p ~D :

J(p) = ~ G(P) ,

multiplicatively to all ideals of K prime to D . Elementary proper-

ties of Gauss sums show that J takes values in K . By a theorem of
Stickelberger and an e x p l i c i t version of the analytic class number for-
mula for K one finds that, if J is an algebraic Hecke character,
then its infinity-type

: K* > K*

is g i v e n by

in+h)/2 -(n-h)/2
X ~ > Z X t

where h is the class number of K . (Note that n and h have the

same parity, by genus theory.) In o t h e r words, if J is a H e c k e cha-
racter, then

(13) j. IN- (n+h)/2 .%,h

for some character ~ of K~ of finite order and some Hecke character

of K of w e i g h t -I

That J is in f a c t a Hecke character, i.e., is w e l l - b e h a v e d at the

places dividing D , if v i e w e d on id~les, was proved by Weil (loc. cit) .
But it c a n also be deduced, e.g., from the following construction of a
motive for J which was given by Greg Anderson, [AI],[A2].

Anderson finds a motive defined over K , with coefficients in K ,

whose l-adic representations are given by J , in H n-2 of the zero-

set Z of the function

 46

~(~D ) > K
x ~----> tr~ (~D) / K(xm)

viewed as a p r o j e c t i v e variety in ]PK (Q(~D)) , the p r o j e c t i v e space

of the K - v e c t o r space ~(bD ) . N o t e that

{ D .+xD 0}cipn-1
Z X K ~ ( D D) = x1+.. n =

Anderson's construction is of c o u r s e motivated by the well-known fact

that Fermat-hypersurfaces contain motives (carved out by the action of
their large automorphism groups) attached to J a c o b i - s u m Hecke characters
of c y c l o t o m i c fields: see [DMOS], pp. 79 - 96. For details of A n d e r s o n ' s
more general construction, we r e f e r to his preprints, or to [Sch].

At any rate, thanks to A n d e r s o n ' s work, we h a v e at our d i s p o s a l a motive

M(J) for the character J , which lies in the c a t e g o r y of m o t i v e s ob-
tained from abelian varieties. (This last fact is p r o v e d by S h i o d a -
induction: [DMOS], p. 217). Thus, by (13), the p e r i o d s of the m o t i v e

M ( J I N - (n+h)/2) : M(J) @ K K ( ( n + h ) / 2 )

will be the same as those of any m o t i v e constructed for the c h a r a c t e r

. ~h

The p e r i o d calculations on F e r m a t - h y p e r s u r f a c e s always reduce even-

tually to B e t a - i n t e g r a l s . For M(J) one e s s e n t i a l l y gets the p r o d u c t

7T F {<~>)-
a I
X (a)=-1

Here, X(P) : (~-

-m) is the D i r i c h l e t character of the q u a d r a t i c field
K , and the p r o d u c t is t a k e n over those a6 (Z/D Z )* for w h i c h
a
X(a) = -I . <~> is the r e p r e s e n t a t i v e of the class ~ mod Z which
lies between 0 and I
 47

A motive for 5.~h can be b u i l t up from elliptic curves with complex

multiplication by K . - Assume for s i m p l i c i t y that ~ONH/K takes
values in K* , for H the H i l b e r t class field of K . Choose any
elliptic curve A/H such that HI(A) is a m o t i v e for ~ONH/K , and
call B = RH/KA its r e s t r i c t i o n of s c a l a r s to K . Calling E the
f i e l d of v a l u e s of ~ , HI(B) ® E h can be shown to be a motive for
~h (viewed as t a k i n g values in E ) : cf. [GS], § 4. U s i n g formulas
derived in [GS], § 9, the p e r i o d s of t h i s m o t i v e c a n be computed in
terms of the p e r i o d s % of the c o n j u g a t e s AJ/H of o u r e l l i p t i c
curve A , for

d 6 GaI(H/K) : CI(K)

Straightening out the twists by the n o r m a n d the finite order character

(cf. § 4.6 and 4.7), one finally obtains the following relation, up to
a factor of K* :

a
6 C£ (K) ~ X (a) =-I r (<~>)

where y generates the a b e l i a n extension of K belonging to ~ .

Multiplying (14) w i t h its c o m p l e x Conjugate, we g e t

(15
d 6 C£ (K) a6 (~/m~) *

for some z with z 4 6 ~* . Except for the d i f f e r e n t interpretation

of z a n d the ~ , this is the e x p o n e n t i a l o f an i d e n t i t y proved
analytically by Lerch in [Le], p.303. The first geometric proof of
(15), up to a f a c t o r in ~* , was given by G r o s s in [Gr], a paper
which in t u r n inspired Deligne's proof of the theorem about absolute
Hodge cycles on a b e l i a n varieties - which again is e s s e n t i a l in

proving uniqueness of the m o t i v e for an a l g e b r a i c Hecke character.

 48

REFERENCES

[All G. A n d e r s o n , The m o t i v i c interpretation of J a c o b i sum Hecke

characters; preprint.

[A2] G. A n d e r s o n , C y c l o t o m y and an extension of the T a n i y a m a

group, preprint.

[B] Don Blasius, On the critical values of H e c k e L-series; preprint

[Da] R.M. D a m e r e l l , L - f u n c t i o n s of e l l i p t i c c u r v e s w i t h c o m p l e x
m u l t i p l i c a t i o n . I, A c t a A r i t h m . 17 (1970), 2 8 7 - 301; II,
A c t a A r i t h m e t i c a 19 (1971), 311 -----317.

[ml] P. D e l i g n e , V a l e u r s de fonctions L et p @ r i o d e s d ' i n t 6 g r a l e s ;

Proc. Symp. Pure Math. 33 (1979), p a r t 2; 313 - 346.

[D2] P. D e l i g n e (texte r @ d i g 6 p a r J.L. B r y l i n s k i ) , C y c l e s de H o d g e

a b s o l u s e t p 6 r i o d e s des i n t 6 g r a l e s d e s v a r i @ t @ s a b 6 1 i e n n e s ;
Soc. Math. France, M~moire n~2 (2 @ m e s~r.) 1980, p. 23 - 33.

[DMOS] P. D e l i g n e , J. M i l n e , A. O g u s , K. Shih, H o d g e C y c l e s , M o t i v e s
and Shimura V a r i e t i e s ; S p r i n g e r Lect. N o t e s M a t h . 900 (1982).

[Ei] G. E i s e n s t e i n , U b e r die I r r e d u c t i b i l i t ~ t u n d e i n i g e a n d e r e
E i g e n s c h a f t e n d e r G l e i c h u n g , y o n w e l c h e r die T h e i l u n g der
g a n z e n L e m n i s c a t e a b h i n g t , - a n d the s e q u e l s to this p a p e r -;
M a t h . W e r k e II, 536 - 619.

[GS] C. G o l d s t e i n , N. S c h a p p a c h e r , S @ r i e s d ' E i s e n s t e i n et f o n c t i o n s
L de c o u r b e s e l l i p t i q u e s ~ m u l t i p l i c a t i o n c o m p l e x e ; J.r. ang.
Math. 327 (1981), 184 - 218.

[GS'] C. G o l d s t e i n , N. S c h a p p a c h e r , C o n j e c t u r e de D e l i g n e e t F-
h y p o t h @ s e de L i c h t e n b a u m sur les c o r p s q u a d r a t i q u e s i m a g i -
n a i r e s . C R A S P a r i s , t. 296 (25 A v r i l 1983), S@r. I, 6 1 5 - 6 1 8 .

[Gr] B.H. G r o s s , O n the p e r i o d s o f a b e l i a n i n t e g r a l s a n d a for-

m u l a of C h o w l a a n d S e l b e r g ; I n v e n t i o n e s M a t h . 45 (1978),
193 - 211.

[Ha] G. H a r d e r , E i s e n s t e i n c o h o m o l o g y of a r i t h m e t i c groups - The

case GL 2 ; p r e p r i n t B o n n 1984.

[Hu] A. H u r w i t z , U b e r die E n t w i c k l u n g s k o e f f i z i e n t e n der l e m n i s -

k a t i s c h e n F u n k t i o n e n ; f i r s t c o m m u n i c a t i o n in: N a c h r . k. Ges.
W i s s . G ~ t t i n g e n , M a t h . Phys. KI. 1897, 273 - 276 = M a t h .
W e r k e II, n ° LXVI, 338 - 341. P u b l i s h e d in e x t e n s o : Math. Ann.
51 (1899), 196 - 226 = M a t h . W e r k e II, n ° L X V I I , 342 - 373.

[KI] N. K a t z , p - a d i c i n t e r p o l a t i o n o f r e a l a n a l y t i c Eisenstein
series; Ann. M a t h . 104 (1976), 459 - 571.

[K2] N. K a t z , p - a d i c L - f u n c t i o n s for CM-fields; Inventiones math.

49 (1978); 199 - 297.
 49

[Ko] B. K o s t a n t , Lie a l g e b r a c o h o m o l o g y a n d the g e n e r a l i z e d

B o r e l - W e i l t h e o r e m ; Ann. M a t h . 74 (1961), 329 - 387.

[LCM] S. Lang, Complex Multiplication; Springer: Grundlehren 255,

1983.

[La] R.P. L a n g l a n d s , A u t o m o r p h i c R e p r e s e n t a t i o n s , S h i m u r a
V a r i e t i e s , a n d M o t i v e s . Ein M i r c h e n ; Proc. Symp. P u r e Math.
3 3 (1979), p a r t 2; 205 - 246.

[Le ] M. Lerch, S u r q u e l q u e s f o r m u l e s r e l a t i v e s au n o m b r e d e s
c l a s s e s ; B u l l . Sc. M a t h & m . (2) 21 (1897), prem. partie,
290 - 304.

[Schl] N. S c h a p p a c h e r , P r o p r i & t 6 s de r a t i o n a l i t 6 de v a l e u r s
s p 6 c i a l e s de f o n c t i o n s L a t t a c h 6 e s a u x c o r p s CM; in:
S 6 m i n a i r e de t h ~ o r i e de n o m b r e s , P a r i s 1 9 8 1 - 8 2 , B i r k h ~ u s e r
(PM 38), 1 9 8 3 ; 267 - 282.

[Sch] N. S c h a p p a c h e r , O n the p e r i o d s of H e c k e characters; in

preparation.

[S£] J-P. S e r r e , A b e l i a n £ - a d i c representations and elliptic

c u r v e s ; B e n j a m i n 1968.

[Shl ] G. S h i m u r a , On the z e t a - f u n c t i o n of a n a b e l i a n v a r i e t y w i t h
c o m p l e x m u l t i p l i c a t i o n ; Ann. M a t h . 94 (1971), 504 - 533.

[Sh2] G. S h i m u r a , A u t o m o r p h i c f o r m s a n d the p e r i o d s of a b e l i a n
v a r i e t i e s ; J. M a t h . Soc. J a p a n 31 (1979), 561 - 592.

[Sh3] G. S h i m u r a , On some a r i t h m e t i c p r o p e r t i e s of m o d u l a r forms

of one a n d s e v e r a l v a r i a b l e s ; Ann. M a t h . 102 (1975), 491-515.

[WIII] A. W e i l , Oeuvres Scientifiques - Collected Papers, vol. III.

Springer 1980.

[WE~] A. Weil, E l l i p t i c f u n c t i o n s according to E i s e n s t e i n and

K r o n e c k e r . S p r i n g e r 1976.
AN I N T R O D U C T I O N TO I N F I N I T E S I M A L VARIATIONS OF HODGE STRUCTURES

Joe Harris
Mathematics Department, B r o w n University
Providence, RI 02912, U.S.A.

The p u r p o s e of this note is to give a simple introduction to the

notion of i n f i n i t e s i m a l variation of Hodge structure. This is an
object first defined and used in [i] (though the u n d e r l y i n g ideas had
been in the air for a while) and more recently the subject of an excel-
lent m o n o g r a p h by Peters and S t e e n b r i n k [2]. Unfortunately, this
theory, which in fact should make life e a s i e r for m a t h e m a t i c i a n s trying
to apply Hodge theory to geometry, gives at first the i m p r e s s i o n of
being complicated and technical. It is my hope here to avoid this
impression by p r e s e n t i n g the b a s i c ideas of the theory in as simple a
fashion as possible.
We b e g i n by r e c a l l i n g the basic set-up of Hodge theory. The goal
of this theory is to a s s o c i a t e to an m - d i m e n s i o n a l complex m a n i f o l d X
(for s i m p l i c i t y we w i l l take X a submanifold of ~N) a linear-alge-
braic invariant, as follows. To b e g i n with, we can for each n assoei-
th
ate to X its n-- t o p o l o g i c a l cohomology group m o d u l o torsion
H~ = Hn(X,~)/~ors , or its c o m p l e x i f i c a t i o n the n th deRham c o h o m o l o g y
group H{ = H ~ ® C = HDRn (X) . We can also associate the cup product in
cohomo!ogy; or rather, since we are only dealing with one group at a
time, the b i l i n e a r pairing

Q : HZZ × HZ~ --> ZZ

d e f i n e d by

wm_n

X
where w is the r e s t r i c t i o n to X of the g e n e r a t o r of H2(~N,~)
Of course, these are invariants of the u n d e r l y i n g differentiable mani-
fold of X , and do not reflect its complex structure. What does
determine the complex structure of X is the d e c o m p o s i t i o n of the com-
plexified tangent spaces to X into h o l o m o r p h i c and a n t i - h o l o m o r p h i c
parts; or, equivalently, the d e c o m p o s i t i o n of the space An(x) of
differential forms of degree n on X by type:

An(x) = AP,q
p+q=n
 52

Naturally, this data is too c u m b e r s o m e to carry around, but here

we are in luck: by the H o d g e theorem, this decomposition descends to
the level of cohomology. Precisely, if we let H p'q = HP,q(x) c H~R(X )
be the s u b s p a c e of classes representable by forms of type (p,q) , we
get a d e c o m p o s i t i o n

H~R(X)w = HC = ~ HP'q
p+q=n

satisfying the obvious relations

HP,q = Hq,P
and
Q ( H P ' q , H p''q') : 0 unless p+p' : q+q' = n .

The package of data introduced so far -- a lattice H~ with integral

bilinear form Q and d e c o m p o s i t i o n H ~ ® ~ = @ H p'q satisfying these
relations -- we call a Hodge structure of w e i g h t n associated to X .
It is an object that is on one h a n d e s s e n t i a l l y finite, and that on the
other h a n d we m a y hope will reflect the g e o m e t r y of X .
Now, w h e n e v e r we associate to a g e o m e t r i c object a (presumably
simpler) invariant, two q u e s t i o n s arise: to w h a t extent does the
invariant actually determine the o r i g i n a l object; and to w h a t extent
can we read off d i r e c t l y from the invariant answers to naive questions
about the geometry of the object. In the p r e s e n t circumstances, the
first q u e s t i o n translates into the Torelli problem, which asks when the
members of a given family of v a r i e t i e s (e.g., curves of genus g ,
hypersurfaces of degree d in ~n) are d e t e r m i n e d by their Hodge
structures; or the "generic" or "birational" Torelli problem, which
asks w h e n this is true for a g e n e r a l member of the family. The Torelli
p r o b l e m has been answered in a number of cases (e.g., for curves of
genus g it was proved classically by Torelli; and the generic Torelli
for h y p e r s u r f a c e s was proved recently by Donagi) ; it remains very much
an open q u e s t i o n in general.
The most famous example of a q u e s t i o n in Hodge theory along the
lines of the second sort above is of course the H o d g e conjecture. It
is not h a r d to see that if y c X is an a n a l y t i c subvariety of codim-
ension k , its fundamental class must lie in the s u b s p a c e
H k,k c H2k(x,~) The Hodge c o n j e c t u r e asks w h e t h e r the converse is
true: that is, w h e t h e r a class y ~ Hk, k n H ~ is n e c e s s a r i l y a
rational linear combination of classes of subvarieties.
The s i m p l e s t case of Hodge theory is its a p p l i c a t i o n to curves,
and here by any standards it is successful. To the Hodge structure
 53

( H ~ , Q , H I'0 • H 0'I) of a curve we associate the p r o j e c t i o n A of H~

to H 0'I (traditionally r e p r e s e n t e d by the p e r i o d matrix: we choose

a basis for H~ n o r m a l i z e d with respect to Q and write out the
(0,1)-components of these vectors in a g×2g matrix ~) and then the
complex torus H0'I/A = J(C) , called the J a c o b i a n of C . This in
turn gives rise to a host of s u b v a r i e t i e s of J(C) and t h e t a - f u n c t i o n s
that reflect and e l u c i d a t e the geometry of C .
For h i g h e r - d i m e n s i o n a l varieties, the a p p l i c a t i o n of Hodge theory
has b e e n less successful, for w h i c h there are perhaps two reasons. The
first of these is the apparent absence of any r e a s o n a b l y natural geo-
m e t r i c and/or analytic object a s s o c i a t e d to a Hodge structure in gene-
ral. Looking at the case of curves, one sees that it is exactly
through the geometry of the Jacobian, and the analysis of the theta-
function, that Hodge theory is useful. Unfortunately, no analogous
objects have been found in general.
The second factor is simply this: that only in a very few cases
can one ever hope to determine e x p l i c i t l y the Hodge s t r u c t u r e of a
given variety. To be specific, consider the case of a smooth h y p e r s u r -
face X c ~n+l given by a h o m o g e n e o u s polynomial F(Z) = 0 of degree
d . By the Lefschetz theorem, all the c o h o m o l o g y of X below the
middle d i m e n s i o n (and hence above it as well) is at most one-dimen-
sional, so we focus on Hn(x) We can immediately identify one of
the Hodge groups: Hn'0(X) , the space of h o l o m o r p h i c n-forms on X ,
may be r e a l i z e d as P o i n c a r e residues of (n+l)-forms on ~n+l with
poles along X ; explicitly,
Q , _/Zl~ ./Zn+~ zn+l~
G(z 0 ..... • ^ok V0/ 0 }
= Res " " F(Z 0,. .],Zn) . . . . . . . . . . . /
/

G(Z0, . ,Zn+l)d(~--@)A..^d(~)A.
. . . . . ^d(Zntl~\'-~G/.ZOn
~Z~i (Z 0 , .... Zn+ 1 )

for G(Z) a h o m o g e n e o u s p o l y n o m i a l of degree d-n-i . Thus

Hn, 0 = Sd_n_ 1

where S is the graded ring C[Z0,...,Zn+I] . Similarly, the other

H o d g e groups of X may be realized as residues of forms on ~n+l
w i t h h i g h e r - o r d e r poles on X (actually, we get in this way just the
primitive c o h o m o l o g y pr(X)
Hn , w h i c h here means the classes o r t h o g o n a l
to ~) . We obtain an i d e n t i f i c a t i o n
 54

Hn-k ,k
pr (X) = (S/J) ( k + l ) d - n - i

where j c S is the J a c o b i a n ideal of X , that is, the h o m o g e n e o u s

ideal generated by the p a r t i a l derivatives of X .
W e h a v e thus f o u n d the v e c t o r space decomposition H pr(X)
n =
Hn-k'k(x) The p r o b l e m is, it is i m p o s s i b l e in g e n e r a l to i d e n t i f y
pr
in t h e s e terms the l a t t i c e H~ of i n t e g r a l classes. Indeed, this has
b e e n done o n l y in the p r e s e n c e of a l a r g e a u t o m o r p h i s m group acting on
X , e.g., for F e r m a t h y p e r s u r f a c e s . Thus, for e x a m p l e , if one is g i v e n
a particular hypersurface of e v e n d i m e n s i o n n = 2k , it is i m p o s s i b l e
to d e t e r m i n e in g e n e r a l Hk'k(x) n Hn(X,~) , or w h e n two s u c h X have
the s a m e H e d g e structure. Simply put, w e c a n n o t find the lattice; but
without the l a t t i c e we h a v e no i n v a r i a n t s .
One s o l u t i o n of this d i f f i c u l t y appears at f i r s t to b e m o v i n g in
the w r o n g direction, toward increased difficulty. One c o n s i d e r s not
just a v a r i e t y X , b u t a f a m i l y of v a r i e t i e s {Xb}bc B parametrized
by a variety B , of w h i c h X = X0 is a m e m b e r ; we assume 0 ~ B is
a s m o o t h point. Locally around X 0 , then, w e can i d e n t i f y the lat-
tices Hn(Xb,~)/tors with a single lattice H~ and the v e c t o r spa-
ces Hn(Xb,~) with H~ correspondingly. We t h e n c o n s i d e r the s p a c e s
H n - k ' k ( X b ) -- or the a s s o c i a t e d

k
Fk = ~ Hn-~,~(~) __
~=0

as v a r i a b l e subspaces of HC The b a s i c facts t h e n are:

i) The map ~k from B (or a n e i g h b o r h o o d of 0 e B) to the

Grassmannian sending b to F k ( X b ) c H~ is h o l o m o r p h i c ; and
ii) In terms of the i d e n t i f i c a t i o n of the t a n g e n t s p a c e to the
Grassmannian at A c H with Hom(A,H/A) , the i m a g e u n d e r 6 k = d~ k
of any t a n g e n t vector to B at 0 carries Fk into Fk+i/Fk . We
thus a r r i v e at a c o l l e c t i o n of maps
~k : T0B --> H ° m ( H n - k ' k ( x ) ' H n - k - l ' k + l ( x ) ) By e q u a l i t y of m i x e d p a r -
tials, they s a t i s f y the r e l a t i o n s

(*) 6k+l(V) O~k(W) = ~k+!(w)O6k(V) V v,w c T

and s i n c e the s p a c e s F k ( X b) satisfy the r e l a t i o n Q ( F k , F n _ k _ I) H 0

for all b , we have
 55

(**)
Q(@k(V) (e) ,~) + Q ( e , @ n _ k _ l ( V ) (B)) = 0

~ e • Hn-k'k(x) , 8 • Hk+l'n-k-l(x) ,v • T

We n o w d e f i n e an i n f i n i t e s i m a l variation of H o d g e structure (IVHS)

to be just this c o l l e c t i o n of data: t h a t is, a quintuple
(H~,Q,HP'q,T,6q) in w h i c h ( H ~ , Q , H p'q) is a H o d g e structure, T a
vector space, and

6 : T--> H o m ( H P ' q , H p-l'q+l)

q

maps satisfying (*) a n d (**) above. By w h a t we h a v e just said, to

every member X = X0 of a f a m i l y of v a r i e t i e s {X b} we h a v e associa-
t e d such an object.

T w o key o b s e r v a t i o n s here are the f o l l o w i n g :

i) The i n f i n i t e s i m a l variation of H o d g e structure associated to

a family is in g e n e r a l computable; or at least as c o m p u t a b l e as the
Hodge structures associated to the m e m b e r s . For example, going back
to our e x a m p l e of h y p e r s u r f a c e s , if we let X c pn+l be s m o o t h w i t h
equation F(Z) = 0 , the t a n g e n t space at X to the f a m i l y of h y p e r -
surfaces of d e g r e e d up to p r o j e c t i v e isomorphism is just the space
Sd of h o m o g e n e o u s polynomials of d e g r e e d , modulo the J a c o b i a n
ideal. (A v a r i a t i o n of X in ~n+l is g i v e n by F+sG for
G E Sd/{F ; if G = E a i j X i ~~-~. this c o r r e s p o n d s to f i r s t o r d e r to the
] tA
motion of X a l o n g the 1-parameter group e of a u t o m o r p h i s m s of
~n+l). Thus T = (S/J) d ; and the maps

@k : (S/J) d --> H o m ( ( S / J ) (k+l) d-n-l' (S/J) (k+2)d-n-1)

t u r n out to be n o t h i n g b u t p o l y n o m i a l multiplication.
It s h o u l d be n o t e d h e r e t h a t this in i t s e l f has some nice c o n s e -
quences: for e x a m p l e , w h i l e w e are as i n d i c a t e d e a r l i e r u n a b l e to
~n+l
determine Hk,k(X) n H 2k (X,~) for any g i v e n h y p e r s u r f a c e in ,
n = 2k , the fact t h a t for d a n+l the m a p

(S/J) d x ( S / J ) k d _ n _ 1 --> (S/J) ( k + l ) d - n - i

is s u r j e c t i v e i m m e d i a t e l y i m p l i e s that for g e n e r a l X ,
Hk'k(x) n H 2 k ( x , ~ ) = 0 , and so Hk'k(x) n H2~(X,~)- = ~ Thus on a
pr
g e n e r a l h y p e r s u r f a c e e v e r y a l g e b r a i c s u b v a r i e t y is h o m o l o g o u s to a
 56

rational multiple of a complete intersection. In p a r t i c u l a r in case

n = 2 this yields the famous

Theorem (Noether; Lefschetz) : a surface S c ~3 of degree d ~ 4 ,

having general moduli, contains no curves other than complete inter-
sections S nT with other surfaces.

2) The s e c o n d key p o i n t is this: that even w i t h o u t the lattice

H~ , a__nni n f i n i t e s i m a l variation of Hodge structure will in general
possess non-trivial invaziants, and will give rise to g e o m e t r i c ob-
jects. These of course come from the maps $k which, being trilinear
objects, have lots of a c c e s s i b l e invariants (e.g. their associated
determinantal varieties).

To illustrate the use of this, c o n s id e r the generic Torelli theo-

rem for h y p e r s u r f a c e s . The a p p l i c a t i o n of IVHS to this p r o b l e m is
based on the following trick: for any map f : X --> Y of varieties,
the condition that f is b i r a t i o n a l onto its image, i.e. that

for general p ~ X , ~ q c X : q ~ p , f(q) = f(p)

is in fact e q u i v a l e n t to the a priori w e a k e r statement

for g e n e r a l p c X , ~ q c X :

q ~ p , f(q) = f(p) and Im(f,)q = Im(f,)p

In our present circumstances, this e q u i v a l e n c e means that

A general hypersurface of degree d in ~n+l is

determined by its H o d g e structure
<~>

A general hypersurface of degree d in ~n+l is

determined by its i n f i n i t e s i m a l variation of Hodge
structure.

Thus, to prove the generic Torelli theorem for h y p e r s u r f a c e s ,

Donagi shows that from the data of the vector spaces

(S/J) (k+l)d-n-i ' (S/J) d

and the m u l t i p l i c a t i o n maps

6k_ 1 : (S/J) d x (S/J)kd_n_l --> (S/J) (k+l)d-n-i

 57

Q : (S/J) ( k + l ] d - n - i × (S/J) ( n - k + ! ) d - n - i --> (S/J) ( n + 2 ) d - n - i

one can r e c o n s t r u c t the e n t i r e ring S/J , and f r o m this the h y p e r s u r -

face X . This suffices to e s t a b l i s h the theorem; again, it s h o u l d be
observed that at no p o i n t in the argument is the lattice H~
mentioned.
Donagi's argument is a b e a u t i f u l one, but this is not the p l a c e
to reproduce it. Let me i n s t e a d conclude by g i v i n g a similar and
easier example of the use of IVHS: to p r o v e the g e n e r i c Torelli
theorem for curves of genus g ~ 5 .
Of course, the Torelli theorem has b e e n proved many times over,
in as s t r o n g a form as one c o u l d wish. One common characteristic of
the p r o o f s , however, is that they all m a k e essential use of the geom-
etry of the J a c o b i a n and its subvarieties. A natural question if one
is s t u d y i n g higher-dimensional Torelli theorems is: does there exist
a proof of the T o r e l l i for curves that avoids the use of the J a c o b i a n ?
The answer to this is u n k n o w n to me; however, using IVHS we can give a
very short proof of the g e n e r i c Torelli in genus g ~ 5 as f o l l o w s .
The tangent space, at a c u r v e C , to the family of all c u r v e s is
dual to the s p a c e H 0 ( C , K 2) of q u a d r a t i c differentials on C . The
IVHS associated to C in this family thus consists of the H e d g e
structure of C , together with a map

: H 0 ( C , K 2) --> H o m ( H I ' 0 , H 0'I) .

Here the r e l a t i o n (*) above is t r i v i a l ; while the relation (**) says

that in terms of the identification of H 0'I with (HI'0) * given by
Q , the image of ~ lies in the subspace
Sym2(Hl'0) * c Hom(H 1 ' 0 , (HI'0) *) , i . e .

: H 0 (C,K 2) * --> Sym2(H0(C,K) *

The transpose of ~ is now easy to i d e n t i f y : it is the m a p

t6 : Sym2H0(C,K) --> H0 (C,K 2 )

that simply takes a quadratic polynomial P(~l,...,~g) in the h o ! o -

morphic differentials on C and e v a l u a t e s it as a q u a d r a t i c differen-
tial on C . In p a r t i c u l a r , the k e r n e l of t0 is just the v e c t o r
space of q u a d r a t i c polynomials vanishing on the image of the canonical
curve C c ~H0(C,K) * = ~g-i ; since it is w e l l known that a general
 58

canonical curve of genus g a 5 is the intersection of the q u a d r i c s

containing it, we can recover the curve C . Explicitly, in terms of
the i n f i n i t e s i m a l variation of H o d g e structure ( H ~ , Q , H I'0 @ H 0 ' I , T , S )
associated to C , we have

C = ]P{Z ( H 0'I : Q(Z,I(Z)) = 0 for all I : H 0'I --> H I'0

such that trace(~.o@(v)) = 0 for all v e T }

and this suffices to e s t a b l i s h generic Torelli.

References

[i] R. Donagi, J. C a r l s o n , M. Green, P. G r i f f i t h s and J. H a r r i s ,

Compositio M a t h 50(1983).
[2] C. P e t e r s and J. S t e e n b r i n k , I n f i n i t e s i m a l v a r i a t i o n s of H o d g e
s t r u c t u r e and the % g e n e r i c T o r e l l i p r o b l e m for p r o j e c t i v e
hypersurfaces, in B i r k h a u s e r P r o g r e s s in M a t h e m a t i c s series
n u m b e r 39. (1984)
 NEW DIMENSIONS IN G E O M E T R Y

Yu. I. M a n i n

Steklov Mathematical Institutc

Moscow USSR

Introduction

Twenty-five years ago Andr6 Weil published a short paper entitled "De
la m@taphysique aux math@matiques" [37]. The mathematicians of the
XVIII century, he says, used to speak of the "methaphysics o f the c a l -
culus" o r the "metaphysics of the t h e o r y of e q u a t i o n s " . By this they
meant certain dim analogies which were difficult to g r a s p a n d to m a k e
precise but which nevertheless were essential for r e s e a r c h and dis-
covery.

The inimitable Weil style requires a quotation.

"Rien n'est plus f6cond, tousles math6maticiens le savent, que
ces obscures analogies, ces t r o u b l e s reflets d'une th6orie ~ une autre,
ces furtives caresses, ces brouilleries inexplicables; rien aussi ne
donne plus de p l a i s i r au c h e r c h e u r . Un j o u r v i e n t od l ' i l l u s i o n se
dissipe, le p r e s s e n t i m e n t se c h a n g e en c e r t i t u d e ; les t h 6 o r i e s jumelles
r6v~lent leur source commune avant de d i s p a r a ~ t r e ; comme l'enseigne la
Git~ on a t t e i n t ~ la c o n n a i s s a n c e et ~ l ' i n d i f f 6 r e n c e en m~me temps. La
m6taphysique est devenu math6matique, prate ~ former la m a t i 6 r e d'un
trait@ d o n t la b e a u t @ froide ne saurait plus nous 6mouvoir".

I think it is t i m e l y to s u b m i t to the 25 th A r b e i t s t a g u n g certain vari-

ations on t h i s theme. The a n a l o g i e s I want to speak of are of the
following nature.

The archetypal m-dimensional geometric object is the space Rm which

is, after Descartes, represented by the p o l y n o m i a l ring ~ [x1,...,Xm].

Consider instead the r i n g Z [ x I ..... Xm;~1 ..... ~n ], w h e r e Z denotes

the integers and ~i are "odd" variables anticommuting among them-
 60

selves and c o m m u t i n g with the "even" variables xK . It is c o n v e n i e n t

to a s s o c i a t e with this ring a certain g e o m e t r i c object of d i m e n s i o n
I + m + n, or better still (];mln) , where I refers to the "arithmetic
dimension" Z , m to the o r d i n a r y geometric dimensions (Xl,...,x m)
and n to the new "odd dimensions" represented by the c o o r d i n a t e s

%i"

Before the advent of ringed spaces in the fifties it would have been
difficult to say p r e c i s e l y what we mean when we speak about this geo-
metric object. Nowadays we simply define it as an "affine superscheme"
Spec Z [xi,~K], an object of the c a t e g o r y of t o p o l o g i c a l spaces local-
ly ringed by a sheaf of Z2-graded supercommutative rings (cf. n°4 be-
low) I have tried to d r a w the "three-space-2000", w h o s e p l a i n x-axis
i~ is s u p p l e m e n t e d by the set of p r i m e s
and by the "black arrow", c o r r e s p o n -
ding to the odd dimension.

. . . . . Spec
Three-space-2000

The message of the picture is intended to be the following methaphysies

underlying certain recent developments in g e o m e t r y : " a l l three types of
geometric dimensions are on an equal footing"

Actually the similarity of Spec Z to Spec k[x], or in general of alge-

braic number fields to algebraic function fields, is a well k n o w n
heuristic principle which led to the most remarkable discoveries in the
dioph a n t i n e g e o m e t r y of this century. This similarity was in fact the
subject matter of the Weil paper I just quoted. Weil likens the three
theories, those of Riemann surfaces, algebraic numbers and a l g e b r a i c
curves over finite fields, to a t r i l i n g u a l inscription with p a r a l l e l
texts. The texts have a common theme but not identical. Also they have
been p a r t l y destroyed, each in d i f f e r e n t places, and we are to d e c i p h e r
the enigmatic parts and to r e c o n s t r u c t the m i s s i n g fragments.

In this talk I shall be c o n c e r n e d with only one aspect of this similarity,

reflected in the idea that one may c o m p a c t i f y a projective scheme over
Z by adding to it a fancy infinite closed fibre. In the r e m a r k a b l e
papers [I], [2] S. A r a k e l o v has shown convincingly that in this way the
arithmetic dimension acquires truly g e o m e t r i c global properties, not
 61

just b y itself, but in its c l o s e interaction with the "functional"

coordinates. G. Faltings [10], [11] h a s pushed through Arakelov's idea
much further and beyond doubt (for me) the existence of a g e n e r a l arith-
metic geometry, or A-geometry. This A-geometry is e x p e c t e d to c o n t a i n
the analogues of all m a i n results of c o n v e n t i o n a l algebraic geometry.
The leading idea for the c o n s t r u c t i o n of the arithmetic compactifications
seems to b e as follows:

K~hler-Einstein geometry = ~-adic arithmetic

I have tried in s e c t i o n s I-3 of t h i s talk to b r i n g together our

scattered knowledge on this subject.

Starting with section 4 the o d d dimensions enter the game. The algebraic
geometers are well accustomed to e n v i s a g e the spectrum of the dual
numbers Spec ~[s] ,s 2 = 0 ~ as t h e infinitesimal arrow and w i l l hardly
object to a s i m i l a r visualization of Spec ~[~]. Still, there is a n
essential difference between these two cases. The even arrow Spec ~[c]
is n o t a m a n i f o l d but only an i n f i n i t e s i m a l part of a m a n i f o l d . This
c a n be s e e n e.g. in the fact that ~I[c]/~ is n o t ~[s]-free, since
2
from s = 0 it f o l l o w s that cds = 0. By c o n t r a s t , the o d d a r r o w
Spec ~[~] is an h o n e s t manifold from this point of v i e w , since the
Z2-graded Leibniz formula for, say, the e v e n differential d~ , is
valid automatically, <.d~ + d ~ - ~ = 0 and one easily sees that QI~[~/~
is R[~]-free.

In s p i t e of t h e e l e m e n t a r y nature of this example it s h o w s w h y the odd

nilpotents in t h e structure sheaf may deserve the n a m e of coordinates.
B u t of c o u r s e this is o n l y a b e g i n n i n g .

The most remarkable result of supergeometry u p to n o w is p r o b a b l y the

extension of the K i l l i n g - C a r t a n classification to the finite dimensional
simple Lie supergroups made in [15], [16]. The Lie supergroups acting
on supermanifolds mix the even and the o d d coordinates, which is one
reason more to c o n s i d e r t h e m on an e q u a l footing.

In s e c t i o n s 5 and 6 we state some recent results of A. V a i n t r o b ,

J. Skornyakov, A. V o r o n o v , I. P e n k o v and the a u t h o r on the geometry
of s u p e r m a n i f o l d s . They refer to the K o d a i r a deformation theory and
 62

the c o n s t r u c t i o n of the S c h u b e r t supercells and show that in t h i s

respect also s u p e r g e o m e t r y is a n a t u r a l extension of the p u r e e v e n
geometry.

The f o l l o w i n g radical idea seems m o r e fascinating:

~ h e even g e o m e t r [ is a c o l l e c t i v e
effect in the ~ - d i m e n s i o n a l odd geometry

There is a v e r y simple algebraic model s h o w i n g h o w this m i g h t h a p p e n .

The h o m o m o r p h i s m of the f o r m a l series in ~ variables

1) ~ [[Xl,X 2 .... ]] ÷ R [ [ . . . < _ i , ~ 0 , ~ i .... ]] ,

oo

Xi ÷ [ <n~n+i
n = _co

zs i n j e c t i v e .

A considerably more refined version of this c o n s t r u c t i o n has r e c e n t l y

emerged in the w o r k on r e p r e s e n t a t i o n s of K a c - M o o d y algebras [17],[7].
This r e s u l t establishes t h a t the two r e a l i z a t i o n s of gl(~) in the
differential operator algebras Dev : D i f f ~ ( ~ [ X l , X 2 ,. .. ]) a n d
Dodd = Diff(~ [ [ . . . ~ _ I , ~ 0 , ~ I .... ]]) are e x p l i c i t l y isomorphic:

(2) 6~i9{j @ij +-~ Zij ( x , ~ )

Here Z.. are defined f r o m the f o r m a l series

z3
-i -i
EZijplq - q ~ 1 ( 1 - q p - l l [ e x p (~ x i ( P l - g l ) ) e x p ( ~ q -P ) -1].
Dx i
I 1

The i s o m o r p h i s m (2) is e s t a b l i s h e d by c o m p a r i s o n of two n a t u r a l repre-

s e n t a t i o n s , that of Dev on R [Xl,X2,...] a n d t h a t of ~odd on
F = ~ [~.,~-~--~]/I, w h e r e I is the left ideal g e n e r a t e d b y ~, i < 0,
and Z~--,j~0~ The g e n e r a t o r I m o d I of the c y c l i c ~[~,~-~n-]-module F
~j 3
can be c o n v e n i e n t l y r e p r e s e n t e d as the i n f i n i t e w e d g e - p r o d u c t
-I
n A: - ~ ~i
and the total m o d u l e F as the span of h a l f - i n f i n i t e m o n o m i a l s
i~J
~i't Jc~' c a r d ~ \ J < ~. The i s o m o r p h i s m

(3) F ~[Xl,X2,...]
 63

may then be considered as the development of the simplistic idea (I).

The investigation of g e o m e t r y with odd coordinates was started by

physicists and is c o n t i n u e d mainly in the p h y s i c a l l y motivated work
[12],[13],[34]. In p a r t i c u l a r , the mathematical foundations of super-
geometry were l a i d b y F.A. Berezin [5] who early understood the role
and the necessity of this extension of o u r g e o m e t r i c intuition. Of
course the general philosophy of a l g e b r a i c geometry is o f g r e a t help.

Odd functions serve for m o d e l l i n g the internal degrees of f r e e d o m of the

fundamental matter fields, leptons and quarks. Their quanta have spin
I
a n d o b e y t h e F e r m i - D i r a c s t a t i s t i c s . O n the o t h e r hand the quanta of
gauge fields (photons, gluons, W±,Z,...) have spin I and are bosons. The
map (I) is a t o y m o d e l of the b o s o n i c collective excitations in t h e c o n -
densate of pairs of fermions. The formulas (2) and (3) a l s o w e r e
essentially known to s p e c i a l i s t s in d u a l strings theory.

The idea that fermionic coordinates are p r i m a r y with respect to the

bosonic ones has been repeatedly advertised in v a r i o u s disguises. It is
still awaiting the p r e c i s e mathematical theory. It m a y w e l l prove true
that our four space-time coordinates (x 0 = c t , x l , x 2 , x 3 ) are o n l y the
phenomenologically effective entities convenient for the d e s c r i p t i o n of
the low energy world in w h i c h our biological life can exist only, but
not really fundamental ones.

Meanwhile physicists are d i s c u s s i n g grand unification schemes and super-

gravity theories which account for all fundamental interactions (or some
o f them) united in a L a g r a n g i a n invariant with respect to a L i e super-
group or covariant with respect to the g e n e r a l coordinate transform in
a superspace.

Section 6 of this talk describes the geometry of simple supergravity from

a new viewpoint which presents superspace as a " c u r v e d flag space"
keeping a part of its S c h u b e r t cells.

The geometry of supergravity being essentially different f r o m the simple-

minded super-riemannian geometry, one is l e d to b e l i e v e that the substi-
t u t e of t h e K i h l e r i a n structure in s u p e r g e o m e t r y must be rather sophisti-
 84

cated. Therefore I do not venture here to m a k e any guesses about the

~-geometry with odd coordinates.

Comparing our present understanding of t h e a r i t h m e t i c dimension with

that of the odd ones we d i s c o v e r that the destroyed texts are r e -
constructed in d i f f e r e n t parts o f the p a r a l l e l texts. Trying to g u e s s
more, we can ask two q u e s t i o n s .

a) Is it p o s s i b l e to c o m p a c t i f [ _ a supermanifold with respect to the

odd dimensions ?

We seemingly need a construction of such a compactification if w e w a n t

to h a v e a cohomology theory in w h i c h the S c h u b e r t supercells would have
nontrivial (i.e. depending essentially o n the o d d part) cohomology
classes.

D. L e i t e s has conjectured that in an a p p r o p r i a t e category an "odd pro-

jective space" might exist, that is the q u o t i e n t of S p e c k[~1 .... ,~n ]~
Speck modulo the m u l t i p l i c a t i v e group action (t,(~i)) ~ (t~i). Of
course, in the o r d i n a r y sense it is e m p t y .

b) Does there exist a group, mixing the a r i t h m e t i c dimension with the

(even) geometric ones ?

There is no such g r o u p naively, but a "category of representations of

this group" may well exist. There may exist also certain correspondence
rings (or t h e i r representations) between Spec ~ and x . A recent
work by Mazur and Wiles [27] shows that the p-adic Kubota-Leopoldt
~-function divides a certain modular p-adic C-function defined in
characteristic p. Such things usually happen if a c o r r e s p o n d e n c e
exists.

Finally, I would like to a c k n o w l e d g e my gratitude to m a n y f r i e n d s w h o s e

ideas helped to c o n s o l i d a t e certain beliefs expressed here. I am
particularly grateful to I.R. Shafarevitch who taught the a r i t h m e t i c -
geometry analogies to h i s students for t h r e e decades, to A.A. Beilinson
who has generously shared his geometric insight with the author.
 85

I. A-manifolds and A-divisors

I. A - m a n i f o l d s . Let K be a finite algebraic number field, R its

ring of integers, S = Sf U S the set of finite and infinite places o f K.
If v 6 S, K denotes the completion of K with respect to t h e valu-
v
ation I[v:K ÷ ~ * . We put la] v : lal if Kv = ~ , lal v :lal 2 , if K v = {-

Then I-~ la[ v = I for all a 6 K. Moreover, R= {aE KI lalv~1 for all v6Sf}.
V

We shall call the following data an A-manifold:

(I) X= (Xf;~v,V 6Soo)

Here Xf is a s c h e m e of finite type, proper, surjective and flat over

S p e c R, with smooth irreducible generic fiber. Furthermore, ~ is a
V
K~hlerian form on the complex variety Xv: (Xf ~ K v ) (~), and ~v =~ v
if Kv = ~ ; if Kv= ~ , then the forms corresponding to the two embeddings
K ÷~ should be conjugate.
V

We shall denote by vol dim X v the corresponding volume forms.

V V

The simplest example of an A-manifold is t h e A-curve Xf = S p e c R

endowed with the volumes of ~ii points v 6 S ; ~v do not exist in t h i s
case.

I want to stress the preliminary nature of the definition (I). First

of all, one should not restrict oneself to t h e relatively proper schemes.
If Xv is n o t proper, the ~v presumably may have logarithmic growth
at infinity, cf. [8]. Furthermore, a very special role is p l a y e d by the
K~hler-Einstein forms ~ , see n°5 below.
v

2. Invertible A-sheaves. An invertible A-sheaf on the A-manifold (I)

is t h e data

(2) L = (Lf; hv , v E S ).

Here if is an invertible sheaf on X[,_ h v - a H e r m i t i a n metric on

i v = if ® K v with the evident reality conditions and the following
R
property:
 86

(3) the c u r v a t u r e form F of the h e r m i t i a n connection

v
correspondin9 to hv i~s ~v-harmonic.

We recall that if s is a l o c a l holomorphic section of iv , then

Fv = ~ log h v ( S , S ) in the d o m a i n of s. F o r an A - c u r v e the condition
(3) is e m p t y .

It is e v i d e n t h o w to d e f i n e the tensor product [ ® i' of two invertible

A-sheaves. The g r o u p of isomorphism classes of invertible A-sheaves is
denoted by PiCAX. The identity is the class of t h e structure sheaf

0X: (0Xf; hv I hv(1,1) : I for all v6 S )

Later o n we shall use the following fact: for a f i x e d ~ , h is de-

v v
fined by the condition (3) u p to a m u l t i p l i c a t i v e constant.

3. S e c t i o n s . Let i be the invertible sheaf (2). Set H (X,[) = Hi(Xf,lf) •

These cohomology groups are the R-moduli of finite type, and the ordinary
Riemann-Roch-Grothendieck theorem for schemes [35] tells much about their
structure. An essentially new object in A - g e o m e t r y is t h e E u l e r A-charac-
teristic XA([) . If a c a n o n i c a l m a p D : X - ~ " A - p o i n t " w e r e to e x i s t , o n e
would define XA(i ) as RD,([). T h i s m e i n g o t h e r w i s e , o n l y c e r t a i n ad
noc definitiol,s of XA(1) in a f e w p a r t i c u l a r cases are k n o w n , which
are reviewed in n°2. The general i~ea is t h a t in c a s e Hi(x,i) = 0,1 > 0,
one must define XA(1) as the covolume o f the image

H°(X,[) + ® H ° ( X v , i v) : H° (X,[)
v6S

relative to a c e r t a i n volume f o r m on H° . The g e n e r a l definition of

this volume form is s t i l l lacking. Following Faltings [10] , one may
conjecture that to c o n s t r u c t it o n e can use a canonical metric on the
bundle on Pic0X v with fiber ® d e t H i ( X v , i v ) (-])i a n d to supplement
i
this by inductive reasoning on the N6ron-Severi group.

A correctly defined XA(I ) s h o u l d be c a l c u l a b l e via an A-Riemann-Roch

theorem so t h a t w e shall need divisors of sections of [ and, more
generally, A-characteristic classes.
 67

4. A - d i v i s o r s . We shall mean b y an A - d i v i s o r on X the following data:

D : (Df;rv I r v 6 ~ , v 6 S ).

where Df is a C a r t i e r divisor on Xf. The following symbolic notation

is m o r e convenient

D : Df * [ rvX(V)
v6S

The A-divisors X(v) (not to be c o n f u s e d with X v) are called the

"closed fibers" of X at i n f i n i t y . The A-divisors form a group DiVAX.

By the A-divisor of a s e c t i o n s 6 H°(X,i) we shall mean the following

element of DiVAX:

(4) div s: divfs -[ (S log ISlv'VOlv) X ( v ) ,

v6S X
v

Isl[¢:Kv]v = hv(S'S) "

Here divfs is t h e C a r t i e r divisor of s . If a r a t i o n a l function g

on Xf is a q u o t i e n t of two sections of i , it is n a t u r a l to d e f i n e
its p r i n c i p a l A-divisor b y the formula

{sl div g : divfg -[ (f log Igl v V O l v ) X ( v ) ,

v6S X
v

which does not depend on i . Finally, the same formula (4) m a y b e u s e d

to d e f i n e that A-divisor of a m e r o m o r p h i c section of i.

N o w we c a n e a s i l y introduce the A-sheaves 0(D) where D is an a r b i -

trary A-divisor, together with the c a n o n i c a l section whose A-divisor
is D . First, for D= Df we set:

(6) 0(Df) = (0xf(Df) ; h v)

where hv is t h e u n i q u e metric on 0 x f ( D f) ® K v satisfying equation

(3) a n d n o r m a l i z e d by

(71 f logliDflv - vol -0, v C S .

Xv V '
 68

where 1Df is the m e r o m o r p h i c section of 0xf(D f) whose Cartier

divisor is Df. Using (4) and (7) we get div ID= divfl D = Df which
justifies (6).

Furthermore, we set

(8) 0( ~ rvX(V)) : (0xf;h v i 111V : exp (-r (f VOlv)-l)) .

v6S v Xv

Again using (4) we o b t a i n

div I = vCS
[ rvX (v)

as is to be expected.

As in the g e o m e t r i c case we can c o n s t r u c t the exact sequence

0 ÷ Div~X÷ DiVAX + PiCAX ÷ 0,

D ÷ class of 0(D)

where Div~X is the g r o u p of p r i n c i p a l A-divisors.

5. G r e e n ' s functions. It is clear from the p r e v i o u s definitions that

the e s s e n t i a l information about the A r c h i m e d e a n part of the A - d i v i s o r s
is e n c o d e d in the f u n c t i o n s

(9) Gv(Dv,X) lIDvlv(X) , x6 X v

On the c o m p a c t Kihlerian manifold (Xv,~ v) they are u n i q u e l y defined

by the f o l l o w i n g conditions:

a) G v ( D v , X } is real analytic for x {supp. D v. The f u n c t i o n

G v ( D v , X ) / I g ( x ) iv , w h e r e gv is a local equation of Dv, is ex-
t e n d a b l e to supp D
V

b) The (1,1)-form ~ log G v ( D v , X ) is ~v-harmonic outside of D v. The

corresponding current is a linear combination of a h a r m o n i c form and
the @Dv ,
 89

c) ~ log Gv(Dv,X)'vol v : 0
Xv

Furthermore,

d) G v ( D v + Dv,X) : Gv(Dv,X) "Gv(D"v,X) .

e) Gv(div gv,X) : Cg- Igv(X) Iv for each meromorphic function gv where

c is d e f i n e d by c) .
V

It is explained in the last chapter of Lang's book [22] how to calculate

Green's functions on abelian varieties and algebraic curves using theta-
functions and differentials of the third kind respectively. The K~hlerian
metric involved is flat in t h e first case and induced by the flat metric
of the Jacobian in the second one. The same metrics are used in the
Arakelov-Faltings-Riemann-Roch theorem on A-surfaces which we shall state
i~ ~°2.

Since the function (9) is n o t constant except in trivial cases, the

closed fibers X(v) , v 6 S , should be imagined as "infinitely degene-
rate". To make it m o r e credible note that if for v 6 Sf the closed fiber
is degenerate then a meromorphic function or a section of an invertible
sheaf can have different orders at different components of X(v) . In
other words, instead of lIDflv one should consider in this case
lIDflv i where the valuations V 1• correspond to the components X(v)
i
of X(v). Finally, one can unify these numbers into a function
lIDfl (x) , x 6 X ( R v) = Xv(Kv) setting lIDfl (x) = lIDflV i if the section
x intersects X(v) .
1

This analogy suggests refining the definition of the divisor supported

by X(v) , v 6 S . Conjecturally, instead of a constant rv one should
consider a volume form @v and delete the integrals from (4) and (5).
A comparison with the Mumford-Schottky curves may serve to clarify the
situation. Meanwhile we shall use the coarse definitions.

6. The intersection index. Let o:Spec R÷ Xf be a section of the

structural morphism z: X f + Spec R . We shall consider the image of @
as the closed A-curve Y lying in X. We define the intersection index
of Y with an A-divisor D such that supp Y n supp D is disjoint from
the generic fiber of ~ :
 70

(10) <Y,D> : ~ (Y'Df)vlOg qv + [ <Y'D>v

vESf vES

Here (Y,Df) for v 6 Sf denotes the sum of the local intersection

indices of Y and D in the closed points of Xf over v, qv is the
order of the r e s i d u e field. Furthermore, for v6 S we set

<Y,Df> v : -log Gv(Df,v,Yv) ,

-I
<Y, ZrvX(V)> = X rv(I v°l v)
vES vES Xv

An e q u i v a l e n t definition is o b t a i n e d if one puts (I0) and (4) together.

Denote by o*(I D) the section of the A - s h e a f o*(0(D)) on Y, induced
by 0(D) . T h e n

(11) <Y,D> : -Zog 77 io*(I D) Iv

V

From the p r o d u c t formula one sees that the r i g h t hand side of (11) re-
mains unchanged if one takes a different non zero section of 0*(0(D))
instead of O*(ID) . This justifies the following general definition of
degree of an i n v e r t i b l e A-sheaf i on the A - c u r v e Y = Spec R:

(12) deg [ : -log ~ Isl v

V

One can take any non zero meromorphic section s of i in (12).

2. The Riemann-Roch theorems.

I. The g e o m e t r i c Riemann-Roch theorems. We shall recall first the

simplest Riemann-Roch-Hirzebruch theorem for p r o j e c t i v e manifolds over
a field. Let X be a d-dimensional manifold, i an invertible sheaf on
it. Set X([) = E(-1) i d i m Hi(i) and d e n o t e by c1(L), tdi (X) the C h e r n
i
and T o d d classes respectively. Then

d
I
(1) x(L) = < [ ~ c1(i)i tdd_i(X) >
i=0

where < > in the right hand side of (I) m e a n s the intersection index
calculated in the C h o w ring or in a c o h o m o l o g y ring with characteristic
zero coefficients. In p a r t i c u l a r
 71

d i
(2) X(L) : [ n i
<c1(1) t d d _ i (X) >
i=0

In this section we shall describe three particular cases of a w o u l d be

Riemann-Roch theorem for A - m a n i f o l d s , for the p r o j e c t i v e space, A-curve
and A-surface respectively, the last case being by far the d e e p e s t one.
As we h a v e said already, the first problem is to d e f i n e XA(L)

2. P r o j e c t i v e A-space. L e t us c o n s i d e r the A - m a n i f o l d , for simplicity

over ~ • pd = ( P i "~)' where ~ is a K i h l e r i a n f o r m on pd(~) . We
shall realize P~ as Proj S(Tz), where TZ is a Z - f r e e module of
the rank d+1 , and we set T = ~ ~ T~ . There is a c a n o n i c a l hermitian
metric on 0f(n) whose curvature from is a m u l t i p l e of ~ . We shall
denote by 0(n) the c o r r e s p o n d i n g A-sheaf. Since

Hi(Ip d ,0(n) ) = Sn(T Z) , H i ( p d, 0(n)) = 0,

(n~d) (Sn(T))
for i >0 , n >_ 0 , we m u s t choose a volume form w n6 Aft
and then define

w
XA(O(n)) = log I~1 n
#

where v E AZ (n~d) (Sn(TZ)) is one of the g e n e r a t o r s of this cyclic

n
group.

The simplest imaginable choice of w is the following one. Consider

n
the isomorphism
n n~d)
(n~d) .d+1
~n:Az (Sn(TZ)) ÷ [I\z (TTz) ]

n (nad,
® ~ J n ,n+d,
which maps vn onto vI (if ~-ft d ~ ~ ~ one can still
correctly define ~n@(d+1) , which suffices for our needs). Now choose
somehow wI and set
® n ,n+d)

wn (<0n~ idlR) \W 1 )

Then

w I ® ~ + 1 ( n d d)
n (n+d, (0(I))
×A(0(n)) = log iV~11 d~l d J XA
 72

In v i e w o f (8), n°1, the tensor multiplication of a n A - s h e a f i by

r r
0(K P(~)) multiplies the metric o n i~ b y e x p ( -' n v o l P ) " A s s u m i n g
the corresponding change of w I , we g e t

r r(d+IL
XA(0(]) ® 0( n ]? (~))) : XA(0(1)) + n vo i P

and finally

n (n+d. .n+d r
(3) XA(0(n) ~0(rm(~))) : d$~ d }XA(0(1)) + ~ d ) vol ~

Comparing (3) w i t h (I) and (2) we see that pd looks like a (d + 1)-
dimensional geometric manifold. We can also guess the Todd A-classes
tdAd+1_i(~d) .

3. A - c u r v e . Let X : Spec R, n = [R:Z] : r I + 2r2, r I : card {v 6 S IKv:R}.

Denote by i = ([f;hv) an invertible A-sheaf on X, L = H°(X,if) .
According to n°.I.3, we have

XA(L) : vol(( @ L ~ Kv)/L) .

v6S

We choose the volume form on ~ L~K v i m p l i c i t in t h i s d e f i n i t i o n

vtS~
following A. Weil [36] and S z p i r o [33] in t h e f o l l o w i n g w a y :

2n
w : q~ wv ,
2 r i T r2 v6S

where w is t h e v o l u m e f o r m c o r r e s p o n d i n g to the Euclidean metric on

v
L defined by h With this choice, the following statements, closely
V V
parallel to t h e case of curves over finite fields, are valid.

The Riemann-Roch theorem:

X A ( L ) = d e g L + X A ( 0 x)

where deg i is d e f i n e d by (12), n°1.

The Euler number of the structure sheaf:

X i ( 0 X) = r 2 log ~ - ~ log IAKi ,

 73

where £K is the d i s c r i m i n a n t of K.

o
Furthermore, set H~(L) = { s £ L I Isl v ~ 1 for all v} Then HA(0):{0]U
{roots of u n i t y in R}, which is the a n a l o g of the constant field. From
the M i n k o w s k i theorem one e a s i l y deduces that XA(L) k 0 implies
o
HA(i) + 0

4. A - s u r f a c e . An A - s u r f a c e X = (Xf,ev), according to A r a k e l o v and

Faltings, is a s e m i s t a b l e family of c u r v e s Xf ~ Spec R with smooth
irreducible generic fiber of g e n u s g > 0 and the following metrics
at infinity:

~v : V ° i v = I k =~1
2zi Vk,v A --
Vk'v

Here (ml,v,...,mg,v) is a b a s e of the d i f f e r e n t i a l s of the first kind

on X orthonormal with respect to the scalar product
V

I
<m,v'> 2"~i S m A ~'.
Xv(~)

We d e n o t e by ~f the r e l a t i v e dualizing sheaf of ~. T h e n Rv : ~I x v .

The canonical A-sheaf ~ = {~f,hv) is u n a m b i g o u s l y defined by the fol-
lowing prescription which normalizes hv . F o r an a r b i t r a r y point
xE Xv the r e s i d u e map res x : ~Ix v ® 0Xv(X) + ~ is an i s o m e t r y of the
geometric fiber of the former sheaf and of { .

5. The E u l e r characteristic. Faltings [10] defines XA(L) for an in-

vertible A-sheaf i on X in the following way.

The decisive step is the definition of c a n o n i c a l metrics on the spaces

d e t K v H ° ( i v) ~ d e t K S H 1 (iv) ~ Kv' v6 S , in the case H°(L v) = HI (Lv) = 0.
This being done, Faltings uses this case as the induction base with
respect to the o r d i n a r y degree of i on the g e n e r i c fiber. To this end
he r e p r e s e n t s i in the form L0(D) , where H°(L0, v) = H I (i0,v) : 0

and D is a h o r i z o n t a l A-divisor which can be t a k e n as a sum of sections

after a base extension. One can then simultaneously define XA(L0(D))
and prove the R i e m a n n - R o c h formula if o n l y one e s t a b l i s h e s the in-
dependence of this construction on the c h o i c e of the isomorphism
i : L0(D) which is h i g h l y non-unique.
 74

This independence is v a l i d for the following particular choice of

metrics on all L0's simultaneously. Set PiCg_IX v= M v and denote
by Iv the u n i v e r s a l sheaf over Xv × Mv . Let IT2:X v x M v ÷ M v by the
projection map. We can c o n s t r u c t the invertible sheaf det R ~ 2 , E v on
M v. Its g e o m e t r i c fiber at a p o i n t y6 M v corresponding to the sheaf
iv(Y) : Iv I (X v x {y}) can be c a n o n i c a l l y identified with

det H ° ( i v ( Y ) ) ® det-]H1{iv(y)) ® Kv

Over the set U = { y 6 Mv I H°(Lv(Y)) : H1(Lv(Y)) = 0} the sheaf

det R ~ 2 , E has a canonical unit section. On the o t h e r hand, M v~U is
the theta-divisor, a n d an e a s y consideration shows that under the suit-
able identification det R ~ 2 , [ = 0My(-8) the u n i t section goes into I.
Therefore, a choice of an A - s t r u c t u r e on 0M(-8) normalizes all
XA(L0). Faltings proves that the 8-polarization induces precisely the
A-structure suitable for the inductive argument.

We can n o w state the R i e m a n n - R o c h .

I
6. T h e o r e m . a) XA(0{D)) : ~ <D,D-K> + XA(~) , where 9 = 0(K) , < > is
the intersection index defined in n°1.6.

b) XA(~) = -i~( < K , K > + 6 ) , 6 = [ 6v(Xv),

yES

where 6v(X v) : log card (singular points of X(v)) for v C Sf ; for

v 6S , 6 is a real analytic function on the m o d u l i space of R i e m a n n
V
surfaces which measures the d i s t a n c e of X to the b o u n d a r y .
V

Of c o u r s e , in the g e o m e t r i c case, the N o e t h e r formula b) follows from

the R i e m a n n - R o c h - G r o t h e n d i e c k theorem applied to the m o r p h i s m ~.

The structure of @ (X v) for v6 S vaguely agrees with our p h i l o s o p h y

v
about X(v) as a d e g e n e r a t e fiber.

Faltings proves the N o e t h e r formula by an a r g u m e n t using the m o d u l i

space of X instead of Picg_1 of the first part.

The governing idea always is to use some canonical A-structures on the

moduli spaces and t h e i r tautological sheaves, to a p p l y the o r d i n a r y
 75

Riemann-Roeh-Grothendieck to the f i n i t e part and then to " c o m p a c t i f y "

this information by the K ~ h l e r i a n geometry.

Hence we n e e d the A - g e o m e t r y of a r b i t r a r y dimension anyway, e v e n to

deal w i t h A - s u r f a c e s only. In the n e x t s e c t i o n we shall d i s c u s s w h a t
is to be done to p u t this p r o g r a m on a f i r m f o u n d a t i o n .

3. P r o s p e c t s and p r o b l e m s of A - g e o m e t r y

I. The p r o b l e m of the d e f i n i t i o n of the fundamental categories.

In the d e f i n i t i o n of A - m a n i f o l d s given in 1 . 1 . no c o n d i t i o n s on the

K~hlerian forms ~ were imposed. However, the A r a k e l o v and F a l t i n g s
v
theorems are p r o v e d for d i s t i n g u i s h e d K i h l e r i a n structures. W e shall g i v e
the t e n t a t i v e definitions in a m o r e g e n e r a l context.

Let Iv be a l o c a l l y free sheaf on a c o m p a c t K i h l e r i a n manifold

(Xv,~V) and hv a Hermitian metric on Iv. We c h o o s e h o l o m o r p h i c
local c o o r d i n a t e s (z ~) on X and a b a s e of local h o l o m o r p h i c
v
sections (si) of Ev and set hij : h v ( S i , S j ) . The c u r v a t u r e tensor
of the c a n o n i c a l connection associated with h is
v

F. = - 92hij + h ab 9hib ~hij

13~8 ~za~[~ ~z ~ ~8

where (h ab} = (hij)-1 . Set ~v = - ~ g~8 dz~ A d~8 and (gy$) = (gas)-1
Then ([v,hv) is c a l l e d a Hermite-Einstein sheaf if

ge~F i jab = ~ h 13 '

where l is a c o n s t a n t . (It can be e x p l i c i t y calculated: setting

n = dim X v we g e t ~ = (2~n ~ c1([v)~n-1)v / (rk [ ~ nv ))"
v v

On the h o l o m o r p h i c tangent sheaf IX there is the h e r m i t i a n metric

v
gv = 2gab d z a d ~ " The m a n i f o l d (Xv,Wv) is called the Hermite-Einstein
manifold if (TXv,g v) is a H e r m i t e Einstein sheaf.

The e x i s t e n c e and u n i q u e n e s s problems for H e r m i t e - E i n s t e i n structures

 78

on a sheaf E were considered by K o b a y a s h i [19] and D o n a l d s o n [9].

v
Kobayashi has shown that the e x i s t e n c e of such structure implies the
semistability of E and t h a t any s e m i s t a b l e H e r m i t e - E i n s t e i n sheaf
v
(Ev,hv) is a d i r e c t sum @ (Ev( i) , hv(i) ) w i t h s t a b l e Ev (i)
(Stability h e r e m e a n s that the f u n c t i o n ~(F) = deg0~ r F/rk F, w h e r e
n-1
deg~v F = f c I (F)~v , is m o n o t o n o u s on subsheaves F c E) .
Xv

On the o t h e r hand, Donaldson proved that on p r o j e c t i v e algebraic sur-

faces (Xv,~v) any stable sheaf has a unique Hermite-Einstein metric
(up to a m u l t i p l i c a t i v e constant). The same is true for a l g e b r a i c
curves and, conjecturally, for all p r o j e c t i v e manifolds.

Deep existence and u n i q u e n e s s properties of K i h l e r - E i n s t e i n metrics

on m a n i f o l d s were obtained by Yau [38] and Aubin [3]. From Yau's
results it f o l l o w s in p a r t i c u l a r that for c](X v) = 0 each cohomology
class of K ~ h l e r i a n metrics contains a unique Kihler-Einstein metric.
Aubin has established the e x i s t e n c e on X of a u n i q u e K~hler-Einstein
v
metric with the c o n s t a n t I = -] under condition that ci (X v) contains
a form with negative definite metric.

Ourlimited understanding of A - g e o m e t r y suggests the special role of those

A-manifolds for w h i c h (Xv,~v) are K ~ h l e r - E i n s t e i n . This condition
appears to be a r e a s o n a b l e analog of the m i n i m a l i t y of Xf over Spec R.
Furthermore, on a g i v e n A-manifold, the f o l l o w i n g d e f i n i t i o n of a local-
ly free A - s h e a f seems plausible enough: it is the data E = ( E f ; h v , V 6 S ~)
for w h i c h Ef is a l o c a l l y free sheaf on Xf and (Ev,h v) are
Hermite-Einstein sheaves on (Xv,~v). For rkEf = I this is our initial
definition.

The category-theoretic aspects of t h e s e definitions need clarification.

Since the Hermite-Einstein property may possibly be r e l e v a n t only for

locally free and semistable sheaves, to d e f i n e a s u b s t i t u t e for c o h e r e n t
A-sheaves one probably is b o u n d to c o n s i d e r something like "perfect
complexes of l o c a l l y free A-sheaves", as in [35]. Unfortunately, the
differential geometry of c o m p l e x e s of sheaves in a d e r i v e d category
is not sufficiently developed.

The c o m p l e x e s of A - s h e a v e s must have a torsion invariant XA. For

 77

v 6S the c o r r e s p o n d i n g component of XA under certain conditions

should be given by the Q u i l l e n construction [30].

I do not k n o w h o w to define m o r p h i s m s of A - m a n i f o l d s . The p r o b l e m seems

to be related to the h y p e r b o l i c i t y t h e o r y by K o b a y a s h i [18]. In fact, it
is based on the study of m o r p h i s m s D÷ X and X + D, w h e r e
v v
D = {zC~, IzbS1} is the analog of Spec Z . T h e r e f o r e it can be con~
P
sidered as the c o u n t e r p a r t of the theory of Zp-models of @p-manifolds.

2. The p r o b l e m of canonical A-structures on moduli spaces and of

A-moduli spaces. The A r a k e l o v and F a l t i n g s work shows the e x i s t e n c e of
distinguished A-structures on the moduli spaces of curves and of in-
verti b l e sheaves on a curve, by which one arrives at good statements
of the principal results. (For the moduli space of curves having a
boundary this statement should be c o n s i d e r e d as heuristic, granting the
existence of good definitions).

One should study from this view point the moduli spaces of stable sheaves
on a curve, with rank and degree relatively prime. The first u n s o l v e d
problem is to g e n e r a l i z e the R i e m a n n - R o c h - A r a k e l o v - F a l t i n g s theorem to
the A - s h e a v e s of a r b i t r a r y rank, where the second Chern A-class c2A(E)
should emerge,an intersecting new invariant.

When a c a t e g o r y of A - s p a c e s is p r o p e r l y defined one w o u l d n a t u r a l l y

hope for e x i s t e n c e of m o d u l i - o b j e c t s in this category. Of course, the
first p r o b l e m s here are again c o n n e c t e d with the situation "at infinity",
i.e. the K ~ h l e r i a n geometry. In this respect a recent work of Koiso [20]
deserves to be mentioned. Koiso shows in p a r t i c u l a r that the base space
of normal and stable family of K ~ h l e r - E i n s t e i n structures carries a
canonical K~hler structure. Unfortunately in most cases it is u n k n o w n
whether it in a d d i t i o n satisfies the E i n s t e i n equation.

3. The p_~obl@m of i n t e r s e c t i o n theory of A - m a n i f o l d s . In the important

paper [4] A. B e i l i n s o n defined regulators for K - t h e o r y and i n t r o d u c e d
the general technique for c o n s t r u c t i o n of i n t e r s e c t i o n theory on
A-manifolds. We shall b r i e f l y describe here a part of his results,
stressing the role of K - t h e o r y as a c o h o m o l o g y theory.

Let Xf be a regular projective scheme flat over Z , dim Xf = d + I

 78

We fix two cycles zi of pure codimension Zi in Xf . Assume that

~0 + ZI = d + I and supp (z 0 A Zl) n Xf,~ = ~. We shall describe Beilinson's
construction of the A - i n t e r s e c t i o n index <z0,zl > . F o r simplicity, we
shall assume that both cycles have zero cohomology classes on the g e n e r i c
fiber. In this case we can forget ~ since the intersection index will
not depend on this metric. We g e t

<z0,zl > = < z 0 , z 1 > + ~ log p ' < z 0 , z 1 > p

p6Sf
and define the fiberwise indices < > with the h e l p of several cohomo-
V
logy theories.

To c a l c u l a t e < >~ we shall use the D e l i g n e - B e i l i n s o n cohomology H~ .

For a smooth compact complex manifold X and a coefficient ring
B c~ , B(j) : (2~ -/~)JB, this cohomology is d e f i n e d as follows:

H~(X ,B(j)) : RKF(x ,B(j)0) ,

B(J) D = Cone (F j @ B(j) ~ ~') [-I],

where F j : ~J (the t r u n c a t e d complex of h o l o m o r p h i c forms), ~ : ~I-e2,

ei the n a t u r a l injections. In the n o n c o m p a c t case, the forms with
logarithmic singularities are used. The now standard homological methods
permit us to d e f i n e the D - c o h o m o l o g y of simplicial schemes, the relative
D-cohomology, to d e f i n e the classes of a l g e b r a i c cycles and to p r o v e the
• /

Polncare duality theorem.

N o w we r e t u r n to the situation described earlier and set Ui:X -supp z itoo .

If the c l a s s e s of z.l,~ in H D21i(X ,IR(li)) vanish, the M a y e r - V i e t o r i s
sequence shows that the classes c/Dzi6 H~ii(X,Ui,IR(ii)) are of the form
~i where ~i C H~ 21i-I (Ui,19(~i)) . We can construct the class

60 U 6 1 6 H~d(u0(] UI)IR ( d + 1))

and its image

2d+I
(z 0 n Zl) ~ = ~(%0L~%1)6 H D (X,IR(d + I) )

Let ~:X~÷ Spec R be the structure morphism. The final formula for
 79

the A - i n t e r s e c t i o n index at t h e a r i t h m e t i c a l infinity is

< z 0 , z 1 > ~ = ~,(z 0 N Zl) ~ 6 H?1 ( S p e c m ,~ ( I ) ) = {/m(1) =

To d e f i n e <zl,z2> p in a s i m i l a r way, Beilinson introduces the

K-cohomology:

where ~P are the Adams operations. In t h i s case also the r e l a t i v e

version and the formalism of the cyclic classes c a n be d e f i n e d . Setting
U i = Xf ~ supp zi as e a r l i e r we c a n n o w c o n s t r u c t the intersection class

(z 0 N z l ) f C H K2 d + 2 ( X f ,U 0 U U 1 ; ~ ( d + I))

Set S' = supp ~f(Xf ~ (U 0 U UI)) , w h e r e [f:Xf+ Spec Z is the structure

morphism. This is a f i n i t e set of p r i m e s . We have <z0,zl > = 0
for
P
for p { S' , and for p 6 S' the index <~0,Z1>p is a sort of d i r e c t
image zf(z 0 N zl) f localized at p.

4. T h e p r o b l e m of t h e Euler A-characteristic a n d of the R i e m a n n - R o c h

A-theorem. I cannot add much to w h a t has been said earlier. Two remarks
may be in o r d e r .

First, granting that the d e f i n i t i o n of XA in a g e n e r a l situation can

be d o n e in t e r m s of the a n a l y t i c torsion of the D a l b e a u l t complexes, we
shall need the r e l a t i v e analytic torsion to t r e a t the g e n e r a l Riemann-
Roch-Grothendieck case.

Second,independently of t h e conjectural general theory, very interesting

and directly accessible problems of A - g e o m e t r y m a y be found, e.g., in
the t h e o r y of flag m a n i f o l d s G/P Z . A recent work by Bombieri-Vaaler
[6] is an e x a m p l e . It s u g g e s t s in p a r t i c u l a r that the classical Min-
kowski "geometry of n u m b e r s " should be interpreted in A - g e o m e t r y as a
theory of c h a r a c t e r i s t i c classes at t h e arithmetic infinity.

4. S u p e r s p a c e

I. E x a m p l e s of superspaces. A smooth or a n a l y t i c manifold can be described

 80

by a f a m i l y of local c o o r d i n a t e systems and transition functions. Be-

fore introducing a formal definition of s u p e r s p a c e , we shall g i v e
several examples of s u p e r m a n i f o l d s with the h e l p of local c o o r d i n a t e s .

a) The mln -dimensional affine s u p e r s p a c e. It has g l o b a l coordinates

(x I ..... X m ; ¢ I , .... ~n ) , w h e r e xi commute among themselves and with
Cj and Cj anticommute. In the c a t e g o r y of s u p e r s c h e m e s over a
commutative ring A the r i n g of f u n c t i o n s on the r e l a t i v e a f f i n e
mln-space is the G r a s s m a n n algebra with generators ~j over polynomial
ring A [ x I .... ,Xm]. In the c a t e g o r y of C -supermanlfolds the ring of
functions is C~ (Xl ,... 'Xm) [¢I ,. ""¢n]

b) The mln-dimensional projective superspace. It is d e f i n e d by the

atlas U i, i = 0 , . . . , m , each Ui being a mln-dimension~l affine space.
It is c o n v e n i e n t to i n t r o d u c e a homogeneous coordinate system
( X 0 , X 1 , . . . , X m ; Z ] , . . . , Z n) and to r e l a t e the c o o r d i n a t e s
(x~, j #i i~ ..... ¢i) by s e t t i n g x!=3 Xj/Xi' cji = Zj/X i"

c) The supergrassmannian of the d0idl-dimensional linear superspaces

in the (d O + c 0 i d I + C l ) - d i m e n s i o n a l linear superspace. W e shall d e s c r i b e
it Dy the f o l l o w i n g standard atlas. Consider matrices of the f o r m
(d O + d I) × (d0+c 0 + d I + c I) divided into four b l o c k s such t h a t the f o r m a t
of the u p p e r left b l o c k is d O x (d O + Co). For each s u b s e t I of c o l u m n s
containina dO columns of the ]eft p a r t and dO columns of the r i g h t
part c o n s i d e r the m a t r i x

cO dO dI c.

I 0
xI 0 ~I dO
0 I
(I) zI =
] 0
~I 0 xI dI
0 I

< )
I

The c o l u m n s I in ZI f o r m the identity matrix. All the r e m a i n i n g

places are f i l l e d by the i n d e p e n d e n t even and odd variables x~b ,¢~d
 81

even places being in the u p p e r left and lower right blocks. These vari-
ables (xi,<i) are the coordinates of the local chart U I. D e n o t e by
BIj the submatrix of ZI formed by the columns with indices in J.
-I
Then the transition rules are Z j : B I j Z I.

Setting in t h i s p r e s c r i p t i o n d 0 1 d I : II0, d O + c 0 1 d I + c I = m + 1in, we g e t

the p r o j e c t i v e superspace. On the o t h e r hand, setting dI = cI= 0 , we
get an o r d i n a r y grassmannian.

Proceeding in a m o r e systematic way, we shall start with several basic

notions of superalgebra and then define superspames by means of a
structure sheaf.

2. S u p e r a l g e b r a . The algebraic composition laws relevant in g e o m e t r y

are n a t u r a l l y divided into additive and multiplicative ones. All additive
groups in s u p e r a l g e b r a are e n d o w e d with a ~ 2-gradation and all m u l t i -
plications are c o m p a t i b l e with it. W e u s e notation A = A0 @ A I and
= s in c a s e a 6 A s , then ~= ~ + ~. The e l e m e n t s of A0 are called
even ones and those of AI o d d ones. The characteristic feature of t h e
superalgebra is t h e a p p e a r a n c e of c e r t a i n signs ±1 in all definitions,
axioms and polynomial identities of fundamental structures.

We shall give a representative l i s t of e x a m p l e s .

Let A = A 0 @ A I' b e an a s s o c i a t i v e ring. The supercommutator of h o m o -

geneous elements a,b 6 A is d e f i n e d by the formula [a,b] = a b - (-1)~ba.
The ring A is c a l l e d supercommutative iff [a,b] = 0 for all a,b
If 2 is i n v e r t i b l e (which we shall always assume), a 2 = ~1[ a , a ] = 0 for
all a 6 A I. T h e supercommutators in g e n e r a l satisfy two identities

[a,b] = -(-1) a'a~ [ b , a ] ,

[a,[b,c]] + (-1) ~(~+~) [b,[c,a]] + (-1} ~(~+~) [c,[a,b]] : 0 .

These identities (together with superbilinearity) are t a k e n as t h e de-

finition of L i e superalgebras. The ring morphisms, by definition, res-
pect the g r a d a t i o n .

Let A be a supercommutative ring. The notions of (Z2-graded, of

 82

course) left, right and bimodule S over A coincide, just as in the

commutative case, left and right multiplications being connected by the
formula as = (-I)~ s a . A new feature is the p a r i t y - c h a n g e functor:
(I I S) 0 = SI,(I I S) I = S O , r i g h t m u l t i p l i c a t i o n by A coincide on S
and TTs . An A-module S is c a l l e d free of rank Plq iff it is
isomorphic to A plq= A P@ (HA)q ~ T h e tensor algebra of A-modules differs
from the ordinary one b y the introduction of Z sign into certain
canonical isomorphisms, e.g. ~ : S ® T~ T® S is d e f i n e d by
~(s® t)= (-1)st t ® s. T h e r e are internal Hom's in t h e c a t e g o r y of
A-modules consisting of o r d i n a r y morphisms a n d a l s o of o d d o n e s , with
the linearity rule f(as) = (-I) ~ a f(s) .

The morphisms between the free A - m o d u l e s c a n be g i v e n by matrices.

One must not forget that the p a s s a g e from the left (row) coordinates
to the r i g h t (column) coordinates of an e v e n element implies sign
change in o d d c o o r d i n a t e s etc. The matrices are o f t e n written in t h e
standard format, like (I) w h e r e the e v e n - e v e n places are k e p t in the
upper left block. The group GL represents the functor of the in-
vertible matrices corresponding to e v e n m o r p h i s m s .

F. B e r e z i n has invented the superdeterminant, or B e r e z i n i a n ,

B e t : G L ( p l q , A ) ÷ A T . It is a r a t i o n a l function of the elements of the
matrix which in the standard format is g i v e n by the formula

I
Ber B 3 B4] = det (B I - B 2 B 4 - 1 B 3 ) d e t B 4

The kernel of B e r is d e n o t e d SL. Since Ber is r a t i o n a l , supergrass-

mannians fail to h a v e PlUcker coordinates, as we shall see l a t e r .

The Berezinian of a free module Ber S is d e f i n e d as a f r e e m o d u l e of

rank Ii0 or 011 (depending on the p a r i t y of q in rkS : p(q)
freely generated by any element of the form D ( s I ..... Sp+q) where (s i)
is a f r e e b a s e of S, w i t h the r e l a t i o n s
m ( f ( s ] ) .... ,f(Sp+q)) = B e r f - D ( s I .... Sp+q). This notion is a s p e c i f i c
substitute for the maximal exterior power in c o m m u t a t i v e algebra.

The bilinear forms on a free A - m o d u l e T with the symmetry conditions

are divided into four main types: OSp (even s y m m e t r i c ) , SpO (even
 83

alternate), I-~O (odd symmetric), ~ S p (odd alternate). The form b~ on

~-~T is d e f i n e d by the formula b~(~tl,]Tt2) = (-])tb(t],t2). This cons-
t r u c t i o n p r e s e r v e s parity but changes the symmetry of the form so that
usually it suffices to consider only types 0Sp and ]~Sp.

The a u t o m o r p h i s m functor of a n o n - d e g e n e r a t e form defines the super-

groups of the c o r r e s p o n d i n g type (we shall consider b e l o w only the split
ones). Besides SL, OSp and ~Sp, there is in superalgebra one more
series Q of supergroups of classical type: the c e n t r a l i z e r group of
an odd involution p : T + T, p = I, p2 = id. The Lie s u p e r a l g e b r a s of
these groups, slightly d i m i n i s h e d i f necessary (to kill a center etc.)
c o n s t i t u t e the classical part of the Kae c l a s s i f i c a t i o n [15] of simple
finite dimensional Lie superalgebras. There are also two exceptional
types (one having p a r a m e t e r s ) a n d C a t z d a n ~ y p e superalgebras of formal
vector fields w h i c h h a p p e n to be f i n i t e - d i m e n s i o n a l when defined on
G r a s s m a n n algebras.

A superderivation X:A÷ A v e r i f i e s the Leibniz formula

~N
X(ab) = (Xa)b + (-I) xa aXb. There are two natural modules of r e l a t i v e
d i f f e r e n t i a l s of a c o m m u t a t i v e A--algebra B:~vB/A and ~ddB/A,
c l a s s i f y i n g even and odd d i f f e r e n t i a l s r e s p e c t i v e l y . Later on we use
mainly ~ddB/A since the c o r r e s p o n d i n g de Rham complex is super-
c o m m u t a t i v e while for ~I B/A it is super anticommutative.
ev

3. Supergeometry. The most general known notion of a "space with even

and odd coordinates" is that of superspace. Superspace is a pair (M,0M) ,
where M is a toplogical space, 0M a sheaf of local supercommutative
rings on it. M o r p h i s m s of superspaces a r e m o r p h i s m s of locally ringed
spaces compatible with the g r a d a t i o n s of structure sheaves.

All objects of the m a i n g e o m e t r i c c a t e g o r i e s , - d i f f e r e n t i a b l e and analytic

manifolds, analytic spaces, schemes, - are t r i v i a l l y superspaces, with
0 M = 0M, 0. Such superspaces we shall call purely even ones.

In the general case we set J M = 0M" 0M,I, Gr i 0 M : j~/j~+In

s . Furthermore,
Mrd: (M,Gr00M), GrM = (M,I~
0': Gri0M) .

The structure sheaf GrM has a natural ~ - g r a d a t i o n . To consider GrM

as a superspace we reduce it modulo 2.
 84

With the help of these c o n s t r u c t i o n s we can define the most simple and
important class of superspaces. We shall call a superspace (M,0 M) a
supermanifold, a n a l y t i c or algebraic, iff a) Mrd is a pure even mani-
fold of the r e s p e c t i v e class; b) the sheaf 0M is locally isomorphic
to the sheaf Gr0M, which is in turn isomorphic to the G r a s s m a n n algebra
of the locally free (over Gr0M) sheaf JM/J~ of finite rank. (Note
that this G r a s s m a n n algebra should be called symmetric in the super-
algebra since JM/J~ is of pure odd rank).

One proves then that C~ and analytic s u p e r m a n i f o l d s can be d e s c r i b e d

by local charts (x~ ..... Xm; ~I ..... ~]n) . The sheaf 0M locally consists
of the expressions E f (x)~ e where f are even functions of the
corresponding type. An e s s e n t i a l feature of s u p e r g e o m e t r i c c o n s t r u c t i o n s
is their invariance with respect to the c o o r d i n a t e changes mixing even
the
and odd functions. Set,goese'g'' y = x + ~ i ~ 2 ,+~ I =~ (I TI1~2)x2)~ I, 8 2 = f~2iy)~h~2
" 12 + x
local function f(x) into f(y-~+xL = f(y)-
I
The appearance of d e r i v a t i v e s in the c o o r d i n a t e change formulas plays
an essential role in the L a n g r a n g i a n f o r m a l i s m of supersymmetric field
theoretic models. It also shows that in c o n t i n u o u s supergeometry a
natural structure sheaf ought to contain certain distributions. The pe-
c u l i a r i t i e s of c o n t i n u o u s s u p e r g e o m e t r y were not studied for this
reason.

The most important superspaces which are not n e c e s s a r i l y s u p e r m a n i f o l d s

can be easily defined in the analytic category. They are the super-
spaces (M,0M) such that (M,0M,0) is an analytic space and 0M, I
is 0M,0-coherent. In the same way one defines superschemes.

The notion of a locally free sheaf of 0M-mOdules is a natural substi-

tute for vector bundles. For s u p e r m a n i f o l d s over a field the tangent
sheaf TM and the c o t a n g e n t sheaf ~ I M = 9~dd M are defined in the usual
way, using s u p e r d e r i v a t i o n s over the ground field. The g e n e r a l i z a t i o n to
the relative case is selfevident. The rank of TM is called the di-
mension of the supermanifold.

Now the reader will easily transcribe the d e s c r i p t i o n s given in n°4.1

into the d e f i n i t i o n s of the s u p e r s p a c e s in the algebraic, a n a l y t i c or
C~ categories. Notice that the g r a s s m a n n i a n is e n d o w e d with the tauto-
logical sheaf, which is g e n e r a t e d by the rows of ZI over U I. Over
 85

the p r o j e c t i v e superspace it is d e n o t e d 0(-I).

4. M e t h o d s of c o n s t r u c t i o n of superspaces, a) Let (M,0 M) be a p u r e

even locally ringed space, [ a locally free sheaf of m o d u l e s of rank
01q over 0M . Then (M,0M, s - S([)) is a s u p e r s p a c e , which is c a l l e d
split. In the C~ category every supermanifold is split, i.e. can be
obtained in this way from a manifold and a vector bundle on it. In the
analytic and algebraic categories this is not true anymore, e.g. the
Grassmannians are n o t analytically split unless the tautological bundle
is of pure even or p u r e odd rank. (cf. b e l o w ) .

b) As in p u r e even analytic and algebraic geometry, very important

superspaces are defined by t h e i r functor of points. We h a v e already
mentioned the a l g e b r a i c supergroups GL. SL , OSp, ~Sp and in the
next section we shall w o r k .with the flag superspaces F(dl,...,dK;T),
where d I < d 2 < ...< d K are the dimensions of the components of a flag
in the l i n e a r superspace T. We e x p e c t that the m a i n theorems on the
representability of v a r i o u s functors and moduli problems admit their
counterparts in s u p e r g e o m e t r y although the systematic w o r k has barely
begun. We shall state two results proved by A. V a i n t r o b which show the
existence of a local deformation theory of K o d a i r a - S p e n c e r type. The
basic definitions are r e a d i l y stated in the context of a n a l y t i c super-
spaces. The infinitesimal deformations are represented by the ring
{[x,~] i (X2,x~) .

Let M be a compact supermanifold, TM its t a n g e n t sheaf.

5. T h e o r e m . a) Let dim HI(M,TM)= alb . If H2(M,TM) = 0, then in the

category of a n a l y t i c supermanifolds there exists a local deformation
of M over B= {alb such that the K o d a i r a - S p e n c e r map
P:ToB ÷ H I ( M , T M ) is an i s o m o r p h i s m .

b) Any deformation of M over a supermanifold with surjective Kodaira-

Spencer map is c o m p l e t e ; in p a r t i c u l a r , it is v e r s a l , if p is
isomorphic, i

Let us g i v e an e x a m p l e . Let M be a c o m p a c t analytic supermanifold of

dimension 111 It is c o m p l e t e l y defined by the Riemann surface
M 0 = Mrd and the invertible sheaf ]~JM= L on it. A s s u m e that
 8@

genus M = g >I, deg i = 0 and i is not isomorphic to 0M@ In this

case rk HI(M,TM) = 4g- 314g- 4. The even part 4g- 3 of this dimension
corresponds to the c l a s s i c a l manifold Z of d e f o r m a t i o n s of the p a i r
(M0,L) which is f i b e r e d by J a c o b i a n s over the c o a r s e moduli space of
curves. Theorem 5 shows that outside of the zero section this manifold
Z is n a t u r a l l y extended to the supermanifold of odd dimension 4 g - 4.
This structure deserves further study.

6. T h e o r e m . Let M be a c l o s e d compact subsupermanifold of the c o m p l e x

supermanifold M'. Let N be a n o r m a l sheaf to M .

a) If d i m HO(M,N) = alb and HI(M,N) = 0, t h e r e exists a versal local

deformation of M in M' over B = {alb

b) A d e f o r m a t i o n of M in M' over B is c o m p l e t e iff the K o d a i r a -

Spencer map P:ToB ÷H0(M,N) is s u r j e c t i v e .

7. E x a m p l e . Let us r e t u r n to the d e f i n i t i o n of a s u p e r g r a s s m a n n i a n and

illustrate certain of our constructions. The supergrassmannian
G = G(II1; ~212) is c o v e r e d by four 212 -dimensional affine super-
spaces. The corresponding Zi-matrices are

ix11 ~i o I I~2 i o ~21

~I 0 Yl I , rl2 0 I Y2 '

11 x3 ~3 0!
0 D3 Y3 I ,
11 x4 o
0 T]4 I
~I
Y4

Using the p r e s c r i p t i o n in the b e g i n n i n g of t h i s section, one calculates

the t r a n s i t i o n functions, e.g.
 87

<x1iI <x4 -I

r~1 Yl q4 Y4

-I -2 -I _x[ly 4 I~4
: I x4 -I+ x4-I Y4 ~4q4_i ' 1
-x4 Y4 84 ' Y4 -x4-1y4-2~4q4 /

It follows immediately, that Grd = p1 x p1 A calculation shows also

that

2 ~ £2(pI p1
JG x )

so that the obstruction ~ to the splitting of G lies in the group

H1(T(p1xp1) ® £2(p1×p1)) = H1(p1,~1p1)2 : { ~ ~ .

One can check directly, using the ~ech cocycle in the standard atlas,
that this obstruction is (I,1). Hence G is not split. (Notice that
the projective superspace is split: pmln = (pm,s(770~m(_1))) "

Moreover, G is not a projective supermanifold. In fact, the image

of the Picard group HI(G,0~, 0) ÷ Pic (pIxpI) consists of the classes
of sheaves 0(a,-a), a6 ~, since the obstruction to extending 0(a,b)
from Grd to G is essentially (a+ b)~. Therefore, any supergrass-
mannian G(alb;~ mln) with 0 < a < m, 0 < b < n is non-projective since
G admits a closed embedding in such a Grassmannian.

This example shows that the use of projective technique in the algebraic
supergeometry is restricted, and one is obliged to generalize those
methods of algebraic and analytic geometry which do not rely upon the
existence of ample invertible sheaves.

For example, P. Deligne conjectured that the dualizing sheaf on a smooth

complex supermanifold X in
Ber ~I X. This was proved by I. Penkov [29]
ev
who has demonstrated that in this case working with g-modules on a
 88

supermanifold permits one to e f f e c t i v e l y reduce the situation to the

pure e v e n one.

5. S c h u b e r t supercells.

I. B a s i c notions. Let G be a s e m i s i m p l e algebraic group, Bc G its

Borel subgroup. The G-orbits of G / B x G/B form a finite stratification
on this manifold whose strata Yw are n u m b e r e d by elements w of the
Weyl group W = W(G) (Bruhat decomposition). We shall call locally
closed submanifolds Xw(b) = ({b} × G/B) N Y c G/B the Schubert cells.
w
In the same way, using G-orbits of G / B x G/P, one defines the Schubert
cells for a parabolic subgroup P c G. The geometry of the S c h u b e r t
cells plays an i m p o r t a n t role in m a n y developments of c h a r a c t e r i s t i c
classes theory and representation theory.

In t h i s section we shall define Schubert supercells for c o m p l e t e flag

superspaces of c l a s s i c a l type and explain how some c l a s s i c a l results
generalize in t h i s context.

Let T be a linear superspace of d i m e n s i o n mLn over a field. We shall

consider the following algebraic supergroups G given together with
their fundamental representation T: a) G = SL(T) ; b) G = 0Sp(T) , the
automorphism group of a n o n d e g e n e r a t e even symmetric form b : T ÷ T* ;
c) G = ~ S p ( T ) , the a u t o m o r p h i s m group of a n o n d e g e n e r a t e odd alternate
form b:T÷ T*; d) G = Q(T), the automorphism group of an o d d involution
2
p:T÷ T , p : id. In t h e cases ~Sp and Q we h a v e m = n.

The counterpart of t h e classical manifold G/B is the supermanifold

F of c o m p l e t e flags in T, w h i c h are in a d d i t i o n invariant with respect
to b or p in the c a s e s G = OSp, ~Sp , Q (the e x a c t definitions are
given below), several differences between this situation a n d the c l A s s i -
cal o n e are worth mentioning.

First, the stabilizers of c o m p l e t e flags B in g e n e r a l are n o t m a x i m a l

solvable subsupergroups. However, they play the same role as the c l a s s i c a l
Borel subgroups both in the t h e o r y of h i g h e s t weight [16] and in the
theory of t h e Schubert supercells. Second, not all subgroups B are
pairwise conjugate, a n d the flag m a n i f o l d s F consist of several
 89

components. Third, the stratification we w a n t to c o n s t r u c t is not

purely set-theoretic. In fact it w i l l be a d e c o m p o s i t i o n of F x F
into a union of l o c a l l y closed subsuperschemes. The flag realization
is s u i t a b l e for this construction.

The first unsolved problem is to d e f i n e a cohomology theory in w h i c h

the c l a s s e s of the Schubert cells would be free generators. As we con-
jectured in the introduction, this may require a sort of c o m p a c t i f i -
cation along the o d d d i m e n s i o n s .

We shall now give some details.

2. The connected components of flag supermanifolds. These connected

components are n a t u r a l l y numbered b y the sets GI which are defined
as follows. Set

m+n

SLI = {(61 , .... 6re+n) I 6i = I}0 or 011, i=~ I 6 i = mln}

' "

Furthermore,

0SPI : {(6 1 ..... 6m+ n) 6 SLI 16 1 : 6 m + n + 1 _ i , i = I ..... m+n} ,

]IspI = {(6q , .... 62m) 6 S L I I 6 i = 6 c2 m + 1 - i . i. =. I., ,2m} ,

m
L ...........
]
where (plq) c = qlp • Finally, QI = {~I11 ..... I11)} (the one element
set). We s a y that the flag f: 0 = SoC S 1 c . . . CSm+ n = T is of type
I 6GI , if 6i(I) = rk Si/Si+ 1 . For groups G = OSp, -~'Sp, Q the
flag f is c a l l e d G-stable if the following conditions are fulfilled:

b(S i) = SI . for OSp, ~Sp ; p(S i) = S. for Q .

re+n-± 1

The functor GF I on the category of superschemes over ground field

associates with a superscheme S the set of flags of t y p e I 6 GI in
the sheaf 0S ® T ; for G # SL the flags should be G - s t a b l e . (A flag
is a f i l t r a t i o n of 0s ® T by subsheaves Si such that all injections
S. c S . locally split; G-stability is d e f i n e d with respect to
id0s ® b or id0s ® p ).
 90

3 6 Theorem. Functor GF I is r e p r e s e n t a b l e by a s u p e r m a n i f o l d which

is i r r e d u c i b l e except for G = 0Sp(2rl2s) , r _->I, in w h i c h case GF I
consists of two i s m o r p h i c components. Furthermore,

dimSLFi = <m(m 2 I ) + n(n-2 I ) 1 mn ) ;

(r 2 + s2 I (2r+ 1)s) for m = 2r + I, n = 2s > 0;

diO S~I = 2
(r 2 - r + s I 2 rs) for m = 2r, n = 2s > 0

dimJ~S~i= <rs + r ( r - I) + s(s + I) rs + r ( r + I) + s(s 2 I)~

2 2 ~ '

m
where (rls) = [ 6i(I) . •
i=I

Of course, the f u n c t o r s of n o n c o m p l e t e flags also are r e p r e s e n t a b l e and

the m o r p h i s m s of p r o j e c t i o n onto a subflag are r e p r e s e n t a b l e by m o r p h i s m s
of s u p e r m a n i f o l d s . It is c o n v e n i e n t to p r o v e theorem 3 by i n d u c t i o n on
the length of a flag, starting with relative Grassmannians as in s e c t i o n
4. The reader shall find most of d e t a i l s in [24].

N o w we shall set GF = I ~ I G F I and d e n o t e by Sic 0F ® T the c o m p o n e n t s

of the t a u t o l o g i c a l flag on FG . There are two n a t u r a l flags on
GF x GF , {p1*(Si) } and {p2*(Sj) } , where PI,2 are the p r o j e c t i o n s .
In the c l a s s i c a l theory every G-orbit consists of t h o s e points of
GF x GF , over w h i c h the type of r e l a t i v e position of flags {p1*(Si) ]
and {p2*(Sj) } is fixed. We can imitate this d e f i n i t i o n in s u p e r g e o m e t r y
taking functor of p o i n t s instead of g e o m e t r i c points.

The type of r e l a t i v e position of c o m p l e t e flags (S i ) and (Sj) in T

is, by d e f i n i t i o n , the m a t r i x di:J = rk(S i + S:) •
3

Let us i n t r o d u c e the Weyl g r o u p s GW , acting on GI :

SLw = Sm+n ; QW = Sm ; GW:{g 6 S m + n l g ( G I ) c GI} for G =0Sp,~Sp.

 91

The r e a d e r will n o t i c e that our GW is in g e n e r a l different f r o m the

W e y l g r o u p of Grd , e.g. the l a t t e r is Sm x Sn for SL a n d not
S . As we shall see in a m o m e n t , in the t h e o r y of S c h u b e r t super-
m+n
cells it is this big group, which contains the odd r e f l e c t i o n s , which
is the r i g h t one.

4. Lemma. The t y p e s of r e l a t i v e positions of c o m p l e t e G-stable flags

are in (1,1)-correspondence w i t h the t r i p l e s (I,J,w) r where I , J ~~ GI r

w 6 GW , J = w(I) • .

The proof is p u r e l y combinatorial.

5. B r u h a t subsets. W e set n o w for w6 GW , I,J = GI :

!Yw,IJ != {x 6 G F I × GFj Irk(P1*Si(x) + p2*Sj(x)) : dij,w}

where (dij,w) is the type of r e l a t i v e position corresponding to the

triple (I,J,w) in v i e w of L e m m a 4 (if J % w(I) we set iYw,iJ I = ~) .
Furthermore, put

iywi : _~__ {Yw,iJ I c GF x GF

I,J

6. T h e o r e m . Each set iYwl carries a canonical s t r u c t u r e of the

locally closed subsuperscheme Y c GF x GF , such that the d e c o m p o s i t i o n
W
I I Yw is the f l a t t e n i n g stratification for the f a m i l y of s h e a v e s
W
Sij = P1* Si + P2 *S j on GF X GF m.

We r e c a l l that by d e f i n i t i o n of a f l a t t e n i n g stratification this con-

dition means that each m o r p h i s m q : X + GF x GF for w h i c h all the sheaves
q*(Sij) are f l a t u n i q u e l y decomposes as X + IjYw+GF × GF. The p r o o f
of the e x i s t e n c e of the f l a t t e n i n g stratification is the same as in the
p u r e e v e n case.

7. S u p e r l e n g t h . In the c l a s s i c a l theory the d i m e n s i o n Qf a S c h u b e r t

cell a s s o c i a t e d with w6 W equals the m i n i m a l l e n g t h of a d e c o m p o s i t i o n
of w into a p r o d u c t of b a s i c reflections. To state the c o u n t e r p a r t of
this fact in s u p e r g e o m e t r y we n e e d several definitions.

W e shall call the f o l l o w i n g elements of GW the b a s i c reflections:

 92

d. : (i,i + I) for G : SL,Q ;

l

di = (i,i+ I) (m+ n+ I - i, m + n - i) , i+ I < [m+n--~--]

m+n
T~ : (£,m+ n+ I - k) , ~ = [--~--] for- G = OSp, ~Sp.

The superlength Gz(w) , w C GW will be defined inductively . This is

a vector of superdimensions (GzIj(W) I I,J 6GI) such that

a) For a basic reflection o 6 GW we h a v e Gzij(a) : 0 if J # d(I) ;

the other possibilities are contained in t h e table:

SL OSp l lSp Q

110 (I=J); 110 (I=J); I 0 (l:J) ; 111

011 (I@J); 011 (I%J); 0 I (I%J) ,

T~ 110 (6~(I) = 1t0, 0 ] (~i(z):110)

mln = 2r + 112s); 0 0 (69 (r):011)

111 (6£(I) = 011,

mln : 2r + 112s) ;

010 (6Z(I) : 110,

mln : 2r12s) ;

110 (6Z(1) : 011,

mln : 2r12s).

b) Let w = o K . .. oi be an irreducible decomposition of w as a

product of b a s i c reflections. Set I = oi... ~I (I) and
l

K-I
Gilj(W) = [ G~{ (o i+I) , if S = w z);
i=0 ~Ii'Ii+l

GZIj(W) : 0 , if J % w(1) .

8. Theorem. The projection map Y GF I is s u r j e c t i v e , and locally

w,IJ
over GFI t h e Bruhat m a n i f o l d YW,IJ is a r e l a t i v e affine superspace
 93

of d i m e n s i o n Gilj(W) .

In o t h e r words, the d i m e n s i o n of a S c h u b e r t cell Yw(b) coincides with

the superlength of w.

From the g e o m e t r i c proof of the theorem some p u r e l y combinatorial facts

follow. For example, G£1j(w) does not d e p e n d on the c h o i c e of an
irreducible decomposition of w; furthermore, G£Ij(W) = G£ij(w-1
for G %~Sp, finally for G =~Sp

G£ij(w) + dim GF I Giji(w-1) + d i m GFj,

if J = w(I).

6. G e o m e t r y of supergravity

I. M i n k o w s k i space and Schubert cells. The objective of this section

is to d e s c r i b e a model of simple supergravity from the v i e w p o i n t which
was introduced in [26] where the k i n e m a t i c constraints of supergravity
were interpreted as the integrability conditions for a c u r v e d version
of a flag superspace.

To e x p l a i n the essence of our approach let us r e c a l l the u s u a l exposition

of g e n e r a l relativity. The space-time without gravitational field is the
Minkowski space of special relativity ~4 with a metric which in an
inertial frame takes the form dx 2 -iE3 I dx i 2 . The g r a v i t a t i o n field
reflects itself in the curvature of space-time which becomes a smooth
four-manifold M4 with the p s e u d o r i e m a n n i a n metric gabdxadxb . The
dynamics is g o v e r n e d by the L a g r a n g i a n (action density) R vol
g
where R is the scalar curvature, vol the v o l u m e f o r m of g.
g

The m o d e l s of supergravity in s u p e r s p a c e studied in m a n y recent works

[12],[13],[34] also start with certain geometric structures on a
differentiable supermanifold M mln which are then used to d e f i n e a
(super) Lagrangian which is a s e c t i o n of the sheaf Ber M : Bet ~I M.
ev
There are physically meaningful cases with m% 4 ,e.g. the c a s e mln=11132
is n o w considered as the m o s t fundamental one.
 94

What is still very much unclear, is the q u e s t i o n what exactly is the

geometry to start with, i.e. the k i n e m a t i c s of supergravity. The naive
suggestion to u s e a s u p e r m e t r i c was quickly seen inadequate. The most
universal known method is t h e C a r t a n approach. One starts with an affine
connection and then painstakingly guesses the so c a l l e d constraints and
the action density. The constraints are the differential equations which
must imply no e q u a t i o n s of m o t i o n . The physical interpretation and
quantization of c o n s t r a i n e d fields is a d i f f i c u l t task and one faces
the p r o b l e m of solving constraints and expressing everything in t e r m s
of f r e e fields. This approach was successful more than once but the
poor command of u n d e r l y i n g geometry hinders the w o r k considerably.

Our approach essentially interprets the constraints as i n t e g r a b i l i t y

conditions ensuring the e x i s t e n c e of certain families of submanifolds
in Mmln , the g e o m e t r y of t h e s e families being a curved geometry of
Schubert supercells.

Let us first describe from this viewpoint the simplest example, the
Pl~cker-Klein-Penrose model of M i n k o w s k i space.

Let T be a four dimensional complex space (Penrose's twistor space).

Let G = G(2;T) b e the G r a s s m a n n i a n of p l a n e s in T , S = Sl the
tautological sheaf on G , Sr =
(T ® OG/S~ )* There is a c a n o n i c a l
isomorphism ~IG : Sl~ S r , and the subsheaf A2S~ ~ A2S r c S2(~]G)
can be interpreted as the h o l o m o r p h i c conformal metric on G. Choose
a big cell UcG. The complement G~ U is a s i n g u l a r divisor, the
light-cone at infinity, and there are sections s£,c r of t h e sheaves
A2SZ,A2S r on U having a pole of first order at this divisor. The
complex metric sZ ® Sr 6 F ( U , S 2 ~ I G ) is w e l l defined u p to m u l t i p l i c a -
tive constant. Now introduce a real structure p on T @ T* inter-
changing T and T*. The involution p acts on G - p o i n t s of G(2,T)
since G(2,T) canonically identifies with G(2;T*) . Let p be com-
patible with (U,sl,Sr) in t h e sense that uP=u, sip = Sr" The
following statements c a n be d i r e c t l y verified.

a) The real (i.e. p-invariant) points of t h e b i g cell U f o r m the

space ~4 . The restriction of s Z ® Sr to it is a M i n k o w s k i metric.

b) The real three-dimensional Schubert manifolds in the G r a s s m a n n i a n

 95

G intersected with U(~) form the s y s t e m of light c o n e s of t h i s

metric.

c) There a r e no r e a l two-dimensional Schubert cells in G. In U(~)

they define two connected families of c o m p l e x planes. These families
play an essential role in t h e t h e o r y of Y a n g - M i l l s fields. In fact,
the integrability of a c o n n e c t i o n along one family means that this
connection is an (anti) self dual solution of the Y a n g - M i l l s equation.

In a c u r v e d space-time of g e n e r a l relativity null geodesics and light

cones still exist and, moreover, define the corresponding metric u p to
a conformal factor. To break the conformal invariance one m a y c h o o s e
metrics si,Cr for t w o - c o m p o n e n t Weyl spinors, left and right.

We d e s c r i b e supergravity along these lines. In n n ° 2,3 a flag m o d e l

of M i n k o w s k i superspace is i n t r o d u c e d , In n°4 w e e x p l a i n that in a
curved superspace two families of 0L2N - dimensional Schubert super-
cells should be p r e s e r v e d . Finally, in nn°5,6 we define the dynamics
by means of a n a c t i o n density, expressed through the Ogievetsky-Sokachev
prepotential [28].

2. M i n k o w s k i superspace. Fix an i n t e g e r N ~ I and a linear complex

superspace T of d i m e n s i o n 414N. Set M = F(2i0, 2iN;T), i.e. a
S-point of M is a f l a g S 2i0 c S 2LN in 0S® T . Moreover, define
the l e f t and right superspaces as G r a s s m a n n i a n s

MZ : G(210;T), M r = G(21N;T) = G(210;T*) .

Denote by S 210cS 21N the tautological flag in 0 M ® T, b y ~210c~21N

the orthogonal flag in 0 M ® T*. Set F~ = s 2 1 N / s 2i~
,
Fr
: ~21N/~210
Let ~i,r:M ÷ MZ, r be t h e c a n o n i c a l maps. Let TzM = TM/M r ,
TrM = TM/M l (recall that we work in t h e category of c o m p l e x super-
spaces). Since M over Mi, r is a r e l a t i v e Grassmannian, a standard
argument gives canonical isomorphisms

TIM= ($210) * ¢ Ft , TrM = E l * ® ( ~ 2 1 0 ) .

Combining this with the map Fl ® Fl ÷ 0M we get a natural map

 96

(i) a: T/M ® T r M ÷ ($210), ® (~210),.

On the other hand, the relative tangent sheaves T/,rM are integrable
distributions, i.e. locally free subsheaves of Lie superalgebras in
TM, of r a n k 012N. Every point of M is c o n t a i n e d in two closed sub-
supermanifolds of d i m e n s i o n 012N tangent to TiM and TrM
respectively. They are the Schubert cells we are interested in. The
supercommutator between T/M and TrM defines the F r o b e n i u s map

(2) b:TiM ® TrM÷ TM/(TIM + TrM) : ToM

The following statements contain the e s s e n t i a l geometric features of

the p i c t u r e we w a n t to k e e p in the c u r v e d case.

3. P r o p o s i t i o n . a) The sum TiM+ TrM in TM is a d i r e c t subsheaf

in TM of rank 014N.

= ® (~210)
b) There is a w e l l defined isomorphism TOM ($210) * *
making the m a p s (I) and (2) to c o i n c i d e . •

Finally, as in n°1, we m u s t introduce a real structure p on T @ T*

(in s u p e r a l g e b r a (ab) 0 ( - 1 ) a b a P b P ; cf. [24] for further details) .
We shall assume that T P = T* , in this case (T/M) p = Yr M,
(S210)p= ~210 . One can c h e c k that over a p-stable big cell in M some
natural sections of TiM, TrM, ToM generate the Poincare superalgebra
introduced by p h y s i c i s t s (see e.g. [34]).

4. C u r v e d superspace. A complex supermanifold M 414N with the following

structures will be c a l l e d superspace of N - e x t e n d e d supergravity.

a) Two integrable distributions T/M,TrM c TM or rank 012N whose

s u m is direct.

b) Two locally free sheaves S l, S r of rank 210 , two locally free

sheaves F1;Fr=F1 * of rank 01N and structure isomorphisms
TIM = Sl* ~ Fl" T r M = ~r ® St*

c) A real structure p on M such that its real points in Mrd form

a four-manifold, and e x t e n s i o n s of this real structure to Sl@ S r ,
 97

F1 @ F r interchanging left subsheaves with right ones.

d) V o l u m e forms v£, r 6 F(Mz,r, B e r MZ, r such that vz p = v r

A choice of these volume forms corresponds to the c h o i c e of spinor
metrics ~i'Sr in nO1.

This data is s u b j e c t e d to o n e axiom. Set ToM - TM/(TIM @ TrM) . T h e n

the F r o b e n i u s map ~: T~M ~ TrM +ToM coincides with the natural map
Sl*®Fl® ~® Sr*÷S~£ Sr* under the appropriate identification
T 0 M = Sl* ® S r * a s in Proposition 3 b) .

5. L a g r a n g i a n . Let M be a superspace of N - e x t e n d e d supergravity.

Using the d a t a above one can construct a canonical isomorphism:
2-N
(3) Bet M = [~i~* B e r M ~ ~ ~r * B e r M r ]4-N ( N % 4)

(From t h i s p o i n t o n we d e f i n e Ber M as Ber*(~1odd M) ) .

Hence the v o l u m e forms vg~ , v r make it p o s s i b l e to d e f i n e a section

(for N • 4)
N-2
N-4
(4) w = (~* v ~ ® ~r*Vr) 6 F(M, B e r M)

In t h i s w a y w e g e t for N = I the c o r r e c t action of sample supergravity.

In the c a s e N = 2 the action is c e r t a i n l y wrong since it g i v e s trivial
equations of m o t i o n . It s e e m s that considering N as a formal parameter
and taking the left (or right) part of the c o e f f i c i e n t of the T a y l o r
expansion of (4) at N = 2 we g e t an a c t i o n suggested b y E. Sokaehev.
Anyway, for N > I one must take into account new constraints which
might take the form of integrability of m o r e Schubert cells.

It is a l s o certain that the o t h e r types of flag supermanifolds and their

curved ~ersions are n e c e s s a r y for a fuller understanding of supergravity
and super Yang-Mills equations. For example, in a r e c e n t paper by
A. Galperin, E. Ivanov, S. K a l y t s y n , V. O g i e v e t s k y a n d E. S o k a c h e v
the m a n i f o l d F(210,2Jl,2L2;T) implicitly appears which in t h e curved
version c a n be d e f i n e d as t h e p r o j e c t i v i z e d bundle P(Fz) = P(Fr)+ M.

In t h e same v e i n , for t h e largest physically acceptable case N = 8,

 98

the 11132-dimensional flag s u p e r m a n i f o l d FI2{0,211,2{8,T) or its

curved version P(F£) ÷ M seems to be the space c o n s i d e r e d in the
context of the so c a l l e d d i m e n s i o n a l reduction, or the g e n e r a l i z e d
Kaluza-Klein model.

6. P r e p o t e n t i a l . To c o n c l u d e , we g i v e some c o o r d i n a t e calculations
which make it p o s s i b l e to i d e n t i f y our g e o m e t r i c picture with that
of the a r t i c l e [28]. Set N = I and c h o o s e in Mi a local c o o r d i n a t e
system (xza,@~). Assume t h a t the f o l l o w i n g properties are true:
functions, (X~)rd are p-stable and f u n c t i o n s (xa. = ~ ( x za + x~),
I
04,8 ~ ) are local c o o r d i n a t e s on M, w h e r e Or~ =(~ ~ ) P, x a
r = (x~)P)
Such coordinates (xz,@ Z) on M£ , (Xr,0 r) on M r and (~,0£,@ r) on M
w i l l be c a l l e d distinguished ones.

N o w we set

a a)
(5) Ha = ~ i (x£ - xr .

These four real n i l p o t e n t superfunctions on M are c a l l e d the O g i e v e t s k y -

Sokachev prepotential. Working locally and i d e n t i f y i n g Fl, r with
,~0 M we can say t h a t the p r e p o t e n t i a l completely defines the g e o m e t r y
of the superspace, except for the forms v£, r w h i c h m u s t be g i v e n
separately :

(6) V£,r 3
: ¢Z,r D * ( d O ~ , r dO~ ,r )

Some calculations (cf. [26] for details) s h o w that the a c t i o n (3) c a n

be e x p r e s s e d through (5) and (6) by m e a n s of the W e s s - Z u m i n o formula

(7) w = ~I Bet (E~) D*(d@~,d@~ ,dr a)

where E AB is the t r a n s i t i o n matrix between the f r a m e s

~, ~ = (~a,~ ,;~) and (~[~ , ~ ] , A a , A ~ ) .

~x a ' 80 Z ~0 r

This last frame can be d e f i n e d in t h r e e steps.

Step I. A = ~ + xa~ and A-= -2.- xa~ are d e f i n e d as local

~ a a ~ e e a
bases for TiM and TrM r e s p e c t i v e l y . F r o m this one f i n d s the
 99

coefficients

xai ~: [ (I- i~)-I] ab ; ~ = -i (I + i )-I a

b ~&

Step 2. The volume forms vz, r define the spinor metrics [£,r 6 BerSl, r:

El : (zi*vi)]/3 ® (~r*Vr)-2/3' [r : (~l*vl)-2/3® (nr*Vr) 1/3

Step 3. The m u l t i p l i e r F, defining ~ : F£ and £. = FP£. , is con-

strutted in such a way, that D*(£ ) = [l' D*(~) = Er .

The structure frame (~[A

, ~ ,A~],~
) _ can be used to describe the
g e o m e t r y of simple supergravity Cartan style. In this approach it
appears as the final product rather than the starting point.

REFERENCES

[i] S.J. Arakelov. An intersection theory for d i v i s o r s on an a r i t h m e t i c

surface. - Izv. Akad. Nauk SSSR, ser. mat. 1974, 38:6, pp. I]79-1192
(in russian).

[2] S.J. Arakelov. Theory of intersections on arithmetic surface. -

Proc. Int. Congr. Vancouver, 1974, vol. I, pp. 405-408.

[3] T. Aubin. Equations du type de M o n g e - A m p ~ r e sur les v a r i e t & s

k ~ h l e r i e n n e s compactes. - C.R. Acad. Sci. Paris, 1976, 283,
pp. 119-]21.

[4] A.A. Beilinson. Higher regulators and the values of L-functions.

- Sovremenny problemy matematiki, vol. 24, Moscow, VINITI 1984
(in russian), pp 181-238.

[5] F.A. Berezin. I n t r o d u c t i o n to algebra and calculus with an-

ticommuting variables. - Moscow U n i v e r s i t y Press, 1983 (in russian).

[6] E. Bombieri, J. Vaaler. On Siegel's lemma. - Inv. math. ]983,

vol. 73:1, pp. 11-32.

[7] E. Date, M. Jimbo, M. Kashiwara, T. Miwa. Solitons, T-functions

and e u c l i d e a n Lie algebras. - in: M a t h @ m a t i q u e et Physique,
S&minaire ENS 1979-82, Birkh~user, Boston 1983, pp. 261-280.

[8] P. Deligne. Preuve des c o n j e c t u r e s de Tate e t S h a f a r e v i t c h (d'apres

G. Faltings).- Sere. Bourbaki, Nov. 1983, N 616.
 100

[9] S. Donaldson. Stable h o l o m o r p h i c bundles and curvature. -

Oxford, p r e p r i n t 1983.

[10] G. Faltings. Calculus on a r i t h m e t i c surfaces.Ann.Math. 119 (1984),

pp. 387-424.

[11] G. Faltings. E n d l i c h k e i t s s i t z e fur a b e l e s c h e V a r i e t ~ t e n ~ber

Zahlk~rpern. - Inv. Math., 1983, 73, pp. 349-366.

[15] S.J. Gates, M.T. Grisaru, M. Rocek, W. Siegel. Superspace or one

thousand and less lessons in supersymmetry. - London, Benjamin,
1983.

[13] G e o m e t r y ideas in physics. Ed. by Yu.I. Manin (in russian). -

Moscow, Mir Publishers, 1983.

[14] K. lwasawa. A n a l o g i e s b e t w e e n number fields and function fields.

- Preprint, 1965, Lecture g i v e n at the Yeshira Science Conference.

[15] V.G. Kac. Lie superalgebras. - Advances in Math., 1977, 26:1,

pp. 8-96.

[16] V.G. Kac. R e p r e s e n t a t i o n s of classical Lie s u p e r a l g e b r a s . - in:

Lecture Notes in Math., N 676, pp. 597-627, Springer 1978.

[17] V.G. Kae, D.A. Kazhdan. S t r u c t u r e of r e p r e s e n t a t i o n s with h i g h e s t

w e i g h t of infinite d i m e n s i o n a l Lie a l g e b r a s . - A d v a n c e s in Math.,
34, (1979), pp. 97--108.

[~8] Sh. Kobayashi. H y p e r b o l i c m a n i f o l d s and h O l o m o r p h i c mappings.-

New York, M. Dekker, 1970.

[19] Sh. Kobayashi. C u r v a t u r e and stability of v e c t o r bundles. -

Proc. Jap. Ac. Sci., 1982, 58:4, pp. 158-162.

[20] N. Koiso. E i n s t e i n m e t r i c s and c o m p l e x strutures.- Inv. Math.,

1983, 73:1, pp. 71-106.

[21] B. Kostant. G r a d e d manifolds, graded Lie theory and p r e q u a n t i z a t i o n .

- in: Lecture N o t e s in Math., N 570, pp. 177-306, Springer 1977.

[22] S. Lang. Fundamentals of D i o p h a n t i n e geometry.- Berlin, Springer,

1983.

[23] D.A. Leites. I n f o r m a t i o n to theory of supermanifolds. (in russian).

Uspekhi, 1980, 35:1, pp. 3-57.

[24] Yu. I. Manin. Gauge fields and complex geometry. (in russian).
Moscow, Nauka, 1984.

[25] Yu. I. Manin. Flag s u p e r s p a c e s and s u p e r s y m m e t r i c Y a n g - M i l l s

e q u a t i o n s . - in: A r i t h m e t i c and geometry. Papers d e d i c a t e d to
I.R. S h a f a r e v i c h on the o c c a s i o n of his sixtieth birthday, vol. 2,
pp. 175-198, Birkh~user, 1983.

[26] Yu. I. Manin. The g e o m e t r y of s u p e r g r a v i t y and the S c h u b e r t super-

cells. (in russian).- Zapiski nachn, sem. LOMI, Leningrad, 1984,
vol. 133, pp. 160-176.
 101

[27] B. Mazur, A. Wiles. A n a l o g i e s b e t w e e n function fields and number

fields. - Amer. Journ. of Math., 1983, ]0%,pp. 507-521.

[28] V.I Ogievetsky, E.S. Sokatchev. The g r a v i t a t i o n a l axial super-

field and the f o r m a l i s m of the d i f f e r e n t i a l geometry. (in russian).
- Yad. fiz., 1980,31:3, pp. 821-839.

[29] I.B. Penkov. B-modules on supermanifolds. -Inv. Math., 1983, 71:3,

pp. 501-512.

[30] D. Quillen. A r i t h m e t i c surfaces and analytic torsion.- Bonn,

A r b e i t s t a g u n g , 1983.

[31] I.R. Shafarevich. A l g e b r a i c number fields.-Proc. Int. Congress

of M a t h . , S t o c k h o l m , 1962.

[32] I.R. Shafarevich. Minimal models of two d i m e n s i o n a l schemes.-

Tata I n s t i t u t e of Fund. Res., 37, B o m b a y ]965.

[33] L. Szpiro. La conjecture de Mordell (d'apres G.Faltings) .- S@m

Bourbaki, Nov. 1983, N 619.

[34] Superspace and s u p e r g r a v ] t y (ed. by S.W. Hawking and M. Rocek).

-Cambridge Univ. Press, 1981.

[35] Th&orie des i n t e r s e c t i o n s et t h 6 o r @ m e de R i e m a n n - R o c h . -

Lecture Notes in Math., N 225, S p r i n g e r 1971.

[36] A.Weil. Sur l ' a n a l o g i e entre les corps de n o m b e r s a l g 6 b r i q u e s et

les corps de fonctions alg&briques.-in: Collected Papers, vol. I,
pp. 236-240. New York, S p r i n g e r ]980.

[37] A. Weil. De la m & t a p h y s i q u e aux m a t h & m a t i q u e s . - i n : C o l l e c t e d

Papers, vol. 2, pp. 408-412. New York, Springer 1980.

[38] S.T. Yau. On C a l a b i ' s c o n j e c t u r e and some new results in algebraic

geometry. - Proc. Nat. Acad. Sci. USA, 1977, 74, pp. ]798-1799.
 C O M M E N T A R Y ON THE A R T I C L E OF M A N I N

Michael Atiyah
Mathematical Institute
O x f o r d OX1 3LB

31. Mathematics and P h y s i c s

Manin's stimulating contribution to the 25th A r b e i t s t a g u n g which,

in his absence, I a t t e m p t e d to present, p r o v i d e d m e w i t h an o p p o r t u n i t y

of a d d i n g s o m e f u r t h e r r e f l e c t i o n s of my own. This c o m m e n t a r y , which

is t h e r e f o r e a v e r y p e r s o n a l r e s p o n s e to M a n i n ' s article, c o n s i s t s of

very g e n e r a l and s p e c u l a t i v e r e m a r k s about large areas of c o n t e m p o r a r y

mathematics. Such s p e c u l a t i o n s are, for good reason, r a r e l y put down

on paper but the r e c o r d of the 25th A r b e i t s t a g u n g provides a rather

singular o c c a s i o n w h e r e ideas of this t y p e may not be out of place.

In r e c e n t years t h e r e has b e e n a r e m a r k a b l e r e s u r g e n c e of the

traditional links b e t w e e n m a t h e m a t i c s and physics. A n u m b e r of

striking ideas and p r o b l e m s from t h e o r e t i c a l p h y s i c s h a v e p e n e t r a t e d

into v a r i o u s b r a n c h e s of m a t h e m a t i c s , i n c l u d i n g areas such as a l g e b r a i c

g e o m e t r y and number theory w h i c h are r a r e l y d i s t u r b e d by such o u t s i d e

influences. Perhaps a few s p e c i f i c e x a m p l e s will i l l u s t r a t e the

point. The K a d o m t s e v - P e t i a s h v i l l i equation, w h i c h a r o s e in p l a s m a

physics, has b e e n s h o w n to be e x t r e m e l y r e l e v a n t to the c l a s s i c a l

S c h o t t k y p r o b l e m a b o u t the c h a r a c t e r i z a t i o n of J a c o b i a n v a r i e t i e s of

algebraic curves (this was e x p l a i n e d in the A r b e i t s t a g u n g l e c t u r e of

van der Geer) . Witten's analytical a p p r o a c h to the M o r s e i n e q u a l i t i e s ,

based on the p h y s i c i s t ' s use of s t a t i o n a r y - p h a s e a p p r o x i m a t i o n , has led

D e l i g n e and others to i m i t a t e his ideas in n u m b e r t h e o r y w i t h g r e a t

success. The Y a n g - M i l l s equations and their 'instanton' solutions

have been brilliantly e x p l o i t e d by D o n a l d s o n to s o l v e o u t s t a n d i n g

p r o b l e m s on 4-manifolds.
 104

All these examples c o n n e c t p h y s i c s w i t h v a r i o u s b r a n c h e s of

geometry, and it is t h e r e f o r e n a t u r a l that M a n i n s h o u l d h a v e a t t e m p t e d

an o v e r v i e w of g e o m e t r y in the w i d e s t sense. The p i c t u r e he d e s c r i b e s

is b e s t i n d i c a t e d by the f o l l o w i n g s c h e m a t i c diagram:

Arithmetic < Algebra Geometry Analysis ~ Physics

Topology

Homology K-theory

As this suggests, ideas from topology, n o t a b l y h o m o l o g y and

K-theory, p r o v i d e a c o m m o n l a n g u a g e and u n d e r p i n n i n g for the w h o l e

structure. The b r i d g e b e t w e e n geolaetry and a r i t h m e t i c was g r e a t l y

e x p a n d e d and d e v e l o p e d d u r i n g the G r o t h e n d i e c k era w i t h the i n t r o d u c t i o n

of 'schemes'. The b r i d g e b e t w e e n g e o m e t r y and p h y s i c s b e g i n s essent-

ially w i t h E i n s t e i n ' s t h e o r y of g r a v i t a t i o n but has b e c o m e m u c h s t r o n g e r

w i t h the r e c e n t d e v e l o p m e n t of g a u g e t h e o r i e s of e l e m e n t a r y p a r t i c l e s .

The p i c t u r e just envisaged is r e s t r i c t e d , on the p h y s i c s side, to

c l a s s i c a l physics. However, one should be m o r e a m b i t i o u s and try to

fit q u a n t u m p h y s i c s into the p i c t u r e also. I w i l l h a v e m o r e to say on

this a s p e c t later.

§2. A r i t h m e t i c laanifolds

An a l g e b r a i c c u r v e d e f i n e d by e q u a t i o n s w i t h i n t e g e r c o e f f i c i e n t s

can be v i e w e d as a s c h e m e o v e r Spec Z. It is the a n a l o g u e of a

s u r f a c e m a p p e d onto a curve, the 'fibre' over a p r i m e p being the

curve reduced mod p. Such an a r i t h m e t i c s u r f a c e c a n be 'compactified'

by a d d i n g the R i e m a n n s u r f a c e of the curve over the 'prime' at ~.

The A r a k e l o v - F a l t i n g s t h e o r y is then c o n c e r n e d w i t h e x t e n d i n g as m u c h

as p o s s i b l e of the usual theory of surfaces to this a r i t h m e t i c case.

 105

For this purpose it t u r n s out t h a t o n e n e e d s to i n t r o d u c e or find

canonical metrics on v a r i o u s objects associated to the R i e m a n n surface.

For example, given a line-bundle ~ on the Riemann surface one has

the one-dimensional complex vector space

det H°(~)/det Hl(~)

(where d e t d e n o t e s the highest exterior power), and one wants a natural

metric on this space.

This particular problem which was solved in o n e w a y by F a l t i n g s

has b e e n examined in a w i d e r context (e.g. replacing ~ by a v e c t o r

bundle) by Quillen. He has shown that a natural definition arises by

using the regularized determinants of L a p l a c e type operators which were

introduced into d i f f e r e n t i a l geometry by Ray and Singer [3]. Such

operator determinants are extensively used b y p h y s i c i s t s in q u a n t u m

field theory, and this link b e t w e e n geometry and physics is c u r r e n t l y

the s c e n e of m a n y investigations. In a n y case it p r o v i d e s a clear

link w i t h quantum and n o t p u r e l y classical physics.

On t h e R i e m a n n surface itself there are two natural metrics (for

g ~ 2), one being the P o i n c a r 6 metric and the other being the metric

induced by the h o l o m o r p h i c differentials. In h i g h e r dimensions the

analogues of t h e P o i n c a r @ metric are the K[hler-Einstein metrics.

Similarly for stable vector bundles there are distinguished metrics,

and M a n i n proposes they should be u s e d for a higher dimensional

theory of a r i t h m e t i c manifolds. It is i n t e r e s t i n g to n o t e t h a t all

of t h e s e m e t r i c s arose in a p h y s i c a l context.

Thus the geometry of K [ h l e r manifolds, and in p a r t i c u l a r the

study of o p e r a t o r determinants on such manifolds, appears as a n a t u r a l

meeting point for a r i t h m e t i c and physics. In this context it is

perhaps worth pointing out that a number of d i f f e r e n t i a l - g e o m e t r i c

invariants constructed from operator determinants have already been

identified with quantities arising in n u m b e r theory. Thus Ray and

 106

Singer [4] m a d e a c o n n e c t i o n w i t h the S e l b e r g z e t a - f u n c t i o n of a

Riemann surface while Millson [2] did s o m e t h i n g similar in h i g h e r odd

dimensions. Also v a l u e s of c e r t a i n L - f u n c t i o n s of t o t a l l y real

number fields h a v e b e e n r e l a t e d in [i] to the e t a - i n v a r i a n t (essentially

the l o g a r i t h m of a c e r t a i n o p e r a t o r determinant) : the e t a - i n v a r i a n t

is also w h a t appears in [2].

§3. Fermions

T h e r e is a basic d i s t i n c t i o n in p h y s i c s between two types of

particles, n a m e l y bosons and fermions. Bosons involve commuting

variables and so are easily u n d e r s t o o d on a g e o m e t r i c level, b u t

fermions involve anti-commuting variables and so are m o r e m y s t e r i o u s

geometrically. On a p u r e l y a l g e b r a i c level of c o u r s e t h e r e is no

mystery: polynomial and e x t e r i o r algebras are b o t h w e l l u n d e r s t o o d

and e x t e n s i v e l y studied. However, the d e v e l o p m e n t of g a u g e t h e o r i e s

in p h y s i c s w h e r e g e o m e t r i c i n s i g h t and i n t e r p r e t a t i o n g r e a t l y a s s i s t

the p u r e l y formal a l g e b r a i c a s p e c t has naturally led to the a t t e m p t to

develop a 'super-geometry' in w h i c h b o t h sets of v a r i a b l e s are

incorporated.

The d e v e l o p m e n t of s u p e r - m a n i f o l d s as o u t l i n e d by M a n i n a p p e a r s

to be an e l e g a n t e x t e n s i o n of c l a s s i c a l g e o m e t r i c ideas and it should

throw light on the a l g e b r a i c c o m p u t a t i o n of p h y s i c i s t s who b u i l d

'super-symmetric' theories. Nevertheless the t h e o r y still a p p e a r s to

lack s o m e e s s e n t i a l i n g r e d i e n t s and M a n i n asks w h e t h e r the f e r m i o n i c

coordinates can s o m e h o w be 'eompactified' so as to make t h e m

topologically more interesting. In this c o n t e x t I w o u l d like to m a k e

a tentative suggestion concerning the right g e o m e t r i c way to i n t e r p r e t

f e r m i o n i c variables.
 107

Consider first a smooth manifold M, its de R h a m complex 2*(M)

and in p a r t i c u l a r the ring ~°(X) of smooth functions. For a super-

symmetric analogue, suppose now that M is a c l o s e d sub-manifold of

another manifold N, and let A = ~M*(N) denote the complex of

currents on N which are supported on M and smooth in the

M-directions (recall that a current is j u s t a d i f f e r e n t i a l form with

distributional coefficients). Locally an e l e m e n t of A can be

expressed in t h e form

E f (x)~G dx~ A dy 5
~BY

where x = (x l...xm) are coordinates on M, Y : (YI'''''Yr) are

normal coordinates to M in N, 6 is t h e D i r a c G-function of M,

~, B are skew-symmetric multi-indices and y is a s y m m e t r i c multi-

index (so t h a t ~Y represents derivatives in y). If we t a k e

Y
a,7 to be e m p t y and B to b e a s i n g l e i n d e x we g e t a subspace RG

of A where R is t h e super-ring of t h e super-manifold given by M

and its n o r m a l bundle in ~. On the other hand A itself should be

viewed as the s u p e r de Rham complex of t h i s super-manifold.

The advantage of t h i s p o i n t of v i e w is t h a t approximating the

G-function by s u i t a b l e smooth functions (e.g. Gaussians) we c a n try to

interpret fermions as b o s o n s on N which are very sharply peaked

along M. More precisely the fermions should appear as 'leading

terms' of such sharply peaked bosons. Geometrically this m i g h t

correspond to p u t t i n g a metric on N which is v e r y sharply curved

along M, so t h a t M is an 'edge of r e g r e s s i o n ' in t h e language of

classical differential geometry.

I am t r y i n g to s u g g e s t that super-geometry should be s o m e kind of

limit of ordinary geometry and not an e n t i r e l y different k i n d of

entity constructed simply by formal analogy.

 108

~4. The quantum level

Quantum theory is c h a r a c t e r i z e d by i n f i n i t e - d i m e n s i o n a l i t y and b y

non-commutativity. When trying to u n d e r s t a n d the p o s s i b l e geometric

counterpart of s o m e a s p e c t of q u a n t u m - t h e o r y this must be b o r n e in

mind.

As I have already mentioned the s t u d y of l i n e a r elliptic operators

provides one bridge between geometry and quantum field theory. For

example ideas from supersymmetric field theories have cast new light

on the index theorem.

In a different direction it is I t h i n k not inappropriate to

consider Connes' non-commutative differential geometry (see t h e survey

t a l k by C o n n e s in this volume) as a v e r s i o n of q u a n t i z e d geometry.

Recall that Connes studies situations s u c h as t h e ergodic action of a

discrete group on a manifold where the geometric quotient does not

exist in a n y w a y as a r e a s o n a b l e space. However, a non-commutative

algebra exists with which various geometric constructions can still be

made.

In the lecture of L a n g he explained a conjecture of V o j t a based

on an i n t e r e s t i n g a n a l o g y b e t w e e n arithmetic surfaces and N e v a n l i n n a

theory. It is p e r h a p s interesting in t h i s connection that John Roe

in h i s Oxford D.Phil. thesis shows how the Nevanlinna theory fits into

Connes' framework. Analysing this situation might shed l i g h t on the

analogy between Connes' theory and q u e s t i o n s in A r i t h m e t i c .

If o n e a s k s for t h e a n a l o g u e of q u a n t u m theory in A r i t h m e t i c one

can hardly avoid considering the whole Langlands programme. Adelic

groups are obvious analogues of g a u g e groups and Hilbert space

representations are the basic objects of t h e t h e o r y . This analogy

deserves closer scrutiny, particularly in view of the fact that non-

abelian dualities, generalizing class-field theory on the one hand

and electric-magnetic Maxwell duality on the other, seem to be a main

 109

objective in both number theory and physics. Perhaps our c l a s s i c a l

diagram should be enlarged to a q u a n t u m diagram in the following way:

Quantum
Langlands ÷ Connes + Field T h e o r y Quantum

Arithmetic + Geometry ÷ Physics Classical

References

i. M.F. Atiyah, H. D o n n e l l y and I.M. Singer, Eta invariants,

signature defects of cusps and values of L-functions, Ann. of

Math. 118 (1983), 131-177.

2. J. Millson, Closed geodesics and the q-invariant, Ann of Math.

108 (1978), 1-39.

3. D.B. Ray and I.M. Singer, R-torsion and the L a p l a c i a n on

Riemannian manifolds, Advances in Math. 7 (1971), 145-210.

4. D.B. Ray and I.M. Singer, Analytic torsion for complex manifolds,

Ann. of Math. 98 (1973), 154-177.

 The Mandelbrot Set in a M o d e l for P h a s e Transitions *)

Heinz-Otto Peitgen and Peter H. Richter **)

INTRODUCTION

According to D. R u e l l e [18] "... the main problem of equilibrium

statistical mechanics is to u n d e r s t a n d the n a t u r e of phases and phase
transitions ...". A remarkable observation of B. D e r r i d a , L. D e S e z e
a n d C. Itzykson [4] h a s put these problems of t h e o r e t i c a l physics in-
to a n e w p e r s p e c t i v e : For a very particular model (the h i e r a r c h i c a l
q-state Potts model on a h i e r a r c h i c a l lattice) they indicated that
the J u l i a set of the corresponding renormalization group transforma-
tion is the zero set of the p a r t i t i o n function in t h e c l a s s i c a l theo-
r y of C° N. Yang a n d T. D. L e e [22]. T h e Y a n g - L e e theory describes a
physical phase as a d o m a i n of a n a l y t i c i t y for t h e free energy, viewed
as a f u n c t i o n of c o m p l e x temperature. The boundaries of t h e s e domains
are given by the zeroes of t h e p a r t i t i o n function. Carrying on these
ideas we show a connection with a discovery of B. M a n d e l b r o t [13].
More precisely, in a d i s c u s s i o n of the m o r p h o l o g y o f the a b o v e zero
sets w e d i s c o v e r a structure which is r e l a t e d to the M a n d e l b r o t set
(see [15])attached to the o n e - p a r a m e t e r family ~ 9 z ~ z2 + c , c 6
a fixed constant. For this we exploit recent results of D. Sullivan
[21] w h i c h classify the stable regions of r a t i o n a l maps on 5 = { U {~}.
Though the p h y s i c a l meaning of the h i e r a r c h i c a l Potts model is c e r -
tainly very questionable it s e e m s that the c l a s s i c a l (see G. J u l i a [12]
a n d P. F a t o u [8]) and recent (see A. D o u a d y a n d J. Hubbard [5,6,7],
D. Sullivan [21], M. Herman [11]) t h e o r y of c o m p l e x dynamical systems
may produce a major step towards a deeper understanding of t h e n a t u r e
of p h a s e transitions. Besides the hierarchical Potts model we have
analyzed I- a n d 2-dimensional Ising models with and without an e x t e r -
nal magnetic field and have found that the t h e o r y of J u l i a sets and

*) T h i s p a p e r s u r v e y s the r e c e n t i n t e r a c t i o n b e t w e e n t h e t h e o r y of
p h a s e t r a n s i t i o n s in s t a t i s t i c a l m e c h a n i c s a n d the t h e o r y of c o m -
plex dynamical systems.

**) Forschungsschwerpunkt Dynamische Systeme, Universit~t Bremen

D-2800 Bremen 33
 112

their typical fractal properties play a very significant role in the in-
teraction b e t w e e n the Yang-Lee theory and the r e n o r m a l i z a t i o n group
approach. None of t h e s e and the findings in [15] w o u l d have been possi-
ble without t h e a i d of e x t e n s i v e computer graphical studies and experi-
ments.

PRELIMINARIES AND NOTATION

The hierarchical Potts model is a s s o c i a t e d with a very particular

and physically unrealistic lattice construction which we introduce
schematically in fig. I.

k = I
<> k = 2 k = 3

Figure I. The diamond hierarchical lattice with

n = n(k) = 4 + 2(4 k-1 -4)/3 atoms (dotts) and 4 k-1

bonds (line s e g m e n t s ) for k > I

For this particular lattice and nearest neighbor coupling an e x p l i c i t

f o r m of t h e renormalization group transformation is k n o w n and that is
why it is v a l u a b l e here. On each lattice site i we assume a spin with
q 6 ~ possible states

(]. = I I*o-i q •
1

The partition function Zk(T) is t h e s u m of B o l t z m a n n factors extended

over all configurations
 113

{o : {I ..... n} ~ {I ..... q}} , n = # of l a t t i c e points,

(I) Zk(T) = Z exp (- I

o k--~ E(°)) '

where E(o) is t h e p o t e n t i a l energy of t h e configuration o .

Assuming t h a t the interaction of d i f f e r e n t lattice sites is r e s t r i c t e d
to nearest neighbors only, i.e. only across a bond indicated by a line
segment in f i g u r e 1, t h e e n e r g y across such a bond for a fixed confi-
guration a is:

I
- U , if O. = 0
(2) E(i,j) = l j
0 , else.

Hence,
(3) E(o) : ~ E(i,j)
bonds

For convenience we introduce new variables

(4) x = exp(U/kB.T )

so t h a t Zn(X) becomes a polynomial in x with integer coefficients.

The coupling constant U is c h a r a c t e r i s t i c for the m a t e r i a l , U > O
for ferromagnetic, and U < O for a n t i f e r r o m a g n e t i c coupling. From
Zk one derives the free energy per atom

kBT
(5) fn - n in Z k , n = n(k)

Thus, zeroes of Zk correspond to logarithmic s i n g u l a r i t i e s of fn

and are reasonable candidates for p h a s e transitions. Note, however,
that Zk(X) # O for any finite lattice with n = n(k) points and for
all x > O , which is t h e p h y s i c a l l y meaningful temperature range.

THE YANG-LEE MODEL OF PHASE TRANSITIONS

In e s s e n c e the i d e a of C. N. Yang a n d T. D. L e e ~, which had a

substantial impact o n the forthcoming attempts to s o l v e p h a s e transi-
tion problems, is a s f o l l o w s :
 114

Let

(6) N k = {x 6 • , Zk (x) = 0] ,

i.e. one embeds the partition function in a c o m p l e x temperature plane.

To make boundary effects negligable one has to p a s s to the thermodyna-
mic limit, i.e. one lets n ~ ~ It is n o t obvious, of course, that
such a limit makes sense and exists. If, however, the potential energy
E admits an a p p r o p r i a t e growth condition and the range of the inter-
action is sufficiently small, which is trivially satisfied in o u r case,
then (see [18]) t h e limit exists and we denote by N the zero-set
of the partition function Z in t h e thermodynamic limit. Now Yang
and Lee postulated that N would distinguish a unique point Xc> O,

(7) N oa 0 ~ + = {x c } ,

so that T x c : exp(U/kB.Tc) , is t h e phase transition point

c I
(see fig. 2).

<<!iii!iii!iiTiiiiiiTi?Cii
:i:!iiii!!i{iiii!!iiii[i!iiiiii!.!.i:!:!:i:{:"Lgneti c

. :i!iiii!!iiiiii!!iiiiii!iiiii!i!iiiii!iiiiii e . . . .

Xo iiii!ii!iiiiiii!ii!iiiiiii!ii!i!ii!iiiiiiiiiiiiii
~ ~ : t i ! ! i ii!iii![iiiiitiii!ii!iiii!ii!iii!i!li :ti:!iiii[!
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::

Figure 2. Note that T = ~ corresponds to x = I

Thus to find and characterize T it r e m a i n s to find x and inter-

c c
prete N in the neighborhood of xc For example the critical in-
dex a , which characterizes the singularity of the specific heat,

(8) C ~ I T- Tc I- ~

can be obtained from the d e n s i t y of the zeroes in the thermodynamic

limit near x (see [9]).
c
 115

THE RENORMALIZATION GROUP APPROACH

In g e n e r a l the p a r t i t i o n function Zk is n o t only a function of

temperature x but also of o t h e r variables like for e x a m p l e an ex-
ternal magnetic field H . In e s s e n c e the i d e a of the r e n o r m a l i z a t i o n
group approach is to r e l a t e

Zk_ I with Zk , i.e.

(9) I Zk(X,H,...) = Z k _ I ( X , H .... ) • ~ ( x , H .... )

(x,H .... ) = R ( x , H .... )

Thus, up to a t r i v i a l factor ~ the partition function of step k is

obtained by that of step k-1 modulo an a p p r o p r i a t e adaption of the
variables (x,H, ...). This determines a map R , the renormalization
transformation.

In our specific hierarchial q-states Potts model Zk is o n l y a

function of x , the temperature variable. However, Zk depends on
the m a t e r i a l constant q . An elementary calculation shows that
(see [4], [I~)

Z k ( X ) = Zk_ I (x) • <D(x) , k~ 2

R(x) = x = ( x2+q-lh
(10)
\2x+q-2/ 2
Z I (X) = q(x+q-1)
2.4 k-2
~0(x) = (2x+q-2)

Thus, the r e n o r m a l i z a t i o n transformation is a r a t i o n a l map of d e g r e e 4.

Actually, as w e let q vary in C we o b t a i n a l-parameter family
R = R For any q we h a v e that
q

Rq(1) = 1 and R'(1)q = O

(11)
Rq(~) = ~ and R'(~)q = O ,

i.e. I and ~ are superstable attractors. Their basins of a t t r a c t i o n

are defined by (~ = ~ U {~})
 116

Aq(1) = {x 6 ~ : Rn(x) ~ 1 as n ~ ~}
q
(12)
Aq(~) = {x 6 ~ : Rn(x) ~ ~ as n ~ ~} i
q

As a c o n s e q u e n c e of the c l a s s i c a l theory of G. J u l i a [12] and P. F a t o u

[8] on the i n t e r a t i o n of r a t i o n a l functions in { we h a v e t h a t

(13) ~Aq(1) = Jq = 3Aq(~)

is the J u l i a set of Rq

J U L I A SETS A N D P H A S E T R A N S I T I O N S

We are n o w in a p o s i t i o n to d i s c u s s the Y a n g - L e e model in t e r m s of

the r e n o r m a l i z a t i o n approach f r o m the p o i n t of v i e w of the t h e o r y of
Julia sets. We b e g i n by listing a few i n t e r e s t i n g conjectures and
problems:

CONJECTURE 1.1.

N = Jq , i .e.

Aq(1) (resp. Aq(~)) corresponds to the non-

magnetic (resp. m a g n e t i c ) phase.

To d i s c u s s this c r u c i a l conjecture the f o l l o w i n g immediate observation

from (10) is of i m p o r t a n c e :

Note that Zk_1(Rq(X)) q(x) for some q(x) and

= /~(x)
therefore
k-1
(14) Nk = {x 6 ~ : Rq (x) = I - q}

Moreover, the free e n e r g y in the t h e r m o d y n a m i c limit f

satisfies the f u n c t i o n a l equation (15) as a c o n s e q u e n c e
of (5) and (10):

(I 5) l f

g(x)
(x) =-~-f (Rq(X))

:
Ii n
(2x+q-2)
+ g(x) , with
 117

PROBLEM 1.2.

(a) In w h a t sense is Nk ~ N as k ~ ~ ?

(b) F o r w h i c h q 6 ~ is N = J ?
q
(c) F o r w h i c h q E • does R admit further
q
attractors, other than I and ~ ?

If R admits a further attractor other than I and ~ t h e n its c o r -

q
r e s p o n d i n g b a s i n of a t t r a c t i o n m a y c h a r a c t e r i z e a t h i r d m a g n e t i c p h a s e
s u c h as for e x a m p l e the antiferromagnetic phase.

In v i e w of (14) and (15) conjecture (1.1) means that the

singularities of f~ , the p h a s e transitions points, are
given by points from Jq , and this is i n t i m a t e l y related
to an u n d e r s t a n d i n g of the forward and backward orbit

(16) Rq(1-q)}i E Z

for q 6 ~ . Thus, the question which remains is: Is

(1-q) E J q o r in w h i c h component of ~ ~ Jq is i t ?

THE CLASSIFICATION OF STABLE REGIONS

This leads us d i r e c t l y i n t o o n e of t h e m o s t celebrated recent re-

sults in the theory of complex d y n a m i c a l systems: The classification
of s t a b l e regions of D. Sullivan [14,21]. Let f b e an a n a l y t i c endo-
morphism of ~ . A point x 6 ~ is stable f o r f if on s o m e n e i g h -
borhood of x the family of i t e r a t e s f,f2,f3, ... is an e q u i c o n -
tinuous family of m a p p i n g s of that neighborhood into ~ . Note that
when x is n o t stable, i.e unstable, for a n y n e i g h b o r h o o d the u n i o n
of images of i t e r a t e s must cover ~ except for t w o p o i n t s at m o s t .
The s e t of u n s t a b l e points for f is t h e J u l i a set J of f . It is
the closure of the expanding periodic points. The open set of s t a b l e
points ~J consists of countably many connected components, the
stable regions of f , which are transformed among themselves by f .
The following three theorems of D. Sullivan [21] and P. F a t o u [8] a r e
crucial for c o n j e c t u r e 1.1 and problem 1.2 . Let f be a r a t i o n a l
mapping with d = deg(f) > 2
 118

THEOREM 1.3. (Sullivan)

Each stable region is e v e n t u a l l y cyclic.

(For a n y c o m p o n e n t C c ~ ~ J there is n 6 ~ such
that D = fn(c) is c y c l i c , i.e. fk(D) = D for some
k 6~ .)

THEOREM 1 . 4. (Sullivan)

The cycles of s t a b l e regions D are classified into five types:

(a) A n attractive basin D arises f r o m an a t t r a c t i v e periodic cycle

¥ with non zero derivative of m o d u l u s less than one,
¥ = {z,f(z) ..... fn-1(z) } , f n ( z ) = z , O < I (fn)' (z) I < I ,
and D consists of c o m p o n e n t s of

U {y : lira d i s t a n c e (fn(y) fn(x))=O }

x6¥ n-~o
containing points of ¥ .

(b) A parabolic basin D arises from a non-hyperbolic periodic cycle

¥ with derivative a r o o t of u n i t y ,
¥ = {z,f(z) ..... f n - 1 ( z ) } , fn(z)=z , ((fn)' (z))m=1 ,
¥ is c o n t a i n e d in t h e b o u n d a r y of D , and each compact in D
converges to ¥ under forward iteration of f .

(c) A superattractive basin D is d e f i n e d just like an attractive ba-

sin b u t n o w (fn), (z) = O .

(d) A Siegel disk D is a s t a b l e region which is c y c l i c and on which

the a p p r o p r i a t e power of f is a n a l y t i c a l l y conjugate to an i r r a -
tional ratotion of the s t a n d a r d u n i t d i s k .

(C.L. Siegel [I~ p r o v e d these occur near a non-hyperbolic fixed

point if the a r g u m e n t ~ of its d e r i v a t i v e satisfies the follo-
wing diophantine condition: there exists c > 0 and ~ > 2 such
that
i a- P/q I ~ c / q~

for a l l relatively prime integers p and q .)

(e) A Herma~ ring D is a stable region similar to a Siegel disk.

Now we have a periodic cycle of a n n u l i and a power of f which
restricted to a n y of t h e s e annuli is a n a l y t i c a l l y equivalent to an
 119

irrational rotation of the standard annulus.

(For a p p r o p r i a t e 8 and a M. Herman [10] f o u n d such regions for

the m a p :

eie ( x-a ) 2
x i ~ )
x I-~x

The f a t e of c r i t i c a l points {c : f' (c) = O} is c r u c i a l in c o n n e c -

tion with theorem (1.4).

THEOREM 1.5. (Fatou)

(a) If D is an a t t r a c t i v e or p a r a b o l i c basin then D contains

at l e a s t o n e critical point of f .

(b) If D is a S i e g e l disk or Herman ring then ~D is c o n t a i n e d

in t h e e-limit sets of c r i t i c a l points.

Thus f can have only finitely many cyclic stable regions. But it
is s t i l l an o p e n p r o b l e m whether 2d-2 (d > 2 the d e g r e e of f) is
a sharp upper bound. Another open problem is w h e t h e r a Siegel disk al-
ways has a critical point on its b o u n d a r y . M. Herman [11] in a r e -
markable paper proved this conjecture recently for f(z) = z2 + I .
Note that theorem 1.5 a n d theorem 1.4 p r o v i d e an e x c e l l e n t basis for
computer experiments. For the detection and characterization of all
cyclic stable regions of a m a p f one simply has to f o l l o w the forward
orbits of a l l c r i t i c a l points. The following example illustrates the
strength of t h e s e results:

x-2 >2
EXAMPLE 1.6. f(x) = ~ , J =

The critical points are: 2,O . Observe that 2 ~ O ~ co ~ 1 ~ I

and f' (I) = - 4 Thus C ~ J = ~ , because none of the cases
(a), (b) in t h e o r e m 1.5 is p o s s i b l e .
 120

THE CRITICAL POINTS OF THE RENORMALIZATION MAP R AND A

q
MORPHOLOGY OF N

Our map
Rq(X) = < x222+xq+-q1- 2 )

h a s the six c r i t i c a l points:

I , ~ , 1-q , ± ~ , (2-q)/2 .

Since I and ~ are attractive fixed points and since (2-q)/2 ~ ~ ,

± V 1-q ~ O it s u f f i c e s to e x a m i n e the o r b i t s of 1-q and O only.
We do this in the s p i r i t of B. Mandelbrot'shistorymakingexperiment:
Let

A I :: {q6{ : Rn
q(1-q) ~ I , n ~}

(17) A := {q6{ : Rn(1-q) ~ ~ , n ~ ~}

q

M R := • ~ (A I u A )

Figures 3,4 and 5 show AI ,A and MR . Figure 6 shows a b l o w up of

a detail of figure 5. S u r p r i s i n g l y it d i s p l a y s a structure which looks
like a copy of the o r i g i n a l Mandelbrot set [13]. I.e. it is e x a c t l y si-
milar to the b i f u r c a t i o n s e t of the q u a d r a t i c family x~x2+c , c6~
It is o b v i o u s that any q such that lql > > I is in A , thus AI
and MR are bounded. Experimentally it t u r n e d out that the f a t e of
the t w o crucial orbits of (l-q) and 0 were related, i.e. whenever

R n(1-q) ~ I then R n(O) ~ ~ , as n ~

q q
(18)
Rn(1-q) ~ ~ then Rn(O) ~ I , as n ~
q q

Indeed, this is a n i m m e d i a t e consequence of the c o m m u t a t i v e diagram

(19)
R

q I ~q

S o S
q q
 q

b~ I
I J

yl rl
>

I L
I

7"

J J
 122

2.5

-2.5
-I
Figure 5. MR in b l a c k

] .70

I .54
0.2 0.36

Figure 6. Detail in M R
D e t a i l of MR ( s e e f i g u r e 6) i n b l a c k s u r r o u n d e d by A I i n y e l l o w and
A~ i n g r e e n .

The M a n d e l b r o t s e t M i n b l a c k t o g e t h e r w i t h i t s e l e c t r o s t a t i c poten-
t i a l g i v e n by t h e Douady-Hubbard c o n f o r m a l homeomorphism ~'.D -~ ~'-M.
 123

I .70

I .54
0.2 0.36

Ato in b l a c k

I .70

1.54
0.2 0.36

Figure 6. (continued) Detail of MR A I and MR in b l a c k

 124

where

x+q-1
~q(X) ........x_ I
(20) and
Sq(X) = x2+q-1
x2-1
This means that

Rq(X) = (~q o Sq o ~q) 2 (x) = D 2 (x) ,

(21)
with (x+ i)2
Dq(X) = x-1

Thus, Mq exchanges the hot phase (x=1) with the cold phase
(x=~) and the two crucial critical orbits of (l-q) and O .
Figures 3-6 are e x p l a i n e d and d e s c r i b e d in greater detail in [15].
In p a r t i c u l a r p r o b l e m 1.2 (c) is answered. R o u g h l y s p e a k i n g the main
body of MR and each of its buds as well as the main body of the
detail in figure 6 and each of its buds identify p a r a m e t e r s q for
which there is a p e r i o d i c attractor. Their basins of a t t r a c t i o n estab-
lish a third magnetic phase and the b o un d a r y of these basins, which
is the J u l i a set of R , b e i n g also the b o u n d a r y of A (I) and
q q
A (~) , is a c a n d i d a t e for a formal locus of p h a s e transitions. Note,
q
however, that even though N m a y be given by Jq , the J u l i a set
of Rq , its points may not be s i n g u l a r i t i e s of the free energy f in
the t h e r m o d y n a m i c limit. This seems to c o n t r a d i c t (5), but note that
in the t h e r m o d y n a m i c limit the free energy may simply allow an analytic
continuation.
In summary our e x p e r i m e n t s leed to the f o l l o w i n g interesting conjec-
tures:

CONJECTURE 1.7.

(I) MR is connected.

(2) The subset of MR shown in figure 6 is h o m e o m o r p h i c

(quasi conformally) to the M a n d e l b r o t set M , where
 125

M = {c6~ : fn
c (0) / , ~ , as n ~ ~}

fc(X) = x2 + c .

(3) N = J for a n y q 6 ( ~ ~ Mq) U Mq

q

Note that according to [5] t h e M a n d e l b r o t set M is c o n n e c t e d .

Actually, Douady and Hubbard showed that • ~ M and ~ ~ D ,
D = {x6~ : Ixl < I} , are homeomorphic subject to a conformal mapping.
Sullivan [21] g a v e an alternative proof which may apply also to o u r
case. To indicate the idea we briefly survey another remarkable result
of J. C u r r y , L. Garnett a n d D. Sullivan [3]:

NEWTON'S METHOD AND THE MANDELBROT SET

Consider the one-parameter family of r a t i o n a l maps

gl(x) = x - pl(x)/p~(x) , where

(22)
pX(x) = x3 + (l-1)x - I .

Note that Newton's method for a n y c u b i c is e q u i v a l e n t by a l i n e a r

change of v a r i a b l e s to at l e a s t one of the gl 's. T h e 4 critical
points of gl are the 3 roots of pl and the d i s t i n g u i s h e d point 0 ,
which in v i e w of t h e o r e m 1.5 is the o n l y non-trivial critical point.
The black regions in the c o m p l e x X-plane in f i g u r e s 7, 8 and 9 were
determined b y the condition of the forward orbit of 0 converging to
the r o o t I of pl(x) . Let

(23) Mg {16~ : g~(O) / , r o o t of P l ' as n ~ ~}

Then Sullivan [21] a r g u e s that the components in ~ ~ M c o r r e s p o n d to

g
quasi-conformal conjugacy classes which are analytically just punctu-
red disks. Hence, M is c o n n e c t e d . T h e s u b s e t of M s h o w n in fi-
g g
gure 9 is a c t u a l l y h o m e o m o r p h i c to the M a n d e l b r o t set M , as A. D o u -
a d y a n d J. H. H u b b a r d s h o w in [6] . Arguments similar to t h o s e in
 126

2.0

-2.0
1.75
-2.25

Figure 7. {l 6 ~ : g~(O) ~ I , n ~ ~} = black

 127

2.0

1.4

-0.20 0.4

Figure 8. (a) Detail of figure 7.

? i i

,D

t9

(b) {I 6 C : g (O) i-, - ~ - n ~ ~}

U{l 6 C : g~(O) ~ root of Pl ' n ~ co}

 128

1 . 70

I .54
0.20 0.36

Figure 9. (a) D e t a i l of f i g u r e 8a.

(b) The M a n d e l b r o t - l i k e set in

{~ E C : g~(O) 7~ r o o t of Pl f n
 129

[6] a n d [21] s h o u l d suffice to e s t a b l i s h conjecture 1.7 (I), (2).

We add in p a s s i n g that figure 9 gives some insight into a completely

different set of q u e s t i o n s : Given a polynomial, describe the set of
initial values in ~ for w h i c h Newton's method converges towards a
root. It is k n o w n , that for a p o l y n o m i a l with real coefficients and
real roots this set is ~ except for a set of Lebesgue measure zero
(see [1,20]). Now figure 9 teaches us that this remarkable result does
not extend to { , because for a n y l in the M a n d e l b r o t - l i k e set (see
figure 9) N e w t o n ' s method allows a periodic attractor with an o p e n b a -
sin of a t t r a c t i o n .

Conjecture 1.7(3) is m e a n t to c o n t r i b u t e to p r o b l e m 1.2 (a) a n d (b).

Note that if o n e k n e w that

(24) ~R 2 hyp (M R ) , ( : hyperbolic part of M R )

i.e. for any q E MR the orbit of (l-q) converges towards a periodic

attractor of Rq , then conjecture 1.7 (3) c o u l d be e s t a b l i s h e d from
classical theory. Note, however, that an i d e n t i t y corresponding to
(24) is n o t even known for the m u c h more fundamental Mandelbrot set M.
On the o t h e r hand it is k n o w n that if M were locally connected then
o

M = hyp (M) (see [7]). For a good visual impression of t h e d i f f i c u l -

ties with regard to t h e last questions we refer t o the p i c t u r e s and
experiments in [16] .

SOME JULIA SETS FOR R

q

Finally we discuss some Julia sets of R for the p h y s i c a l l y m e a -

q
ningful choices q = 2,3,4; see figure 10. F i r s t l y , o n e s h o w s t h a t

2 C AI , 3 6 A1 , 4 E A

Furthermore, for q = 4 o n e has that A*(1) = Aq(1) and

q
A~(~) = Aq(~) , where A* denotes the i m m e d i a t e b a s i n of a t t r a c t i o n ,
i.e. the c o m p o n e n t which contains the a t t r a c t o r . Hence, it f o l l o w s from
[2] that the Julia set Jq , q = 4 , is a J o r d a n curve, which, due
to the symmetry with respect to c o n j u g a t i o n , must intersect ~+ in a
unique point x c , the ferromagnetic transition point.
 I I
U~

! -I
I

e:

C~
~Q

0 C~
o

.,'a

I-'-
M

L~

0
II il

~Q
°
L¢I
 I

L i

II II

0 0

0 ~D
 I I

It II

0
 133

Remarkably, also the Julia sets for q < 4 in figure 10 d i s t i n g u i s h

a unique phase transition on ~+

A c k n o w l e d g e m e n t : The color plates were o b t a i n e d by D. Saupe and the

authors on an AED 767 while figures 3-10 were o b t a i n e d by H.W. Ramke
and the authors on a laser printer. All pictures were p r o d u c e d in our
"Graphiklabor Dynamische Systeme - U n i v e r s i t ~ t Bremen".

REFERENCES

I. B. Barna: Uber die D i v e r g e n z p u n k t e des N e w t o n s c h e n V e r f a h r e n s zur

B e s t i m m u n g von W u r z e l n A l g e b r a i s c h e r Gleichungen. II, P u b l i c a t i o n e
Mathematicae, Debrecen, 4, 384-397 (1956).

2. H. Brolin: Invariant sets under iteration of rational functions,

A r k i v f~r Math.., 6, 103-144 (1966).

3. J. Curry, L. Garnett, D. Sullivan: On the iteration of rational

functions: C o m p u t e r e x p e r i m e n t s w i t h N e w t o n ' s method, Commun. Math.
Phys., 9! , 267-277 (1983).

4. B. Derrida, L. De Seze, C. Itzykson: Fractal structure of zeroes

in hierarchical models, J. Statist. Phys. 3_~3, 559 (1983).

5. A. Douady, J. H. Hubbard: Iteration des p o l y n o m e s q u a d r a t i q u e s

complexes, CRAS Paris, 294, 123-126 (1982).

6. A. Douady, J. H. Hubbard: On the dynamics of p o l y n o m i a l - like

mappings, preprint, 1984.

7. A. Douady, J. H. Hubbard: Etude d y n a m i q u e des p o l y n o m e s complexes,

P u b l i c a t i o n s M a t h e m a t i q u e s D'Orsay, 1984.

8. P. Fatou: Sur les @quations f o n c t i o n n e l l e s , B u l l . Soc. Math. Fr.,

417, 161-271 (1919), 48, 33-94, 208-314 (1920).

9. S. Grossmann: A n a l y t i c P r o p e r t i e s of T h e r m o d y n a m i c F u n c t i o n s and
Phase Transitions, in: F e s t k ~ r p e r p r o b l e m e IX, Ed. O. M a d e l u n g
V i e w e g 1969.

10. M. Herman: E x a m p l e s de fractions r a t i o n n e l l e s ayant une o r b i t e

dense sur la sphere de Riemann, to appear in Bull. Soc. Math. Fr.

11. M. Herman: Are there c r i t i c a l points on the b o u n d a r i e s of singular

d o m a i n s ? Report 14, I n s t i t u t M i t t a g - L e f f l e r , 1984.

12. G. Julia: Sur l ' i t e r a t i o n des fonctions r a t i o n n e l l e s p J o u r n a l de

Math. Pure et Appl., 8, 47-245 (1918).

13. B. Mandelbrot: Fractal aspects of the iteration of z~lz(1-z) for

complex I , z, A n n a l s N.Y. Acad. Sciences, 357, 249-259 (1980).
 134

14. R. Mane, P. Sad, D. Sullivan: On the dynamics of rational maps,

Ann. scient. Ec. Norm. sup., 16, 193-217 (1983).

15. H.-O. Peitgen, M. Prefer, P. H. Richter: Phase transitions and

Julia sets, Report 118, FS "Dynamische Systeme", Universit~t
Bremen, Mai 1984, Proceedings 5th Meeting of the UNESCO Working
Group on Systems Theory 'Lotka - Volterra Approach in Dynamic
Systems', Wartburg, Eisenach, 1984.

16. H.-O. Peitgen, P. H. Richter: Die unendliche Reise, GEO June

1984, pp. 100-124, Gruner + Jahr, Hamburg.

17. H.-O. Peitgen, D. Saupe, F. v. Haeseler: Cayley's problem and

Julia sets, The Mathematical Intelligencer, 6, (2), 11-20 (1984).

18. D° Ruelle: Thermodynamic Formalism, Encyclopedia of Mathematics

and its Applications, Volume 5, 1978, Addison-Wesley Publishing
Company, Reading, Massachusetts.

19. C. L. Siegel: Iteration of analytic functions, Ann. Math., 4~3,

607-612 (1942).

20. S. Smale: The fundamental theorem of algebra and complexity

theory, Bull. Amer. Math. Soc., 4, 1-36 (1981).

21. D. Sullivan: Quasi conformal homeomorphisms and dynamics I, II,

III, preprint 1982-1983.

22. C. N. Yang, T. D. Lee: Statistical theory of equations of state

and phase transitions.
I. Theory of Condensation,Phys. Rev. 87, 404-409 (1952),
II. Lattice Gas and Ising Model~Phys. Re---v. 8_/7, 410-419 (1952).
 RECENT DEVELOPMENTS IN R E P R E S E N T A T I O N THEORY

Wilfried 5chmid*
D e p a r t m e n t of M a t h e m a t i c s
Harvard University
Cambridge, MA 02138

For the purposes of this lecture, "representation theory" means

representation theory of Lie groups, and m o r e specifically, of s e m i s i m p l e
Lie groups. I am i n t e r p r e t i n g my a s s i g n m e n t to give a survey rather
loosely: while I shall touch upon various major advances in the s u b j e c t ,
I am c o n c e n t r a t i n g on a s i n g l e development. Both order and e m p h a s i s of
my p r e s e n t a t i o n are m o t i v a t e d by e x p o s i t o r y considerations, and do not
reflect my view of t h e relative importance of v a r i o u s topics.

Initially G shall denote a locally compact topological group which

is u n i m o d u l a r -- i.e., left and right Haar measure coincide -- and H C G
a closed unimodular subgroup. The quotient space G/H then carries a G-
invariant measure, so G acts unitarily on the Hilbert space L2(G/H).
In essence, the fundamental problem of h a r m o n i c analysis is to d e c o m p o s e
L2(G/H) into a direct "sum" of irreducibles. The quotation marks allude
to the fact that the decomposition typically involves the c o n t i n u o u s ana-
logue of a sum, namely a direct integral, as h a p p e n s already for non-
compact Abelian groups. If G is of t y p e I -- loosely speaking, if t h e
unitary representations of G behave reasonably -- the a b s t r a c t Plan-
cherel theorem [12] asserts the e x i s t e n c e of such a decomposition. This
existence theorem raises as m a n y questions as it a n s w e r s : to make the
decomposition useful, one w a n t s to know it e x p l i c i t l y and, most impor-
tantly, one w a n t s to u n d e r s t a n d the structure of the irreducible sum-
mands. In p r i n c i p l e , any irreducible unitary representation of G can
occur as a c o n s t i t u e n t of L2(G/H), for some H C G. The Plancherel
problem thus leads naturally to the study of the irreducible unitary
representations.

To what extent these problems can be s o l v e d depends on o n e ' s know-

ledge of the structure of the group G and on the nature of the s u b g r o u p
H. Lie groups, p-adic groups, and algebraic groups over finite fields
constitute the most interesting and b e s t understood large classes of

*Supported in part b y NSF grant DMS 831743%.

 136

groups. Although the formal similarities are b o t h striking and instruc-

tive, the technical aspects of t h e representation theory for these three
classes diverge -- hence the limitation to the c a s e of L i e groups.

5emisimple groups play a distinguished role among all Lie groups,

since they come up f r e q u e n t l y in p h y s i c a l , geometric, and number-theore-
tic problems. The special emphasis on semisimple groups can also be
justified on other grounds: one of t h e aims of M a c k e y ' s theory of induced
representations is to r e d u c e the h a r m o n i c analysis on general Lie groups
to that on s e m i s i m p l e groups; recently Duflo [13] has w o r k e d out the re-
duction step quite concretely, at least for algebraic groups of t y p e I.

From the point of view of h a r m o n i c analysis, irreducible unitary

representations are the m a i n objects of interest. Nevertheless, there
are important reasons for b e i n g less restrictive: non-unitary r e p r e s e n -
tations not only occur naturally in t h e i r own right, for example as
solution spaces of linear differential equations invariant under the
action of a s e m i s i m p l e group, but t h e y arise even in t h e context of u n i -
tary representations -- a hint of t h i s phenomenon will become visible
below. Once one leaves the class of u n i t a r y representations, one should
not insist on irreducibility; various known constructions produce irre-
ducible representations not directly, but as q u o t i e n t s or s u b s p a c e s of
certain larger representations.

After these preliminaries, I let G denote a semisimple Lie group,

connected, with finite center, and K a maximal compact subgroup of G.
The choice of K does not matter, since a n y two m a x i m a l compact sub-
groups are conjugate. By a r e p r e s e n t a t i o n of G, I shall mean a conti-
nuous representation on a complete, locally convex Hausdorff space, of
finite length - every chain of closed, G-invariant subspaces breaks
off after finitely many steps - and "admissible" in the sense of Harish-
Chandra: any irreducible K-module occurs only finitely often when the
representation is r e s t r i c t e d to K. This latter assumption is automati-
cally satisfied by unitary representations, and c o n s e q u e n t l y G is of
type I [19]. No e x a m p l e s are known of Banach representations, of f i n i t e
length, which fail to be admissible.

To s t u d y finite dimesional representations of G, one routinely

passes to the a s s o c i a t e d infinitesimal representations of the Lie alge-
bra. Infinite dimensional representations are generally not differen-
tiable in t h e naive sense, so the notion of infinitesimal representation
requires some care. A vector v in the representation space Vw of a
representation ~ is said to be K-finite if its K-translates span a
finite dimensional subspace. By d e f i n i t i o n , v e V~ is a d i f f e r e n t i a b l e
 137

vector if the assignment g --+ ~ ( g ) v maps G into V~ in a C" fashion.

Differentiable and K-finite vectors can be c o n s t r u c t e d readily, by aver-
aging the translates of a r b i t r a r y vectors against compactly supported C®
or K-finite functions [17]. One may conclude that the K-finite vectors
make up a d e n s e subspace V C Vw , which consist entirely of d i f f e r e n t i -
able vectors - at t h i s point the standing assumption of a d m i s s i b i l i t y
plays a crucial role. In particular, the complexified Lie algebra g of
G acts on V by d i f f e r e n t i a t i o n . The subgroup K also acts, by t r a n s -
lation, but G does not. Partly for trivial reasons, and p a r t l y as c o n -
sequence of the original hypotheses on the representation ~, the g- and
K-module V satisfies the following conditions:

a) as K-module, V is a d i r e c t sum of f i n i t e d i m e n s i o n a l
irreducibles, each o c c u r i n g o n l y f i n i t e l y often;
(I) b) the actions of g and K are compatible;
c) V is f i n i t e l y generated over the universal enveloping
algebra U(g).

Here b) simply means that the g-action, restricted to the complexified

Lie algebra k of K, coincides with the derivative of the K-action.
This definition of the infinitesimal representation, which was introduced
by Harish-Chandra [19], has the very desirable feature of associating
algebraically irreducible g-modules to t o p o l o g i c a l l y irreducible repre-
sentations of G; by c o n t r a s t , g acts in a h i g h l y reducible fashion on
the s p a c e s of all differentiable or a n a l y t i c vectors of an infinite di-
mensional representation ~.

A simultaneous g- and K-module V with the properties (la-c) is

called a Harish-Chandra module. All Harish-Chandra modules can be lifted
to representations of G [10,41], not uniquely, but the range of pos-
sible topologies is now w e l l - u n d e r s t o o d [45,54]. If V arises from an
irreducible unitary representation ~, it i n h e r i t s an inner product which
makes the action of g skew-hermitian. An irreducible Harish-Chandra
module admits at m o s t one such inner product, up to a p o s i t i v e factor; if
it does, the c o m p l e t i o n becomes the representation space of a u n i t a r y
representation of G [19,39]. In other words, there is a o n e - t o - o n e
correspondence between irreducible unitary representations of G and
Harish-Chandra modules which carry an inner product of the appropriate
type. The problem of describing the irreducible unitary representations
thus separates naturally into two s u b - p r o b l e m s : the description of all
irreducible Harish-Chandra modules, and secondly, the determination of
those which are "unitarizable'. Of t h e two, the latter seems conside-
rably more difficult, and has not yet been solved, except in s p e c i a l
cases -- more on t h i s below.
 138

The irreducible Harish-Chandra modules of a general semisimple Lie

group were classified by Langlands [33] and Vogan [48]; one of the ingre-
dients of Langlands' c l a s s i f i c a t i o n is due to Knapp-Zuckerman [31]. To
describe the classification in geometric terms, I introduce the flag
variety of g,

(2) X = set of Borel subalgebras of q

It is a complex projective variety and a homogeneous space for the com-

plex Lie group

(3) GC = identity component of Aut(g) .

In t h e case of the prototypical example G = 51(n,R), X can be identi-

fied with the variety of all "flags" in C n, i.e. chains of subspaces
0 C V1 C V2 C . .. C V n = C n, with dim Vk = k : every Hotel subalgebra of
the complexified Lie algebra g = sl(n,C) stabilizes a unique flag
(Vn}. The group G acts on the flag variety via the adjoin% homomor-
phism. There are finitely many G-orbits -- for G = 51(n,R), for
example, these are characterized by the position of flags relative to the
real structure R n C C n. Now let D C X be a G-orbit, and L --+ D a
homogeneous line bundle - a line bundle with a G-action compatible
with that on the base D. A cohomological construction, which I shall
describe next, if only in rough outline, associates a family of Harish-
Chandra modules to the pair (D,L).

At one extreme, if G contains a compact Car%an s u b g r o u p I, and if

D is an open orbit, the homogeneous line bundles over D are paramet-
rized by a lattice. As an open subset of X , D has the structure of
complex manifold. Every homogeneous line bundle L --~ D can be turned
into a holomorphic line bundle, so that G acts as a group of holomor-
phic bundle maps. Thus G acts also on the sheaf cohomology groups of
L. The differentiated action of the Lie algebra g turns

(4) H*(D,O(L))(K ) = { ~ e H~(D,O(L)) J ~ is K-finite } ,

into Harish-Chandra modules. Whenever the line bundle L is negative in

the appropriate sense -- for example, if L extends to a line bundle over
the projective variety X whose inverse is ample -- the cohomology ap-
pears in only one degree and is irreducible:

HP(D,O(L)) = 0 if p ~ s ,
(5)
HS(D,O(L))(K) is a non-zero, irreducible Barish-Chandca module

[43]; here s denotes the largest dimension of compact subvarieties of

lequivalently, a torus which is a maximal Abelian subgroup.

 139

D. These modules can be m a p p e d (g,K)-equivariantly into L2(G) H C ' ( G ) ,

and are consequently unitarizable. The unitary structure is v i s i b l e also
in t e r m s of the geometric realization: the L2-cohomology of L injects
into the D o l b e a u l t cohomology, its image is dense, has a natural Hilbert
space structure, and contains all K-finite cohomology classes [3,44].
The resulting unitary representations make up the discrete series, which
was originally constructed by Harish-Chandra via c h a r a c t e r theory [21].

The opposite extreme, of a t o t a l l y real G-orbit D C X, occurs only

when g contains Borel subalgebras defined over R, as is the case for
G = Sl(n,R). In t h i s situation D is n e c e s s a r i l y compact, and the c o h o -
mological construction collapses to t h a t of the single Harish-Chandra
module

(6) C'(D,L)(K ) = space of K-finite, C" sections of L •

The m o d u l e (5) need not be irreducible, but it has a unique irreducible

quotient, provided L satisfies a suitable negativity condition. Harish-
Chandra modules of t h i s type are induced from a Borel subgroup of G ;
they belong to the principal series.

The construction for a general G-orbit D combines elements of

"complex induction , as in (4), and ordinary induction, as in (6). It
can be v i e w e d as a c o h o m o l o g i c a l form of g e o m e t r i c quantization. There
is c o m p l e t e l y parallel, algebraic version of the construction, due to
Zuckerman, which offers certain technical advantages. It is t h i s version
that has been studied and used extensively [49]. Subject to c e r t a i n
hypotheses on the pair (L,D) , Zuckerman's "derived functor construc-
tion" -- equivalently, the geometric construction -- produces cohomology
in o n l y one degree, a Harish-Chandra module that arises also by induction
from a discrete series module of a s u b g r o u p of G. Under more stringent
assumptions, the Harish-Chandra module corresponding to (L,D) has a
unique irreducible quotient [38]. Every irreducible Harish-Chandra mo-
dule can be realized as s u c h a quotient in a d i s t i n g u i s h e d manner --
this, in effect, is L a n g l a n d s ' classification [33]. The problem of
understanding the irreducible Harish-Chandra modules does not end here,
however. The irreducible quotient may be all of the original module, or
m a y be m u c h smaller. In p r i n c i p l e , the Kazhdan-Lusztig conjectures for
Harish-Chandra modules, proved by Vogan [51] in the generic case, provide
this type of information, but not as e x p l i c i t l y or c o n c r e t e l y as one
might wish.

I now t u r n to a different, more recent construction of Harish-

Chandra modules, that of B e i l i n s o n - B e r n s t e i n [6]; similar ideas, in t h e
context of V e r m a modules, can be found also in the w o r k of B r y l i n s k i and
 140

Kashiwara [9]. 5ome preliminary remarks are necessary. The flag variety
X may be thought of as a quotient GC/B; here B is a particular Borel
subgroup of G C, i.e~ the normalizer of a Borel subalgebra b C g. The
differentials of the algebraic characters of B constitute a lattice A.
Each x ¢ A -- or more precisely, the corresponding character ex of B
-- associates a Gc-homogeneous, holomorphic line bundle Lx --~ X to the
principal bundle B --~ G C --+ X. Its cohomology groups are finite dimen-
sional Gc-modules, which are described by the 8orel-Weil-Bott theorem
[8]. In p a r t i c u l a r ,

H*(X,O(Lx)) vanishes except in one degree p = p(x) ,

(7)
HP(x,O(Lx)) is irreducible •

The center Z(g) of the universal enveloping algebra U(q) acts on the
cohomology by scalars (Schur's lemma! ), so

(8) Ix = annihilator of B*(X,O(LI)) in Z(g)

is a m a x i m a l ideal in Z(g). As a result of its construction, the line

bundle Lx carries an algebraic structure, and it m a k e s sense to define

Dx = sheaf of linear differential operators, with

(9)
algebraic coefficients, acting on the sections of Lx ;

the notion of sheaf is taken with respect to the Zariski topology, as

befits the algebraic setting. To picture D x, one should n o t e t h a t it
is locally isomorphic to the sheaf of scalar differential operators on
X. The Lie algebra g operates on sections of Lx by infinitesimal
translation. This operation extends to a homomorphism from U(g) into
FD, ( = algebra of global sections of Dx ), w h i c h in turn drops to an
isomorphism

(10) FDx s Ox =def O(g)/IxU(g)

[6]. This is the point of departure of the Beilinson-Bernstein construc-

tion.

The passage from U(g) to t h e sheaf Dx has a counterpart on t h e

level of U(g)-modules: a pair of functors between

(Ii) M ( U x) = category of Dx-modules

-- equivalently, the category of U(g)-modules on which the center Z(g)
acts as it does on the cohomology groups (7) - and

(12) M ( D x) = category of quasi-coherent sheaves of Dx-modules •

Quasi-coherence means simply that the sheaves admit local presentations

in t e r m s of generators and relations, though not necessarily finite pre-
sentations. In one direction, the global section functor
 141

(13) F : M ( D x) --+ M U x)

maps sheaves of Dx-modules to modules over F D x s U x, Extension of

scalars from the algebra of global sections Ux to the stalks of Dx
determines a functor in t h e opposite directlon,

A : M(U x ) ) M(D x ) ,
(14)
AV = D,@ U V ;
%

the sheaves AV are quasi-coherent because every V e M ( U x) can be

described by generators and relations.

Those parameters x e A which correspond to ample line bundles Lx

span an open cone C C R@zA . One calls x dominant if it lies in the
closure of C, dominant nonsingular if x lies in C itself. The
inverse of the canonical bundle is ample, and is t h e r e f o r e indexed by a
particular dominant nonsingular quantity, customarily denoted by 2p.
With these conventions it possible to state the following remarkable ana-
logue of Cartan's theorems A and B:

(15) Theorem (Beilinson-Bernstein [6]) A) The global sections of any

quasi-coherent sheaf of D~-modules generate its stalks, provided x + p
is dominant and nonsingular. B) If ~ + p is dominant, the sheaf coho-
mology groups HP(X,V) vanish, for every V e M ( D z) and p > 0.

As a direct consequence, Beilinson-Bernstein deduce:

(16) Corollary In t h e situation of a dominant nonsingular x + p , the

functor F defines an equivalence of categories M(Ux) ~ M ( D x) , with
inverse A.

Perhaps surprisingly, the equivalence of categories implies properties of

general Ux-modules that were previously unknown. The most fruitful ap-
plications, however, occur in the context of certain smaller categories,
in particular the category O of Bernstein-Gelfand-Gelfand [7] and the
category of Harish-Chandra modules.

Irreducible modules in e i t h e r of these categories are annihilated by

maximal ideals in Z(g), but not every maximal ideal is of the form (8),
with x e A . According to Harish-Chandra [20], the correspondence
z --+ I x extends naturally to a sur]ective map from the vector space
C@zA onto the set of all maximal ideals; any two of the ideals Ix, I~,
for x, ~ e C@zA , coincide precisely when x + p and ~ + p belong to
the same orbit of the Weyl group W -- a finite group which acts line-
 142

arly on C@2A . I want to emphasize one consequence of Harish-Chandra's

description of t h e maximal ideal space:

every maximal ideal in Z(g) can be realized as I x , with

(17)
x ~ C@zA having the property that Re(x + p) is dominant.

Although the bundle L, ceases to exist as soon as t h e parameter x

leaves the lattice A , there are "phantom line bundles" attached to all
x e C@ZA , locally defined line bundles to which the action of g on X
lifts. The construction (9) of the sheaf of algebras Dx continues to
make sense in t h i s wider setting, as do t h e isomorpism (I0), the catego-
ries M(Dx), M(Ux), and the functors F, A Most importantly, theorem
(15) and its corollary remain valid [6], with one minor adjustment: the
phrase "x + p is d o m i n a n t " should be replaced by "Re(x + p) is
dominant'. Different values of x may correspond to identical maximal
ideals Ix and quotients U x = U ( g ) / I x , but an appropriate choice of x
will bring any such quotient within the scope of p a r t B of the theorem --
this follows from (17). The same x makes part A and the corollary
apply at least generically, for parameters outside a finite number of
hyperplanes. The equivalence of categories breaks down in t h e remaining,
singular cases only because certain sheaves fail to have global sections.

The maximal compact subgroup K C G possesses a complexification, a

complex algebraic group KC , defined over R , which contains K as the
group of real points. If V ~ M(U x) is a Harish-Chandra module, the K-
action induces an algebraic Kc-action on t h e sheaf AV . The differen-
tial of this action agrees with the multiplication action of k, viewed
as Lie subalgebra of FD x s U x -- in short, KC and k act compatibly.
For the purpose of t h e preceeding discussion, the finiteness condition
(la) in the definition of Harish-Chandra module becomes irrelevant. It
is n e c e s s a r y only that K act locally finitely, i.e. the K-translates
of any v e V must span a finite dimensional subspace. The passage from
locally finite K-actions on Ux-modules to algebraic Kc-actions on
sheaves of Dx-modules can be reversed; in other words, both F and
restrict to functors between

M(Ux,K) = category of Ux-modules with a compatible,

locally finite K-action , and
(18)
M ( D , , K C) = category of sheaves of quasi-coherent D,-modules
with a compatible, algebraic Kc-action

Whenever x + p is n o n s i n g u l a r and Re(x + p) is d o m i n a n t , the equi-

valence of categories (16) identifies these two subcategories,
 143

(19) F : M ( D ~ , K c) ~ M(Ux,K) •

A theorem of Harish-Chandra asserts, in e f f e c t , that finitely generated

modules in t h e category M(Ux,K) satisfy the finiteness condition (la)
[19], hence

the irreducible objects in t h e categories M(Ux,K) , x e CSzA ,

(20)
exhaust the class of irreducible Harish-Cbandra modules

In particular, the identification (19) relates irreducible Harish-Chandra

modules to irreducible sheaves V e M(Dx,Kc).

Geometric considerations suggest how to find such sheaves. The

support of any V e M ( D x , K C) is invariant under the translation action
of KC on X , via the adjoint homomorphism. If V is irreducible, the
support must be an irreducible variety -- n e c e s s a r i l y the closure of an
orbit, since

(21) KC acts on X with finitely many orbits

[36]. Now let Y C X be a Kc-orbit, Y its closure. The operation of

pushforward yields irreducible sheaves with support in Y , as I shall
explain next.

Ordinarily the D-module pushforward of a sheaf exists only as an

object of the derived category. In t h e situation at hand it can be des-
cribed quite explicitely. The analogy with the C" case is instructive.
Linear differential operators on a C" manifold M cannot be applied
naturally to the C~ functions on a closed submanifold N C M . How-
ever, after the choice of a smooth measure, functions in C'(N) may be
viewed as distributions on M , and the sheaf of differential operators
DM -- here in the C® sense - does act on these. The D-module push-
forward of the sheaf of smooth measures on N is t h e sheaf generated by
that action; in other words, the sheaf of distributions on M , with
support in N , which are smooth along N . Formally, measures and dis-
tributions must be t r e a t e d as sections not of the trivial bundle, but the
top exterior power of the cotangent bundle. For this reason the pushfor-
ward from N to M involves a twist by the quotient of the two deter-
minant bundles, i.e., a twist by the top exterior power of the conormal
bundle. The preceeding discussion can be expressed in t e r m s of local co-
ordinates, and then makes sense equally in t h e algebraic setting.

Back to the Kc-orbit Y C X ! Under an appropriate integrality

condition on t h e parameter x , the "phantom line bundle" corresponding
to , extends to the orbit as a true Kc-equivariant line bundle, pos-
sibly in s e v e r a l different ways. I let Ly, x denote a particular such
extension, tensored by the top exterior power of the normal bundle. Its
 144

sheaf of s e c t i o n s Oy(Ly, x) is a m o d u l e for a twisted sheaf of d i f f e r e n -

tial operators Dy, x on Y . The complement of t h e boundary ~Y in X
contains Y as a smooth, closed subvariety. Because of the shift built
into the definition, the D-module pushforward of Oy(Ly, x) from Y to
X - aY is a sheaf of m o d u l e s over the sheaf D x , restricted to X - ~Y.
It b e c o m e s a sheaf of Dx-modules over all of X when pushed forward
once more -- naively, simply as a s h e a f -- from the open subset X - ~Y
to X . The r e s u l t i n g sheaf, which I denote by Vy, x , b e l o n g s to the
category M(Dx,Kc) , since KC operates at each step of its c o n s t r u c t i o n .
A basic result of K a s h i w a r a , on s h e a v e s of D-modules supported by s m o o t h
subvarieties, implies

a) the s h e a f of Dx-modules Vy, ~ has a finite composition

series and contains a unique irreducible subsheaf ;
(22)
b) e v e r y irreducible sheaf in the category M(Dx, K C) arises in
this manner, for some Kc-orbit Y and line bundle Ly, x

Under the hypotheses of the equivalence of c a t e g o r i e s , this statement

translates immediately into a classification of the irreducible Harish-
Chandra modules which are a n n i h i l a t e d b y the m a x i m a l ideal I x C Z(g) :

FVy, x has a unique irreducible submodule; the assignment

of that module to the d a t u m of the orbit Y and line bundle
(23)
Ly, x establishes a bi3ection between such pairs ( Y, Ly, x)
and irreducible Harish-Chandra modules in M(Ux, K)

When x + p is s i n g u l a r , the situation becomes more complicated, as it

does also from the p o i n t of view of the L a n g l a n d s classification. Irre-
ducible Harish-Chandra modules in M(Ux, K) can still be r e a l i z e d as sub-
modules of some VVy, x , but not always as an o n l y irreducible submodule,
nor in a u n i q u e manner.

The reducibility or irreduciblity of Vy, x , in the category of

sheaves of D x - m o d u l e s , is a local phenomenon. All stalks at p o i n t s of
the c o m p l e m e n t of ? vanish, and a s m a l l calculation s h o w s t h o s e over
points of Y to be a u t o m a t i c a l l y irreducible. If a n o n - t r i v i a l quotient
of Vy, x exists, it a l s o b e l o n g s to t h e category M ( D x , K C) and has sup-
port in the b o u n d a r y . In particular, the sheaf Vy, x cannot possibly
reduce unless there is a n o n - e m p t y boundary: subject to the usual p o s i -
tivity condition on x ,

the Harish-Chandra modules FVy, x associated

(24)
to c l o s e d Kc-orbits are irreducible

Non-trivial quotients of Vy, x do e x i s t whenever the line bundle Ly, x

extends, g-equivariantly, across some Kc-orbit in ~Y . Matsuki [37]
 145

and Springer [47] have worked out the closure relations between K c-
orbits; their results make it p o s s i b l e to interpret this geometric irre-
ducibility criterion quite explicitely.

The proof of the original Kazhdan-Lusztig conjectures was the first

triumph of D-modules in r e p r e s e n t a t i o n theory. Irreducible modules in
the category 0 arise from orbits of a B o r e l subgroup B C G C , via the
same process of pushforward, t a k i n g s e c t i o n s , and p a s s i n g to t h e u n i q u e
irreducible submodule. Kazhdan and L u s z t i g [29] had already related
their conjectured composition multiplicities for V e r m a modules to the
intersection cohomology of c l o s u r e s of 8-orbits, i.e., of S c h u b e r t varie-
ties. Both Brylinski-Kashiwara [9] and Beilinson-Bernstein [6] saw t h e
connection with the t h e o r y of D-modules; they independently established
the conjectures, by r e l a t i n g the intersection cohomology to the composi-
tion multiplicities of the a p p r o p r i a t e sheaves. This second step carries
over, essentially unchanged, to the setting of H a r i s h - C h a n d r a modules.
The paper [35] of L u s z t i g and V o g a n contains the a n a l o g u e of the first
ingredient, namely the combinatorics of the intersection cohomology of
closures of Kc-orbits. Vogan [51], finally, deduces multiplicity for-
mulas for the Langlands classification, which he had conjectured earlier
[50]. I should point out that the original Kazhdan-Lusztig conjectures
cover only Ux-modules with x e A, as does the known version for Harish-
Chandra modules; Vogan's conjectures, by c o n t r a s t , apply to the general
case.

At first glance, the Beilinson-Bernstein construction appears far

removed from the construction of H a r i s h - C h a n d r a modules in t e r m s of line
bundles on G-orbits. The former leads quickly to g e o m e t r i c reducibility
criteria, as we just saw, and opens a path towards the Kazhdan-Lusztig
conjectures for Harish-Chandra modules. It also has points of c o n t a c t
with Vogan's classification via K-types [48]; indeed, it p r o b a b l y im-
plies the results of [48]. The G-orbit construction, on the other hand,
is c l o s e l y tied to L a n g l a n d s " classification, which in t u r n relates it to
certain analytic invariants of H a r i s h - C h a n d r a modules: the asymptotic
behavior of m a t r i x coefficients, for example, and the global character
[I0, II,24]. Since the two constructions complement each other, the p o s -
sible connections between them merit attention.

Results of M a t s u k i [36], on o r b i t s in flag varieties, provide an

important clue. For each G-orbit D C X , there exists a unique K C-
orbit Y , such that K acts transitively on the intersection DDY ;
conversely every Kc-orbit intersects a unique G-orbit in this manner,
The correspondence D ÷--~ Y reverses the relative sizes of orbits, as
measured by their dimensions -- I shall therefore call D "dual" to t h e
 146

orbit Y. Once the parameter x has been fixed, the duality D ~--+ Y
extends to the line bundles which enter the two constructions: a homo-
geneous line bundle L --+ D is dual to Ly, x --+ Y if the tensor product
L@Ly, Z restricts to a trivial K-homogeneous vector bundle over DDY .

It is instructive to examine the special case of SI(2, R) , or equi-

valently, its conjugate SU(I, I). The diagonal matrices in G = 5U(I,I)
constitute a maximal compact subgroup K ~ U(1). Both G and K C ~ C*
act on the flag variety X ~ C P 1 s C U {-} , as groups of M6bius trans-
formations. The duality relates the three Kc-orbits {0}, {®}, C ~,
in t h e given order, to the G-orbits A ( = unit disc), 6" ( = complement
of the closure of A ), 5 I. A homogeneous line bundle over A is deter-
mined by a character of the isotropy subgroup at 0 , i.e. a character
eX of K. Dually, a Kc-homogeneous line bundle over the one-point
space (0) has a single fibre, on which Kc acts by an algebraic cha-
racter e x. The differentials X, x may be viewed as linear functions
on k s C , whose values on Z C C are integral multiples of 2~i ; here
the duality reduces to I ÷--+ x = -X . The situation for the orbits A',
(-}, is entirely analogous. At points z e 51 , the isotropy subgroup
Gz C G has two connected components. Its characters are parametrized by
pairs ( I , ~ ) , consisting of a complex number 1 and a character
of the center {±i} C G , which meets both connected components of G z.
The corresponding G-equivariant line bundle over S1 extends holomor-
phically at least to the Kc-orbit C~ , as does the dual, or inverse
line bundle. If t h e line bundle is to extend even across {0} or {-} ,
the pair ( X , ~ ) must lift to a character of the complexification of
Gz -- this happens precisely when X/2~i is integral and ~ trivial or
non-trivial, depending on the parity of X/2~i.

To a Kc-homogeneous line bundle Lx--~ (0} , the Beilinson-Bernstein

construction assigns the Harish-Chandra module of "holomorpbic distribu-
tions" supported at 0 , with values in the bundle Lz -- in other words,
the U(g)-submodule generated by "evaluation at 0" in t h e algebraic dual
of the stalk O{0}(L~®T ~) ; the formal duality between functions and dif-
ferentials accounts for the appearance of the cotangent bundle T~. By
its very definition, this module is dual, in the sense of Barish-Chandra
modules, to H0(A,O(L~@T~))(K) , the module associated to the G-orbit
and the line bundle L~@T~. For non-negative values of the integral
parameter x/2Ti , the resulting Ha~ish-Chandra modules are irreducible
and belong to the discrete series. They become reducible if z/2Ti < -i ;
in t h i s situation the equivalence of categories (19) no longer applies.
The preceeding discussion carries over, word-for-word, to the pair of
orbits {-}, A'. As for the orbits 51 and C ~, the two constructions
 147

start with the choice of a G-homogeneous line bundle LA, ~ --+ 5 I. Its
extension to C ~, w h i c h I denote by the same symbol, comes equipped with
an a c t i o n of g and an a l g e b r a i c structure. 5ince C~ is open in X ,
the pushforward construction attaches the space of a l g e b r a i c sections
H O ( c ~ , O ( L x , ~)) to the datum of the Kc-orbit C~ and line bundle LX, ~
Integration over 51 pairs this Harish-Chandra module nondegenerately
with C®(SI,L*x, ~@T~) (K) • the m o d u l e corresponding to the G - o r b i t S 1 and
line bundle L'X,~@T~. The hypothesis of the equivalence of c a t e g o r i e s
translates into the inequality Re k > -I . On the Beilinson-Bernstein
side, this implies the existence of a unique irreducible submodule: the
entire module generically, when Lk, ~ cannot be c o n t i n u e d across 0
and - , otherwise the finite dimensional submodule consisting of s e c -
tions regular at the two punctures. The realization of the dual module
C'(51, LI, ~ @ T ~ ) ( K ) exhibits both Harish-Chandra modules as m e m b e r s of the
principal series.

One phenomenon that does not show up in the case of G = SU(I,I)

is t h e occurence of h i g h e r cohomology. For general groups, without any
positivity assumption on the parameter x , the sheaves Vy, x can h a v e
non-zero cohomology groups above degree zero, but these are still Harish-
Chandra modules. Zuckerman's derived functor construction also produces
a family of H a r i s h - C h a n d r a modules IP(D,L) , for each G-orbit D and
G-equivariant line b u n d l e L --+ D , i n d e x e d by an integer p 2 0 . The
example of 5U(1,1) suggest a duality between the two constructions, and
indeed this is the case. I fix p a i r s of data (Y, Vy, x), (D,L), which
are dual in the s e n s e described above, and define s = dimR(YnD) - dimcY ,
d = dim C X . Then

there exists a natural, nondegenerate pairing between

(25) the Harish-Chandra modules HP(Y, Vy,,) and Is-P(D,L®AdT~) ,

for all p ~ Z , with no restriction on x (Hecht-Milidit-Schmid-Wolf

[23]). In b o t h constructions homogeneous vector bundles can be s u b s t i -
tuted for line b u n d l e s . The duality carries over to t h i s wider setting,
and t h e n becomes compatible with the coboundary operators. Earlier,
partial results in t h e direction of (25) appear in V o g a n ' s proof of t h e
Kazhdan-Lusztig conjectures for Harish-Chandra modules [51]; there Vogan
identifies certain Beilinson-Bernstein modules with induced modules, by
explicit calculation.

The duality does not d i r e c t l y relate the 8eilinson-Bernstein classi-

fication to t h a t of L a n g l a n d s : in t h e language of g e o m e t r i c guantization,
the latter uses partially real polarizations, whereas the former works
with arbitrary, mixed polarizations. This problem can be d e a l t with on
 148

the level of Euler characteristics, and the known vanishing theorems for
the t w o constructions are sufficiently complementary to p e r m i t a compari-
son after all. In p a r t i c u l a r , it is p o s s i b l e to c a r r y techniques and
results back and forth between the two constEuctions [233.

By d e f i n i t i o n , the discrete series is the family of irreducible uni-

tary representations which occur discretely in L2(G). It was remarked
earlier that G has a non-empty discrete series if it c o n t a i n s a compact
Caftan subgroup; these representations then correspond to open G-orbits,
and their unitary structures are related to the geometric realization.
Open G-orbits are dual to c l o s e d Kc-orbits, so the observation (24)
"explains" the irreducibility statement (5). The discrete series lies at
one extreme of the various non-degenerate series of irreducible unitary
representations -- the other series consist of r e p r e s e n t a t i o n s unitarily
induced from discrete series representations of p r o p e r subgroups. These
are p r e c i s e l y the representations which occur in t h e decomposition of
L2(G) [22]. Roughly speaking, they are p a r a m e t r i z e d by hermitian line
bundles over G-orbits. Here, too, the inner products have geometric
meaning [56]. As for the rest of the unitary dual, the picture remains
murky, though substantial progress has been made during the past few
years. I shall limit myself to s o m e brief remarks, since more detailed
summaries can be f o u n d in the articles [30,52] of K n a p p - S p e h and Vogan.

A unitarizable Harish-Chandra module V is n e c e s s a r i l y conjugate

isomorphic to its own dual, a property which translates readily into a
condition on the Harish-Chandra character, or on the position of V in
the L a n g l a n d s classification. If the condition holds, V admits a non-
trivial g-invariant hermitian form -- only one, up to s c a l a r multiple,
provided V is i r r e d u c i b l e . The real difficulty lies in d e c i d i n g whether
the h e r m i t i a n form has a definite sign. For a one p a r a m e t e r family Vt
of irreducible Harish-Chandra modules of t h i s type, the form stays defi-
nite if it is d e f i n i t e anywhere: not until the family reduces at s o m e
t = to can the hermitian form become indefinite; even the composition
factors at t h e first reduction point are u n i t a r i z a b l e . In t h e case of
SI(2, R), such deformation techniques generate the complementary series
and t h e trivial representation -- in o t h e r words, all of t h e unitary
dual outside the discrete series and t h e unitary principal series [5].
Examples of K n a p p and 5pet [30] suggest that the analogous phenemenon for
general groups can b e c o m e quite complicated.

Neither induction nor deformation techniques account for isolated

points in the u n i t a r y dual. Typically isolated unitary representations
do exist, beyond those of the discrete series, but with certain formal
similarities to the discrete series. Zuckerman's derived functor con-
 149

struction, and the Beilinson-Bernstein construction as well, extends to

orbits in g e n e r a l i z e d flag vartieties, i.e., quotients Gc/P by p a r a b o -
lic s u b g r o u p s P C G C. The G-invariant hermitian line b u n d l e s over an
open G-orbit D C Gc/P are parametrized by the character group of the
center of the isotropy subgroup Gz C G at some reference point z e D.
Whenever that center is c o m p a c t , the derived functor construction pro-
duces a discrete family of irreducible Harish-Chandra modules. According
to a c o n j e c t u r e of Z u c k e r m a n , which was recently proved by V o g a n [53],
these modules are unitarizable. Vogan actually proves more; in g e o m e t r i c
language, the cohomology of G-invariant vector bundles, modeled on irre-
ducible unitary representations of the isotropy group G z , vanishes in
all but one d e g r e e and can be made unitary, again under an a p p r o p r i a t e
negativity assumption on the b u n d l e s . The proof consists of an a l g e b r a i c
reduction to t h e case of the non-degenerate series: Vogan introduces a
notion of s i g n a t u r e for g-invariant hermitian forms on H a r i s h - C h a n d r a
modules, formal sums of irreducible characters of K with integral coef-
ficients; he t h e n calculates these signatures for the derived functor
modules, in t e r m s of t h e K-multiplicities of induced modules. Because
of t h e origin of Z u c k e r m a n ' s conjecture, one m i g h t hope for a geometric
proof. Earlier attempts in this direction were only marginally success-
ful, but give a hint of a p o s s i b l e strategy [42].

A complete description of the u n i t a r y dual exists for groups of low

dimension, for groups of real rank one [4,27,32], and t h e family 50(n,2)
[i]. Vogan has just announced a classification also for the special
linear groups over R , C , H -- a big step, since there is no b o u n d on
real rank. In effect, the methods of u n i t a r y induction, degeneration,
and V o g a n ' s proof of t h e Zuckerman conjecture generate all irreducible
unitary representations of the special linear groups. One feature that
makes these groups more tractable is a h e r e d i t a r y property of t h e i r para-
bolic subgroups: all simple factors of the Levi component are again of
type S1 n. In t h e general case, conjectures of A r t h u r [2] and V o g a n [52]
predict the u n i t a r i t y of certain highly singular representations. There
are also results about particular types of u n i t a r y representations [14,
15,28], but a definite common pattern has yet to e m e r g e ~

I close my lecture by r e t u r n i n g to its s t a r t i n g point, the decompo-

sition of L2(G/H). The solution of t h i s problem for H = {e) -- the
explicit Plancherel formula [22] - was aim and crowning achievement of
Harish-Chandra's work on real groups. A discussion of his proof would
lead t o o far afield. However, I should mention a recent elementary',
though not simple, argument of H e r b and W o l f [26]. It is b a s e d on H e r b ' s
formulas for the d i s c r e t e series characters [25], and emulates Harish-
 150

Chandra's o r i g i n a l proof in the case of SI(2, R), by i n t e g r a t i o n by parts

[18].

The d e c o m p o s i t i o n p r o b l e m has been studied systematically for two

classes of s u b g r o u p s H , besides the identity group: arithmetically de-
fined subgroups, and symmetric subgroups, i.e. groups of fixed points of
involutive automorphisms. The symmetric case c o n t a i n s the case of the
trivial group, since G can be i d e n t i f i e d with GxG/diagonal Oshima
and M a t s u k i [403, building on a r e m a r k a b l e idea of F l e n s t e d - J e n s e n [16],
have d e t e r m i n e d the d i s c r e t e summands of L2(G/H) , for any s y m m e t r i c
H C G ; these representations are p a r a m e t r i z e d by h o m o g e n e o u s line bun-
dles ove~ certain orbits in g e n e r a l i z e d flag varieties. Oshima has also
described a notion of induction in the context of s y m m e t r i c quotients.
Presumably L2(G/H) is made up of r e p r e s e n t a t i o n s which are induced in
this sense, from d i s c r e t e summands b e l o n g i n g to smaller quotients, but
the explicit decomposition remains to be worked out. The case of a r i t h -
metically defined subgroups is the most interesting from m a n y points of
view, and the most difficult. Again the d i s c r e t e s u m m a n d s constitute the
"atoms" of the theory, as was shown by L a n g l a n d s [34] -- the E i s e n s t e i n
integral takes the place of induction. There is an e x t e n s i v e literature
on the d i s c r e t e spectrum, too e x t e n s i v e to be s u m m a r i z e d here, yet a full
understanding does not seem within reach.

References

[i] E. Angelopolos: 5ur les r e p r e s e n t a t i o n s de SOo(p,2). C.R. Acad.

Sci. Paris 292 (1981), 469-471
[2] J. Arthur: On some problems s u g g e s t e d by the trace formula. In
Lie Group R e p r e s e n t a t i o n s II. Springer L e c t u r e Notes in M a t h e m a -
tics 1041 (1984), pp. 1-49
[3] M. F. Atiyah and W. 5chmid: A geometric c o n s t r u c t i o n of the dis-
crete series for s e m i s i m p l e Lie groups. I n v e n t i o n e s Math. 42
(1977), 1-62
[4] M. W. Baldoni-Silva: The u n i t a r y dual of 5p(n, I), n~2. Duke
Math. J. 48 (1981), 5 4 9 - 5 8 4
[5] W. 8argman: Irreducible unitary representations of the L o r e n t z
group. Ann. of Math. 48 (1947), 5 6 8 - 5 4 0
[6] A. B e i l i n s o n and J. Bernstein: Localisation de g-modules. C.R.
Acad. 5ci. Paris, 292 (1981), 15-18
[7] J. Bernstein, I. M. G e l f a n d and S. I. Gelfand: D i f f e r e n t i a l ope-
rators on the base affine space and a study of g-modules. In Lie
Groups and their R e p r e s e n t a t i o n s . A k a d ~ m i a i Kiadb, Budapest 1975
[8] R. Bott: Homogeneous vector bundles. Ann. of Math. 66 (1957),
203-248
 151

[91 J.-L. Brylinski and M. Kashiwara: Kazhdan-Lusztig conjectures and

h o l o n o m i c systems. I n v e n t i o n e s Math. 64 (1981), 387-410
[I0] W. Casselman: Jacquet modules for real r e d u c t i v e groups. In Pro-
ceedings of the I n t e r n a t i o n a l C o n g r e s s of Mathematicians. Helsinki
1978, pp. 557-563
[Ii] W. Casselman and D. Milidit: Asymptotic behavior of matrix coef-
ficients of a d m i s s i b l e representations. Duke Math. J. 49 (1982),
869-930
[12] J. Dixmie~: Les C ~ - a l g ~ b r e s et Leur Representations. Gauthier-
Villars, Paris 1964
[13] M. Duflo: Construction de r e p r e s e n t a t i o n s unitaires d'un groupe
de Lie. Preprint
[14] T. J. Enright, R. Howe and N. R. Wallach: A c l a s s i f i c a t i o n of
u n i t a r y highest weight modules. In R e p r e s e n t a t i o n T h e o r y of
Reductive Groups. Progress in M a t h e m a t i c s 40 (1983), pp. 97-144
[15] T. J. Enright, R. Parthasarathy, N. R. Wallach and J. A. Wolf:
Unitary derived functor modules with small spectrum. Preprint
[16] M. Flensted-Jensen: Discrete series for s e m i s i m p l e symmetric
spaces. Ann. of Math. III (1980), 253-311
[17] L. G~rding: Note on continuous r e p r e s e n t a t i o n s of Lie groups.
Proc. Nat. Acad. Sci. USA 33 (1947), 331-332
[18] Harish-Chandra: Plancherel formula for the 2×2 real unimodular
group. Proc. Nat. Acad. 5ci. UZA 38 (1952), 337-342
[19] Harish-Chandra: R e p r e s e n t a t i o n s of semisimple Lie groups I.
Trans. Amer. Math. Soc. 75 (1953), 185-243
[20] Harish-Chandra: The characters of semisimple Lie groups. Trans.
Amer. Math. Soc. 83 (1956), 98-163
[21] Harish-Chandra: Discrete series for semisimple Lie groups If.
Acta Math. 116 (1966), i-Iii
[22] Harish-Chandra: Harmonic analysis on real reductive groups III.
Ann. of Math. 104 (1976), 117-201
[23] H. Hecht, D. Milidit, W. 5chmid and J. Wolf: L o c a l i z a t i o n of
H a r i s h - C h a n d r a modules and the derived functor construction. To
appear
[24] H. Hecht and W. 5chmid: Characters, asymptotics, and n - h o m o l o g y
of H a r i s h - C h a n d r a modules. Acta Math. 151 (1983), 49-151
[25] R. Herb: D i s c r e t e series characters and Fourier inversion on
semisimple real Lie groups. Trans. Amer. Math. 5oc. 277 (1983),
241-261
[26] R. Herb and J. Wolf: The P l a n c h e r e l theorem for general semi-
simple Lie groups. Preprint
[27] T. Hirai: On irreducible r e p r e s e n t a t i o n s of the Lorentz group of
n-th order. Proc. Japan. Acad. 38 (1962), 83-87
[28] R. Howe: Small u n i t a r y r e p r e s e n t a t i o n s of c l a s s i c a l groups. To
appear in the P r o c e e d i n g s of the M a c k e y Conference, B e r k e l e y 1984
[29] D. Kazhdan and G. Lusztig: R e p r e s e n t a t i o n s of Coxeter groups and
Hecke algebras. Inventiones Math. 53 (1979) 165-184
[30] A. Knapp and B. Speh: Status of c l a s s i f i c a t i o n of irreducible
u n i t a r y representations. In Harmonic Analysis, Proceedings, 1981.
Springer Lecture Notes in M a t h e m a t i c s 908 (1982), pp. 1-38
 152

[31] A. K n a p p and G. Zuckerman: C l a s s i f i c a t i o n of i r r e d u c i b l e t e m p e r e d

representations of s e m i s i m p l e groups. Ann. of Math. 116 (1982),
389-501
[32] H. K r a l j e v i t : R e p r e s e n t a t i o n s of the u n i v e r s a l c o v e r i n g group of
the g r o u p S U ( n , l ) . G l a s n i k Mat. 8 (1973), 2 3 - 7 2
[33] R. L a n g l a n d s : On the c l a s s i f i c a t i o n of i r r e d u c i b l e r e p r e s e n t a -
t i o n s of real a l g e b r a i c groups. M i m e o g r a p h e d notes, I n s t i t u t e for
A d v a n c e d S t u d y 1973
[34] R. L a n g l a n d s : On the F u n c t i o n a l E q u a t i o n s S a t i f i e d by E i s e n s t e i n
series. S p r i n g e r L e c t u r e N o t e s in Mathematics 544 (1976)
[35] G. L u s z t i g and D, Vogan: S i n g u l a r i t i e s of c l o s u r e s of K orbits
on flag m a n i f o l d s ~ I n v e n t i o n e s Math. 71 (1983)
[36] T. M a t s u k i : The o r b i t s of a f f i n e s y m m e t r i c s p a c e s u n d e r the
a c t i o n of m i n i m a l p a r a b o l i c s u b g r o u p s . J. Math. 5oc, J a p a n 31
(1979), 3 3 1 - 3 5 7
[37] T. Matsuki: Closure relation for K-orbits on c o m p l e x flag mani-
folds. Preprint
[38] D. Mili~i6: A s y m p t o t i c b e h a v i o r of m a t r i x coefficients of the
d i s c r e t e series. D u k e Math. J. 44 (1977), 59-88
[39] E. Nelson: Analytic vectors. Ann. of Math. 70 (1959), 572-615
[40] T. O s h i m a and T. M a t s u k i : A d e s c r i p t i o n of d i s c r e t e s e r i e s for
s e m i s i m p l e s y m m e t r i c spaces. To a p p e a r in Adv. S t u d i e s in Math.
[41] S, J. P r i c h e p i o n o k : A n a t u r a l t o p o l o g y for l i n e a r r e p r e s e n t a t i o n s
of s e m i s i m p l e Lie a l g e b r a s . S o v i e t Math. Dokl. 17 (1976), 1 5 6 4 - 6 6
[42] J. R a w n s l e y , W. 5 c h m i d and J. A. Wolf: Singular unitary represen-
t a t i o n s and i n d e f i n i t e h a r m o n i c theory. J. Funct. Anal. 51 (1983),
1-114
[43] W. Schmid: Homogeneous c o m p l e x m a n i f o l d s and r e p r e s e n t a t i o n s of
semisimple Lie groups. Thesis, UC B e r k e l e y 1967
[44] W. 5chmid: L2-cohomology and the discrete series. Ann. of Math.
102 (1975), 535-564
[45] W. Schmid: Boundary value p r o b l e m s for g r o u p i n v a r i a n t differen-
tial e q u a t i o n s . To a p p e a ~ in P r o c e e d i n g s of the C a f t a n Symposium,
L y o n 1984
[46] B. S p e h and D. Vogan: R e d u c i b i l i t y of g e n e r a l i z e d p r i n c i p a l se-
ries r e p r e s e n t a t i o n s . A c t a Math. 145 (1980), 2 2 7 - 2 9 9
[47] T. A. S p r i n g e r : Some results on a l g e b r a i c groups with involu-
tions~ Preprint
[48] D. Vogan: T h e a l g e b r a i c s t r u c t u r e of the r e p r e s e n t a t i o n s of s e m i -
s i m p l e Lie g r o u p s I. Ann. of Math. 109 (1979), 1-60
[49] D. Vogan: Representations of R e a l Reductive Lie Groups. Birkhiu-
set, B o s t o n 1981
[50] D. Vogan: Irreducible c h a r a c t e r s of s e m i s i m p l e Lie g r o u p s II. The
Kazhdan-Lusztig c o n j e c t u r e s . D u k e Math. J. 46 (1979), 6 1 - 1 0 8
[51] D. Vogan: I r r e d u c i b l e c h a r a c t e r s of s e m i s i m p l e Lie groups III.
Proof of Kazhdan-Lusztig c o n j e c t u r e in the i n t e g r a l case. Inven-
t i o n e s Math. 71 (1983), 3 8 1 - 4 1 7
[52] D. Vogan: U n d e r s t a n d i n g the u n i t a r y dual. In Lie G r o u p R e p r e s e n -
t a t i o n s I. S p r i n g e r L e c t u r e N o t e s in M a t h e m a t i c s 1024 (1983), pp.
264-288
 153

[53] D. Vogan: U n i t a r i z a b i l i t y of certain series of representations.

Ann. of Math. 120 (1984), 141-187
[54] N. Wallach: A s y m p t o t i c expansions of g e n e r a l i z e d matrix entries
of r e p r e s e n t a t i o n s of real reductive groups. In Lie Group Repre-
sentations I. Springer Lecture Notes in M a t h e m a t i c s 1024 (1983),
pp. 287-369
[55] N. Wallach: On the u n i t a r i z a b i l i t y of derived functor modules,
Preprint
[56] J~ A. Wolf: Unitary R e p r e s e n t a t i o n s on P a r t i a l l y H o l o m o r p h i c Co-
h o m o l o g y 5paces. Amer. Math. 5oc. Memoir 138 (1974)
 LOOP GROUPS

G.B. Segal,
St. C a t h e r i n e ' s College,
Oxford.

§I General remarks

In this talk a loop group LG will mean the group of smooth maps
from the circle S I to a compact Lie group G. One reason for study-
ing such groups is that they are the simplest examples of infinite
dimensional Lie groups. Thus LG has a Lie algebra L~ - the loops
in the Lie algebra ~ of G - and the e x p o n e n t i a l map L ~ ÷ LG is a
local d i f f e o m o r p h i s m . Furthermore LG has a c o m p l e x i f i c a t i o n LG~,
the loops in the c o m p l e x i f i c a t i o n of G. Neither of these properties
is to be e x p e c t e d of infinite dimensional groups: neither holds, for
example, for the group of d i f f e o m o r p h i s m s of the circle [17].
From this point of view the group Map(X;G) of smooth maps
X ÷ G, where X is an a r b i t r a r y compact manifold, seems almost as
simple as LG. Such groups are of great importance in q u a n t u m theory,
where they occur as "gauge groups" and "current groups"; the m a n i f o l d
X is p h y s i c a l space. Thus loop groups arise in q u a n t u m field theory
in t w o - d i m e n s i o n a l space-time. In fact it is not much of an
exaggeration to say that the m a t h e m a t i c s of t w o - d i m e n s i o n a l quantum
field theory is almost the same thing as the r e p r e s e n t a t i o n theory
of loop groups.
If dim(X) > 1, however, surprisingly little is k n o w n about the
group Map(X;G). Essentially only one irreducible representation of
it is known - the r e p r e s e n t a t i o n of Vershik, Gelfand and Graev [9] -
and that r e p r e s e n t a t i o n does not seem relevant to q u a n t u m field
theory. For loop groups, in contrast, there is a rich and
extensively developed theory. They first became popular because of
their c o n n e c t i o n with the intriguing combinatorial identities of
Macdonald [16]. They are the groups whose Lie
 156

algebras are the "affine algebras" of K a c - M o o d y - roughly speaking,

the algebras associated to p o s i t i v e - s e m i d e f i n i t e Cartan matrices.
From that point of v i e w the groups have been d i s c u s s e d in Tits's talk
In this talk I shall keep away from the Lie algebra theory, of w h i c h
there is an e x c e l l e n t exposition in the recent book of Kac [11], and
instead shall attempt to survey what is known about the global
geometry and analysis connected with the groups.

From any point of v i e w the crucial property of loop groups is

the e x i s t e n c e of the o n e - p a r a m e t e r group of a u t o m o r p h i s m s which
simply rotates the loops. It permits one to speak of r e p r e s e n t a t i o n s
of LG of p o s i t i v e energy. A representation of LG on a t o p o l o g i c a l
vector space H has p o s i t i v e energy if there is given a positive
action of the circle group T on H w h i c h intertwines with the action
of LG so at to provide a representation of the s e m i d i r e c t product
T ~ LG, where T acts on LG by rotation. An action of T on H is
p o s i t i v e if e i8 E T acts as e iAe, w h e r e A is an o p e r a t o r with
positive spectrum. It turns out that representations of LG of
positi v e energy are n e c e s s a r i l y projective (cf. (4.3) below).
The theory of the p o s i t i v e energy representations of LG (or,
more accurately, of T ~ LG) is strikingly simple, and in strik-
ingly close analogy with the r e p r e s e n t a t i o n theory of compact
groups.(*) Thus the irreducible representations
(i) are all unitary,
(ii) all extend to h o l o m o r p h i c representations of LG{, and
(iii) form a c o u n t a b l e discrete set, parametrized by the points
of a p o s i t i v e cone in the lattice of c h a r a c t e r s of a torus.
None of these p r o p e r t i e s holds, for example, for the r e p r e s e n t a t i o n s
of S L 2 ( ~ ) •
The p o s i t i v e energy c o n d i t i o n is strongly motivated by q u a n t u m
field theory: the circle action on H c o r r e s p o n d s to the time
evolution on the Hilbert space H of states. It w o u l d be very
interesting if one could formulate an analogous condition for more
general groups Map(X;G) . Certainly in q u a n t u m field theory one
might expect such a gauge group to act on a state space on w h i c h
time e v o l u t i o n was d e f i n e d by a p o s i t i v e Hamiltonian operator, and

(*) We are t h i n k i n g of c o n t i n u o u s representations on a r b i t r a r y

compl e t e locally convex topological vector spaces. But we do not
distinguish between representations on H and ~ if there is an
injective intertwining operator H ÷H with dense image.
 157

the g a u g e transformations should intertwine in some perhaps com-

licated way with the t i m e evolution. But there has been no p r o g r e s s
on this front, and the a t t e m p t may well be m i s c o n c e i v e d . (Cf. §3
below.)

To c o n c l u d e these introductory remarks I should say that the

material I am g o i n g to p r e s e n t is all e s s e n t i a l l y well-known, and
has b e e n w o r k e d out independently by m a n y people in s l i g h t l y different
contexts. As r e p r e s e n t a t i v e treatments of v a r i o u s aspects of the
subject from standpoints somewhat different from mine let me refer
to G a r l a n d [8], Lepowsky [15], Kac and P e t e r s o n [12], Goodman and
Wallach [10], Frenkel [5]. More details of m y o w n approach c a n be
found in [18], [19] and [20].

§2 The fundamental homogeneous space X

In the study of LG the h o m o g e n e o u s space X = LG/G (where G is

identified with the c o n s t a n t loops in LG) plays a central role.
One can t h i n k of X as the space QG of b a s e d loops in G; b u t we
prefer to r e g a r d it as a h o m o g e n e o u s space of LG. I shall list its
most important properties.

(i) X is a c o m p l e x manifold, and in fact a homogeneous space of

the c o m p l e x group LG~:

+
X = LG/G ~ LG~/L G~ . (2.1)

Here L + G~ is the g r o u p of s m o o t h maps y : S I ÷ G~ w h i c h are the

boundary values of h o l o m o r p h i c maps

y : {z 6 • : Izl < I} ÷ G~ .

The isomorphism (2.1) is e q u i v a l e n t to the assertion that any loop y

in L G ~ can be factorized

y = yu.y+

+
with ¥u 6 LG a n d Y+ 6 L G ~ . This is analogous to the factorization
of an element of GL (~) as (unitary) x (upper triangular).
n
 158

(ii) For each invariant inner product < , > on the Lie algebra
of G there is an invariant closed 2 - f o r m ~ on X w h i c h makes it a
symplectic manifold, and even fits t o g e t h e r with the complex structure
to make a Kahler manifold. The tangent space to X at its b a s e - p o i n t
is L ~ /~ , and ~ is given there by
2n
(2.2)
0

(iii) The energy function ~ : X + ~+ defined by

2~

0
is the H a m i l t o n i a n function c o r r e s p o n d i n g in terms of the symplectic
structure to the c i r c l e - a c t i o n on X w h i c h rotates loops. The c r i t i c a l
points of ~ are the loops w h i c h are h o m o m o r p h i s m s qr ÷ G. Downwards
g r a d i e n t t r a j e c t o r i e s of ~ emanate from every point of X, and travel
to c r i t i c a l points of ~ . The g r a d i e n t flow of ~ and the H a m i l t o n i a n
circle action fit t o g e t h e r to define a h o l o m o r p h i c action on X of the
multiplicative semigroup ~I = {z 6 • : 0 < Iz[ ~ I}.
The c o n n e c t e d c o m p o n e n t s C[l] of the critical set of ~ are the
c o n j u g a c y classes of h o m o m o r p h i s m s I : T ÷ G. They c o r r e s p o n d to
the orbits of the Weyl group W on the lattice z1(T), where T is a
m a x i m a l torus of G. The g r a d i e n t flow of ~ stratifies the m a n i f o l d
X into locally closed complex submanifolds X[I], where X[k] consists
of the points which flow to C[I]. Each stratum X[I ] is of finite
codimension.

Proposition (2.3). The s t r a t i f i c a t i o n c o i n c i d e s w i t h the d e c o m p o s i -

tion of X into orbits of L-G~; i.e. X[l] = L-G~.k.

Here L-G~ is the group of loops in G~ which are b o u n d a r y values

of h o l o m o r p h i c maps D ÷ G~, where D = {z 6 S 2 : Izl > I}.
Proposition (2.3) is the classical B i r k h o f f f a c t o r i z a t i o n theorem:
a loop y in G~ can be f a c t o r i z e d

y = y_.l.T+ ,

+
with ¥± 6 L-G~, and I : S I ÷ G a homomorphism. This is the analogue
 159

of factorizing an element of GLn(~) as

(lower t r i a n g u l a r ) x ( p e r m u t a t i o n m a t r i x ) × ( u p p e r triangular).

There is one dense open s t r a t u m X 0 in X. It is contractible,

and can be i d e n t i f i e d w i t h the n i l p o t e n t group
LoG~ = {y 6 L-G~ : y(~) = 1}.

(iv) The complex structure of X can be c h a r a c t e r i z e d in another

way, pointed out by A t i y a h [I]. To give a h o l o m o r p h i c map Z ÷ X,
where Z is an a r b i t r a r y complex manifold, is the same as to give a
h o l o m o r p h i c p r i n c i p a l G ~ - b u n d l e on Z × S 2 together w i t h a trivial-
ization over Z × D . If Z is compact it follows that the space of
based maps Z ÷ X in a given h o m o t o p y class is finite dimensional; for
the moduli space of G ~ - b u n d l e s of a given t o p o l o g i c a l type is finite
dimensional. This is a rather striking fact, showing that X, a l t h o u g h
a rational variety, is quite unlike, say, an infinite d i m e n s i o n a l
c o m p l e x p r o j e c t i v e space: for in X the set of points w h i c h can be
joined to the b a s e - p o i n t by h o l o m o r p h i c curves of a given degree is
only finite dimensional.

§3 The G r a s s m a n n i a n e m b e d d i n g of X

Let us choose a finite d i m e n s i o n a l u n i t a r y r e p r e s e n t a t i o n V of

compact group G, and let H denote the Hilbert space L2(S I;V) .
E v i d e n t l y LG~ acts on H, and we have a h o m o m o r p h i s m i : LG~ ÷ GL(H)
an e m b e d d i n g if V is faithful.
To make a more refined statement we write H = H+ (9 H_, where
in@
H (resp. H ) consists of the functions of the form E v e
(resp. Z v e In@) w i t h v 6 V. The r e s t r i c t e d 9eneral linear group
n<0 n n . . . . . .

GLres(H) is defined as the subgroup of GL(H) c o n s i s t i n g of e l e m e n t s

Ia b] (3.1)
c d
whose o f f - d i a g o n a l blocks b,c (with respect to the d e c o m p o s i t i o n
H+ @ H_) are Hilbert-Schmidt. The blocks a and d are then auto-
m a t i c a l l y Fredholm.

Proposition (3.2!. i(LG~) c GLres(H ) .

 160

The set of c l o s e d subspaces of H o b t a i n e d f r o m H+ b y t h e action

of G L = GL (H) w i l l b e c a l l e d the G r a s s m a n n i a n Gr(H). It is
res res
naturally a Hilbert manifold, and has the homotopy t y p e of t h e space
known to topologists as ~ × BU. The homomorphism

i : LG~ ÷ GLres(H) (3.3)

induces a smooth map (again an e m b e d d i n g if V is f a i t h f u l )

i : X ÷ Gr(H)

This map is c l o s e l y connected with the B o t t p e r i o d i c i t y theorem. In

fact Bott's theorem asserts that when G = U a n d V = { n the m a p is a
n
homotopy equivalence u p to d i m e n s i o n 2n-2.
It s h o u l d b e r e m a r k e d that i(X) is far from being a closed sub-
manfold o f Gr(H) . Indeed it is so h i g h l y curved that its c l o s u r e is
not a submanifold of Gr(H).

There is a h o l o m o r p h i c line b u n d l e D e t o n Gr(H) whose fibre at

W c H can be thought o f as the renormalized "top e x t e r i o r power" of
W. Because of t h e renormalization needed to d e f i n e it it is n o t a
homogeneous bundle under G L r e s, b u t its g r o u p of h o l o m o r p h i c auto-
morphisms is a c e n t r a l extension ~L of G L b y ~x. The
res res
homomorphism (3.3) then gives us a c e n t r a l e x t e n s i o n of L G ~ b y {×;
u p to f i n i t e - s h e e t e d coverings, all e x t e n s i o n s of L G { b y ~x a r e
obtained in t h i s way. (The L i e algebra cocycle of t h e extension is
given by (2.2), where < , > is the trace f o r m of V.)
The line bundle Det has no holomorphic sections, but its d u a l
Det* has an infinite dimensional space of sections F on which ~L
res
acts irreducibly. Just as t h e space of sections of t h e d u a l of t h e
determinant bundle on the Grassmannian Gr(E) of a f i n i t e dimensional
vector space E can be identified with the exterior algebra A(E*) we
find

Proposition 3.4. F ~ A(H+ @ H_)

This space is v e r y familiar in q u a n t u m field theory as the

"fermionic Fock space" got by quantizing a classical state space H
 161

(e.g. the space of s o l u t i o n s of the D i r a c equation) in w h i c h H÷ and

H_ are the states of p o s i t i v e and n e g a t i v e energy.
F r o m the p o i n t of v i e w of loop groups the importance of F is t h a t
when G = U n and V = {n the p r o j e c t i v e action of LU n on F via (3.3) is
the "basic" irreducible representation of LU n (cf. §4 below). It
even extends f r o m LU n to LO2n, for F is m o s t correctly regarded as
the spin representation of the restricted orthogonal group of the
real Hilbert space underlying H.

Let us b r i e f l y consider generalizing the foregoing discussion

to the g r o u p Map(X;G), where X is a c o m p a c t odd-dimensional Riemannian
manifold. If H is the space of spinor fields on X t h e n M a p ( X ; U n)
acts naturally on H ® {n. We can decompose

H 8 ~n = (H+ (~ ~n) @ (H_ O {n) ,

where H± are the p o s i t i v e and n e g a t i v e eigenspaces of the D i r a c

operator. We get an e m b e d d i n g

M a p ( X ; U n ) + GL(m) (H 8 ~n) , (3.5)

where GL(m ) denotes the group of o p e r a t o r s of the form (3.1) in

which the o f f - d i a g o n a l blocks belong to the S c h a t t e n ideal ~ m with
m-1 = dim(X). (Cf. [21],[4].)
The homomorphism (3.5) is v e r y interesting: topologically it
represents the index map in K - t h e o r y [4]. On the o t h e r hand no
representations of G L ( m ) are known, and one even feels that rep-
resentations are not the natural thing to look for, as the two-
dimensional cohomology class which forces GLre s = GL(2 ) to have a
projective rather than a genuine representation is r e p l a c e d by an
m-dimensional class for G L ( m ) . Alternatively expressed, on the
Grassmannian Gr(m) (H) a s s o c i a t e d with GL(m ) there is a t a u t o l o g i c a l
infinite dimensional bundle with a connection. The "determinant"
line b u n d l e of this - i.e. its first Chern class - cannot be d e f i n e d ,
but nevertheless the h i g h e r components of its C h e r n character do m a k e
geometric sense.
 162

§4 The Borel-Weil theory

(i) The basic representation

To simplify the d i s c u s s i o n we shall assume f r o m n o w on that the

compact group G is s i m p l y connected and simple. Then H2(X;~) ~ ~ ,
and so the c o m p l e x line b u n d l e s L on X are classified by an i n t e g e r
invariant c1(L). In fact each bundle has a unique holomorphic
structure, and has n o n - z e r o holomorphic sections if and o n l y if
c1(L) ~ 0. The space of h o l o m o r p h i c sections of the b u n d l e L 1 with
c 1 ( L I) = I is c a l l e d the b a s i c representation of LG~: we h a v e
remarked that when G = SU n the b u n d l e L I is the restriction of
Det* on Gr(H). As we saw in that case, L 1 is not quite homogeneous
under LG{. The holomorphic automorphisms of L I w h i c h cover the
action of LG~ on X f o r m a group ~G~ which is a c e n t r a l extension of
LG~ by ~× - in fact its u n i v e r s a l central extension. It c o r r e s p o n d s
to the Lie algebra cocycle (2.2) for an i n n e r product < , > on
which I shall also call "basic".
One reason for the n a m e "basic" is p r o v i d e d by

Proposition (4.1). If G is a s i m p l y - l a c e d group and F is the b a s i c

representation of LG{ then any irreducible representation of p o s i t i v e
energy is a d i s c r e t e summand in p'F, where p : LG~ ÷ L G ~ is an e n d o -
morphism.

(ii) The Borel-Weil theorem

To d e s c r i b e all the p o s i t i v e energy irreducible representations

of LG we m u s t consider the larger complex homogeneous space
Y = LG/T, where T is a m a x i m a l torus of G. This manifold Y is f i b r e d
over X with the finite dimensional complex homogeneous space G/T as
fibre. Complex line b u n d l e s on Y are c l a s s i f i e d topologically by

H2(y;zz) ~ZZ • H2(G/T;ZZ) ~ @ T ,

where T is the c h a r a c t e r group of T. Once again each bundle has a

unique holomorphic structure, and is h o m o g e n e o u s under ~G~. If we
denote the b u n d l e corresponding to (n,l) 6 ~ @ T by Ln, l t h e n we
have the following "Borel-Weil" theorem.
 163

Proposition (4.2).

(a) The space F(Ln,I) of h o l o m o r p h i c sections of Ln, 1 is either

zero or an irreducible r e p r e s e n t a t i o n of LG~ of p o s i t i v e energy.
(b) Every p r o j e c t i v e irreducible r e p r e s e n t a t i o n of LG of
positive energy arises in this way.
(c) F(Ln, I) # 0 if and only if (n,l) is positive in the sense
that

0 < l(h ) < n<h ,h >

for each positive coroot h of G, where < , > is the basic inner
product on ~ .

It should be e m p h a s i z e d that except for the "if" part of (c)

this p r o p o s i t i o n is quite elementary, a m o u n t i n g to little more than
the o b s e r v a t i o n s that (i) any r e p r e s e n t a t i o n of positive energy
contains a ray invariant under L-G{, and (ii) L-G~ acts on Y with a
dense orbit. Thus the e l e m e n t a r y part already yields

Corollary (4.3). For positive energy r e p r e s e n t a t i o n s of LG:

(a) each r e p r e s e n t a t i o n is n e c e s s a r i l y projective,
(b) each r e p r e s e n t a t i o n extends to a h o l o m o r p h i c r e p r e s e n t a t i o n
of LG{, and
(c) each irreducible r e p r e s e n t a t i o n is of finite type, i.e. if
it is d e c o m p o s e d into energy levels H = ~ H , where H is the part
• q q
where the rotation e i@ 6 ~- acts as e lq8, then each H has finite
q
dimension.

Assertion (c) holds because a h o l o m o r p h i c section of Ln, 1 is

d e t e r m i n e d by its Taylor series at the base-point. That gives one
an injection

F(Ln, I) ÷ S(Ty) , (4.4)

where Ty is the tangent space to Y at the base-point, and S denotes

the c o m p l e t e d symmetric algebra. The injection is c o m p a t i b l e with
the action of T , and the right hand side of (4.4) is of finite type.
 164

(iii) Unitarity

We have mentioned that all positive energy representations of

LG are unitary. In fact a s i m p l e formal argument shows that each
irreducible representation has a non-degenerate invariant sesquilinear
form, but it is n o t so s i m p l e to s h o w t h a t it is p o s i t i v e definite.
By (4.1) it is e n o u g h to consider the b a s i c representation. When
G = SU n or S 0 2 n the u n i t a r i t y is t h e n c l e a r f r o m the d e s c r i p t i o n
(3.4) o f the b a s i c representation; and one can deal similarly with
all simply laced groups by the method of §5 b e l o w . The only proof
known in t h e g e n e r a l ease is a n inductive argument in t e r m s of
generators and relations, due to G a r l a n d [7].
It w o u l d obviously be v e r y attractive to p r o v e the unitarity
directly by putting an i n v a r i a n t measure on the infinite dimensional
manifold Y and using the standard L 2 inner product. That has not
yet been done, though it s e e m s to be p o s s i b l e . The measure will be
supported not on Y but on a thickening Y*, to w h i c h the h o l o m o r p h i c
line bundles L extend. One expects to h a v e an L G - i n v a r i a n t measure
on s e c t i o n s of L @ L f o r e a c h p o s i t i v e bundle L. There is n o d i f -
ficulty in f i n d i n g a candidate f o r Y*: the m a n i f o l d Y is m o d e l l e d on
the Lie algebra N ~ ~ of holomorphic maps ~ : D ÷ ~{ (with ~(~)
lower triangular) which extend smoothly to the b o u n d a r y of D ; the
thickening is m o d e l l e d o n the d u a l space, i.e. the holomorphic maps
with distributional boundary values o n S I. (*)

(iv) T h e K a c character formula and the Bernstein-Gelfand-

Ge!fand resolution

Because each irreducible representation of T ~ L G is of finite

type it m a k e s sense to s p e a k of its formal character, i.e. of its
decomposition under the t o r u s T × T. This is g i v e n by the Kac
character formula, an exact analogue of t h e c l a s s i c a l Weyl character
formula.
Thinking of Y = L G / T as T ~ LG/ T × T, w e o b s e r v e that the
torus ~ × T acts on Y with a discrete set of fixed points. This
set is t h e a f f i n e W e y l group War f = N(T × T)/(T x T). If o n e
ignores the infinite dimensionality of Y and writes down formally

(*) A n i n t e r e s t i n g f a m i l y of m e a s u r e s o n Y is c o n s t r u c t e d in [5],
but it d o e s n o t i n c l u d e the m e a s u r e n e e d e d to p r o v e u n i t a r i t y .
 165

the Lefschetz fixed-point formula of A t i y a h - B o t t [2] for the character

of the torus action on the h o l o m o r p h i c sections of a p o s i t i v e line
bundle L on Y then one obtains the Kac formula, at least if one
assumes that the c o h o m o l o g y groups Hq(Y; ~(L)) v a n i s h for q > 0.
(Here ~(L) is the sheaf of h o l o m o r p h i c sections of L.) Unfortunately
it does not seem possible at present to prove the formula this way.
One can do better by using more information about the g e o m e t r y
of the space Y. It p o s s e s s e s a s t r a t i f i c a t i o n just like that of X
d e s c r i b e d in §2. The strata {Ew } are complex affine spaces of finite
codimension, and are indexed by the elements w of the group Waff:
indeed Z is the orbit of w under N-G~ = {7 6 L-G~ : 7(~) is lower
w
triangular}.
Let Y denote the union of the strata of complex c o d i m e n s i o n p.
P
The c o h o m o l o g y groups H*(Y; ~(L)) are those of the cochain c o m p l e x
K" formed by the sections of a flabby r e s o l u t i o n of ~(L). Filtering
K" by d e f i n i n g K" as the subcomplex of sections with support in
P
Yp gives us a spectral sequence c o n v e r g i n g to H*(Y; ~(L)) w i t h
E~
q_ = HP+q(Kp/K~+I
. ~) B e c a u s e Yp is affine and has an open n e i g h b o u r -
hood U isomorphic to Y x {P the spectral sequence collapses, and
P P
its E l - t e r m reduces to

E? ° = IUp;
P

E pq
!
= 0 if q ~ 0

In other words H*(Y; ~(L)) can be c a l c u l a t e d from the cochain complex

{H~ (Up; ~(L))}. Here H~ (Up; ~(L)) means the e o h o m o l o g y of the sheaf
~ ( L ) IUp with' supports Pln Yp. It is simply the space of h o l o m o r p h i c
sections of the bundle on Y whose fibre at y is
P

Ly 0 H~O } ( N y ; ~ ) ,

where Ny ~ {P is the normal space to Yp at y; furthermore, H~0}(Ny; ~ )

is the dual of the space of h o l o m o r p h i c p-forms on Ny. Thus as a
r e p r e s e n t a t i o n of ~ × T

E pO m (~ S(T* @ N w) ~9 det(N w) @ L w ,
w

where w runs through the elements of Waf f of c o d i m e n s i o n p, and T w

and N are the tangent and normal spaces to Z at w. If we k n o w
W W
that Hq(Y; ~ (L)) = 0 for q > 0 then we can read off the Kac formula.
 166

The cochain complex E~ 0 is the dual of the B e r n s t e i n - G e l f a n d -

Gelfand resolution, described in the finite dimensional case in [3]
(cf. also [13]). Its e x a c t n e s s can be p r o v e d by standard algebraic
arguments, and one can deduce the v a n i s h i n g of the higher cohomology
groups Hq(Y; ~ ( L ) ) . But it w o u l d be a t t r a c t i v e to reverse the
argument by p r o v i n g the v a n i s h i n g theorem analytically.

§5 "Blips" o r "vertex operators"

The B o r e l - W e i l construction of r e p r e s e n t a t i o n s is quite

inexplicit. I shall c o n c l u d e with a very brief d e s c r i p t i o n of an
interesting explicit construction of the basic representation of LG,
for simply laced G, w h i c h was independently extracted from the
physics literature in [14], [6] and [19].
The idea is to start w i t h a standard irreducible projective
representation H of LT, and to extend the action from LT to LG.
The a b e l i a n group LT is e s s e n t i a l l y a vector space, and for H we
take its "Heisenberg" representation. To make the Lie a l g e b r a
L~ act on H amounts to defining, for each basis element ~i of ~6,
an " o p e r a t o r - v a l u e d distribution" B i on $I: for then an e l e m e n t Efi~ i
of L ~ will act on H by

Z I fi(8)Bi (8)d8
i S1

We must c o n s t r u c t B i for each basis element of ~{/t 6" These are

indexed by the roots of G, and the r e m a r k a b l e fact about simply-
laced groups (i.e. those for w h i c h all the roots have the same
length) is that the roots correspond precisely to the set of all
I
homomorphisms ~ : ~ + T of minimal length. N o w for each @ E S
and each small positive a let us c o n s i d e r the b l i p - l i k e element
B ,8,£ of LT such that

B ,9,s(8') = 1 if 18' - 8 I > S ,

while on the interval (8 - s, 8 + e) of the circle B ~,@,S describes

the loop ~ in T. When Bd,@, s is r e g a r d e d as an o p e r a t o r on H it
turns out that the r e n o r m a l i z e d limit

lim e-IB
s÷0 ~,@,s
 167

exists in an appropriate sense, and is the desired Be(@). Such

o p e r a t o r s have been called "vertex operators" in the physics
literature.
E x t e n d i n g the r e p r e s e n t a t i o n from theLie algebra to LG p r e s e n t s
no problems.

REFERENCES

[I] M.F. Atiyah, Instantons in two and four dimensions. To appear

[2] M.F. Atiyah and R. Bott, A. Lefschetz fixed point formula for
elliptic complexes: II. Applications. Ann. of Math., 8 8 (1968),
451-491

[3] I.N. Bernstein, I.M. Gelfand, and S.I. Gelfand, D i f f e r e n t i a l

operators on the base affine space and a study of ~ -modules.
In Lie groups and their representations. Summer School of the
Bolyai Janos Math. Soc., ed. I.M. Gelfand. Wiley, New York,
1975.

[4] A Connes, Non c o m m u t a t i v e d i f f e r e n t i a l geometry.

C h a p t e r I, the Chern c h a r a c t e r in K homology. IHES preprint,
1982.

[5] I.G. Frenkel, Orbital theory for affine Lie algebras. To appear.

[6] I.G. Frenkel and V.G. Kac, Basic r e p r e s e n t a t i o n s of affine Lie

algebras and dual resonance models. Invent. Math., 62 (1980),
23-66.

[7] H. Garland, The arithmetic theory of loop algebras. J. Algebra,

53 (1978), 480-551.

[8] H. Garland, The arithmetic theory of loop groups. Publ. Math.

IHES, 52 (1980), 5-136.

[9] I.M. Gelfand, M.I. Graev, and A.M. Vershik, R e p r e s e n t a t i o n s of

the group of smooth m a p p i n g s of a m a n i f o l d into a compact Lie
group. C o m p o s i t i o Math., 35 (1977), 299-334.

[10] R. Goodman and N. Wallach, Structure and unitary cocycle rep-

resentations of loop groups and the group of d i f f e o m o r p h i s m s of
the circle. To appear.

[11] V.G. Kac, Infinite d i m e n s i o n a l Lie algebras. Birkhauser, 1983.

[12] V.G. Kac and D.H. Peterson, Spin and wedge r e p r e s e n t a t i o n s of

infinite d i m e n s i o n a l Lie algebras and groups. Proc. Nat. Acad.
Sci. USA, 78 (1981), 3308-3312.

[13] G. Kempf, The G r o t h e n d i e c k - C o u s i n complex of an induced rep-

resentation. A d v a n c e d in Math., 29 (1978), 310-396.

[14] J. Lepowsky, C o n s t r u c t i o n of the affine algebra A~ 1) Comm.

Math. Phys., 62 (1978), 43-53.
 168

[15] J. Lepowsky, G e n e r a l i z e d Verma modules, loop space cohomology,

and M a c d o n a l d type identities. Ann. Sci. Ec. Norm. Sup., 1 2
(1979), 169-234.

[16] I.G. Macdonald, Affine Root Systems and the D e d e k i n d H-function.

Invent. Math., 15 (1972), 91-143.

[17] J. Milnor, On infinite d i m e n s i o n a l Lie groups. To appear.

[18] A.N. Pressley and G.B. Segal, Loop groups. Oxford Univ. Press,
to appear.

[19] G.B. Segal, Unitary r e p r e s e n t a t i o n s of some infinite d i m e n s i o n a l

groups. Comm. Mlath. Phys., 80 (1981), 301-342.

[20] G.B. Segal and G. Wilson, Loop groups and equations of KdV type.
Publ. Math. IHES, to appear.

[21] B. Simon, Trace ideals and their applications. London Math.

Soc. Lecture Notes NO.35, C a m b r i d g e Univ. Press, 1979.
 Some Recent Results in C o m p l e x M a n i f o l d Theory Related to
V a n i s h i n g Theorems for the S e m i p o s i t i v e Case

Yum-Tong Siu
Department of M a t h e m a t i c s
Harvard University
Cambridge, MA 02138, U.S.A.

To put this survey in the proper perspective, let me first make

some rather general remarks. To study c o m p l e x manifolds or in general
complex spaces, one works with holomorphic objects like holomorphic
maps, holomorphic functions, holomorphic vector bundles and their
holomorphic sections. One has to construct such objects. For
example, to p r o v e that a c o m p l e x manifold is b i h o l o m o r p h i c to £ n one
tries to p r o d u c e n suitable holomorphic functions. To p r o v e that a
complex manifold is b i h o l o m o r p h i c to Pn' one tries to produce n+l good
holomorphic sections of a s u i t a b l e holomorphic line bundle. To p r o v e
that two complex manifolds are biholomorphic, one tries to produce a
biholomorphic map. How does one p r o d u c e such h o l o m o r p h i c objects? So
far we h a v e m a i n l y the f o l l o w i n g methods:

(i) The method of c o n s t r u c t i n g harmonic objects first and then getting

holomorphic objects from them. An e x a m p l e is the use of the D i r i c h l e t
principle to c o n s t r u c t harmonic functions on open Riemann surfaces and
then obtaining holomorphic functions from them. Examples of the
construction of h a r m o n i c objects are the r e s u l t s of Eel ls-Sampson [i0]
and S a c n s - U h l e n b e c k [40] on the e x i s t e n c e of h a r m o n i c maps. However,
unlike the one-dimensional case, in the higher-dimensional case the
gap between a harmonic object and a holomorphic object is v e r y wide
and, except for some special cases, is i m p o s s i b l e to bridge.

(2) The method of using the vanishing theorem of Kodaira to construct

holomorphic sections of high powers of p o s i t i v e line b u n d l e s [24].

(3) Grauert's bumping technique to c o n s t r u c t holomorphic functions on

strongly pseudoconvex domains [13].
 170

(4) The method of using L 2 estimates of ~ to construct holomorphic

functions on strongly pseudoconvex domains (Morrey [30], Andreotti-
Vesentini [i], K o h n [25], H ~ r m a n d e r [20]).

These methods produce holomorphic objects from scratch so to

speak. There are also other methods like t h e u s e of T h e o r e m s A and B
of C a r t a n - S e r r e to c o n s t r u c t holomorphic objects, but one has to h a v e
a Stein manifold (i.e. a complex submanifold of £n) or a S t e i n space to
apply Theorems A and B and on such manifolds previously existing
global holomorphic functions are e s s e n t i a l for the c o n s t r u c t i o n .

Let me b r i e f l y explain the n o t i o n s of p o s i t i v e line bundles and

strongly pseudoconvex domains and h o w t h e y are r e l a t e d . A holomorphic
line bundle with a Hermitian metric along its fibers is said to be
positive if the c u r v a t u r e form associated to the H e r m i t i a n metric is a
positive-definite quadratic form. A relatively compact domain with
smooth boundary in a complex manifold is said to be strongly
pseudoconvex if it is defined near its boundary by r < 0 for some
smooth function r with nonzero gradient such that the complex Hessian
of r as a H e r m i t i a n form is positive-definite. If L is a Hermitian
ho]omorphic line bundle over a compact complex manifold, then the
set ~ of all vectors of the dual bundle L* of L w h o s e lengths are less
than 1 is a strongly pseudoconvex domain in L* if a n d only if L is
positive. Grauert [14] observed that a holomorphic function on
gives rise to holomorphic sections of powers of L because its k th
coefficient in the power series expansion along the fiber~ of L* is a
section of t h e k t h p o w e r o f L. So producing holomorphic sections of
powers of a positive line bundle is a special case of producing
holomorphic functions of a s t r o n g l y pseudoconvex domain.

In the above methods of producing holomorphic objects some

positive-definite quadratic form is used, be it the c u r v a t u r e form in
the case of a positive line bundle or the complex Hessian of the
defining function in the case of a strongly pseudoconvex domain. In
the method of using harmonic objects to c o n s t r u c t holomorphic objects
no positive-definite quadratic form is used. However, in the h i g h e r -
dimensional case there is a w i d e gap between harmonic and holomorphic
objects and m e t h o d s known up to n o w [45, 46, 47, 51, 52] to b r i d g e the
gap require positive-definiteness of a certain quadratic form coming
 171

from the curvature tensor. This survey talk discusses the situation
when the q u a d r a t i c forms used in p r o d u c i n g holomorphic objects are
only positive semidefinite instead of strictly positive-definite. In
certain c a s e s we m a y e v e n a l l o w certain benign negativity. One may
wonder w h y one should bother to s t u d y the s e m i d e f i n i t e case. There
are a number of reasons. Let me give two here. One is that some
situations are n a t u r a l l y semidefinite, like the s e m i n e g a t i v i t y of the
sectional curvature for a bounded symmetric domain. Another is that
when limits of h o l o m o r p h i c objects are used in p r o o f s (like in the
continuity method), the l i m i t of s t r i c t l y positive definite objects
can only be assumed first to be semidefinite though in the final
result it may turn out to be strictly positive definite. The
semidefinite case is by far m u c h m o r e c o m p l i c a t e d t h a n the d e f i n i t e
case.

In this talk we will survey some recent results concerning

vanishing theorems for the semidefinite case and their applications.
M o r e s p e c i f i c a l l y we w i l l d i s c u s s the f o l l o w i n g three topics:

(i) The construction of h o l o m o r p h i c sections for line bundles with

curvature form not strictly positive or even with bengin negativity
somewhere. An application is a proof of the Grauert-Riemenschneider
conjecture characterizing Moishezon manifolds by s e m i p o s i t i v e line
bundles [49, 50].

(ii) The strong r i g i d i t y of c o m p a c t K~hler m a n i f o l d s with seminegative

curvature, in p a r t i c u l a r the r e s u l t s of J o s t - Y a u [22] and Mok [29] on
the s t r o n g r i g i d i t y of i r r e d u c i b l e compact quotients of p o l y d i s c s .

(iii) Sube] liptic estimates of Kohn's school [26, 6] and their

applications to v a n i s h i n g theorems for s e m i p o s i t i v e bundles.

I. Producing_ S e c t i o n s for S e m i p o s i t i v e Bundles

We w a n t to d i s c u s s how one can p r o d u c e holomorphic sections for

a Hermitian line b u n d l e whose curvature form is o n l y semipositive or
may e v e n be n e g a t i v e somewhere. The o r i g i n a l m o t i v a t i o n for this kind
of study is to prove the so-cal led Grauert-Riemenschneider
 172

conjecture[15, p.277]. Kodaira[24] characterized projective algebraic

manifolds by the e x i s t e n c e of a Hermitian holomorphic line bundle
whose curvature form is p o s i t i v e definite. The c o n j e c t u r e of G r a u e r t -
Riemenschneider attempts to g e n e r a l i z e Kodaira's r e s u l t to the case of
M o i s h e z o n manifolds. A M o i s h z o n m a n i f o l d is a c o m p a c t c o m p l e x m a n i f o l d
with the property that the transcendence degree of its meromorphic
function field equals its c o m p l e x dimension. Moishezon showed [28]
that such m a n i f o l d s are precisely those w h i c h can be transformed into
a projective algebraic m a n i f o l d by p r o p e r modification. The c o n c e p t of
a M o i s h e z o n space is s i m i l a r l y defined.

The c o n j e c t u r e of G r a u e r t - R i e m e n s c h n e i d e r asserts that a c o m p a c t

complex space is Moishezon if there exists on it a torsion-free
coherent analytic sheaf of rank one with a Hermitian metric whose
curvature f o r m is p o s i t i v e definite on an o p e n d e n s e s u b s e t . Here a
Hermitian metric for a s h e a f is d e f i n e d by g o i n g to the linear space
associated to the sheaf and the c u r v a t u r e form is d e f i n e d only on the
set of points where the sheaf is locally free and the space is
regular. The difficulty with the proof of t h e c o n j e c t u r e is h o w to
p r o v e the f o l l o w i n g special case.

Conjecture of Grauert-Riemenschneider. Let M be a compact complex

laanifold w h i c h admits a Hermitian holomorphic line bundle L whose
curvature form is p o s i t i v e definite on an o p e n d e n s e s u b s e t G of M.
Then M is Moishezon.

Since the conjecture of G r a u e r t - R i e m e n s c h n e i d e r was introduced,

a number of other characterizations of Moishezon spaces have been
obtained [38,57,53,12,35] which circumvent the difficulty of proving
the Grauert-Riemenschneider conjecture by stating the
characterizations in such a way that a proof can be o b t a i n e d by using
blow-ups, Kodaira's vanishing and e m b e d d i n g theorems, or L 2 e s t i m a t e s
of ~ for c o m p l e t e K~hler m a n i f o l d s . If the m a n i f o l d M is a s s u m e d to
be K~hler, then R i e m e n s c h n e i d e r [39] observed that Kodaira's original
proof of his vanishing and embedding theorems together with the
identity theorem for solutions of second-order elliptic partial
differential equations [2] a l r e a d y y i e l d s right away the c o n j e c t u r e of
Grauert-Riemenschneider. If the set of points where the c u r v a t u r e form
of L is n o t p o s i t i v e definite is of c o m p l e x dimension z e r o [38] or o n e
 173

[44] or if some additional assumptions are imposed on the e i g e n v a l u e s

of the curvature form of L [47] , the conjecture of Grauert-
Riemenschneider can rather easily be proved. Recently Peternell [33]
used degenerate K~hler metrics to o b t a i n some p a r t i a l results about
the G r a u e r t - R i e m e n s c h n e i d e r conjecture. However, all the a b o v e r e s u l t s
fail to deal with the fundamental question of how to produce in
general holomorphic sections for a line b u n d l e not strictly positive
definite.

Recently a new method of obtaining holomorphic sections for

nonstrictly positive line b u n d l e s was introduced [49]. T h e r e it was
used to give a proof of the conjecture of G r a u e r t - R i e m e n s c h n e i d e r in
the special case where M-G is of measure zero in M. It was later
refined to g i v e a proof of the general case and a stronger version of
the conjecture of Grauert-Riemenschneider [50]. The method imitates
the familiar technique in a n a l y t i c number theory of using the Schwarz
lemma to p r o v e the identical vanishing of a function by e s t i m a t i n g its
order and making it v a n i s h to h i g h order at a s u f f i c i e n t number of
points. Such a technique applied to the holomorphic sections of a
holomorphic line bundle was used by Serre [41] and also iater by
Siegel [43] to obtain an alternative proof of Thimm's theorem [54]
that the transcendence degree of the meromorphic function field of a
compact complex manifold cannot exceed its complex dimension. In
[49, 50] the technique was applied to h ar m o n i c forms with coefficients
in a h o l o m o r p h i c line b u n d l e and its use was coupled with the theorem
of Hirzebruch-Riemann-Roch [19, 3].

We give a more precise brief description of the method of

[49,50]. To make the d e s c r i p t i o n easier to understand, we first impose
the condition that M-G is of measure zero in M. By the theorem of
Hirzebruch-Riemann-Roch (which for the case of a general compact
complex manifold is a c o n s e q u e n c e of the index theorem of A t i y a h -
Singer [3]), ~=o(-l)q dim Hq(M,L k) h ckn for some positive cortstant
c w h e n k is s u f f i c i e n t l y large, w h e r e n is the c o m p l e x dimension of M.
To prove that Lk admits enough holomorphic sections to give
sufficiently many meromorphic functions to m a k e M Moishezon, it
suffices to s h o w that d i m H 0 ( M , L k) > ckn/2 for k sufficiently
large. Thus the problem is reduced to proving that for any given
positive number ~ and for q ~ 1 one has dim Hq(M, L k) ~ ~k n for
 174

k sufficiently large. Give M a Hermitian metric and represent elements

of H q ( M , L k) b y Lk-valued harmonic forms. By using the L 2 estimates of
T one obtains a linear map from the space of h a r m o n i c forms to the
space of c o c y c l e s . Take a lattice of p o i n t s with distances k -I/2 apart
in a s m a l l neighborhood W of M-G. Then one uses the usual technique of
Bochner-Kodaira for the case of a compact Hermitian (not necessarily
K~hler) manifold [16, p.429, (7.14)] and uses the Schwarz lemma to
show that any cocycle coming from a harmonic form via the linear map
and vanishing at a l l the lattice points to an a p p r o p r i a t e fixed order
must vanish identically, otherwise its n o r m is so s m a l l that the ~-
closed form constructed frora it b y u s i n g a partition of unity would
have a norm smaller than that of the harmonic form in its cohomology
class, contradicting the minimality of the norm of a h a r m o n i c form in
its cohomology class. It f o l l o w s that dim H q ( M , L k) is d o m i n a t e d by
a fixed constant times the number of lattice points (which is
comparable to the volume of W t i m e s kn), otherwise there is a nonzero
combination of cocycles coming from a basis of harmonic forms via the
linear map and having the r e q u i r e d vanishing orders. Since M-G is of
measure zero in M, w e c a n m a k e the volume o f W as s m a l l as we please
and therefore can choose E smaller than any prescribed positive
number after making k sufficiently large. The reason why such a
lattice of points is c h o s e n is that the pointwise square norm of a
local holomorphic section o f L k is o f t h e f o r m IfI2e -k% , where f is
holormorphic function and ~ is a plurisubharmonic function
corresponding to the Hermitian metric of L. The factor e -k# is a n
obstacle to a p p l y i n g the S c h w a r z lemma. To o v e r c o m e this obstacle, one
chooses a local trivialization of L so t h a t ¢ as w e l l as de vanishes
at a point. The on the ball of r a d i u s k -I/2 centered at that point,
e -k# is bounded below from zero and from above by constants
independent of k. T h e reason why one uses cocycles instead of d e a l i n g
directly with harmonic forms is that the Schwarz lemma is a
consequence of t h e log p l u r i s u b h a r m o n i c property of the a b s o l u t e value
of holomorphic functions and there is no corresponding Schwarz lemma
for harmonic forms.

The method outlined above can be refined in t h e f o l l o w i n g w a y so

that it works in the general case where G is only assumed to be
nonempty. Let R be the set of points of M where the smallest
eigenvalue of the curvature form ~ of L does not exceed some
 175

positive number k . For every point 0 in R one can choose a

coordinate polydisc D with coordinates Zl,...,z n c e n t e r e d at O and can
choose a global trivialization of L o v e r D s u c h t h a t for s o m e c o n s t a n t
C > 0

n
I $(Pi ) - $(P2) I <_ C ( x Izi(Pi) - zi(P 2) I2 + ~ Izi(Pi) - zi(P212)
i:2

for PI' P2 in D. Moreover, both C and the polyradius of D can be

chosen to be the same for all points O of R. Cover R by a finite
number of such coordinate polydiscs so that for some constant m
depending only on n no more than m of them intersect. Then one
chooses the lattice points so that they are ( k k ) -I/2 apart along the
z I direction but are k -I/2 apart alon 9 the directions of z2,...,z n.
Now the total number of lattice points is no more than a constant
times k kn times the volume of R. By c h o o s i n g ~ sufficiently small,
we conclude that for any given positive number ~ and for q > i one
has d i m H q ( M , L k) < ~k n and therefore d i m H 0 ( M , L k) is n o l e s s than
ck n for some positive number c when k is s u f f i c i e n t l y large. Thus we
have the following theorem [49, 50].

Theorem 1. Let M be a compact complex manifold and L be a

Hermitian holomorphic line bundle over M whose curvature form is
everzwher 9 semipositive and is s t r i c t l Z p o s i t i v e at s o m e point. Then M
is a M o i s h e z o n manifold.

By the result of G r a u e r t - R i e m e n s c h n e i d e r , one h a s as a c o r o l l a r y

the vanishing of H q ( M , L K M) f o r q > I , where K M is t h e canonical line
bundle of M.

In c o n j u n c t i o n with the c h a r a c t e r i z a t i o n of M o i s h e z o n manifolds,

I would like to m e n t i o n the r e c e n t result of P e t e r n e l l [34] that a 3-
dimensional Moishezon manifold is p r o j e c t i v e algebraic if i n it n o
positive integral linear combination of irreducible curves is
homologous to zero. Together with Hironaka's example [18 andl7,
p.443] of a 3 - d i m e n s i o n a l non-projective-algebraic Moishezon manifold
Peternell's result gives us the complete picture of the difference
between projective-algebraic threefolds and Moishezon threefolds.

The noncompact analog of Theorem 1 is the following conjecture

 176

which is s t i l l open.

Conjecture. Let fl be a relatively compact open subset of a complex

manifold such that its b o u n d a r y is w e a k l y pseudoconvex at e v e r y point
and is strictly pseudoconvex at some point P. T h e n there exists a
holomorphic function on ~ going to i n f i n i t y along some sequence in fl
approaching P.

Theorem 1 corresponds to the case where is the set of

vectors in t h e d u a l bundle of L w i t h length < i.

The method used in t h e p r o o f of T h e o r e m 1 c a n be f u r t h e r refined

to y i e l d results about the e x i s t e n c e of holomorphic sections for line
bundles whose curvature form is a l l o w e d to be n e g a t i v e somewhere [50].
An example of s u c h r e s u l t s is the following.

Theorem 2__
t F__o~ ~ y e r y positive integer n there exists a constant Cn
de__2mgnding o n l Z o n n w i t h the following ~r_~o~r_~ty.- L e t M be a compact
K~hler manifold of complex dimension n and L be a Hermitian line
bundle over M. L e t G b e a n o ~ _ n subset of M and a, b ~[ positive
numbers such that the curvature form of L admits a as a l o w e r bound
at ever Z point of G a n d admits - b as a l o w e r bound at ever Z point of
M-G. Assume that

Cn (i + log+(b/a)) n (b2/a)n(volume of M-G) < Cl(L)n

where Cl(L) is t h e f i r s t Chern class of L. T h e n dim H0(M, L k) is >

Cl(L)n k n / 2 ( n !) for k sufficientl~ large.

When the m e t r i c of the m a n i f o l d is H e r m i t i a n instead of K~hler,

there is a c o r r e s p o n d i n g theorem with the c o n s t a n t C n depending o n the
torsion of the Hermitian metric. The inequality in the assumption of
Theorem 2 is not natural. There should be better and more natural
formulations of this kind of r e s u l t s .

We describe below the refinement needed to get a proof of

Theorem 2. T h e method described above can readily yield Theorem 2 if
we allow the constant C n to d e p e n d o n M, b u t then Theorem 2 would be
far less interesting. The r e a s o n w h y the a b o v e method can only yield a
 177

C n depending o n M is t h a t in c o n s t r u c t i n g a correspondence from the

space of h a r m o n i c forms to the space of cocycles, besides solving the
equations, one has to use a partition of unity and the constants
obtained in the process depend very heavily on the manifold M. T o
solve this problem, we make use of the e s t i m a t e of the (0,1)-covariant
derivative of the harmonic form from the Bochner-Kodaira formula. We
locally solve with estimates the inhomogeneous Y equations with
of the coefficients of harmonic form on one side so that the
differences between the coefficients of the harmonic form and the
solutions are holomorphic and then apply the Schwarz lemma to the
differences. This way we avoid passing from the Dolbeault cohomology
to the C e c h cohomology and c a n m a k e the c o n s t a n t C n independent of M.

If. S t r o n g Rigidity of S e m i n e g a t i v e l y Curved Compact K~dller M a n i f o l d s

A compact K~hler manifold is s a i d to be stronq~ Z rigid if a n y

other K~hler manifold homotopic to it is biholomorphic or
antibiholomorphic to it. Strong rigidity can be regarded as the
complex analog of Mostow's strong rigidity [31]. Compact K~hler
manifolds M with curvature tensor negative in a s u i t a b l e sense are
known to be strongly rigid [45, 46, 47]. The way to obtain the strong
rigidity of M is to c o n s i d e r a harmonic map f to M from the compact
K~ller manifold N homotopic to M "which is a h o m o t o p y equivalence. The
existence of s u c h a h a r m o n i c m a p is g u a r a n t e e d b y the result of E e l l s -
Sampson[10] because of the n o n p o s i t i v i t y of the s e c t i o n a l curvature of
M. As a s e c t i o n of t h e t e n s o r product of the bundle of (0,1)-forms of
N and the pullback under f of the (l,0)-tangent bundle of M, ~f is
harmonic. By using the technique of Bochner-Kodaira we conclude that
either ~f or Yf vanishes because of the curvature condition of M.
The reason why we can only conclude the vanishing of either ~f or
~f is that the curvature term from the Bochner-Kodaira formula is
homogeneous of degree two in ~f and of degree two in ~f because
it comes from pulling back of the curvature tensor of M under f.
Actually the B o c h n e r - K o d a i r a technique is a p p l i e d to the image of Yf
under the complexified version of the Hodge star operator. In o t h e r
w o r d s w e are a p p l y i n g the B o c h n e r - K o d a i r a technique to the d u a l of the
bundle. That is the reason why the curvature tensor of M has to be
assumed negative instead of positive and also that is the reason why
the Ricci tensor of N does not enter the picture. The most general
 178

formulation of this kind of results on strong rigidity is the

following theorem [47].

Theorem 3. A compact K~hler manifold M of complex dimension n ~[

stronq! ~ rigid if t h e r e exists a positive number p less than n with
the following ~roBerties: (i) The bundle of (l,0)-forms on M is
positive semidefinite in the s e n s e of N a k a n o [32] a n d the b u n d l e of
(p,0)-forms on M !~ positive definite i__nnth___~es e n s e 2 ~ ~a__~k~2 [32].
(ii) At a n [ p o i n t of M the c o m p l e x tangent s p a c e of M d o e s n o t c o n t a i n
two n o n t r i v i a l orthogonal subspaces with combined dimension exceeding
p such that the b i s e c t i o n a l curvature of M in the d i r e c t i o n of two
vectors one from each subspace vanishes.

As a c o r o l l a r y any compact quotient of an i r r e d u c i b l e bounded

symmetric domain of c o m p l e x dimension at least twc is s t r o n g l g rigid,
because we have the following table giving the c o m p l e x dimension and
the smallest p satisfying the assumptions of Theorem 3 for each
bounded symmetric domain.

Type Complex Dimension Smallest p

Ira,n mn (m-L) (n-1)+l
II n n(n-l)/2 (n-2) (n-3)/2 +i
III n n(n+l)/2 n(n-l)/2 +i
IV n n 2
V 16 6
VI 27 ii

The v a l u e s of the smallest p for the two e x c e p t i o n a l domains

were computed by Zhong [58].

This method also yields the ho ] omorphicity or

antiholomrorphicity of any h a r m o n i c map from a compact K~hler manifold
into M whose rank o v e r ]]9 i s > 2 p +I at some point [47].

This method can be r e g a r d e d as an a p p l i c a t i o n of the q u a s i l i n e a r

version of K o d a i r a ' s vanishing theorem. Though strict negativity of
the c u r v a t u r e tensor is not needed for this method, this method should
be c o n s i d e r e d as c o r r e s p o n d i n g to the s t r i c t l y definite case rather
 179

than the semidefinite case of the vanishing theorem, because through

the u s e of the c o m p l e x i f i e d Hodge star operator the vanishing required
is in c o d i m e n s i o n one rather than in d i m e n s i o n one.

The only case of c o m p a c t quotients of b o u n d e d symmetric domains

which are expected to enjoy the property of strong rigidity as
suggested by Mostow's result [31] and which are not covered by the
results of [47] is t h e c a s e of an irreducible compact quotient of a
polydisc of complex dimension at least two. This remaining case
corresponds to the semidefinite case of the vanishing theorem. Jost-
Yau [21] first considered this remaining case and obtained some
partial results. Recently Jost-Yau [22] and M o k [29] c o m p l e t e l y solved
this case. We would l i k e to s k e t c h a s l i g h t l y more streamlined version
of the proof in [29]. F i r s t we make some general observations about
the application of the Bochner-Kodaira technique to the case of a
compact quotient of a polydisc and discuss a simple but rather
surprising theorem about the existence of holomorphic maps from
compact K~hler manifolds into compact hyperbolic Riemann surfaces.

Let f be a harmonic map from a compact K~hler manifold M to a

compact quotient Q of a polydisc Dn of complex dimension n. The
following conclusions are immediate from the Bochner-Kodaira
technique.

(i) f is p l u r i h a r m o n i c in t h e sense that the restriction of f to any

local complex curve in M is h a r m o n i c .

(ii) ~fi ^ ~ f--~ is z e r o for 1 <__i <__n, w h e r e f i is t h e i th component

of f when it is e x p r e s s e d in t e r m s of local coordinates along the n
component discs.

From conclusion (ii) above it follows that the pullback f * T ~ '0

under f of the (l,0)-tangent bundle T I'0 of Q can be endowed with the
structure of a h o l o m o r p h i c vector bundle in the f o l l o w i n g way. A l o c a l
section is d e f i n e d to be holomorphic if i t s c o v a r i a n t derivative in
the (0,i) direction is identically zero. This c a n be d o n e because (ii)
implies that the (0,i) covariant exterior differentiation composed
with itself is identically zero, which is the i n t e g r a b i l i t y condition
for such a holomorphic vector bundle structure. The same argument can
 180

be applied to the Dullback f*T Q

0'I under f of the (0,1)-tangent bundle
T~'lof Q to g i v e it a h o l o m o r p h i c vector bundle structure. Moreover,
if e v e r y element of the f u n d a m e n t a l group of Q m a p s e a c h i n d i v i d u a l
component disc of D n to itelf, then each of these two holomorphic
vector bundles are the d i r e c t sum of the n h o l o m o r p h i c line bundles
which are the p u l l b a c k s of the line s u b b u n d l e s of the tangent bundle of
Q defined by the d i r e c t i o n s of the i n d i v i d u a l component discs. In such
a case l e t L i be the line subbundles of f*T I'0Q a n d L i' be the line
subbundles of f*T~'t

Because of c o n c l u s i o n (i) ~f is a h o l o m o r p h i c section of

f,T~,0 × ~i and 3f is a h o l o m o r p h i c section of f*T$'l × ~I, where ~i
is the bundle of holomorphic 1-forms on M. Assume that every element
of the fundamental group of Q maps each individual component disc of
D n to itself. Then for each fixed 1 <i< n,

(iii) ~fi is a holomorphic section of Li O ~ and afi is a

holomorphic section of ~ O ~.

For any local holomorphic section s i of the dual bundle of L i,

si ~fz is a (local) holomorphic 1-form on M whose exterior
derivative equals its product with some 1-form. By the theorem of
Frobenius, near points where ~fi does not vanish we have a
holomorphic family of local complex submanifolds of complex
codimension one whose tangent spaces annihilate afi. Such a
holomorphic foliation of codimension one defined by the kernel of
af z exists also at points where afz can be divided by a local
holomorphic function to g i v e a nowhere zero h o l o m o r p h i c local section
of Li O ~- If in a d d i t i o n the rank of df i is two over ~ at the
points under consideration, because of (ii) the local leaves of the
holomorphic foliation agree with the fibers of the locally defined map
--w-

fi. T h e same consideration c a n be a p p l i e d to ~fz. A l s o because of

(ii) the holomorphic foliation defined by the kernel of ~fl agrees
with the h o l o m o r p h i c foliation of ~fl at points where both ~fi and
~fl can be d i v i d e d by local holomorphic functions to g i v e nowhere zero
holomorphic local sections of L i ~ ~ and L [ ~ ~ respectively. These
rather straightforward discussions lead us immediately to the
following theorem [48].
 181

Theorem 4. L e t M be a compact K~hler manifold and R be a compact

hyperbolic Riemann surface such that there exists a continuous map f
from M t_~o R w h i c h is n o n z e r o on the second homology. Then there exists
a holomor~hi~ ma_~ g from M into a compac~ h~perbolic Riemann
surface S and a harrnonic ma[ h from S ~o R such t h a t hog is h o m o t o p i [
to f.

The Riemann surface S is constructed from the holomorphic

foliation described above in the following way. By the result of
Eells-Sampson we can assume without loss of generality that f is
harmonic and therefore real-analytic. Let Z be the complex subvariety
of complex codimension h 2 in M c o n s i s t i n g of a l l points where either
or ~f c a n n o t be d i v i d e d by a n y local holomorphic function to g i v e a
nowhere zero holomorphic local section of the tensor product of ~
with the pullback under f of t h e (I,0) or (0,i) tangent bundle of R.
Let V be the set of points of M where the rank of df over ~ is <
2. O n M - Z we have a holomorphic foliation described above with the
property that whenever a leaf of the foliation has a point in c o m m o n
with M-V, the leaf agrees with the real-codimension-two branch of the
fiber of f passing through that point and therefore c a n be e x t e n d e d to
a complex-analytic subvariety of c o d i m e n s i o n one in M. Because of the
K~hler metric of M, by using Bishop's theorem [4] on the limit of
subvarieties of bounded volume and by passing to limit, we conclude
that every leaf of the holomorphic foliation can be extended to a
complex-analytic subvariety of codimension one in M. Since Z is of
complex codimension > 2 in M, by u s i n g the t h e o r e m of R e m m e r t - S t e i n on
extending subvarieties [37] we conclude that M is covered by the
holomorphic family of subvarieties consisting of the e x t e n s i o n s of the
leaves of the holomorphic foliation. The Rieraann surface S is now
obtained as the nonsingular model of the quotient of M w h o s e points
are the branches of the extensions of t h e leaves of t h e h o l o m o r p h i c
foliation.

The rather surprising aspect of Theorem 4 is that from the

existence of a continuous map from a compact K~hler manifold to a
compact hyperbolic Riemann surface nonzero on the second homology we
can conclude the existence of a n o n t r i v i a l holomorphic map from the
K~hler manifold to a c o m p a c t hyperbolic Riemann surface. In p a r t i c u l a r
by going'to the respective universal covers we obtain a nontrivial
 182

bounded holomorphic function on the univeral cover of the K~hler

manifold. So far there is no known general method of constructing
bounded holomorphic functions on complex manifolds which are expected
to a d m i t a large number of b o u n d e d holomorphic functions, such as the
universal cover of compact K~hler manifolds of negative curvature.
Here to conclude the existence of a nontrivial bounded holomorphic
function we do not use any curvature property of the compact K~hler
manifold. Instead the existence of a continuous map to a compact
hyperbolic Riemann surface is used. Since until now there is no
general way of constructing nontrivial bounded holomorphic functions,
this could only mean that the existence of the k i n d of c o n t i n u o u s map
we want is rather rare and if such a continuous map exists, its
existence would be r a t h e r difficult to e s t a b l i s h . Even for negatively
curved compact K~hler manifolds in general we do not expect such
continuous maps to exist. As a matter of fact, for compact quotients
of a b a l l of c o m p l e x dimension at l e a s t t w o the o n l y known examples so
far that admit nontrivial holomorphic maps into any compact hyperbolic
Riemann surface are the ones constructed by Livn~ [27] by taking
branched covers of certain elliptic surfaces. It is n o t k n o w n whether
in o t h e r d i m e n s i o n s there are similar examples of m a p s b e t w e e n compact
quotients of b a l l s of d i f f e r e n t dimensions besides the o b v i o u s ones.

Problem. Suppose 1 < m < n are integers. Let M a n d N be r e s p e c t i v e l y

compact quotients of the b a l l s of c o m p l e x dimensions m a n d n.
(a) Is it true that there exists no surjective holomorphic map from N
to M?
(b) Is it true that every holomorphic embedding of M in N m u s t have a
totally geodesic image?

Yau conjectured that Problem (b) should be a consequence of

uniquenss results for proper holomorphic maps between balls of
different dimensions. For n ~ 3 Webster [56] showed that the only
proper holomorphic maps f r o m the n - b a l l to the (n+l)-ball C 3 u p to the
boundary are the obvious ones. Faran [ii] showed that, up to
automorphisms of the two balls, there are only four proper holomorphic
maps from the 2-bal 1 to the 3-bal 1 C 3 up to the boundary
Unfortunately until now there are no general results about proper
holomorphic maps between balls of d i f f e r e n t dimensions without any
known boundary regularity. In o u r case the proper map, though without
 183

any known boundary regularity, has the additional property that it

comes from maps between compact quotients. Hopefully this additional
property may be used instead of b o u n d a r y regularity.

We now introduce the theorem on the strong rigidity of

irreducible compact quotients of p o l y d i s c s and sketch its proof.

Theorem 5 (Jost-Yau [22] and Mok [29]). S u p p o s e Q is a n irreducible

colnp_act ~ u ~ t i e n t of an n-disc D n with n h 2, M is a c o m p a c t K~hler
manifold, and f is a harmonic ma~ from M t__oo Q w h i c h is a homotopy
equivalence. Let M be the universal cover of M and F: M + D n with
components (F 1 ..... F n) b e i n d u c e d b ~ F. T h e n for each 1 ~ i ! n, F i i__{s
either holomorphic o_[r a n t i h o l o m o r p h i c .

Here an irreducible quotient means one that cannot be d e c o m p o s e d

as a p r o d u c t of two lower-dimensional quotients of polydiscs. For the
proof of t h i s theorem, by replacing both M and Q by finite covers, we
can assume without loss of g e n e r a l i t y that the fundamental group of Q
is a product of n groups GI,...,Gn, each of which is a (nondiscrete)
subgroup of the automorphism group of the 1-dimensional disc D. We
regard aF i and aF i as holomorphic sections of the holomorphic
vector bundles on M described above (rather as (l,0)-forms on M). A
consequence of the irreducibility of Q is that for each 1 ! i ! n
every orbit o f G i is d e n s e in D. W e h a v e to s h o w that for every 1 < i<n
either aF i o r ~ F 1 vanishes identical ly o n M. W i t h o u t loss of
generality we a s s u m e t h a t the a s s e r t i o n fails for i = 1 and try to g e t
a contradiction. Since f is a homotopy equivalence, the length of
8F 1 a n d t h e length of a F-l-cannot a g r e e at every point. Without loss
of g e n e r a l i t y we can assume that the length of a F 1 is g r e a t e r than
the length of a F 1 at some point. By (ii) and (iii) we can write
~F 1 = g aF
--f so that l o c a l ly g is t h e p r o d u c t of a nowhere zero smooth
function and a meromorphic function. Thus the pole-set V of g is a
complex-analytic hypersurface in M if it is n o n e m p t y . The pole-set V
cannot be empty, otherwise by considering the Laplacian of the log of
the absolute value h of g we get a contradiction at a maximum point of
h. Since f is a homotopy equivalence, the real rank of f on the
regular points of V must be precisely 2n - 2, otherwise the homology
class represented by V would be m a p p e d to z e r o b y f. Let p:M + M be
the projection of the universal cover and q:D n + D be the p r o j e c t i o n
 184

onto the first component. Because of the holomorphic foliation

discussed above the function hop on M must be constant along the
components of the fibers of FI: M ÷ D. T h e proper closed subset
F(p-I(v)) of D n contains an entire fiber of q w h e n e v e r it c o n t a i n s
one of its points. It follows that q(F(p-I(v))) is a proper closed
subset of D w h i c h is i n v a r i a n t under the group G I. T h i s contradicts
the density of e v e r y orbit of G 1 in D.

Mok [29] also showed that for any harmonic map from a compact
K~hler manifold to an irreducible compact quotient of the n-disc
(n h 2) w i t h real rank 2n s o m e w h e r e , each of t h e n c o m p o n e n t s of t h e
map between the universal covers induced by it is either holomorphic
or antiholomorphic.

Ill. Vanishing Theorems Obtained b~ Subelliptic Estimates

So far all the vanishing theorems for bundles with curvature

conditons make use of the pointwise property of the curvature form.
The recent theory of subel liptic multipliers developed by Kohn,
Catlin, and others [26, 5, 6] makes it p o s s i b l e to g e t vanishing thoerems
based on the local property of the curvature form when the curvature
form is semidefinite. Kohn developed his theory to deal with the
question of boundary regularity for solutions of the ~ equation in
the c a s e of a w e a k l y pseudoconvex boundary.

Let ~ be an open subset of £n whose boundary is smooth and

weakly pseudoconvex at a b o u n d a r y point x 0. L e t 1 ~q ~n be an integer.
We say that a subelliptic estimate holds for (0,q)-forms at x 0 if
there exists a neighborhood U of x 0 and constants ~ > 0 and C > 0
such that

Ir~fl 2 < c l l l ~ t l 2 + IIY%ll 2 , 11~tl21

C

for all smooth (0,q)-form ~ on U{~[ with compact support belonging to

the domain of ~*, where II II m e a n s the L 2 norm and II II~ m e a n s the
Sobolev E-norm. In o r d e r to o b t a i n subelliptic estimates Kohn intro-
duced the concept of a subelliptic multiplier. A smooth function f on
U is said to be a subelliptic multiplier if there exist positive ¢
 185

a n d C so t h a t

IIf~l12 < ctlt~-~tl2 + 11-#'9112 + 11~112)

E

for all 9 . The subelliptic multipliers form an ideal lq. Let cij
( l ! i,j ~n-l) be the L e v i f o r m of t h e b o u n d a r y of ~ n e a r x in t e r m s of
an o r t h o n o r m a l frame field of (i,0) vectors tangential to the b o u n d a r y
of ~ . The starting point of K o h n ' s theory is the f o l l o w i n g results
concerning the ideal lq of s u b e l l i p t i c multipliers. For notational
simplicity we describe the case q = 1 and the general c a s e is s i m i l a r .

(i) A s m o o t h function r with nonzero gradient whose zero-set is the

boundary of ~ belongs to I I.

(ii) The determinant of the matrix (cij)l < i , j <__n_l belongs to I I.

(iii) W h e n e v e r fl,...,fk b e l o n g t o If, the determinant formed in the

following way belongs to I I. T h e i th c o l u m n consists of the c o m p o n e n t s
of ~fi i n t e r m s of the frame field of (I,0) vectors tangential to
the boundary of ~ The other n-l-k columns are any n-l-k columns of
the matrix (cij)l < i , j < n - l "

(iv) I 1 equals to its real radical in t h e s e n s e that if f b e l o n g s to

I 1 and g is a smooth function with Igl m £ Ifl for some positive
integer m, then g also belongs to I I.

Kohn [26] s h o w e d that if t h e b o u n d a r y of ~ is r e a l - a n a l y t i c

near x 0 and contains no local complex-analytic subvariety of complex
dimension q, then the constant function 1 belongs to the ideal Iq o f
subelliptic multipliers and as a consequence a subelliptic estimate
for (0,q)-forms holds a t x 0. ( D i e d e r i c h - F o r n a e s s [9] c o n t r i b u t e d to
the formulation of the a s s u m p t i o n s in Kohn's result.) Recently Catlin
[5,6] carried out the investigation for the case of smooth boundary
and showed that a subelliptic estimate for (0,1)-forms holds at x 0 if
and only if t h e b o u n d a r y of ~ is of f i n i t e type a t x 0 in t h e s e n s e
of D'Angelo [7,8]. (Similar statements hold for subelliptic estimates
for (0,q)-forms.l D'Angelo's definition of finite type is as follows.
The boundary of ~ is of t y p e < t at x 0 if f o r e v e r y holomorphic map
h = (hl,...,h n) from the open 1-dimensional disc D to ~n with h(0) =
 186

x 0 the vanishing order of roh at 0 does not exceed t times the

minimum of the vanishing orders of hl,...,h n at 0. A t every point x of
the boundary of ~ let t(x) be the smallest number such that the
boundary of £ a t x is o f t y p e ~ t(x). D'Angelo showed that t(x) in
general is not upper semicontinuous, but satisfies t(x) ~ t ( x 0 ) n - i / 2 n - 2
fo~ x near x0 . The order E in the subelliptic estimate at x 0
is e x p e c t e d to be the reciprocal of the m a x i m u m of t(x) for x n e a r x 0.
Catlin's result showed that ¢ cannot be bigger than the expected
number b u t he c a n so far o n l y show that subelliptic estimates hold for
2
an ¢ of the order of t(x 0) r a i s e d to the power -t(x0)n

We now study how the subelliptic estimates can be used to get

vanishing theorems. We follow Grauert's approach to v a n i s h i n g theorems
[14]. A n u m b e r of v a n i s h i n g theorems c a n be f o r m u l a t e d from the m e t h o d
of subelliptic estiraates. Some of them can readily be derived by
other means. We illustrate here by an example of such vanishing
theorems. Let M be a compact complex manifold and L be a Hermitian
holomorphic line bundle over M whose curvature form is s e m i p o s i t i v e .
Let V be a holomorphic vector bundle over M. L e t p:L* + M be the
dual bundle o f L. L e t ~ be the open subset of L* c o n s i s t i n g of all
vectors of L* of length < i. If s u b e l l i p t i c estimates for (0,q)-forms
hold for the boundary of ~ at every one of its points, then one
concludes that Hq(£,p*V) is f i n i t e - d i m e n s i o n a l by representing the
cohomology by harmonic forms. It f o l l o w s that Hq(M,V ® L k) v a n i s h e s
for k sufficiently large, because the k th c o e f f i c i e n t in t h e power
series expansion in the fiber coordinate of L* of a l o c a l holomorphic
function defined near a point in t h e zero-section o f L* is a local
section of L k.

When the Hermitian metric of L is real-analytic, by Kohn's

result subelliptic estimates for (0,q)-forms hold if the boundary of
contains no local q-dimensional complex-analytic subvariety. If t h e r e
is s u c h a subvariety, its p r o j e c t i o n under p is a local q-dimensional
subvariety W with the property that with respect to some local
trivialization of L the Hermitian metric of L is represented by a
function which is c o n s t a n t o n W. If we g i v e M a Hermitian metric, then
all covariant derivatives of the curvature form of L along the
directions of W m u s t vanish. Hence we have the following theorem.
 187

Theorem 6. L e t M be a compact com[lex manifold with a Hermitian

metric, L a holomorphic line bundle over M with a real-analytic
Hermitian metric, and V a holomorphic yector bundle over M. Let 0 be
the curvature form of L. Let q be a positive integer. Sup~gse 0
i~s p o s i t i v e semidefinite and suppose at ever Z point x 2~ M the
following is true. If E is a q - d i m e n s i o n a l comple ~ linear subspace
of the s~ace all (l,0)-vectors at x such that the restriction o__ff 0
t_oo E ×E is zero (where E is the c o m p l e x conjugate o f E), then for some
positive integer m the m th covariant derivative of 0 evaluated at
some m+2 vectors from E and E is not zero. Then Hq(M,V O L k) = 0 for k
sufficientl~ large.

By using Catlin's result [6] for weakly pseudoconvex smooth

boundary, one can drop the real-analytic assumption on the Hermitian
metric of L. This kind of result tells us that in the case of a
semipositive line b u n d l e we can s t i l l get v a n i s h i n g of the c o h o m o l o g y
if the d e r i v a t i v e s of the c u r v a t u r e form satisfy certain nondegeneracy
conditions. Similar theorems can be formulated for h o l o m o r p h i c vector
bundles and noncompact pseudoconvex manifolds. When q = i, Theorem 5
c a n be p r o v e d by u s i n g the m e t h o d of p r o d u c i n g holomorphic sections
for semipositive line bundles described above and Grauert's criterion
of a m p l e n e s s [14, p.347, Lemma] to show that the line b u n d l e L must be
ample. Though for lack of known examples there is no a p p l i c a t i o n yet
for the kind of v a n i s h i n g theorems derived from s u b e l l i p t i c estimates,
hopefully in the future this a p p r o a c h may turn out to be fruitful.

We w o u l d like to r e m a r k that on c o m p a c t projective algebraic

manifolds there is a n o t h e r kind of v a n i s h i n g theorems motivated by
Seshadri's criterion of ampleness [42, p.549] and obtained by
Ramanujam [36] , K a w a m a t a [23] , a n d Viehweg [55] for line bundles
satisfying conditions weaker than ampleness. An e x a m p l e of such a kind
of vanishing theorems is the following. If L is a h o l o m o r p h i c line
bundle over a compact projective algebraic manifold M of complex
dimension n such that Cl(L)n > 0 and Cl(LIC) > 0 for every complex-
analytic curve C in M, then H q ( M , L k K M) = 0 for q > i, where Cl(-)
denotes the first C h e r n c l a s s and K M denotes the c a n o n i c a l line b u n d l e
of M. The assumptions involved are weaker than local curvature
conditions. However, such results apply only to the projective
algebraic case.
 188

References

i. A. Andreotti & E. Vesentini, Carleman estimates for the Laplace-

Beltrami equation on c o m p l e x manifol~]s, Inst. Hautes Etudes Sci. Publ.
Math. 25 (1965), 91-130.

2. N. Aronszajn, A unique continuation theorem for solutions of

elliptic partial differentia] equations or inequalities of second
order, J. Math. Pures AppI. 36 (1957) 235-249.

3. M. F. A t i y a h & I. M. S i n g e r , The index of e l l i p t i c operators III,

Ann. of Math. 87 (1968) 546-604.

4. E. Bishop, Conditions for the a n a l y t i c i t y of c e r t a i n sets, Michigan

Math. J. ii (1964) 289-304.

5. D. C a t l i n , Necessary conditions for subellipticity of the ~-

Neumann problem, Ann. of Math. 117 (1983), 147-171.

6. D. C a t l i n , in p r e p a r a t i o n .

7. J. D ' A n g e l o , Subelliptic estimates and failure of s e m i c o n t i n u i t y

for orders of contact, Duke Math. J. 47 (1980), 955-957.

8. J. D'Angelo, Intersection theory and the ~-Neumann problem, Proc.

Syrup. P u r e Math. 41 (1984), 51-58

9. K. D i e d e r i c h & J. E. Fornaess, Pseudoconvex domains with real-analytic

boundary, Ann. of Math. 107 (1978), 371-384.

i0. J. E e l ls & J. H. Sampson, Harmonic mappings of Riemannian

manifolds, Amer. J. Math. 86 (1964), 109-160.

Ii. J. F a r a n , Maps f r o m the t w o - b a l l to the t h r e e - b a l l , Invent. Math.

68 (1982), 441-475
 189

12. R. Frankel, A differential geometric criterion for Moi~ezon

spaces, Math. Ann. 241 (1979), 107-112; Erratum, Math. Ann. 252
(1980), 259-260.

13. H. G r a u e r t , On Levi's problem and the imbedding of real-analytic

manifolds, Ann. of Math. 68 (1958), 460-472.

14. H. Grauert, Uber Modifikation und exzeptionel le analytische

Mengen, Math. Ann. 146 (1962), 331-368.

15. H. Grauert & O. Riemenschneider, Verschwindungss~tze fur

analytische Kohomologiegruppen auf komplexen R~umen, Invent. Math. ii
(1970) 263-292.

16. P. Griffiths, The extension problem in complex analysis If:

embeddings with positive normal bundle, Amer. J. Math. 88 (1966), 366-
446.

17. R. Hartshorne, A!9_ebraic G e o m e t r Z, New York: Springer-Verlag

1977.

18. H. H i r o n a k a , On the theory of b i r a t i o n a l blowing-up, Ph.D. thesis,

Harvard (1960).

19. F. H i r z e b r u c h , Topological methods in algebraic geometry. New

York: Springer-Verlag 1966.

20. L. H~rmander, L 2 estimates and existence theorems of the T operator,

Acta Math. 113 (1965}, 89-152.

21. J. Jest & S.-T. Yau, Harmonic m a p p i n gs and K~nler manifolds, Math.
Ann. 262 (1983), 145-166.

22. J. Jost & S.-T. Yau, Strong rigidity theorem for a certain class
of compact complex surfaces.

23. Y. K a w a m a t a , A generalization of K o d a i r a - R a m a n u j a m ' s vanishing

theorem, Math. Ann. 261 (1982), 43-46.
 190

24. K. K o d a i r a , On K ~ h l e r varieties of r e s t r i c t e d type (an i n t r i n s i c

characterization of a l g e b r a i c varieties), Ann. of Math. 60 (1954), 28-
48.

25. J. J. Kohn, Harmonic integrals on strongly pseudoconvex

manifolds, I, II, Ann. of Math. 78 (1963), 112-148; 79 (1964) 450-472.

26. J. J. Kohn, Subellipticity of the ~-Neumann problem on w e a k l y

pseudoconvex domains: Sufficient conditions, Acta Math. 142 (1979),
79-122.

• I
27. R. A. L z v n e , On c e r t a i n covers of the u n i v e r s a l elliptic curve,
Ph.D. thesis, Harvard University 1981.

28. B. Moishezon, On n-dimensional compact varieties with n

algebraically independent meromorphic functions, Amer. Math. So,
Translations 63 {1967), 51-177.

29. N. Mok, The holomorphic or a n t i h o l o m o r p h i c character of h a r m o n i c

maps into irreducible compact quotient of polydiscs.

30. C. B. M o r r e y , The analytic embedding of a b s t r a c t real analytic

manifolds, Ann. of Math. 68 (1958), 159-201.

31. G. D. M o s t o w , Strong rigidity of locally symmetric spaces, Ann.

of Math. Studies 78, Princeton: Princeton University Press 1973.

32. S. Nakano, On c o m p l e x analytic vector bundles, J. Math. Soc. Japan

7 (1955), 1-12

33. T. P e t e r n e l l , Fast-positive Geradenbundel auf kompakten komplexen

Manifaltigkeiten.

34. T. Peternell, Algebraic criteria for c o m p a c t complex manifolds.

35. J. H. R a b i n o w i t z , Positivity notions for coherent sheaves over

compact complex spaces, Invent. Math. 62 (1980), 79-87; Erratum,
Invent. Math. 63 (1981), 355.
 191

36. C. P. Ramanujam, Remarks on the Kodaira vanishing theorem, Ind. J.

of Math. 36 (1972), 41-51; Supplement, Ind. J. of M a t h . 38 (1974),
121-124.

37. R. Remmert & K. Stein, Uber die wesentlichen Singularit~ten

analytischer Menger, Math. Ann. 126 (1953), 263-306.

38. O. Rimenschneider, Characterizing Moi~ezon spaces by almost

positive coherent analytic sheaves, Math. Z. 123 (1971), 263-284.

39. O. Riemenschneider, A generalization of Kodaira's embedding

theorem, Math. Ann. 200 (1973), 9 9 - 1 0 2 .

40. J. Sachs & K. Uhlenbeck, The existence of minimal immersion of 2-

spheres, Ann. of Math.ll3 (1981), 1-24.

41. J.-P. Serre, Fonctions automorphes: quelques majorations dans

le cas ou X/G est compact, Semlnalre Caftan 1953-54, 2-1 to 2-9.

42. C. S. Seshadri, Quotient spaces modulo reductive algebraic groups,

Ann. of Math. 95 (1972), 511-556.

43. C. L. Siegel , Meromorphe Funktionen auf kompakten

Mannifaltigkeiten. Nachrichten der Akademie der Wissenschaften in
G~ttingen, Math.-Phys. Klasse 1955, No.4, 71-77.

44. Y.-T. Siu, The Levi problem, Proc. Symp. P u r e Math. 30 (1977),
45-48.

45. Y.-T. Siu, The c o m p l e x - a n a l y t i c i t y of h a r m o n i c maps and the strong

rigidity of c o m p a c t K~hler manifolds, Ann. of Math. 112 (1980), 72-
iii.

46. Y.-T. Siu, Strong rigidity of compact quotients of e x c e p t i o n a l

bounded symmetric domains, Duke Math. J. 48 (1981), 857-871.

47. Y.-T. Siu, The c o m p l e x - a n a l y t i c i t y of h a r m o n i c maps, vanishing and

Lefschetz theorems, J. Diff. Geom. 17 (1982) 55-13.
 192

48. Y.-T. Siu, Lecture at the c o n f e r e n c e in honor of the 60th b i r t h d a y

of G. D. M o s t o w .

49. Y.-T. Siu, A vanishing theorem for semipositive line b u n d l e s over

non-K~hler manifolds. J. Diff. Geom. 1984.

50. Y.-T. Siu, A method of p r o d u c i n g holomorphic sections for line

bundles with curvature of mixed sign. In preparation.

51. Y.-T. Siu, Curvature characterization of hyperquadrics, Duke Math.

J. 47 (19801, 641-654.

52. Y.-T. Siu & S.-T. Yau, Compact K~hler manifolds of positive
bisect i o n a l curvature, Invent. Math. 59 (1980), 189-204.

53. A. J. S o m m e s e , Criteria for quasi-projectivity, Math. Ann. 217

(1975), 247-256.

54. W. Thimm, Uber algebraische Relation zwischen meromorphen

Funcktionen abgeschlossenen R~umen. Thesis, K~nigsberg, 1939, 44 pp.

55. E. V i e h w e g , Vanishing theorems, J. r e i n e angew. Math. 335 (1982),

1-8.

56. S. W e b s t e r , On m a p p i n g an n - b a l l into an n + l ) - b a l 1 in c o m p l e x
space, Pacific J. Math. 81 (1979), 267-272.

57. R. O. W e l l s , Moishezon spaces a n d the K o d a i r a embedding theorem,

Value Distribution Theory, Part A, Marcel Dekker, N.Y. 1974, pp.29-
42.

58. J.-q. Zhong, The degree of strong nondegeneracy of the b i s e c t i o n a l

curvature of exceptional bounded symmetric domains, Proc. 1981
Hangzhou Conference on S e v e r a l Complex Variables, Birkhauser 1984, pp.
127-139.
 GROUPS AND GROUP FUNCTORS ATTACHED TO K A C - M O O D Y DATA

Jacques Tits
CollAge de F r a n c e
11PI Marcelin Berthelot
75231 Paris Cedex 05

1. The finite-dimenSional complex semi-simple Lie algebras.

To start with, let us r e c a l l the c l a s s i f i c a t i o n , due to W. Killing

a n d E. C a r t a n , of all c o m p l e x semi-simple Lie algebras. (The p r e s e n -
tation we a d o p t , for l a t e r purpose, is of c o u r s e not t h a t of t h o s e
authors.) The isomorphism classes of s u c h algebras are in o n e - t o - o n e
correspondence with the systems

(1.1) H , (ei)1~iSZ , (hi) 1~i$ Z ,

where H is a f i n i t e - d i m e n s i o n a l complex vector space (a C a r t a n

subalgebra of a r e p r e s e n t a t i v e G of the isomorphism class in q u e s t i o n ) ,
(~i) i ~ i $ ~ is a b a s i s of the dual H* of H (a b a s i s of the r o o t system
of @ relative to H ) and (hi) i$i~ ~ is a b a s i s of H indexed by the
same set {I ..... i} (h i is the c o r o o t associated with di )' such
that the m a t r i x ~ = (Aij) = (~j(hi)) is a C a r t a n matrix, which means
that the following conditions are satisfied:

(C1) the A.. are integers ;

1]
(C2) A.. = 2 or ~ 0 according as i = or ~ j ;
13
(C3) A. ~ = 0 if a n d o n l y if A.. = 0 ;
13 31
(C4) is t h e p r o d u c t of a positive definite symmetric matrix
and a diagonal matrix (by a b u s e of l a n g u a g e , we shall
simply say that ~ is p o s i t i v e definite).

More correctly: two such data correspond to the same isomorphism class
of a l g e b r a s if a n d o n l y if t h e y differ only by a p e r m u t a t i o n of the
indices I,...,~ . Following C. C h e v a l l e y , Harish-Chandra a n d J.-P.
Serre, one can give a simple presentation of the algebra corresponding
 194

to the system (1.1): it is g e n e r a t e d by H a n d a set of 2Z elements

e I .... ,ez, f1' .... fz subject to the following relations (besides the
vector space structure of H ):

[H,H] : {0} ;

[h,e i ] = ei(h) .e i (hE H) ;

[h,f i ] = -~i(h) .fi ( h 6 H) ;

[ei,f i ] = -h ;
1

[ei,f.] = 0 if i ~ j ;
3
-A. + I
(ad e i) 13 (ej) = 0 if i ~ j ;

-A. + I
(ad fj) 13 (fj) = 0 if i ~ j

If o n e does no longer assume that the e and the h generate H*

l 1
and H respectively, one obtains in t h a t same way all c o m p l e x reductive
Lie algebras.At this point, the g e n e r a l i z a t i o n is r a t h e r harmless
(reductive = semi-simple × commutative), but it b e c o m e s more signifi-
cant at the g r o u p level and will turn out to be q u i t e essential in
the K a c - M o o d y situation.

2. Reductive algebraic groups and Chevalley schemes.

It is w e l l known that a complex Lie algebra determines a Lie group

only up to l o c a l isomorphism. Thus, in o r d e r to c h a r a c t e r i z e a reductive
algebraic group, over ~ , say, an e x t r a - i n f o r m a t i o n , besides the data
(1.1), is n e e d e d . It is p r o v i d e d by a lattice i in H (i.e. a
-submodule of R generated by a b a s i s of H ) such that h. 6 A
1
and e. 6 A* (the Z - d u a l of A ) namely the lattice of rational co-
characters of a maximal torus of the g r o u p one considers. To summarize:
the isomorphism classes of complex reductive groups are in o n e - t o - o n e
correspondence (again up to p e r m u t a t i o n of the indices) with the systems
 195

(2.1) S = (A, (~i) i$i~ £ , (hi)1$i~Z) ,

where i is a f i n i t e l y generated free Z -module, ~ i C A* , h.E i

and ~ = (ej(hi)) is a C a r t a n matrix.

A remarkable result of C. Chevalley [Ch2] is t h a t the same classi-

fication holds when one replaces ~ by a n y algebraically closed field.
Furthermore, to a n y system (2.1), Chevalley [Ch3] and D e m a z u r e [De2]
associate a group-scheme over ~ , hence, in p a r t i c u l a r , a group
functor GS on the c a t e g o r y of r i n g s . Thus, the m a i n result of [Ch2]
asserts that the r e d u c t i v e algebraic groups over an a l g e b r a i c a l l y
closed field K are p r e c i s e l y the g r o u p s Gs(K ) , where S runs over
the systems (2.1) described above.

Question: what happens if, in the above considerations, one drops

Condition (C4) (in w h i c h case, the m a t r i x ~ is c a l l e d a generalize_dd
Cartan matrix, or GCM )? T h i s is w h a t the K a c - M o o d y theory is a b o u t .

3. Kac-Mood~ Lie algebras.

F r o m n o w on, when talking about a system (1.1), we o n l y assume

that ~. 6 H* , h. 6 H (the ~. and h. need not generate H* and
i i l 1
H ) and that ~ = (~j(hi)) is a GCM. To such a system, the p r e s e n t a t i o n
(1.2) associates a Lie algebra which is i n f i n i t e - d i m e n s i o n a l whenever
is n o t a Cartan matrix. The Lie algebras one o b t a i n s that way are
called Kac-Moody algebras. A large part of the c l a s s i c a l theory - root
systems, linear representations etc. - extends to them, with a bonus:
the study of r o o t m u l t i p l i c i t i e s (roots do h a v e multiplicities in the
general case) and of character formulas for l i n e a r representations
with highest weights have a number-theoretic flavour which is n o t
apparent in the finite-dimensional situation. For those questions, which
are o u t s i d e the subject of the p r e s e n t survey, see [Ka3] and its
bibliography.

In g e n e r a l , Kac-Moody algebras are e n t i r e l y new objects, but there

is a case, besides the p o s i t i v e definite one, where t h e y are still
closely related to finite-dimensional simple Lie algebras, namely the
"semi-definite" case: b y the same abuse of language as a b o v e , we say
that the m a t r i x ~ is s e m i - d e f i n i t e if it is the p r o d u c t of a
 196

semi-definite symmetric matrix and a diagonal matrix.

The simplest example of K a c - M o o d y algebras of s e m i - d e f i n i t e type

is p r o v i d e d by the so-called loop a l @ e b r @ 9. Let @ be a complex semi-
simple Lie algebra, H a Cartan subalgebra of @ , (~i) 1~iS £ a basis
of the r o o t system of @ relative to H , e0 the o p p o s i t e of the
dominant root and h. , for 0$j~ , the coroot corresponding to ~. .
3 3
Then, the system

H , (~j)0$jSZ ' {hj)0$j~£

satisfies our c o n d i t i o n s and the c o r r e s p o n d i n g Kac-Moody algebra turns

out to be the "loop algebra" @ ~[z,z -I] . In this case, the GCM
= (ek(hj))1$j,k$1 is d e s c r i b e d by the well-known extended Dynkin
diagram ("graphe de D y n k i n compl&t&" in [Bo]) of @ ; we shall call it
the e x t e n d e d Cartan matrix of @

Let us m o d i f y the p r e v i o u s example slightly: instead of H , we

take a direct sum H = 0_-<~_<I[ C.~j , where the ~.3 's are "copies" of the

h.'s , and we c h o o s e the e l e m e n t s ~ of H* in such a w a y t h a t the

3 3
matrix (~k(~j)) be the same ~ as before. Then, ~ is the e x t e n s i o n
of H by a one-dimensional subspace c = _C" (Zdj~j) (where the d.'s]
are nonzero coefficients such that Zd.h. = 0 ), and it is r e a d i l y
3 3
seen that the K a c - M o o d y algebra d e f i n e d by the s y s t e m (H, (~j), (~j))
is a p e r f e c t central extension of @ ® C [ z , z -I] by the one-dimensior{al
algebra c . In fact, it is the u n i v e r s a l central extension Of
6®C[z,z -I] : this is a s p e c i a l case of the following, rather easy
proposition, proved independently by Kac ([Ka3], exercise 3.14) , M o o d y
(unpublished) and the author ([Ti4]) :

PROPOSITION 1. - If the h. 's f o r m a b a s i s of H , the K a c - M o o d y

1
algebra defined by (1.2) (for (~j(hi)) a__nn a r b i t r a r y GCM) has no
nontrivial central extension.

The existence of a nontrivial central extension of G ® ~ [ z , z -I]

by ~ plays an important role in the applications of the Kac-Moody
theory for instance to p h y s i c s and to the theory of d i f f e r e n t i a l
 197

equations (cf. e. g. [Vel], [SW] and the literature cited in those

papers). It is worth noting that the K a c - M o o d y presentation provides
a natural approach to that e x t e n s i o n and a very simple proof of its
universal property, which is much less e v i d e n t when one uses direct
(e. g. cohomological) methods (cf. [Ga], [Wi]) . (NB. In the literature,
the e x p r e s s i o n "Kac-Moody algebras" is f r e q u e n t l y used to designate
merely the loop algebras and/or their universal central extension;
this unduly restrictive usage explains itself by the importance of
those special cases for the applications.)

Here, a GCM will be called "of affine t _ ~ " if it is semi-definite,

nondefinite and indecomposable; we say that it is of s t a n d a r d (resp.
twisted) affine type if it is (resp. is not) the e x t e n d e d Cartan matrix
of a f i n i t e - d i m e n s i o n a l simple Lie algebra. (In the literature, one
often finds the words "affine" and "euclidean" to mean "standard affine"
and "twisted affine" in our terminology.) In rank 2, there are two GCM

of affine type, one standard (22 - 2 2 ) a n d one twisted (_24 - ~ ) ( u p

to p e r m u t a t i o n of the indices). When the rank is > 3 , the c o e f f i c i e n t s

of a GCM (Aij) of affine type always satisfy the r e l a t i o n AijAji <
=3
(for i ~ j ), so that the m a t r i x can be r e p r e s e n t e d by a Dynkin diagram
~n the usual way (cf. e. g. [BT3], 1.4.4 or [Bo], p. 195); then, it turns
out that the diagrams representing the twisted types are o b t a i n e d by
reversing arrows in the diagrams representing standard types (i.e. in
extended Dynkin diagrams of f i n i t e - d i m e n s i o n a l simple Lie algebras).
For instance,

(Z41 I ~ i ..... I

is standard, whereas

is twisted.

The most general Kac-Moody algebra of standard affine type is a

semi-direct product of an abelian algebra by a central extension of a
loop algebra. There is a similar description for the algebras of
twisted affine type, in which the loop a lg e b r a s must be r e p l a c e d by
suitable twisted forms. For instance, if @ is a complex Lie algebra
 198

of type E6 and if J denotes an involutory automorphism of the

loop algebra @ ® ~ [ z , z -I] operating on the first factor b y an o u t e r
automorphism a n d o n the second by z I > -z , then the fixed point
algebra ( @ ® ~ [ z , z -I ] )d is a K a c - M o o d y algebra of type 2E 6 above
(hence the notation !) . The c o n n e c t i o n between Kac-Moody algebras
of a f f i n e type a n d the loop algebras and their twisted analogues was
first made explicit in [Ka2], but the c o r r e s p o n d i n g relation at the
group level had been known for some time: cf. [IM] and [BT2] (where,
however, a local f i e l d - s u c h as ~((z)) - replaces ~[z,z-1]).

4. Associated groups: introductory remarks.

In the c l a s s i c a l , finite-dimensional theory, the Lie algebras

often appear as intermediate step in the study of L i e g r o u p s . It is
therefore natural to try similarly to "integrate" Kac-Moody Lie algebras
a n d to define "Kac-Moody groups". More precisely, let S be as in (2.1)
except that, now, the m a t r i x (~j (hi)) is o n l y assumed to b e a GCM.
To s u c h a system S , one wishes to a s s o c i a t e an " i. n f .l n m. t e .- d l. m e n s l o n a 1

group over C " , let us c a l l it Gs(C) , or, more ambitiously, a

group functor GS o n the c a t e g o r y of rings.

Before passing in q u i c k review the m e t h o d s that have been used to

define such groups, let us m a k e a preliminary comment. As m a y be e x p e c -
ted, since the g r o u p s in q u e s t i o n are "infinite-dimensional", one is
led, for a g i v e n S , to define n o t one b u t several groups which are
various completions of a s m a l l e s t one (those completions corresponding
usually to various completions of the K a c - M o o d y Lie algebra). Thus,
the g r o u p theory c a n be d e v e l o p e d at d i f f e r e n t levels (or, if one
prefers, in d i f f e r e n t categories); roughly speaking, one m a y d i s t i n g u i s h
a minimal (or p u r e l y algebraic) level, a formal level and an a n a l y t i c
level, with many subdivisons.

Instead of trying to d e f i n e those terms formally, I shall just

illustrate them with one e x a m p l e . Let @ be a complex, quasi-simple
s.±mply connected algebraic group (Lie a l g e b r a s will now play a minor
role, a n d we are free to use g o t h i c letters for o t h e r purposes !),
A* the lattice of r a t i o n a l characters of a m a x i m a l torus of G ,
i its ~ -dual, (~i)]~i~ i a basis of the r o o t s y s t e m of @ with
respect t o the t o r u s in q u e s t i o n , ~0 the o p p o s i t e of the dominant
 199

root, hj (for 0~j5£) the c o r o o t corresponding to ~j and

S : (i, ((zj)0~j~ £ , ( h j ) 0 ~ j ~ £) . In § 3, we have seen that the Lie algebra
associated w i t h S (in w h i c h i is r e p l a c e d by ~® i) is L i e @ ®~[z,z-1].
Clearly, the group most naturally
associated with S over
-I
must - and will - be the g r o u p @(~[z,z ]) o f all " p o l y n o m i a l maps"
×
--> @ . In t h a t special case, this is the answer to o u r q u e s t i o n at
the m i n i m a l level. At the formal level, we find @(~((z))) . Now, the
points of @(C[z,z-1]) can also be v i e w e d as c e r t a i n special loops
S I --> @ (by r e s t r i c t i n g x --> @ to the complex numbers of a b s o l u t e
value one) and this opens the w a y to a g r e a t variety of c o m p l e t i o n s of
~(~(z,z-1)) , l e a d i n g to g r o u p s of l o o p s S I --> @ in v a r i o u s c a t e -
g o r i e s (L 2 , c o n t i n u o u s , C , e t c . ) : t h i s is the a n a l y t i c level.

In the case of the a b o v e system S , there is no d i f f i c u l t y in

guessing what should be the g r o u p functor GS : at the m i n i m a l level,
we shall have Gs(R) = @(R[z,z-1]) , where @ now denotes the
Chevalley scheme corresponding to the system (£, (~i) 1 ~ i s £ , (hi)1~i~ £) ,
and the c o r r e s p o n d i n g formal group will be @ ( R ( ( z ) ) ) . (In t h i s
generality, I do n o t know what "analytic" s h o u l d mean.) As one sees,
all those groups c a n be d e s c r i b e d with elementary means, without re-
ference to K a c - M o o d y algebras. But things change as s o o n as one slightly
modifies the system S as in § 3 by t a k i n g for instance

A = A~]0~ j (and k e e p i n g the G C M u n c h a n g e d , as b e f o r e ) . The c o r r e s -

J
ponding group is t h e n a central extension of the loop g r o u p (whichever
category one is in) by ~× or, in the r i n g situation, by R × . As in
the L i e algebra case, the e x i s t e n c e of that extension comes out of
the g e n e r a l theory quite formally, but in the loop group case, it
reflects rather deep properties of those groups (cf. e. g. [SW]) , a n d
direct existence proofs are n o t easy. Note that if R is a f i n i t e
field k , one gets a central extension of @(k((z))) by k× which
appears in the w o r k of C. M o o r e [Mo2] and H. M a t s u m o t o [Ma3].

Here, we shall most of the time adopt either the m i n i m a l or the

formal viewpoint (the a n a l y t i c ones are usually deeper and m o r e impor-
tant for the applications, but unfortunately less familiar to the
speaker). L e t us b r i e f l y mention some contrasting features of t h o s e .
The formal groups are u s u a l l y simpler to handle (as a r e local fields
compared to g l o b a l ones !). T h i s is due in p a r t i c u l a r to the fact that
 200

they contain "large" proalgebraic subgroups (cf. e. g. [BT2], § 5, and

[SZ], Kap. 5). Also, they seem to be the r i g h t category for simplicity
theorems (cf. [Mol]; observe that if @ denotes a complex simple Lie
group, then @(~((z))) is a s i m p l e group, which is far from true for
@(~[z,z-1]). On the other hand, the m i n i m a l theory presents a certain
symmetry (the symmetry between the e.'s and f.'s or, in the e x a m p l e
of @(~[z,z-1]), the symmetry between z and z ), w h i c h gets lost
in the formal completion.

Let us m e n t i o n an important aspect of t h a t symmetry. All the groups

G = GS(~) we are t a l k i n g about (and, in fact, the groups Gs(K) ,
for K a field), whether minimal or f o r m a l , are e q u i p p e d with a BN-
pair (B,N) (or T i t s system: cf. [Bo]) whose Weyl group W = N/B DN
is the C o x e t e r group W(~) defined as follows:

C..
W(A) = <ri [ 1<i<Z ; r2 = I ; (rir j) 13 : I if i ~ j ,

A..A.. ~ 3 , and c . = 2,3,4 or 6 according

53 31 13

as A. A = 0, 1, 2 o r 3 >
13 ]l

(cf. [MT], [Mal], [Ti3] a n d also, for the a f f i n e case, [IM], [BT2] and
[Ga]) . In p a r t i c u l a r , G has a Bruhat decomposition G = U BwB ,
w 6 W
leading to a " c e l l decomposition" of G/B : the q u o t i e n t s BwB/B have
natural structures of finite-dimensional affine spaces. Now, in the
minimal situation, the same N is the g r o u p N of a n o t h e r BN-pair
(B-,N) , not conjugate to the p r e v i o u s one except in the finite-dimen-
sional case (i.e. when ~ is p o s i t i v e - d e f i n i t e ) . Furthermore, one
a l s o has a partition G = U B wB , called the B i r k h o f f decomposition
w 6W
of G (because of the special case considered in [Bi]; for the g e n e r a l
result, cf. [Ti4]) . W h i l e the c e l l s BwB/B are finite-dimensional,
the "cells" B-wB are finite-codimensional in G , in a s u i t a b l e
sense, and, unlike the B r u h a t decomposition, the B i r k h o f f decomposition
always has a big cell, namely B-B if one chooses B- in its c o n j u g a c y
class by N so t h a t the intersection B N B is m i n i m u m with respect
to the inclusion (we t h e n say that B and B are o p p o s i t e ) . In the
formal situation, a Birkhoff decomposition (and h e n c e a b i g cell) still
 201

exists, but here, the groups B and B play completely asymmetric

roles: B is m u c h smaller than B in that, for instance, B-\G/B-
is n o w h i g h l y u n c o u n t a b l e (always e x c e p t i n g the case where ~ is
positive-definite). We c a n be m o r e explicit: if G = U B wB is the
A
Birkhoff decomposition of the minimal group G , and if G denotes
A
the formal completion of G , then the B i r k h o f f decomposition of G
is U B wB , where B is the closure of B in ~ ; the g r o u p B
A
is c l o s e d (and e v e n discrete) in G .

Different methods have been used to a t t a c h groups to K a c - M o o d y

data. Roughly, one c a n classify them into four types, according to
which techniques they are b a s e d upon, namely:

linear representations (cf. § 5 below);

generators and relations (cf. § 6);
Hilbert manifolds and line bundles;
axiomatic (cf. [Ti4]) .

About the t h i r d approach, which is h a n d l e d in G r a e m e Segal's lecture

at t h i s A r b e i t s t a g u n g , let us just say t h a t it g i v e s a deeper geometric
insight in the situation than the other methods, but that, at p r e s e n t ,
it c o n c e r n s only the affine case. Also the axiomatic approach has been
used only in the affine case so far: we shall briefly indicate below
(§ 6 a n d A p p e n d i x 2) to w h i c h purpose.

5. Construction of the groups viarepresentation theory.

One of the simplest way to p r o v e the e x i s t e n c e of a L i e group with

a given (finite-dimensional) Lie algebra L consists in e m b e d d i n g L
in the e n d o m o r p h i s m algebra End V of a vector space V (by A d o ' s
theorem) and then considering the group generated by exp L .

If L is a K a c - M o o d y algebra, linear representations are infinite-

dimensional and exp L is n o l o n g e r defined in g e n e r a l . However,
suppose that the linear representation L c_--> E n d V is s u c h t h a t the
elements ei,f i , considered as e n d o m o r p h i s m s of V , are locally
nilpotent (an e n d o m o r p h i s m ~ of V is s a i d to be locally nilpotent
if, for a n y vE V , ~n(v) = 0 for a l m o s t all n6~ ). T h e n , if the
 202

ground field K has characteristic zero, say, exp Ke and exp Kf

i l
are w e l l - d e f i n e d "one-parameter" automorphism groups of V which
generate the g r o u p Gs(K) one is l o o k i n g for, at l e a s t if the h.'sl
generate A . Otherwise, one must also require that, as a A-module
(remember that AcL ), V is a d i r e c t sum ~V s of one-dimensional

modules on which A operates through "integral characters" X s 6 A* ;

then, one adds to the above generators the "one-parameter groups"
I(K ×) , with ~6A , where, by d e f i n i t i o n , ~(k) operates on V
s
via the m u l t i p l i c a t i o n by k <l'Xs> . An L-module V is s a i d to be
integrable if it s a t i s f i e s the a b o v e conditions (local nilpotency of
ei,f i , plus the e x t r a - r e q u i r e m e n t on A , which however follows from
the first condition when the h '±s generate i ).

That method for integrating L , inspired b y C. Chevalley's

TohSku paper [Chl], was first devised b y R. M o o d y and K. T e o [MT],
who used the adjoint representation of L . In t h a t way, of course,
they only g e t the m i n i m a l a__djoint g r o ~ . (More p r e c i s e l y , the group
they construct is the analogue of C h e v a l l e y ' s simple group, namely the
subgroup of the a d j o i n t group generated by the e x p Ke i and exp Kf i ;
here, we say t h a t the system S defines an a d j o i n t group if the
~i's generate A* and if =Q@ A is g e n e r a t e d as a _n-vect°r space by
the h's .) On the other hand, a suitable variation of the m e t h o d
i
described above enables t h e m to i n c l u d e the c a s e of a g r o u n d field with
sufficiently large characteristics. L a t e r on, Moody [Mol] has applied
the same ideas at t h e formal level, starting from a suitable completion
o f the K a c - M o o d y algebra.

In [Mal ], R. M a r c u s o n works with highest weight modules, at the

formal level. His m e t h o d requires the c h a r a c t e r i s t i c to be zero.

In [Ga], H. Garland also uses highest weight representations. He

restricts himself to the standard affine case - and makes heavy use
of the relation between L a n d the loop algebra -, b u t in t h a t special
case, his results go much beyond those of Marcuson in t h a t he e s s e n t i a l l y
works over ~ (with ~--forms o f the universal enveloping algebra of L
a n d o f the representation space), which enables him to d e f i n e groups
over arbitrary fields.

One drawback of the approach by m e a n s of linear representations

is t h a t it is not clear, a priori, h o w the group one associates to a
 203

given Kac-Moody algebra (over ~ , say) varies with the c h o s e n repre-

sentation. In [Mal], this q u e s t i o n is ].eft o p e n . Garland answers it
by u s i n g the fact that the g r o u p s he c o n s t r u c t s are central extensions
of loop groups, and computing a cocycle which describes the e x t e n s i o n .

V. K a c a n d D. P e t e r s o n [KP] o b v i a t e that inconvenient of the

method by c o n s i d e r i n g all integrable modules simultaneously. They start
f r o m the free p r o d u c t G* of the additive groups Kei,Kf i for all i .
For any integrable module V , the m a p s te l• - - > exp te i ,
tf i - - > exp tfl extend to a r e p r e s e n t a t i o n eXPv : G* - - > GL(V) , and
the g r o u p they consider is G*/ ~ (Ker eXPv) , where V runs through
all integrable representations. This is the m i n i m a l group, in the sense
of § 4, a n d c o r r e s p o n d s t o the c a s e where the h's form a basis of
1
i . (An o t h e r , earlier approach of t h a t same group, but without this
last r e s t r i c t i o n on the h.'s , c a n be found in [Ti3] : cf. § 6).
1

R. Goodman a n d N. W a l l a c h [GW] are c o n c e r n e d with the standard

affine case over ~ . Working within the theory of Banach Lie algebras
and groups, they consider a large variety of Banach completions of the
Kac-Moody algebras, and integrate them by u s i n g suitable topologizations
of c e r t a i n highest weight (so-called standard) modules. One of t h e i r
×
purposes is to d e f i n e the central extension of loop groups by ~ at
various analytic levels. An a l t e r n a t i v e , more elementary approach to
that problem (not t o u c h i n g , however, the m a i n body of r e s u l t s of [GW])
may possibly be suggested b y the r e m a r k of A p p e n d i x I below.

6. Generators and relations.

In a c o u r s e of lectures summarized in [Ti3] (cf. also [S£] and

[Ma2]) , I g a v e another construction for groups associated with Kac-
Moody data. In o r d e r to s k e t c h the m a i n idea, let us r e t u r n to the
case of a f i n i t e - d i m e n s i o n a l complex semi-simple Lie group G . Such
a group is k n o w n t o be the a m a l g a m a t e d product of the normalizer N
of a maximal torus T a n d the p a r a b o l i c subgroups PI,...,P£ contai-
ning properly a given Borel subgroup B containing T and m i n i m a l
with that property, with amalgamation of the intersections P n P. = B
l 3
and P ~N. (cf.[Ti2],]3.3) . Furthermore, P. is the s e m i - d i r e c t p r o d u c t
1 1
of its L e v i s u b g r o u p L i containing T by a unipotent group Ui
Thus, we h a v e a presentation of G whose ingredients are the subgroups
 204

N,Li,U i . The groups N and L1 c a n be r e c o n s t r u c t e d f r o m the system

S of (2.1) in a uniform way, without reference to the positivity of
the m a t r i x ~ : the group N is g e n e r a t e d by T = H o m ( A * , C ×) and
£ elements m. (Igi$£) submitted to the relations
1

(6.1) m. normalizes T , a n d the automorphism of T it

l
i n d u c e s is t h e adjoint of the r e f l e c t i o n

l ~--> i - <l,h.>-@. of A* ,
i 1

(6 • 2 ) m2 = (-1 <l,hi>
i = ~i 6 T = H ° m ( i * ' C × ) ' with 0i(

for I E A*
and
(6 .3) if A, .A.. : 0 (resp. I;2;3), then m.m. = m m.
13 3x 13 3 1

(resp. m i m j m i : m j m i m j ; ( m i m j ) 2 = ( m j m i ) 2 " (mimj)3= (mjmi)3) ,

whereas L. is n o t h i n g else but the r e d u c t i v e group of s e m i - s i m p l e

I
rank one corresponding to the system (A,hi,~i) . As for the UiI s ,
being unipotent, t h e y are e a s i l y described in t e r m s of their Lie alge-
bras Lie U. , either by m e a n s of the C a m p b e l l - H a u s d o r f f formula or,
1
more conceptually, by e x p o n e n t i a t i n g Lie U in the c o m p l e t i o n (for
1
the natural filtration) of its u n i v e r s a l enveloping algebra U(Lie Ui) -

All this c a n be c a r r i e d over to an a r b i t r a r y system S , with the

only difference that, now, Lie U is i n f i n i t e - d i m e n s i o n a l a n d no
i
longer nilpotent but only pro-nilpotent (more p r e c i s e l y , the Lie U . ' s
l
are c e r t a i n s u b a l g e b r a s of c o d i m e n s i o n I of the L i e a l g e b r a g e n e r a t e d
by the e . ' s , a n d the l a t t e r has a p r o - n i l p o t e n t c o m p l e t i o n ) . Moreover,
3
by u s i n g a s u i t a b l e Z=-form of the u n i v e r s a l e n v e l o p i n g a l g e b r a of the
Kac-Moody algebra (generalizing the k - f o r m u s e d by H. Garland in the
affine case: cf. § 5), one is a b l e to do e v e r y t h i n g over ~ and, by
reduction, over an arbitrary ring R . Thus, one is l e d to a t t a c h to
S a group functor on the c a t e g o r y of r i n g s , call it ES . But this
group functor ES is n o t the "good" functor GS one is l o o k i n g for:
indeed, if ~ is a C a r t a n matrix, that is, in the positive definite
case, GS should of course be the C h e v a l l e y group-scheme corresponding
to the C h e v a l l e y - D e m a z u r e data S , and o n e finds that the functor ES
 205

coincides with that scheme only over the p r i n c i p a l ideal domain. This
suggests that, in g e n e r a l , ES m a y be the g o o d functor when restricted
to t h o s e rings. This is u n d o u b t e d l y so in the affine case. Indeed, in
that case, one can characterize the functor ES restricted to p r i n c i p a l
ideal domains - call it E (pid) - by a s y s t e m of v e r y natural axioms
which, it seems, s h o u l d be satisfied b y the "good" functor GS (cf.
[Ti4], 7.6 b)). Another application of t h o s e axioms is t h a t they enable
one t o d e t e r m i n e explicitly the functor E~pid)- (whereas the m o r e
abstract definition by generators and r e l a t i o n s is m u c h less m a n a g e a b l e ) ,
and that the r e s u l t one obtains suggests (always in the a f f i n e case)
what must be the functor GS for a r b i t r a r y rings. We shall come back
to t h a t q u e s t i o n in t h e next section (and in A p p e n d i x 2), b u t l e t us
first conclude the p r e s e n t one by t w o r e m a r k s .

The above considerations c a n be d e v e l o p e d both at the m i n i m a l and

the formal level. In fact, the c o n s t r u c t i o n of [Ti3] depends o n the
choice of a c e r t a i n subgroup X (subject to s o m e simple conditions)
of the m u l t i p l i c a t i v e group of the c o m p l e t e d universal enveloping
algebra of the L i e algebra generated by the e's . Among the possible
3
X , there is a m i n i m a l one, leading to the m i n i m a l g r o u p GS(~) (and
functor GS ), and a m a x i m a l one (which h a s been determined by
O. M a t h i e u [Ma2]), leading to the formal group (and functor) associated
with S .

The g r o u p s we h a v e been considering are the generalizations, in

the K a c - M o o d y framework, of the s_]21it r e d u c t i v e groups but, as in the
finite-dimensional case, one can define ~on-split forms of t h o s e groups.
In p a r t i c u l a r , over ~ , there is a " c o m p a c t " form (which is b y no
means compact in the t o p o l o g i c a l sense !) : in the m i n i m a l set-up, it
can be d e f i n e d as the fixed-point group of the "anti-analytic" involu-
t i o n of GS(~) induced by the semi-linear involution of the K a c - M o o d y
algebra which permutes e. and f~ a n d i n v e r t s the e l e m e n t s of i
l 1
(Another definition, i n v o l v i n g h e r m i t i a n f o r m s in r e p r e s e n t a t i o n
spaces, also works at the formal level: cf. [Ga], [KP]) . G e n e r a l i z i n g
a result which was known in t h e finite-dimensional case [Kal], V. K a c
[KDP] has observed that that compact f o r m c a n be d e f i n e d as the
amalgamated product of its r a n k 2 subgroups corresponding to t h e p a i r s
of indices i,j, 6 {I ..... ~} , with amalgamation o f the r a n k I sub-
groups (of t y p e SU2) corresponding to the indices (here, one m u s t
 206

assume that the h.'s form a basis of A , or add a c o m p a c t torus to

1
the a m a l g a m ) . As for the r a n k 2 groups, which are the ingredients of
that definition, they are k n o w n in case they are finite-dimensional
(i.e. when ei(hj).~j(hi) $ 3) ; otherwise, they are shown to be amal-
gamated products of two g r o u p s of type U2 with suitable amalgamation
of a t w o - d i m e n s i o n a l torus. (N.B. A r e s u l t similar to the a b o v e is
known to h o l d for finite-dimensional split groups or, more generally,
for g r o u p s having a BN-pair with finite Weyl group: cf. [Ti2], 13.32,
and, for e a r l i e r versions and special cases, [Cu] and [Til], 2.12.)

The fact that the d e f i n i t i o n by g e n e r a t o r s and r e l a t i o n s does not

provide the "good" functor GS for r i n g s that are n o t p r i n c i p a l ideal
domains probably lies in the n a t u r e of t h i n g s (as K - t h e o r y suggests).
A more likely way to get at the "right" GS would consist in e x h i b i -
ting a suitable Z--form of the affine algebra of GS(~) (cf. § 8
below).

7. An example: groupsof type 2E 6

In this section, we a d o p t the f o r m a l v i e w p o i n t ; to e m p h a s i z e the

^
fact, we shall use the n o t a t i o n GS , i n s t e a d of GS as above.

Let S be the system ( i , ( ~ j , h j ) 0 ~ j $ 4) , where the m a t r i x

(aj(hi)) is of type
2~6 (ef. § 3), and where the ~i generate A*
whereas the h. generate i : these properties characterize S . Our
l
A
purpose is to d e s c r i b e the g r o u p s GS(K) when K is a field. We
discuss only this special example for the sake of c o n c r e t e n e s s , but
similar results hold for any other twisted type (the type

~<I I... ~ i<J

is b r i e f l y examined in [Ti4], 7.4 , and g e n e r a l statements, concerning

all affine types and arbitrary rings, will be g i v e n in A p p e n d i x 2).

From the e x p l i c i t description of the Kac-Moody algebras of type

2~ 6 given in § 3, one readily guesses what must be the g r o u p A
Gs(K)
when K is a f i e l d of c h a r a c t e r i s t i c not 2, n a m e l y
 207

A
GS(~) = ~(K((z))) ,

where G is a q u a s i - s p l i t algebraic group of t y p e 2E 6 , d e f i n e d

over K((z)) and whose splitting field is K((/z)) If K : ~ , one
proves this by straightforward "integration" (for a r b i t r a r y affine
types, t h i s p a r t of the work is d o n e in [Mo3]), and the g e n e r a l case
ensues via the a x i o m a t i c method mentioned above (cf. § 6 and [Ti4],
7.6 b)).

Now, suppose that car K = 2 . The above description cannot hold

in t h a t case since the e x t e n s i o n K((/z)/K((z)) is n o t separable,
hence is i m p r o p e r for the d e f i n i t i o n of a q u a s i - s p l i t group. But there
is a c i r c u m s t a n c e which enables one again to g u e s s the result, at l e a s t
when K is p e r f e c t . Indeed, one knows that, in the f i n i t e - d i m e n s i o n a l
theory, the arrows carried b y the d o u b l e bonds of Dynkin diagrams
"disappear" over perfect fields of c h a r a c t e r i s t i c 2: m o r e precisely,
reversing such an a r r o w c o r r e s p o n d s to an i n s e p a r a b l e isogeny which is
bijective on rational points. Here, the d i a g r a m becomes "the same as"
F4 = } ] I > I l , hence the (correct) guess

Gs(K) = F4(K((z)) )

But how can it be t h a t a 7 8 - d i m e n s i o n a l (quasi-split) group of

type E6 suddenly degenerates into a 52-dimensional (split) group of
type F 4 ?! The answer is s i m p l e : F4(K((z))) must be v i e w e d as the
group of r a t i e n a l points of a s u i t a b l e 78-dimensional group defined
over K((z)) . The existince of such a group is not so s u r p r i s i n g when
one considers the isomorphism (a,b) I > a2+zb 2 of K((z))×K((z))
onto K((z)) , hence of a 2 - d i m e n s i o n a l group onto a l-dimensional
group ( K perfect).

To be m o r e specific, set L : ~2((z)) , L' : ~((/~)) and denote

by F a split group of t y p e F4 over L' , by o : F --> F a
special isogeny of F into itself whose square is t h e F r o b e n i u s endo-
morphism (cf. [BTI], 3.3), by RL,/L the restriction of s c a l a r s and
by @ the image of RL,/L~ : RL,/L ~--> RL,/LF . Then

thee a l g e b r a i c group @ i ss 7 8 - d i m e n s i o n a l . For any (non n e c e s s a r i l y

perfec t ) field K o_ff c h a r a c t e r i s t i c 2, o n e has GS(K) = G (K((z))) ;
 208

if K i_ss p e r f e c t , the m a ~ RL,/L ~ : RL,/LF -->@ i_~s b i j e c t i v e on rati-

onal points, therefore G s ( K ) ~ (RL,/LF) (K((z))) : F(K((/~) ) ~ F(K((z))).

Let us e x p l a i n briefly where the 78 d i m e n s i o n s of @ come from.

The g r o u p F4 has an o p e n set ~ which is the p r o d u c t , in a s u i t a b l e
order, of 48 a d d i t i v e groups Ua corresponding to the 48 roots a
and a 4 - d i m e n s i o n a l torus T . The isogeny ~ induces a bijection
a~--> c(a) of the root system into itself which maps short roots onto
long roots and vice versa. The groups RL,/LU a are 2-dimensional and
dim R L , / L T = 8 . Now, it is r e a d i l y checked that:

if a is short, RL,/LO maps RL,/LH a isomorphically

onto RL,/LHo(a)

if a is long, RL,/LO maps RL,/LU a onto a

one-dimensional subgroup of RL,/LH (a) ;

RL,/LO maps RL,/LT onto a six-dimensional subtorus of

itself.

Thus, dim (RL,/LO) (~) = 2.24 + 24 + 6 = 78.

We p r o p o s e the following exercise to the interested reader: for

K perfect of c h a r a c t e r i s t i c 2, w r i t e SL2(K((z))) as the group of
rational points on K((z)) of an 8 - d i m e n s i o n a l algebraic group. This

arises when one studies the case of the G C M (_~-~) ; the 8-dimensional

group in q u e s t i o n appears as a c h a r a c t e r i s t i c 2 "degeneracy" of

SU3(K((/z))) . Cf. [Ti4], 7.4, for m o r e details.

8. The algebro-geometric nature of the groups GS(~)

What kind of a l g e b r o - g e o m e t r i c objects are the functors GS and,

in p a r t i c u l a r , the groups GS(~) ? Little is k n o w n for GS is g e n e r a l ,
but something can be said about GS(~) (here, ~ could be r e p l a c e d
by any field of c h a r a c t e r i s t i c zero).
 209

Set G = GS(~) . We have already mentioned the Bruhat decomposi-

tion G = U BwB , where B is a c e r t a i n subgroup of G , which we m a y
call "Borel subgroup", and w runs over a Coxeter group W . Coxeter
groups, endowed as u s u a l with a distinguished generating set
S = {r] ..... r~} (cf. § 4), have a natural ordering: for w,w' 6 W ,
one sets w ~ w' if t h e r e exists a reduced expression w : Sl...s n
(s i 6 S) and a subsequence (ii, .... im) of (1 ..... n) such that
w' : s . . . . s. . Then:
11 zm

for any w6 W , the subset Schub w = w' ~ w (Bw'B)/B of G/B ,

=
called the Schubert variety c__orresponding to w , has a natural struc-
ture of p r o j e c t i v e manifold (ef. [Ti4]); thus, G/B is a d i r e c t
limit of projective manifolds.

(In [Ti4], the projective structure of Schub w is d e f i n e d by m e a n s

of a highest weight representation of G , and is t h e n shown not to
depend on the choice of that representation. It w o u l d be desirable to
have a more intrinsic definition, using for instance the big cell of
the Birkhoff decomposition, as w a s suggested to the speaker by
G. Lusztig.) The set G/B , and its description as a limit of p r o j e c t i v e
varieties, does not depend on whether one adopts the minimal or the
formal viewpoint (more precisely, the formal group is the completion
of the minimal one for a topology for which B is an o p e n subgroup).
Also, since B contains H = B A N , the quotient G/B and the varieties
Schub w depend only on the GCM ~ , and not on A ; when the choice
of ~ needs to be specified, we shall write SchUbA__W instead of
Schub w .

If we n o w take the formal viewpoint, the Borel subgroup, or rather,

to r e m a i n consistent with the notation of t h e end of § 4, the closure
g of B in ~ = GS(~) , is a p r o a l g e b r a i c group, semi-direct exten-
sion of a torus by a prounipotent group (cf. [BT2], [Si]).

A --
Having thus described both G/B = G/B and B , we h a v e g a i n e d
^
some understanding of the algebro-geometric nature of G itself. But
a more direct and promising picture is g i v e n b y V. Kac and D. Peterson
[KPI] who attach to G (the m i n i m a l group) a "coordinate ring", or
rather two rings, the ring ~[G] of "strongly regular" functions, and
the ring ~[G] r of "regular" functions. The first one is g e n e r a t e d by
the coefficients of all highest weight representations (in [ K P 1 ] , t h i s
 210

is not chosen as definition of ~[G] but is p r o v e d to be a p r o p e r t y of

the ring, defined in a d i f f e r e n t way) and p r o v i d e s a Peter-Weyl type
theorem. That ring is not invariant by the map i : g ~__> g-1 (under
that map, highest weight representations become lowest weight represen-
tations !); for a s u i t a b l e topology, ~[G] r is t o p o l o g i c a l l y generated
by C[G] and i*(C[G]) It is shown in [KP] that G is an a f f i n e
(infinite-dimensional) algebraic group with coordinate ring C[G] ,
-- r

in the sense of S h a f a r e v i t c h [Sh] : this implies, in p a r t i c u l a r , that

G can be identified with a subset of C=~= in such a way that ~[G] r
is the r e s t r i c t i o n to G of the ring ~[~---~]r-- of r e g u l a r f u n c t i o n s on
C~
~T (i.e. the ring of functions whose restriction to ~[0,n] is p o l y -
nomial for all n ), and that G is the v a n i s h i n g set of an ideal
of ~[c_m=]
r

9. Applications.

"Kac-Moody groups" have been used in a v a r i e t y of d o m a i n s such as

topology, differential and p a r t i a l differential equations, singularity
theory, etc. Those applications, a fast growing subject, are b e y o n d both
the scope of this survey and the competence of the speaker. Let me just
unsystematically list a few b a s i c references, which will give access to
at least part of the literature on that topic: [SW] (cf. also the
reference[5] of [SW]), [Vel], [Ve2] (these concern applications of Kac-
Moody Lie algebras, rather than groups), [RS], [Si].

Most applications so far use only groups of a f f i n e type, and

there may still be d o u b t s about the u s e f u l n e s s of the g e n e r a l theory.
To finish with, I would like to give an a r g u m e n t in f a v o u r of it. We
have seen that to e v e r y GCM ~ = (~ij) and e v e r y element of the corres-
ponding Coxeter group W(~) , the t h e o r y associates a certain complex
projective variety SchUbAW . If w is one of the canonical generators
ri of W , SchUbA__W is just a p r o j e c t i v e line. The next simple case
is W = rl.r.3 ; then, SchUbA__W is a r a t i o n a l ruled surface, i.e. a sur-
face fibered over ~i (~) wTth projective lines as fibers. It is well
known that such a s u r f a c e X is c h a r a c t e r i z e d up to isomorphism by a
single invariant ~(X) which is a n e g a t i v e integer (if v(X) ~ 0 , X
is o b t a i n e d by b l o w i n g up the vertex of a cone of d e g r e e m(X) in a
(~(X)+1)-dimensional projective space). Now, one shows that
~(Schub~(rirj)) = Aij . This gives a geometric interpretation of the
 211

matrix A . Moreover, observe that, if one accepts only to c o n s i d e r

GCM of a f f i n e type, only the surfaces X with v(X) 6 [-4,0] , among
the rational ruled surfaces, have the right to be called "Schubert
varieties", which seems rather unnatural ! I should think that the class
of all Schub~w , for all ~ and w , will turn out to be a very
natural and interesting class of p r o j e c t i v e varieties to c o n s i d e r .
 2';2

A_p_pendix I. Central extension.

For arbitrary S , the "minimal group" GS(~) c a n be c o n s t r u c t e d

by the methods described in §§ 5 a n d 6. In p a r t i c u l a r , those methods
provide very simple, purely formal existence proofs for a n o n t r i v i a l
central extension of the "polynomial" loop groups by C x . The situation
is q u i t e different when one starts from loop groups defined by analytic
conditions. However, the following rather trivial considerations may
conceivably enable one to e x p l o i t the r e s u l t known for p o l y n o m i a l loops
in the analytic case. Here, all topological spaces are a s s u m e d to be
Hausdorff.

Let ~ : G' - - > G be a c e n t r a l group extension and let U',H'_ ,U+'

be t h r e e subgroups of G' such that Ker nc H' , that H' normalizes
U±' and that the product mapping U'_ × H' x U'+ - - > G' is i n j e c t i v e .
Thus, ~+ = ~IU+' and ~_ = ~[U'_ are isomorphisms of U'+ and U'_

onto two subgroups U+ and U_ of G . We set H = z(H')

A
Now, let us e m b e d G in a c o m p l e t e topological group G and
suppose that, if U and U denote the c l o s u r e s of U and U+
A -- + A --
in G , the p r o d u c t mapping in G defines a h o m e o m o r p h i s m of
^
_ x H x ~+ onto a dense open subset ~ of G . L e t us a l s o e n d o w
H' with a topology making it into a c o m p l e t e topological group, such
that Ker z is c l o s e d in H' and t h a t the c a n o n i c a l algebraic iso-
morphism H'/Ker ~--> H is an i s o m o r p h i s m of t o p o l o g i c a l groups as
well (observe that, by hypthesis, H is locally closed, hence closed
^
in G , and is t h e r e f o r e a complete topological group).

Set X = {(u,u')6 U+ x U_ I uu' 6 ~} . This is a d e n s e open subset

of U+ x U_ (endowed with the t o p o l o g y induced b y t h a t of U+ x U_ ).
For (u,u') 6 X , there is a u n i q u e element ~ ( u , u ' ) C H' such that

~+-I lu) • ~_-I (u') cu_' .~(u,u') .u~

A
The t o p o l o g y of G induces a topology on U± which we lift to U~
-I
by m e a n s of ~± , a n d we e n d o w ~' = U'_ H' Ui with the p r o d u c t topology.
The following proposition is e a s y .
 213

PROPOSITION 2. If the function ~ :X --> H' is c o n t i n u o u s , there is a

unique topolo_s ~ on G' making G' into a topological group and 9' ,
topologized as above, into a dense open subset of G' . Suppose further
that there is a n e i g b o r h o o d Xl of (1.1) i_nn U+ × U such that the
restriction of ~ t_~o X n Xl extends to a c o n t i n u o u s map X1 --> H'
A
Then, the to_gpolo~ical group G' admits a completion G' , Ker z is
A
a closed subgroup o_~f ~' and the h o m o m o r p h i s m &' - - > G extending
A
factors through a__nni s o m o r p h i s m of topological ~roups ~'/Ker 7--> G .

Note that the left (or right) translates of all open subsets of
obviously form a basis of the topology of G' (hence the u n i q u e n e s s
assertion).

In the application I have in mind, G would be a " p o l y n o m i a l "

A
loop group, G some other loop group, z : G' --> G the "natural"
x
central extension of G by Ker ~ ~ C (whose existence is e a s i l y
;
proved by any of the m e t h o d s described in §§ 5 and 6), Ui and U+
the (non c o m p l e t e ) "prounipotent radicals" of two o p p o s i t e Borel sub-
groups of G' (ef. § 4) and H' the intersection of those Borel sub-
x
groups, a direct product of c o p i e s of C which one e n d o w s with its
natural topology. The m a i n problem, which I have not investigated, is
of c o u r s e to p r o v e (in the interesting cases) that ~ is c o n t i n u o u s
and e x t e n d s to a n e i g h b o r h o o d of (I,I) in U+ × U_

A
Appendix 2. The group functor GS in the affine case.

In this appendix, we shall use the techniques and terminology of

A
[BT4] to describe the formal functors GS for all systems

S = (A, (~i)0 < i < Z , (hi)0 < i < Z)

satisfying the following conditions:

(AI) the m a t r i x ~ = (~j(hi)) is of irreducible, affine type;

(A2) the set {hil0 ~ i S Z} generates i ;

(A3) the set {~iI0 ~ i £ £} contains a ~- basis of ~® A*

 214

More precisely, for any such S , we shall describe a topological

A
group functor GS having the following properties.

A
(P0) There is a Lie algebra functor Lie GS defined as f o l l o w s
(compare [DG], pp. 209-210). For any r i n g R , set
R(e,g') = R [ t , t ' ] / ( t 2 , t '2) , where s,s' are the canonical images
of t,t' in the q u o t i e n t ; in o t h e r words, R(e,e') is the tensor
product of two a l g e b r a s R(e), R(e') of dual numbers. For r 6 R ,
let ~ : R(S) --> R , 1 : R(e) --> R(E') , o : R(~) --> R(c,E') and
~r : R(e) --> R(s) be the R-homomorphisms sending £ onto 0,s',ee'
A
and re respectively. Then, the additive group (Lie GS) (R) is the
kernel of the h o m o m o r p h i s m

A A A
GS(~n) : Gs(R(e)) ----> Gs(R) ,

the s c a l a r m u l t i p l i c a t i o n by r is i n d u c e d by the a u t o m o r p h i s m
A A
GS(~ r) of Gs(R(s)) and the c o m m u t a t o r of two e l e m e n t s
A A
x , y £ (Lie GS} (R) E G s ( R ( ~ ) ) is the o n l y e l e m e n t Ix,y] such that

A A
GS(O) ([x,y]) = (X,Gs(1) (y))

A
where ( , ) stands for the usual commutator in the group Gs(R(£,e')).

A
(PI) (Lie GS) (~) is the Kac-Moody algebra associated to the system
( ~ ® A, (~i)0 ~ i ~ i' (hi)0 ~ i ~ i ) completed with respect to the natural
gradation (deg e i = I, d e g fi = -I, deg h i = 0)

A
(P2) The group Gs(C) coincides with the formal group over C
attached to S by any one of the construction processes described in
§§ 5 a n d 6; in p a r t i c u l a r , it c o n t a i n s (a c a n o n i c a l image of)
w x
Hom(A ,~ ) and its c e n t e r consists of all ~ 6 H o m ( A * , ~ ×) such that
~(~i ) = 0 for all i

A
(P3) Modulo its center, Gs(C) is the subgroup of Aut((Lie GS) (C))
generated by all converging e x p ad g , with g E (Lie GS) (~) (this
turns out to be identical with the adjoint group considered by
R.V. Moody [Mol] and J.I. Morita [Mo3]; about this group, cf. also the
 215

last sentence of this appendix).

A
(P4) The functor GS restricted to p r i n c i p a l ideal domains, together
with suitably defined functorial homomorphisms
A A
~i : SL 2 - - > GS , B : Hom(A*,? x) --> GS (which we leave as an exercise
to d e t e r m i n e explicitly), satisfies the axioms (i') to (iv') of
^
[Ti4], 7.5, and is c h a r a c t e r i z e d by them, once Gs(C),= ~i(~) , ~ (C)=
are given.

A
Those properties clearly indicate that the functor GS which we
are going to d e f i n e is the "right one", at least when restricted to
principal ideal domains but maybe also for general rings, considering
its fairly simple and natural definition (though it is c o n c e i v a b l e
that some algebro-geometric invariants of the ring, such as Pic R ,
should be brought into play).

Let e be an integer and let G be a q u a s i - s p l i t , simply

connected absolutely almost simple group defined over the
field K : Q(Z), whose splitting field over K is g e n e r a t e d by
the e-th roots of Z ; thus, e = I, 2 or 3 and, in the latter
case, G is of type D4 . Let S be a maximal split torus of G ,
the system of roots of G with respect to S , ~ the system of
root rays ("rayons radiciels": cf. [BT4], 1.1.2), i.e. of h a l f - l i n e s
R
~+
.a with a 6 • , T the centralizer of S in G and Ua (for
a 6 ~ ) the root subgroup corresponding to a . We also denote by
G,S,T,..., the groups of K - r a t i o n a l points of G,S,T, ....

Let now S = (A'(~i)0 ! i Z z'(hi)0 ~ i ~ Z) be defined as follows:

A = X,(S) = Hom(Mult,S) is t h e group of c o c h a r a c t e r s of S ;

(~1,...,el) is a b a s i s of © and -anv is t h e maximal root

if e = I or if e = 2 and G is of type A2n , and it is the
maximal "short" root in t h e remaining cases; h. is the coroot
i
associated ~.
1
 216

Varying e and the type of G , one gets all systems S satisfying

the conditions (At) to (A3) above in this way. If G has type
X , we say that S has type e~ . The Dynkin diagram representing
the GCM (aj(hi)) is g i v e n by the following table:

type of S diagra m

I~ extended Dynkin diagram of X

2A2n ,~7--1 I -'" 1 ~°,~ (n+1 vertices)

(n+1 vertices)

(n vertices)

3D4 I

We shall now choose a system of "&pinglages" of the Ua'S (cf.

[BT4], 4.1). This is a s y s t e m ( X a ) a 6 ~ where, for all a , xa is one
of t h r e e things:

(i) an i s o m o r p h i s m K --> Ua ;

(ii) an i s o m o r p h i s m K(Z I/e) --> Ua (here Z I/e denotes any

e-th root of Z and, when e = 3 , all cubic roots involved
m a y be c h o s e n equal);

(iii) an i s o m o r p h i s m H --> U a , where H is the p r o d u c t

K(Z I/2)" × ZI/2.K e n d o w e d w i t h the group structure

(*) (u,v) • (u',v') = (u+u', v+v'+(u°u ' - u'Ou))

in w h i c h o represents the nontrivial K-automorphism of K(Z I/2)

(observe that H is n o t h i n g else but the group Hi of [BT4], 4.1.15,
for ~ = I/2 ,) t r a n s f o r m e d by the automorphism (x,y) ~> (x,2y) of
the underlying variety).

In all c a s e s except 2A2n (i.e. when G is of type A2n and

 217

e = 2), we take for (x a) a cohere[:t system of "&pinglages" deduced

from a Chevalley-Steinberg system (cf. [BT4], 4.1.16). In o r d e r to
describe the system (x a) in the case 2A2n , let us c h o o s e an o r t h o -
gonal basis (ai)1 ~ i ~ n of the (relative) root lattice: thus,
© = {± a i, ± 2a i, ± a i ± a j with i ~ j] . For a E ~ , let @a denote the
automorphism of the source of xa defined as follows: if a contains
a l.+a3 (resp. a.-al3 ; resp. -a.-a.l3 )' Ca(k) = 2k (resp. k/2 ) and
if a contains ai (resp. -ai), @a(U,V) = (2u,4v) (resp. (u,v))
Finally, we set x a = x'a o @a ' where (Xa)a 6 9 is a c o h e r e n t system
of "~pinglagles", as in loc. cit.

Let us n o w d e s c r i b e a certain schematic root datum (T, (Ua)a 6 ~)

in G over the r i n g K = ~ [ Z , Z -I] (cf. BT4 , 3.1.1). The scheme
is the "canonical group-scheme associated with the torus T" , defined
as in [BT4], 4.4.5 (as in [BT4], it can be shown that T does not
depend on the w a y T is e x p r e s s e d as a p r o d u c t of tori of the form

Mult ) and the scheme U is the "imaoe by x "of:

L/K L a ~ a

the additive group-scheme canonically associated with the m o d u l e K

(resp. K[Z1/e])" in case (i) (resp. (ii)) (cf. [BT4], 1.4.1);

the group-scheme whose underlying scheme is c a n o n i c a l l y associated with

the m o d u l e H = K[Z I/2] x z l / 2 . K and w h o s e product operation is o i v e n
by (*) in case (iii) .

It is r e a d i l y verified, using the appendix of [BT4], that the system

(T, (Ua)) is i n d e e d a schematic root datum. By S e c t i o n 3.8.4 of [BT4],
there exists a unique smooth connected group-scheme G with generic
fibre G containing the d i r e c t product

aE~_
77
Ua x T x a6=~ + Ha

as an o p e n s u b s c h e m e ("big cell") (here, ~+ c ~ denotes a system of

positive root rays and ~_ = - ~ + ) . Finally, S being as above, the
announced functor ~S is d e f i n e d by

A
Gs(R) = @(R((Z))) ,

this group being given the n a t u r a l topology, induced by that of R((Z))

 2';8

Suppose now that R is a p e r f e c t field of c h a r a c t e r i s t i c e

(which implies that e = 2 or 3). There is a " n a t u r a l " isomorphism of
each Ua(R((Z))) onto R((Z)) , namely

-I
xa in case (i) ,

x a (r) ~ - > re in case (ii) ,

x a (r,r') ~--> r ' 2 + r 4 in case (iii) ,

and T(R((Z))) , which is a p r o d u c t of g r o u p s of the form R((Z)) × and

R ( ( z l / e ) ) × , is c l e a r l y isomorphic to the group T' (R((Z))) of r a t i o n a l
points of a s p l i t torus T' . It is then readily verified (using [BT3],
§ 10, and the a p p e n d i x of [BT4]), that, via those isomorphisms, the
system (T(R((Z))), ( U a ( R ( ( Z ) ) ) ) a 6 ~) "is" the standard root datum of
the g r o u p of r a t i o n a l points of an R((Z))-split simple group of type

Cn if A= = (ej(hi)) has type 2A2n

Bn if ~ has type 2~2n_1 ,

2~
Cn_ I if A has type Dn ,

F4 if A has type 2E 6 ,

G2 if A has type 334

This is the p h e n o m e n o n already mentioned in § 7 for the special case

of type 2E 6 .

Let us r e t u r n to the group-scheme @ . In the classical cases

2~ and 23 it can be g i v e n a more direct and m o r e elementary des-
m n'
cription. Here, we shall only briefly treat the types 2~ (the case
m
of 23 is s l i g h t l y m o r e c o m p l i c a t e d b e c a u s e one m u s t w o r k w i t h the
n
spin g r o u p ) . A c c o r d i n g as m = 2n-] or 2n , set I ={±],±2,...,±n}
or I = {0,±I, .... ±n} . Let V be the K[Z1/2]-module (K[ZI/2]) I
endowed with a coordinate system ~ = (zi)i 6 I ' let T denote the
K-automorphism of K[Z I/2] defined by T(Z I/2) = -Z 1/2 and c o n s i d e r
the h e r m i t i a n form

iT
h(z;z') = Z(Z'i T Z i + Z i Z_i) ,
 219

where i runs from I to n or from 0 to n according as m = 2n-I

or 2n . We represent by VK the module V considered as a K-module;
in it, we use the coordinate system (~,Z) = (xi,Yi)i61 , where

xi' Yi 6 K and z i = x i + Yi . ZI/2 . Separating the "real and imaginary

parts" of h , we get h = s + Z I/2 • a , where s and a are a
symmetric and an alternating bilinear form in VK respectively. Similar-
ly , the determinant in End V can be written d e t 0 + Z I/2 • d e t I ,
where det 0 and det I are K-polynomials in End V considered as a
K-module. Let q be the quadratic form q(~,y) = ~Is (_~ , y_ ; x ,y) in VK
The multiplication by Z I/2 is an automorphism J of the K-module VK .
Finally, the group-scheme @ (corresponding to the type 2~ ) can
- m

be described as the subgroup-scheme of G L ( V K) defined by the equations

g • a = a , g. q = q (hence g • s = s ), gJ=J, det0g = I, d e t l g = 0
In other words, if R is a K-algebra, G(R) is the subgroup of all
elements of SL(V®R[ZI/2]) preserving the (R-valued) "forms" a and
q . (For the case m = 2n , see [Ti4], 7.4.)

Now, consider again the case R : R((Z)) , where R is a perfect

field of characteristic 2 (in fact, any ring R such that the map
x F--> x 2 is a bijection of R [ Z I/2] onto R would do) Let V' (resp.
V") denote the R[Z1/2]-module, product of 2n+I (resp. 2n ) factors
R [ Z I/2] indexed by { 0 , ± I , .... +_n} (resp. {_+I, .... +-n ). In those modules,
we use again coordinates z where i runs through the same index
i n I/2 2
sets . In V' , consider the quadratic form q' (_z)_ = i__Z1 Z_l.Z + Z I "z0 '
n
and in V" , the alternating bilinear form a' (_z;_z') : i=E1 ( z ' i z i - z l z _ i )

If m = 2n-I ,V®R[Z I/2] can be identified with V" , hence with a

quotient of V' , the "bilinearization" and the "real part" (K-part) of
q' are the inverse images in V' of the "forms" aR and qR (with
obvious notational conventions), and it is easy to verify that the
projection V' --> V" induces an isomorphism SO(q') ~> G(R) . If
m = 2n , V@R[Z1/2]can be identified with V',the bilinear form h
R [ Z I/2 ]
is t h e inverse image of a' by the projection V' --> V" and, this
time, the latter induces an isomorphism @(R) -~> Sp(a') . Thus we have
found again the two isomorphisms obtained earlier in a different way.

A
The description of the functor GS associated to an arbitrary
system S of affine type, i.e. a system satisfying (At) but not
necessarily (A2) and (A3) now amounts to a combination of extension
 220

problems. In p a r t i c u l a r , when i :~Z.h i , one m u s t d e f i n e a c e n t r a l

A
extension of the a b o v e functor G by the m u l t i p l i c a t i v e g r o u p - s c h e m e
mult (I)., t h i s is r e l a t e d t o w o r k o f C. M o o r e [Mo2],
H. M a t s u m o t o [Ma3] a n d P. D e l i g n e [Dell. Note that if, w i t h the n o t a t i o n
used throughout this appendix, we a s s u m e e = I , we denote by Sad
the system obtained in t h e same w a y as S but replacing A by the
dual of the lattice of r o o t s and by @ad the split adjoint group-scheme
A
of the same type as G , t h e n the functor GSa d is n o t e q u a l to

R ~--> G a d ( R ( ( Z ) ) ) in g e n e r a l ; for instance, ~Sad(~) is the image of

the c a n o n i c a l map

G(C((Z))) --> @ad(C((Z))) ,

whose cokernel is i s o m o r p h i c to the center of G .

(1)As P. D e l i g n e p o i n t e d o u t to me, the w o r d " e x t e n s i o n " m u s t be u n d e r -

s t o o d h e r e in a " s c h e m a t i c s e n s e " ; one s h o u l d n o t e x p e c t the e x t e n s i o n
m a p to be s u r j e c t i v e for r a t i o n a l p o i n t s o v e r an a r b i t r a r y r i n g R .
 221

REFERENCES

[Bi] G.D. BIRKHOFF, A theorem on matrices of analytic function, Math,

Ann. 74 (1913), 122-133.

[Bo] N. BOURBAKI, Groupes et Algebras de Lie, Chap. IV, V, vI, Hermann,

Paris, 1968.

[BT]] A. B O R E L et J. TITS, Homomorphi~mes "abstracts" de groupes alg[b*~iques

simpl66, Ann. of Math., 97 (1973) , 499-571.
[BT2] F. BRUHAT et J. TITS, Groups a~g~briques s i m p l ~ sw% un corps local,
Proc. Conf. on local fields (Driebergen, 1966), Springer,
1967, 23-36.

[BT3] F. B R U H A T et J. TITS, Groupes r~ductifs s ~ un corps l o c ~ , I: Donn~es

r a d i c i e ~ e s valu~6s, Publ. Math. I.H.E.S. 41 (1972), 5-251.

[BT4] F. BRUHAT et J. TITS, Groupes r ~ d u c ~ f s sur un corps local, II: Schemas

en groupes, Existence d'une donn~e r a d i c i e l l e vm£u~e, P u b l . M a t h .
I.H.E.S. 60 (1984), 5-184.

[Chl] C. CHEVALLEY, Sur certains groupes simples, Tohoku Math. J. (2) 7

(1955), 14-66.

[Ch2] C. CHEVALLEY, S~minaise sur la c £ a s s i f i c a t i o n des group~ de Lie

alg~briques,2 vol., Inst. H. Poincar~, Paris, 1958,
mimeographed.

[Ch3] C. CHEVALLEY, C ~ a i ~ sch~m~ de group~ s e m i - s i m p l ~ , S@m.

N. Bourbaki, expos~ 219 (1980-1981), Benjamin, N e w York, 1966.

[Cu] C.W. CURTIS, Grou~ with a Bruhatdecomposition, Bull. Amer. Math.

Soc. 70 (1964), 357-360.

[D21] P. DELIGNE, Exte~tion de G(K) par K2(K) , S@minaire I.H.E.S.,

1979, non publi6.

[De2] M. DEMAZURE, Seh~m~ en group~ r ~ d u c t i f s , Bull. Soc. Math. Fr. 93

(1965), 369-413; cf. aussi le S~minaire de G~om~tg~e alg~brique,
de M. D E M A Z U R E er A. GROTHENDIECK, vol. 3, Springer Lecture
Notes in Math. n ° 153, 1970.

[DG] M. D E M A Z U R E et P. GABRIEL, Groupes alg~briques, I, Masson, Paris,

1970.

[Ga] H. GARLAND, The a ~ h m e t i c theory of loop groups, Publ. Math.

I.H.E.S. 52 (1980), 5-136.

[GW] R. G O O D M A N and N.R. WALLACH, S~%ucture and u~ttary cocyle r e p r ~ e n -

t~o~ of loop groups and the group of diffeomorphisms of the c i r c l e ,
Jour. r. angew. Math. 347 (1984), 69-133.
 222

JIM] N. IWAHORI and H. MATSUMOTO, On some Bruhat decomposition and the

structure of the Hecke ring of p - a d k c C h e v ~ e y groups, P u b l . M a t h .
I.H.E.S. 25 (1965), 5-48.

[Kal] V. KAC, An algebraic d e f i n ~ i o n of compact Lie groups, Trudy M.I.E.M.

5 (1969), 36-47.

[Ka2] V. KAC, Simple i~reducible Lie algebras of fim6te growth, I z v e s t i j a

Akad. Nauk S.S.S.R. (ser. mat.) 32 (1968), 1923-1967; E n g l i s h
transl., Math. USSR I z v e s t i j a 2 (1968), 1271-1311.

[Ka3] V. KAC, Infim6te dimensional Lie algebras, Birkh~user, Boston, 1983.

[KPl] V. KAC and D. PETERSON, Regular functions on c ~ a i n i n f i n i t e dkmen-

sionc~ groups, in A r i t h m e t i c and G e o m e t r y (ed. M. A R T I N and
J. TATE) , Birkh~user, Boston, 1983, 141-166.

[KP2 ] V. KAC and m. PETERSON, Cohomology of infinite-dimensional groups and

their flag varieties: algebraic poAt, Colloque Elie Caftan, Lyon
June 1984, to appear.

[Mal] R. MARCUSON, Tire'S system in generalized non-adjoint Chevalley groups,

J. of A l g e b r a 34 (1975), 84-96.

[Ma2] O. MATHIEU, S~% la cons$tuc~@n de groupes associ~s aux alg[bres de

Kac-Moody, C.R. Acad. Sci. s6rie I, 299 (1984), 161-164.
[Sa3] H. MATSUMOTO, Sur les sous-groupes arithm[tiques des groupes semi-simples
d~ploy~s, Ann. Sci. Ec. Normale Sup. 2 (4 - s6rie) (1969),
1-62.

[Mol] R.V. MOODY, A s i m p l i c i t y theorem for Chevalley groups defined by

generalized C~tan m o ~ c e s , p r e p r i n t , 1982.

[Mo2] C. MOORE, Group extensions of p - a d i c and adelic linear groups, P u b l .

Math. I.H.E.S. 35 (1968), 5-70.

[No3] J. MORITA, On adjoint Chevalley groups asso~ated with completed Euclidean

Lie ~gebras, Comm. in A l g e b r a 12 (1984) , 673-690.
[MT] R. MOODY a n d K. TEO, Tits systems with crystallographic Weyl groups,
J. Algebra 21 (1972) , 178-190.

[RS] Reduotion of Hmmi2~o~an

A. REIStAN and M. S E M E N O V - T J A N - S H A N S K I I ,
systems, affine Lie algebras and Lax e q ~ o n s , I, Invent. Math.54
(1979), 81-100; II, Invent. Math.63 (1981), 423-432.

[sw] G. SEGAL and G. WILSON, Loop groups and equations of KdV type, Publ.
Math. I.H.E.S. 61 (1985).

[Sh] I. SHAFAREVICH, On some infinite dimensional groups, II, I z v e s t i j a

Akad. Nauk S.S.S.R. (ser. mat.) 45 (1981); E n g l i s h transl.,
Math. USSR, I z v e s t i j a 18 (1982), 214-226.

[Sl] P. SLODOWY, S i n g ~ m ~ t ~ t e n , Kac-Moody Liealgebren, a s s o z i i e r t e G~tuppen

und Verallgemeinerungen, H a b i l i t a t i o n s s c h r i f t , Bonn, 1984.
 223

[Til] J. TITS, Groupes semi-simples isotopes, Coll. t h @ o r , gr. alg.,

C.B.R.M. ( B r u x e l l e s , 1962), L i b r a i r i e U n i v e r s i t a i r e , L o u v a i n ,
et Gauthier-Villars, P a r i s , 1962, 1 3 7 - 1 4 7 .

[Ti2] J. TITS, Buildings of sph~%ical type and finite BN-pairs, Springer

L e c t u r e N o t e s in M a t h . n O 386 (1974).

[Ti3] J° T I T S , R@su/a& de c o u r s , Annaire du Col!~e de F r a n c e , 81 e a n n & e

( 1 9 8 0 ) - 1 9 8 1 ) , 75-86.

[Ti4] J. TITS, R @ s u m @ de c o u r s , A n n u a i r e du Coll~ge de F r a n c e , 82 e an

82 e a n n @ e (1981-1982) , 9 1 - 1 0 5 .

[Vel] J.-L. VERDIER, A19~bres de Lie, systgmes hamiLto~er~, courbes alg~briqc~es

(d'aprgs M. A D L E R et P. v a n M O E R B E K E ) , S&m. N. B o u r b a k i ,
e x p o s @ n ° 566 ( 1 9 8 0 - 1 9 8 1 ) , S p r i n g e r L e c t u r e N o t e s in M a t h .
n ° 901 (1982), 85-94.

[Ve2] J.-L. VERDIER, Les repr~sent~io~ des alg~br~ de Lie affines: Applica-
tions ~ quelques problgmes de physique (d'aprgs E . DATE, M. J I M P B ,
M. K A S H I W A R A , T. M I W A ) , S@ra. N. B o u r b a k i , e x p o s @ n ° 596
(1981-1982) , A s t @ r i s q u e 9 2 - 9 3 (1983) , 3 6 5 - 3 7 7 .

[wi] R. W I L S O N , Euclidean Lie algebras are universal central extensions, in

Lie a l g e b r a s a n d r e l a t e d t o p i c s , S p r i n g e r L e c t u r e N o t e s in
Math. n ° 933, 2 1 0 - 2 1 3 .
 MODULAR POINTS, MODULAR CURVES, MODULAR SURFACES AND M O D U L A R FORMS

D. Zagier
U n i v e r s i t y of M a r y l a n d Max-Planck-Institut fur M a t h e m a t i k ,
C o l l e g e Park, MD 20742 D-5300 Bonn, FRG

This talk, instead of b e i n g a survey, will concentrate on a single

example, using it to i l l u s t r a t e two themes, each of w h i c h has been a
leitmotif of m u c h r e c e n t work in number theory and of much of the w o r k
report e d on at this A r b e i t s t a g u n g (lectures of Faltings, Manin, Lang,
Mazur-Soul6, Harder). These themes are:
i) special values of L-series as r e f l e c t i n g geometrical relation-
ships, and
ii) the analogy and interplay between classical algebraic geometry
over ~ and a l g e b r a i c geometry (in one dimension lower) over
Z, and more especially between the theory of complex surfaces
and the theory of a r i t h m e t i c surfaces ~ la A r a k e l o v - F a l t i n g s .
In particular, we will see that there is an intimate relation-
ship b e t w e e n the p o s i t i o n s of m o d u l a r curves in the h o m o l o g y
groups of m o d u l a r surfaces and the p o s i t i o n s of m o d u l a r points
in the M o r d e l l - W e i l groups of the J a c o b i a n s of m o d u l a r curves.
The p a r t i c u l a r example we will treat is the elliptic curve E
defined by the d i o p h a n t i n e equation

y(y - i) = (x + l)x(x - i) ; (I)

m o s t of w h a t we have to say applies in m u c h g r e a t e r generality, but

by c o n c e n t r a t i n g on one example we will be able to simplify or sharpen
many statements and make the essential points emerge more clearly.
The e x p o s i t i o n has been d i v i d e d into two parts. In the first
(§§1-5), which is e n t i r e l y expository, we d e s c r i b e various theorems
and c o n j e c t u r e s on e l l i p t i c and m o d u l a r curves, always centering our
discussion on the example (i). In particular, we e x p l a i n how one can
construct infinitely many rational solutions of (i) by a c o n s t r u c t i o n
due to Heegner and Birch, and how a result of Gross and the a u t h o r
and one of W a l d s p u r g e r lead one to surmise a relationship between
these solutions and the c o e f f i c i e n t s of a m o d u l a r form of h a l f - i n t e g r a l
weight. The second part (§§6-9) is d e v o t e d to a proof of this relation-
ship.
I would like to thank G. van der Geer and B. Gross for useful
discussions on some of the m a t e r i a l in this talk.
 226

I. The elliptic curve E and its L - s e r i e s .

Multiplying bo~h sides of (i) by 4 and adding 1 we obtain the
Weierstrass form

2
Yl = 4x3 4x + i (Yl = 2y - i); (2)

from this one calculates that the curve E has d i s c r i m i n a n t A = 37

and j-invariant j = 21233/37. Of c o u r s e , (i) and (2) are affine
equations and w e should really work with the p r o j e c t i v e equations
y 2 z - yz 2 = x 3 - xz 2 and yl2 z = 4x 3 - 4xz 2 + z 3 whose points are the
points of (i) or (2) t o g e t h e r with a "point at i n f i n i t y " (0:i:0). The
points of E over any field k form a group with the p o i n t at i n f i n i t y
being the o r i g i n and the g r o u p law d e f i n e d by P + Q + R = 0 if P,Q,R
are collinear; the n e g a t i v e of a p o i n t (x,y) of (i) or (x,y I) of
(2) is (x,l-y) or (x,-Yl) , r e s p e c t i v e l y . In a c c o r d a n c e with the
philosophy of m o d e r n geometry, we try to u n d e r s t a n d E by looking at
the g r o u p s E(k) of k-rational points for v a r i o u s fields ~.

k = ~ : The set of real solutions of (i) is easily sketched; it

consists of two c o m p o n e n t s , ~ ~ x ~ ~ and y f x, w h e r e ~ = -1.107...,
= 0 . 2 6 9 5 .... y = 0.8395... are the roots of 4x 3 - 4x + 1 = 0 (the
group E(~ ) is i s o m o r p h i c to S 1 x ~/2Z). We h a v e the real period

~i =

!
E ~
dx

Yl
- 2
S Y /4x3_ 4x+ 1
dx
- 2.993458644...; (3)

the n u m e r i c a l value is o b t a i n e d by u s i n g the formula ~i = ~/M(/7"i,

Y/~-8), where M(a,b) denotes the arithmetic-geometric mean of G a u s s
(M(a,b) = lira a n = l i m b n for a,b > 0, w h e r e {a0,b 0} = {a,b},

a +b
{an+ I, bn+ I} = {--~-~, n/~n}).

k = ~: As w e l l as the real period we have the imaginary period

~2 = 2
I Y

6 /4x3-4x+l
dx
- 2.451389381...i (4)

(which c a n be c a l c u l a t e d as i~/M( 8/~/~-~, /y-s)). The set of c o m p l e x

points of the (projective) curve E is i s o m o r p h i c to the complex torus
C/Z~ 1 + Z ~ 2 via the W e i e r s t r a s s p-function:
 227

¢/Ze I + Z~ 2 * E({)

p ' (z) +i) ,

z , • (p (z) , 2

1 ' 1 i
p(z) = -~+ Z ( 2-
z m,n (z-m~l-nw 2) (m~l+n~02) 2)
!

(Z means [ ), w h i c h satisfies
m,n (m,n)~(0,0)

p.2 = 4p3 - g2 p - P3'

g2
=
60 [ , ........ 1
4 -
4~ 4
~
(i + 240
oo n3 )
.2zin~i/~ 2 : 4,
m,n (m~l+n~ 2) 3~ 2 n=l e -i

, ~ 5
g3 = 140 [ 1 _ 8w6 ( n )
m,n (m~l+nw2)[ 27~J2~ 1 - 504 n=l~ e 2 Z i n ~ i / e 2 - 1 =-I.

k = ~: The M o r d e l l - W e i l group E(~) is infinite cyclic with generator

P0 = (0,0), the first few m u l t i p l e s being

15
2P 0 = (l,0), 3P 0 = (-i,i), 4P 0 = (2,3), 5P0=(~,~), 6P 0 = (6,-14)

and their negatives -(x,y) = (x,l-y). If we w r i t e nP 0 as (Xn,Y n)

and xn as Nn/D n with (Nn,D n) = i, then

log max(INnl , IDnl ) r~ cn 2 (Inl _~ ~)

with a certain positive constant c (in other words, the number of

solutions of (i) for w h i c h x has n u m e r a t o r and d e n o m i n a t o r less than
B is a s y m p t o t i c to 2c-i/2(iog B) I/2 as B + co). This constant is
called the height of P0 and d e n o t e d h(P0) ; it can be c a l c u l a t e d via
an a l g o r i t h m of Tate (cf. [14], [ 2]) as
co

h(P 0) = [ 4-i-ilog(l+ 2t. 2 - 2 t . 3 + t.4),

i=l l 1 1

where the t i (=l/x2i) are d e f i n e d inductively by

t I . i, ti+. 1 .(I + 2ti. 2 2ti 3 + ti4)/(4ti 4t'31 + t'4)'i

and we find the n u m e r i c a l value

h(P 0) = 0.0511114082... (5)

Similarly one can define h(P) for any P 6 E(~); clearly h(nPo) =
 228

n2h (P0) .

k : Z/pZ: Finally, we c a n look at E o v e r the f i n i t e field k = ~/pZ,

p~37 prime. Here E(k) is a f i n i t e g r o u p of o r d e r N(p) + I, w h e r e

N(p) = #{x,y(mod p) [y2 _ Y ~ x 3 _ x (mod p)}.

We c o m b i n e the i n f o r m a t i o n contained in all t h e s e n u m b e r s into the

L-series

LE(S) = II 1 1 (6)
pfi37 l + N P ) - P + ~ s i+ 1
P P 37 s

the s p e c i a l behavior of 37 is due to the fact t h a t A ~ 0 (mod 37),

so that the r e d u c t i o n of E over Z/37Z is s i n g u l a r . Multiplying
out, we o b t a i n LE(S) as a D i r i c h l e t series

LE(S ) = [ a(~), (7)

n=l n

the f i r s t few a(n) b e i n g g i v e n by

n Ii~ 2 3 4 5 6 7 8 9 i0 ii 12 13 14 15
-Z (8)
a(n) Ii ~ 2 .........-3
....... 2 -2 6 -i 0 6 4 -5 -6 -2 2

Since clearly N(p) S 2p, the p r o d u c t (6) and the s u m (7) c o n v e r g e

absolutely for Re(s) > 2; in fact, IN(p)-pl is less t h a n 2/p
(Hasse's theorem), so we h a v e a b s o l u t e convergence for Re(s) > 3/2.
We will see in §3 that LE(S) extends to an e n t i r e function of s
and s a t i s f i e s the f u n c t i o n a l equation

LE(S) := (2~)-s37s/2F(S)LE(S) = -LE(2-S); (9)

in p a r t i c u l a r , LE(S) vanishes at s = i. The B i r c h - S w i n n e r t o n - D y e r

conjecture relates the i n v a r i a n t s of E over ~ , ~ and Z/p~
by p r e d i c t i n g that

o r d s = 1 LE(S) = rk E(~) = 1

and that

d (i0)
d--s LE(S) [s=l = 2h(P0)'~I'S

with a certain positive integer S which is s u p p o s e d to be the o r d e r

of the m y s t e r i o u s Shafarevich-Tate group ]]I. S i n c e the f i n i t e n e s s of
 22g

iLL is not known (for E or any other elliptic curve), this last
statement cannot be checked. However, L~(1) can be computed numer-
ically (cf. §3), and its value 0.3059997738... strongly suggests
(cf. (3), (5)) the equation

L~(1) = 2h(P0)w I, (Ii)

i.e. (i0) with S = I; the truth of this equation follows from equation
(18) below.

2. Twists of LE; the numbers A(d) .

Let p be a prime congruent to 3 (mod 4) which is a quadratic
residue of 37 and consider the "twisted" L-series

L E p(S) = ~ (~) a(~) (12)

' n=l n

((~) = Legendre symbol). The proof of the analytic continuation of

LE will also show that each LE, p continues analytically and has a
functional equation under s ~ 2-s. Now, however, the sign of the
functional equation is +, so we can consider the value (rather than
the derivative) of LE at s = i, and here one can show that
,P
2w 2
= ~ A(p)
LE'p(1) i/p

with ~2 as in (4) and some integer A(p). The value LE,p(1) can
be calculated numerically by the rapidly convergent series LE,p(1) =

2 [ (~) a(n) e -2~n/p 3/~ (cf. §4) so we can compute A(p) for small
n=l P n
p. The first few values turn out to be

p } 3 7 ii 47 67 71 83 107 127 139 151 211 223 (13)

A(p) 1 1 1 1 36 1 1 '0 1 0 4 9 9 "

More generally, LE,d(S) can be defined for all d satisfying (~7)=I,

-d = discriminant of an imaginary quadratic field K (just replace
(~) in (12) by (~), the mirichlet character associated to K/~),
and we still have

2e2 A(d) (14)

LE,d(1) -
i/J

for some A(d) 6 Z; the first few values not in (13) are
 230

d 4 40 84 95 104 IiI 115 120 123 136 148

A(d) 1 4 ............1........ 0 0 1 36 4 9 16 9 (15)

The most striking thing about the v a l u e s in (13) and (15) is t h a t they
are all squares. This is e a s i l y understood from the B i r c h - S w i n n e r t o n -
Dyer conjecture: the D i r i c h l e t series LE, d is j u s t the L-series of
the "twisted" elliptic curve

E<d>: -dy 2 = 4x 3 - 4x + i, (16)

so A(d) should be e i t h e r 0 (if E<d> has a rational point of

infinite order) or (if E<d>(~) is finite) the o r d e r of the Shafare-
vich-Tate group of E<d> and hence a perfect square (since this group,
if finite, has a non-degenerate (~/~)-valued alternating form).
Surprisingly, even though we are far from knowing the B i r c h - S w i n n e r t o n -
Dyer conjecture or the finiteness of lll(E<d>), we can p r o v e that A(d)
is a s q u a r e for all d, and in fact p r o v e it in two d i f f e r e n t ways:
On the one hand, a theorem of W a l d s p u r g e r leads to the f o r m u l a

2
A(d) = c(d) , (17)

where c(d) (c Z) is the d th F o u r i e r coefficient of a c e r t a i n modular

3
f o r m of w e i g h t ~. On the o t h e r hand, a theorem of Gross and m y s e l f
gives the formula

4Wl~ 2
L E,( 1 ) L E ' d ( 1 ) - i/d h(Pd) (18)

for a c e r t a i n explicitly constructed point ("Heegner point") Pd in

E(Q); writing Pd as b(d) times the generator P0 of E(~) and
comparing equation (18) with (14) and (ii), we o b t a i n

A(d) = b(d) 2 (19)

We thus have two c a n o n i c a l l y given square roots b(d) and c(d) of

the integer A(d), and the q u e s t i o n arises whether they are equal.
The o b j e c t of this paper is to g i v e a geometrical proof of the fact
that this is so. First, however, we m u s t define b(d) and c(d) more
precisely, and for this we n e e d the m o d u l a r description of the elliptic
curve E, to w h i c h we n o w turn.

3. The modular curve E.

The essential fact about the elliptic curve E is t h a t it is a
modular curve. More precisely, let F be the subgroup of SL2(~)
 231

g e n e r a t e d by the group

F0(37 ) = {(2 ~ ) 6 SL2(Z) Ic H 0 (mod 37)}

and the m a t r i x w37 =

0
(3/~7
-1/3/]7,
0 J" This group acts on the upper half-

plane ~ in the usual way and the q u o t i e n t ~/F can be c o m p a c t i f i e d

by the a d d i t i o n of a single cusp ~ to give a smooth c o m p l e x
curve of genus i. We c l a i m that this curve is isomorphic to E({) ;
more precisely, there is a (unique) isomorphism

~/F O { ~ } >~ E(C) (20)

sending ~ to 0 (E(~) and such that the p u l l - b a c k of the canonical

dx dx
differential ~ - Yl is -2~if(T)dT, where

f(T) = q-2q2-3q3+2q4-2q5+6g6-q7+6q9+... (q = e 2~iT) (21)

is the unique n o r m a l i z e d cusp form of w e i g h t 2 on F, i.e. the

unique h o l o m o r p h i c f u n c t i o n f on ~ satisfying

.at+b. b
ftc--~) = (cT+d) 2f(T) (T ~ ~,(~ d ) 6 F) (22)

and f(T) = q + O(q 2) as Im(T) + ~. This c l a i m is simply the asser-

tion of the W e i l - T a n i y a m a c o n j e c t u r e for the elliptic curve under con-
sideration, and it is w e l l - k n o w n to specialists that the W e i l - T a n i y a m a
c o n j e c t u r e can be checked by a finite c o m p u t a t i o n for any given elliptic
curve; moreover, the p a r t i c u l a r curve E was treated in detail by
Mazur and S w i n n e r t o n - D y e r in [ii]. Nevertheless, for the b e n e f i t of
the reader who has never seen an example of a modular p a r a m e t r i z a t i o n
worked out, we will give the details of the proof of (20); our treat-
ment is somewhat d i f f e r e n t from that in [Ii] and may make it clearer
that the a l g o r i t h m used would apply equally well to any elliptic curve.
The reader who is a c q u a i n t e d with the c o n s t r u c t i o n or who is w i l l i n g
to take (20) on faith can skip the rest of this section.
We have two quite d i f f e r e n t d e s c r i p t i o n s of the i s o m o r p h i s m (20),
d e p e n d i n g whether we use the algebraic model (i) or the a n a l y t i c
model {/~i + Z~2 for E(f). We start with the a l g e b r a i c model. The
p r o b l e m is then to show the existence of two F - i n v a r i a n t and holomor-
phic functions ~(T) and n(T) satisfying

2 3 ~' (~)
q(T) - ~(T) = ~(T) - ~(T), -2~if(T) (23)
2q(T)-I

dx
(this gives a map as in (20) with the right p u l l - b a c k of 2--~L~; that
 232

it is an i s o m o r p h i s m is then e a s i l y c h e c k e d ) . Equations (23) imply

that ~ and n h a v e p o l e s of o r d e r 2 and 3, r e s p e c t i v e l y , at ~,
and r e c u r s i v e l y determine all c o e f f i c i e n t s of their L a u r e n t expansions.
Calculating o u t to 9 terms, w e see that t h e s e e x p a n s i o n s must begin

{(T) = q-2+2q-l+5+9q+18q2+29q3+51q4+82q5+131q 6+ . . . .
(24)
~(T) = q-3+3q-2+9g-t+21+46q+92q2+t80q3+329q4+593q5+-.-
So far we h a v e not used the fact t h a t f is a m o d u l a r f o r m on F;
we c o u l d h a v e t a k e n any p o w e r series f(T) = q+... and u n i q u e l y solved
(23) to get L a u r e n t series ~(T) = q-2+..., ~(T) = q-3+... However,
since ~ and ~ are s u p p o s e d to be F-invariant functions w i t h no
poles in ~, and since f is a m o d u l a r f o r m of w e i g h t 2, the two
functions f4 = f2~ and f6 = f3~ m u s t be h o l o m o r p h i c modular forms
on ? of w e i g h t 4 and 6, r e s p e c t i v e l y . B u t the s p a c e Mk(F) of
modular forms of w e i g h t k on F is f i n i t e - d i m e n s i o n a l for any k
and o n e can o b t a i n a b a s i s for it by an a l g o r i t h m i c procedure (e.g.,
u s i n g the E i c h l e r - S e l b e r g trace formulas, but we w i l l find a s h o r t c u t
here), so w e c a n i d e n t i f y f4 and f6 f r o m the b e g i n n i n g s of t h e i r
Fourier expansions. O n c e one has c a n d i d a t e s f4 and f6' one d e f i n e s
= f4/f 2 and ~ = f6/f3; t h e s e are then a u t o m a t i c a l l y modular func-
tions on F, and the v e r i f i c a t i o n of (23) reduces to the v e r i f i c a t i o n
of the two f o r m u l a e

f62 _ f6f3 = f43 f4f4 , f(2f 6 _ f3) _ i i(2f4f

2~ ' _ ff~) , (25)

which are identities between modular forms on F (of w e i g h t s 12 a n d

8, r e s p e c t i v e l y ) and h e n c e can be p r o v e d by c h e c k i n g finitely many
terms of the F o u r i e r expansions. In our c a s e the d i m e n s i o n of Mk(F)
equals [5k]
6 + 3[ ] for k > 0, k even, so M2(F) is g e n e r a t e d by
f while M4(F) and M6(F) have dimension 6 a n d 8, r e s p e c t i v e l y .
However, we w i l l be a b l e to i d e n t i f y f4 and f6 without calculating
bases for t h e s e spaces. The s p a c e M2(F0(37)) is the d i r e c t s u m of
M2(?) = ~f and the 2 - d i m e n s i o n a l s p a c e of m o d u l a r forms F of w e i g h t
1
2 on F0(37) satisfying F(-~-~) = -37T2F(T). AS a b a s i s of this
latter s p a c e we can c h o o s e the t h e t a - s e r i e s

@(T) = [ qQ(a,b,c,d) = l+2q+2q2+4q3+2q4+4q5+Sq6+4q7+10qS+... ,

a,b,c,d6~

Q(a,b,c,d) = ~(4b+c-2d) + (2a+c+d) 2 2

= a 2 + 2b 2 + 5c 2 + 10d 2 + ac + a d + b c - 2bd
 233

and the c u s p form

h(T) : ~@(~) - ~E2(T)

1 - = q + q3 _ 2q4 _ q 7 +... ,

3 dqnd The four f u n c -

where E2(T) = ~ + [ is an E i s e n s t e i n series.
d,n>0
3V}d
tions f2, @2, @h and h2 lie in the s p a c e

U = {F 6 M4(F) lordT=A(F) c 2, o r d T = B ( F ~ 4},

where A and B a r e the f i x e d p o i n t s in ~/F0(37) of o r d e r 2 and 3,

respectively, because any f u n c t i o n in M2(F0(37)) must vanish at A
and v a n i s h d o u b l y at B. For the same reason, f4 = f2 lies in U
(recall t h a t ~ has no p o l e s in ~) ; and s i n c e U has c o d i m e n s i o n
2 in M4(F) (a g e n e r a l function in M4(F) satisfies O r d A F = 2r,
OrdBF = 3s+l for some r,s a 0), t h e s e five f u n c t i o n s m u s t be l i n e a r l y
dependent. Looking at the f i r s t few F o u r i e r coefficients, we f i n d
that f4 m u s t be g i v e n by

As to f6' we o b s e r v e t h a t the f u n c t i o n

@ (T) = ~(T)2/~(37T) 2 + 37~(37T)2/n(T)

is F-invariant and h o l o m o r p h i c in ~' and has a t r i p l e p o l e at ~,

so m u s t be a l i n e a r c o m b i n a t i o n of ~,~ and i; l o o k i n g at the f i r s t
few F o u r i e r coefficients we find that ~ = ~ - 55 + 6, so f6 must
be ~f3 + 5f4f _ 6f 3. As e x p l a i n e d above, o n c e w e have our c a n d i d a t e s
f4 and f6 it is a f i n i t e c o m p u t a t i o n to c h e c k (25) and thus establish
2
that T ~ (f4(T)f(T) : f6(T) : f(T) 3) maps ~/F U {~} to E(~) c ~ (~)
as c l a i m e d .
For the s e c o n d d e s c r i p t i o n of the m a p (20), we d e f i n e a function
@:~ ÷ ~ by

~(T) = 2~i f(~')dT' = _ q + q2 + q3 -~ql 4 +5q2 5 - ... . (26)

Y

.aT+b.
From @' = - 2 ~ i f and (22) it f o l l o w s t h a t the d i f f e r e n c e ¢~c--~-~J-%(T)
ab
is a c o n s t a n t for all Y = (c d ) 6 F. C a l l this c o n s t a n t C(y); clearly
C:F ÷ ¢ is a h o m o m o r p h i s m . The t h e o r y of E i c h l e r - S h i m u r a implies
that the image i = C(F) is a l a t t i c e in { with g2(i) and g3(A)
rational integers. S i n c e we can c a l c u l a t e @(T) and h e n c e C(y)
 234

numerically (the series in (26) converges rapidly), we can c a l c u l a t e a

basis of i n u m e r i c a l l y and get g2 and g3 exactly. The result
g2 = 4, g3 = -i shows that i is the lattice ~w I + Zw 2 of §i, and
the identity ¢(yT) - ~(T) = C(Y) shows that ~ ~ {/i factors
through F. We thus obtain a map ~/F-~E({) = f/i such that the
pull-back ¢*(dz) equals -2zif(T)dT, as asserted. In practice, it
is easier to c a l c u l a t e the image in E of a p a r t i c u l a r point T 6
by using (26) and r e d u c i n g modulo i than by using the first d e s c r i p -
tion of the map (20).

4. Modular forms a t t a c h e d to E
The most important c o n s e q u e n c e of the modular d e s c r i p t i o n of the
elliptic curve E is that the L-series of E equals the L-series of
the m o d u l a r form f, i.e. that the numbers a(n) in (7) are p r e c i s e l y
the Fourier c o e f f i c i e n t s in (21). This follows from the E i c h l e r -
Shimura theory (cf. [13]). As a consequence, the function LE defined
in (9) has the integral r e p r e s e n t a t i o n

LE(S ) f tS-ldt f ( i t ) (ts-i tl-S)dt,

3~7 3~7
0 1

from w h i c h the analytic c o n t i n u a t i o n and functional equation are obvi-

ous. Differentiating and setting s = 1 we find

foo oo
/~7
~-~ L{(1) = ~L~ (i) = 2 f( it) log t dt = 2 [ a(n)G(2zn),
3/77 n=l 3/~7
1

with

e-Xtlog t dt = i eiU d u ¢
G(x) =
I1
I x
U

and since there are w e l l - k n o w n expansions for G(x), this can be used
!
to c a l c u l a t e LE(1) = 0.30599... to any desired d e g r e e of accuracy,
as m e n t i o n e d in §i. Similarly, if -d is the d i s c r i m i n a n t of an
imaginary q u a d r a t i c field in w h i c h 37 splits, then the "twisted" form
f*(~) = [(--~d)a(n)qn-- is a cusp form of w e i g h t 2 and level 37d 2 satisfy-
n 2 22,
ing f * ( - i / 3 7 d T) = -37d T f (T), so
(.co .
LE,d(S) := (2~)-s37s/2dSF(s)LE d(S) = ] f*(~) ( t s - l + tl-s) dt,
' /
1

from which we deduce the functional equation LE,d(S) = LE,d(2-s and

 235

the f o r m u l a LE,d(1) = 2 ~ (~)a(n) e - 2 Z n / d 3/37 mentioned in §2.

n=l
In p a r t i c u l a r , we c a n c a l c u l a t e the n u m b e r s Aid) defined by (14)
approximately a n d hence, since they are integers, exactly.
The o t h e r m o d u l a r form which will be i m p o r t a n t to us is the f o r m
of w e i g h t 3/2 associated to f under Shimura's correspondence.
Around ten y e a r s ago, Shimura [12] d i s c o v e r e d a relationship between
modular forms of a r b i t r a r y even w e i g h t 2k and m o d u l a r forms of h a l f -
integral weight k + 1/2. This was studied subsequently by m a n y o t h e r
authors. In p a r t i c u l a r , Kohnen (in [ 8] for f o r m s of l e v e l 1 and in
[ 9] for forms of odd s q u a r e f r e e level) showed how Shimura's theory
c o u l d be r e f i n e d by i m p o s i n g congruence conditions modulo 4 on the
Fourier expansion so as to g e t a p e r f e c t correspondence between appro-
priate s p a c e s of forms of w e i g h t s 2k and k + 1/2. The result in the
case k = 1 and p r i m e level is the f o l l o w i n g ([ 9], T h e o r e m 2) :

Theorem 1 (Shimura; Kohnen). For N prime and s e {il} let $3/2

d e n o t e the space of all f u n c t i o n s g(T) satisfying
i) g(T)/8(T) 3 , w h e r e @(T) is the s t a n d a r d t h e t a - s e r i e s [ q n2 ,
n6Z
is i n v a r i a n t u n d e r F0(4N) ;
ii) g(T) has a F o u r i e r development ~ c(d)q d with c(d) = 0
d>0
if -d - 2 o__rr 3 (rood 4) or (~) = -~.
Let S2(F0(N) ) denote the s p a c e o_ff c u s p f o r m s o_~f weigh_______tt 2 o__nn F0(N)
satisfying f(-I/NT) = sNT2f(T) . Then d i m $3/2(N) = dim $2(0(N))
and for e a c h H e c k e eigenform f = [ a ( n ) q n e S~(F0(N)) there is a l-
dimensional space of g • S~/2(N) whose Fourier coefficients are re-
lated to t h o s e of f b_yy

~ _ _ 2
a(n) c(d) = [ (~)c(~d 2 ) ( n e ~ , -d a f u n d a m e n t a l disorim- (27)
rln r inant).
r>0

In our c a s e N = 37, e = +i and the s p a c e $2(F0(37)) is o n e -

dimensional, s p a n n e d by the f u n c t i o n f of (21). Theorem 1 therefore
asserts that there is a u n i q u e function

2~idT
g(~) = ~ c(d)e
d>0
-d ~ 0 or l ( m o d 4)
(-d/37) = 0 or 1

such that g(T)/e(T) 3 is F0(148)-invariant and the F o u r i e r coeffi-

cients c(d) (normalized, say, by c(3) = i) s a t i s f y (27). It is not
an e n t i r e l y trivial matter to c a l c u l a t e these coefficients; a method
 236

for d o i n g so, and a t a b l e up to d = 250, w e r e g i v e n in [ 3, pp. i18-

120, 145] in c o n n e c t i o n with the t h e o r y of "Jacobi forms-" W e give
a short table:

d 3 4 7 ii 12 16 27 28 36 40 44 47 48 63 64 67 71 75 83 ... 148
(28)
c(d) 1 1 ~ 1 -i -2 -3 3 -2 2 -i -i 0 2 2 6 1 -i -i ... -3

W e now c o m e to the t h e o r e m of W a l d s p u r g e r [15], m e n t i o n e d in §2,

which relates these coefficients to the v a l u e s at s = 1 of the t w i s t e d
L-series LE,d(S). A g a i n we need a r e f i n e m e n t due to K o h n e n [i0, T h e o -
r e m 3, Cor. i] w h i c h gives a p r e c i s e and simple identity in the s i t u a -
tion of T h e o r e m I:

T h e o r e m 2 (Waldspurger; Kohnen). Let S(Fo(N)), g =

f = ~a(n)q n e S 2
~c (d) qd e $3/2(N)
t c o r r e s p o n d as in T h e o r e m i. Let -d be a f u n d a -
mental discriminant with (~) = 0 __°r s and let Lf,d(S) be the
associated convolution L-series [(~)a(n)n -s. Then

Lf,d(1) = 3z I]flI2 Ic(d) 12 (29)

IIgJl2 /~

where

rlfJi = I If()12dudv' NgH2 =I Lg()12v-1/2dudv (3O)

~/F0(N 1 ~/F0(4N) (T = u + iv)

are the norms of f and g in the P e t e r s s o n metric. (Note that the

identity is i n d e p e n d e n t of the c h o i c e of g, s i n c e r e p l a c i n g g by
Ig (I e ~*) multiplies both l;glI* and Ic(d) I2 by 11[ 2.)

Actually, the e x a c t c o e f f i c i e n t in (29) is not too r e l e v a n t to us,

for k n o w i n g that Lf,d(1) is a f i x e d m u l t i p l e of c(d) 2 / / d implies
that the n u m b e r s A(d) d e f i n e d by (14) are p r o p o r t i o n a l to c(d) 2,
2
and c a l c u l a t i n g A(3) = c(3) = 1 we d e d u c e (17). Then going back
and s u b s t i t u t i n g (17) and (14) into (29) we d e d u c e 3~]IfIl2/;]g]!2 = 2~2/i.
w e n o w show (since the r e s u l t w i l l be n e e d e d later) that

Ilfll2 = ~i~2/2~2i, (31)

it t h e n f o l l o w s that IIgl]2 = 3~i/4~. To p r o v e (31), w e r e c a l l from

§3 t h a t t h e r e is an i s o m o r p h i s m @ from ~/F U {~} to E(~) = ~/A
with @*(dz) = 2~if(T)dT. Since IF:F0(37)] = 2 we h a v e
 237

2~211f!I 2 = 4~ 2 / If(T) 1 2 d u d v = I l-2nif(T) 1 2 d u d v

]

dx dy = ~iw2/i
J
{/A

as c l a i m e d .

5. Heeqner points on E
In this section we d e s c r i b e a construction which associates to
each integer d > 0 a point Pd e E(~). These are the "modular
points" of the title, since their construction depends o n the m o d u l a r
description of E given in §3.
We assume first that -d is a f u n d a m e n t a l discriminant, i.e. the
discriminant of an imaginary quadratic field K. We consider points
b+i/d
T e .]( of the form '< - with
2a

a,b { Z, a > 0, 371a, b °" - -d (mod 4a). (32)

-d
If (~7) = -i, there are no s u c h T and w e set Pd = 0; o t h e r w i s e
the set of ~ is i n v a r i a n t under F and there are h distinct points
71,...,Th modulo the a c t i o n of F, w h e r e h = h(-d) is the c l a s s
number of K. The theory of c o m p l e x multiplication shows that these
points are individually defined over a finite extension H of
(the H i l b e r t class field of K) and collectively over @ (i.e. their
images in ~6/F are p e r m u t e d by the a c t i o n of the G a l o i s group of H
over ~). Hence the sum ~(T I) + . . . + ~(Th) , w h e r e ~:~{/F ÷ E(~) is
the m a p constructed in §3, is in E(~). M o r e o v e r this s u m is d i v i s -
1
ible by u, w h e r e u is ~ the n u m b e r of u n i t s of K (= 1,2 or 3)
if 37~d and u = 2 if 371d ; this is b e c a u s e each point 7j ~
is the fixed point of an e l e m e n t of F of o r d e r u. We define
Pd e E(~) by
h
up d = [ ~(~j) ; (33)
j=l

this is w e l l - d e f i n e d because E(~) is t o r s i o n - f r e e . If d is not

f u n d a m e n t a l , we d e f i n e Pd0 the same way but with the e x t r a condition
, b2+d,
(a,D,--~) = 1 in (32) (now h(-d) is the c l a s s number of a c e r t a i n
non-maximal order of K, and the p o i n t s ~l,...,Th ~ ~/F are d e f i n e d
over the c o r r e s p o n d i n g ring class field), and then set Pd = ~ d P d / e 2"
el
The d e f i n i t i o n of Pd just given is a s p e c i a l case of a c o n s t r u c -
 238

t i o n d u e to H e e g n e r and B i r c h (cf. [i]) and in g e n e r a l would yield

rational points in the J a c o b i a n of X0(N)/w N (X0(N) : XU(cusps)/F0(}~)).
From a modular p o i n t of view, a point T e ~/F0(N)/w N classifies
isomorphism classes of u n o r d e r e d p a i r s of N-isogenous elliptic curves
{EI,E 2} over { (namely E 1 = {/~+ZT, E 2 = {/~+N2Z<, w i t h the i s o g e n i e s
E 1 + E2, E2 + E1 i n d u c e d by N.id~ and id{, respectively), and the
points ~l,...,Th correspond to t h o s e w i t h c o m p l e x m u l t i p l i c a t i o n by
an o r d e r @ of ~(_/L-~) (namely E 1 = ~/~, E 2 = ~/~, where ~ = ~+~<
is a f r a c t i o n a l @-ideal and n an i n t e r g r a l O-ideal of n o r m N). A
general formula for the h e i g h t s of t h e s e "Heegner points" was p r o v e d
recently by B. Gross and m y s e l f [ 4]; the r e s u l t in our special case
becomes

Theorem 3(Gross-Zagier) : Suppose -d is a f u n d a m e n t a l discriminant

with (~7) = 1 and let Pd c E(~) b e the p o i n t d e f i n e d by (32). Then
the h e i g h t of Pd is g i v e n b y
/£
h(P d) - Li(1) LE, d (i) •
8~2prfH 2

(To get this s t a t e m e n t from [ 4], take X = 1 in T h e o r e m 2 there,

noting that vf, 1 = uP d and L(f,l,s) = L E ( S ) L E , d ( S ) ; the h e i g h t
in [ 4] is o n e - h a l f t h a t on E because it is c a l c u l a t e d on X0(37)
which is a d o u b l e c o v e r of E.)
In v i e w of e q u a t i o n (31), T h e o r e m 3 is e q u i v a l e n t to the f o r m u l a
(18) g i v e n in §2. As e x p l a i n e d there, this f o r m u l a gives both equation
!
(i0) for LE(1) and the r e l a t i o n s h i p (19) b e t w e e n A(d) and the
integers b(d) defined by Pd = b(d)P0" The e q u a l i t y b(d) 2 = c(d) 2
suggested comparing the v a l u e s of b(d) and c(d). Note that the
numbers b(d) are n u m e r i c a l l y calculable: o n e finds the h F-ineguiv-
alent solutions of (32) by r e d u c t i o n theory, computes the c o r r e s p o n d i n g
.b+i/d.
values %[--~---] by (26), adds the r e s u l t i n g complex numbers; modulo
i = Z ~ I + Z ~ 2 , the r e s u l t m u s t be a m u l t i p l e of the p o i n t
1
P0 = - . 9 2 9 5 9 . . .
+ 2~2" Thus for d : 67 we have h = 1 and

P67 = % ( ~ ) = .40936... s 6P0(mod i),

so b(67) = 6; for d = 83 we h a v e h = 3,

55+i8<~, 19+i~, ~) = (.5~i . + 1.225...i)

P83 = 9[ 22-2----; + %( 7q~---) + %(55+

+ (.194... - .570...i) + (.194... + .570...i) ~ -Po(mod i),

 239

so b(83) = -1; for d = 148 we have h = 2,

2P148 = ~( ) + <b(7 + = .19189... - .60125... 5 -6P0(mod A),

/37

so b(148) = -3. In this way one can make a table of the multiples
b(d). Such a table (up to d = 150) was computed b y B. Gross and
J. Buhler, while I was independently computing the Fourier coefficients
c(d) by the m e t h o d mentioned in §4; the letter with their data
arrived in G e r m a n y on the v e r y morning that I had completed my com-
putations and d r a f t e d a letter to them, and the p e r f e c t agreement of
the two t a b l e s gave ample reason to c o n j e c t u r e the followin~:

Theorem 4. b(d) = c(d) for all d.

The remainder of this paper is d e v o t e d to the p r o o f of this result.

6. Curves on H i l b e r t modular surfaces

In v i e w of the uniqueness{ clause in T h e o r e m i, w h a t we n e e d to do
to p r o v e T h e o r e m 4 is s i m p l y to show that ~ b(d)q d belongs to
+
S 3 / 2 ( 3 7 ) , i.e. that the p o s i t i o n s of the H e e g n e r points in the M o r d e l l -
Weil group of E are the F o u r i e r coefficients of a m o d u l a r f o r m of
weight 3/2. This statement is r e m i n i s c e n t of a t h e o r e m of H i r z e b r u c h
and the a u t h o r [ 7] a c c o r d i n g to w h i c h the p o s i t i o n s of c e r t a i n modular
curves in the h o m o l o g y group of a m o d u l a r surface are the F o u r i e r co-
efficients of a m o d u l a r f o r m of w e i g h t 2. Since this result is not o n l y
very analogous to the one we want, but w i l l actually be u s e d to p r o v e
it, we recall the exact statement.
Let p be a p r i m e congruent to 1 (mod 4) and let 0 = ~ +~l+/P
2
be the r i n g of integers in ~(/p). The group PSL2(O) (Hilbert modular
group) acts on ~C × 7C bv

(a~l+b a'T2+b' = ~ ((0), '~,~2 ~ ~),

Mo(TI,<2) = c~1+d' c ' < 2 + d ') (M !(ca b) PSL2

where ' denotes conjugation in Q(/p)/~. The quotient ~ × ~/SL2(@)

can be n a t u r a l l y compactified by the a d d i t i o n of finitely many points
("cusps"), and w h e n the singularities thus introduced are resolved by
cyclic configurations of r a t i o n a l curves according to H i r z e b r u c h ' s
recipe [ 6] the r e s u l t i n g surface Y = Y is a n e a r l y smooth compact
P
algebraic surface (it still has quotient singularities coming from the
points in ~£ × ~' with a non-trivial isotropy group in PSL2(O), so
it is a r a t i o n a l homology manifold or "V-manifold"). The middle homo-
logy of Y splits as
 240

c
H2(Y ) = H2(Y ) ~ <Sl> ~ . . . ~ <Sr> , (34)

where SI,...,S r are (the h o m o l o g y classes of) the curves used in the
r e s o l u t i o n s of the cusp singularities and H~(Y) consists of h o m o l o g y
classes o r t h o g o n a l to the Sj; the homology groups in (34) are taken
with c o e f f i c i e n t s in Q.
For each integer N > 0 there is an a l g e b r a i c curve TN c y
defined as follows. C o n s i d e r all equations

ATIT2 + % %'

with A,B c Z, % e @, and %%' + ABp = N. Each one defines a curve in

× ~ isomorphic to ~ and the union of these curves is i n v a r i a n t
under SL2(O); TN is d e f i n e d as closure in Y of the image of this
N) =
u n i o n in ~ x ~/SL2(O ) . If (~ -I, there are no solutions of
%%' + ABp = N and TN is empty. If (~) = +i then TN is irreduc-
ible (all equations (35) are e q u i v a l e n t under PSL2(e)) and isomorphic
to the m o d u l a r curve X0(N). The main result of [ 7] is

T h e o r e m 5 (Hirzebruch-Zagier). Let [T~] denote the p r o j e c t i o n t__oo

c
H2(Y) of the h o m o l o g y class of TN in the s p l i t t i n g (34). Then the

power series ~ - c~ 2~iNT

[TN]e is a modular form of w e i g h t 2, level p
N=l
and N e b e n t y p u s (~).

Here "modular form of w e i g h t 2, level p and Nebentypus" means a

modular form F(T) satisfying F,
taT+b"
c~) = (~) (cT+d) 2F(T) for
a d
(c b) £ F0(p); w h e n we say that a power series w i t h c o e f f i c i c e n t s in

H~(Y) is such a form we m e a n that each c o m p o n e n t (with respect to a

basis of H~(Y) over ~) is. Alternatively, if [X] is any h o m o l o g y

class in H2(Y) , then the power series [ ( X o T ~ ) ~ ~iNr , w h e r e (XoT~)

denotes the i n t e r s e c t i o n p a i r i n g of IX] and [T~], is a m o d u l a r
form of the specified type, now with o r d i n a r y n u m e r i c a l Fourier coeffic-
ients. In particular, this is true for X = T M, one of our special
curves on Y. In fact the proof of T h e o r e m 5 in [ 7] c o n s i s t e d in
c a l c u l a t i n g the i n t e r s e c t i o n numbers (TMoT ~) e x p l i c i t l y and showing
that they were the F o u r i e r c o e f f i c i e n t s of a m o d u l a r form. The formula
o b t a i n e d for (TMOT~), in the case when N and M are coprime, was
2
o c (4NM-x) (MN) (36)
(TM T N) = [ H ---7--- + Ip
X2<4NM
2
x ~4NM(mod p)
 241

where

H(d) = [ h' (-d/e 2)

e 2 [d

(h'(-d) = h(-d) for d > 4, h' (-3) = 1/3, h' (-4) = 1/2) and I (n)
P
is a c e r t a i n arithmetical function whose definition w e do n o t repeat.
The proof of (36) was geometrical: the p h y s i c a l intersection points
of TM and TN in X × ~/PSL2(~ ) are in i:I correspondence with
certain equivalence classes of b i n a r y quadratic forms and are counted
by the first term in (36), while the t e r m Ip(MN) counts the inter-
section points of TM and TN at i n f i n i t y and the intersection of
TM with the c o m b i n a t i o n of c u s p - r e s o l u t i o n curves S. which was
c 3
removed from TN to get T N.

7. Heegner points as intersection points of m o d u l a r curves on m o d u l a r

surfaces
Now suppose that p is a p r i m e satisfying p H 1 (mod 4), (3~7) = i,
and (for l a t e r purposes) p > 2"37, say p = i01. As a l r e a d y mentioned,
the c u r v e T37 on Yp is in this case isomorphic to X0(37) =
~U{cusps}/r0(37). For instance, if p = i01 we c a n get an e q u a t i o n
(35) for T37 by t a k i n g A = B = 0 and I = 21+2 I/~-i, an e l e m e n t of
0 of n o r m 37; then the solution of (35) is g i v e n parametrically by
{(IT,I'T), T e ~} and the m a t r i c e s M e SL2(@) w h i c h p r e s e r v e this
set are those of the form a
(c/l bl)
d with a db ) E r0(37), so we get
(c
a degree 1 map ~/r0(37) ÷ ~ × ~/SL2(@) and hence a map X0(37) ÷ YI01"
On y we h a v e an e x t r a involution t which is i n d u c e d by the invol-
P
ution (TI,T2) ~ (Tz,T~) of ~ × ~, and this induces the involution
w37 on T37 = X0(37), so our curve E ~ ~U{cusps}/r can be found on
the q u o t i e n t surface Y/t. However, since all TN are invariant under
t and there is no d i f f e r e n c e (except a factor of 2) b e t w e e n the inter-
section theory of l-invariant curves on Y or of t h e i r images in
Y/l, we will continue to w o r k on the surface Y rather than the q u o -
tient surface Y/l, which has a one-dimensional singular locus.
In §5 we constructed for each d > 0 a set of (i+ (~-7))
-d H (d) points
in X0(37), namely the set of r o o t s of q u a d r a t i c equations aT2+bT+c=0
with b2-4ac=-d and 371a. (If d is of the form 3n 2 or 4n 2
then H(d) is n o t an integer a n d we are using the c o n v e n t i o n that a
fixed point
of an e l e m e n t of o r d e r u in r0(37) is to be c o u n t e d
1
with multiplicity q in ~/r0(37); f r o m n o w on w e w i l l ignore this
technicality.) Call this set Pd" The point Pd { E(@) was (one-
half of) the s u m of the images of the p o i n t s of Pd in E. If we
 242

worked on X0(37), or on some other X0(M) of h i g h e r genus, we w o u l d

have to take the s u m in the J a c o b i a n of the c u r v e rather than on the
curve itself, i.e. Pd would be the p o i n t of Jac(X0(M)) represented
by the d i v i s o r Pd - deg(Pd)" (~) of d e g r e e 0.
The geometric content of (36) is that the intersection
points of TN and TM in X × ~/PSL(0) are the p o i n t s of Pd for
148N-x 2
certain d, namely those of the form - - , i.e.
P

T37 n TN = O P U D (37)
txl< 14/Y~Z~ (14SN-x2)/P
x2~148N (mod p)

where D is c o n t a i n e d in the part of Yp at infinity (resolutions

of the c u s p singularities); here when we w r i t e union we of c o u r s e
mean for the p o i n t s to be c o u n t e d with appropriate multiplicities, i.e.
we are w o r k i n g with divisors rather than just sets of points. If we
simply count the p o i n t s in (37), i.e. replace each Pd by its
degree, we o b t a i n the n u m b e r s (36), and Theorem 5 tells us that these
are the F o u r i e r coefficients of a m o d u l a r f o r m of w e i g h t 2, level p,
and N e b e n t y p u s (~). If i n s t e a d we add the p o i n t s in (37) in the
Jacobian of T37, i.e. replace each Pd by Pd' t h e n we w i l l
deduce f r o m this that the c o r r e s p o n d i n g statement holds:

Proposition: For N > 0 define B(N) bz

148N-x 2
B(N) = ~ b ( - - )
x2<148N P
x2Z148N (mod p)

N
with b(d) as in §5. Then ZB(N)q is a m o d u l a r form of weight 2,
level p and N e b e n t y p u s (~ .

Proof. Let M denote the set of all m o d u l a r forms of the specified

type, so that Theorem 5 asserts

~(T~oX)q N ~ M for all [X] ~ H2(Y) . (38)

The space M is f i n i t e - d i m e n s i o n a l and has a basis consiting of m o d -

ular forms with rational Fourier coefficients. Hence there is an in-
finite set R of finite relations over Z defining M, i.e. a set
R whose elements are sequences

R = (r0,rl,r 2 .... ), r N ~ Z, rN = 0 for all but finitely many N

 243

and such that

C(N)q N c M ~ ~ rNC(N) = 0 (VR c R). (39)

N=0 N=0

(For instance, one could find integers NI,...,N d with d = dim M

th
and such that the Nj Fourier coefficients of forms in ~4 are
J

l i n e a r l y i n d e p e n d e n t ; then for e a c h N we h a v e a relation C(N) =

d
IjC(Nj) with rational numbers ll,...,ld, and we c o u l d take for
j=l
R the set of t h e s e relations, each multiplied by a c o m m o n denominator.)
Equation (38) now implies that

rN(T~oX) = 0
N=I

for all R e R, and since this holds for all homology classes X, we
must have [r N [ T N ] = 0 in H 2(Y,@) . Since T Nc is a l i n e a r c o m b i n a t i o n
of T and curves S. coming f r o m the c u s p r e s o l u t i o n s , this m e a n s
n 3
co r
[ r N [ T N] + [ sj [S~] = 0 (40)
3
N=I j =i

in H2(Y,~) for some rational numbers S l , . . . , s r. Multiplying by a

further common denominator we can assume that thes. are also integers
]
and that the r e l a t i o n (40) holds in i n t e g r a l homology. But the H i l b e r t
modular surface Y is k n o w n to be s i m p l y connected, so the e x a c t
sequence

0 = HI(y,0) ÷ H I ( y , 0 *) ÷ H2(y,z) (0 = s t r u c t u r e sheaf of Y)

induced from 0 ÷ ~ ÷ 0 + 0* + 0 shows that any divisor on Y which

is h o m o l o g o u s to 0 is l i n e a r l y equivalent to 0. Hence the relation
(39) implies that the d i v i s o r [rNT N + ~sjSj is the d i v i s o r of a
meromorphic function on Y, i.e. there is a m e r o m o r p h i c function
on Y which has a zero or p o l e of o r d e r rN on e a c h TN (resp. sj
on e a c h Sj) and no o t h e r zeros or poles. If we r e s t r i c t ¢ to T37,
then it f o l l o w s that the zeros and poles of # occur at the intersec-
tion points of T37 with other TN and at the cusps, and in f a c t
(by (3S~) that

divisor of ¢ T37 = [ rN [ P + d
N{I x2<148N (148N-x2/p)

x2{148N (mod p)
 244

where d is a d i v i s o r with support concentrated at the cusps. Take

the image in E, o b s e r v i n g that the c u s p s map to 0, and a d d the p o i n t s
obtained; since the p o i n t s of a p r i n c i p a l divisor s u m to zero and the
points of Pd s u m to b(d)P0, we deduce [rNB(N) = 0 with B(N) as
in the P r o p o s i t i o n . The desired result now follows from equation (39).

8. Completion of the p r o o f
We are now nearly done. For each N > 0 define

C(N) = ~ c(148N-x2),
P
x2<148N
x2~148N (mod p)

where c(d) are the F o u r i e r coefficients defined in §4. Then

(pd+x2)/148
G(z) := ~ C(N)q N = [ c(d)q
N>0 d>0
xeZ
pd+x2~0 (mod148)

= g(pz) @ (z)IU148,

2 th
where @ = [qX and U is the m a p w h i c h picks out every m coef-
m
ficient of a F o u r i e r e x p a n s i o n , i.e.

1
%(z) IUm - m j(mod m)

Since g is a m o d u l a r f o r m of w e i g h t 3/2 and 8 one of w e i g h t 1/2,

and since U maps modular forms to m o d u l a r forms of the same w e i g h t ,
m
it is c l e a r that G(z) is a m o d u l a r f o r m of w e i g h t 2; a r o u t i n e cal-
culation shows that it has level p and N e b e n t y p u s (~). Hence both
G(z) and F(z) = [B(N)q N belong to the finite-dimensional space M.
Moreover, since b(d) = c(d) for small d by the c a l c u l a t i o n s men-
tioned in §5, the first Fourier coefficients of F and G agree, and
this suffices to s h o w F = G. Specifically, with p = i01 the a g r e e -
ment of c(d) and b(d) for d < 150 implies the a g r e e m e n t of B(N)
and C(N) for 1 Z N S i00, and this is m o r e than enough to e n s u r e
that F = G (it w o u l d suffice to h a v e agreement up to N = 9). Hence
B(N) = C(N) for all N. We claim that this implies b(d) = c(d) for
all d. Indeed, s u p p o s e i n d u c t i v e l y t h a t b(d') = c(d') for all
-d
d' < d. If ( 7 ) = -i or -d ~ 2 or 3 (mod 4) then c(d) and b(d)
are b o t h zero and there is n o t h i n g to prove. Otherwise we can find an
 245

integer n with

2
n E -pd (mod 148), InT ~ 37.
pd+n 2
Take N - 148 " Then in the equations

B(N) = Z b(-148N-x2) , C(N) 148N-x 2

P
Z c(--) P
x2<148N X2<I48N
x2EI48N (mod p) X2=-I48N (mod p)

the n u m b e r s ±n occur as v a l u e s of x a n d all o t h e r values of x

are larger in a b s o l u t e value because Inl S 37 < ½ p by assumption.
Thus B(N) equals 1 or 2 times b(d) plus a certain linear
combination of b(d') with d' < d, and C(N) equals the same multiple
of c(d) plus the same linear combination of lower c(d'), so the
equality B(N) = C(N) and the inductive assumption b(d') = c(d')
imply that b(d) = c(d) as d e s i r e d .

9. Generalization to o t h e r modular curves

Our exposition so far was simplified by s e v e r a l special properties
of the elliptic curve E: that it was actually isomorphic to a m o d u l a r
curve rather than just covered by one, that its M o r d e l l - W e i l group
had rank one and no torsion, etc. We end the p a p e r by d i s c u s s i n g to
what extent the results proved for E generalize to o t h e r curves.
First, we c o u l d replace E by an a r b i t r a r y elliptic curve whose
L-series coincides with the L - s e r i e s of a m o d u l a r form f of w e i g h t
2 and some (say, prime) level N, w i t h f(-~)-- = NT2f(T). Then we
would again have a covering map % : X 0 ( N ) / w N + E, Heegner points
Pd E E(~) for all d > 0 (with Pd = 0 if -d ~ 0 (rood 4N) ) , and
a relationship c(d) 2 ~ h(P d) for the F o u r i e r c o e f f i c i e n t s c(d) of
+
a modular form in S3/2(N ) corresponding to f as in T h e o r e m i. We
could then ask whether all the Pd belong to a o n e - d i m e n s i o n a l sub-
space <P0 > of E(~)/E(~)tor s and, if so, whether the c o e f f i c i e n t s
b(d) defined by Pd = b(d)P0 '{ E(~) ® @ are p r o p o r t i o n a l to the
Fourier coefficients c(d). More generally, we c o u l d forget elliptic
curves entirely and simply start with a modular curve X0(N) or
X0(N)/w N (still, say, with N prime). The construction of §5 y i e l d s
Heegner points Pd in the J a c o b i a n of this curve over @. To a v o i d
torsion we tensor with ~ and w r i t e V = Jac(X0(N)/WN) (~) ~2Z ~" The
Hecke algebra acts on V the same w a y as it a c t s on cusp forms of
weight 2, so V ~ ~ splits as @f Vf, where the f are Hecke eigen-
 246

forms f = [ a(n) g n in M~(F0(N)) (normalized by a(1) = I) and Vf

is the subs~ace of V ® ~ on which the n th Hecke operator acts as
multiplication by a(n). For each f we define Pd,f as the compon-
ent of Pd in Vf. The Fourier coefficients a(n) will be in Z if
f corresponds to an elliptic curve E defined over @; in that case
Vf is isomorphic to E(~) ® ~ and we are back in the situation de-
scribed before. In general the a(n) will be integers in an algebraic
number field Kf c ~, the Fourier coefficients c(d) of the form in
+
S3/2(N) corresponding to f can also be chosen to lie in Kf, and
the main theorem of [4] combined with Theorem 2 tells us that h(P d f)
2
is proportional to c(d) This suggests that the right g e n e r a l i z a t i o n
of T h e o r e m 4 is:

Theorem 6. Let f, c(d) be as above. Then Pd,f = c(d)P0 for all

d and some P0 ~ Vf. In particular, the projections Pd,f of the
Heegner points all lie in a one-dimensional subspace o_~f Vf.

Theorem 6 is equivalent (because of the uniqueness clause in Theorem

1 and the way the Hecke operators act on Heegner points) to the follow-
ing apparently weaker theorem:

Theorem 6' The powe~ series d~0 pdqd __is_a modular form of weight
3/2 and level N.

(As with T h e o r e m 5, this means that [ Pd qd ~ V[[q]] belongs to the

+
subspace V ® $3/2(N) or, in more d o w n - t o - e a r t h terms, that each
component of this power series, with respect to a fixed basis of V
over is a modular form in S~/2(N).)
How can we prove these theorems? The argument of §§6-7 permits
us to embed our modular curve in the Hilbert modular surface Y for
P
any prime p H 1 (mod 4) with (~) = 1 and to prove that the power
series

~ ( ~ P(4NM_x2)/p )q M
M x2<4NM
x2~4NM (mod p)

is a modular form (with coefficients in V) of weight 2, level p and

Nebentypus. To deduce T h e o r e m 6' we would need the following asser-
tion:

Let h(T) b_~e a power series o_~f the form

 247

b(d)q d
d>0
-d~square (mod 4N)

with N ~rime, and suppose that the power series

= ( q M - ~ -
(4NM-x2))
h(pT) 8 (T){U N [ I b
M>0 x2<4NM
X 2 ~ 4 N M ( m o d p)

is a modular for_~m o f w e i g h t 2, level p and N e b e n t y p u s (~) for every

prime p ~ 1 (mod 4) with ( ) = 1. Then h belongs to S3/2(N).

This a s s e r t i o n is e x t r e m e l y likely to be true. The a r g u m e n t of §8

proves it -- even if the hypothesis on h(pT)8(T)IU N is made for only
one prime p > 2N -- under the a d d i t i o n a l a s s u m p t i o n that one p o s s e s s e s
a candidate g = [ c(d)q d e S;/2(N) for h with c(d) = b(d) for
s u f f i c i e n t l y many Values of d. Thus the m e t h o d of proof we used for
N = 37 can be used for any other fixed v a l u e of N if we do a finite
a m o u n t of computation. To get a general proof of T h e o r e m s 6 and 6'
along these lines one would need either to prove the a s s e r t i o n above
or else to g e n e r a l i z e the geometric proof in some way (perhaps by
u s i n g Hilbert m o d u l a r surfaces of a r b i t r a r y discriminant, for w h i c h
the i n t e r s e c t i o n theory has been w o r k e d out by H a u s m a n n [5]).
In any case, however, we would like to have a proof of T h e o r e m 6
using only intrinsic p r o p e r t i e s of the m o d u l a r curve, rather than its
g e o m e t r y as an embedded s u b m a n i f o l d of an a u x i l i a r y m o d u l a r surface.
Such a proof .has been given by B. Gross, W. Kohnen and myself. It is
a direct g e n e r a l i z a t i o n of the m a i n result of [4]: instead of a for-
mula for the height h(P d) of a Heegner point, we give a formula for
the height p a i r i n g (Pd,Pd,) of two H e e g n e r points, w h e r e
( , ): V × V ÷ ~ is the bilinear form a s s o c i a t e d to the q u a d r a t i c
+
form h. The formula implies that ~ (Pd,Pd,)q- belongs to $3/2 (N)
d>0
for each d i s c r i m i n a n t d', and T h e o r e m 6' follows.
Finally, we m e n t i o n that the correct g e n e r a l i z a t i o n of T h e o r e m 4
to c o m p o s i t e levels N should be f o r m u l a t e d u s i n g the theory of
"Jacobi forms" d e v e l o p e d in [3] rather than the theory of m o d u l a r
forms of h a l f - i n t e g r a l weight. This, too, will be c a r r i e d out in the
joint work w i t h Kohnen and Gross m e n t i o n e d above.
 248

Bibliography

[1] B. Birch, Heegner points of elliptic curves, Symp. Mat., Ist. di

Alta Mat. 15(1975), 441-445.

[2] J. Buhler, B. Gross and D. Zagier, On the c o n j e c t u r e of B i r c h and

S w i n n e r t o n - D y e r for an elliptic curve of rank 3, to appear in
Math. Comp. (1985).

[3] M. Eichler and D. Zagier, The T h e o r y of Jacobi Forms, .... to appear

in P r o g r e s s in Mathematics, Birkh~user, B o s t o n - B a s e l - S t u t t g a r t
(1985).

[4] B. Gross and D. Zagier, Points de Heegner et d 6 r i v 6 e s de fonctions

L, C.R. Acad. Sci. Paris 297(1983), 85-87.

[5] W. Hausmann, Kurven auf H i l b e r t s c h e n Modulflichen, Bonnet Math.

S c h r i f t e n 123(1980).

[6] F. Hirzebruch, H i l b e r t modular surfaces, L'Ens. Math. 19(1973),

183-281.

[7] F. H i r z e b r u c h and D. Zagier, I n t e r s e c t i o n numbers of curves on

H i l b e r t modular surfaces and m o d u l a r forms of Nebentypus,
Inv. Math. 36(1976), 57-113.

[8] W. Kohnen, Modular forms of h a l f - i n t e g r a l w e i g h t on F0(4),

Math Ann. 248(1980), 249-266.

[9] W. Kohnen, New forms of h a l f - i n t e g r a l weight, J. reine Angew.

Math. 333(1982), 32-72.

[i0] W. Kohnen, Fourier c o e f f i c i e n t s of m o d u l a r forms of h a l f - i n t e g r a l

weight, to appear in Math. Ann. (1985).

[ii] B. Mazur and H.P.F. Swinnerton-Dyer, A r i t h m e t i c of Well curves,

Inv. Math. 25(1974), 1-61.

[12] G. Shimura, On m o d u l a r forms of h a l f - i n t e g r a l weight, Ann. of

Math. 97(19731, 440-481.

[13] H.P.F. S w i n n e r t o n - D y e r and B. Birch, E l l i p t i c curves and modular

functions, in M o d u l a r F u n c t i o n s of One V a r i a b l e IV, Lecture
Notes in Math. 476, Springer, B e r l i n - H e i d e l b e r g - N e w York (1975),
2-32.

[14] J. Tate, Letter to J-P. Serre, Oct. I, 1979.

[15] J.-L. Waldspurger, Sur les c o e f f i c i e n t s de Fourier des formes

m o d u l a i r e s de poids demi-entier, J. Math. pures et appl. 60
(1981), 375-484.
 EIGENVALUES OF THE DIRAC OPERATOR

Michael Atiyah
Mathematical Institute
O x f o r d O X I 3LB

§i. The Theorems

In r e c e n t years mathematicians have learnt a great deal from

physicists and in p a r t i c u l a r from the work of Edward Witten. In a

recent preprint [3], Vafa and Witten have proved some striking results

about the eigenvalues of the Dirac operator, and this talk will present

their results. I shall concentrate entirely on the mathematical parts

of their preprint leaving aside the physical interpretation which is

their main motivation.

The mathematical context is t h e following. We fix a compact

Riemannian spin manifold M of dimension d, and denote by D the

Dirac operator of M acting on the spin bundle S. In a d d i t i o n if

we are given a hermitian vector bundle V with a connection A we

can define the extended Dirac operator:

DA : S 8 V + S ~ V.

In t e r m s of
an orthonormal basis e of tangent vectors DA is g i v e n
d 3
l o c a l l y by E e~V~, where V~ is t h e covariant derivative in the
3 3 3
j=l
e.-direction and e. a c t s on s p i n o r s b y Clifford multiplication. In
3 3
particular DA depends on A only in t h e O-order term, i.e. if

B is a second connection on V, then DA - DB is a multiplication

operator not involving derivatives.

The operator DA is s e l f - a d j o i n t and has discrete eigenvalues

lj, both positive and negative, which we will suppose indexed by

increasing absolute value so t h a t

l~ll ~ i~21 ~ ....

The questions which Vafa and Witten address themselves to c o n c e r n

 252

the way in w h i c h the ~ d e p e n d on A (and V) : the metric on M

3
is a s s u m e d fixed throughout. More precisely they are interested in

getting uniform upper bounds. The simplest and most basic of t h e i r

results is

THEOREM I. There is c o n s t a n t C (__depending on M but not on V o__rr

A), such that Illl _< c.

More generally there is a u n i f o r m estimate for t h e n-th eigen-

value:

THEOREM 2. There is a c o n s t a n t C' (depend_ ing q n M but n o t on V,

A o__rr n) such that Ilnl s C ' n I/d.

Remarks. i) The asymptotic formula i n 1/d is a v e r y general

n

result for e i g e n v a l u e s of elliptic operators, but Theorem 2 is m u c h

more precise.

2) Theorem 1 does not hold for the L a p l a c e operator AA of V. To

see this just consider d = 2 and V to be a line-bundle of c o n s t a n t

curvature F : then I1 = IF] ~ ~ with the Chern class of V. This

emphasizes that the uniformity in T h e o r e m s 1 and 2 is w i t h respect to

the continuous parameter A and a l s o w i t h respect to the d i s c r e t e

parameters describing the t o p o l o g i c a l type of V.

3) The inequalities in Theorems 1 and 2 go in t h e opposite direction

to the Kato inequalities for eigenvalues of L a p l a c e type operators.

This had, in principle, been conjectured by physicists on t h e g r o u n d s

of F e r m i o n - B o s o n duality.

For odd-dimensional manifolds there are even stronger results,

namely:

THEOREM i*. If d is odd, there exists a constant C, so t h a t every

interval of length C, contains an e i g e n v a l u e of D A.

 253

THEOREM 2*. If d is odd, there exists a constant C~ so t h a t everj

interval of lengt~ C ~ n I/d contains n eigenvalues.

Note that Theorems i* a n d 2* a r e d e f i n i t e l y false in e v e n

dimensions. To see this recall that, when d is even, S decomposes

as S + ~ S- and DA is of the form

0 A
(i.i)

so t h a t
DA :
I
D A2 = D A D A+ @ D A+ DA, and t h e non-zero eigenvalues of the t w o

factors +
DAD A and DA+ DA coincide. If V has large positive

curvature then typically DAD~ will have a zero-eigenvalue of l a r g e

multiplicity while D~D~ will be a 'large' positive operator. Hence

D A2 will have a large gap between its O - e i g e n v a l u e and its first non-

zero eigenvalue. Moreover this gap tends to i n f i n i t y with the size

of the c u r v a t u r e of V. When d = 2 and V is a l i n e - b u n d l e of

constant curvature it is just the first Chern class of V which

determines the s i z e of t h e first gap.

§2. The even proof

Although the theorems we have just stated appear purely analytical

results, involving upper bounds on e i g e n v a l u e s , it is a r ~ a r k a b l e

feature of t h e w o r k of V a f a and Witten that the proofs are essentially

topological. To understand how this comes about I will consider

first the case when the d i m e n s i o n d is even. Then, as o b s e r v e d at

the end of §i, the spinors decompose and DA takes the form given in

(i.i). In p a r t i c u l a r a O-eigenvalue of DA arises whenever either

D A+ or DA has a non-trivial nullspace N A+ or NA respectively.

Next recall that the i n d e x of D A+ is d e f i n e d as

 2~

index D A+ = d i m N A+ - d i m NA

so t h a t a non-zero value for index D A+ forces DA to have a

O-eigenvalue. On the other hand index D A+ is a purely topological

invariant, given by an explicit formula [i] involving characteristic

cohomology classes of V and M. Hence, whenever the index,

computed topologically, is non-zero we have a O-eigenvalue for DA

(for all connections A on the given bundle V) and so trivially

Theorem 1 holds.

For d even Theorem 1 therefore has significant content only for

those bundles V for which the index formula gives zero. To treat

these the key idea is now the following. Suppose we can find a

connection A on V so that
o

(i) DA has a O-eigenvalue.

o

(ii) IIDA- DA II ~ C
o

then it w i l l follow that the smallest eigenvalue of DA does not

exceed C. Now we cannot actually find such a connection on V

itself but we can find one on some multiple NV = V ~ CN of V, and

this will do equally well since the only effect of taking multiple

copies of DA is to increase the multiplicity of each eigenvalue.

We now proceed as follows. First choose a bundle W' so that

the index of D+ on S + ~ V ~ W' is non-zero. From the index

+
formula (and the assumption that the index of D on S ® V is

zero) it is enough to take W' to be the pull-back to }4 of a

generating bundle on S 2d (i.e. with C2d : (d - I)!) by a map

M ÷ S 2d of degree i: this makes the index equal to dim V. Thus

for any connection B' on W' (which combines with A to give a

connection say A' on V ® W') the operator DA~

+ has a non-zero

index. Hence DA, has a zero-eigenvalue.

 255

Next choose an orthogonal complement W" to W', i.e. a bundle

so t h a t

(2.1) W' @ W" --- M × CN

and fix a connection B" on W" (defining, together with A, a

connection A" on V ~ W"). The operator

DA,~A,, = DA, ~ DA,,

still of course has a zero-eigenvalue (since DA, has). On the

other hand A = A' @ A" is a c o n n e c t i o n on

o

V ~ (W' ~ W") = V ~ C N = NV

and so c a n be compared with the connection NA (once w e have fixed

the isomorphism (2.1)). Comparing the corresponding Dirac operators

we find

(2.2) DA - DNA = B
o

where B is the matrix valued 1-form which describes the connection

B' ~ B" in the trivialization given by (2.1). Since B is q u i t e

independent of V and A we get a uniform constant C = [IBII and

this completes the proof of Theorem 1 in t h e even case.

Note that the simple formula (2.2), which is essential for the

proof, depends on the fact that the highest order part of DA is

independent of A.

To prove Theorem 2 (for even d) we proceed in a similar manner

but this time we pull back the bundle W' (and its complement W")

from Sd by using maps of d e g r e e n. The index formula then shows

that +
DA, has index n dim V. Theorem 2 then follows easily if one

can show that the constants C : lIB11 g r o w like n I/d. When M is

a torus T and n = rd (with r an integer) this follows by using

 256

the covering map Td ÷ Td given by x + rx : since B is a 1 - f o r m

(with matrix values) it p i c k s up a factor r. For general M one

applies this construction to a s m a l l box in M and the case of g e n e r a l

n follows by interpolation.

§3. The odd proof

If w e replace M by M x S I, where S1 is t h e circle, the

eigenvalues lj of mA get replaced by ±/15 + m2 where m runs

]
over the integers. The smallest eigenvalues are therefore the same

on M and on M × S I. This means that theorem 1 for d even, when

applied to M × S I, immediately yields Theorem 1 for d odd. A

similar but more careful count of eigenvalues shows that Theorem 2 for

d even also implies Theorem 2 for d odd.

Notice also that conversely, if w e first establish Theorems 1 and

2 for d odd, they then follow for d even. In fact for d odd we

want to establish directly the much stronger results given by Theorems

i* and 2*. The reason why the odd case yields stronger results is

roughly the following. In §2, for d even, we used the index theorem,

together with a deformation argument relating a connection A to

another connection A o. In t h e odd case the analogue of t h e index

theorem is itself concerned with 1-parameter families, as w e shall now

recall.

Suppose that Dt is a p e r i o d i c one-parameter family of self-

adjoint elliptic operators, with the parameter t ~ S I. The eigen-

values 1 are now functions of t and when t goes once round the
3
circle the 1 have, as a set, to r e t u r n to their original position.

However I. need not return to I : we may get a shift, e.g. i.

3 3 J
might return to lj+ n for some integer n. This integer n is c a l l e d

the spectral flow of the family and it is a t o p o l o g i c a l invariant of

 257

the family. It r e p r e s e n t s the number of negative eigenvalues which

have become positive (less the number of positive eigenvalues which

have become negative).

The spectral flow, like the index, is g i v e n by an explicit

topological formula [23. Moreover, for the first order differential

operators (e.g. Dirac operators) this formula is actually related to

an index formula as follows. If Dt is the family, defined on a

manifold M, consider the single operator

= - -
St + Dt

defined on M × S 1. Note that

* : _ ~__
~t + D t

so that ~ is n o t self-adjoint. Then one has [2]

(3.1) spectral flow of {D t %

J = index of ~.

As an i l l u s t r a t i v e example consider the case when M is also a circle

with angular variable x and take

D t = -i ~ + t.

The eigenvalues are n + t with n integral and so, as t increases

from 0 to i, we get a spectral flow of precisely one. The

periodicity of Dt is expressed by the conjugation property:

-ix ix
Dt+ 1 = e Dt e

The operator ~ acts naturally on the functions f(x,t) such that

(3.2) f(x + l,t) = f(x,t)

f(x,t + l) = e -ix f(x,t).

In fact these equations describe sections of a certain line-bundle

 258

on the torus S 1 × S 1.

Functions satisfying (3.2) have a Fourier series expansion

(3.3) f(x,t) = Zf (t)e ± n x

n

where fn(t + i) = fn+l(t).

Solving the equation ~f = 0 leads to the relations

fn(t) + (n + t ) f n ( t ) = 0

and so

2
fn(t) = C n ex~ (n+t) }
2

In view of the conditions (3.3) C is independent of n. Thus

n
has a one-dimensional null-space spanned by the theta function

f(x,t) = exp ~ exp(inz - n /2)

where z = x - it. A similar calculation shows that ~f = 0 has

no L2-solution, so t h a t index ~: 1 which checks with the spectral

flow.

After this digression about spectral flow we return to c o n s i d e r

the Dirac operators DA on a manifold M of odd dimension d. Let

S d ÷ U(N) be a generator of ~d(U(N)), where we take N in the

stable range, i.e. d+l

N > -~--, and now compose with a map M + Sd of

degree one to give a map F : M ÷ U(N). Consider F as a multi-

plication operator on the bundle S @ NV = S ® V ~ C N, on w h i c h the

Dirac operator DNA is d e f i n e d . Since the matrix parts of F and

DNA act on different factors in the tensor product they commute, and

so

[DNA,F ]= X

is independent of A. This multiplication operator X acts

 259

essentially on S ~ CN (trivially extended to S ~ V @ C N) , and is

locally given by

X = Ze F -1 $F.
1 1

In particular Ilxll : c is a u n i f o r m constant independent of V and A.

Consider now the linear family of connections

At = (l - t ) h + t F(A)

joining A to its gauge transform F(A). The corresponding family

of Dirac operators is

(3.4) D t = D A + tX.

By construction Do = DA and D 1 : F-IDA F is u n i t a r i l y equivalent

to Do . Thus we have a periodic family of self-adjoint operators with

a spectral flow. Moreover the general formula for the spectral flow

(e.g. via the index formula on M x S I) shows that in our case,

because of the construction of F, we have spectral flow equal to one.

It follows that, for some value of t, the operator Dt has a zero-

eigenvalue. Hence as b e f o r e (3.4) shows that the smallest eigenvalue

of DA does not exceed C.

The use of spectral flow to p r o v e Theorem 1 for odd d is so far

quite similar to the use of t h e index to prove Theorem 1 for even d.

However, spectral flow has the advantage that 0 is not a disting-

uished point of the spectrum, i.e. the spectral flow of a family is

unchanged by adding a constant. Replacing 0 by some other value

and repeating our argument then shows that there is an eigenvalue

of DA within C of B, and this is t h e content of Theorem I*.

Theorem 2* follows by extending the argument using maps

F : M ÷ U(N) of h i g h e r degree, on the same lines as T h e o r e m 2 was

proved in the even case.

 260

Finally it is worth pointing out that the upper bounds on the

eigenv a l u e s of Dirac operators given by those methods are fairly

sharp. In fact Vafa and W i t t e n actually determine the best b o u n d

when M is a flat torus. For this they use the index t h e o r e m for

multi-parameter families of e l l i p t i c operators - not just the s p e c t r a l

flow of a o n e - p a r a m e t e r family.

References

I. M.F. Atiyah and I.M. Singer, The index of elliptic operators III,

Ann. of Math. 87 (1968), 546-604.

2. M.F. Atiyah, V.K. Patodi and I.M. Singer, Spectral asymmetry and

Riemannian geometry III, Math. Proc. Camb. Phil. Soc. 79 1976),

71-99.

3. C. Vafa and E. Witten, Eigenvalue inequalities for fermions mn

gauge theories. Princeton University preprint, April 1984.

[Commun. in Math. Physics, 95, No.3 (1984) , 257-276.]

 MA~IFOLDS OF NON POSITIVE CURVATURE

W. Ballmann
Mathematisches Institut
WegelerstraBe 10
5300 B o n n I

This is m a i n l y a report on r e c e n t and rather recent work of the a u t h o r

and others on Riemannian manifolds of n o n p o s i t i v e sectional curvature.
The names o f the o t h e r people involved are M. Brin, K. B u r n s , P. E b e r -
l e i n a n d R. Spatzier.

Denote by Mn a complete connected smooth Riemannian manifold, by

KM the sectional curvature of M a n d by d the d i s t a n c e on M induced
by the R i e m a n n i a n metric. We always assume KM ~ 0 , that is, KM(O) ~ 0
for every tangent plane 0 of M.

O n e of the significant consequences of the a s s u m p t i o n KM ~ 0 is as

follows. Let ¥I and T2 be unit speed geodesics in t h e u n i v e r s a l co-
vering space M of M such that TI(0) = Y2(0) . Then for t,s ~ 0

d2(T1(t), T2(s)) ~ t 2 + s 2 - 2ts • cos(~1(0), ~2(0))

with equality if a n d o n l y if TII [0,t] and T21 [0,s] belong to t h e

boundary of a t o t a l l y geodesic and flat triangle. It f o l l o w s that the e x -
ponential map exp: T M > M is a d i f f e o m o r p h i s m for e a c h p 6M. In
P
particular, M is a K(z,1) ; t h e h o m o t o p y t y p e of M is d e t e r m i n e d by
F = ~I(M) . As we will see b e l o w , there are also strong relations bet-
ween the structure of F and the g e o m e t r y of M.

One of the principal aims in t h e study of n o n p o s i t i v e l y curved mani-

folds is to s p e c i f y the circumstances under which assertions about nega-
tively curved manifolds become false - if t h e y b e c o m e false - under the
weaker assumption of n o n p o s i t i v e sectional curvature. For example, a
theorem of M i l n o r [Mi] asserts that F has exponential growth if M is
compact and negatively curved. As for the w e a k e r assumption K M ~ 0,
Avez [Av] showed that F has exponential growth if a n d o n l y if M is
 262

not flat.

In g e n e r a l , one expects some kind of flatness in M if some proper-

ty of negatively curved manifolds is n o t shared by M . Hence it is o n l y
natural to try to m e a s u r e the flatness of M . In t h e case of locally
symmetric spaces, the rank is such a measure. The question arises, whether
such a notion can be introduced in a meaningful way for general manifolds
of nonpositive sectional curvature. This is indeed the content of Problem
65 in Y a u ' s list [Y]. We state this problem in a slightly modified form
and in two parts.

a) DEFINE THE RANK OF M AND SHOW THAT F CONTAINS A FREE

ABELIAN SUBGROUP OF RANK k IF M IS C O M P A C T OF RANK k .

Note that in t h e case M is compact and locally symmetric, the (usual)

rank of M is given by the maximal number k such that r contains a
free abelian subgroup of rank k . See also Theorem 1 below.

b) SHOW THAT F CONTAINS A FREE ABELIAN SUBGROUP OF RANK 2

IF M HAS A 2-FLAT.

Here a k-flat is d e f i n e d to b e a totally geodesic and isometrically im-

mersed Euclidean space of dimension k .

As in t h e case of locally symmetric spaces, the rank of M should

be an integer between I and n = dim (M) . Further properties of this
notion, which one expects, are as follows.

PI) IF M IS L O C A L L Y SYMMETRIC, THEN THE RANK OF M SHOULD CO-

INCIDE WITH ITS USUAL RANK.

P2) FLAT MANIFOLDS OF DIMENSION n SHOULD HAVE RANK n. NEGATI-

VELY CURVED MANIFOLDS SHOULD HAVE RANK ONE.

Vice versa, manifolds of rank one should resemble negatively curved mani-
folds.

P3) THE RANK OF M SHOULD BE EQUAL TO THE RANK OF M . THE RANK

OF A RIEMANNIAN PRODUCT MI ×M2 SHOULD BE THE SUM OF THE
RANKS OF MI AND M2 .

Note that M1 ×M2 still has nonpositive sectional curvature. If M I and

M2 are compact, then M1 ×M2 does not carry a metric of negative sec-
tional curvature, see Theorem I below.

Of course, there may be different satisfactory solutions to problem

a). One candidate for the rank of M , and maybe the most obvious one,
 263

is t h e following:

Rank (M) = max { k I M contains a k-flat } .

At this point it is o n l y conjectural that this notion of r a n k solves

problem a) ° A l s o note that with this definition of rank, problem b) is
p a r t of p r o b l e m a) . W i t h respect to R a n k (M) , the following results are
known.

Theorem I (Gromoll-Wolf [GW], Lawson-Yau [LY]) . If Mn is c o m p a c t , then

every abelian subgroup of F is f r e e a b e l i a n of r a n k a t m o s t n . If F
contains a free abelian subgroup of rank k , then M contains a totally
geodesic and isometrically immersed flat k-torus.

This result is the e x t e n s i o n of the theorem of P r e i s s m a n n [Pr] which sta-

tes that every abelian subgroup of F is i n f i n i t e cyclic if M is c o m -
pact and negatively curved. Theorem I implies that

Rank (M) ~ m a x { k I F c o n t a i n s a free abelian subgroup of r a n k k }

if M is c o m p a c t . Problem a) n o w consists in s h o w i n g that equality holds.

We say t h a t M satisfies the v i s i b i l i t y axiom if a n y two distinct

points in t h e ideal boundary of M can be joined by a geodesic [E0]. For
example, compact negatively curved manifolds satisfy the visibility axiom.

Theorem 2 (Eberlein [El]). If M is c o m p a c t , then M satisfies the

visibility axiom if a n d o n l y if M does not contain a 2-flat, that is,
Rank (M) = I.

Thus problem b) can be reformulated as s a y i n g that M satisfies the v i -

sibility axiom if a n d o n l y if e v e r y abelian subgroup of F is i n f i n i t e
cyclic.
 264

We now discuss a different notion of rank which was introduced in

[BBE]. We need some definitions. Denote by SM the u n i t tangent bundle
of M . For v6SM , let Yv be the g e o d e s i c which has v as initial
velocity vector. Along ¥v consider the space JP(v) of all p a r a l l e l
Jacobi fields. Note t h a t b y the a s s u m p t i o n KM & 0 , a parallel field X
along Yv ' which is l i n e a r i l y independent of ~v ' is s u c h a p a r a l l e l
Jacobi field if a n d o n l y if KM(~v(t) AX(t)) = 0 for all t . Now set

rank (v) = dim (JP(v)) and

rank (M) = m i n { r a n k (v) I v 6 S M } .

Note that rank (M) = I if M has a point p such that the sectional
curvatures of a l l tangent planes at p are negative. In p a r t i c u l a r ,
rank (M) = 1 if M is a c o m p a c t surface of n e g a t i v e Euler characteristic.

The above definition of rank was motivated by the results in the

papers [BI], [B2], and [BB] which deal primarily with geodesic flows on
manifolds of r a n k one. (Formally, the g e n e r a l assumption in [BI] and
[B2] is t h a t M has a geodesic which does not bound a flat half plane,
but in v i e w of T h e o r e m 4 below this is e q u i v a l e n t to rank (M) = 1.) The
geodesic flow gt operates on SM, and by definition gt(v) = ~v(t)
The geodesic flow leaves invariant the Liouville measure of SM.

We now state s o m e of the p r o p e r t i e s of manifolds of r a n k one.

Theorem 3. Suppose rank (M) I.

i) [BB] If M is c o m p a c t , then gt is e r g o d i c .
ii) [BI] If M has finite volume, then gt has a dense orbit.
iii) (Eberlein [B2]) If M has finite volume, then tangent vectors
to c l o s e d geodesics are dense in SM.

Part i) of this theorem generalizes, at the same time, the c e l e b r a t e d

theorem of A n o s o v that the g e o d e s i c f l o w on a c o m p a c t negatively curved
manifold is e r g o d i c [An] and the r e s u l t of P e s i n that the g e o d e s i c flow
on a c o m p a c t surface of n e g a t i v e Euler characteristic is e r g o d i c [Pe].
The proof of p a r t i) m a k e s essential use of the r e s u l t s of Pesin [Pe] and
of the results in [BI].
 265

As for manifolds of h i g h e r rank, the following result is o n e of t h e

basic ingredients in a l l the further developments.

Theorem 4 [BBE]. If the v o l u m e of M is f i n i t e or if M is a n a l y t i c ,

then

rank (M) = max { k I each geodesic of M is c o n t a i n e d in a

k - f l a t }.

In p a r t i c u l a r , rank (M) £ R a n k (M) . There are examples where this in-

equality is strict, see the introduction of [BBE]. In an e a r l i e r version
of T h e o r e m 4, B u r n s proved that each geodesic in M bounds a flat half
plane if rank (M) ~ 2 , see [Bu]

The counterpart to T h e o r e m 3 in t h e h i g h e r rank case is as f o l l o w s .

Theorem 5. Suppose that rank (M) = k a 2 and that KM has a l o w e r

2
bound -a .
t
i) [BBE] If M has finite volume, then g is n o t ergodic.
ii) [BBS] If M has finite volume, then gt has k-1 inde-
pendent differentiable first integrals on an open, dense,
and gt-invariant subset of SM.
iil) [BBS] If M is c o m p a c t , then tangent vectors to t o t a l l y ge-
odesic and isometrically immersed flat k-tori are dense
in SM.

It f o l l o w s from iii) that F contains free abelian subgroups of r a n k

k if rank (M) = k . In p a r t i c u l a r , problem a) is s o l v e d w i t h this no-
t i o n of rank.

There are some immediate questions related to the a s s u m p t i o n s in

Theorems 3 and 5. N a m e l y , is it p o s s i b l e to d e l e t e the a s s u m p t i o n that
K has a lower bound in T h e o r e m 5 and the assumption that M is c o m p a c t
in p a r t i) of T h e o r e m 3? I b e l i e v e that the answer is y e s in b o t h cases.
That the compactness assumption can be d e l e t e d in p a r t iii) of T h e o r e m 5
is a c o n s e q u e n c e of the f o l l o w i n g result.
 266

T h e o r e m 6 [B3, BS]. Suppose that r a n k (M) ~ 2 , K has a lower bound

2 M
-a and M h a s f i n i t e v o l u m e . If M is i r r e d u c i b l e , t h e n M is a lo-
cally symmetric space of n o n c o m p a c t type.

Actually, Burns-Spatzier [BS] need the stronger assumption that M is

compact. Under the further assumptions M compact and dim (M) ~ 4, T h e -
orem 6 was proved earlier b y the author in j o i n t work with Heintze [BH].
All these proofs are along completely different lines, up to the fact
that they are b a s e d o n the r e s u l t s in [BBE] and [BBS].

The use of T h e o r e m 6 lies in the fact that, for many purposes, it

will be sufficient to p r o v e a given assertion in the r a n k one case and
the symmetric space case separately in o r d e r to g e t a conclusion in the
general case. Using this device and results of P r a s a d - R a g h u n a t h a n [PR],
the a u t h o r in c o l l a b o r a t i o n with Eberlein defined algebraically a number
rank (r) , the r a n k of the fundamental group r of M , and showed that
rank (r) = rank (M) . Using various other previous results of E b e r l e i n
and a recent result of Schroeder one obtaines the following conclusion.

Theorem 7 [BE]. Suppose that K M ~ -a 2 and M has finite volume. Then

M is a n irreducible locally symmetric space of noncompact type of rank
k ~ 2 if a n d o n l y if the following three conditions are satisfied:

i) F does not contain a normal abelian subgroup (except {e} )

ii) no f i n i t e index subgroup of r is a p r o d u c t
iii) rank (r) = k .

Here a Riemannian manifold N is c a l l e d irreducible if no f i n i t e covering

of N is a R J e m a n n i a n product. Theorem 7 c a n be u s e d to e x t e n d the rigi-
dity results of M o s t o w [Mo] a n d M a r g u l i s [Ma]. Namely, using their results
and Theorem 7 we o b t a i n :

2
Theorem 8. Suppose that K M ~ -a and M has finite volume. Suppose
M* is an irreducible locally symmetric space of n o n c o m p a c t type and high-
er r a n k with finite volume. If t h e fundamental groups of M and M* are
isomorphic, then M and M* are isometric up to n o r m a l i z i n g constants.

Under the stronger assumption that M is c o m p a c t , Theorem 8 was proved

earlier by G r o m o v [GS] and, in a s p e c i a l case, by Eberlein [E2].
 267

References

[An] D.V. Anosov, G e o d e s i c F l o w s on C l o s e d R i e m a n n i a n M a n i f o l d s with

N e g a t i v e Curvature, Prec. Steklov Inst. Math. 90, Amer. Math. Soc.,
Providence, Rhode Island, 1969.

[Av] A. Avez, "Vari~t~s r i e m a n n i e n n e s sans points focaux", C.R. Acad.

Sc. Paris 270 (1970), 188 - 191.

[BI] W. Ballmann, "Einige neue R e s u l t a t e ~ber M a n n i g f a l t i g k e i t e n nicht

p o s i t i v e r Kr~mmung", B o n n e r math. S c h r i f t e n 113 (1978), 1 - 57.

[B2] W. Ballmann, "Axial i s o m e t r i e s of m a n i f o l d s of n o n - p o s i t i v e cur-

vature", Math. Ann. 259 (1982), 131 - 144.

[B3] W. Ballmann, in preparation.

[BB] W. B a l l m a n n and M. Brin, "On the e r g o d i c i t y of g e o d e s i c flows",

Erg. Th. Dyn. Syst. 2 (1982),311 - 315.

[BBE] W. Ballmann, M. Brin and P. Eberlein, "Structure of m a n i f o l d s of

n o n p o s i t i v e curvature. I", Preprint, Bonn - C o l l e g e Park - C h a p e l
Hill 1984.

[BBS] W. Ballmann, M. Brin and R. Spatzier, "Structure of m a n i f o l d s of

n o n p o s i t i v e curvature. II", Preprint, Bonn - C o l l e g e Park - Ber-
k e l e y 1984.

[BE] W. B a l l m a n n and P. Eberlein, in preparation.

[BH] W. B a l l m a n n and E. Heintze, unpublished.

[Bu] K. Burns, "Hyberbolic b e h a v i o r of geodesic flows on m a n i f o l d s with

no focal points", Erg. Th. Dyn. Syst. 3 (1983), I - 12.

[BS] K. Burns and R. Spatzier, in preparation.

[El] P. Eberlein, "Geodesic flow in certain m a n i f o l d s without conjugate

points", Trans. AMS 167 (1972), 151 - 170.

[E2] P. Eberlein, " R i g i d i t y of lattices of n o n p o s i t i v e curvature", Erg.

Th. Dyn. Syst. 3 (1983), 47 - 85.

[E0] P. E b e r l e i n a n d B. O'Neill, "Visibility manifolds", Pac. J. Math. 46

(1973), 45 - 109.

[GW] D. Gromoll and J. Wolf, "Some r e l a t i o n s between the m e t r i c struc-

ture and the a l g e b r a i c structure of the f u n d a m e n t a l groups in
m a n i f o l d s of n o n p o s i t i v e curvature", Bull. AMS 77 (1971), 545-552.

[GS] M. G r o m o v and V. Schroeder, Lectures o__nnM a n i f o l d s of Nonpositive

Curvature, in preparation.

[LY] H.B. Lawson and S.-T. Yau, "Compact m a n i f o l d s of n o n p o s i t i v e cur-

vature", J. D i f f e r e n t i a l G e o m e t r y 7 (1972), 211 - 228.
 268

[Ma] G.A. M a r g u l i s , " D i s c r e t e g r o u p s of m o t i o n s of m a n i f o l d s of n o n -

positive curvature", AMS T r a n s l a t i o n s 109 (1977), 33 - 45.

[Mi] J. M i l n o r , "A n o t e o n c u r v a t u r e a n d fundamental group", J. D i f f e r -

e n t i a l G e o m e t r y 2 (1968), I - 7.

[Mo] G.D. M o s t o w , S t r o n g R i g i d i t y of L o c a l l y S y m m e t r i c S p a c e s , A n n a l s
of M a t h . S t u d i e s No. 78, P r i n c e t o n U n i v e r s i t y Press, P r i n c e t o n ,
N e w J e r s e y , 1973.

[Pe] Ja.B. P e s i n , " G e o d e s i c f l o w s o n c l o s e d R i e m a n n i a n m a n i f o l d s with-

o u t f o c a l p o i n t s " , M a t h . U S S R Izv. 11 (1977), 1195 - 1228.

[mR] G. P r a s a d a n d M.S. R a g h u n a t h a n , " C a r t a n s u b g r o u p s a n d l a t t i c e s in

s e m i - s i m p l e g r o u p s " , Ann. of M a t h . 96 (1972), 296 - 317.

[mr] A. P r e i s s m a n n , " Q u e l q u e s p r o p r i ~ t ~ s g l o b a l e s d e s e s p a c e s de R i e -
m a n n " , C o m m e n t . M a t h . H e l v e t i c i 15 (1943), 175 - 216.

[Y] S . - T . Yau, S e m i n a r o n D i f f e r e n t i a l G e o m e t r y , A n n a l s o f M a t h .
S t u d i e s No. 102, P r i n c e t o n U n i v e r s i t y P r e s s a n d U n i v e r s i t y o f
T o k y o Press, P r i n c e t o n , N e w J e r s e y , 1982.
 Metrics with Holonomy G2 or Spin (7)

by

Robert L. Bryant

%1. The Holonomy of R i e m a n n i a n Manifolds

In this section, all objects are assumed smooth unless stated

otherwise, M will denote a connected, simply connected n-manifold and

g will denote a Riemannian metric on M. If ~: [0,i] ~ M is a p a t h

in M, then the Levi-Civita connection of g induces a well-defined

parallel translation along ~, p : T (o)M ~ T (1)M which is a n iso-

merry of v e c t o r spaces. For every x ~ M, we let Hx denote the set

of a l l P where ~ ranges over all paths with ~(0) = ~(i) = x. It

is w e l l - k n o w n , see [i], that the simple connectivity of M implies

that Hx is a c o n n e c t e d , closed Lie subgroup of SO(TxM), the group

of oriented isometries of TxM with itself. Moreover P (H (0)) =

H (i) for any path ~. It follows that by choosing an isometry

i: T M = ~n, we can identify H with a subgroup H c SO(n). The

x x
conjugacy class of H in O(n) is independent of the choice of x

or i. By abuse of language we speak of H as the holonomy of g.

The holonomy group is a m e a s u r e of the curvature of g. For

n2
example, if H preserves an orthogonal decomposition ~n = ~nl • ~ ,
n.
then g = gl + g2 locally where gi is a local metric on ~ ~. It

follows that, in o r d e r to d e t e r m i n e which subgroups of SO(n) can be

holonomy groups of Riemannian metrics, it s u f f i c e s to d e t e r m i n e the

subgroups H c SO(n) which act irreducibly on ~n and are holonomy

groups of Riemannian metrics. By examining the Bianchi identities and

making extensive use of representation theory, Berger [2] proved the

following classification theorem.

 270

Theorem (Berqer): Let (Mn,g) be a connected, simply connected

Riemannian n-manifold and suppose that its holonomy group H c SO(n)

acts irreducibly on ~n. Then either (M,g) is locally symmetric or

else H is one of the following subgroups of SO(n)

(i) SO(n)

(ii) U(m) if n = 2m > 2

(iii) SU(m) if n = 2m > 2

(iv) Sp(1)Sp(m) if n = 4m > 4

(v) Sp(m) if n = 4m > 4

(vi) G2 if n = ?

(vii) Spin(?) if n = 8

(viii) Spin (9) if n = 16

After noting that the above list is exactly the list of subgroups

of SO(n) which act transitively on S n-I c ~n, Simons [3] gave a

direct proof that the holonomy of an irreducible non-symmetric metric

on Mn acts transitively on S n-l.

It is natural to a s k which of the possibilities on Berger's list

actually do occur. It is easy to show that the "generic" metric on

Mn has holonomy SO(n). If n = 2m, a matric with holonomy a sub-

group of U(m) is, of course, a Kahler metric. Such a metric is

given in local coordinates on £m in the form

gf = ( ~2f/~zi a~J)dz i o di~

where f is a s m o o t h function on cm satisfying the condition that

its complex hessian Hf = (~2f/~zi ~zJ) be positive definite. For a

"generic" f with Hf > O, the metric gf will have holonomy U(m).

Every metric on M 2m with holonomy H c SU(m) can be put in the

above form locally where f satisfies the complex Monge-Ampere equa-

tion det(Hf) = I. Again, the "generic" solution of this equation

 271

yields a metric whose holonomy is e x a c t l y SU(m). Since Sp(m) c_

SU(2m), we ca e v e n construct metrics whose holonomy is Sp(m) on

M 4m (m > I) locally by selecting a linear map j: c 2 m ~ £ 2 m satis-

fying j2 = -I and j = tj and considering the gf where f

satisfies the s y s t e m of e q u a t i o n s Hf J Hf = J. Even though this is

an overdetermined system of e q u a t i o n s for f, e n o u g h solutions c a n be

found to e x h i b i t local metrics with holonomy exactly Sp(m). A simi-

lar c o n s t r u c t i o n with complex contact structures o n £2m+i allows one

to e x h i b i t metrics locally on ~4m with holonomy Sp(1) Sp(m). It

must be e m p h a s i z e d that it is the e n c o d i n g of h o l o n o m y properties into

the C a u c h y - R i e m a n n equations (which are completely understood locally)

that allows the c o n s t r u c t i o n of m e t r i c s in c a s e s (ii)-(v) to be re-

duced to a m a n a g a b l e partial differential equations problem.

There remain the "exceptional" cases (vi)-(viii). In a s u r p r i s -

ing p a p e r , Alekseevski [4] s h o w e d that a n y m e t r i c on M 16 with holo-

nomy Spin(9) was necessarily locally symmetric. Thus, case (viii)

c a n be r e m o v e d from Berger's list. It is w o r t h remarking that cases

(vi) and (vii) do not occur as s y m m e t r i c spaces [6]. This raises the

possibility that these two c a s e s do not occur at all. As of this

writing, no examples of c a s e s (vi) or (vii) are known. Nevertheless,

there is e x t e n s i v e literature o n the p r o p e r t i e s of these elusive

metrics. See [7], [8], and [9] a n d the bibliographies contained

therein.

In this lecture, we shall outline a proof of the e x i s t e n c e of

local metrics in c a s e s (vi) and (vii). The details, which involve an

analysis of a d i f f e r e n t i a l system to be c o n s t r u c t e d below will be p u b -

lished elsewhere. For the a p p r o p r i a t e concepts from differential sys-

II

tems and Cartan-Kahler theory, the r e a d e r may consult [i0].

%2. Linear Algebra, H-structure, a n d D i f f e r e n t i a l System ~

Our strategy will be to d e s c r i b e a set of d i f f e r e n t i a l equations

whose solutions will represent metrics on Mn with the d e s i r e d holo-

 272

nomy. We begin by giving a somewhat non-standard description of G2 .

Let 1 2 , . .., 7 be an oriented orthonormal coframing of ~7

We define the 3-form

= 123 + 145 + 167 + 246 _ 257 _ 356 _ 347

where iJk is a n a b b r e v i a t i o n for wiA~JA~k.

Proposition I: G2 = (A • GL(7) IA*(P) = ~} where G2 is the 14-

dimensional simple Lie group of c o m p a c t type.

We will not p r o v e Proposition 1 here. It is i n t e r e s t i n g to n o t e

that a dimension count shows that the o r b i t of ~ in A 3 ( ~ 7) under

GL(7) is open. (In fact, there are exactly two o p e n GL(7) orbits

in A3(~?). The stabilizer of a form ~ in the o t h e r open orbit is

the s i m p l e Lie group of n o n - c o m p a c t type of d i m e n s i o n 14.) The form

was discovered by Chevalley [5]. Bonan [7] s h o w e d that

G
2
A = span{l,P,*F,*l = (i/7)~ /x *~}

G2
where A c A ( ~ 7) is the s u b r i n g of G2-invariant exterior forms.

If V is a s e v e n dimensional vector space, we will say that

a • A 3 ( V *) is p o s i t i v e if t h e r e exists a linear isomorphism

3 *
L: V ~ ~7 so that a = L*(~). The set A + ( V ) c A 3 ( V *) of p o s i t i v e

forms is c l e a r l y an open subset of A3(V*). If a • n 3 ( M 7) we say

that a is p o s i t i v e iff alx is p o s i t i v e for all x • M 7. We let

E c A3(T~M) denote the o p e n s u b m a n i f o l d of p o s i t i v e 3-forms.

~: E ~ M is a s m o o t h fiber bundle with fibers isomorphic to

G L ( 7 ) / G 2. The sections of E are the p o s i t i v e forms on M and are

also obviously in I-i correspondence with the set of G2 reductions

of the tangent bundle of M, i.e., G2-structures on Mo Since

 273

G 2 ~ SO(7), it follows that each G2-structure on M induces a

canonical underlying orientation and Riemannlan metric.

On the other hand, if (M7,g) is an oriented Riemannian manifold

with holonomy G2, it is easy to see that there is a u n i q u e parallel

positive 3-form ~ on M whose underlying orientation and metric

g
are the given ones.

Proposition 2: Let ~ be a positive 3-form on M, and let ~a be

the dual 4-form with respect to the underlying metric and orientation.

Then a is parallel with respect to the underlying metric's Levi-

Civita connection iff d~ = d*~ = O.

Proposition 2 is due to Gray [8] in the context of vector cross pro-

ducts. It follows from this that every positive 3-form ~ which

satisfies the system of partial differential equations da = d*~ = 0

has an underlying metric whose holonomy is a s u b g r o u p of G2 and

conversely every metric whose holonomy is a s u b g r o u p of G2 arises

from such an ~.

The conditions d~ = d~ = 0 form a quasi-linear first order

system for the 35 (= d i m A 3+( V ~ ) ) unknown coefficients of ~. The

system is q u a s i - l i n e a r because coefficients of *a are algebraic

functions of the coefficients of a. A priori, this appears to be 56

(= d i m ( A 4 ( ~ 7) • A5(~7)) equations for the 35 u n k n o w n s . However,

there is a (miraculous) identity

(~d/3) A /3 + (*d*/3) /k */3 = 0

valid for any positive /3 where the ~ is the Hodge star of the

underlying S0(7) structure. It can be shown that the remaining 49 =

56 - 7 equations are independent.

This overdetermined system is invariant under the diffeomorphism

group of M and hence cannot be elliptic. However, it can be shown

 274

that it is transversely elliptic, i.e., elliptic when restricted to a

local slice of the a c t i o n of Dill(M) on n~(M).

Our first main result is

Theorem I: The system da = d * a = 0 for a • R3+(M) is i n v o l u t i v e

with Caftan characters (Sl,S 2 ..... s?) = (0,0,1,4,10,13,7). In

particular, the "generic" solution has the p r o p e r t y that its u n d e r -

lying metric has holonomy exactly G 2.

We remark that Theorem I is e s s e n t i a l l y a calculation. One

describes the a p p r o p r i a t e differential system with independence con-

dition on E c n3(T*M) and calculates both the integral elements and

the C a r t a n characters to a r r i v e at the r e s u l t . Note that this system

is real analytic in l o c a l coordinates. The transversality property

actually implies that any solution is real analytic in s o m e coordinate

system anyway, so the a p p l i c a t i o n of Cartan-Kahler theory is v i n d i -

cated. Details will appear elsewhere.

We now turn to the a n a l o g o u s case H = Spin(7). Write ~8 = ~I $

~7 and augment the g i v e n coframing of ~7 by an O. We then define

the 4 - f o r m on ~8

0
~ = ~ A Io + ~ = *~

where ~ = *~ e A4(~7).

Proposition 3: Spin(?) = {A • G L ( 8 ) I A (~) = qb} where Spin(Y) c_

SO(8) is i s o m o r p h i c to the u n i v e r s a l cover of SO(7).

Proposition 3 is not d i f f i c u l t to p r o v e assuming Proposition i.

The form ~ was discovered by Bonan [7] w h o showed that

A spin(7) = {i,~ = *~,'i = ( I / 1 4 ) @ 2}

 275

where A Spin(g) c A ( ~ 8) is the s u b r i n g of Spin(g)-invariant exterior

forms on R 8. The GL(8)-orbit of ~ e A 4 ( ~ 8) is n o t open but is, of

course, a smooth submanifold of A 4 ( ~ 8). We shall say that an a e

~8
A 4 ( V ~) is a d m i s s i b l e if t h e r e exists a linear isomorphism L: V

so that a = L*(4~). If a e 24(M8), we shall say that a is a d m i s -

sible if alx is a d m i s s i b l e for all x e M 8. We let F c_ M 4 ( T ~ M )

denote the s u b m a n i f o l d of a d m i s s i b l e 4-forms. ~: F ~ M 8 is a s m o o t h

fiber bundle with fibers isomorphic to GL(8)/Spin(Y). Clearly the

space of s e c t i o n s of F, i.e. the s p a c e of a d m i s s i b l e 4-forms on M,

is in i-I c o r r e s p o n d e n c e with the s p a c e of Spln(g)-structures on M.

Since Spin(?) c SO(8), we see that each admissible a on M canoni-

cally induces an orientation and metric on M.

O n the o t h e r hand, if (M8,g) is a n o r i e n t e d Riemannian manifold

with holonomy Spin(7), it is e a s y to s e e that there is a u n i q u e

parallel admissible 4-form a on M whose underlying orientation

g
and metric are the g i v e n ones.

Proposition 4: Let a be an admissible 4-form on M. Then a is

parallel with respect to the L e v i - C i v i t a connection of the u n d e r l y -

ing metric iff da = 0.

Proposition 4 is a c t u a l l y more elementary than the corresponding

Proposition 2, b u t seems to h a v e been overlooked. It f o l l o w s from

this that every admissible 4-form a which satisfies da = 0 has an

underlying metric whose holonomy is a s u b g r o u p of Spin(g) and con-

versely every metric whose holonomy is a s u b g r o u p of Spin(?) arises

from such an ~.

Since F is n o t an open subset of a v e c t o r bundle over M, the

condition da = 0 is o n l y a q u a s i - l i n e a r first order system of 56

(= d i m A 5 ( ~ 7 ) ) equations for the 43 (= d i m ( G L ( 8 ) / S p i n ( 7 ) ) unknown

coefficents of the s e c t i o n a: M ~ F. It c a n b e s h o w n that these 56

 276

equations are algebraically independent. Again, this over-determined

system is i n v a r i a n t under the d i f f e o m o r p h i s m group of M a n d c a n be

shown to b e transversely elliptic.

The analogue of T h e o r e m I for S p i n ( 7 ) is

Theorem 2: The system da = 0 for s e c t i o n s a: M ~ F is i n v o l u t i v e

with Cartan characters (sl,s 2 ..... Ss) = (0,0,0,I,4,10,20,8). In

particular, the "generic" solution has the p r o p e r t y that its u n d e r l y -

ing metric has holonomy exactly Spin(7).

Theorem 2 is a l s o a c a l c u l a t i o n with the a p p r o p r i a t e differential

system with independence condition on F c A4(T~M). Details will

appear elsewhere.

~3. Closing Remarks

The methods of %2 o n l y y i e l d the w e a k e s t positive result. Name-

ly, that there exist local metrics on ~7 and ~8 which are not

locally symmetric and have holonomy equal to G2 and Spin(7) re-

spectively. This at least shows that Berger's list cannot be s h o r t e n -

ed a n y further. Of c o u r s e , in m a n y respects this is q u i t e unsatis-

factory.

In the first place, w e d o not know a single example of s u c h a

metric in e i t h e r case. The search for s u c h m e t r i c s is led b y G r a y [8]

but has s o far p r o v e d fruitless.

In the s e c o n d place, we do not know if t h e r e exists a complete

metric even on ~7 or ~8 with holonomy G2 or Spin(7). This

problem reminds us, in s o m e respects, of the c o n j e c t u r e that a com-

plete Kahler metric on £m which has holonomy a subgroup of SU(m)

is a c t u a l l y flat [II].

Finally, we do not know if t h e r e exists a compact example of

either kind. Nevertheless, the d e s c r i p t i o n s of s u c h m e t r i c s afforded

by Theorems I and 2 allow one to p r o v e a good number of theorems about

 277

possible examples. In a f o r t h c o m i n g joint work b y the a u t h o r and

Reese Harvey it is s h o w n that a compact (MT,g) with holonomy G2

must be orientable, spin, and have finite fundamental group. The

first Pontriagin class of M7 must be n o n - z e r o and the d e f o r m a t i o n

theory of the s o l u t i o n s of da = d ~ a = 0 is u n o b s t r u c t e d , the d i m e n -

s i o n of the local moduli space being b3 > 0 where b3 is the third

Betti number of M. Similar results are obtained for 8-manifolds with

holonomy Spin(7). The difficulty of e x p l i c i t l y writing down such a

metric can be appreciated by contemplating the fact that no e x p l i c i t

example of a C a l a b i - Y a u metric on a K-3 surface is k n o w n as of this

writing.

BIBLIOGRAPHY

I. K o b a y a s k i , S. a n d N o m i z u , K., F o u n d a t i o n s o_f_fD i f f e r e n t i a l
G e o m e t r y , W i l e y a n d Sons, N e w York, 1963 a n d 1969.

2. Berger, M., Sur les G r o u p e s d'Holoonomie Homogene des Varietes a

C o n n e x i o n A f f i n e et d e s V a r i e t e s R i e m a n n i e n n e s , Bull. Soc.
Math. F r a n c e , 83 (1955), 2 7 9 - 3 0 0 .

3. S i m o n s , J., O_n_nT r a n s i t i v i t y Holonomy systems, Ann. of M a t h . , 76

(1962), 213-234.

4. A i e k s e e v s k i , D. V., R i e m a n n i a n S p a c e s with Unusual Holonomy.

G r o u p s , F u n k c i o n a l Anal. i P r i l o v e n 2 (1968), i-i0. Trans-
l a t e d in F u n c t i o n a l Anal. Appl.

5. Chevalley, C., Algebraic Theory of__fS p i n o r s , 1954.

6. Helgason, S., D i f f e r e n t i a l Geometry_, L i e G_rrouPs, and,.Symmetric

spaces, A c a d e m i c Press, 1978.

7. Bonan, E., Sur les V a r i e t e s R i e m a n n i e n n e s a Groupe d'Holonomie

G2 ou S p i n ( 7 ) , C. R. A c a d . Sci. P a r i s 262 (1966), 1 2 7 - 1 2 9 .

8. Gray, A., Weak Holonomy Groups, Math. Z. 123 (1971), 290-300.

9. F e r n a n d e z , M~ a n d Gray, A., Riemannian Manifolds with Structure

GROUP G 2, A n n a l i di Mat. p u r a e d appl. (IV), 32 1982, 19-45.

i0. Chern, S. S. et al, Exterior Differential Systems, to a p p e a r .

ii. Yau, S. T., ed., P r o b l e m S e c t i o n in S e m i n a r o n D i f f e r e n t i a l

G e o m e t r y , A n n a l s of M a t h . S t u d i e s , no. 102, P r i n c e t o n ,
U n i v e r s i t y P r e s s , 1982.
 ON RIEMANNIAN METRICS ADAPTED TOO THREE-DIMENSIONAL
CONTACT MANIFOLDS

by

S,S. Chern
Mathematical Sciences Research Institute
1000 Centennial Drive
Berkeley, California 94720

R.S. Hamilton
Department of Mathematics
University of California, San Diego
La Jolla, California 92093

0. Introduction It was proved by R. Lutz and J. Martinet

[8} that every compact orientable three-dimensional manifold M has a
contact structure. The latter can be given by a one-form co, the
contact form. such that coAdco never vanishes; ~0 is defined up to a
non-zero factor. A Riemannian metric on M is said to be adapted to
the contact form ~ if: 1) co has the length I; and 2) dco=2.,.~,.
being the Hodge operator. The Webster curvature W, defined below in
|9], is a linear combination of the sectional curvature of the plane co
and the Ricci curvature in the direction perpendicular to co.
Adapted Riemannian metrics have interesting properties. The
main result of the paper is the theorem:

Every contact structure on a compact orientable

three-dimensional mani]old has a contact ]orm and an
adapted Rie~annian metric whose Webster curvature ia
either a constant ~ 0 or is everywhere stricttV
positive.

The problem is analogous to Yamabe's problem on the conformed

transformation of Riemannian manifolds Most recently, R. Schoen has
proved Yamabe's conjecture in all cases, including that of positive
scalar curvature 19|. It is thus an interesting question whether in the
second case of our theorem the Webster curvature can be made a
positive constant.

1),2) Research supported in part by N S F grants DMS84-03201 and

DMS84-01959.
 280

AfLer our theorem was proved, we learned that a similar

theorem on CR-manifolds of any odd dimension has been proved by
Jerison and Lee. [7] As a result, our curvature was identified with
the Webster curvature. We feel that our viewpoint is sufficiently
different from Jerison-Lee and that the three-dimensional case has so
m a n y special features to merit a separate treatment.
In an appendix, Alan Weinstein gives a topological implication
of the vanishing of the second fundamental form in {54}. For an
interesting a c c o u n t of three-dimensional c o n t a c t manifolds, cf. [2].

1. C o n t a c t Structure~. Let M be a manifold and B a

subbundle of the t~ngent bundle TM. There is a naturally defined
anti-symmetric bilinear form A on B with values in the quotient bundle
TM/B

(I} A: B × B --* T M / B

defined by the Lie bracket;

(2) A(V,W) .~ [V,W] rood B.

It is easy to verify that the value of A(V,W) at a point peM

depends only on the values of V and W at p. The bundle B defines a
foliation if and only if it satisfies tile Frobenius integrability condition
A = 0. Conversely, a c o n t a c t s t r u c t u r e on M is s subbundle B of the
tangent bundle of codimension I such that A is non-singular at each
point pcM. This can only o c c u r when the dimension of M is odd.
It is an interesting problem to find some geometric s t r u c t u r e
which can be put on e v e r y three-manifold, since this would be helpful
in studying its topology. Along these lines we have the following
remarkable theorem o f Imtz and Martinent {see [8], [I0]).

1.1 TheoreL EverF compact orientable

three-manifola possesses a contact structure.
 281

There are many different contact structures possible, since the

set of B with A ~ 0 is open. Even on S3 there are contact
structures f o r w h i c h t h e b u n d l e s B 1 and B 2 a r e t o p o l o g i c a l l y d i s t i n c t .
N e v e r t h e l e s s t h e n o t i o n of a c o n t a c t structure is r a t h e r f l a b b y , in t h e
following sense. We say B is conjugate to B, if there is a
d i f f e o m o r p h i s m ~ : M --m* M w h i c h h a s ~ (B) = B=. Then we have the
f o l l o w i n g r e s u l t due t o G r a y ( s e e [4]).

1.2 Theorem. Given a contact structure B, a n y

other contact structure B, close enough to B ~s
conju~Tate to it.

2. Metrics adapted to contact structures. A c o n t a c t form to

is a 1 - f o r m on M w h i c h is n o w h e r e z e r o and has t h e c o n t a c t bundle B
for its null s p a c e . In a three-manifold a non-zero 1-form to is a
contact form for the contact structure B = Null co ff and only i f
toAdw=0 at every point. The contact structure B determines the
c o n t a c t form up t o a s c a l a r muItiple. The c h o i c e o f a c o n t a c t f o r m to
a l s o d e t e r m i n e s a v e c t o r f i e l d V in t h e f o l l o w i n g w a y .

2.1 [,emma. There exists a unique vector field

V such that t~{V) = 1 a n d dtofV,W} = 0 f o r all WeTM.

Proof. Choose V0 with co{V0) = 1. Since dtoAco ~ 0, the

form dto is n o n - s i n g u l a r on B. Therefore there exists a unique VleB
with

d ~ f V l , W ) = dco(V0,W)

for all WeB. Let V = V 0 - V 1. Then w(V) = to(V 0) - co(V 1) = 1,

and dcoW,W) = 0 for all WeB. Since V is transverse to B and
dco{V,V) = 0, we h a v e dco(V,W) = 0 f o r all W e T M .
Locally any two non-zero vector fields are conjugate by a
diffeomorphism. However, this fails globally, since a vector field may
have closed orbits while a nearby vector field does not. It is a
c l a s s i c a l r e s u l t t h a t l o c a l l y a n y t w o c o n t a c t forms a r e c o n j u g a t e by a
 282

diffeomorphism. But globally two n e a r b y c o n t a c t forms may not be

conjugate, since t h e v e c t o r f i e l d s t h e y d e t e r m i n e may not be.
A choice of a Riemannian metric on a contact manifold
determines a choice of the contact form t~ up to sign by the
condition that co have length 1. Let = denote the Hodge star
operator. We make t h e following definition.

2.2 Definition. A Riemannian metric on a contact

three-manifold is said to be adapted to the contact form co if co is
of length one and satisfies the structural equation

{3) dt~ = 2 =w.

Such metrics have nice properties with respect to the contact

structure. For example, w e have the following results.

2.3 Lemma. If the metric is adapted to the

form to, then the vector field V determined by to i s
the unit vector field perpendicular to B.

Proof. Let V be the unit v e c t o r field p e r p e n d i c u l a r to B. Then

to(V) = 1, and f o r all v e c t o r s W in B we have dto(V,W)=2=to(V,W)=0.
Hence V is t h e v e c t o r field d e t e ~ i n e d by t h e c o n t a c t form to.

2.4 bemma. If the metric is adapted to the

contact form ~, t h e n the area form on B is given by
1
~i dto.

Proof. The a r e a form on B is =~.

A CR s t r u c t u r e on a mainfold is a c o n t a c t s t r u c t u r e t o g e t h e r
with a complex structure on the contact bundle B; that is, an
involution J:B--*B with j2 = -I where I is the identity. If M has
dimension 3 then B has dimension 2, and a complex structure on B is
equivalent to a conformal structure; that is knowing h o w to rotate by
90 °. Hence, a Riemannian metric on a contact three-manifold also
produces a C R structure. C R structures have been extensively studied
 283

since they arise naturally on the boundaries of complex manifolds.

The following observation will be basic to our study.

2.5 T h o o ~ m . Let M be an oriented three-manifold

with contact structure B. For every choice of
contact form w and a CR structure J there exists a
unique Riemann~an metric g adapted to the contact
form w and inducing the CR s t r u c t u r e J.

Proof. The form ¢o determines the unit vector field V

perpendicular to B. The metric on B is determined by the conformal
structure J and the volume form ,o~IB = 1 d w I B '

3. Structural equations. W e beg:m with a review of the

structural equations of Riemannian geometry. Let
w a':l<~a, ~,<dimM, be an orthonorm al basis of :l-forms on a
Riemannian manifold M. Then there exists a unique anti-symmetric
matrix of l-forms ~ u ~ such that the structural equations

(4) dwa + ~aB A w ~ = 0.

hold on M. The forms ~ a ~ describe the Levi-Civita connection of

the metric in the moving frame w a. We can also view the coa
as intrinsically defined l-forms on the principal bundle of orthonormal
bases. Then the ~ a B are also intrinsically defined as 1-forms on this
principal bundle, and the collection {c0a, ~a~} forms an
orthonormal basis of l-forms in the induced metric on the principal
bundle. The curvature tensor R a ~v 5 is defined by the
structural equation

(5) dea~eavAev~s+Ra~vswvAws,l<a,~,v,~<dimM,

where the summation convention applies.

In three-dimensions it is natural to replace a pair of indices in
an anti-symmetric tensor by the third index. Thus w e will write ~12
 284

= W3 and R1212 = K33, etc. Here K a $ are the components of the

Einstein tensor

{6) Ka~ = ~ } ~ o e - RaB'

which has the property that, for any unit v e c t o r V, K{V,V) is the
Riemannian sectional curvature of the plane V ±. The structural
equations then take the following form.

3.1 Structural equations in three ~mensions.

d~ 1 = WzA~3 - W3A~2,

{7} dw 2 = w3Awl - ~lAto3 ,

dw 3 = WlAW2 - ~2Aw1,
and

d~ 1 = ~2A~o3+K11~o2ALo3+K12Lo3A~o1+K13~o1A~02,

{8) d~o2 = ~3A~1+K21~2Aw3+K22~3Aw1+K23~o1Aw2,

d~3 = W1A~ 2 +K31 Lo2 A~o3+K32 w3 A~ 1+K 33 ~ 1Aw2 ,K a ~ = K ~a"

If the metric is adapted to the c o n t a c t from w, we choose

the frames such that w3 = to. As a consequence K33 is the sectional
curvature of the plane V ± and ~ (K11+K22) is the Ricci curvature in
the direction V. The Webster curvature is defined by

(9} W = (K11+K22+2K33+4)

and has remarkable properties.

We proceed to illustrate these equations with three examples
which are very relevant to our discussion, the sphere S 3, the unit
tangent bundle of a compact orientable surface of genus > 1, and the
 285

Heisenberg group H 3.

3.2 Example. The sphere S 3 is defined by the equation

(10) x2+y2+z2+w 2 = 1

in R 4. Differentiating we get

(11) ~ 0 = xdx + y dy + z dz + w dw = 0.

A specific choice of an orthonormal basis in the induced metric is

e I = x dy - y dx + z d w - w dz,

02) w2 = x dz - z dx + y dw - w dy,

= x dw - w dx + y dz - z dy.
~3

The reader can verify that i f <dx,dx> = 1, <dx,dy> = 0, e t c . , then

<Wl,Wl > = 1, <~1,w2 > = 0, etc., and that <¢~0,w0 > = 1, <w0,Wl>
= 0. e t c . T a k i n g e x t e r i o r d e r i v a t i v e we h a v e

(13) dw I = 2 w 2 A ~ 3, d ~ 2 = 2w3Aw 1, d ~ 3 = 2 ~ l A ~ 2,

and h e n c e m t h i s b a s i s

(14) ~Pl = w1' ~'2 = w2' ¢P3 = t°3"

which makes

(15) Kll = 1, K22 = 1, K33 = 1.

a n d t h e o t h e r e n t r i e s a r e zero. T h e W e b s t e r c u r v a t u r e W = 1.

3.3 Example. The unit tangent bundle of a compact

o r i e n t a b l e s u r f a c e o f g e n u s ;~1.
 286

Let N be a compact orientable surface of genus g. If N is

equipped with a Riemannian metric, its orthonormal coframe 0 I, 02,
and the connection form 012 satisfy the structural equations

(16) dO 1 = 012A02, dO 2 = 01A012, d012 = -KO1AO 2,

where K is the Gaussian curvature. Suppose g g l . We can choose

the metric such that

P
K - c - ~ +I , when g-0,
(17)
t -I , when g>l .

The unit tangent bundle T I N of N, as a three-dimensional manifold,

has the metric

1
(18) ~" (e 2I + 022 + 012).

Putting

(19) Wl = ~ el' WZ = T e2, w3 = - ~ ¢e12,

we find

(20} d~ 1 = 2 ¢ w 2 A w 3, dw 2 = 2c~3A~01, dw 3 = 2 W l A W 2,

and

(21) ~1 = Wl' ~2 = w2' ~3 = (2¢-1)~3"

It follows that

(22} Kll = K22 = I, K33 = 4¢-3,

all other Kcl$'s being zero. By (9) we get

W ~ c.
 287

This includes the example in ~3.2 w h e n g = 0, for the unit tangent

bundle of S 2 is the real projective space R P 3, which is covered by S 3,
and our calculation is local. On the other hand, TIN, for g > I, has
a contact structure and an adapted Riemannian metric with W = -I.

3.4. Example. The Heisenberg group.

We can make C 2 into a Lie group by identifying (z,w) with the
matrix

(23)

The subgroup given by the variety

(24) ~+w+W=O

is the Heisenberg group H 3. The group acts on itself by the

translations

Z --.# Z + 8,

{25)
w'-'* w - K z + b.

which leave invariant the complex forms

{26) dz and d w + f dz.

Hence an invariant metric is given by

{27) ds 2 = i !2+ Idw f d 12.

+

Introduce the real coordinates

{28) z = x + iy w = u + iv.

Then the v a r i e t y (24) is

 288

(29) x 2 + y2 + 2 u = 0

and differentiation gives

(30) du + x dx + y dy = 0.

Then an orthonormal basis of 1-forms in the metric above is given by

(31) Wl = dx, =2=dY, to3=dv + x dy - y dx,

and we compute

I
dwliO, dw2=O, d w 3 = 2 ~ i A w 2,
(32) ~ I = I ' ~ 2 = ~ 2 " ~3=-w3 '
K11 I , K 2 2 = I , K33=-3,

and the other entries are zero. By (9) we have W=0. All these
examples give metrics adapted to a c o n t a c t form w=w3, since in an
orthonormal basis =w3 = wlAw 2.
In general, given a metric adapted to a c o n t a c t form ~, we
shall r e s t r i c t our attention to orthonorma] bases of 1-forms Wl' w2'
~3 with w3=co. Considering the dual basis of vectors, we only need to
choose a unit v e c t o r in B. These form a principal circle bundle, and
all of our structural equations will live naturally on this circle bundle.
I t turns out to be advantageous to compare the general situation to
that on the Heisenberg group. Therefore, we introduce the forms
q/l' ¢/2' ¢3 and the matrix L l l , L12 ..... L33 defined by

~i=#i+~i " ~2=~2+~2 • ~3=#3-~3 ,

(33) KI I = L I i +I , K 2 2 = L 2 2 +I , K 3 3 = L 3 3 - 3 ,
KI2=L12, KI3=LI3, K23=L23.

Thus the ~ and L all v~ni~h on the Heisenberg group. We then

compute the f o U o ~ .
 289

3.5. Structure equations for an adapted metric. T h e y are:

d~ 1 = ¢2A~3-~3Aw2,
(34) dw 2 = @ 3 A W l - C l A W 3 ,
d~ 3 = 2~1A~2,

and

{ ~lAw2 - W2A~I = o,
(35)
¢IA~I + ~2Aw2 = o,

and

de 1 = @2A@3+LllW2Aw3+L12w3A~I+L13WlAW2 ,
(36) de 2 = ¢3A¢l+L21w2A~3+L22~3A~I+L23w1Aw2 ,
de 3 = ~lA¢2+L31~2A~3+L32~3A~l+L33wIA~2 .

Proof. The equation dw 3 = 2WlAW2 comes from the

condition dw = 2.~0 t h a t t h e metric is a d a p t e d to the c o n t a c t form
t~. Then the corresponding structural equation yields ~lAW2 -
¢2A~1 = 0. Using ddw 3 = 0 we compute ¢1Awl + ¢,2Aw2 =
0 also.

3.6. Corollary. We can find functions a and b on the

principal circle bundle so that

@I = aWl+bW2'
(37)
~2 = b W l - a ~ 2 "

Proof. This follows a l g e b r a i c a l l y from the e q u a t i o n s (35).

It is even more c o n v e n i e n t to write t h e s e equations in complex
form. We make t h e following substitutions.
 290

3.7. Complex substitutions.

On account of the complex structure in B it is convenient to
use the complex notation. W e shall set:

Q = wl+i~ 2, W = W 3,

t = @1+i~2 , ~ = @3'
Z = a+ib,

p - (LII +L22 ) , q
~l . ~1 (LII-L22), r = L 12'
(38) s q+ir,
=
1
z - ~ (LI3+iL23) ,
t=L
33'
w = ~l ( t _ a 2 b 2 ),

where W is the Webster curvature, to be verified below. Note that

= L Q. Thus D and ~ give a basis for the l-forms on M, while z
and @ define the connection.

3.8 Complex structur~ ~uations.

(39) dO = i(@AO-zQA~),

and

{ d@ = i[2wQA~+(zO-iQ)A~],
(4O) dz ~ i(2Z@+zO-sw) mod O,
p+l~l 2 = O.

Proof. This is a direct computation. Note that the real

functions p,W and the complex functions z,s give the curvature of the
metric.
The equation p+[ z [2 = 0 has the important consequence that w e
can compute the Webster curvature W from the KaB. The result is
the expression for W in (9).
 291

The following notation will be useful. If f is a function on a

Riemannian manifold with frame ~a' then

(41) df = D a f • w a,

where D a f is the derivative of f in the direction of the dual v e c t o r

field V a. If f is a function on the principal bundle then we can still
define D a f as the derivative in the direction of the horizontal lifting
of V a. In this case we will have

(42) df _= D a f - ~ a rood ~ a ~ "

I f the function f represents a tensor then Daf are its covariant

derivatives, and the extra terms in ~ a B depend on what kind of
tensor is represented. In the example if T is a c o v a r i a n t 1-tensor
and

(43) f = T(Vv),

then,

(44) df = Daf-~o a + T(V$) ~ $ v '

while if T is a covariant 2 - t e n s o r and

(45) f = T(V v, Vs),

then

{46) df = D a f - • a + T(V~, VS) W ~ . + T(Vv, V~) ~ B ~ '

and so on. In the complex notation we write

{471 df = a ¢ . O + ~ f - O ÷ D v f . W
 292

as the definition of the differential operators ~f, ~f, and Dvf As

usual

~f = ~ (Dlf - iD2f),
(48) ~f - ~ (Dlf + i D 2 f ) ,
Dvf " D3f ,

reflecting the transition from real to complex notation. If f is a

function on the principal circle bundle coming from a symmetric
k-tensor on B then

(49) df = ~f-t~ + ~ f - Q + Dvf - ~o + ikfq.

For example, the function z represents a trace-free symmetric 2-form

on B, and the structural equation for L tells us

3.9. [,emma.

(50) ~L = iz and DvZ = -is.

4. Change of basis. W e start with the simplest change of

basis, namely rotation through an angle 0. We take 0 to be a
function on M and study what happens on the principal circle bundle.
The new basis ~ , w[, w~ is given by ~=~3=~ and

w ~ = cos.8 Wl-sin, e w 2,
(51)
~ = sin.e Wl+COS.e co2

or in complex terms ~= = ~ and

(52) Q* = e iO O.

Then from the structural equations w e immediately fred that

 293

4.1. Lemma.

¢= = ¢ + dO,
(53)
z" = z e 2i~.

Now a function or tensor on the principal circle bundle comes from

one on M by the pull-back if and only if it is invariant under rotation
by @. Thus w e see that the curvature form de = = d¢~ is invariant
and hence lives on M. The form OAO is also invariant, so W =
W = is invariant and W is a function on M. This W is the scalar
curvature introduced by Webster (see [11]}. Likewise [z [2 is invariant
s

and hence a function on M. The function z defines a tensor zO 2

which is invariant. Hence its real and imaginary parts

2 2
a{~l-W 2) + 2b ca)-]t~ 2 ,

(54}
- 2°

define trace-free symmetric bilinear forms on B {they differ by

rotation}. This form is called the torsion tensor by Webster (see [11]};
it is analogous to the second fundamental form for a surface.
We now consider more interesting changes of basis. First w e
change the C R structure while leaving the contact form w fixed. In
order to keep the metric adapted to the contact form we must leave
W l A ~ 2 invariant. This gives a new basis

w~ = Au)I+B~2,
(55) ,~ = C¢~I+DW2,
w ~ = t~3

with A D - B C = I. An infinitesimal change of basis is given by the

tangent to a path at t = 0. Thus an infinitesimal change of the basis
which changes C R structure but leaves the contact form invariant and
keeps the metric adapted is given by
 294

~ i = gWl+hC~ 2,
~ 2 = kwl+lw2'
w3=0

with g+l = O. Since t h e r o t a t i o n s are trivial we may as well t a k e

h=k. This gives

~ 1 = gc°l+h~2'
(56) ~2 = h~l-g~2'
to3 = 0.

In complex notation if f = g+ih then

{57) f~ = fQ and ~o = O.

For f u t u r e use we c o m p u t e the infinitesimal change ¢ in ¢

J

and L in z from t h e s t r u c t u r a l e q u a t i o n s {39t, i40}. We find t h a t f

transforms as a 2-tensor

(58) df = Eft. O + ~ f . ~ + D v f . u~+2if¢

and that

4.2. Lemma.

z = -i Dvf,
(59)
J

= i(Df-CY-~f-Q) - {zf+~f)w

using the fact that w e know 4~AQ and ¢ is real.

On the other hand w e may wish to fix the C R structure and
change the contact form while keeping the metric adapted. In this
case let ¢ ~ = f2~ 3 where f is a positive real function. Excluding
rotation w e find that to keep the metric adapted w e need
 295

w~ = f- ~l-D2f- uJ3,
(60) ~ = f . w 2 + D l f . w 3,
uJ~ = f2u>3,

In c o m p l e x n o t a t i o n

O" = fO + 2i ~ f ' w ,

(61)
~, = f2~.

F o r an i n f i n i t e s i m a l v a r i a t i o n w e d i f f e r e n t i a t e t o o b t a i n

/ / /

0 = f 0+2i ~ f -~,
{62)
/ /

w = 2f w.

Hence changes of metric fixing the CR structure are given by a

p o t e n t i a l f u n c t i o n f, much t h e s a m e w a y a s c h a n g e s o f m e t r i c fixing a
conformaJ s t r u c t u r e . T h e main d i f f e r e n c e is t h a t t h e d e r i v a t i v e s o f f
e n t e r t h e f o r m u l a f o r t h e new basis.
As a c o n s e q u e n c e of d d f = 0 w e h a v e

(63) D~f - ~Df + iDvf = 0.

W e also define the sub-Laplace operator

(64) l::If = 2(~)B'f+BDf} = (DIDlf + DBD2f).

Then a straightforward computation substituting in the structural

equations yields

(4.3. L,emma.)
 296

4.3.

(65)
z* = z - 2 " ~ ' - 6 2

Differentiating the first we get

dq* - dqt-2i l ~ oAH rood co,

which shows the remarkable relation given by

4.4 [,emma.

(66) f3W* = fin - Eft.

4.5. Corollary. In an infinitesimal variation

/' s s

(67) W = -of - 2 f W.

5. Energies. Let ~ be the measure on M

(68) U = WlhW2hw 3 = ~ rlhHhw

induced by the metric. Here are two interesting energies which we

may form. The first is

(69) EW =I MW~t,
which is analogous to the energy

(70) E = f R
J M

in the Yamabe problem. The second is

 297

(71) Ez = f M Izl2 /~'

which is a kind of Dirichtet energy.

In this section we shall study the critical points of these
energies.
First we observe that for computational reasons it is easier to
integrate over the principal circle bundle P. The measure there is

(72) r : ~IAWZA~3A¢3 : ½ QAQA~A¢.

If f is a function on the base M then

(73) jrP f v = 2K ,[M f U,

so nothing is lost.
Next we observe that we can integrate by parts.

5.1. Lemma. For any f on P

{74) ~pOf" ~ = 0 and ~ p D V f - v = 0.

Proof. The first follows from

~p d(fQA~A~} = 0
and the second follows from

~ p d{fOA~q~} = O,

since dQ~O rood ~, ~ and dw~O rood Q, ~ and d#~O rood Q. ~,

5.2. TheoreL The emero~ EW i s critical over a~t

contact ]orms with a ]ixed CR structure and ]ixed
votume i1 a~d o~ty i] W is constant. It is criticat
 298

over all CR s t r u c t u r e s with a ]ixed contact 1arm i1

and only ~f L = O.

S
Proof. W e compute the infinitesimal variation E W . Fixing the
CR s t r u c t u r e and varying the potential f of the c o n t a c t form with w s
= f2w gives ~ = 4f ~ and

EW = {-Elf + 2f W) v = 2 f W y,
P

since o integrates away. The volume is fixed when ~if" v = 0.

Thus, F_~ = 0 precisely when W is constant.
Fixing the contact form and varying the C R structure we use
the following.

5.3 Lemma.

(75) ~ = ~ Ipd¢AwA¢.

Proof. We use the s t r u c t u r a l equation to see

dv~A~ = 2iWOAQA~a

and i n t e g r a t e by p a r t s to g e t the result. Then we have

EW = A~A¢ + dCA~A¢

(using ta = 0), and this gives

F.w " _ _ . ~~p i ,/, "AD~SA~,.

Then using Lemma 4.2 we g e t

EW" = - ½ I p (Lf+~f) v.

so that the C R structure is critical for fLxed w precisely when z=O.

 299

Next we c o n s i d e r t h e e n e r g y E L .

5.4. Theorem. The energy E z is critical over

all CR structures with fixed contact form if and
only if DV~ = o, w h i c h is equivalent to s = O, o r Kll =
K22 and K12 = 0. The energy E L is critical over all
contact forms with fixed CR structure and fixed
volume if and only if

(76) 2i(~z - ~z-) + 3p = c o n s t a n t .

Proof. T h e e n e r g y E L is g i v e n by

Ez
_-f p
so its f i r s t v a r i a t i o n is

E z = (L ~ + L 2 )p + z y .
P

When w is fixed, ~ = 0 and p = 0. By Lemma 4.2 we have the

P

r e s u l t t h a t ff C/ = f ~ then L = - i D v f , and this gives

5.5. Lemma.

E~
-= 2 Im
f P {Dv~ ~ .
S i n c e f is a n y r e a l f u n c t i o n on M, we s e e Ez = 0 w h e n DVL - 0.
T h e n s = 0 by Lemma 3.9 and K l l = K22 a n d K12 = 0 by s u b s t i t u t i o n
{38).
This c o n d i t i o n says that, at each point of M, the sectional
c u r v a t u r e of all p l a n e s p e r p e n d i c u l a r t o t h e c o n t a c t p l a n e B a r e equal.
If, on t h e other hand, we fix t h e CR s t r u c t u r e and v a r y t h e
c o n t a c t form by a p o t e n t i a l f, we have from Lemma 4.3
 300

Z'= ~: - 2 -6 2.

Taking an infinitesimal variation

L " = - 2~" . ~ " = 3f " ~'

Then the variation in E z is

Ez
"
=
fP
(-2(~ a~f
"
+ ~f'} + 3
IIL 2 f ' } ~,

from which we see that E z = 0 precisely when

2( a, + 3 l,j 2

is constant. Since ~z = iz by L e m m a 3.9, and I z 12+p = 0, this gives

the equation {76).

6. Changing Webster Scalar Curvature. The problem of

fixing the C R structure and changing the Webster scalar curvature is
precisely analogous to the Y a m a b e problem of fixing the conformal
structure and changing the scalar curvature, except the problem is
subelliptic, and the estimates and constants for the 3-dimensional C R
case look like the 4-dimensional coP,formal case. The first result is
the following.

6.1. Theo~m. Let M be a c o m p a c t orientable

three-manifoZd with fixed CR structure. Then we can
change the contact ]orm so that the Webster scalar
cur~ature W o1 the adapted Riemannian metric is
either positive or zero or ~egative everywhere.

Proof. We have f3W" = f W - Elf from [,emma 4.4. We take

f to be the eigenfunction of W - Q with lowest eigenvalue k I. By
the strict maximum principle for subelliptic equations (see Bony [1]) w e
conclude that f is strictly positive. Since W f - of = X1f w e have
 301

f2W* = k 1. Hence W* always has the same sign as X].

Next we show that in the negative curvature case we can make
W whatever we want, in particular, a negative constant.

6.2. Theorem. Let M be a compact orientable

three-maniYold with a lixed CR structure, lI some
contact form has negative Webster scalar curvature,
then every negative function W<0 is the Webster
scalar curvature of one and only one contact form ~.

Proof. Let C be the space of all c o n t a c t forms and let Y be

the space of functions. We define the operator P by

P:C --=* Y, P(w) = W.

Let Y- be the space of negative functions and let C- be the space of

contact forms with negative Webster curvature. Then

(77) P:C- -'* Y-

is also defined. We claim the P in (77) is a global dfffeomorphism.

This follows from the following observations.

a) C- is not empty.

b) P is locally invertible.

c) P is proper (the inverse image of a compact set is

compact).

d) if- is simply connected.

We then argue that (a) allows us to start inverting somewhere,

(b) allows us to continue the inverse along paths, (c) says that the
inverse doesn't stop until w e run out of Y-, and (d) tells us that the
inverse is independent of the path and hence unique.
 302

Before we start the proof we remark on a few technical

details. There are two possible approaches to the proof. One is to
work with C °o functions and quote the Nash-Moser theorem (see [5] for
an exposition) using the ideas in [6] to handle the subelliptic
estimates. The other is to work with the Folland-Stein spaces S~
{see [3]) which measure k derivatives in the direction of the c o n t a c t
structure in Lp norm. We can take ¢~¢8k÷
p 2 and W¢S p
provided pk>8 so that W e C ° by the appropriate Sobolev inclusion.
The easiest case analytically is to take p = 2, which necessitates
k~>5.
We proceed with the proof. Observation {a} follows from the
hypothesis. To see {b) we compute the derivative of P, and apply the
inverse function theorem.
In fact, from Corollary 4.5 w e write

I S I

f + 2CNf = -W ,

by putting dashes on the original metric. The operator C] + 2~/ has

zero null space by the maximum principle, since W < 0. Since it is
self-adjoint, it must also be onto and hence invertible. This proves
that D P is invertible when ~" < 0, and so P is locally invertible on all
of C-.

To see assertion {c) that P is proper, we apply the maximum

prineipie to the equation

Where f is a maximum Clf~<0, and where f is a minimum ~¢~0.

Since W and W are both negative w e get the estimate

1 1

(78) /W)min
1 ~< f min ~< f max ~< /W) max
1
Notice that the estimate fails ff W and ~N are positive. Having
control of the m a x i m u m and minimum of f, it is easy to control the
 303

higher derivatives using the equation and the subelliptic Garding's

inequality

(79) Ilflls~¢+ 2 ~< C {llC-2flls[ + ilfllLp).

In the C °o c a s e this shows P is proper. For given any c o m p a c t s e t of
M, we have uniform bounds on Wma x and
iIWII~p. Wmi n and all
~k
This gives bounds on fmax and fmin and all Ilflls~¢ for all f in the

preimage, so the preimage is compact since Coo is a Montel space.

To work in the Banach space Skp we also need the following
observation. Suppose we have a sequence of c o n t a c t forms ton with
W n --* W < 0 in S p. The previous e s t i m a t e s give bounds on ~n in
S k2 +, p which implies convergence of a subsequence in S p. Let
2-
tan - - , G, and write w n = fnCa. The maximum principle estimate
shows fn --* 1 in C °. Then using the equation we get the e s t i m a t e

(80) Ilfn-ltis~+2 .< C IIWn - ~ q l t s [ ;

.__,
this shows wn & in S kP + 2, and proves P is proper.
The assertion (d} that ~- is simpb" connected follows by
shrinking along straight line paths to W = - I . This completes the
proof of the theorem.

7. ]~inimimng Torsion. We consider finally the problem of

minimizing the energy

(81) Ez P

r e p r e s e n t i n g the L 2 norm of the torsion by the h e a t equation with the

c o n t a c t form ~ fixed. From Lemma 5.5 we have the result t h a t if we
J

t a k e a path of O's depending on t with O = fO then

E z'= 2 lm J'p f ' D v L )J.

Following the gradient flow of E z we let f = i I ~ . z . This gives

 304

h e a t equation for E z. Since DVL = - is by Lemma 3.9 we g e t the

following results.

7.1. Heat Equation Formulas.

o = i Dr,-5,

(82) Ez =-2 p s 2~,

2
z = DvL.

T h e s e equations show t h a t if the solution e x i s t s f o r all time then t h e

e n e r g y E z d e c r e a s e s and the c u r v a t u r e s ~ 0. The equation
2
z =Dv z is a highly d e g e n e r a t e p a r a b o l i c equation, since the right
hand" side involves only the second d e r i v a t i v e in t h e one d i r e c t i o n V.
Nevertheless, it is not a bad equation, since t h e maximum principle
applies. This shows that the maximum absolute value of z
decreases. The equation is in f a c t just the o r d i n a r y heat equation
r e s t r i c t e d to each orbit in t h e flow of V. P h y s i c a l l y we can imagine
the manifold P to be made of a bundle of wires i n s u l a t e d from each
other, with the h e a t flowing only along t h e wires. When the orbits of
V are closed, the analysis should Be f a i r l y s t r a i g h t f o r w a r d . When the
o r b i t s of V are dense, things a r e much more c o m p l i c a t e d , and probably
lead to small divisor problems.
A r e g u l a r f o l i a t i o n is one where each l e a f is compact and the
s p a c e of l e a v e s is Hausdorff. In this c a s e we always have a S e i f e r t
foliation, one where each l e a f has a neighborhood which is a f i n i t e
quotient of a bundle. In three dimensions the Seifert foliated
manifolds a r e w e l l - u n d e r s t o o d by t h e t o p o l o g i s t s , and provide many of
t h e nice examples. Wc c o n j e c t u r e t h e following result.

7.2. Conjecture. Let M b e a c o m p a c t t h r e e - m a n i f o l d with a

fixed c o n t a c t form Lo whose v e c t o r field V induces a Self e r r foliation.
There there exists a CR structure on M such that the associated
metric has s = 0, i.e., the s e c t i o n a l c u r v a t u r e of all planes a t a given
point p e r p e n d i c u l a r to the c o n t a c t bundle B = Null ¢a are equal. The
 305

metric is obtained as the limit of the heat equation flow as t --* **.

REFERENCES

J s

[i] J.M. Bony, Principe du maximum, inegalite de Harnack, et

unicite du prob|eme de Cauchy pour les o p e r a t e u r s elliptiques
• / I

degeneres, Ann. Inst. Fourier 19(1969), 2?7-304.

[21 A. Douady, Noeuds et s t r u c t u r e s de c o n t a c t en dimension 3,
d'apres Daniel Bennequin, Seminaire Bourbaki, 1982/83, no.
604.
[3] G.B. Folland and E.M. Stein, Estimates for the ~b-complex
and analysis on the Heisenberg group, Comm. Pure and App.
Math 27(1974), 429-522.
[4] J.W. Gray, Some global properties of contact structures,
Annals of Math 69(1959), 421-450.
[5] R. HamiIton, The inverse function theorem of Nash and Moser,
Bull. Amer. Math. Soc. 7(1982), 65-222.
16] R. Hami]ton. Three-manifolds with positive Ricci curvature, J.
Diff. Geom. 17(1982). 255-306.
[7] D. Jerison and J. Lee, A subelliptic, non-linear eigenvalue
problem and scalar curvature on CR manifolds, Microlocal
Analysis, Amer. Math Soc. Contemporary Math Series,
27{1984), 57-63.
[8] J. Martinet, Formes de c o n t a c t sur Ies v a r i e t e s de dimension
3, Proc. Liverpool Singularities Syrup II, Springer Lecture
Notes in Math 209(1971), 142-163.
I9] R. Schoen, Con_format deformation of a Riemannian metric to
constant s c a l a r curvature, preprint 1984.
[lO1 W. Thurston and H.E. Winkelnkemper, On the existence of
c o n t a c t forms~ Proc. Amer. Math, Soc. 52(1975), 345-347.
[11] S.M. Webster, Pseudohermitian structures on a real
hypersurface, J. Diff. Geom. 13(1978), 25-41.
 306

APPENDIX

by
Alan Weinstein

THREE-DIMENSIONAL CONTACT MANIFOLDS

WITH VANISHING TORSION TENSOR

In a lecture on some of the material in the preceding paper,

Professor Chern raised the question of determining those 3-manifolds
admitting a c o n t a c t s t r u c t u r e and adpated Riemnnnian metric for which
the torsion invariant c2=a2+b 2 is identically zero. (See ~3. A
variational characterization of such structures is given in Theorem
5.2.) The purpose of this note is to show that the class of manifolds
in question consists of certain Seifert fiber manifolds over orientable
surfaces, and that the first real Betti number b]{M) of each such
manifold M is even. These results are not new; see our closing
remarks.
By a simple computation, it m a y be seen that the matrix

b -ab1 (see Corollary 3.5) represents the Lie derivative of the

ihduced inetric on the c o n t a c t bundle B with r e s p e c t to the c o n t a c t
vector field V. We thus have:

A.]. [..emma. The invariant c2 is identically

zero i] and only i] V is a killing vector ]ield. (In
other words, M is a "k-contact manifold"; see [I]3

We would like the flow generated by V to be periodic. If this

is not the case, w e can m a k e it so by changing the structures in the
following way. Let G be the closure, in the automorphism group of M
with its contact and metric structures, of the l-parameter group
generated by V. G must be a torus, so in its Lie algebra w e can find
Killing vector fields V' arbitrarily close to V and having periodic flow.
Let • be the ]-form which snnihilates the subbundle B' perpendicular
I

to V' and which satisfies w {V') - 1. For V' sufficiently close to V,

I

w will be so close to the original contact form ~ that it is itself a

 307

contact form. Since the flow of V' leaves the metric invariant, it
leaves the invariant the form to , from which it follows that V' is the
P
contact vector field associated with ~o .
Having made the changes described in the previous paragraph,
we may revert to our original notation, dropping primes, and assume
that the flow of V is periodic. A rescaling of • will even permit us
to assume that the least period of V is 1. {Note that, by Gray's
theorem [2], w e could actually assume that the n e w contact structure
equals the one which was originally given.)
Suppose for the moment that the action of S] = IR/Z
generated by V is free. Then M is a principal S 1 bundle over the
surface M / S 1. The form w is a connection on this bundle; since
is a contact form, the corresponding curvature form on M / S I is
nowhere vanishing. Thus M / S 1 is an orientable surface, and the
Chern class of the flbration M --* M/S I is non-zero. By the
classification of surfaces, bl(M/SI) is even; by the Gysin s e q u e n c e ,
b1(M) = bI{M/SI) and is therefore even as well.
We are left to consider the case where the action of S I,
although locally free, is not free. The procedure which w e will follow
is that of [8]. Let r~s I be the {finite) subgroup generated by the
isotropy groups of all the elements of M. Then M is a branched
cover of M/r, and M / r is a principal bundle over M / S I with fiber
the circle S1/r. The branched covering map M --, M / r induces
isomorphisms on real cohomology, so it suffices to show that b1(M/r)
is even. To see this, we consider the fibration
s 1 / r - - . M / r - - , M / S 1. The quotient spaces M/r and M/S I are
V-manifolds in the sense of [4], and w e have a fibre bundle in that
category. The base M / S I is actually a topological surface which is
orientable since it carries a nownere-zero 2-form on the complement
of its singular points. Now the contact form may once again be
considered as a connection on our V-fibration, and so, just as in the
preceding paragraph, w e m a y conclude that b1(M/r) is even.

Remarks. A K-contact manifold is locally a 1-dimensional

bundle over an almost-K~hler manifold. When the base is Kahler, the
contact manifold is called Sasakian. Using harmonic forms, Tachibana
 308

[5] has shown that the first Betti number of a compact Sasakian
manifold is even. On the other hand, since every almost complex
structure on a surface is integrable, every 3-dimensional K-contact
mainfold is Sasakian, and hence our result follows from Tachibana's
theorem. In higher dimensions, compact symplectic manifolds with odd
Betti numbers in even dimension are known to exist [3] [7], and circle
bundles over them will carry K-contact structures, while having odd
Betti numbers in even dimension.
The paper [6] contains a study of which Self err fiber manifolds
over surfaces actually admit sl-invariant contact structures.
Acknowledgments. This research was partially supported by
NSF Grant DMS84-03201. I would like to thank Geoff Mess for his
helpful advice.
R E F E R E N C E S

. D. Blair, Contact Manifolds in Riemannian Geometry, Lecture

Notes in Math., vol. 59 {t976}.
2. J. Gray, Some global properties of contact structures, Ann. of
Math. 69 {1959}, 421-450.
3. D.McDuff, Examples of simply connected symplectic manifolds
which are not Kahler, preprint, Stony Brook, 1984.
4. I. Satake, The Gauss-Bonnet theorem for V-manifolds, J.
Math. Soc. Japan 9 (1957), 464-492.
. S. Tachibana, On harmonic tensors in compact Sasakian
spaces, Tohoku Math. J. 17 {1965}, 271-284.
. C.B. Thomas, Almost regular contact manifolds, J. Diff. Geom.
11 (1976), 521-533.
. W. Thurston, Some simple examples of symplectic manifolds.
Proc. A.M.S. 55 {1976}, 467-468.
. A. Weinstein, Symplectic V-manifolds, periodic orbits of
hamiltonian systems, and the volume of certain riemannian
manifolds, Comm. Pure Appl. Math. 30 (1977), 265-271.
 4-MANIFOLDS WITH INDEFINITE INTERSECTION FORM

S.K. Donaldson
All Souls College
Oxford, England

In w r i t i n g up this lecture I shall not concentrate so m u c h on des-

cribing problems of 4-manifold topology; instead I shall explain how a
simple topological construction has applications in t w o different direc-
tions. First I will recall that, just as b u n d l e s over a single space have
homotopy invariants, so do families of b u n d l e s , and that these define cor-
responding invariants in f a m i l i e s of connections. Next I will sketch the
way in w h i c h such a topological invariant, when endowed with a geometric
realisation, becomes important for studying holomorphic bundles over al-
gebraic varieties. Last I will indicate how this same homotopy invariant
of families of connections, combined with arguments involving moduli spa-
ces of self-dual connections over a Riemannian 4-manifold, gives restric-
tions on the possible homotopy types of smooth 4-manifolds and I will
speculate on possible future progress in t h i s area.

Topology of b u n d l e s .

This is standard material that may be found in [2] for example. Con-
sider a fixed manifold X and a family of b u n d l e s over X parametrised
by some auxiliary space T , so w e have a bundle P over the product
X × T with structure group G (compact and connected, say). Take first
the case when T is a point so we h a v e a single bundle over X , deter-
mined up to equivalence by a homotopy class of m a p s from X to BG .
This may be non-trivial, detected for example by characteristic classes
in t h e cohomology of X . If w e choose a connection A on the bundle
the real characteristic classes can be represented by explicit differen-
tial forms built from the curvature of the connection. Equally if D is
an elliptic differential operator over X then using a connection it m a y
be extended to act on objects (functions, forms, spinors etc.) twisted by
a vector bundle associated to P . This has an integer valued index:

index (DA) = dim ker D A - dim coker DA

 310

which is a rigid i n v a r i a n t of the bundle, i n d e p e n d e n t of the c o n n e c t i o n .

So these are two ways in w h i c h the u n d e r l y i n g h o m o t o p y may be r e p r e s e n -
ted g e o m e t r i c a l l y , by c u r v a t u r e and by d i f f e r e n t i a l operators. The C h e r n -
Weil and A t i y a h - S i n g e r t h e o r e m s then give f o r m u l a e r e l a t i n g the three.

In the same way for a g e n e r a l f a m i l y p a r a m e t r i s e d by T the b u n d l e

P is c l a s s i f i e d by a h o m o t o p y class of m a p s f r o m T to the m a p p i n g space
Maps(X,BG) , and at the o t h e r e x t r e m e f r o m the case T = point we h a v e
a universal f a m i l y p a r a m e t r i s e d by this m a p p i n g space. A g a i n we m a y al-
w a y s c h o o s e a c o n n e c t i o n over X × T , w h i c h we may think of as a f a m i l y
of c o n n e c t i o n s p a r a m e t r i s e d by T, and c o n v e r s e l y any family of e q u i v a -
lence c l a s s e s of c o n n e c t i o n s on some b u n d l e e s s e n t i a l l y a r i s e s in this
way. (This is p r e c i s e l y true if we w o r k w i t h b a s e d m a p s and b u n d l e s , re-
moving base points gives small t e c h n i c a l d i f f e r e n c e s w h i c h can s a f e l y be
i g n o r e d here). Equivalently we have the infinite d i m e n s i o n a l space B of
all e q u i v a l e n c e c l a s s e s of c o n n e c t i o n s o b t a i n e d by d i v i d i n g the a f f i n e
space of c o n n e c t i o n s A by the b u n d l e automorphism group G . B has the
h o m o t o p y type of Maps(X,BG) .

A g a i n we may c o n s t r u c t topological invariants of such f a m i l i e s of

bundles. In c o h o m o l o g y we can use the c h a r a c t e r i s t i c classes again. There
is a slant product:

H p+q(x × T) ® H (X) - - > H p(T)

q

so that c h a r a c t e r i s t i c c l a s s e s of b u n d l e s over X xT c o n t r a c t e d with, or

i n t e g r a t e d over, homology classes in the base m a n i f o l d X yield coho-
mology classes in f a m i l i e s of c o n n e c t i o n s . In p a r t i c u l a r if G is, say,
a u n i t a r y group we o b t a i n in this way a map:

: H 2 (X) > H 2 (T)

~(~) = c2 (P)/~

(A s i m p l e r e x a m p l e is to take the J a c o b i a n p a r a m e t r i s i n g c o m p l e x line bund-

les o v e r a R i e m ~ n n surface. O p e r a t i n g in the same w a y w i t h the first C h e r n

class g i v e s the usual c o r r e s p o n d e n c e b e t w e e n the 1 - d i m e n s i o n a l h o m o l o g y
of the s u r f a c e a n d the c o h o m o l o g y of the J a c o b i a n ) . We can do the c o r r e s -
 311

ponding thing in K - t h e o r y and realise the r e s u l t i n g elements in the K-

theory of T by using differential operators again. For e x a m p l e if the
base m a n i f o l d X is the 2-sphere then a u n i t a r y bundle over S2 x T de-
fines an e l e m e n t of K(S 2 × T) w h i c h maps to K(T) by the inverse of the
Bott p e r i o d i c i t y map. If we take the Dirac operator D over S2 then a
family of c o n n e c t i o n s gives a family of Dirac operators {D t} parame-
trised by T and, after suitable stabilisation the index of this family
[2] defines the r e q u i r e d class:

index D t = [Ker D t] - [coker D t] 6 K(T)

Of c o u r s e we o b t a i n other classes in this way and the A t i y a h - S i n g e r index

theorem for f a m i l i e s gives formulae relating these to the u n d e r l y i n g ho-
motopy. In p a r t i c u l a r we may u n d e r s t a n d our class above from e i t h e r point
of view via the formula:

c1(index D t) : ~(fundamental class of S 2)

S_table bundles on algebraic curves and surfaces ~

Here I only want to say enough to fit into our overall theme; more
details and r e f e r e n c e s may be found in [4], but I learnt the p o i n t of
v i e w we are adopting now from lectures of Quillen.

There is a g e n e r a l algebraic theory dealing with the action of a

complex reductive group G~ on a v e c t o r space ~n+1 via a linear repre-
sentation. Equivalently we may take the induced action on ~n and the
hyperplane bundle H over it. In that theory there is a d e f i n i t i o n of a
"stable" point. Now suppose that ~n+1 has a fixed H e r m i t i a n metric, in-
ducing metrics on H and on ~pn , and p i c k i n g out a maximal compact
subgroup GcG ~ whose action preserves these metrics. There is a general
theory dealing w i t h the m e t r i c a l properties of these actions and r e l a t i n g
t h e m to the p u r e l y complex algebraic p r o pe r t i e s . Roughly speaking if we
restri c t to the stable points then a t r a n s v e r s a l to the G~-action on
~n is induced by taking the points in ~n+1 , or e q u i v a l e n t l y H -I ,
which m i n i m i s e the norm in their G~ orbits. The c o r r e s p o n d i n g variati-
 312

onal e q u a t i o n s c u t t i n g out the t r a n s v e r s a l take a simple form and are the

zeros of a map:

m : E~n > j, [7] , [8]

Large p a r t s of this t h e o r y can be d e v e l o p e d a b s t r a c t l y from g e n e r a l pro-

p e r t i e s of Lie g r o u p s and the fact that the c u r v a t u r e f o r m of the H e r m i -
lian line b u n d l e H g i v e s the K ~ h l e r symplectic f o r m on ~n

A t i y a h and B o t t [I] o b s e r v e d t h a t the t h e o r y of h o l o m o r p h i c struc-

tures on a v e c t o r b u n d l e E over an a l g e b r a i c curve C c o u l d be cast in
the same form, e x c e p t w i t h an infinite d i m e n s i o n a l a f f i n e space in p l a c e
of a p r o j e c t i v e space. For a h o l o m o r p h i c s t r u c t u r e on E is g i v e n by a
F-operator and these are p a r a m e t r i s e d by a c o m p l e x affine space A . The
infinite dimensional group G~ of c o m p l e x linear a u t o m o r p h i s m s of E
acts by c o n j u g a t i o n and the q u o t i e n t set is by d e f i n i t i o n the set of e q u i -
valence c l a s s e s of h o l o m o r p h i c (or algebraic) bundles, topologically equi-
v a l e n t to E Independently, and from a n o t h e r p o i n t of view, stability
of a l g e b r a i c b u n d l e s had been d e f i n e d in a l g e b r a i c g e o m e t r y ; the defini-
t i o n uses the n o t i o n of the degree of a bundle - the integer o b t a i n e d by
e v a l u a t i n g the first C h e r n class on the f u n d a m e n t a l cycle.

If n o w E has a f i x e d H e r m i t i a n m e t r i c t h e n a ~-operator induces

a unique unitary connection. R e g a r d e d as c o n n e c t i o n s the s y m m e t r y g r o u p
of the a f f i n e space A is r e d u c e d to the s u b g r o u p GcG ~ of u n i t a r y
a u t o m o r p h i s m s , and this s u b g r o u p p r e s e r v e s the n a t u r a l m e t r i c form on the
space of c o n n e c t i o n s A derived from i n t e g r a t i o n o v e r C . We w o u l d have
all the ingredients for the a b s t r a c t t h e o r y d e s c r i b e d above if we had a
Hermitian line b u n d l e i over A with c u r v a t u r e g e n e r a t i n g this m e t r i c
form, and a c t e d on by G~

It was e x p l a i n e d above that over a space of c o n n e c t i o n s we o b t a i n vir-

tual b u n d l e s from the a s s o c i a t e d e l l i p t i c o p e r a t o r s . In p a r t i c u l a r we can
take the D i r a c o p e r a t o r o v e r the a l g e b r a i c c u r v e C , w h i c h is the same
as the ~-operator after t e n s o r i n g w i t h a square root ~I/2 of the ca-
~C
n o n i c a l bundle, so the k e r n e l and c o k e r n e l form the usual sheaf c o h o m o l o -
gy. M o r e o v e r we get a g e n u i n e line b u n d l e if we take the h i g h e s t e x t e r i o r
p o w e r or d e t e r m i n a n t of the r e l e v a n t v e c t o r spaces. Thus we get a c o m p l e x
line b u n d l e iC over A :
 313

ic = X ( E ® K I/2) = det H0(E®K I/2 ) ® d e t H I (E®K I/2 ) -I

C C C

acted upon by G~ , and realising via the first Chern class the cohomo-
logy class obtained under our map ~ from the fundamental cycle of the
curve C , as in S e c t i o n I.

N o w Quillen h a s defined Hermitian metrics [9] on such determinant

line bundles and computed the associated curvature to b e precisely the
metric form above. Thus all the ingredients for applying the general the-
ory are present - the map m cutting out a transversal to the stable
orbits is g i v e n by the curvature of a connection and the preferred points,
minimising Quillens analytic torsion norm, are given by the projectively
flat unitary connections.

We can study algebraic bundles over any projective variety; in p a r -

ticular over an algebraic surface X . Now the definition of stability
requires the choice of a polarisation - the first chern class of an am-
ple line bundle L over X . This means that the degree of a bundle is
defined, in t h e normal way. We can represent this polarising class by a
K~hler form ~ , the curvature of some metric on L . Then the same the-
ory holds; we do n o t find flat connections on stable bundles but connec-
tions whose curvature is o r t h o g o n a l to the K~hler metric at each point.
The relation with metrics on cohomology is less well established but the
relevant line bundle should probably be of a form such as:

iX = X(E®KI/2x ® LI/2) ® X(E®~I/2 @ L-I/2)-I

Suppose that L has a section s cutting out a curve CcX , we can

think in t h e sense of currents of C as a degenerate form of a metric.
There is an e x a c t sequence:

0 > E ® K I/2 ~ L I/2 s > E ® KI/2 ® LI/2

.... > E! ®K > 0
X X C C

whose long exact sequence in c o h o m o l o g y gives an isomorphism iX ~ iC ;

moreover one can compute formulae for the difference in n o r m s , one defined ~
 314

relative to C and one to the m e t r i c ~ on X , compared under this iso-

morphism, with an e x p l i c i t difference term given by integrals involving
Chern-Weil polynomials in t h e c u r v a t u r e . These are u s e f u l for t h r o w i n g
problems back to t h e c u r v e f r o m the surface.

All this should probably be u n d e r s t o o d in t h e following way. Topo-

logically we have a map ~ from H2(X) to t h e c o h o m o l o g y of any family
of c o n n e c t i o n s over X . If w e w i s h to d e f i n e stable bundles t h e n we n e e d
a polarisation [~] of X which via this map ~ and Poincar6 Duality
induces a corresponding "polarising class" in t h e infinite dimensional
space of connections. We may represent the original class in v a r i o u s ex-
plicit ways; by a metric or by a line bundle or by a curve, and to e a c h
such representation on X w e get a corresponding representation in t h e
space of connections. The usual formulae for h o m o l o g i e s between the re-
presentations on X go over to c o r r e s p o n d i n g formulae on the connections
which we can use in o u r arguments involving stable bundles. But the ex-
istence of these formulae underlines that the basic correspondence bet-
ween the geometry of the base manifold and its stable bundles is g e n e -
rated by our simple construction of S e c t i o n I.

Connections over smooth 4-manifolds.

Self dual connections are solutions to a d i f f e r e n t i a l equation which

is s p e c i a l to 4 - d i m e n s i o n s . O n an o r i e n t e d Riemannian 4-manifold the 2-
forms decompose into the ±I eigenspaces of the star operator; the s a m e
is t r u e for bundle valued forms, and a connection is s e l f - d u a l if its
curvature lies in t h e +I eigenspace. If t h e m a n i f o l d is a n a l g e b r a i c
surface with the standard orientation reversed these are the connections
whose existence characterised stable bundles in the p r e v i o u s section.
(For on a K ~ h l e r surface the self dual 2-forms a r e m a d e u p of the (0,2)
and (2,0) forms and the s p a n of the K ~ h l e r form). Correspondingly these
solutions of d i f f e r e n t i a l equations in R i e m a n n i a n geometry behave rather
like objects in a l g e b r a i c geometry; in p a r t i c u l a r the solutions, u p to
equivalence by bundle automorphisms, are parametrised by finite dimensi-
onal moduli spaces rather as t h e J a c o b i a n parametrises line bundles over
a Riemann surface. Moreover these moduli spaces have applications in d i f -
ferential topology.

At present there is no g e n e r a l theory of smooth 4-manifolds. A cen-

 315

tral p r o b l e m is to u n d e r s t a n d the r e l a t i o n s h i p between homotopy and dif-

ferentiable structures and to q u a n t i f y the gap b e t w e e n them. For simply
connected 4-manifolds the h o m o t o p y type is easily understood - the sole
invariant is the i n t e r s e c t i o n form on the integral 2-dimensional homolo-
gy. L i k e w i s e the c l a s s i f i c a t i o n of t o p o l o g i c a l 4-manifolds up to homeo-
morphism has been established by F r e e d m a n [6], and is v i r t u a l l y the same
as that up to homotopy. Now while there are many integral definite forms,
and so c o r r e s p o n d i n g topological 4-manifolds, it was p r o v e d by m e t h o d s
similar to those d e s c r i b e d below [5] that none of these arise from smooth
manifolds beyond the o b v i o u s examples given by d i a g o n a l i s a b l e forms. The
interesting remaining class of forms are the even (which c o r r e s p o n d s to
spin manifolds) indefinite forms w h i c h are all of the shape:

n Es+m <0 11

For smooth m a n i f o l d s n m u s t be even by R o h l i n ' s Theorem and the simp-

lest known example, beyond S2 × S2 which has form (~ ~ , is the smooth

k--
4-manifold underlying a complex K3-surface, having 2 E%s and 3 s

in the intersection form. By taking connected sums with S2 x S2 one can

always increase m so that the p r o b l e m of r e a l i s a t i o n of these forms
is to d i s c o v e r the m i n i m a l value of m for each given n. It is hoped
that a proof that for p o s i t i v e n the value of m must be at least 3
(implying in p a r t i c u l a r that the K3 surface is s m o o t h l y indecomposable,
hence genuinely the simplest "non-obvious" smooth 4-manifold) using the
methods described below, will appear very shortly.

First a word on the formal structure of these proofs. We need some

way of d i s t i n g u i s h i n g the forms w h i c h are o b v i o u s l y realised when n is
zero from the case when n is positive. The r e l e v a n t property that emer-
ges is that a direct sum H1 @ H2 @ .... @ H k of copies of the "hyperbo-

lic" form (~
\, !h is d i s t i n g u i s h e d by the fact that the symmetric power:
U/

k+1
(H 1 @ ... @ H k)
 316

is i d e n t i c a l l y zero mod 2 . For example when k = I this says that for

any four integral elements ~i,...u,4 :

(~I "~2 ) (~3"~4) + (~I"a3) (~2"e4) + (~I"e4) ( ~ 2 " ~ 3 ) = 0 rood 2

So o u r proofs are really to e s t a b l i s h such identities when ~ are in-

l
tegral homology classes and (.) is the intersection pairing.

These identities are obtained by pairing two kinds of information

and, since we are interested in the d i f f e r e n c e s between homotopy and
differentiable structures, it is p r o b a b l y important to s t r e s s the c o n -
trast in t h e w a y s that these arise. By definition our moduli space M
of self dual connections on some bundle parametrises a family of c o n n e c -
tions a n d we h a v e seen in the first section above t h a t we c a n p r o d u c e
cohomology classes in s u c h p a r a m e t e r spaces. Alternatively we can think
of the m o d u l i space as a s u b s e t of the infinite dimensional space B of
all equivalence classes of c o n n e c t i o n s , cut out by the non-linear dif-
ferential equations giving the self duality condition. Since we r e g a r d
the homotopy type of the b a s e m a n i f o l d X4 as k n o w n we may regard the
homotopy type of this infinite dimensional parameter s p a c e of c o n n e c t i o n s
as k n o w n . For example w e h a v e o u r map; defined in a n e l e m e n t a r y way:

: H 2(X 4) > H 2(Bx)

and in f a c t t h i s generates a c o p y of t h e p o l y n o m i a l algebra on H 2 ( X 4)

within H*(Bx) .

Our moduli space M sits within this infinite dimensional space.

At present we may regard this as largely unknown and mysterious, except
for properties t h a t c a n be u n d e r s t o o d by linearisation, for example the
dimension o f the space. What w e do k n o w is t h a t the m o d u l i space carries
a fundamental class in h o m o l o g y ; or r a t h e r , as w e shall see, that it m a y
be t r u n c a t e d , t y p i c a l l y , to a m a n i f o l d with boundary ~M so we m a y assert:

< ~, [~M] > = 0 for a n y ~ in H * ( B X)

 317

To p r o d u c e a suitable cohomology class # we may use o u r m a p ~ - this

builds in the two d i m e n s i o n a l homology that we w i s h to study; likewise
we m a y p r o d u c e more subtle cohomology classes coming, from our present
point of view, from the index of the 4-dimensional Dirac operator on a
spin 4-manifold. But all this is h o m o t o p y , the smooth structure and the
difference between differentiable and topological manifolds enters by the
existence of the r e l a t i v e homology class carried by the m o d u l i space of
solutions to the d i f f e r e n t i a l equation.

Here is an e x p l i c i t example, directly relevant to the c a s e when we

study 4-manifolds with one negative eigenvalue in t h e i r intersection form.
Take the complex projective plane with its standard orientation reversed;
then we m a y study the self-dual connections via the stable holomorphic
bundles as above, and in p a r t i c u l a r if we c o n s i d e r rank 2 bundles with
c I = 0 ; c 2 = -2 then the a p p r o p r i a t e moduli space has been described
by B a r t h [3] as follows. TO the o r i g i n a l projective plane P we m a y as-
sociate the d u a l plane P* , so p o i n t s of one plane are lines in the
other. The conic curves in P* are p a r a m e t r i s e d by a copy of ~5 ; the
non-singular conics form an o p e n subset, the complement of a d i v i s o r
which is n a t u r a l l y identified with the symmetric product sym2(p) (since
a singular conic is m a d e up of two lines). According to B a r t h the m o d u l i
space of a l g e b r a i c bundles may be i d e n t i f i e d with these non-singular co-
nics, which we may obviously truncate by r e m o v i n g an o p e n neighbourhood
of sym2(p) to get a m a n i f o l d M with boundary ~M made up; loosely
speak£ng, of a c i r c l e bundle over sym2(p) with fibre L say.

We can u n d e r s t a n d our map u very easily in this example, and d o i n g

so e x p l i c i t l y will illustrate the general case. Let Z be a line in P
(so r e p r e s e n t i n g a generator of H2(P)) . T h e n it f o l l o w s essentially im-
mediately f r o m our d i s c u s s i o n of the p r e v i o u s sections and the d e s c r i p -
t i o n by B a r t h of t h e "jumping lines" of a b u n d l e that a representative
for the c o h o m o l o g y class ~[Z] is g i v e n by the h y p e r p l a n e V in ~5
consisting of c o n i c s through the p o i n t z in P*. Consider four general
lines £I,£Z,£3,£4 in P . The eight dimensional cohomology class
~(£i ) • ~(£2 ) • ~(Z 3) • ~(£4 ) is r e p r e s e n t e d by the p r o j e c t i v e line
N N D . The intersection of this with our truncated moduli
V~ I V~ 2 V~ 3 V~ 4

space M is a s u r f a c e with boundary three copies of the loop L , corres-

ponding to the three point-pairs:
 318

( ( ~ 1 N £2), (Z 3 A ~4)), ((£ 1 A £3), (% 2 N Z 4 ) ) , ( ( Z I A Z 4 ) , ( £ 2 N %3 ) )

in P . If w e p r o c e e d analogously on a n y (simply connected) 4-manifold

with one negative part of the intersection f o r m t h e n we h a v e a broadly
similar moduli space - a non-compact manifold of r e a l dimension 10 . If
we consider a cup product ~ ( ~ 1 ) ~ ( ~ 2 ) ~ ( ~ 3 ) ~ ( ~ 4) for a n y 4 surfaces
~.l t h e n w e are led in t h e same way to c o n s i d e r a set of point pairs of
the form:

((~i 0 ~j) , (~k N ~i ))

a n d the n u m b e r of such pairs, modulo 2, is j u s t the expression in t e r m s

of the intersection pairing given above. The key additional fact is t h a t
for a manifold with a spin structure (unlike {~2) the c o r r e s p o n d i n g
loop L is e s s e n t i a l in the s p a c e of c o n n e c t i o n s , detected by a m o d 2
cohomology class w I , thus we argue in t h e m a n n e r above with the c o h o -
mology class %q = w I • ~(~i ) • ~(~2 ) • ~(a3 ) • ~(~4 )

Finally I will make two g e n e r a l remarks. Following Taubes [10] the

structure of t h e s e boundaries to m o d u l i spaces can be understood reaso-
nably explicitly in t e r m s of a number of "instantons" - connections con-
centrated near a finite set of p o i n t s on the manifold. In t h e c o m p l e x
algebraic version we should probably think of t h e s e as b e i n g bundles ob-
tained from deformations of ideal sheaves, rather in the w a y that the
symmetric products of an a l g e b r a i c curve map into the J a c o b i a n . The ways
that these instantons can b e o r i e n t e d relative to e a c h o t h e r give the
structure of the "link" L in t h e m o d u l i space itself and this depends
upon the values of t h e a n t i self-dual harmonic 2-forms a t the p o i n t s .
The possibilities become rapidly more complicated as the n u m b e r m of
negative parts of t h e intersection form grows larger, and roughly spea-
king what distinguishes the cases m = 0,1,2 is t h a t the c o d i m e n s i o n
of t h e "special divisors", on which the forms are aligned in e x c e p t i o n a l
ways, is s u f f i c i e n t l y high. It w o u l d s e e m to b e p o s s i b l e that the b e h a -
viour of t h e s e harmonic forms ( w h i c h of c o u r s e globally reflect the c o -
homology, via Hodge Theory) contains differential topological informati-
on about the 4-manifold. In t h e c o m p l e x case these anti self-dual forms
 319

are made up of the K ~ h l e r symplectic form and the h o l o m o r p h i c 2-forms

and these are well known to carry a lot of i n f o r m a t i o n about the complex
structure. Rather similarly the "periods" of the h a r m o n i c forms, the
relation with the integral structure, also enter into the R i e m a n n i a n the-
ory via line b u n d l e s and Hodge Theory.

I have e m p h a s i s e d here that no global properties of these m o d u l i

spaces beyond existence are really used in these arguments, and indeed
the number of e x p l i c i t examples that are known is rather small. On the
other h a n d we have seen that we may e a s i l y construct cohomology classes
over these m o d u l i spaces and that we have obtained information by p a i r i n g
these with the r e l a t i v e homology class carried by the manifold. It seems
that the m o d u l i spaces should carry an a b s o l u t e homology class with re-
spect to c o h o m o l o g y with sufficiently small support, which can then be
p a i r e d to give integer valued invariants. Moreover these should be inde-
pendent of the R i e m a n n i a n metric on the 4-manifold in the usual way that
the h o m o l o g y class carried by the fibre of a map is a d e f o r m a t i o n invari-
ant.

Of course there are m a n y ways in w h i c h r i g i d integer v a l u e d invari-

ants can be p r o d u c e d by analytic methods - integration of forms or in-
dices of operators; but as I r e c a l l e d in the first section these can all
be u n d e r s t o o d entirely from homotopy, via the usual formulae. This is not
obviously the case for our m o d u l i spaces. For example if we take the case
described above on the p r o j e c t i v e plane then we see that
(~[£])5 [M] = I , given by the intersection of five hyperplanes. It is
not clear that this c o u l d be p r e d i c t e d from the h o m o t o p y type of ~2
alone. Again, the fact that these cohomology classes appear so n a t u r a l l y
in the c o m p l e x algebraic theory gives e x t r a m o t i v a t i o n in this direction.
 320

References:

[i] Atiyah, M.F. and Bott, R. "The Y a n g - M i l l s e q u a t i o n s over Rie-

mann surfaces". Trans. Roy. Soc. London A 308 (1982) 523-615.

[2] Atiyah, M.F. and Singer, I.M. "The index of elliptic operators
IV". A n n a l s of Math. 93 (1971) 119-138.

[3] Barth, W. "Moduli of vector bundles on the p r o j e c t i v e plane". In-

ventiones math. 42 (1977) 63-91.

[4] Donaldson, S.K. "Anti self dual Y a n g - M i l l s c o n n e c t i o n s over com-

plex a l g e b r a i c surfaces and stable vector bundles". To appear in
Proc. Lond. Math. Soc.

[5] Donaldson, S.K. "An a p p l i c a t i o n of gauge t h e o r y to four dimen-

sional topology". J o u r n a l Diff. Geom. 18 (1983) 279-315.

[6] Freedman, M.H. "The t o p o l o g y of four d i m e n s i o n a l manifolds". Jour-

nal Diff. Geom. 17 (1982) 357-453.

[7] Kirwan, F.C. "The c o h o m o l o g y of q u o t i e n t s in symplectic and al-

gebraic geometry". P r i n c e t o n U.P. to appear.

[8] Ness, L. "A s t r a t i f i c a t i o n of the null cone by the moment map".

To appear in the Amer. Journal Math.

[9] Quillen, D. Lecture at 1982 A r b e i t s t a g u n g , Bonn.

[10] Taubes, C.H. "Self-dual c o n n e c t i o n s on m a n i f o l d s with indefinite

intersection form". To appear in Journal of Diff. Geom.
 ARITHMETISCHE KOMPAKTIFIZIERUNG DES M O D U L R A U M S

DER ABELSCHEN VARIET~TEN

G. F a l t i n g s
Fachbereich Mathematik
Universit~t-Gesamthochschule Wuppertal
GauSstr. 20
5600 Wuppertal I

INHALTSVERZEICHNIS

§ I EINLEITUNG
§2 DEGENERIERENDE ABELSCHE VARIET~TEN
§3 MUMFORD'S KONSTRUKTION
§4 KONSTRUKTION VON AG
§ 5 LEVEL-N-STRUKTUREN
§ 6 MODULFORMEN UND MINIMALE KOMPAKTIFIZIERUNG
§ 7 ETALE GARBEN
§8 DIE TORELLI-ABBILDUNC
§ 9 DIE KOMPLEXE THEORIE

§ I EINLEITUNG

Die K o n s t r u k t i o n des M o d u l r a u m e s A der prinzipal polarisierten

g
abelschen Variet~ten der D i m e n s i o n g ~ber ~ ist seit langem bekannt.
Man erh~it je n a c h Geschmack einen groben Modulraum oder ein algebra-
isches Feld, nach Einf~hrung yon Level-Strukturen sogar einen feinen
Modulraum. Es sind auch Methoden der Kompaktifizierung bekannt ~ber
den komplexen Zahlen (siehe [AMRT], [N] ), d o c h fehlte bis jetzt die
Beschreibung einer solchen ~ber ~ . Dies geschieht in d i e s e r Arbeit.
Genauer gesagt, konstruieren wir tin algebraisches Feld, welches
eigentlich ~ber ~ ist, d a s A als o f f e n e T e i l m e n g e e n t h ~ i t , u n d
g
Ober dem eine universelle semiabelsche Variet~t existiert. Der Rand
wird ziemlich genau beschrieben, und man erh~it f~r L e v e l - n - S t r u k t u r e n
sogar einen algebraischen Raum.
 322

Dabei w i r d fur unsere Zwecke e i n a l g e b r a i s c h e s Feld g e g e b e n durch ein

Schema S , y o n e n d l i c h e m Typ ~ber ~ , sowie eine e n d l i c h e A b b i l d u n g
R+ S × ~ S , welche R zu e i n e m G r u p p o i d ~ber S macht, und f~r die
die P r o j e k t i o n e n v o n R auf S 6tale sind. Man e r k e n n t l e i c h t die
~quivalenz zur D e f i n i t i o n in [DM], und wet w i l l , k a n n sich nach E i n f ~ h -
r u n g yon L e v e l - n - S t r u k t u r e n darauf beschr~nken, dab R abgeschlossenes
U n t e r s c h e m a von S × ~S ist, wobei m a n d a n n bei a l g e b r a i s c h e n R ~ u m e n
landet ([A]). Bei der K o n s t r u k t i o n y o n S b e n u t z t m a n M. A r t i n ' s
Deformations-Theorie ([A]) sowie eine leichte V e r a l l g e m e i n e r u n g yon
D. M u m f o r d ' s Konstruktion degenerierender abelscher Variet~ten. Als
R
--
n i m m t man e i n f a c h die N o r m a l i s i e r u n g des yon dem M o d u l p r o b l e m Ag
gelieferten Gruppoids. DaB dies die g e w O n s c h t e n E i g e n s c h a f t e n hat, folgt
aus e i n e r B e t r a c h t u n g d e g e n e r i e r e n d e r abelscher Variet~ten, i n d e m man
zeigt, dab m a n die in M u m f o r d ' s K o n s t r u k t i o n a u f t r e t e n d e n P e r i o d e n aus
den K o e f f i z i e n t e n der e-Reihe a b l e s e n kann.

SchlieBlich sei n o c h erw~hnt, dab anders als im Fall der K u r v e n die

Kompaktifizierung nicht kanonischist, s o n d e r n y o n der Wahl e i n e r Kegel-
z e r l e g u n g der p o s i t i v semidefiniten quadratischen Formen in g Variablen
abh~ngt. Dies ist a u c h bei der k o m p l e x e n t o r o i d a l e n K o m p a k t i f i z i e r u n g
der Fall, und in der Tat l i e f e r n unsere M e t h o d e n ~ber { gerade diese
Modelle.

Der A u f b a u der A r b e i t ist wie folgt:

Z u n ~ c h s t b e t r a c h t e n wir d e g e n e r i e r e n d e a b e l s c h e V a r i e t ~ t e n und o r d n e n
ihnen q u a d r a t i s c h e F o r m e n zu. Dies w i r d zum einen benutzt, um sp~ter
die K o m p a k t h e i t zu zeigen, und m o t i v i e r t zum a n d e r e n die Wahl der Daten,
w e l c h e bei der v e r a l l g e m e i n e r t e n M u m f o r d - K o n s t r u k t i o n eingehen.

Diese folgt dann im n ~ c h s t e n Kapitel. Dabei sind alle a u f t r e t e n d e n

Schwierigkeiten im w e s e n t l i c h e n sehon yon M u m f o r d in [M4] g e l ~ s t worden.
Wit b r a u c h e n dies nur n o c h von Tori auf s e m i a b e l s c h e Variet~ten zu ver-
allgemeinern.

D a n a c h b e r e i t e t die K o n s t r u k t i o n von S und R keine g r o B e n P r o b l e m e

mehr. Ihr ist das v i e r t e K a p i t e l gewidmet, w o r a u f d a n n die A n w e n d u n g e n
folgen:
 323

Wir betrachten Level-Strukturen, Modulformen (unter a n d e r e m eine arith-

metische Behandlung der m i n i m a l e n Kompaktifizierung), 6tale Garben und
Kohomologie, die T o r e l l i - A b b i l d u n g sowie die Beziehungen zur k o m p l e x e n
Theorie. Eine weitere Anwendung wire es, den ersten Teil des Beweises
der M o r d e l l - V e r m u t u n g zu v e r e i n f a c h e n (siehe [F]), u n d es b l e i b t zu
hoffen, dab e i n e arithmetische Theorie der Siegel'schen Modulformen
in Z u k u n f t noch einiges Sch~ne hervorbringt.

Der Leser wird bemerken, da~ alle wesentlichen Grundideen y o n D. M u m f o r d

~bernommen worden sind, und dieser h~tte sicher auch n o c h die Resultate
dieser Arbeit erhalten, wenn er sich nicht anderen Interessen zugewandt
h~tte. Einer seiner Sch~ler, Ching-Li Chai, hat k~rzlich ebenfalls eine
arithmetische Kompaktifizierung des M o d u l r a u m s A beschrieben. (Siehe
g
[C]) . N a c h den mir voliegenden Informationen hat er auch Mumford's
Konstruktion verallgemeinert (entsprechend unserem § 3), benutzt aber
for die K o n s t r u k t i o n der Kompaktifizierung Theta-Funktionen und Auf-
blasungen. Dies hat den Vorteil groSer Explizitheit und den Nachteil,
dab m a n k e i n et u n i v e r s e l l e
~
semiabelsche Variet~t erh~it, und dab m a n
nur [ b e r ~ [I/2] kompaktifiziert. Auf jeden F a l l h a t er seine Resul-
tate unabh~ngig yon mir und fr~her erhalten, so dab ihm b e i Uberschnei-
dungen der Vorrang geb~hrt. Da er sehr viel mehr Sorgfalt auf die
Ausarbeitung der Details verwendet als der V e r f a s s e r dieser Arbeit,
konnten seine Ergebnisse bisher noch nicht erscheinen.

§ 2 DEGENERIERENDE ABELSCHE VARIETiTEN

a) Sei R ein normaler kompletter lokaler Ring mit maximalem Ideal

m , Restklassenk~rper k = R/m und Quotientenk~rper K . Wir nehmen
an, dab K eine Charakteristik verschieden von z w e i hat, doch ist e s
durchaus zugelassen, dab die C h a r a k t e r i s t i k von k zwei ist. s und
n seien der spezielle und der generische Punkt von Spek(R)

G sei e i n e semiabelsche Variet~t ~ber Spek(R) , d.h. G ist e i n

glattes algebraisches Gruppen-Schema ~ber R , yon endlichem Typ,
dessen Fasern zusammenh~ngend sind und Erweiterungen von abelschen
Variet~ten d u r c h Tori. Die Darstellung vereinfacht sich sehr, wenn die
spezielle Faser Gs selbst ein Torus ist. Wir empfehlen, sich die
Argumente zuerst an diesem Spezialfall klar zu m a c h e n . Der allgemeine
 324

Fall erfordert keine neuen Ideen, sondern nut e i n e Reihe von Notationen
und Definitionen. Wir setzen voraus, dab G eine abelsche Variet~t
q
ist, und dab der maximale Torus yon Gs zerfNllt. Dann ist die formale
A
Kemptettierung G eine Erweiterung einer formalen abelschen Varietfit
A
A (entsprechend einem A ~ber R ) durch einen formalen Torus
~ ~ G~ . Es gibt eine Gruppe G , mit ~ ~ , so dag G eine Erwei-
terung von A durch T = Gr ist.
m

0~T~G~A~0

Sei X = X(T) ---Z~r die C h a r a k t e r g r u p p e yon T . Dann wird ~ gegeben

durch einen Morphismus X~Pic 0(A) (R)

~F--> 0 ,

welcher jedem ~ 6 X das zugeh6rige GeradenbUndel auf A zuordnet.

Es g i b t k a n o n i s c h e Isomorphismen

0 ®0 ---0
w Z+~

b) Wir nehmen weiter an, dab auf G ein Geradenb~ndel L gegeben

ist, welches auf der generischen Faser G eine prinzipale Polarisation
n
definiert. ([MI], Ch. 6, § 2). Dann besitzt L eine kanonische kubische
Struktur, oder iquivalent dazu, definiert m,(im) ® P r l (L)-I ® P r 2 ( L ) - 1
eine Biextension von G xG dutch G (siehe [MB], I, § 2)
m

A
Das formale Geradenb~ndel L ist d a n n samt seiner kubischen Struktur
A A
Pullback eines M auf A , welches eine prinzipale Polarisation f~r
A A
A definiert. M kommt von einem M auf A , und ~ auf ~ sei das
Pullback von M . Dann ist L isomorph zu L , wobei der Isomorphismus
die k u b i s c h e Struktur respektiert. Allerdings ist d i e s e r Isomorphismus
nicht eindeutig, sondern kann mit einem Charakter X : ~ ~G
modifi-
m
ziert werden. Es sei n o c h b e m e r k t , dab e i n solcher Charakter eindeutig
bestim~t ist durch seine Einschr~nkung ~ C X auf T , und dab man auf
diese Weise genau alle ~'s erh~it, welche im K e r n der Abbildung
X ~ Pic0(A) liegen.

Bisweilen werden wir voraussetzen, dab L symmetrisch ist,d.h., dab

[-I]*(L) =-L ([-I] = -id : G~G) . Dann ist a u c h [-I]*(M) = M , doch
 325

sind diese Isomorphismen im a l l g e m e i n e n nicht miteinander vertriglich.

Wenn man sie so n o r m a l i s i e r t , dab sie auf d e r F a s e r in N u l l die Iden-
A
tit~t sind, so u n t e r s c h e i d e n s i c h die Symmetrien auf L und dem
A
Pullback yen M um einen Charakter X wie oben.

c) M definiert einen Isomorphismus A ~> PicO(A) , und somit erh~it

man eine AJ3bildung

C : X~A(R) ,

mit c(~)*(M) ~M - - M ® 0~ (Schnitte yon M ~ber einer offenen Tell-

- -- --~a -- A--p~
menge von A entsprechen Schnitte von L ~ber dem Urbild, welche
sich unter T gem~B ~ transformieren). F~r d a s folgende m~ssen wir
diese Isomorphismen geeignet normalisieren:

Definition:

Ein zul~ssiges System von Isemorphismen besteht aus Isomorphismen

i) c(~)*(0v) ~ 0w
ii) M ~ c(~)*(M) , so dab
a) Die Isomorphismen in i) sind linear in b und
b) FOr ~,w 6 X kommutiert das D i a g r a m m

c(~+~)*(M) Z c(~)*Cc(v)* (M)) Zc(~)*(M )

M
-~+w
<---~ -M~ ~0 v
<---~-C(~)*(~)
-
ec(~)*(0
--
v)

Man sieht leicht, dab zul~ssige Systeme von Isomorphismen existieren.

Je zwei unterscheiden sich dadurch, d a b m a n die Isomorphismen in ii)
mit einem q(~) ff R* multipliziert, q : X --> R* muB die E i g e n s c h a f t
haben, dab b(~,v) = q(b+w)/(q(~).q(v)) bilinear i s t in ~ und v .
Die Isomorphismen in i) w e r d e n dann mit b(~,u) multipliziert. Wir
w~hlen von nun an ein festes zul[ssiges System von Isomorphismen.

c) Da L auf G eine prinzipale Polarisation definiert, ist F(G,~)

ein R-Modul vom Rang I (in d e r T a t sogar ein divisorielles Ideal).
 326

Sei 0 L 6 F(G,L) ein nicht verschwindendes Element. Genauso ist F (A,M)

frei, mit einem Erzeugenden @M " F~r ~ 6 X erzeugt dann
@MZ : c ( g ) * S M 'F(A,M ) = F(A,s(~)*(}j))

A A A
Den formalen Schnitt (~r, 6 I" ( G , L ) kann m a n nun nach T-Eigenfunktionen
entwickeln :

@L = ~EX
Z a(~). @M ~

Dabei sind die K o e f f i z i e n t e n a(~) E R , und sie k o n v e r g i e r e n gegen Null

in der m - a d i s c h e n Topologie. Bei W e c h s e l des zul~ssigen Systems yon
Isomorphismen e r h a l t e n sie e i n e n Faktor q(~) . Damit ist klar, dab
der Inhalt des folgenden Satzes nicht von dieser Wahl abh~ngt:

Satz I:

i) a(~) % 0 f~r alle ~ 6 X

ii) b(~,~) = a(~+m)a(0)/(a(~)a(~)) 6K* ist b i l i n e a r in ~ und
iii) Falls ~ * 0 , so ist b(z,~) 6 m
iv) Wenn ein Geradenb~ndel LI A auf^ G dieselbe prinzipale Polari-
sation definiert, so dab ~i = L , so l i e f e r t ~I dieselbe
Bilinearform b(~,v)

Bemerkung:

Die A u s s a g e iii) folgt aus i) u n d ii) : 2

Diese liefern, dab a(n~)a(-n~) = a(0)2, b(n~) n . Da die linke Seite
in R liegt und fur n÷~ m-adisch gegen Null konvergiert, liegt auch
b(~,~) in R und ist k e i n e Einheit.

Beweis yon Satz I:

Man bettet R geeignet in e i n e n kompletten diskreten Bewertungsring

ein u n d darf dann annehmen, dab dim(R) = I . Es steht uns frei, den
Grundk~rper zu e r w e i t e r n , d. h., R durch die N o r m a l i s i e r u n g in e i n e r
 327

endlichen Erweiterung zu e r s e t z e n . Wir dQrfen dann zum Beispiel annehmen,

da~ a l l e 2-Teilungspunkte yon G K-rational sind. Wir behandeln
q
zun~chst den folgenden Spezialfall:

~ [-I]*L ist s y m m e t r i s c h , und die Symmetrie ist die Identit[t auf

allen 2-Teilungspunkten von T . (Wir i d e n t i f i z i e r e n
A
T[2] = T[2] ~ [ 2 ] = ~[2] ~ G [ 2 ] die 2-Teilungs-Untergruppen.)

Dann ist a u c h M symmetrisch, u n d der C h a r a k t e r X von G , welcher

A A
den Unterschied der Symmetrien zwischen L und M beschreibt, ist
gleich Eins auf T[2] Somit ist XIT = 2#o , mit einem ~o 6 X .

Dann ist [-I]'8L : ±SL , und so e r g i b t sich, dab a(2#0-~) = a(b)"

(Einheit aus R*--) . Wir--zeigen zun~chst, dab Funktion b,c : X~K
existieren mit
(*) a(~)a(~) = b(~+~)c(~-~) . Dies ist ~ q u i v a l e n t zu d e r folgenden
Behauptung:
(*)' Sei p 6 X . Dann gibt es Funktionen bp,C P auf X mit
a(#+v+p)a(~-v) = bp(~)c (v)
P

Dazu zun~chst etwas Terminologie:

Ein Schnitt f 6 r(~- × ^

G , L^2 @_{2 ) h e i s t e i n P r o d u k t , f a l l s f = g@ h
A A
mit g,h6 F ( G , L 2) . A n a l o g f[ir S c h n i t t e y o n _~2 . D a s f u n d a m e n t a l e
Beispiel eines zerlegten Schnittes ergibt sich wie folgt:
Betrachte die Isogenie

~>G : G × G - ~ G x G

(x,y) ~--> (x+y, x-y)

Bekanntlich ist #G * ( L @ L ) : _L 2 @ _L 2 Sei H~G[2] eine endliche

flache Untergruppe der Ordnung 2d (d = d i m ( G ) ) , so dab H(K) E G [ 2 ] ( K )
ein maximal isotroper Unterraum f~r d i e durch die Polarisation gegebene
symplektische Form ist. Dann kann man H [quivariant auf ~2 operieren
lassen, und descente liefert ein GeradenbNndel ~I auf GI = G/M ,
welches auf GI eine prinzipale Polarisationdefiniert. Jeder
,q
H × H-invariante Schnitt yon L2 e L2 liefert dann einen globalen
Schnitt von ~1 @ ~ I und ist damit ein Produkt. Beispiele fNr H × H-
invariante Schnitte erhalten wir wie folgt:
 328

~ G * ( e L ® G L) (x,y) : ~L(x+y) ®~_L(x-y)

ist schon invariant u n t e r der D i a g o n a l - A k t i o n von H . Dann ist

zEH(R) [L(x+y+z) ®~L(X-y+z)

H x H-invariant, u n d d a m i t ein Produkt. Man k a n n die H - A k t i o n auf L

noch m i t e i n e m Charakter c : H +{±1} twisten, und e r h ~ i t , d a B auch

z6EH(R) s (Z)SL(x+y+z) ®8_L(x-y+z)

ein P r o d u k t ist°

Wir w~hlen nun ein H , welches T[2] umfaBt. Dann ist H/T[2] = HI~A[2]
maximal isotrop, und jedes solche HI kann man auf dies® Weise erhalten.
M a n l~Bt n u n H so auf L ® 2 operieren, dab m a n den ~ b e r b l i c k [ber
die z u l ~ s s i g e n Isomorphismen nicht verliert. D a z u gehe m a n f o l g e n d e r -
m a B e n vor:

Lasse HI auf M2 operieren. Dies l i e f e r t ein ~I auf A I : A/H I

Die A b b i l d u n g

c
c I : X:A+A I

definiert dann eine Erweiterung

0+T : T I +~-1 + A 1 + 0

via Geradenb~ndeln 01,~ und M] ~I @ 01 auf A I . Beim Pullback

nach A geht M],b ~ber in _M2 J #02~: , u n d '#0I,~ in 02~ . Man e r h ~ I t
ein kommutatives Diagramm

0 --> T --> G --> A --> 0

0 -->T=TI--> ~-I - - > AI - - > 0

 329

und es ist GI ~ G I / H . Weiter sieht man sofort, dab m a n ein zul~ssiges

System yon Isomorphismen for GI,MI w[hlen kann, welches bei Pullback
vertr~glich ist m i t dem fHr G,M (z. B. ist das P u l l b a c k yon
cI(~)*(01,~) gleich _c(~)*02~ und isomorph zu 02v , u.s.w.)

Die Operation yon HI auf M2 liefert eine Operation von H auf ~2

und ~2 . Diese Operation ist a l g e b r a i s c h , d. h., kommt yon e i n e r
Operation von H auf L2 : Je zwei Operationen yon H auf L2 oder
~2 differieren um e i n e n Charakter 8 : H(R) ÷ {±I} , und mindestens
eine formale Operation ist a l g e b r a i s c h .

Wir k~nnen nun (*)' zeigen:

Sei p 6 X . Wihle ~ : H(R) ÷ {±I} mit sIT[2] : piT[2] . Dann ist der
folgende Schnitt von £2 @ £ 2 ein Produkt:

A A
Z 8(z) OL(x+y+z) @ 8L(x-y+z)
z6HIR)
A~ A
zCH(R)~ (z) a(R)a(v) 0 M ( x + y + z ) @ OM (x-y+z)
~,vOX

Bei festem b,v verschwindet die Summe ~ber H(R) (sogar schon ~ber
T[2]) , auBer wenn es ~,5 6 X gibt mit b = p+~+B , v = ~-B • Also
ergibt sich

P+~+[{(x+y+z) ® eMa-5(x-y-z)
~, BOX a(p+~+~) a(~-~) z~H(R)~(Z)UM

Die inhere Summe l~Bt sich u m s c h r e i b e n als

(c(~)* ® c(B)*) ~] (z) z* * P

-- -- zOH I (R) s (~A (@M @ b_M))

mit
¢
A : AxA+A×A

(x,y) ~--> (x+y,x-y)

Dabei ist e A 0 ® 0A der e i n z i g e Schnitt von c(p)*(_M) @ M, 6A * ( %A p ® 0A)

ein Schnitt yon (M 2 @ 0 9 ) ® (M 2 @ 0 0) , und die innereSumme wieder ein
Produkt, etwa vonder Form g ®h . Wir e r h a l t e n schlieBlich, dab
 330

~,86X a(p+e+8) a(e-~) c(~)*(g) 8c(8)*(h)

ein Produkt ist.

Da c(~)*(g) ein Schnitt ist von M 2 ® 02~+p , und c_(B)*(h) ein

Schnitt yon M2 ® 02~+ p , folgt daraus die Behauptung (*) ' , i n d e m man
obige Summe schreibt als

(o.6EX b p ( e ) ~ ( ~ ) * ( g ) ) ® (B6ZX c p(B) c ( 8 ) * ( h ) )

Damit sind (*) ' und (*) gezeigt, d. h.

a(~) a(%~) = b(#+v) c(#-v)

d) Wir zeigen zun~chst Teil i) y o n Satz I:

Wir wissen schon, da6 a(~) = 0 <=> a ( 2 #0-~) = 0 . Wit behaupten zu-
n[chst, da6 eine Untergruppe y cX existiert mit

a(~) # 0 < = > ~ 6 #0 + Y "

Ersetzt man a(~) dutch a(~+~0) , so d a r f m a n annehmen, da~ ~0 = 0 .

A
Sei Y = { ~ l a ( g ) ~ 0} . Da ~G % 0 , i s t Y n i c h t leer, und es ist
y = -y .

i) 06 Y : Sei ~ 6 Y => a(~) a(~) : b(2~) c(0)@ 0 ,

a(~) a(-~) = b(0).c(2~)% 0 , somit ist
b(0) % 0 , c(0) % 0 und a(0) 2 = b(0) • c(0) # 0

ii) ~,~6Y :> b ( 2 ~ ) % 0 , c(2~) t 0 (siehe i)

=> a ( ~ + ~ ) a(~-~) = b(2~) c(2~) % 0 :>~±~6Y

iii) Y : X :
Andernfalls g~be es ein endlich flaches Untergruppenschema NcT,
N #(0) , so d a b a l l e ~ 6Y auf N identisch den Wert I annehmen,
A A
und 8G ist ein Eigenvektor fir die Aktion von N auf L . F~r
n
Elemente x I ..... x n £ G(R) mit jZI
= xj = 0 ist j~1= x j * ( 0 G ) ein
 331

* ®n
globaler Schnitt von ® xj (L) --L , welcher ein Eigenvektor fur N
ist. G(R) ist Zariski-dicht in G , und es ist w o h l b e k a n n t , dab fur
@n
n_> 3 die o b e n definierten Schnitte von L eine projektive Einbet-
tung von G liefern. Andererseits muB diese Einbettung ~ber (G/N)
n n
f a k t o r i s i e r e n , w a s e i n W i d e r s p r u c h ist.

Damit ist zun~chst i) g e z e i g t , ii) ist n u n g a n z einfach:

Aus der Identit~t

a(~) a(v) : b(~+~) c (b-v)

folgt fur ~,~,~ 6 X :

a(l+}~+~) . a ( l + ~ ) - 1 . a ( l + ~ ) - 1 . a ( b + v ) - 1

-I
-a(l) .a(~) .a(m) a(O) : I

(Berechne a(l+~+v).a(1), a(l+~) a(l+m) , a(~) a(~) und a(~+v)-a(0)

nach obiger Identit~t), und dies ist B e h a u p t u n g ii).

e) Wir ko~men nun zu b e l i e b i g e n L's . Diese erh~it man durch Trans-

lation mit einem Element aus G(K) aus e i n e m L der b i s h e r betrachte-
ten A r t (symmetrisch, Symmetrie = I auf T[2]) . Wenn dieses Element
A
in G(R) liegt, so i n d u z i e r t es einen Automorphismus von G , und
man rechnet alles direkt nach. Im a l l g e m e i n e n k a n n m a n es j e d e n f a l l s
ausdehnen zu e i n e m Element aus G*(R) , wobei G* das N @ r o n - M o d e l l von
G bezeichne. G ist die Zusammenhangskomponente der Eins von G* ,
und ~ und 0G dehnen s i c h aus auf G* . Allerdings h a t die A u s d e h n u n g
L* yon L im a l l g e m e i n e n keine kubische Struktur mehr.

A A
Wie bisher kann man 0G, auf jeder Komponente yon G* nach T-Eigen-
funktionen entwickeln. Man erh~it dann Koeffizienten { ~ ( ~ ) , b 6 X} ,
welche vonder Komponente abh~ngen. Wir mUssen zeigen, dab sie a l l e
verschieden yon Null sind, u n d daft ~(~+~) ~(0)/(a(~)a(v)) = ~(~,m)
bilinear und unabhingig vonder Komponente ist. Wir wissen s chon, daft
nicht alle ~(~) verschwinden. Wir schlieBen mit unserem alten Trick:
W~hle HcGcG* wie vorher. Dann ist fur jeden Charakter s :H(R) + {±I ]
und jedes x 0 6 G*(R)
 332

Z s(z)(~G.(x+Xo+Y+Z)
z6H ® GG.(X-y+z )

wieder ein Produkt, u n d es e r g i b t sich, dab ~(~) a(v) = b(~+m) ~ (~-~),

mit geeigneten Funktionen b,c auf X . Dabei sind a,a die Koeffi-
zienten zu v e r s c h i e d e n e n Komponenten. Da f~r die Funktion a zur
Einskomponente schon alles n~here bekannt ist, folgt leicht die Be-
hauptung, und Satz I ist v o l l s t ~ n d i g bewiesen.

f) SchlieBlich ben~tigen wir noch ein Resultat, nach dem b(~,~) und
die P o l a r i s a t i o n auf A die P o l a r i s a t i o n
von G bestimmen. Es seien da-
A An
ZU g e g e b e n zwei G's,G I und G 2 , so dab G I = G2 und damit
AI ~ A2 ~ A . Weiter nehmen wir an, dab Geradenb~ndel ~I und ~2 auf
GI bzw. G2 existieren, welche prinzipale Polarisationen auf den ge-
nerischen Fasern liefern und a u c h dieselbe Polarisation auf A ergeben.
(d. h. ~I und ~2 unterscheiden sich um eine Translation). SchlieBlich
sollen ~I und ~2 dieselbe Bilinearform b liefern.

Satz 2:

A A
Unter diesen Umst[nden ist der formale Isomorphismus GI --~> G2 alge-
braisch, d. h., er w i r d induziert yon e i n e m Isomorphismus polarisierter
abelscher Variet~ten G1,B ~_T_> G2,n

Beweis: Es ist stets erlaubt, zu e i n e r endlichen Erweiterung von K

~berzugehen. Weiter d~rfen wir annehmen, dab ~I~2/ ~ . Dann sind
A A A
~I und ~2 isomorph zum P u l l b a c k von M auf G I ~ G2 = G . Diese
Isomorphismen respektieren die k u b i s c h e Struktur, sind a b e r nicht unbe-
dingt eindeutig. Sie liefern aber kanonische Isomorphismen

A A A A
L I ~ [-I]*_L I ---L2@ [-I ]*L 2 = Pullback

von M@ [-I]*M

Wir zeigen, dab sich bei diesem Isomorphismus die algebraischen Schnitte
F(GI,LI @ [-I]*~i ) und F(G2,~2 @ [-I]*L2 ) entsprechen. Genauer gesagt
 333

zeigen wir, dab man ein Erzeugendensystem der algebraischen Schnitte von
~I @ [ - I ] * L oder von [2 ~ [ - I ] * ~ 2 erh~it durch die Reihen

86ZX b(p+@,@) c(B)*(f)

Dabei durchliuft p6 X ein Vertretersystem f~r X/2X , und f eine

A
Basis der globalen Schnitte von M @ [-I]*M . W~hle wie bisher H~G[2]
endlich und flach, maximal isotrop, mit H ~T[2] . Dann operiert H
auf ~I @ [-I]*~i und ~2 @ [-I]*~2 ' ihnlich wie bisher. Es ist bekannt,
dab f~r j = 1,2 F(Gj,~j @ [-I]*Lj) eine Basis aus H-Eigenvektoren
besitzt, wobei jeder Charakter s: H(R) ÷ {£I} genau einmal vorkommt.
Wenn 0L 6F(Gj,~) ein nicht verschwindendes Element ist, so liegt
-j
f~r y 6 G. (R)
3

6(z)6 (x+y+z)O (-x+y+z)

zEH(R) L. L.
--3 --3

im s-Eigenraum, und man kann durch Wahl yon y in einer Zariski-dichten

Menge yon G. (R) erreichen, dab dies #0 wird. Rechnen wir nun formal:
3

^@L. (x) L aj (b)o-~-(x)M (dabei L, = Pullback von M )

--7 ~6X --3 --

w~hle p6X mit pl~[2] = sIT[2]

=>

Z
zCH(R) E ( z ) ^~L.(x+y+z)OL. (-x+y+z)
--3 --3

~,B6XL s (z)aj(0+~+6)aj(~-~) 8MP+~+B(x+y+

~ z)8 M ~-5(-x+y+z)
z 6 H(R)

~,~6X s (z) aj (p+~) aj (~) b (p, @) b (~, B) •

z6H (R)

-c(8)*(8MP+C~(x+y+z) @ QM(-x+y+z))

E b(P+8'8) c(8) . (~£X

Z
s(z)aj(p+e)aj(~)0MP+e(x+Y+Z)eM s(-x+y+z))
56x - - _

z 6 H (R)
 334

Die innere Summe ist e i n Schnitt yon M ® [-I]*M , (welcher noch von
y abh~ngt), der sich unter H (bei e i n e r geeignet zu d e f i n i e r e n d e n
Operation von H auf d i e s e m B@ndel) als E i g e n v e k t o r transformiert. Der
zugeh6rige Charakter ist u n a b h ~ n g i g von j = 1,2 , und somit sind die
inneren Summen Vielfache voneinander, f~r j = 1,2 . Es folgt, da~ der
A ~ A
formale Isomorphismus ~1 ® [ - I ] * L ] ~ ~ 2 ® [ - I ] * ~ 2 einen Isomorphismus

F(GI,q' ~1 ® [-I]*~1) Z r ( G 2 , ~ ' ~ 2 @ [-I]*~2)

induziert, und damit auch einen Isomorphismus

F(GI ' n ® [-I] n) ~ ,L2n @ [ _ I ] , L 2 n)

n~0 O'~1 *LI n~0F(G2,~] --

Dies liefert aber unmittelbar die Behauptung.

§ 3 Mumford's Konstruktion

a) In d i e s e m Kapitel liefern wir eine Art Umkehrung der v o r h e r g e h e n d e n

Betrachtungen. Dazu sei R ein exzellenter normaler Ring, I c_R ein
Ideal, so dab R komplett ist in der I-adischen Topologie.

Ferner geben w i r vor.

a) eine abelsche Variet~t A Nber R , zusammen mit einem amplen

Geradenb0ndel M auf A .

b) Eine Erweiterung G yon A durch einen Torus T---G r :

m
0 ~ T ~ A ~ 0

c) Eine Bilinearform b auf X = X(T) , mit Werten in K* (K =

Quotientenk@rper von R ):
b : X×X~K* .
Dabei sei b(p.,~) 6 I , falls ~ # 0 .

Unser Ziel ist es, eine semiabelsche Variet~t G ~ber R zu k o n s t r u -

ieren, so d a b b die Koeffizienten der zugeh~rigen 8-Reihe liefert
(Falls M eine prinzipale Po!arisation definiert). Etwas allgemeiner
ist d a s D a t u m c) folgendes:
 335

i) eine Untergruppe Y~X yon endlichem Index

ii) eine lineare Abbildung i : Y~(K)

iii) eine lineare Abbildung c : X~A(R) ,

so d a b das folgende Diagramm kommutiert:

Y --> ~(K)

A(R) --> A(K)

iv) Ein System von Isomorphismen

_c(~)*(00) ~ 0v , ~ 6 Y,'~ 6 X ,
linear in ~ und ~

v) Ein System von Isomorphismen

M
--~ = M
--
@ 0 ~ ~ -c- ( ~ ) * ( M_) , ~6 Y ,
so dab for ~ , v6 Y das folgende Diagramm kommutiert:

E(~+v)*(M) -~>E(~)*(c(~)*(_M))~c(~)*(M v)

M <~ M @ 0 <-T-- c(b)*(M) ®c(~)*(0 )

-~+~ -~ v - - --

vi) Eine Bil inear form

b : YxX~K*
symmetrisch auf Y , mit b(~,~) 6 1 , falls ~ 6 Y , b % 0 .

Die Abbildungen in ii) , iv) und vi) sollen kompatibel sein in d e m fol-
genden Sinne: b entspricht einer Abbildung y b> T(K) . Dann sei f~r
6 Y i ( y ) b ( y ) -I 6 G(R) , und die Isomorphismen in iv) werden definiert
durch Translation mit diesem Element.

Wir definieren auf den Daten noch eine ~quivalenzrelation, wie folgt:
Sei c : Y x X~R* eine Bilinearform, so dab eine Funktion q : Y ~R*
existiert mit

c(~,~) : q(b+~) q(0)/(q(~).q(~)) b,~ 6 Y

Dann erlauben wir, dab man die Isomorphismen in v) mit q(~) , die in
iv) mit c(z,v) multipliziert, und schlieBlich b durch b - c ersetzt.
Aus der Definitheit der Form b folgt sofort, dab i : Y ~G(K) eine
Injektion ist.
 336

b) Wir werden einen Quotienten G = ~/i(Y) definieren, so dab G

eine abelsche Variet~t ist, mit einer Polarisation vom Grad(M)-[X:Y]
Dies wurde von Mumford in [M4] durchgefOhrt f~r den Fall, dab ~ = T
ein Torus ist.

Yon nun an f o l g e n wir den A u s f O h r u n g e n in [M4] , § 2,3,4:

Definition: ([M4], Definition 2.]). Ein relativ komplettes Modell

besteht aus

a) Ein Schema P ~ber A , integer, lokal von e n d l i c h e m Typ

b) Eine offene Einbettung i : G~P
c) Ein invertierbares Geradenb~ndel L auf
d) Eine Aktion von ~ auf (P,L) , welche die T r a n s l a t i o n s o p e r a t i o n
von G auf G fortsetzt. Bezeichnung: (Tg,Tg*)
e) Eine Operation von Y auf (P,L8 Pullback von M )
Bezeichnung: (S ,Sg*)

Diese m~gen erf~llen:

i) ES g i b t Uc~ offen, von endlichem Typ, so dab ~ : U S (U)

~6Y
ii Sei v6~ (~) eine Bewertung auf d e m F u n k t i o n e n k ~ r p e r K(~) yon
, welche ~0 ist auf R . Sei x6 A das Zentrum von v auf
A (A ist k o m p l e t t ) , und f0r u6X sei X~ ein lokales (in x)
Erzeugendes yon 0 X~ ist e i n d e u t i g bestimmt bis auf eine
Einheit in 0A, x ,und ein E l e m e n t aus K(G) . Dann gilt:

v hat Zentrum auf P <:> V ~6X, 3~6Y mit v(x -b(~,v)) ~ 0

(es r e i c h t dabei, ~ C Y zu b e t r a c h t e n ) .

ii ) Auf ~ operieren Y und G durch Translationen. (Es folgt, dab

S und Tg auf P kommutieren, und dab S *(M) ~ c ( ~ ) * ( M ) )

iv) Sei ~6Y . Dann ist c(~)*(M)

_ _
~ M® 0
_ ~
, und somit induziert S *
einen Isomorphismus S * : S *(L) ~---> L ® 0 . Weiter ist fir
g6~ , mit Bild x6A , kanonisch x*(0_9 ~ 0_~ , via Translation
mit g . Dann kommutiert das folgende Diagramm
 337

Sb*
Tg *S ~ *(L)
_ ~ > T g *(L_ @ 0_~) = Tg* (L)
_ @ Tg* (0 - ~)

II Tg
Sg*Tg*(L)_ _L®Tg*(0-~)

S *(L)
_ > _L @ 0 - P < L@x*(0
_ -~ )

v) L @M ist ample auf P .

Bemerkung:

Die B e d i n g u n g iv) wird etwas einfacher, falls g 6 T , also x = 0 .

-I
Der Isomorphismus x*(0_9 ~> 0_~ ist M u l t i p l i k a t i o n mit ~(g)
und m a n k a n n die K o m m u t a t i v i t ~ t schreiben als

Tg*S~x* : ~ (g) ~B * T g*

Bemerkung: Man kann die Kompatibilit~t iv) umformulieren: Es reicht,

sie auf ~ c~ c~ zu v e r i f i z i e r e n . Zun~chst liefern die T* eine
n- n - g
~ q u i v a r i a n t e O p e r a t i o n von ~ auf A I ~ , und damit wird AI~ ~-linear
trivial: AI~ = 0~ . (kanonisch bis auf E i n h e i t aus R* ). E b e n s o
operiert G auf dem P u l l b a c k yon 0v auf ~ , und es ist k a n o n i s c h
Pullback (0 v) = 0~ . (Achtung: TL~ operiert kanonisch auf dem Pull-
back von 0v , doch unterscheidet sich diese Operation um den C h a r a k t e r
vonder Einschr[nkung der ~-Operation). Sei n u n b6Y, i(~) 6 ~(K)
Dann ist auf Gq S = Ti(z) , somit S (Ll~q) ~ L I G n . Der Isomorphismus
S * : S~*(L)_ ~> _L@ 0 ~ --~-~> _L wird auf ~n gegeben durch eine globale
Einheit auf GN . Dann bedeutet Bedingung iv), dab diese Funktion kon-
stant ist, also gegeben durch ein a(~) 6 K* . Analog ist
S * : a(~).c(b)*_ :

S~*(_M~=LI~r~} ~ S~*(_Mffnl---->s(~)*(_MIZnl
----> (M_~0~]Zn)~ (M~Le0~)IZn-~-> (_Me_~)IZn
Es ist a(0) : I , a(b+v) : a(~) a(v) b(b,v) (Man b e a c h t e , dab auf
 338

Gn die b e i d e n Isomorphismen P u l l b a c k ( c- ( ~ ) * ( 0 v) -~> T*l(~) (Pullback

(0v)) - ~ > P u l l b a c k (0v) sich um b(~,~) unterscheiden: Der e i n e
Isomorphismus ist P u l l b a c k des e n t s p r e c h e n d e n Isomorphismus auf A ,
der a n d e r e k o m m t yon der O p e r a t i o n yon i(~) auf P u l l b a c k (0))

Beispiel ([M4], 2.3-2.5.)

Wir k o n s t r u i e r e n unter bestimmten Voraussetzungen ein r e l a t i v komplettes

Modell. Diese sind:

Sei Z cX ein E r z e u g e n d e n s y s t e m , 0 6 Z : -Z . W~hle eine F u n k t i o n

a : Y~K* mit a(0) = I , a(p+v) = b(~,m) a(~)a(v) (Man z e i g t leicht,
dab ein solches a e x i s t i e r t ) . Es g e l t e

(*) a(~) b(p,~) 6 R f~r ~6 Y,~6 E.

Im a l l g e m e i n e n kann man kein solches a( ) finden. Dies h [ n g t d a m i t

zusammen, dab w i r a u c h e i n G e r a d e n b ~ n d e l auf der a b e l s c h e n Variet~t
konstruieren wollen, welches die P o l a r i s a t i o n induziert. In den uns
interessierenden F ~ l l e n w i r d dies aber k e i n P r o b l e m sein. Zum B e i s p i e l
kann man (*) stets e r f ~ l l e n , wenn R faktoriell ist, und man Y durch
n •Y ersetzt, n gen~gend groB.

Betrachte die b e i d e n folgenden quasikoh~renten graduierten Algebren

SI und S2 Ober A :

}1 : 0A ® n>1@ 0~ 8 n : 0 A • n ~ 1 0 ~ • 8 n

_ "O n
~2 =
0 A e n91 0 ~ _Nn-0 n = _
0 A ® n>~1 (~n
± @ 0))

vCX

~- operiert offensichtl,ich auf S I (@ b l e i b t fest) , und Y operiert

auf $2 ® R K , nach der R e g e l

Sp*(~'fm~9 n) : a ( b ) n b ( ~ , ~ ) c (~)*(%) c (~)*(f~).8

= a(p)n_c ( ~ ) * ( % ) i ( b ) * ( f j 0 n) 6 M n ® 0 v + n ®RK'8 n

(¢6~n' fv 6 0 lokale Schnitte)

 339

RI ~ $I ® R K sei der Unterring, welcher erzeugt wird von

{a(#) b(~,e) 0 b-~ .@Ib6 Y , ~ 6 Z}

und R2 ~ $2 ®R K werde erzeugt yon

{ S *(M®
_ Oc~'O ) I ~6 Y,c~6 E}

Dann ist ~ = ProJA ( R ) = ProJA(R2) ein relativ komplettes Modell:

o p e r i e r t auf (ProJA(R1), 0(I)) = (P,L) , und Y auf (~,L@M)
F~r ~ 6 ~,8 6 F ( A , M ® 0 ) ist ~.@ ein g l o b a l e r Schnitt in
F(~;L_@M) , und U 0,~,8.@ = P-V(e-@) ist affin und von e n d l i c h e m Typ
/R . Da die 8's M ® 0 2 erzeugen, ~ b e r d e c k e n die o f f e n e n Mengen
U ,d,0. @ = S (U0,~,8.@) ganz p , und U0,0,8. G = G-V(0)

W e n n man 0 eine Basis von F(A,M® 0 ) d u r c h l a u f e n l~Bt, e r h ~ i t man

auf diese Weise eine ~ b e r d e c k u n g von ~ wie in i). Die B e d i n g u n g ii)
zeigt man wie in [M4], iii) ist leicht, v) schon gezeigt, und iv)
rechnet man e i n f a c h nach.

Von nun an b e z e i c h n e (P,~ .... ) ein relativ k o m p l e t t e s Modell. Es

folgen nun eine Reihe yon Tatsachen, welche den S[tzen aus [M4], § 3,
und 4 entsprechen:

[M4], 3.1:

Sei ~ 6 Y , und f = b(~,~) 6 R . Das P u l l b a c k von 0 auf G besitzt

ein k a n o n i s c h e s E r z e u g e n d e s h E F(~,0 ) (Das direkte Bild von 0~
bei G ~ A ist die direkte Summe aller 0v,~ E X) . Dann dehnt sich hz
aus zu e i n e m r e g u l ~ r e n Schnitt des Pullbacks yon 0 auf ~f = ~ ® R R f ,
w e l c h e r dort 0~ erzeugt.

Beweis: Aus den V e r t r ~ g l i c h k e i t s b e d i n g u n g e n zu A n f a n g dieses K a p i t e ! s

folgt i(~) 6 G(Rf) . Auf Pf stimmen dann S und Ti(~) ~berein,
und die I s o m o r p h i s m e n

~ * : S ~ *(L)
-- ~ > _L ® 0 -~ und Ti(~) * : Ti(~) *(L)
-- ~ > _L
 340

liefern einen globalen Schnitt von 0 ~ber ~f , welcher das B~ndel

dort erzeugt.

Die Einschr~nkung dieses Schnittes auf ~ transformiert sich unter T

gem~8 -~ , genauso wie h . Also stimmen die beiden bis auf eine
Einheit ~berein.

[M4], 3.2:

q q

[M4], 3.3:

Jede irreduzible Komponente von ~0 : ~®R (R/I) ist eigentlich ~ber

R/I .

Beweis: Sei Z eine irreduzible Komponente von ~0' v eine Bewertung

des Funktionenk~rpers K(Z) , v ~ 0 auf R . W~hle eine Bewertung vI
von K(~) (v I ~ 0 auf R ) mit Zentrum Z , und sei v2 das Kompositum
von v und v I . FOr p 6 Y sei hp C F(~,0 ) das kanonische erzeugende
Element. Nach unserem Analogon zu [M413.1 ist for n >> 0 b ( p , p ) n . h
regul~r im g e n e r i s c h e n Punkt von Z , und verschwindet dort. Sei x 6 A
das Zentrum von v2 auf A , und ~ ein lokales Erzeugendes yon 0
nahe x . Dann ist b(~,~)n-h -(Pullback von ~ ) regul~r und gleich
P
Null im generischen Punkt von Z , hat also bei v2 Bewertung >0 .
Somit ist auch v2 (X~ b ( n ~ , ~ ) ) > 0 , im S i n n e der Bedingung ii) bei der
Definition eines relativ kompletten Modells. (h { R*.X -~ p) . Dies
gilt f~r alle p C Y . Somit hat v2 ein Zentrum auf P und v eins
auf Z . Mache weiter wie in [M4].

[M4], 3,5:

Sei U 0 = U@R R/I ' (U wie in B e d i n g u n g i ) an ~ ) U0 ist eigentlich

~ber R/I .

[M4], 3.6:

Es gibt eine endliche Teilmenge S_cY , so dab fNr p,m 6 Y , p-~ ~ S

 341

s (U0) n s (U 0) :

Beweis:

Seien Fc~ die F i x p u n k t e unter ToG . F~r jede zusammenh~ngende

Teilmenge F' ~ F operiert T auf LIF' via e i n e n Charakter v 6 X ,
und auf LIS (F ') via ~+~ . Also ist for b % 0 F' n S (F') =
Weiter wie in [M4].

[M4], 3.7:

Y operiert frei auf P0

[M4], 3.8:

P0 ist zusammenhangend

Beweis:

Genauso wie in [M4]: t0 ~ 0 definiert eine Zusammenhangskomponente

von g0 . W e n n es e i n e zweite gibt, w~hle eine diskrete Bewertung v
von K(~) mit Zentrum in d i e s e r Zusammenhangskomponente. Sei
R' {f 6 K(A) Iv(f) ~ 0} . Ersetze P durch den A b s c h l u B der generischen
Faser von ~ × A S p e k ( R ') , und w e n d e [M4] , 3.9. an.

[M4], 3.10:

FOr n ~ I existiert ein Schema P , p r o j e k t i v Ober A/I n , m i t amplem

n
Geradenb~ndel 0(I) , u n d ein 6taler surjektiver Morphismus

: ~@R (R/In) ~ Pn '

welcher (Pn,0(1)) zum Q u o t i e n t e n unter Y macht yon

(~®R (R/In) , L @ M )

Die P definieren ein formales Schema F Ober R , welches algebraisch

n ^
ist. Wir e r h a l t e n a l s o ein p r o j e k t i v e s P ~ber R , mit P = P . P
tr~gt ein amples Geradenb~ndel 0(1) . AuBerdem hat m a n e i n
 342

A
abgeschlossenes Unterschema B~P , so d a b B : Quotient von
A ~ A
(P-~Yt S (G))A~ . Sei G : P-B . Es ist G = G Q

[M4], 4.2:

G ist glatt ~ber R .

[M4], 4.3:

P ist irreduzibel

Definition ([M4], 4.4)

Eine semiabelsche Untergruppe GI ~ ~ ' 0 ~T I ~ A I ~ 0 heiBt integrabel,

falls gilt:

i) YI = i-I (GI(R)) hat denselben Rang wie der Torus TI yon ~1

ii) ~(YI) EAt (R)

iii) FHr ~ 6 X,~IT I = I und # 6 YI ist der Isomorphismus

c(~)*0 ~ 0~ auf AI die Identit~t (0 ]A I ~ 0)

Beispiele integrabler Untergruppen erh~it man etwa durch Graphen der

Multiplikation m : G x G~G oder der Inversenabbildung [-I] : ~ .

Es gilt:

Jede integrable Untergruppe GI ~ ~ definiert ein abgeschlossenes

Unterschema G I ~G , wie folgt:

a) Sei WI der Abschlu~ von GI in ~ . Dann ist W1 Y1-invariant.

b) Sei WI die I-adische Komplettierung von W I WI ist ebenfalls

Y1-invariant, und die Vereinigung W2 = ~EY/yIU S (W I) ist lokal

endlich (dies definiert W2 als reduziertes Unterschema von P ).

c) Sei W3 = W2/Yc P
A
d) Sei W3 E P definiert durch W3 = W3 .

e) GI = W3 D G .

Nur Schritt b) ist nicht trivial. Er folgt aus einer Variante yon [M4];
 343

Prop. 4.5, wobei man im B e w e i s benutzt, dab fur ~6 Y und

n > >0 b ( ~ , ~ ) n . h ( h = kanonisches Erzeugendes auf G des P u l l b a c k s
yon 0 ) auf U ein regul[rer Schnitt des Pullbacks von 0 ist,
welcher auf U0 verschwindet. Man beachte auch, dab f0r ~ 6 X mit
~IT I = I das P u l l b a c k des Ceradenb~ndels 0 auf WI kanonisch
trivial ist, wobei die T r i a l i s i e r u n g auf GI dutch h gegeben wird.

Es folgt:

[M4], 4.8, 4.9:

G ist e i n G r u p p e n s c h e m a ~ber R , und G = P ist a b e l s c h e Variet~t.

Welter k6nnen wir die Struktur der Torsions-Untergruppen yon G be-

stimmen:

Die Multiplikation auf G setzt sich fort zu e i n e r Multiplikation auf

~* : U S (~) c P und jedes ~ 6 Y definiert ein ~ E G*(R)
~6Y -- '
FUr n ~ I sei z(n) ~ * das vaserprodukt

Z (n) - - > G*

I ~n
{~} --> G*

FUr m E Y liefert Translation mit ~ einen Isomorphismus

Z (n) _T_~> Z (n)
~+n~ ' u n d die disjunkte Vereinigung II Z (n) wird
~6Y/nY
zu e i n e m Gruppenschema ~ber R .

[M4], 4.10:

Der Kern G (n) der Multiplikation mit n auf G ist isomorph zu

li z (n)
;cY/nY ~

Beweis:

Wie in [M4]
 344

[M4], 4.11:

Sei ~R ein P r i m i d e a l , YI : { ~ C Ylb(~,b) @ p } , s I 6 Spek(R) der zuge-

h~rige Punkt. Dann ist YI eine U n t e r g r u p p e yon Y' Y/YI ist t o r s i o n s -
frei, und es g i b t e x a k t e Sequenzen yon Gruppenschemata ~ b e r dem K ~ r p e r
k(s I ) :

0~ -(n) ~ G (n) ~ YI/nYI ~ 0

sI sI

0 ~ t ° r sI ~ Gt°rs
I ~ YI ~(~/2Z) ~ 0

[M4], 4,12:

G ist s e m i a b e l s c h

A A A A
C) Das ample Geradenb~ndel 0(I) auf P erf~llt 0(]) IG
Da ~ auf L[~ operiert, ist L[G kanonisch trivial, und somit er-
A A A
h~it L®M~ M eine k u b i s c h e Struktur. Wir zeigen, dab diese m i t der
A
kubischen Struktur auf 0(I) Obereinstimmt. Dies ist der Fall, wenn
A
die k u b i s c h e Struktur auf 0(I) vertr~glich ist m i t der T - O p e r a t i o n
A A
(T operiert auf L und (trivial) auf M ) . Dies ergibt sich aus den
nun folgenden 0berlegungen:

FOr J ~ {I,2,3} erh~it man d u t c h A d d i t i o n der K o o r d i n a t e n in J einen

Morphismus mj : ~3 ÷ ~ , u n d z u s a m m e n ein

: ~mj : GI = ~3 ~ 8

Der G r a p h v o n m ist e i n e integrable Untergruppe von ~11 . Man f i n d e t

dann ein relativ komplettes Modell ~1 f0r GI ' so dag m sich
fortsetzt zu ~ : ~I ~ 8

Wegen der k u b i s c h e n Struktur auf M ist m.(~M±I) trivial (die Ex-

ponenten ergeben sich als (-I) IJ[ auf dem F a k t o r J ). D a n n ist

m * ( ®j0 ( 1 )
--
-+I) ~ m- -* ( ® j (^L ® M^) ~:I) ~ m- -* ( ® Lj -+I)
 345

Darauf operieren G1 und YI : y3 . Es g i l t nun, dab die O p e r a t i o n e n

von TI = T3 und Y] kommutieren:

F~r (gl,g2,g3) E TI und (~i,~2,P,3) 6 Y I ist

Z( Z ~j(gk)) ±I
J j,k6J = I

A
Da die kubische Struktur auf 0(I) durch ihre Y-Invarianz eindeutig
bestimmt ist, folgt die Behauptung.

Weiter k6nnen wir den Grad der durch 0(I) auf Gq definierten Pola-
risation berechnen:

Er ist g l e i c h dem Rang des torsionsfreien R-Moduls

A A
F(P, 0(I)) = F(~,{@M) Y (Y-Invarianten) . Wir ersetzen zun~chst P
durch seine Normalisierung, was nichts an d i e s e n Invarianten ~ndert.
A ~ A A
Sei ~EF(P,L®M) . Da T_cC- auf L und (trivial) auf M operiert,
A
kann man ~ nach T-Eigenfunktionen entwickeln:

A
O = Z O (~)
'~EX

Aus den Kommutationsregeln zwischen Y und T folgt, dab fur

b6Y S*(8(v)) 6 L @ M--~+v
- -

Wenn also e Y-invariant ist, mu~ gelten:

e(#+,J) = s *(0(v))

Andererseits kann man fur jedes v 6 X den R a n g des v-Eigenraums in

A
F(~,L ® M ) absch~tzen :

A A A A A A
F (~,n ® M) Vc I'(G,n ® M) v = F (A,M ® % )

(Da ~I~ trivial)

Insgesamt folgt :

Rang (F(P,0(1))) _< [ X : Y ] . G r a d (_M) (Grad (M)2=Grad der yon M gelieferten

 346

dual
Abbildung A ~ A )

Wir zeigen, dab h i e r Gleichheit gilt. Dazu bezeichne H = Ker(Y) ~ T

das durch X/Y definierte multiplikative Unterschema von T . F
A
operiert auf (P,L ® M) , und diese Operation kommutiert mit Y . Wir
k6nnen dann zum H-Quotienten ~bergehen. (P/H, G/H u.s.w.) und anneh-
A A A
men, dab Y = X . Auf G ~G sind 0(%) und der Pullback von
isomorph als k u b i s c h e B~ndel. Dann ist b e k a n n t , dab d e r G r a d d e r P o l a -
risation auf G mindestens so g r o B ist w i e der von M auf A . Wir
hatten jedoch schon eine Absch~tzung in die andere Richtung.

Es f o l g t insgesamt (f~r b e l i e b i g e s Y ):

0(I) definiert eine Polarisation vom Grad [X:Y]. Grad(M)

Da in a l l e n unseren Absch~tzungen nun die Gleichheit gilt, k~nnen wit

auch eine Basis von F(G ,0(I)) angeben: Sei v 6 X,8(~J)6 F ( A , _ M ® 0 )
Dann k~nnen wir 9(v) auffassen als Schnitt von L®M ~ber G ,
welcher sich unter T gem~B m transformiert. Es folgt: Es gibt ein
r 6 R, r % 0 , so d a b sich r . 9('~) ausdehnt zu e i n e m regul~ren globalen
Schnitt aus F(P,L®M) ~

Dann existiert ein 8 6 F(P,0(1)) mit

= ~[Y S * ( r . 8 ( v ) )

Wenn ~ ein Vertretersystem f~r X/Y durchl~uft, und 9(u) eine Basis
yon F(A,M® 0 ) , so e r h ~ I t man auf diese Weise eine Basis yon

r(Gn,0(1))

Eine andere Schreibweise ist ~ b r i g e n s

A
8 = ~ y a(~) b(~,'o) c (~)* (r.9(~)) ,

mit einer Funktion a : Y~K* ,

a(0) = I, a(~+~) = a (~) a(~) b(~,v)

Dies erinnert schon an das v o r h e r i g e Kapitel. Es b l e i b t uns noch eine

 347

Kleinigkeit:

Wir haben bis jetzt ein relativ komplettes Modell nur unter der Annahme
konstruiert, dab

(*) M ist sehr ample, und a(~) b(b,~)C R f~r

6 Y, ~ 6 ~ . Wir wollen noch aufzeigen, wie man dies fallen lassen kann:

Wir setzen voraus, dab eine Funktion a( ) : X~K* existiert mit

a(0) : I , a(~+v) = a(~)a(~)b(~,v) . AuBerdem existiere ein r 6 R ,
r % 0 , so dab r.a(~) 6 R f~r alle ~ 6 X . Ein solches a( ) l~8t sich
in d e n for uns wichtigen F~llen finden, zum Beispiel wenn R regul~r
ist.

(*) ist dann immer erfNllt, wenn man for ein genOgend groBes n
ersetzt durch Mn t Y dutch n.Y t i : Y~G(K dutch i -~ : nY~(K),
-- n

und a( ) und b( ) durch ihre Einschr~nkungen auf nY bzw. (nY)×X .

Falls (*) schon erf~llt ist, l~uft dies darauf hinaus, 0(I) durch
0(n) zu e r s e t z e n .

W~hle nun nl,n 2 gen~gend groB, so d a b man zwei semiabelsche Variet~ten

GI und G2 erh~it, mit Geradenb[ndeln ~I und ~2 . Dann ist
n2) n
(GI,NI ~ (G2,NI I) , und man erh~It

G--G 1 ~G 2

n n2
mit Geradenb~ndel N, N = N I,N2 = N Wir ben~tigen noch Information
iiber d i e globalen Schnitte F(G,N) . Wir kennen schon die globalen
nI n2
Schnitte von ~I = ~ und ~2 = ~ : Man erh[it zum Beispiel eine
n
Basis von F ( G q , N I) wie folgt: Durchlaufe m ein Vertretersystem von
-- nI
X/nIY , und 8(~) eine Basis von F(A,M ® 0 ) . Dann gibt ~? f~r ein
passendes r 6 R , r # 0 , eine Basis aus Elementen 9 6 F(G,N ) mit

A
0 = r- Z a ~)b(~,~) c(~)*(8(~))
~6niY

Es liegt dann nahe, dab man eine Basis von F(G,}j) erh~it aus 8's mit
 348

A
0 = r- ~ cyZ a(p)b(~,v) c (p)*(0(v))
-- ,

~6X/Y, % (v) 6 F ( A , M ® 0 )

A
Und in d e r T a t r e c h n e t man nach, da6 diese 0 folgende Bedingung er-
f~llen: FOr ml,m 2 ganz und positiv mit mlnl-m2n 2 > 0 ist

A m nl-m2n 2 . F(G;N m2n2 m n

(e) I )_cr(G,N_ 1 1)

Daraus folgt, dab d i e s algebraische Schnitte sind.

Bemerkung:

Die Funktion a(p) h~ngt yon der Wahl des relativ kompletten Modells
ab (siehe die Bemerkung nach der D e f i n i t i o n eines solchen). Wirklieh
wichtig ist n u t die Bilinearform b mit a(~+m) = a(p)a(v)b(~,m)
Der Leser wird sich leicht ~berzeugen, dab w i t in d e r T a t g e z e i g t haben,
dab m a n b e i passender Wahl des kompletten Modells alle Funktionen a( )
erh~it, welche dieser Gleichung gen~gen, und fur d i e e i n r 6 R, r # 0
existiert mit r.a(p) 6 R. Zwei verschiedene unterscheiden sich um
einen Homomorphismus Y ~ K* .
Wir formulieren nun das Hauptergebnis dieses Kapitels. Der Einfachheit
halber betrachten wir nun prinzipale Polarisationen.

Satz 3:

Sei R exzellent normal, I-adisch komplett, Quotientenk6rper K ,

Ober R eine semiabelsche Variet~t,

0 ~ T ~ A ~ 0 ,

T ~ Gr zerfallender Torus, Charaktergruppe X ~zr . A = abelsche

m
Variet~t.

Das Geradenb~ndel M auf A definiere eine prinzipale Polarisation,

mit charakteristischer Abbildung c : X~A(R) . W~hle ein zul~ssiges
System von Isomorphismen daf~r, sowie eine symmetrische Bilinearform
 349

b : X × X~K* , so dab b(~,~) 6 I f~r ~ % 0 . Dann existiert eine semi-

abelsche Variet~t G ~ber R , so dab die g e n e r i s c h e Faser G abelsch
ist, und e i n Geradenb~ndel N
-- auf G , welches auf Gq eine prinzi-
pale Polarisation definiert.

Es gilt:

i) (G,N) --- ( , Pullback (M)) mit kubischer Struktur.

ii) Sei 8N 6 F ( G , N ) ~ N # 0 , 8 M 6 F(A;M) ein erzeugendes Element.

Dann ist
A
8N = p6XZ a(~) _c (Z)*(%M) , mit a(z) C R , a(~) # 0 , und

a(~+m) a(0) : a(p) a(~9)b(~,'~)

iil) Im g e e i g n e t zu e r k l [ r e n d e n Sinne ist

G : ~/i(X) , wobei
i : X~G(K) wie folgt zu e r k l ~ r e n ist:

Das zul[ssige System von Isomorphismen liefert eine Liftung von ~ zu

einer linearen Abbildung X~G(R) , und b liefert X~T(K) i ist
das P r o d u k t dieser beiden Abbildungen.

d) AbschlieSend benStigen wir noch einige Anmerkungen zur K o d a i r a -

Spencer Klasse:

Die e x a k t e Sequenz auf G

I I (~ 1
0 ~[~ROR 0G~Q~ ~G/R~0 : ~R/~)

liefert eine Abbildung

< : : IG, /RI ÷H I

I
W e i t e r gibt die e r s t e C h e r n - K l a s s e c(N) 6 HI (G,~G/R) einen Morphismus
I I
! G ® ~R ÷ H I ( G ' ~ R @ R 0 ~ ) , w e l c h e r im g e n e r i s c h e n P u n k t ein Isomorphismus
wird. Man kann dann K auffassen als B i l i n e a r f o r m

I
 350

Es ist b e k a n n t , dab K symmetrisch ist. AuBerdem enth~It t~ = t~ t*

--A '
den dualen Tangentialraum zu A . </t~ x t~ ist d i e K o d a i r a - S p e n c e r
Klasse zu A , und < / t ~ ×t~_ beschreibt zus~tzlich die Deformation
der Erweiterung 0 ~ T ~ A ~ 0 . Sie entspricht dem Problem, eine
translationsinvariante Differentialform aus t~ zu e i n e m T-invarianten
Schnitt aus F(G,~I~) zu l i f t e n .

Wir nehmen n u n an, da~ (~,M) und das vertr~gliche System von Isomor-
phismen schon 0ber einem Unterring R0 ~R definiert sind, und betrach-
ten,statt der absoluten Differentiale, Differentiale relativ R0
K : ~ × ~*G ~RK/R 0 verschwindet dann auf ~A x ~*G ' und definiert eine

Bilinearform t~ x ~ ~/R0 . Es ist ~ ~ X@ R , wobei p 6 X dem

Differential d log(~) = d~/~ auf T entspricht. K wird also zu

einer symmetrischen Bilinearform auf X xX .

Lemma :

K(~,v) = d log(b(~,v))

Beweis:

Zun~chst kann man die durch cI (N) vermittelte Abbildung

~ T ~ ! G ~ HI (G'0G)

noch etwas anders beschreiben: Es ist die Tangentialabbildung zur d u r c h

definierten Abbildung G~Pic0(G) . Die entsprechende Abbildung
TE~ +Pic(~) ist t r i v i a l , da T auf L ®M operiert, also
g*(L®M) ~ L®M fur g 6 T . Allerdings ist d i e s e r Isomorphismus nicht
invariant unter Y = X . Vielmehr f~hrt ein b E X einen Faktor ~(g)
ein. Man erh~it dann leicht die folgende Beschreibung von

C I (N) I_tT : t T = Hom(X,R) ~ H I (G,0 G) :

A A
Sei 1 : X~R eine Linearform. Da P : ~/X , definiert 1 eine Klasse
in HI (P,0~)-
= = HI (P,0p) . Deren Einschr~nkung auf G ist d a s Bild von
1 . Andererseits sei ~ 6 X .
 351

W~hle ~ 6~G* = ~* mit lIT : d(log(Z) . Da ~ ~ber R0 definiert

ist, liftet man 1 kanonisch zu e i n e r ~(R0)-invarianten Form in
F(~, I /R ) , u n d w e g e n ~ : ~ auch zu e i n e r solchen Form in

F(P-~/R~) @R K .

Diese Form ist n i c h t notwendig invariant unter Translation mit i(v) ,

fur m 6 X . Vielmehr ~ndert sie sich u m d l o g ( b ( z , m ) ) , da i : X÷~(R)
bis a u f Faktoren aus ~ ( R 0) ~bereinstimmt mit der durch b definierten
Abbildung X~T(K) . Das Lemma folgt nun leicht.

§ 4 KONSTRUKTION VON AG

a) Wir kommen nun zum Hauptziel unserer Bem~hungen, n~mlich der Kon-
struktion eines Ober ~ eigentlichen algebraischen Feldes, welches
A als offene Teilmenge enth[it. Wie schon in d e r E i n l e i t u n g erw~hnt,
g
ist fur uns e i n algebraisches Feld eine Art Quotient S/R , wobei
ein Schema von endlichem Typ Ober Z ist, und R~S x ~S eine endliche
Abbildung, welche R zu e i n e m Gruppoid macht ~ber S . AuBerdem wird
vorausgesetzt, dab die P r o j e k t i o n e n von R auf S @tale sind. Man
Oberzeugt sieh leicht yon der ~quivalenz dieser Definition mit der in
[DM] :

Wenn man jedem Schema T die "descente-Daten zu Hom(T,S/R) " zuordnet,

bestehend aus @talen Uberdeckungen T' ~ T und Abbildungen
T' ~ S , T' x T' ~ R , m i t g e e i g n e t e n K o m p a t i b i l i t ~ t s b e d i n g u n g e n , so e r -
T
h~it man ein Gruppoid ~ber T u n d e i n a l g e b r a i s c h e s F e l d im S i n n e y o n
[DM]

Zur K o n s t r u k t i o n von S benStigen wir einige zus[tzliche Daten:

b) Sei X = Z g , B(X) bezeichne die symmetrischen Bilinearformen

auf X (B(X) = H o m ( S 2 ( X ) , % ) ) , und B+(X)~B(X) ®Z~ den Kegel der
positiv semi-definiten Formen.

Wir fixieren eine Zerlegung

+
B (X) = U~
 352

+
mit rationalen Kegeln ~ ~B (X)~ (Die d sind d i e konvexe H~lle
endlich vieler rationaler Halbgeraden in B+(X)~ )
Diese erf~lle

i) Jede Seite eines c in d e r Zerlegung kommt ebenfalls vor.

ii) Die Inneren (im S i n n e konvexer Mengen, nicht der Topologie!)

sind d i s j u n k t .

iii) Unter GL(g,g) gibt es nur e n d l i c h viele Konjugationsklassen

von o's .
Die Zerlegung heist glatt, wenn zus~tzlich gilt:

iv) Jedes u wird aufgespannt von einer Teilmenge einer Basis yon
B(X)

Es ist b e k a n n t , daS m a n durch weiteres Unterteilen aus jeder Zerlegung

eine glatte machen kann. In u n s e r e m Fall werden glatte Zerlegungen zu
glatten Kompaktifizierungen f~hren.

Weiter setzen wir voraus:

v) FUr jedes u existiert eine lineare Abbildung I~ : X + S2(X)

so dab fur a l l e b 6 X ra(~) = b ® ~ + i0(~) 6 2.$2(X) ,
und so dab f~r fast alle ~ r (~) £ o v

Bedingung v) ist a u t o m a t i s c h , wenn die Zerlegung glatt ist (also iv) ~ v ) ,

und kann sonst durch Unterteilen realisiert werden. Sie w i r d sp~ter die
Existenz einer quadratischen Funktion au sicherstellen, welche die
Bilinearform bo (weiter unten) liefert:

I
adIz) = ~ I z ® b+id(~) 6~. [S2(X) ]

Weiter gilt f~r jedes Quotientengitter X ~X dab m a n d u r c h Schneiden

+I '
mit B ( X I) ~ B ( X ) eine Kegelzerlegung von B (XI) ~ erhilt.

Sei S der Torus mit Charaktergruppe S2(X) : B(X)* . Dann definiert

jedes a eine Torus-Einbettung S ~S , wobei S affin ist m i t A l g e -
bra ~ [ B ( X ) * N u v] . S operiert auf S , und besitzt einen einzigen abge-
d
schlossenen Orbit. Dessen S t a b i l i s a t o r ist d e r U n t e r t o r u s v o n S ,
 353

welcher zu <0> c B ( X ) geh6rt. Dabei sei <~> das von aufgespannte

Untergitter. Wenn m eine Seite von d ist, so ist S -[ eine offene
Teilmenge von S

Unsere Kompaktifizierung wird die Eigenschaft haben, dab sie lokal in

der @talen Topologie isomorph ist zu e i n e m S
0

F~r jedes o bezeichne X~X den maximalen Quotienten mit

0
S B+(Xd)~ . Dann ist <o>~B(X ) ~B(X) S(d) bezeichne den Torus
mit Charaktergruppe B(Xd)* , und S(a) c S ( d ) g
_ die Torus-Einbettung
zu O ~B(X )~ . Die universelle symmetrische Bilinearform

Xd x X d ~ B ( X d ) * : S2(X )

definiert eine symmetrische Bilinearform

b : X ×X ~K*
0 o @ d

(K = Quotientenk~rper von R , R G = affiner Ring zu

s{a) a : ~ [ B { X a ) * n o ~ ] ) ,

so d a b bd(~,~) 6 R~ f~r ~ 6 Xo , und f~r ~ ~ 0 bd(~,Z) auf dem ab-

geschlossenen S(u)-Orbit von S(o) verschwindet.
d
Sei r der Rang von X
d d

c) W~hle einen abgeschlossenen Punkt s 6 S(d) @ , w e l c h e r im a b g e -

schlossenen Orbit von S(a) liegt, und eine prinzipal polarisierte
abelsche Variet[t der Dimension g-r ~ber dem algebraischen AbschluB
d
von k(s) . Diese besitzt eine verselle Deformation, definiert ~ber der
strikten Henselisierung eines Polynomrings ~ber Z . Die Erweiterungen
r
dieser versellen Deformation durch T = { ~ werden parametrisiert
o m
durch das r a -fache Produkt des Duals der universellen abelschen Va-
riet~t. Sei R0 die strikte Henselisierung in e i n e m abgeschlossenen
Punkt dieses Produkts, und R die strikte Henselisierung von R0 ® z R o
im P u n k t e s .

Dann ist R 0 c_R , ~ber R0 existiert eine semiabelsche Variet~t

~-, 0 ~ T ~G~A~0 , so dab die Kodaira-Spencer Abbildung einen
 354

Isomorphismus definiert t~x_t A "> i~10/Z . Wenn K° den Quotienten-

k~rper von R bezeichnet, so e x i s t i e r t weiter die Bilinearform

b :X xX ~K* ,
G O (7 C

b (~,~) 6 m = maximales Ideal, falls ~ ~ 0 Sei M ein amples Geraden-

b~ndel auf A , welches dort die prinzipale Polarisation definiert, und
w~hle ein zul[ssiges System yon Isomorphismen f~r G , definiert ~ber

R 0 •

A
Satz 3 liefert dann ~ber R (:m-adische Komplettierung yon R ) eine
A A A
abelsche Variet~t G mit einem Geradenb~ndel N , so dab (G,N) --- ( ,M) ,
und dab die zugeh~rige Bilinearform (nach Satz I) gleich b ist. Nach
dem Approximationssatz yon M. Artin (siehe z. B. [A]) kann man annehmen,
dab G und N schon [ber R definiert sind.

d) Aus der Toruseinbettung S(o) ~ S ( o ) o erhZlt man eine Stratifika-

tion Spek(R) = U = U U , wobei 7 ~ber die Seiten von o l~uft.
T~O T
F~r jedes vEo ist Xv ein Quotient von X° , und S(z) ein Unter-
torus yon S(o) . W~hle ein Komplement S' , so dab S(o) = S(T) x ST
Dann ist S(T)T x S'7 eine offene Teilmenge von S(o) , somit RO ein
Unterring von R T ®Z[S'] . Weiter erh~it man eine Zerlegung b° = b ®b',
wobei b' : X x X ~[S'] als Werte nur Einheiten annimmt.
o

Sei s I 6 UT ein Punkt. Die Faser von ~ in s1,Gsl , ist Erweiterung

einer abelschen Variet~t AI durch einen Torus TI . Der Torus T1

ist in nat~rlicher Weise ein Untertorus yon T , zerf~llt also, und
U
die Charaktergruppe XI von TI ist ein Quotient von X0 . Es liegt
nahe zu vermuten, dab XI = X (also TI = T ) . Wir werden gleich sehen,
dab dies in der Tat der Fall ist. Auf jeden Fall schlieBt man schon aus
unserer Variante von [M4], 4.11 (siehe § 3), dab XI den gleichen Rang
hat wie X
T

Es reicht dann, die vermutete Gleichheit XI : X fur den Fall zu zei-

gen, dab sI einer der generischen Punkte von U7 ist. Wegen UT
normal kann man dann ~hnlich wie in IF], § 2, Lemma I, TI als Unter-
torus in GI~ T einbetten. Aufgrund der bekannten Starrheitseigen-
schaften von Tori kann man diese Einbettung auf die formale Komplettie-
 355

rung von U l~ngst U fortsetzen.

T

Sei IcR das Ideal, welches den AbschluB von U Y

definiert. Dann
kann man formal l~ngs I
und l~ngs m komplettieren. Wir unter-
I 2
scheiden dies durch Indizes: und . Dann gibt es exakte Sequenzen

^I ^1
O _ , T I _,~I _,GI -*0 ,

0 * ;2 0 ,
und

0~ (9 2 / 912) ~ ~1 2 ~ AA~ 0

Die formale Komplettierung von N ist jeweils Pullback eines Geraden-

bOndels auf den letzten Termen dieser Sequenzen.

Sei 8 6 F (G,N) , @ , 0 . Dann k6nnen wir @ formal entwickeln:

i) huf ~I :

~1 = E
# E X I @I (#)

A
Dabei sind die 8 1 (~) ~-Eigenfunktionen unter T 1 , und konvergieren
gegen Null in der I-adischen Topologie

A
ii) Auf G2 :

~.2 = Z a(N) c ( ~ ) * ( G M)
~EX -- "
O

wobei a(#+v) a(0) = a(#) a(~) b(~,v)

Durch Vergleich folgt:

A
012(v) : p,+vZ a(#) _c(P')*(0 M)

fur ~ 6 XI (man s u m m i e r t ~ber das Urbild von ~ )

Angenommen nun, es sei X I . XT . Dann gibt es ~ 6 Xo , welches Bild

0 in X I , aber Bild 0 2in X hat. Damit konvergiert einerseits
a(n~)a(-n~) = a(0)2.b(~,~) n I-adisch gegen Null, so d a b b(~,~) 6 I .
 356

Andererseits ist b(~,~) = b' (b,#) = Einheit mod I . Also ist tats~ch-
lich XI = X

Da b = b T .b' , kann man a( ) analog zerlegen:

a(#) : a I (#) a' (#) ,

wobei:

i) a I (p) h~ngt nur vom Bild yon # in XI = XT ab, und

a I (~+v) a I (0) = a I (~}a I ( v } b T ( # , v )

ii) a ' (0) = 1, a' (p+v) = a' (#) a ' (~) b' (p,v)

Damit folgt:

A2
e 1 (v) = a 1 (v) Z
I/+~0
a ' (;z) --
c (#)*(6 M) , v E XT

Man Oberlegt sich ~brigens leicht, dab man erreichen kann, dab
A A
a' (g) E K regulfir ist auf dem P u l l b a c k yon U
c Spek(R) in Spek(R)
o T--
(Dies gilt schon, falls # C Kern(X ~ X) , und sonst modifiziere man
A
a3 und a' mit einer linearen Abbildung X ~ K* . )

Wir wenden dies wie folgt an:

Sei wieder sI 6 UT . Durch das Degenerieren Yon (G,N) in sI erh~it

man nach Satz I eine symmetrische Bilinearform bI auf XT x XT , mit
Werten im Quotientenk6rper der Komplettierung des lokalen Ringes von
U in sI . Wir behaupten, dab bl/b 7 als Werte Einheiten (in sI )
hat:
^
Dazu darf man zun~chst R durch R ersetzen. Sei RI dann der lokale
A
Ring zu sI , mit Komplettierung R1 Die Reihe

Z e 1 (#)
Y
A
induziert dann die entsprechende Zerlegung ~ber RI . Da

A2
e t(v) = a~,(v) p÷vZ a ' ( ~ . ) _c ( ~ ) * ( e M) ,
 357

A
und a'(~) = Einheit in R I , folgt dab 01(v) = a1(~) (regul~rer
Schnitt, % 0 in s I ) . Es folgt, dab bI und die Bilinearform zu
aI (dies ist b7 ) sich nur um E i n h e i t e n unterscheiden.

Schlieglich liefert GI auf der formalen Komplettierung yon U l~ngs

UT eine Kodaira-Spencer Abbildung

x t* ÷ ~/R T ® A
~*G -- G I R RI

Diese induziert einen Isomorphismus.

A $2 A
Bild von (~*G ® t* ) in (t* G) ®RRI
-- G I --
I A
~> ~R/R ® RRI

Beweis:

Sie induziert einen Isomorphismus

A I A
Bild(t* G ®~*A ) ~> ~ R / R o ® RRI

Die induzierte Abbildung

I A
S2 (~*(T/T]) ) ÷ ~ R / R O ® RRI

ist gegeben durch

~Sv ~--> d(log(bo(~,~3)) = d log(b' (~,v)) ,

fNr ~, v6 X ( T / T I) = Kern(X ° + X )

I
Die d log(b' (~,v)) bilden aber eine Basis von 9Z[s,]/Z oder auch
I A
~R / R T ®RR1 , und es folgt alles.

e) Bis jetzt war R einfach die strikte Henselisierung von R 0 @ ~R ° .

Es ist dann induktiver Limes von endlich erzeugten ~-Algebren, und es
ist "alles" schon ~ber einer solchen definiert. Wir erhalten dann ein
Paar von endlich erzeugten Z-Algebren, welches wir wieder R 0 ~R nennen,
so dab R ~tale ist ~ber R 0 ®R O , so dab (G,M) definiert sind ~ber
 358

R0 , und so dab (G,~) ~ber R existieren. AuBerdem erh~it man durch

strikte Lokalisierung von R in e i n e m abgeschlossenen Punkt
s 6 U cU = Spek(R) die bisherige Situation.

e) Lemma:

Indem man U gegebenenfalls durch eine kleinere 6tale Umgebung von s

ersetzt, kann man folgendes erreichen:

i) Sei Spek(R) = U = U U die Stratifikation, und OT die formale

T~@ T
Komplettierung von U lfings U . Dann e x i s t i e r t auf ~T eine
exakte Sequenz

T T T
und ~T ist Pullback eines Geradenb~ndels ~T auf ~T
-- - 7 7

ii) Sei 6 6 F(G,N),0 ~ 0 . Auf ^

UT entwickelt man e nach
AT
T -Eigenfunktionen:

~6X
T
Welter zerlege man b = b -b' Dann existiert ein
T
A7
ao 6R7 = F(U ,00,[) , ao % 0 ,
At ^
SO dab 8y(V)87(-v) = b (v,v).a "f , wobei der Schnitt
T ) auf U in keinem Punkt identisch auf der Faser
-- T

verschwindet.

iii) Die Kodaira-Spencer Klasse von ~< induziert auf einen

T T
Isomorphismus

Bild yon ~*G @~*G T in S 2 ( --

g ' G ) ® 0 ^UT
I
~> ~R/R @ 0A
T U
T

Beweis:

Punkt i) ist klar. F~r ii) w~hlt man a0 so, dab die Bedingung f~r
= 0 erf~llt ist. Dann gilt sie auch in allen Punkten s] 6 U ,
welche s in ihrem AbschluB enthalten (nach den vorherigen ~berlegun-
gen ~ber Bilinearformen). AuSerdem reicht es, sie f~r eine gewisse
endliche Anzahl von ~'s zu v e r i f i z i e r e n . (Wegen des Zusammenhangs
 359

mit Bilinearformen).

Damit l~Bt sich auch ii) erledigen, iii) geht genauso.

Korollar:

Sei sI E U ein abgeschlossener Punkt, sI 6 UT . Dann ist d i e K o m p -

lettierung der strikten Henselisierung von U in s isomorph zur
Komplettierung eines der v o r h e r konstruierten Ringe R (fOr % statt
), w o b e i s i c h die G's und N's entsprechen.

Beweis:

A
Sei RI der lokale Ring in s1,R I seine Komplettierung. Betrachte die
A
Zerlegung in G in s 1 :

A A A A A
0~T I ~G~A~0 , T I : TT .

A
Wenn R2 eine verselle Deformation ist v o n A (mit P o l a r i s a t i o n ) und
der Erweiterung durch T I , so ist d i e A b b i l d u n g R2 ~ R7 ~ RI 6tale,
nach Teil iii) des Lemmas. Weiter gilt fur die zur D e g e n e r a t i o n geh~-
A
rige Bilinearform bI : X ×X ~K~ , dab bl/b T Werte in d e n E i n -
A T • A
heiten RI* annimmt. Wenn man die Abbildung R T ~R I mit einem geeig-
A
neten Element aus S(T) (RI) twister, darf man annehmen, dab b I = bT ,
und alles hat seine Ordnung.

f) Jetzt k~nnen wir den Hauptsatz zeigen:

Satz 4:

Es e x i s t i e r t ein Schema S , yon endlichem Typ ~ber ~ , glatt und mit

geometrisch normalen Fasern, und ein Gruppoid R~S x ~S , endlich,
mit 6talen Projektionen auf S , so d a b gilt:

i) Uber S existiert eine semiabelsche Variet~t G ,abelsch in d e n

generischen Punkten, und ein Geradenb0ndel N , welches in d e n ge-
nerischen Punkten eine prinzipale Polarisation definiert.

ii) Seien GI und G2 die beiden semiabelschen Variet~ten der Dimen-

sion g , die m a n dutch Pullback mit den Projektionen auf
 360

erh~it. Dann gibt es einen Isomorphismus G I ~ G2 , welcher in

den generischen Punkten die Polarisation respektiert und mit
der Gruppoid-Struktur vertr~glich ist.

iii) Sei S 0 ~S die dichte offene Teilmenge, dber der G abelsch

-I -1
ist, R0 : Prl (S0) : Pr2 (S0} . D a n n ist
R0 -~-~> I s o m ( G 1 , G 2 ; P o ] a r i s a t i o n ) . Das d u r c h (S0,R
0)_ _ definier-
te algebraische Feld ist A
g

iv) Das durch (S,R) definierte algebraische Feld ist eigentlich

@ber ~ .

v) S und R besitzen Stratifikationen S : U S , R = U R , para-

metrisiert durch die Konjugationsklassen der o _c B + ( X ) ~ unter
GL(g,~) . F@r einen abgeschlossenen Punkt s 6 --0
S ist die strikte
Henselisierung von S in s isomorph zur strikten Henselisie-
rung yon s in e i n e m Punkt des
abgeschlossenen S-Orbits. Dieser
o . sO
Isomorphismus erh~it die Stratifikationen ist das Stratum
zu ~ : (0)

Beweis:

Wir w~hlen S als endliche Vereinigung von den vorher konstruierten

U's . Jedes solche U geh~rte zu einem c , und es gab eine @tale
Abbildung U+S(0) x (Modulfeld der semiabelschen 0 +Grm*G~A÷0 ).
Wir fordern, dad f(]r je e~n u in e i n e r GL(g,~)-Konjugationsk]asse
die zugeh~rigen U's die Menge (abgeschlossener S(0)-Orbit) x (Modul-
feld) ~berdecken. Da das Modulfeld quasikompakt ist, und da es nur
endlich viele Konjugationsklassen yon o's gibt, reichen dazu endlich
viele U's . Wir haben nun S , und erhalten zwei semiabelsche Varie-
t~ten GI und G2 durch Pullback auf S x~S . Uber S0 x S 0 ist
R0 : isom S (GI,G2; Polarisation) endlich. Sei R die Normalisierung
von _R ° --o
~ber S xZ S Dann ist ~ normal und endlich ~ber S x S .
AuBerdem existiert QDer R ein Isomorphismus GI z G2 (siehe [F], § 2,
Lemma I), und die Aussagen i) , ii) , iii) und v) folqen, wenn wir zeigen,
dab di~ Projektionen yon R auf S @tale sind und die Stratifikati-
onen respektieren. Dazu sei s 6 R ein abgeschlossener Punkt, mit Pro-
jeKtionen sI und s2 auf S . Sei R die Koraplettierung der strikten
A
Henselisierung von R in s , und entsprechend fQr R I ® R2 . Dann ist
 361

A
R endlich ~ber R I ®R 2 (das k o m p l e t t e Tensorprodukt ist zu n e h m e n ~ber
der Komplettierung der strikten Hense]ierung yon ~ in e i n e m Primideal
p~, p % 0 ), n o r m a l , enthilt RI und R2 , und es ist
GI ® R R = G 2 @R R = G12 . Zu d e n degenerierenden polarisierten abelschen
I 2
Variet~ten geh6ren Gitter X I = X 2 : X12 und symmetrische Bilinear-
formen.

b I : X I x X I ~ KI*

b2 : X 2 × X 2 ~ K2*

b12: X12 x X12 ~ K *

(K],K2~ sind die Quotientenk6rper). Es ist b12 = b ] - b I' = b 2- b 2'

wobei b I ' , b 2 ' : X12 × X12 ~ R * als Werte Einheiten annehmen.

Welter hat man Polyeder ~1 ~ B + ( X ] )~ und 0 2~B+(X2 )~ . Es ist

°I : ~2 :
Angenommen o12 = oIRO 2 ist eine echte Seite von 01 (oder analog
von d 2 ). D a n n gibt es endlich viele Elemente ~ j , v j 6 X I , so d a b
fir jede Bilinearform bE0]2 E b(~j,vj) = 0 , abet dab diese Summe
positiv wird fur jedes b in oi-d. 2 , und negativ fur b £ o2-o12
Es sind dann [ b 1 ( ~ j , ~ j) und Hb2-1 (Zj,v j) Elemente aus mI bzw.
R2 , wobei das erste Produkt keine Einheit ist. Ihr Produkt ist aber
eine Einheit in R , und das geht nicht.

Sei also 01 = o2 = o , und damit X I : X 2 = X12 = X

Es sei w i e d e r R0 die Basis einer versellen Deformation von Gs Ober

k(s) (d.h., Deformation des abelschen Tells und der Erweiterung durch
den Torus). Dann sind RI und R2 Komplettierungen yon strikten Hen-
selisierungen gon R 0 @R O t und ~I und ~2 sind Pullback eines
~ber R0 . Man w~hle ein zul[ssiges System von Isomorphismen fur
~ber R 0 , und erh[it dann einen Isomorphismus RI m R2 , bei dem sich
b] und b2 entsprechen. Nach Satz 2 ist dann auch G I ~ G2 , und der
Isomorphismus ~ber R entsteht einfach durch Basiserweiterung. Da R
Ober die Normalisierung yon Isom S (GI,G2; Polarisation) definiert
----O
wurde, ist R = R I : R2 . Die Projektionen sind also e£ale in s 6R .
Es b l e i b t die Aussage iv) . Dazu benutzt man ein Bewertungs-Kriterium:
Sie V ein kompletter diskreter Bewertungsring, mit algebraisch abge-
schlossenem Restklassenk~rper k und Quotientenk6rper K , und
 362

%1 : Spek(K)+

ein Morphismus der den einzigen Punkt yon Spek(K) in e i n e n der g e n e -

rischen Punkte von S abbildet. Dann gibt es e i n e endliche Erweiterung
K' m K , mit Normalisierung V' yon V , und

~2 : Spek(V') ÷S ,

so dab m a n ~i × ~2 : Spek(K') ~S x ~S zu e i n e m K'-wertigen Punkt yon

R liften kann.
Es ist a l s o folgendes zu z e i g e n :
Das Pullback ~I*(G) yon G unter ~I ist e i n e prinzipal polarisierte
abelsche Varietat. Wit brauchen #2 ' so d a b ~2*(G)I K' isomorph zu
~I*(G) ist.

Dazu w~hle man K' so g r o B , dab ¢I *(G) semistabile Reduktion hat

~ber K' • und ersetzt V dutch V' t K durch K'

Sei v : K* ÷ ~ die Bewertung. Das N ~ r o n - M o d e l l yon ~I*(G) definiert

nach Satz I eine symmetrische Bilinearform b : ~ x XV~K x auf e i n e m
Gitter XV , und v o b ist p o s i t i v definit. Es g i b t also ein ~ in
der Kegel-Zerlegung, so dab v o b im I n n e r e n yon ~ liegt, und
XV ~ X ° (~,X) ist e i n d e u t i g bestimmt bis auf K o n j u g a t i o n mit GL(g,Z).

Sei w i e d e r R0 eine verselle Deformation der speziellen Faser des

N~ron-Modells. Wit erhalten eine Abbildung R 0 +V , so dab die u n i v e r -
selle Uberlagerung des N ~ r o n - M o d e l l s Ober R0 definiert ist. Wenn man
dann ~ber R0 ein zul~ssiges System von Isomorphismen w~hlt, definiert
b einen Morphismus R 0 ® R° ÷V bzw. Spek(V) Spek(R0) x~ S(o)o , wo-
bei der abgeschlossene Punkt yon Spek(V) in d e n abgeschlossenen Orbit
von S(o) abgebildet wird. Nach Konstruktion von S kann man diese
o
Abbildung liften in e i n e s der U's .

Dies liefert ~2 " so dab d a s N ~ r o n - M o d e l l von ¢1"(G) und ~2*(G)

dieselbe formale Komp!ettierung in V haben und dieselbe symmetrische
Bilinearform definieren. Nach Satz 2 sind sie isomorph.

g) Damit ist d e r B e w e i s von Satz 4 beendet. Wir notieren hier nur

noch das Korollar, dab die geometrischen Fasern von A ~ber Z irre-
g
duzibel sind: Dies folgt aus der analytischen Theorie in C h a r a k t e r i s t i k
0 , und der Rest ist genauso wie in [DM].
 363

§ 5 LEVEL-N-STRUKTUREN

a) Alle unsere Oberlegungen lassen sich auch mit Level-Strukturen

durchf~hren. Wir w ~ h l e n eine nat~rliche Zahl n . Da sich b e k a n n t l i c h
Level-n-Strukturen schlecht mit Charakteristiken vertragen, welche n

teilen, arbeiten wir ~ber ~[I/n,e2~ . Uber diesem Grundring hat

L J

man eine kanonische symplektische Form <,> auf (~/nZ) 2g , mit

Werten in Pn = n.te Einheitswurzeln. Eine Level-n-Struktur auf
einer prinzipal polarisierten abelschen Variet[t der D i m e n s i o n g ist
ein Isomorphismus A (n) ~ > (~/n~) 2g (A (n) = n - T e i l u n g s p u n k t e ) , welcher
die symplektische Struktur erh~it. Die v e r s c h i e d e n e n Level-n-Strukturen
auf A sind k o n j u g i e r t unter Sp(2g,Z/nZ) . Es g i b t ein algebraisches

Feld Ag,n ~ber Bit/n, e 2~i/n]j , welches die abelschen Variet[ten

mit Level-n-Struktur klassifiziert. Fir n ~ 3 ist A sogar ein

g,n
algebraischer Raum.

Sei A die N o r m a l i s i e r u n g yon A in A . A wird gegeben

g,n g g,n g,n
durch ein Paar (Sn,R
n)_ _ - . Dabei ist S die N o r m a l i s i e r u n g v o n S
--n --n
in der d u r e h Hinzuf~gen von Level-n-Strukturen ~ber S0 definierten
0
Uberlagerung, und R wieder die N o r m a l i s i e r u n g yon R
--n --n

b) Wir wollen ein lokales Modell finden f~r die ~berlagerung S ÷ S .

--n --
Dazu betrachtet man wieder den T o r u s S mit Charaktergruppe B(X)* ,
und die T o r u s - E i n b e t t u n g e n SmS Die M u l t i p l i k a t i o n mit n auf
S induziert eine verzweigte Uberlagerung n : S d --> S

Es g i l t nun:

Satz 5:

i) Die ~berlagerung S +S ist lokal in der 6 t a l e n T o p o l o g i e iso-

--n --
morph zu n : S o --> S O . I n s b e s o n d e r e e r h ~ i t a u c h --nS eine
Stratifizierung, hat geometrisch normale Fasern, und fihnliches •

ii) F~r n >- 3 ist --nR÷_nS ×Z[rl/n,e2~i/n 71|--nS eine abgeschlossene

J
Einbettung, und A ein algebraischer Raum.
g,n
 364

Beweis:

i) Sei R die K o m p l e t t i e r u n g der strikten Henselisierung von S in

einem abgeschlossenen Punkt, G/R die universelle semiabelsche Variet~t,
G, 0 + T +~+A÷ 0 , X und b
: X ×X ÷K* w i e ~ b l i c h . Die B e h a u p t u n g
C O G ~ C
i) l a u f t d a r a u f h i n a u s zu z e i g e n , da8 d i e n - T e i l u n g s p u n k t e v o n G den

K~rper K(n/b) erzeugen. Aus unserem Analogon zu [M4], 4 . 1 1 ( s i e h e § 3)

ergibt sich eine exakte Sequenz yon Gruppenschemata ~ber K

0 +~(n) ÷ G(n) + X o / n . X ÷ 0
@

Dabei ist ~(n) ~tale vom Rang n 2g-r@ , u n d die Faser ~ber der Klasse
modulo n.X von ~ C X ist isomorph zu
o

Z (n) (~) = { g 6 G ( K ) , n.g : i(~)] " Wenn b d : X + T @ (K) die zu b o (,)

geh~rige Injektion ist, so ist i(~) f ~(R) "b0(b) . Da G(R) n-divisibel
ist, ist der v o n den Koordinaten der n-Teilungspunkte erzeugte KSrper
gleich K(n/bd(x)) , u n d es folgt die Behauptung.

ii) Wie bisher sind die Projektionen yon R auf S 6tale, und
--n --n
somit ist --nR~ Sn x ~ [ ] / n , e 2 ~ i / n ] ~n unverzweigt.

Es r e i c h t dann zu zeigen, dab ~ b e r zwei geometrischen Punkten sI und

s2 von --nS h ~ c h s t e n s ein geometrischer Punkt von --nR l i e g e n kann.
Es m~gen wieder RI,R 2 und R die Komplettierungen der strikten
Henselisierungen von S bzw. R in s s2 und s bezeichnen.
--n --n I'
Es ist RI ~ R und R 2 ~ R , und der induzierte Isomorphismus R I = R2
ist u n a b h ~ n g i g v o n d e r Wahl von s (und R ), da s c h o n b e k a n n t ist,
0 0 0
dab R s i c h a b g e s c h l o s s e n e i n b e t t e t in S × S
--n --n S p e k ( ~ [ I / n , e 2 ~ i / n ] ) --n .

Es folgt, dab es n u r e i n s in der F a s e r ~ber (Sl,S 2) geben kann.

Korollar:

Die geometrischen Fasern von A

g,n
,[lJno2 iJn] sind
irreduzibel.
 365

§ 6 MODULFORMEN UND MINIMALE KOMPAKTIFIZIERUNG

a) Auf A e r h ~ I t m a n in n a t ~ r l i c h e r W e i s e e i n e R e i h e v o n V e k t o r -
g
b~ndeln. D a b e i ist e i n V e k t o r b ~ n d e l auf A gegeben durch ein Vektor-
g
b~ndel auf S , d e s s e n b e i d e P u l l b a c k s zu R i s o m o r p h sind (unter
ErfOllung geeigneter Bedingungen).

I.) ~G' ~*G ' d.h. Tangential und KotangentialbOndel des u n i v e r s e l l e n

G's und ihre Tensorpotenzen u.s.w. Sei ~ G = ig t*
t -- G

I
2.) Im a l l g e m e i n e n ist 9S/~ nicht lokal frei. Als Ersatz dient
besser das B~ndel @ , definiert wie folgt: F~r jede Torus-Einbettung
S o--
S o sei 0So das Unterb~ndel des direkten Bildes von ~ SI/ ~ , welches
von den d~/b erzeugt wird (b = C h a r a k t e r von S ). Da S lokal in
der 6talen Topologie isomorph ist zu e i n e m S O , erh~it man 0 auf
Wenn die Kegelzerlegung glatt ist, so ist @ die Garbe der D i f f e r e n -
tialformen mit logarithmischen P o l e n in ~ . D i e K o d a i r a - S p e n c e r K l a s s e
liefert einen Isomorphismus K : S 2 (~*G) ---~> @ (Dies f o l g t aus d e r
ziemlich expliziten Bestimmung von < ).

3. Lokal existiert ein Geradenb~ndel N auf G , welches Nber SO

die prinzipale Polarisation definiert. Wenn e : S --> G den Null-
schnitt bezeichnet, kann man annehmen, dab e*(N) ~ ~ trivial ist.

N® [-I]*N = H ist d a n n e i n w o h l d e f i n i e r t e s GeradenbNndel auf G , mit

e*(H) ~ 0s . Das direkte Bild p,(H) auf S (p : G + S die P r o j e k t i o n )
ist e i n e - k o h [ r e n t e reflexive Garbe vom Rang 2g . Ihre lokale Struktur
ist r e c h t interessant, siehe z. B [MB] . Man kann 0brigens ihre lokale
Struktur mit Hilfe unserer Uberlegungen beim Beweis yon Satz 2 aufhellen.

Auf jeden Fall ist p,(H) lokal frei auf S0 . Wenn wir Level-2-Struk-
turen einfNhren, also zu S~ ~bergehen, so e r g i b t sich eine irreduzible
Darstellung der @-Gruppe auf p,(I~) , und p,(H) ~ K 2 g , f~r e i n Ge-
radenbNndel K auf S~ oder auf Ag,2_i Aus dem Satz von miemann-Roch
folgt, dab in P i C ( A g , 2 ) @~ K und ~ dasselbe Bild haben. Eine
-I
Potenz von K stimmt somit mit einer Potenz yon ~ ~berein.

b) Eine Modulform vom Gewicht k zur G r u p p e S (2g,Z) ist e i n g l o -

k P
baler Schnitt yon ~ ~ber A . Man definiert entsprechend Modul-
g
 366

formen ~ber {' ~ / p Z o d e r a l l g e m e i n e r ~ b e r e i n e m b e l i e b i g e n Ring. Man

k
kann auch Schnitte von ~ ~ber A b e t r a c h t e n . D i e s e lassen sich wie
g
folgt interpretieren:

Jeder abelschen Variet[t A vonder Dimension g w i r d ein

f A 6 F(A, ( ~ ) k) zugeordnet. Die verschiedenen fA's entsprechen sich
k
bei B a s i s w e c h s e l oder Automorphismen von A . Jeder Schnitt von
~ber A besitzt dann Fourierentwicklungen:
g

Sei w i e d e r X = Z g, o ~ B + ( X ) ~ ein k o n v e x e r rationaler Polyeder stabil

unter Homothetien. Dabei wird nicht vorausgesetzt, dab g in der vor-
her g e w ~ h l t e n Kegelzerlegung auftaucht. Wir k S n n e n jedoch wieder
X a , r a : R a n g ( X u) , S,S(~) und S(g) a definieren.

Sei A eine abelsche Variet~t der Dimension g-r . A sei p r i n z i p a l

polarisiert, mit Dual A v . Uber (AV) ro existiert d a n n die u n i v e r s e l l e
Erweiterung von A durch T : ~ r 0 :
a m

0+T ÷~ ~A÷0

A
Sei R die K o m p l e t t i e r u n g des R i n g e s R~ zu S(~)u in der I - a d i s c h e n
Topologie, wobei das Ideal I den a b g e s c h l o s s e n e n S(0)-Orbit in S(U)
beschreibt. Dann l i e f e r t die M u m f o r d - K o n s t r u k t i o n eine semiabelsche
Variet~t G fiber Spek(Rj) x (AV) ~ , (zun~chst ~ b e r dem f o r m a l e n
Schema, aber man m a c h t a l l e s algebraisch, da die K o n s t r u k t i o n gleich
eine Kompaktifizierung dieser Variet[t liefert), welche gute R e d u k t i o n
hat auf d e m U r b i l d des o f f e n e n S(o)-0rbits S(J) ~ S ( o )
o

Wenn ~1,...,~r eine Basis yon X ist, so ist

o

Ag~*G ~ R~(d(l°g(~1)A---Ad~ l°g(~ r )) ® A g - r ~ ( ! * A )

und fG hat eine E n t w i c k l u n g

fG = 71CB
Z (X u ) ,7.fx. (d l o g ( ~ 1 ) A . . . A d log(~ r a ))k

mit

f 6 (Ag-ro(t*A) )k
×
 367

k
Die f liefern globale Schnitte yon ~ ~ber A u n d es g i b t
X g-r o
v
ein 7
_o , so dab fx verschwindet fur X { Xo + ~

Wenn oI ~ ein in ~ enthaltener Polyeder ist, so s t i m m e n die fx

fur ~1 mit denen fur ~ ~berein. SchlieSlich sind sie n o c h invariant
unter der Gruppe GL(X ) , genauer gesagt, f~r a 6 G L ( X o) ist
fa(x) = det(a)k.f X (beachte die Operation yon a auf

d iog(~1)A...Ad 1og(~ r ) !)
O

Wenn r > I , so f o l g t d a n n s c h o n a u t o m a t i s c h , dab f % 0 nur gelten

o X
kann, wenn X 6 B+(Xo)~ v ( K o e c h e r - P r i n z i p ) : X d e f i n iert eine Linear-
+
form auf B(Xo) ~ . FUr jeden rationalen konvexen Polyeder ~I ~ B (Xo) ~
g i b t es X I , so dab a(X) 6 X I + c vI fur a l l e a E G L ( X o) Wenn
C E o] ( ~ ) eine kompakte Teilmenge ist, so i s t d a n n die d u r c h X de-
finierte Linearform nach unten beschr~nkt auf a6~L(Xo) a(C)

Wenn r > I ist, so k a n n m a n a b e r d u r c h g e e i g n e t e W a h l v o n C errei-

c
chen, dab die k o n v e x e H ~ l l e d e r o b i g e n M e n g e g l e i c h B+(Xo)~ ist.
Also liegt X in B+(Xo) v

Aus u n s e r e r Konstruktion yon

A folgt, dab f s i c h g e n a u d a n n zu
kg
einem regul~ren Schnitt von ~ auf A a u s d e h n t , w e n n f u r alle d
g
in d e r g e w ~ h l t e n K e g e l z e r l e g u n g die K o e f f i z i e n t e n fx verschwinden,
falls X { ov - F~ir j e d e s einzelne d ist d i e s [quivalent zur R e g u l a r i -
t~t von f auf d e m S t r a t u m S , und f ist s c h o n auf ganz S re-
gul~r, wenn dies in e i n e m Punkt yon S gilt.
--o

Daraus folgt, dab m a n n u r die o's mit dim(u) = g(g+1) maximal

2
betrachten muS. Dort erhilt man eine Entwicklung

f : X ZB(X) * ' X ' f -(d l o g ( ~ q ) A . . . ^ d log(~g)) k ,

X

mit f 6
X

f definiert eine Modulform <:> f : 0 fur X { B+ (X) v (Dies ist

X
automatisch, falls g ~ 2 )
 368

Man kann statt Z auch andere Grundringe w~hlen, wie {, Z/nZ u.s.w.
Beim G r u n d r i n g { erh~it man bis auf einen Faktor (2~i) gk die klas-
sisehe F o u r i e r e n t w i c k l u n g einer Modulform, indem man etwa B+(X) v ]
identifiziert mit den h a l b g a n z e n s y m m e t r i s c h e n p o s i t i v d e f i n i e r t e n
Matrizen. Es folgt zum Beispiel, dab der Raum der M o d u l f o r m e n Hber
eine Basis besitzt, deren Elemente ganze F o u r i e r k o e f f i z i e n t e n haben,
und dab eine ~ - M o d u l f o r m genau dann ~ber ~ d e f i n i e r t ist, wenn alle
Fourierkoeffizienten in (2~i)gk-z liegen.

Wir n o t i e r e n noch eine weitere E i g e n s c h a f t der Modulformen: Sei wieder

f = X6B*Z ( X ) v X - f x ( d l o g ( ~ 1 ) ^ . . . ^ d log(~r ))k

Sei <o>~B(X ) das yon ~ a u f g e s p a n n t e Untergitter. Dann ist

B+(X )v N < ~ > i = {0} , da ~ eine positiv definite F o r m enth~it. Es
folgt, dab f k o n s t a n t ist auf dem a b g e s c h l o s s e n e n S(~) --Orbit in
S(~)~
Der konstante Wert wird gegeben durch die M o d u l f o r m f0 ' vom Gewicht
k , auf A . Man Oberlegt sich leicht, dab die F o u r i e r k o e f f i z i e n t e n
g-r~
der E n t w i c k l u n g yon f0 p a r a m e t r i s i e r t werden durch
X 6 B + ( K e r n ( X + X ) )v c B + ( X ) v , und dab man fir solche X dieselben
K o e f f i z i e n t e n fur f und f0 erh~it. (Betrachte f auf P r o d u k t e n
AI x A2 der D i m e n s i o n e n (r ,g-r ))

E n t s p r e c h e n d e s gilt auch f~ir M o d u l f o r m e n mit L e v e l - S t r u k t u r (~iber

Z ~I/n, e2Zi/n I zum Beispiel) : Dort w i r d die F o u r i e r - E n t w i c k l u n g para-

m e t r i s i e r t durch X 6 1--Bn (X~)v , und zu jedem 0 g e h ~ r e n mehrere

Fourier-Reihen.

c) Beispiele fur M o d u l f o r m e n erh~it man durch @-Reihen : W~hle

I g
a , b 6 (n Z/•.) . Dann ist bis auf 4.-te E i n h e i t s w u r z e l n

8(Z;a,b) = e i~atb Z e i~((m+a)zt(m+a)) e 2i~mtb

m£~g
 369

eine Modulform vom Gewicht I/2 zum Level 2n 2 , u n d ein P r o d u k t 4k

solcher Q's l i e f e r t eine M o d u l f o r m vom Gewicht 2k . Die F o u r i e r -

koeffizienten liegen in Z [ e i Z / n 2] , so dab die e n t s p r e c h e n d e n Modul-

formen ~ber diesem Ring definiert sind. (Bis auf 2~i's ). Die F o u r i e r -
koeffizienten von 8(Zja,b) werden parametrisiert durch
X 61 B(X)* = I S2(X) Sie sind v e r s c h i e d e n y o n Null nur f~r X
2n 2 2n 2 "
vonder Form

1
{~(_m+a) ® (m+a), mEx : zg].

Es folgt, dab fNr e i n e n K e g e l ~B+(X) der k o n s t a n t e Term eines Produktes

yon 8's verschwindet, a u B e r w e n n alle v o r k o m m e n d e n a'S im K e r n v o n

I/n-X/X +I/n X ~ O liegen. In d i e s e m Fall ist der k o n s t a n t e Term wieder

ein P r o d u k t solcher 8's , mit g-r statt g .
G

Die 8(Z;a,b) h ~ n g e n m i t den 6-Nullwerten zusammen: Wir a r b e i t e n von

nun an in C h a r a k t e r i s t i k #2 . B e i m a) unter 3.) k o n s t r u i e r t e n Iso-

morphismus p.(H) ~ K 2g , mit K ~ -11 modulo Torsion in PiC(Ag,2) ,

entsprechen die 8(Z;a,b) mit a,b C j X / X im g e e i g n e t zu de f i n i e r e n d e n
Sinne einer Basis yon p,(H) . Entsprechendes g i l t f~r die d i r e k t e n

Bilder p , ( H 21)

Es ist nun b e k a n n t , dab d i e s e Basen H erzeugen, u n d dab die 0-Null-

werte sogar eine p r o j e k t i v e Einbettung yon Ag definieren (genauer
gesagt des g r o b e n M o d u l r a u m s ) . Da wir das V e r h a l t e n der 8-Reihen am
R a n d e a u c h kennen, so f o l g t leicht:

Satz 6:

W~hle n , u n d sei A = m~0 F(Ag,n,~m)@z[I/2] • Dann wird for g e n ~ g e n d

groBes m m ~ber Ag,n@7.[I/2] yon seinen globalen Schnitten erzeugt.

Die d a d u r c h definierte Abbildung

 370

hat als Bild ein p r o j e k t i v e s normales Schema A* [ber Zte2~i/n,I/2n].[]

g,n
Sie d e f i n i e r t eine offene E i n b e t t u n g des groben M o d u l r a u m s zu
A
g,n
in A* , und das K o m p l e m e n t hat auf jeder Faser D i m e n s i o n g(g-1)/2 .
g,n
Genauer hat das Bild jedes Stratums A n , o Dimension (g-ro) (g-ro+1)/2 .

Das w e i t e r e Studium der a r i t h m e t i s c h e n Theorie der S i e g e l ' s c h e n Modul-

formen v e r d i e n t sicher noch einige Aufmerksamkeit: V e r m u t l i c h g e l t e n die
o b i g e n Resultate auch in C h a r a k t e r i s t i k zwei, und es sollte auch Anwen-
dungen auf K o n g r u e n z e n geben. Dies w0rde aber wohl den Rahmen der hie-
sigen A u s f O h r u n g e n sprengen.

Eine weitere V e r f o l g u n g der Ans~tze von L. M o r e t - B a i l l y ([MB]) scheint

hier geboten.

§ 7 ETALE GARBEN

a) Durch die m o d u l a r e I n t e r p r e t a t i o n erh~lt man sofort ~tale Garben

auf A , namlich die direkten Bilder RJp.(Z/m~) ( p : G + A die univer-
g g
selle semiabelsche Variet~t). Der E i n f a c h h e i t halber f o r m u l i e r e n wir
die A u s s a g e n fdr ~i-Garben, d o c h gelten e n t s p r e c h e n d e V a r i a n t e n f~r
Z1 oder ~/i~{~-- . A u B e r d e m nehmen wir an, dab unsere K e g e l z e r l e g u n g
glatt ist, so dab alle Strata A glatt 0ber Z sind.
g,o

W~hle eine Primzahl 1 und einen Level n . Wir arbeiten g r u n d s ~ t z l i c h

~ber Z[e2~i/n, I/nl]

Auf Ag,n ist die Garbe R1p,(~ I) lokal k o n s t a n t und besitzt eine
nicht a u s g e a r t e t e symplektische Form mit Werten in 91(-I) ((-I) =
Tate-Twist). Am Rand ist sie zahm verzweigt: A -A ist ein Divisor
g,n g,n
mit n o r m a l e n ~berkreuzungen. Seine i r r e d u z i b l e n K o m p o n e n t e n sind die
Strata Ag,n,o f~r ocB+(X)~ ein e i n d i m e n s i o n a l e r Kegel der Zerlegung.
(FHr n = I e n t s p r e c h e n sie sogar e i n d e u t i g den K o n j u g a t i o n s k l a s s e n
dieser o unter GL(X)) . Die O p e r a t i o n der zugeh~rigen M o n o d r o m i e
e r h ~ i t man aus der B e s c h r e i b u n g der l - T o r s i o n s p u n k t e einer d e g e n e r i e r e n -
den a b e l s c h e n Variet~t, die wir in § 3 (entsprechend [M4], 4.11) g e g e b e n
haben: Sei so ein E r z e u g e n d e s der H a l b g r u p p e <o> A B+(X). so ist eine
p o s i t i v definite symmetrische B i l i n e a r f o r m s o : X~ x X ÷~ und
 371

definiert ein unipotentes Element aus Sp(2g,Z) (Wenn s o durch

I S
eine symmetrische Matrix S gegeben wird, ist die (0 1 ) . Dieses
Element liefert die gew0nschte Monodromie-Transformation.

b) Da RIp* (~l) zahm verzweigt ist im Unendlichen, sind die direkten

Bildgarben (~ber X[e2~i/n,I/in]) lokal konstant. (Siehe zum Beispiel

[L]). Dasselbe gilt for aus R1p*(~ I) abgeleitete Garben:

Satz 7:

Sei p : Csp(2g,~) ÷ G L ( V ) eine algebraische Darstellung der Gruppe

Csp der symplektischen ~hnlichkeiten auf einem endlich dimensionalen
~-Vektorraum V . F~r n ~ 3 bezeichne F
-p die zugeh~rige ~tale ~l-

Garbe auf Ag,n, ~ : A g , n ~ S p e k ( Z ~ e_

)2 ~i/rn., ~ ] die Projektion. Dann sind

alle direkten Bilder Rq~,(F- - p ), Rq~,* (Fp)

--
lokal konstant.

Bemerkung:

Es w~re w ~ n s c h e n s w e r t , a u c h die E i c h l e r - S h i m u r a Relation zu verallgemei-

nern. Dies scheint jedoch sehr k o m p l i z i e r t zu sein. Auch hier ist noch
ein weiteres Feld fur zuk~nftige Untersuchungen.

§ 8 DIE TORELLI-ABBILDUNG

a) Zu jeder glatten Kurve vom Geschlecht g geh~rt kanonisch ihre

Jacobische, eine prinzipal polarisierte abelsche Variet~t der Dimension
g . Wenn man die glatte Kurve in eine singul~re stabile Kurve degene-
rieren ! ~ t , ergibt sich eine semiabelsche Variet~t, und es liegt nahe,
die G e g e b e n h e i t e n der allgemeinen Theorie in § 2,3 bier n~her zu be-
schreiben, Wir u n t e r s u c h e n dabei zugleich das V e r h a l t e n der Torelli-
Abbildung M ÷A zwischen den Modulr~umen am Rande.
g g

b) Sei wieder R ein kompletter normaler lokaler Ring, K der Quo-

tientenk~rper, C ÷Spek(R) eine stabile Kurve, so dab die generische
Faser C glatt ist. Zur speziellen Faser C konstruiert man einen
s
 372

Graphen G ,dessen Ecken V den irreduziblen Komponenten yon C

s
entsprechen, und dessen Kanten E die singul~ren Punkte auf Cs para-
metrisieren. Eine Kante hat als Endpunkte die beiden Ecken, die den
irreduziblen Komponenten entspreehen, auf denen der singul~re Punkt
i ie gt.

Wir nehmen weiter an, dab alle irreduziblen Komponenten yon Cs geome-
trisch irreduzibel sind und alle Doppelpunkte rational 0bet dem Rest-
klassenk~rper k von R . Dies gilt immer, wenn k algebraisch ab-
geschlossen ist.

Jeder Kante e 6 E ordnet man ein Hauptideal Ie mR zu: Die Komplet-

tierung ~C,e des lokalen Rings im P u n k t zu e ist isomorph zu
R [ [ S , T ] ] / ( S T - f e) , und Ie werde erzeugt von fe

e) Sei F = ~I(G) die Fundamentalgruppe des Graphen @ , und

~ --> G die universelle Uberlagerungl mit Gruppe F . Dann gibt es
A
eine Uberlagerung formaler Schemata C ~ C , ebenfal!s mit Gruppe F .

Sei X = Fab = HI(G,~), Y : X* = HI(G,~) , und T der zer fallende

Torus mit Charaktergruppe X . Die Elemente aus T(R) entsprechen den
Homomorphismen F +R* . Jeden solchen Homomorphismus kann man benutzen,
tun e i n e ~quivariante Operation yon F auf dem trivialen GeradenbOndel
A
0~ zu definieren, und damit ein Geradenb~ndel auf C oder auch C .
c

Da F(C,0~) = R , erh~it man so e i n e n Isomorphismus

C

A
T(R) "~ > K e r n ( P i e (C) - - > P i c (~))

c) Sei G : P i c 0 (C/R) . Dann ist G eine semiabelsche Variet~t Hber

R , und der eben angegebene Isomorphismus stammt aus einer exakten
Sequenz

o -~ ~'-~ ~-~ ~.--, o ,

mit A abelsch ~ber R . Sei wieder G die entsprechende Erweiterung

0~T~G~A~0
 373

Es ist G(R) = G(R) . Man erh~It auch eine Abbildung T(K)G(R) ~Pic0(CQ) :
Es r e i c h t , die R e s t r i k t i o n T(K) ÷ P i c 0 ( C q ) anzugeben. Ein Element aus
T(K) wird gegeben durch einen Homomorphismus F ÷K* , welcher eine
~quivariante Operation yon F auf 0~ ® R K definiert. W~hle ein F-in-
C
variantes gebrochenes koh~rentes Ideal Jc0~®R K . Dies ergibt eine
-- C
koh~rente torsionsfreie Garbe auf ~ oder auch C , u n d auf d e r g e n e -
rischen Faser ein Geradenb~ndel vom Grad 0

Wir definieren nun eine Abbildung yon X in d e n Kern des obigen Mor-
phismus: FUr e 6 [ sei 0C, e der zugeh~rige lokale Ring, und fe
ein erzeugendes Element von Ie . Ein X 6 X w£rd gegeben durch ganze
Zahlen x 6Z , f~r a l l e e 6 E , so d a 6 f~r alle Z ±x p E V
= 0 .
e e÷p e
Dabei w~hlt man eine Orientierung aller e 6 [ , und die Summe geht ~ber
alle e's mit Anfangs-oder Endpunkt p , wobei das V o r z e i c h e n je n a c h
Orientierung zu w [ h l e n ist. Wir definieren dann eine symmetrische Bi-
linearform

b : XxXcK*

dutch
XeY e
b(x,y) = e~[ fe

Dann ist b(x,x) 6 m = m a x i m a l e s Ideal R , falls x % 0 b( ) ent-

spricht einem Homomorphismus

b : X~T(K)

Es gilt nun

Satz 8:

Es sei C h a r (K) % 2

i) Es g i b t e i n e n Homomorphismus c : X~G(R) = Pic0(C) , so dab

fur a l l e x6 X b(x) c (x) 6 G(K) das triviale Geradenb~nde! auf
C definiert.

ii) Sei M ein amples GeradenbUndel auf A , welches die p r i n z i p a l e

Polarisation definiert. Dann ist die Zusammensetzung

X _c> G(R) ~ A ( R ) die zur E r w e i t e r u n g ~ yon A durch T

 374

geh~rige Abbildung. Ihre Liftung via c definiert ein zul~s-

siges System von Isomorphismen fur (G,A,M)

iii) G ist isomorph zu der semiabelschen Variet~t, welche die M u m -

ford-Konstruktion mit den o b i g e n Daten liefert.

Beweis:

i) Sei x6 X . b(x, ) definiert eine Darstellung p : F ÷X+K* .

Wir konstruieren ein p-invariantes gebrochenes Ideal jc 0~®K ,

-- C
welches lokal prinzipal ist, und dessen Einschr~nkung auf jede Komponen-
te v o n ~ den Grad Null hat. Die Komponenten C von ~ werden
s p s
parametrisiert durch p E~ = Ecken (7) . Wir w~hlen ~ so,daB es im
Inneren von Cp von einem gp E K* erzeugt wird, u n d dab for e i n e
X e
Seite e 6 ~ , welche Pl und P2 verbindet, gP2 = fe gPl . Es

gibt sicher solche gp'S , und das Ideal ~ ist a u c h p-invariant. Es

bleibt zu zeigen, dab m a n J auch in d e n Doppelpunkten lokal prinzipal
w~hlen kann: Sei e 6 ~ , und betrachte fur n ~ I den Ring
(04 ) ® (R/mn) . Er b e s i t z t zwei m i n i m a l e P r i m i d e a l e ~I und ~2 '
c,e
entsprechend pl,P2 6~ , welche durch e v e r b u n d e n w e r d e n . ~I und ~2
sind H a u p t i d e a l e , und ~i.~2 = (fe)

In der Komplettierung wird

04c,e ~ R [ [ S ' T ] ] / ( S T - f e )

A A
und [I und P2 werden durch S bzw. T erzeugt.

Man kann dann J so w ~ h l e n , dab es in 0~ ® (R/ n) isomorph wird zu

-- C , V m
einer Potenz yon ~I oder ~2 (etwa zu ~i ~v) ,-und es f o l g t die
Behauptung. Der Grad von J auf jedem C ist Null, da Z ±x = 0
-- p e÷p e

Damit ist Teil i) b e w i e s e n . Die A b b i l d u n g c ist n a t ~ r l i c h eindeutig.

d) Es f o l g t schon aus den S~tzen 1 und 2, dab m a n G durch die

Mumford-Konstruktion mit Hilfe einer Bilinearform b* aus ~ erh~it.
Wir m ~ s s e n nun noch nachweisen, dab c ein zul~ssiges System yon Iso-
morphismen definiert, und dab dann b = b* . Man bettet R in e i n e n
diskreten Bewertungsring ein, und reduziert sich d a m i t auf den Fall,
 375

dab R s c h o n ein solcher ist. Dann ist das P r o b l e m fur den Fall e i n e r
voll d e g e n e r i e r e n d e n Kurve schon in [MD] b e h a n d e l t worden, und wit
f o l g e n den d o r t i g e n A u s f ~ h r u n g e n : Wit k~nnen C,~ , G und G als
rigid-analytische Objekte ~ber K auffassen. Es ist d a n n C = ~/F ,
und G/i(X) ist r i g i d - a n a l y t i s c h e abelsche Variet~t
(i = b.c : X~G(K)) . Wit definieren zun~chst eine rigid-analytische
Abbildung ¢ : C x C~G/i(X) , via ~ : C x C~G . D a z u m ~ s s e n wir f0r
jeden endlichen Erweiterungsk6rper L von K eine A b b i l d u n g
: ~(L) × ~(L) ~ G ( L ) definieren. Da alle u n s e r e Konstruktionen inva-
riant unter Grundk~rpererweiterung sein w e r d e n , r e i c h t es, ~ auf
~(K) x ~(K) zu d e f i n i e r e n . Dazu ersetzen wir zun~chst C und ~ durch
ihre r e g u l ~ r e n semistabiien Modelle ~ber R . D i e s ~ n d e r t n i c h t s an
allen Definitionen und Behauptungen. Dann ist ~(K) = ~(R) . Seien also
Zl,Z 2 E C(R) zwei Punkte. Der D i v i s o r D = y ~ F y ( ( z I) - (z2)) ist d a n n
F-invariant auf ~ , d o c h hat die E i n s c h r ~ n k u n g von 0(D) auf die
irreduziblen Komponenten C im a l l g e m e i n e n n i c h t den G r a d Null. Dies
P
w i r d nun k o r r i g i e r t :
e) W~hle ein P0 , und definiere eine A b b i l d u n g

f : p ~ - > f(p0,p)
V --> K*

wie folgt: Orientiere F-invariant die K a n t e n E . Es gilt:

i) f(p0,P0 ) = I
i±) W e n n der k O r z e s t e Weg in G yon p nach P0 der K o m p o n e n t e als
n~chste Ecke Pl trifft, so sei

f(p0,p) = f(p0,pl) w e n n die O r i e n t i e r u n g der K a n t e e zwischen

p und Pl so ist, dab p Anfangs- und
Pl Endpunkt ist.

f(p0,p) = f(p0,Pl)fe , bei anderer Orientierung.

Definiere

I ~ (YPl 'p) f (YP2'P0) 1

g(p1,P2;p) = H E K*
yEF (7P2'P) f(YPl 'P0 )

Dabei seien p1,P2 E~ . Im P r o d u k t ist ein F a k t o r nur d a n n v e r s c h i e d e n

 376

von Eins, w e n n die k [ r z e s t e n Verbindungen (in ~ )von 7p I und YP2

--> >
m i t dem W e g p0 p verschiedene Fu~punkte auf p0 p haben. Dies gilt
aber nur fur e n d l i c h v i e l e T's (Die D i s t a n z von Pl naeh P2 mug
gr~Ber sein als K o n s t a n t e + Distanz (p0,TPl) )

Es ist g(pl,P3;p) = g ( p l , P 2 ; p) g ( p 2 , P 3 ; p )

W e n n m a n von p zu e i n e m b e n a c h b a r t e n p' 6 ~ ~bergeht, und e die

Kante zwischen p und p' bezeichnet, mit p=Anfangspunkt (e) , und
p'=Endpunkt (e), so ist

g(Pl "P2 'p' ) f (YPl,P') f (YP2 ,P)

= I]

g ( P I ' P 2 "p) TEF f (yP2,P') f (YPI ,P)

Die F a k t o r e n sind v e r s c h i e d e n y o n Eins nut dann, wenn e auf d e m Weg

von YPl nach Tp 2 liegt, und zwar e r h ~ I t m a n d a n n fe ' w e n n die
-I
Orientierung yon e m i t der des W e g e s ~bereinstimmt, sonst f
e
Es folgt:

i) Sei 6 6 F , entsprechend x : (x e) E X .

Dann ist

f ( P I ' P 2 ; ~ (P)) = ][ f _+x e

......>....... e
f ( P 1 ' P 2 ;p) e 6 piP2

- - >
Der E x p o n e n t ist + I , w e n n die O r i e n t i e r u n g e n v o n e und plP2 Ober-
einstimmen, sonst - 1 .

ii) Sei J(pl,P2 ) ~0~®K das i n v e r t i e r b a r e Ideal, welches auf Cp

von f(pl,P2; p) e r z e u g t wird. (Die E x i s t e n z folgt [hn!ich wie bei der
Konstruktion yon c )
Dann ist

grad(J(pl,P2) / Cp) =

0 , falls p ~ F.pl U F.p2 , oder p 6 F'pIDF.p 2

-I , falls p 6 ~'Pl' P { FP2

+I , falls p 6 F.p2, p ~ Vpl

 377

S o m i t hat fir z I E Cpl , z 2 6 Cp2 die E i n s c h r ~ n k u n g yon

L ( Z l , Z 2) = ~ ( p l , P 2 ) @ 0(D) Grad 0 auf a l l e n K o m p o n e n t e n . F~r £6F ,

entsprechend s = (x e) 6 X , ist

±x e
6*(L(zl,z2) = ( ~--> fe (PI'P2))~(z1'z2)
e 6 plP2

W e n n m a n also e i n e n M o r p h i s m u s

p : F+X÷K*

+x e
definiert durch p(6) = H > f - , entsprechend einem p 6 T(K) ,
eEPlP 2 e
so k a n n m a n F via p ~quivariant operieren lassen und erh[it ein
Geradenb~ndel aus G(R) = Pic0(C) . Wenn Pl = P2 ist, so w i r d d i e s e s
GeradenbQndel d u r c h den D i v i s o r Zl-Z 2 auf C gegeben. Wenn
z 2 = 7(z I) mit y 6 F , entsprechend y = (ye) £ X , so e r h ~ I t m a n das
Geradenb~ndel zu c(y) 6 G ( R ) . W e n n m a n das o b i g e E l e m e n t aus G(R)
noch mit p 6 T(K) multipliziert, ergibt sich s c h l i e B l i c h eine Abbildung

~ : ~(R) x ~(R) ~ T ( K ) G ( R ) : G(K)

mit

i) ~ ( Z l , Z 2) + ~ ( z 2 , z 3) = ~ ( Z l , Z 3)
ii) ~(zl,7(zl)) = i(y) (y = Bild(7) E X )

iii) Wenn Zl,Z 2 in d e r s e l b e n Komponente liegen, so ist

~ ( Z l , Z 2) 6 ~ ( R ) : Pic°(C) gegeben durch 0 ( z l - z 2)

f) AuBerdem ist ~ vertr~glich mit Erweiterungen des G r u n d k 6 r p e r s .

Andererseits hat m a n ~ber K eine kanonische Abbildung

: Crl x C q ~ G q = P i c 0 (C n)

(Zl,Z 2) I > 0(Zl-Z 2)

Es ist G der r i g i d - a n a l y t i s c h e Quotient G/i*(X) , mit einer Gruppe

 378

yon Perioden i*(X) ~ G ( K ) . Es ergibt sich dann eine rigid-analytische

Abbildung der universellen Uberlagerungen

CxC~G

Da diese Abbildung mit ~ nahe der Diagonale ~bereinstimmt, ist sie

nach dem Identit~tssatz gleich ~ .

Wir erhalten also ein kommutatives Diagramm

I i*(x)
C x C----~> G

Da ~ ( Z l , 6 Z 1) 6 i(X) , ist i(X) ~ i * ( X ) . Aus einer Betrachtung der

Bewertung folgt, dab i(X) endlichen Index in i*(X) hat. ¢ fakto-
risiert dann ~ber ~/i(X) Dies ist e i n e endliche Dberlagerung yon
G , damit algebraisch, und wegen der bekannten Eigenschaften der
Jacobi'schen ist n o t w e n d i g e r w e i s e i(X) = i*(X) . Wir m~ssen nun noch
zeigen, dab i = i* . Auf jeden Fall stimmen sie schon ~berein bis auf
einen Automorphismus yon X . Da i* ebenso wie i yon einer definiten
symmetrischen Bilinearform stammt, wird dieser Isomorphismus in e i n e r
geeigneten Basis durch eine positiv definite symmetrische Matrix
definiert.

g) Alles in a l l e m haben wir den Satz 8 bewiesen bis auf die Tatsache,
dab m a n fur ii) und iii) i durch einen Automorphismus von X ab~ndern
muB. Wir wollen zeigen, dab dieser Automorphismus die Identit~t ist.

ES r e i c h t , dies im " u n i v e r s e l l e n " Fall zu tun, das heiBt,wenn R die

Basis einer versellen Deformation der speziellen Faser Cs ist. Dann
ist R regul~r, und die f fur e 6 E bilden einen Teil eines regu-
e
l~ren Parametersystems. F~r jedes e sei ~e ~R das zugeh~rige Prim-
ideal der H@he I und Ge die Faser von G ~ber k(Ee) . Dann ist Ge
semiabelsch, mit einem Torusteil der Dimension ~ I . Die Dimension
ist g e n a u dann gleich I , wenn die L i n e a r f o r m x = (x e) ~--> x e auf
X nicht verschwindet. Falls dies der Fall ist, so b e s t e h t der Kern
dieser linearen Abbildung aus den x , die auf d e m T o r u s t e i l von G
e
verschwinden. (Der T o r u s t e i l von G ist in n a t N r l i c h e r Weise ein
e
 379

Untertorus von T ). A u s der Mumford-Konstruktion folgt, dab dies genau

dann f~r x zutrifft, wenn i*(x) ganz ist in R
Re
Wenn I: X --~> X der Automorphismus mit i* = iol ist, so g i l t dann

x e : 0 <=> l(x) e = 0

I respektiert also alle Hyperebenen {x : 0} , kann trigonalisiert

e
werden und hat somit Eigenwerte ± I . Da man 1 aber auch durch eine
symmetrische positiv definite Matrix darstellen kann, ist I = id .
Dies beendet den Beweis von Satz 8.

Wir erhalten auch Informationen ~ber das Verhalten der Abbildung

M ~A am Rande: Bei vorgegebenem g gibt es n u t endlich viele M~glich-
g g
keiten f~r den Graphen G einer stabilen Kurve vom Geschlecht g . Zu
jedem solchen G erh~it man symmetrische Bilinearformen (x,y) ~ X e Y e
auf X = HI(G,Z) fur e 6 E . Dann bildet sich die verselle Deformation
einer stabilen Kurve mit Graph G genau dann in d i e mit der Kegelzer-
legung {o} definierte Kompaktifizierung A ab, wenn es ein o gibt,
g
welches alle Bilinearformen XeY e auf X enthilt. Wenn dies nicht der
Fall ist, so m u B man zun~chst noch die verselle Deformation durch eine
Modifikation ersetzen.

§ 9 DIE KOMPLEXE THEORIE

a) Nach Basiserweiterung zu d e n komplexen Zahlen erh~lt man aus

Ag, S u.s.w, komplexe R~ume. Es ergeben sich die toroidalen Kompakti-
fizierungen aus [AMRT] oder auch [N] . Da sich auch die Mumford-Konstruk-
tion ins analytische ~bersetzt, kann man auch die komplex-analytische
Version der universellen semiabelschen Varietit beschreiben:

Der Torus S wird ~ber • gegeben dutch

S(~) = ~g(g+1)/~g(g+1)/2 = B(~g)/B(~g)

Wenn ~ c{ g(g+1)/2 die Siegel'sche Halbebene bezeichnet

(~ = {~ = X + i Y / Z = tz, Y > 0} ) , so e r h ~ i t S({) als offene
g
 380

Teilmenge ~ g/B(Z g )

Wenn d c B + ( ~ g) ein konvexer rationaler Polyeder ist, so b e s t i m m t

eine Torus-Einbettung Sc S . S ist a f f i n algebraisch, und eine
Basis des R i n g e s der algebraischen Funktionen wird gegeben durch

2~i S p u r (MZ)
ZM(Z) : e

M eine halbganze symmetrische Matrix, mit ganzen Diagonalelementen,

welche im D u a l v liegt (Y 6 o ~ Sp(MY) ~ 0) . Weiter existiert auf
S eine "universelle" Bilinearform

b : ~g x~zg~0
s
b(x,y) : XM ,

I
wobei M die Matrix ist mit Eintrigen mjk~ = ~ ( x iky~ + X k Y j) . b(x,x)
setzt sich fort zu e i n e r regul~ren Funktion auf S
d

Sei XcZ g die 6tale Untergruppe, deren Faser ~ber s 6 So aus d e n

x E~e besteht mit b(x,x) (s) # 0 . Dann setzt sich b fort zu e i n e r
Bilinearform

b : xxzg~0 * ,
-- S
o

und definiert damit b : --

X~ (~*)g = T . b ist ~ b e r ~ g
/B(~ g) eine
Einbettung, und der Quotient G = T/b(X) ist e i n e semiabelsche Varie-
t~t ~ b e r e i n e m offenen StUck von S . Dort ist S lokal isomorph zu
o
Ag(~) , und G liefert die universelle semiabelsche Variet~t.

b) Das obige G besitzt ein GeradenbHndel L , welches [ber ~ g / B ( Z g)

eine Polarisation definiert. Das Pullback von L nach T ist k a n o n i s c h
trivial, und ein globaler Schnitt von L wird geliefert durch die
8-Reihe auf T :

2ziz I 2~iz
8(e ..... e g) =
g t
= ~ ei~(mZtm+j~imjZjj)e2~im z
= (ml, .... mg) 6 ~ g
 381

b ist die zu d e n Koeffizienten von ~ geh~rige Bilinearform (im S i n n e

von§ 2).

c) Das Quadratintegral von g-Formen liefert eine kanonische hermite'-

sche Metrik auf ~ = igt* G

]]~]t2(~) = f I1~II 2
G
s

Man rechnet aus, dab f~r Z = X+iY6 ~ , entsprechend s 6 S ,

g

dz I dz 2
-- A . . . A
z I Zg

bis auf einen konstanten Faktor gegeben ist durch det(Y) . Da d i e

Eintr~ge von Y sich aus den Logarithmen der absoluten Betr~ge der
XM berechnen, hat die Metrik auf ~G am R a n d e nur eine logarithmische
Singularit~t.
 382

ANHANG:

Dieses M a n u s k r i p t gibt den K e n n t n i s s t a n d zur Zeit der A r b e i t s t a g u n g

wieder (Juni 1984). Inzwischen (September 1984} gab es die folgenden
Entwicklungen:

I.) Die Thesis von C.L. Chai liegt mir vor.

2.) Die m i n i m a l e K o m p a k t i f i z i e r u n g l~Bt sich auch in C h a r a k t e r i s t i k

2 behandeln:

Betrachte Ag,n ~ber X[I/n,e 2'~i/n]

Sei R = m~= 0 F(Ag,n, ®m) , A*g,n = Proj(R)

Dann gilt:

i) Eine geeignete Potenz von ~ wird yon g l o b a l e n Schnitten erzeugt,

so dab man eine A b b i l d u n g des groben M o d u l r a u m s (zu A g , n ) A g , n nach
A * erh~it:
g,n

: A ÷ A*
g,n g,n

-I
ii) %IAg,n ist eine offene Einbettung, und Ag,n = ¢ (%(Ag,n))

iil) A* -A hat K o d i m e n s i o n g in A*
g,n g,n g,n

Der Beweis benutzt e-Funktionen. Einige Andeutungen: Man b e t r a c h t e t

die U b e r ! a g e r u n g M ÷ A , welehe die s y m m e t r i s c h e n G e r a d e n b ~ n d e l in der
g
Polarisationsklasse liefert (nicht zu v e r w e c h s e l n mit level-2-Strukturen) .
F~r m ungerade (der E i n f a c h h e i t halber) erh~it man dann ein Geraden-
b~ndel L auf MxA A , welches f~r m ~ 3 von seinen g l o b a l e n
--m g,m
g m
Schnitten erzeugt wird, und so dab --mL = ~.~ in Pic ® ~

Dies liefert globale E r z e u g t h e i t ~ber A g . Man muB nun noch den Rand
betrachten,sowie zeigen, dab die Fasern yon ¢ : A g ÷ A *g endlich ~ber
A sind.
g
 383

Literatur

[A] M. Artin, A l g e b r a i z a t i o n of Formal Moduli I,in Global Analysis,

papers in honor of K. Kodaira, 2 1 - 71, P r i n c e t o n U n i v e r s i t y
Press, P r i n c e t o n 1969.

[A~IRT ] A. Ash, D. Mumford, M. Rapoport, Y. Tai, Smooth C o m p a c t i f i c a t i o n

of Locally Symmetric Varieties, Math. Sci. Press, Brookline
1975.

[C] C.-L. Chai, Thesis, Harvard 1984.

[DM] P. Deligne, D. Mumford, The i r r e d u c i b i l i t y of the space of

curves of a given genus, Publ. Math. IHES 36 (1969), 7 5 - 110.

[DR] P. Deligne, M. Rapoport, Les sch6mas de modules de courbes

elliptiques, Springer Lecture Notes 349 (1973) 1 4 3 - 316.

[F] G. Faltings, E n d l i c h k e i t s s ~ t z e f~r abelsche V a r i e t ~ t e n ~ber

Zahlk~rpern, Invent. Math. 73 (1983), 3 4 9 - 366.

[SGA7] A. Grothendieck, Groupes de m o n o d r o m i e en G6ometrie A l g e b r i q u e

(SGA 7 I), Springer Lecture Notes 288 (1972).

[K~{S] G. Kempf, F. Knudsen, D. Mumford, B. Saint-Donat, Toroidal

Embeddings I, Springer Lecture Notes 399 (1972).

ILl G. Laumon, S e m i - C o n t i n u i t 6 du Conducteur de Swan (d'apr~s P.

Deligne) S6minaire E.N.S. (1978/79), Expos6 9.

[MD ] Yu. Manin, V. Drinfeld, Periods of p-adic Schottky groups,

Crelles Journal 262 (1973), 2 3 9 - 247.

[MB ] L. Moret-Bailly, Familles de variet6s a D 6 1 i e n n e s , T h ~ s e , Orsay

(7984).

[MI ] D. Mumford, Geometric Invariant Theory, Springer-Verlag,

H e i d e l b e r g 1965.

[M2 ] D. Mumford, On the Equations D e f i n i n g A b e l i a n Varieties,

Invent. Math. I (1966), 2 8 7 - 354.

[MS ] D. Mumford, An A n a l y t i c C o n s t r u c t i o n of D e g e n e r a t i n g Curves over

Complete Local Rings, Comp. Math. 24 (1972), 1 2 9 - 174.

[M4 ] D. Mumford, An A n a l y t i c C o n s t r u c t i o n of D e g e n e r a t i n g A b e l i a n
V a r i e t i e s over Complete Rings, Comp. Math. 24 (1972),
2 3 9 - 272.

IN] Y. Namikawa, Toroidal c o m p a c t i f i c a t i o n of Siegel spaces,

Springer Lecture Notes 812 (1980).
 THE S C H O T T K Y PROBLEM

Gerard van der Geer

By a s s o c i a t i n g to a (smooth irreducible) curve C of g e n u s g > 0

its J a c o b i a n Jac(C) one obtains a morphism M ~ A f r o m the m o d u l i

g g
space of c u r v e s of g e n u s g to the m o d u l i space of p r i n c i p a l l y polari-

zed A b e l i a n varieties of d i m e n s i o n g. A well-known theorem of T o r e l l i

says that this m o r p h i s m is i n j e c t i v e . The i m a g e of ~{g in Ag is n o t

c l o s e d , it is o n l y closed inside Ag
°, the s e t of p o i n t s of Ag that

correspond to i n d e c o m p o s a b l e principally polarized abelian varieties

(i.e. that are not products). For g=1,2,3 the c l o s u r e of the image

of Mg equals Ag . Since dimA g =g(g+l)/2 , dim Mg =3g-3 (for g > i)

one sees that for g > 3 dimAg > dim Mg , a n d so t h e q u e s t i o n arises

how we can characterize the image of M in A . This question goes

g g
back to R i e m a n n , but is u s u a l l y called Schottky's problem.

In "Curves and their Jacobians" Mumford treats the S c h o t t k y pro-

blem and the c l o s e l y related question how to d i s t i n g u i s h Jacobians from

general principally polarized abelian varieties. In h i s review of the

situation at t h a t m o m e n t (1975) he d e s c r i b e s four approaches and their

merits. He concludes that none of t h e s e seems to h i m a d e f i n i t i v e

solution. In the m e a n t i m e the s i t u a t i o n has c h a n g e d a lot. S o m e of

the a p p r o a c h e s have been worked out more completely, while new and

successfull approaches have appeared. This paper deals with them. I

hope to c o n v i n c e the reader that Mumford's statement that problems in

this corner of n a t u r e are subtle and worthy of h i s time still very much

holds true.
 386

The ingredients.

To begin with, some standard notations.

: the S i e g e l upper half space of d e g r e e g,

g
Fg : S p ( 2 g , 2 ) the s y m p l e c t i c group acting on ~g,
A B : A ~ D 5 i (mod n)
£g(n,2n) = {(C D ) CFg C B 0 (mod n) " d i a g t A C ~ d i a g t B D ~ 0(mod2n)}

Ag = Fg\~g , the m o d u l i space of p r i n c i p a l l y polarized abelian

varieties of d i m e n s i o n g over ~,

Ag(n,2n) = Fg(n,2n)\~g , a Galois cover of Ag .

If X is a p r i n c i p a l l y polarized abelian variety over { we

denote by LX (or s i m p l y L) a symmetric invertible ample sheaf of

degree 1 defining the polarization and by 0 the d i v i s o r of a n o n -

zero section of L . We put X = {x E X : nx = 0}. If X = ~g/~g+ T ~g

X n

(~ ~ ~ ) as a c o m p l e x torus then we write X = X

g
o2
The space F (X,L X ) has dimension 2 g. A b a s i s is d e f i n e d by the

functions

02[o](~,z) = ! exp 2~i(t(m+2)~(m+2 ) + 2(m+~)z)

m ~gg
z ~ {g, o c (zzg/2z~ g)

Here o is v i e w e d as a v e c t o r of length g with zeroes and ones as

entries. A different set of g e n e r a t o r s of F(X,Lx2) is g i v e n by the

squares of
i

e[~,]('c,z) = m! ~g exp 7ri(t(m+2)-r(m+2) + 2(m+~)(z+ 2 ))

with ~,~, ~ (2Z/2~)g, t , =0 (mod 2) . These are related by

e2[ ~,](T,z) = [ <o,~'>02[o+~3(<,0)@2[o](<,z) (i)

O
.t
<o, s'> = e x p z l os'

We call a principally polarized abelian variety indecomposable if

it is n o t a p r o d u c t of t w o p r i n c i p a l l y polarized abelian varietieS, i.e.

if its theta divisor is i r r e d u c i b l e .

The functions @2[o](T,z) define for X = X a morphism

Y
 387

CX: X ~N N = 2g-1
+
Z ÷ ( .... @2[0] ( ~ , z ) , . . . ) = @2(~,z)

which factors through z ÷ -z and is of d e g r e e 2 for indecomposable

X. The image is the K u m m e r variety of X. By t a k i n g z:0 and v a r y -

ing X we get a morphism

¢: Ag(2,4) ~ ~N
+
+ ( .... e 2 [ o ] ( T , 0 ) .... )= @2(s,0)

which is g e n e r i c a l l y of d e g r e e i. We also define

~: A (2,4) ~ ~ M M = 2g-l(2g+])-i
g
÷ ( .... e2E ~ 3(~,0) .... ).
E
and ~ are c o n n e c t e d by the special Veronese V defined by (i) :

~pN
V
Ag (2,4) ~ +
]pM

The m o r p h i s m s % and c a n be e x t e n d e d to m o r p h i s m s of the S a t a k e

compactification Ag(2,4) of Ag(2,4).

The functions @2[o] satisfy the d i f f e r e n t i a l equations

@2[o] = 4~Ti (i + & 4 ) ~ @2[o] , I < i,j s g.

~z. ~z.
1 ] 1]
(6ij: Kronecker 6 )

which are c a l l e d the }{eat E q u a t i o n s .

If M is the m o d u l i space of c u r v e s of g e n u s g then the m a p

g
M ~ A defined by C + Jac(C) is i n j e c t i v e . The closure of the image
g g
in Ag (or Ag) is c a l l e d the J a c o b i a n locus. Notation : Jg.

APPROACH 1 : ALGEBRAIC EQUATIONS.

This is S c h o t t k y ' s original approach for c h a r a c t e r i z i n g the J a c o -

bian locus. It is b a s e d on the c o n s t r u c t i o n of P r y m v a r i e t i e s . F o r an

excellent treatment of P r y m v a r i e t i e s , see M u m f o r d [13].

Suppose we start with a curve C of genus g and a non-zero point

of o r d e r 2 on J : Jac(C) . This determines an u n r a m i f i e d covering

 388

~: C ÷ C of d e g r e e 2 and an induced map Nm : J = Jac(C) ÷ J and

gives us a diagram

¢ t $ 6 = mm (2)
J ~ J
21

where % = ~*, - denotes transpose and ~,I are the principal polari-

zations. One defines the Prym variety of ~: C ÷ C as the identity

component of the kernel of Nm : P = (ker Nm) O, It is a n a b e l i a n

variety of d i m e n s i o n g. Mumford showed that from a diagram (2) it

follows that there exist a symplectic isomorphism Hl/{0'n} + P2 with

HI = { ~ (J2 : e2(~'n) = i} (e2: W e i l - p a i r i n g ) such that

= J x p / { (~,~ (~)) : ~ ~ H I}
2~ 0)
Let o: Jxp + J be the natural isogeny. Then the polarization (0 p

is t h e pull back under ~ of the polarization ~ and this implies

that p is t w i c e a principal polarization. So P carries a principal

polarization ~ For these facts, see [13],§2.

Now use the elementary

(i.i) Lemma. If D is a d i v i s o r of degree g-I on C, then

h°(~*(D)) # 0 if a n d only if h°(D) # 0 or h°(D+ n) # 0.

One finds (using that for Jacobians the theta divisor in J a c g-I con-

sists of the effective divisor classes of degree g-l)

0 - 1 ( 0 o ) n (Jac(C) x (0)) = 0 + @
0 O,~

Here ~o (resp. Oo ) denotes the theta divisor on Jac g-l(C) (resp.

Jac2g-2(C), i.e, ® = {x { J a c g - i (C) : h°(x) > 0} If o n e now chooses

o
(Jac(C) 4 such that 2~ = T) a n d a theta characteristic ~ on C,

then ~ = -i(~ +~) is a t h e t a characteristic on C. If 0 =

{x c J : h°(x÷ ~) > 0} , 0= {x e: J : h°(x+ ~) > 0 } are the theta divisors

on J and on J then (since 01Jx(0) = ~*)-

(~.)-i~ = 0 + e -~ . (3)
 389

(1.2) The link b e t w e e n the K u m m e r variety of P and that of J is

obtained as follows. There is a m o r p h i s m

6: P ~ 12oji
p + (~*) -i ( ~ j ,-p) -

Mumford shows in [MI] that 6 is the u s u a l K u m m e r m a p f o l l o w e d by an

inclusion

~p/ ~ (Ho (L~2) v)

p /

~ 12%1

For any p r i n c i p a l l y polarized abelian variety X the R i e m a n n theta

formula

0(u+v)0(u-v) : Z c~Bs~(u)s~(v)

with 8 a non-zero section of LX and {s } a basis of F(X,Lx2)

e2
gives us a n o n - d e g e n e r a t e form B on F (X,L X ) via the (caB) and

gives rise to a d i a g r a m

~X ®2 v)
~ ]P(H°(X,Lx )
X ~ ~ ~ B '
~ 12°x[
where %i(x) = 0X, x + O X , _ x and B' is i n d u c e d by B.

Formula (3) thus implies the f u n d a m e n t a l relation

i(¢p(0)) = B' (¢X(O,)). (4)

(1.3) For any i n d e c o m p o s a b l e principally polarized X the t h e t a g r o u p

G ( L ~ 2) acts on F ( X , L ~ 2) and this d e f i n e s an a c t i o n of G ( L ~ 2) m o d u l o

scalars ~ X2 on ~(F(X,L~2)v). If ~ ~ X 2, ~ # 0 , then ~ defines

a projective involution i of ~N with

i (%x(X)) = }x(X+~) . (5)

It is a c l a s s i c a l fact t h a t the i n v o l u t i o n i on ~N= ~(F(X,L~2)v))

has as its f i x e d p o i n t set two linear subspaces V + , V[ , e a c h of d i m e n -

 390

sion 2g-l-i and each intersecting the Kummer variety of X in

22 (g-l) points; moreover,

(v+~v~)n~x(X) =~x({X~X4: 2x=~ }).

The linear spaces V ±e cut out on the m o d u l a r variety %(Ag(2,4))

the b o u n d a r y components. To be precise,

(1.4) P r o p o s i t i o n . Let Ag_l(2,4) be one of the 2(22g-i) boundary

components of Ag(2,4) of m a x i m a l dimension. The image #(Ag_l(2,4))

~2g-I
in is the i n t e r s e c t i o n of }(Ag(2,4)) w i t h one of the linear
+
spaces V-.

It follows from (4) and p r o p o s i t i o n (1.4) that for a J a c o b i a n X

the i n t e r s e c t i o n }x(X) n #(Ag_l(2,4)) is not empty (Here w e v i e w

Ag_l(2,4) as a b o u n d a r y component of Ag(2,4).) : the i n t e r s e c t i o n

contains the image of a point of order 4 of X.

(1.5) Definition. The S c h o t t k y locus S &X is the s m a l l e s t closed

g g
subset of A containing the points IX] with X indecomposable for
g
which ~x(X) n ~(Ag_l(2,4)) ¢ ~ for a l l boundary components Ag_l(2,4).

By c o n s t r u c t i o n S contains J , the J a c o b i a n locus (cf. (4)).

g g
S
g can be d e s c r i b e d in terms of theta c o n s t a n t s as well. The p o i n t is

that P can be w r i t t e n as

p = {g-i/~ g-l+ Pg_l~g-i for some Pg-i e ~ g - i

and that after suitable normalizations

02 [ s ,](Pg_l,0) = cSE s , ] (~g,0) 8 [ s ,l](Tg, 0) (6)

with a constant c ~ ~* independent of e,~' ~ (Tz./2Z~)g-I Thus (6)

is a t r a n s l a t i o n of (4).

Let

Tg c~[ X[ ~'e] : e,s' c (Tz/2[Z) g, t , = 0 ]

be the ideal of ~(Ag(2,4)). To an e l e m e n t f ~ Tg_l we associate

 391

o(f) = f( .... 8[ Es, ~38[~'

I s 0] (~g,0) ,...)

by s u b s t i t u t i n g 8[~,0]
8 0 [~, 01] (Tg,0) for X ~,] The g r o u p Fg/Fg(4,8)

acts on ~[ 8[ a ,] (~g,0) : t a' = 0]. Let Zg be the s m a l l e s t

Fg/Fg(4,8)-invariant ideal of this ring c o n t a i n i n g all a(f) with f

in Tg_l. Then Sg is the z e r o - l o c u s of Zg in Ag.

Of c o u r s e , this d e s c r i p t i o n is e x p l i c i t only if w e k n o w Tg_l and

in g e n e r a l the s t r u c t u r e of this ideal is not known.

For g=4 o n e finds that ~ is the ideal g e n e r a t e d by a S i e g e l

g
modular f o r m of w e i g h t 8 as S c h o t t k y showed.

The important question about S is w h e t h e r S = J and if not,

g g g
what the c o m p o n e n t s of Sg are. For g=4 Igusa p r o v e d that S4 is

irreducible. This implies $4 = J4" Recently van Geemen proved

(1.6) T h e o r e m . (van G e e m e n [ 6 ]) J g is an i r r e d u c i b l e component of Sg .

His p r o o f uses an i n d u c t i o n argument and an a n a l y s i s of the inter-

s e c t i o n of the S c h o t t k y locus w i t h b l o w - u p of a b o u n d a r y component of

(4,8).
g
It is a r e c u r r i n g phenomenon in the h i s t o r y of the S c h o t t k y problem

that o n e finds a l g e b r a i c subsets of A that c o n t a i n J as as irred-

g g
ucible component but that m a y h a v e o t h e r components as well. Another

example is the A n d r e o t t i - M a y e r approach. Since it is k n o w n that for a

Jacobian one has d i m S i n g 0 > g-4 one looks at

N m = {[X] e Ag : Sing@ # ~, d i m S i n g @->m }

g

Andreotti and M a y e r proved that J is an i r r e d u c i b l e component of

g
N g-4 , g >_4. However, N g-4 contains other components.
g g
 392

APPROACH 2 : TRISECANTSo

One of the r e m a r k a b l e features of J a c o b i a n s is that their Kummer

varieties possess trisecants :

(2.1) Proposition. Let C be a n o n - s i n g u l a r curve and let a,b,c,d be

points of C. If r { X=Jac(C) is such that 2r = a + b - c - d , then

%x(r), Cx(r-b+c) and %x(r-b+d) are collinear.

Fay's trisecant identity [ 3 ] implies this fact. Gunning [ 9 ]

has g e n e r a l i z e d this identity. The idea b e h i n d it is e s s e n t i a l l y the

following.

Let N be a line bundle on XxX such that

N]xxt ~ T_[(0(28)) ( Tt : translation by t).

Fix a point p of C. This defines ¢: C + X = Jac(C) by c ~c-p.

Let A be a d i v i s o r of d e g r e e g on C such that 9*0(@) ~ 0(£). We

let M be the v e c t o r bundle on X whose fibre at t is

H°(C,0(2A+2t)). Pull back of sections via

H° (X,T_< (0 (20)) ~ H° (C,0 (2£+2t)) (7)

gives rise to a b u n d l e map

~: (pl),N --~ M.

(2.2) Lemma. The m a p ~ is s u r j e c t i v e .

Proof. The m a p H°(X,T_t0(0)) + H°(C,0(A+t)) is s u r j e c t i v e if the

divisor A+t is n o n - s p e c i a l . Therefore, if D { 12A+tl c a n be w r i t t e n

as D = DI+ D 2 , DiEiA+tii with £+t i non-special, then D is the

zero d i v i s o r of a s e c t i o n in the image of (7). Define a non-empty

open set in the s y m m e t r i c product C (g) by (K = c a n o n i c a l divisor)

U = {Zl+...+Zg 6 C (g) : h ° (D- ~ zi)=l, h ° ( < - [ zi)=0 } .

If D = Zl+...+Zg+Z{+...+z' with [z i { U, then h°( [ z i) = i, h e n c e

g
zi and ~ z! are b o t h non-special. This shows that the image of
1
 393

P(H°(X,T*t(0(2@)) + IP ( H ° ( C , 0 ( 2 A + t ) ) ) contains a non-empty open set.

We put as u s u a l

W dr = { x { Sac(d) (C) • h °(x) >_ r+l }.

(2.3) Theorem. (Gunning [i0]) If Zl,...,z n are distinct points of C,

then
n
n-~ + n <
Wn 2 - ~ z. + 2p = {2t ~ Sac(C) : rank @2(t+~(zi))i:] ~ }
- i=l l

Proof. By the lemma, the rank of this 2gxn-matrix is less than n ¢=~

h ° ( 2 A + 2 t - ~ z i) > g + l - ~ ~=~ 2 t + 2 £ - ~z i c W 2gg+_ln- ~ Applying Serre duality

g+l-~
K - W2g_n : W~_~ and the fact that ~(K)=¢(2A) gives the result.

The special case n=~=3 gives proposition (2.1).

We can generalize this by allowing the points zi to c o i n c i d e . If

Zl+...+z n = m]xl+...+meX e with x i # x 3. if i # j then in the rank

condition the mj vectors @2(t+%(xj)) have to be replaced by

O ~ t + ~ (xj)) AlO2(t+ ~ (xj)) ... A m _l@2(t+~ (xj)) ,

3

where the £k are differential operators defined as follows. The curve

#(C) c o n t a i n s at %(xj) an a r t i n i a n subscheme Spec C[c]/(Emj) and this

is g i v e n by a local homomorphism

{[c]/(smJ)
0X,%(xj)

f f(yj) + AIf(Yj)E + ... + ~m _ i f ( Y j ) ~ m J -I , y:% (xj)

3

The special case n=~=3 is important since it g i v e s us b a c k the

curve C : Note that TW o1 ~ C and

3
WlO _ ~ zi + 2p = {2t { S a c ( C ) : rank (~2(t+%(zi)) (t) < 2 } '
i=l

Gunning's idea in F 8 ] was to u s e this property to c h a r a c t e r i z e

Jacobians. Gunning used distinct points z. but Welters has infini-

l
tesimalized Gunning's case to include the case of coinciding points and

transformed it into the following beautiful criterion :

 394

(2.4) Theorem. (Gunning-Welters [19]) Let X be an i n d e c o m p o s a b l e prin-

cipally p o l a r i z e d abelian v a r i e t y and let y cX be an a r t i n i a n sub-

scheme of length 3. Assume that

~ N }.
V = {2t c X : t + Y c #~i(/) for some line lc
A

has p o s i t i v e d i m e n s i o n at some point. Then V is a smooth i r r e d u c i b l e

curve and X is its Jacobian.

v
(2.5) The property of having flexes is closely related to the Kadomcev-

Petviashvili equation (K-P-equation) ,a fourth order partial d i f f e r e n t -

ial equation satisfied by the theta functions of Jacobians. In [16 ]

Mumford noticed that if the points a,b,c,d in p r o p o s i t i o n (2.1) coin-

cide, Fay's trisecant identity leads to the K-P-equation.

To get the link, note that an inclusion Spec ~[~]/(a N+I) ÷(X,0)

is given by a local h o m o m o r p h i s m

CEa]/(c N+I )
0X'0 N
i
f + [ A i (f)
i=l

where the 4. are differential operators satisfying

1

A = id ,
O Ai(gh) = k+~:iAk(g)~/(h)

One can show that this is equivalent to the existence of translation

invariant vector fields DI,...,D N on X such that

' ' -l-hl .D

hv
A = ~ (h t .... h.) b 1 .. v ,
hl+2h2+...+~h > 0

or formally

eJ~imj ~j
~ Ak ak (mod aN+l).
k=0

We apply this to c r i t e r i o n (2.4). Note that V is defined by the van-

ishing of the 3x3 minors f~ , ~ ~ ( ~ g / 2 ~ g ) 3, of (~+ AI%÷ A2%)÷ at

some point. If we assume that this point is the origin and that Y :
 395

Spec~[c] /(E 3) ~ (X,0) is g i v e n by DI,D 2 one finds (using the fact

+ +
t h a t the r a n k of (02 (~iSj @2) i, j) equals g(g+l)/2+ 1 at (~,0) ; ~i =

~/$zi)) that (V) 2=y. Then, as W e l t e r s noticed, o n e has

13+
(V) 3 = S p e c ~ [ ~ ] / ( ~ 4) ~=~ 3 m 3 such that (~m I D I D 2 + D 3 ) f =0 (all v)

~=~ rank(( 1 D1
24 DI+3
D - 3DID3)_2) (~,0) ~ 2.

Without changing Spec { E ~ ] / ( E 4) ~ (X,0) one m a y effect the c h a n g e

D 1 ÷ aD I, D 2 ÷ a 2 D 2 + b D I, D 3 ÷ a 3 D 3 + a 2 b D 2 + c D I, a # 0,b,c, h e n c e we c a n re-

write this as

32 ÷
((D - DID3 + ~D 2 + d) ~ ) (~,0) = 0 (8)

This is the K-P e q u a t i o n . By (2.3) the t h e t a functions @2[6] of a

Jacobian yield solutions. (Usually, the K-P e q u a t i o n is w r i t t e n

U y y + ( U t + U x x x + U U x ) x = 0. It is s a t i s f i e d on a J a c o b i a n by
2
U = D l l O g 8 ( z + x a 1 + Y a 2 + t a 3) + c for some a l , a 2 , a 3 c c g , c ~ {, see [15 ].

Dubrovin formulated the e q u i v a l e n t form (8).) That theta functions

yield solutions was noticed by K r i c h e v e r , who arrived at it in a com-

pletely different way. Novikov conjectured t h e n t h a t this s h o u l d c h a r -

acterize Jacobians :

(2.6) N o v i k o v ' s Conjecture. An i n d e c o m p o s a b l e principally polarized

abelian variety X is a J a c o b i a n if and o n l y if t h e r e e x i s t c o n s t a n t

vector fields DI,D2,D 3 on X and a c o n s t a n t d such that

32 ÷
((D - DID 3 + ~D 2 + d ) 82 ) (T,0) = 0. (9)

Dubrovin proved in [2] that the locus of IX] in A for w h i c h

g
(9) h o l d s for some DI,D2,D 3 and d contains the J a c o b i a n locus as

an i r r e d u c i b l e component.

(2.7) Soon after a weaker version of (2.4) had a p p e a r e d Arbarello and

De C o n c i n i realized that one d o e s not need the p o s i t i v e dimensionality

of V ,but o n l y the fact that 0V, 0 contains an a r t i n i a n s u b s c h e m e of

 396

sufficiently big length, i.e. the c o n d i t i o n is t h a t there exist constant

vector fields DI,..,D M for s o m e b i g M such that

M
X nj~J
eJ:l
(02A AI~2A £2~2) ~ 0 (mod s M) at (~,0).

In t h i s w a y they were the first to w r i t e down equations that characte-

rize the Jacobian locus, see [ i ]. Using the version of (2.4) given

here one can take M= 6gg~ + I.

Recently, Shiota showed that if o n e m a k e s a minor technical assum-

ption on X then Novikov's conjecture is true, see section 4.

APPROACH 3 : THE GEOMETRY OF T H E M O D U L I SPACE.

The approach here, worked out in j o i n t w o r k w i t h van Geemen [ 6 ],

is b a s e d on the observation that under # and ~X both the moduli

space A (2,4) a n d t h e Kunumer v a r i e t y of X are mapped to the same

g
projective space, so t h a t w e can compare their positions in t h i s space.

It w a s m o t i v a t e d by the special case g:2 studied in [ 7 ] and a

paper of F r o b e n i u s dealing with g:3,[ 4 ].

(3.1) We first l o o k at t h e tangent space to %(Ag(2,4)) at %([X]).

i.e. we look at t h e h y p e r p l a n e s

[ ~ 02[a](~,0) = 0 ( x = x )T (10)
O
satisfying

~ij ( ~ ~a@2[a])(T,0) = 0 for all i,j.

By a p p l y i n g the Heat Equations this is transformed into

;2
( [ ~a ~z.?z. 92[a]) (T,0) = 0. (ii)
a l 3
®2
So let us look at the sections of F(X,L x ) satisfying (10) and (ii),

i.e. define

F o o ( X , L x 2) = { s {r(X,Lx2) : mo(S) > 4} .

with m° the multiplicity of a s e c t i o n at zero. Note that for s ~ 0

 397

m (s) is e v e n ) . If X is indecomposable, then

o
~2 2
rk
÷
( 02
÷ 8
3 z l ~ z l @ 2 ' ' ' ' ' 3 z g ~z g
°02
)

in (T,0) equals ~g(g+l) + i, so the codimension of Foo in F ( X , L ~ 2)

equals ½g(g+l) + I. If ¢(Ag(2,4)) is n o n - s i n g u l a r at ¢([X]) and if

Ig c ~ [ . . . , X o , . . . ] is t h e ideal of ¢(Ag(2,4)), then F o o ( X , L ~ 2) eauals

( [ ~~f
0 (.... 0 2 [ 0 ] (~,O) .... ) @ 2 [ o ] (~,z) : f ~ Ig } .

(3.2) AS an example we take g=3. The theory of theta functions gives us

a relation

-000 000 000 000 ~001 , ~001 ~ 001 001

81000181100101010181110 ] 6 ~ O 0 0 J e t l O O] [ O l O ] e [ l l O ] +
o000~ r000 ~000 00O
0 001JSll01]SL0!l]e[lll] : 0

between the 8[s, ] (~,0). We write this as r I- r 2 - r 3 = 0 This implies

the relation

4 4 4 22 22 22
r I + r 2 + r 3 - 2rlr 2 - 2rlr 3 - 2r2r 3 : 0

between the squares of the even thetas. Using (i) this gives an equa-

tion

F( .... @ 2 [ o ] (t ,0) .... ) = 0

of d e g r e e 16 defining a hypersurface in ]p7. Hence

¢= ~ ~F
(" "" /92[°] (T'0) .... ) @ 2 [ o ] (<,z)
O

belongs to Poo and one can check that for indecomposable X~ it is

non-zero. It g e n e r a t e s Foo. In fact, when expressed in t h e theta

squares this is t h e function studied by Frobenius in [ 4 ].

The first question about F is its zero locus. Define

o@
®2
FX = { x { X : s(x) = 0 for all s e Foo(X,L x ) } .

(3.3) Proposition. If X=Jac(C) then F x n { (x-y) { J a c ( C ) : x , y ~ C].

 398

Proof. Use (2.1) and put a=b, c=d there. One finds a relation

~ 2 (a-b) = ~2 (0) + ~ ~ij~i~j~2(0).

(3.4) For a Jacobian one can use the geometry of C to c o n s t r u c t ele-

ments of Foo. Let 1201o O = { D c 1201 : mo(D) a 4 } . If x c Sing 0

then @x u @ - x { 12@Ioo" Define

@ = { ~ (Jacg-l(c) : h°(~) > 0 }

o

Sing @o = { ~ ~ 0o : h°(~) > I }

and define for ~ { J a c g - I (C) :

0 = {x e J a c ( C ) : e-x e@ } .
o

Then obviously,

F X ~_ n (e uG ). ( K: c a n o n i c a l divisor)
C~ K--C~
c Sing 0

If C is h y p e r e l l i p t i c then

1 W o
Sing 00 = g2 + g-3 '

hence if f e Fx one has +f- + g~ + W°g-3 c W g,_°l so + f + g21 c W 2° and

this implies f = (a-b) for some a , b ~ C. So for hyperelliptic C one

finds

Fx = { (x-y) ~ J a c ( C ) : x,y c C } .

By semi-continuity it f o l l o w s that for general X dim FX s 2 and for a

general Jacobian dim F x= 2. We conjectured in [6 ] that for every

Fjac(C) = { (x-y) c J a c ( C ) : x,y c C }

and provided a l o t of evidence for it. Independently, the conjecture

was formulated by Mumford [15] (in a d u a l form) and by Gunning [i@ •

This conjecture has now been proved by Welters. However, there is o n e

exceptional case, namely g=4, where

Fjac(C) -- {(x-y) ( J a c ( C ) : x,y ~ C } u {+(f-f') } ,

 399

with f,f' the two g~ 's on C, see [20]

(3.5) Conjecture. Let X be a principally polarized abelian variety of

dimension g ~ 2. Then X is a Jacobian if and only if dim F X a 2.

An infinitesimal form of this conjecture is related to the Novikov

conjecture. Note that

x E FX ~=~ B l'~ij ~ { such that G2(x) =I@2(0) + [ ~ij$i~j ~2(0)

Now, if Y=Spec ~[c]/(s N+I) is contained in X at 0 via the local

N
homomorphism 0X, 0 ~ Spec ~ [ ~ ] / ( N + I ) , f ÷ ~ Ai(f)~i, we have
i=0

Y c FX •=~ B l,~ij c {[El such that

N
Ak ~ 2 ( T , 0 ) c k = ~2(~,0) + ~ ~ij?i~j~2(~,0).
k=0

Working out the condition for N=4 gives

1 4 1 2 ÷ ÷
(( ~-~DI + ~ D 2 - DID 3) @ 2 ) ( ~ , 0 ) = d ~ (~,0) + ~ eij~i~j@2(~,0) ,

+
where eij is the coefficient of 4 in ~ij" If ~ eij~i~j@2(~,0)
2+
is a multiple of DI@2(~,0), then we can change coordinates such that

this relation becomes the K-P equation.

Gunning has studied the generalizations of Fay's identity. This

leads to interesting analogues of (3.3) involving higher derivatives at

zero, cf [ii]

Instead of intersecting the Kummer variety with the tangent space

of the moduli space we can also intersect the Kummer variety with the

moduli space itself. As an analogue to (3.3) we find

(3.6) Proposition. If X = Jac(C) then Cx(X) n #(Ag(2,4)) contains

~X({¼(x-y) : x , y c C }).

Here ~1 means the inverse image under m u l t i p l i c a t i o n by 4.

 400

Proof. A divisor class a with 2a = x + y defines a (2:l)-covering

T: C ~ C. The Prym variety P : k e r ( Nm: Jac(C) ~ Jac(C) } is a p r i n -

cipally polarized abelian variety of dimension g for general x , y c C.

There exist theta structures on P and X such that

1
%(~(x-y)) = O([P]),

1 1 1
cf. [13],p.340, where ~(x-y) 6 Jac(C) is s u c h that 2(~(x-y)) =~(x-a) .
1
By symmetry it t h e n follows that all of ~ X ( ~ (x-y)) lies in ~ (Ag(2,4))

for general x,y, hence for all x,y.

We made two conjectures in r e l a t i o n to this. First, for a Jacobian

we conjectured that

~x(X) n ~ ( A g ( 2 , 4 ) ) : ~X ( { ~ < X : 4~ : x - y , x,y < C )

and we proved this for g=3. Secondly, we hope that this characterizes

Jacobians :

(3.7) Conjecture. Let X be an indecomposable principally polarized

abelian variety of d i m e n s i o n g > 2. Then X is a J a c o b i a n if a n d only

if dim ~x(X) n % ( A g ( 2 , 4 ) _> 2 .

(3.8) The preceding sections suggest to look at the morphism

-:: F (2,4) ~ ~ 2 g \ ] H g × ~ g : U (2,4) ~ ]pN

g g

(~,z) ÷ ( .... @ 2 [ a ] ( r , z ) .... )

Is it e v e r y w h e r e of maximal rank ? Since the Kummer variety of an inde-

composable X is singular at the images of the points of o r d e r 2 of

X the rank is c e r t a i n l y not maximal at those (T,z) for which 2z

zzg+ ~ZZ g . Using the Heat Equations the question becomes whether the

rank of

( ~ i ~.3 ~ 2 1 ~i~ j -<g ~k~2 1 <k<g 02 )

at (7,z) is m a x i m a l .
 401

Suppose that 2 g a lg(g+l) + g + 1 i.e. g > 4 and that X = X is

2 i - T

indecomposable. If there exist a relation

(( ~ ~ i j 3 i S j + [ 6k8 k + 7 )02) (<, z) : 0

for z such that 2z i ~g+ T~ g then Cx(X) possesses a flex at

~x(Z).
A Jacobian is known to possess a lot of such flexes : if X =

Jac(C), then applying (2.3) with Zl:Z2:Z3= p we find that all points
1
of { ~(x-y) E X : x,y ~ C } are flex points. Hence the rank of z is

not maximal at these points.

(3.9) Question. In v i e w of (2.4) we can ask whether for an indecompo-

sable Jacobian X= X with g ~ 4 the only points (T,z) where Z

%

1
is not of maximal rank are those corresponding to {~(x-y) : x,y< C }

and whether one could use this to c h a r a c t e r i z e Jacobians.

APPROACH 4 : RINGS OF DIFFERENTIAL OPERATORS

As mentioned above Shiota has settled Novikov's conjecture up to

a technical assumption.

(4.1) Theorem. (Shiota [17]) An indecomposable principally polarized

abelian variety X of dimension g is the Jacobian of a complete

smooth non-singular curve C over ¢ of genus g if and only if

+
i) the vector @2(~,z) satisfies the K-P equation (8) for some

DI,D2,D 3 and d, and

ii) no translate of the theta divisor of X contains an abelian

subvariety of X which is tangent to DI(0).

Shiota's approach incorporates ideas of Mulase and is b a s e d on

Krichever's dictionary. Let D be the non-commutative ring which as

an additive group equals ¢[[x]][~] with ~= d__ and with multipli-

dx
 402

cation such that

~.f = f ~ + f' for f c {[[x]][$]. (i2)

If R is a commutative subring of D containing • and two ele-

ments A,B with A= sn+ ... (... = lower order terms), B = ~m+ ... w i t h

(n,m) = i, then any element of R can be written as C = ~r + ... with

c { , r c ~a0 (Proof: work out the commutator [A,C] ). If R n=

{ C ~R : C = e ~ r + ... with r s n one has dim Rn/Rn-i s 1 and =I

for n >> 0.

(4.2) Theorem. (Krichever) There is a natural bijection between the

following two sets of data:

I) C an irreducible curve, P a smooth point of C, a tangent vector

at P and a torsion free rank 1 0c-mOdule F with h°(F) = hl(F) = 1 .

2) R c D a commutative subring containing ~ and two elements A,

B as above.

Let us sketch how to go from i) to 2). Choose a neighbourhood U

of P such that the local coordinate z at P is a unit on U-P.

Let x be the standard coordinate on { . We now glue Fe0{ on U x

and F®0~ on (C-P) x { by multiplication with e x/z This defines a

sheaf F* on C × { . If V is a suitably chosen neighbourhood of

0 6 • then Hi(c x V, F*) = 0 i=0,1. Define now

V : F*(ZP.) ~ F*((Z+I)P)

d 1 d
by taking ~ on C-P and z + ~ on U. A non-zero section so

H °(F*(P)) generates H O(F*(P)) as a H O(V,0~)-mOdule. We normalize

so such that so = 1 + O(z) at p ×V , i.e. ds

~ O = (z-l+o(z))So . Put

s n= vns. The sections s o .... ,s n generate H°(F * (n+l)P) .

If a ~ F(C-P,0c) then aSo e H°(F*(nP)), hence

n-I
as O = ~ a i(x) v l s .
i=0 o
 403

This gives us a m a p

F (C-P,0 C) ~ D
n-i
a ÷ ~ ai(x) ~i.
i=0

The image is a c o m m u t a t i v e subring R of D.

(4.3) In o r d e r to o b t a i n Jacobians one observes that F defines a

point of Jac(C). So let us d e f o r m F. Choose variables tl,...,t N

and consider instead of F®0~ now F®0 N on U × ~N and (C-P) × ~N

N
and glue n o w by exp( [ t.z-J). We introduce formally a variable x
j--i ]
by r e p l a c i n g tI by tl+x. This now gives us F* as above. Define

V as a b o v e and d e f i n e

F* F*(nP)
n

by t a k i n g ~ on C-P. We c h o o s e a normalized so again as above.

n
We n o w o b t a i n

F (C-P,0 C) ~ D

and the image is a c o m m u t a t i v e subring Rt depending on t=(tl,...,tn).

The question arises : h o w does Rt deform with t ? If T denotes the

the tangent space we get a map

~: T t ~ N ~ D

+ B (t)
~t n
n

where Bn(t) is d e f i n e d as follows. By the n o r m a l i z a t i o n ~ sO =

4 n

( z - n + o ( z ) ) s O, so ~nSo = [ b i V l s o • We put B n (t) = [ bi ~i.

We need some notation. Let ~ be the n o n - c o m m u t a t i v e ~-algebra

d -i
whose elements are formal Laurent series in (~-~) with coefficients

from ~ E [ x ] ] ® 0~N ° The m u l t i p l i c a t i v e structure is d e f i n e d by extending

the r u l e (12). Let

'9 = {P c ~ : ordP _<-i } .

-i
So the e l e m e n t s of ~ are expressions a i ( x ) ~i , where we sup-
 404

press the dependence of t in the notation. By extending the map

F ( C - P , 0 c) ~ D to Q ( F (C-P,0c) ~ ~ ( Q : quotient field) we see that

I/z corresponds to an element of + ~ which we call L. From the

5 -
normalization we obtain B n= (Ln)+ , where ( )+ means taking the

differential operator part (non-negative powers of 9 ).

The dependence of Rt on t is now expressed by the following

d
deformation equations for L c ~x + ~ :

(~)L= [ (Ln)+,L] n = l .... ,N

n

Take now infinitely many variables tl,t2,... , i.e. te{ =

d -
lim÷ ~N and consider the equations for L ~ + ~ :

(~-) L : [(Ln)+,L] n = l , 2 ....

n

This set of equations is called the K-P hierarchy. We do not explain

here the translation of solutions to this hierarchy of equations into

differential equations satisfied by theta functions, but we refer to

Shiota's paper and the references there.

If L is a solution to the K-P hierarchy then consider

dL : Tt{ ~ Cn~ t + [[ C n B n , L ] ,
n

the tangent map of the map t ÷ L(t) at t. We call L a finite

dimensional solution if dL is of finite rank. Shiota considers for

a finite dimensional L

RL = ¢(ker dL) •

with ¢: To{ ~ D, ~--~- + B n. He proves

n

RL = {P ~ D : [P,L] = 0}

and that RL is a m a x i m a l commutative subring of D if R L ~ ~ . Thus

a finite dimensional solution to the K-P hierarchy yields a curve by

(4.2). Moreover, it turns out that ker dL can be identified with the
 405

tangent space of t h e J a c o b i a n of this curve at a c e r t a i n point.

Basically, this is t h e w a y Mulase arrived at his theorem which

states that the w h o l e K-P-hierarchy characterizes Jacobians. Both

Mulase and Shiota then noticed that in f a c t finitely many equations

f r o m this hierarchy suffice,arriving thus at a t h e o r e m very similar to

the r e s u l t of A r b a r e l l o and De C o n c i n i . Shiota then continued by show-

ingthat under condition 2) of (4.1) one can e x t e n d a solution to the

K-P-equation (the f i r s t of the K - P - h i e r a r c h y ) to a s o l u t i o n of the w h o l e

hierarchy, see [17]. Namikawa informed me that Mulase now a l s o obtain-

ed s u c h a reduction.

A FINAL REMARK.

Our summaryof recent attacks o n the S c h o t t k y problem is n o t intended to

be complete. O n e of the a p p r o a c h e s that should be mentioned also is

the approach that uses the r e d u c u b i l i t y Of @ n0 a It is c l o s e l y re-

lated to a p p r o a c h 2 and was suggested by Mumford in [14]. For a Jaeobian

X with theta divisor 0 o n e has : if x ~X and x# 0 then there exist

u,v in X with {0,x] n {u,v} : ~ such that 0 n 0 c @ u @ if a n d o n l y

X U V

if x belongs to { (a-b)c X=Jac(C) : a , b ~ C} (Note t h a t o n e implica-

tion follows from (2.1) by u s i n g X~ ~N , x÷ @x u @ - x ~ 1281 .) W e l t e r s

proved the following theorem : Let X be a complex principally polari-

zed a b e l i a n variety of d i m e n s i o n g. A s s u m e i) d i m S i n g @ s g-4, 2) t h e r e

exist a one-dimensional subset y c X such that for g e n e r i c y { Y one

has : @n@y c @u n @v for s o m e u,v c X with {0,y} n {u,v} = ~. Then

X is the p o l a r i z e d Jacobian of a n o n - h y p e r - e l l i p t i c curve, see [21]

References.

Arbarello, E.,De Concini, C.: O n a set of e q u a t i o n s characterizing

Riemann matrices. Preprint 1983.
Dubrovin, B.A.: theta functions and non-linear equations. Uspekhi
Mat. Nauk. 36:2 (1981),11-80 = Russian math. surveys 36,(1981),11.
 406

3 Fay, J.D.: Theta functions on Riemann surfaces. Lecture Notes in

Math. 352. Springer Verlag,Berlin etc. 1973.
4 Frobenius, F.: Uber die Jacobischen Functionen dreier Variabelen.
Journal f~r d i e reine und angewandte Mathematik 105(1889),35-100.
5 van Geemen, B.: Siegel modular forms vanishing on the moduli space
of c u r v e s . Preprint Univ. of U t r e c h t , to a p p e a r in I n v e n t . Math.
6 van Geemen, B.,van der Geer, G.: Kummer varieties and the m o d u l i
spaces of a b e l i a n varieties. Preprint Univ. of U t r e c h t 1983.
7 van der Geer, G.: O n the g e o m e t r y of a S i e g e l modular threefold.
Math. Annalen 260 (1982),317-350
8 Gunning, R.C.: Some curves in a b e l i a n varieties. Invent. Math. 66
(1982),377-389.
9 Gunning, R.C.: On generalized theta functions. ~er. J. M a t h . 104
(1982),183-208.
i0 Gunning, R.C.: Riemann surfaces and their associated Wirtinger
varieties. Preprint 1983.
ii Gunning, R.C.: Some identities for a b e l i a n integrals. Preprint 1983.
12 Mulase, M.: Cohomological structure of s o l i t o n equations, isospec-
tral deformations of o r d i n a r y differential operators and a charac-
terization of Jacobian varieties. Preprint MSRI 003-84.
13 Mumford, D.: Prym varieties I. In: Contributions to A n a l y s i s , p.
325-350. Academic press, London, New-York 1974.
14 Mumford, D. : C u r v e s and their Jacobians. Univ. of M i c h i g a n Press,
Ann Arbor 1975.
15 Mumford, D.: Tata lectures on t h e t a II, to a p p e a r .
16 Mumford, D.,Fogarty, J.: Geometric invariant theory. Ergebnisse
der Mathematik 34, Springer Verlag 1982.
17 Shiota, T.: Characterization of J a c o b i a n varieties in t e r m s of
soliton equations. Preprint 1984.
18 W e l t e r s , G. : O n t h e flexes of the K u m m e r variety. Preprint 1983.
19 Welters, G. : A c r i t e r i o n for J a c o b i varieties. Preprint 1983.
20 Welters, G. : P r e p r i n t 1984.
21 Welters, G.: A characterization of n o n - h y p e r e l l i p t i c Jacobi varie-
ties. Invent. Math. 71 (1983),437-440.

Gerard van der Geer

Mathematisch Instituut
Universiteit van ~sterdam
Reetersstraat 15
1018 W B A m s t e r d a m .
 VOJTA'S CONJECTURE

Serge Lang
Department of Mathematics
Yale University
New Haven, CT 06520
U.S.A.

§ I. Nevanlinna theory

Let f : ~d > X be a holomorphic map, where X is a complex non-

singular variety of dimension d . Let D be an effective divisor on
X , with associated invertible sheaf £ . Let s be a meromorphic sec-
tion of £ , with divisor (s) = D . We suppose that f is non-dege-
nerate, in the sense that its Jacobian is n o t zero somewhere. For po-
sitive real r we define

m(D,r,f) : S -logl f * s I 2 @ .
B d B (r)

where o is the natural normalized differential form invariant under

rotations giving spheres area I . When d = I , then ~ = de/2~ . Ac-
tually, m(D,r,f) should be written m(s,r,f) , but two sections with
the same divisor differ by multiplication with a constant, so m(s,r,f)
is determined modulo an additive constant. One can select this constant
such that m(s,r,f) ~ 0 , so by abuse of notation, we shall also write
m(D,r,f) ~ 0 .
We also define

N(D,r,d) = normalized measure of the analytic divisor in t h e ball

of radius r whose image under f is c o n t a i n e d in D ;
(Cf. Griffiths [Gr] for the normalization.)

r
N(D,r,f) = S [N(D,r,f)-N(D,0,f)] dr + N(D,0,f)log r .
0 r

T(D,r,f) = m(D,r,f) + N(D,r,f)

 408

Remark. If d = I then N(D,r,f) = n(D,r,f) is the number of points

in the disc of radius r whose image under f lies in D

One formulation of the FIRST MAIN THEOREM (FMT) of Nevanlinna the-

ory runs as follows. The func~kon T(D,r,f) depends o n l y on t h e l i n e a r
e q u i v a l e n c e c~ass o f D , moduZo bounded funet%o~s O(I)

The first main theorem is relatively easy to p r o v e . More important is

the SECOND MAIN THEOREM (SMT) , w h i c h we state in t h e following form:

Let D be a d i v i s o r on X w i t h sZmple normal ~ o s s i n g s (SNC, meaning t h a t

the i~reducible componemO~ of D c~e n o n - s i n g u l a r , and i n t e ~ e c t t~amsver-
sally]. Let E be an ample d i v i s o r , and K t h e c a n o n i c a l c l a ~ s . Given s ,
there exists a s e t of f i n i t e mem~ure z(s) such that for r not in this set,

m(D,r,f) + T(K,r,f) ~ ~ T(E,r,f)

This is an improved formulation of the statement as it is given for in-

stance in Griffiths [Gr] , p. 68, formula 3.5.

§ 2. Weil functions

Let X be a projective variety defined over { or { (p-adic complex

P
numbers = completion of the algebraic closure of ~p) . L e t £ be an
invertible sheaf on X and let p be a smooth metric on £ . If s
is a meromorphic section of £ with divisor D , we define the associ-
ated Weil function (also called Green's function)

I(P) = - l o g [ s(P) I for P { supp (D)

If we change the metric or s with the same divisor, I changes by a

bounded smooth function, so is determined mod O(I) . We denote such a
function by 1D . It has the following properties:
The association D I > 1D is a homomorphism mod O(I)
If D = (f) on an open set U (Zariski) then there exists a
smooth function ~ on U such that
 409

ID(P) = -log L f(P) I + 5 .

If D is e f f e c t i v e , then i D ~ -O(I) (agreeing that values of

1D on D are then ~ ).

If v denotes the absolute value on { then we write

v

v(a) = -loglal
v

for any element a 6 {v' sO we can write

~D = v ° f +

In the sequel, metrics will not be used as such; only the associated
Weil functions and the above properties will play a role. Note that
these Well-Green functions need not be harmonic. In some cases, they
may be, for instance in the case of divisors of degree 0 on a curve.
But if the divisor has non-zero degree, then the Green function is not
harmonic.

In the sequel, we shall deal with global objects, and then the Weil
functions and others must be indexed by v , such as ID,v' ~v' etc.

§ 3. Heights (Cf.[La])

Let K be a number field, and let {v} be its set of absolute values
extending either the ordinary absolute value on ~ , or the p-adic ab-
solute values such that Ipl v = I/p . We let Kv be the completion, and
Ka its algebraic closure. Then we have the product formula
v

Z dvv(a) : 0
v

dv
where dv - [ K[K
v :
: ~v]
~] and a 6 K , a ~ 0 . We let llalIv = ialv

Let (x ° ..... X n ) E ~ n ( K ) be a point in p r o j e c t i v e space over K .

We define its height

 410

h(P) : ~ log max II xillv

v i

If K : ~ and xo,...,XnE ~ are relatively prime, then

h(P) : log max Ix il

where the absolute value is t h e ordinary one. From this it is immediate

that there is o n l y a finite number of points of b o u n d e d height and boun-
ded degree.

Let

: X --> ]pn

be a morphism of a projective non-singular variety into projective space.

We define

h (P) = h(~(P)) for P 6 V ( K a)

The basic theorem about heights states:

There e x i s t s a unique homomorphism c t ........ > h

c
Pic(X) - - > functions from X ( K a) to
modulo bounded functions

such t h a t i f D i s very ample, and O(D) : <o*Oip I) , then

H 1. h = h + 0(I)
c <0

In the above statement, we denote by hc any one of the func-

tions in its class mod bounded functions. Similarly, if D lies in
c , we also write hD instead of hc . This height function also satis-
fies the following properties:

H 2. If D is e f f e c t i v e , then h D ~ -O(1)
 411

H 3. If E is a m p l e and D any d i v i s o r , then

h D : O(hE)

In particular, if E I, E 2 are ample, then

hE1 >> << hE2

We are using standard notation concerning orders of m a g n i t u d e . Since

according to our conventions, a given height hE is d e f i n e d only mod
bounded functions, the notation h D = O(h E) or h D << h E means that
there exists a constant C such that for all p o i n t s P with hE(P)
sufficiently large, we h a v e lhm(P) I K ChE (P)

Essential to the e x i s t e n c e and uniqueness of such height functions h

c
is the p r o p e r t y of e l e m e n t a r y algebraic geometry that given any divisor
D , if E is a m p l e , then D + mE is v e r y ample for all m ~ m
o
A fundamental result also states t h a t one can choose metrics Pv "uni-
formly" such that

h D = E dv ID,v + O(I)
V

The r i g h t hand side depends on G r e e n - W e i l functions lD,v' and so is a

priori defined only for P outside the support of D . Since hD de-
pends o n l y on the linear equivalence class of D mod 0(1) , we can
change D by a linear equivalence so as to m a k e the r i g h t hand side
defined at a g i v e n point.

Now let S be a f i n i t e set of a b s o l u t e values on K . We define, rela-

tive to a given choice of W e l l - G r e e n functions and heights:

m(D,S) = Z d v 1 D ,v
v6S

N(D,S) v$S dv ID'v

Then
 412

h D = m(D,S) + N(D,S) ,

a n d one b a s i c property of h e i g h t s says that hD depends o n l y o n the

linear equivalence class of D . This is V o j t a ' s translation of F M T
into the n u m b e r theoretic context, with the h e i g h t hD corresponding
to the function T(D) of N e v a n l i n n n a theory.

Remark. The p r o p e r t i e s of h e i g h t s listed above also hold for T ,

as w e l l as o t h e r s listed for instance in [La], e.g. if D is a l g e -
braically equivalent to 0 , then T(D) = O(T(E)) for E ample. As
far as I can tell, in the analytic context, there has been no such
systematic listing of the p r o p e r t i e s of T , similar t o the listing of
the p r o p e r t i e s of h e i g h t s as in n u m b e r theory.

Vojta's translation of S M T y i e l d s his c o n j e c t u r e :

Let X be a p r o j e c t i v e non-singu~ variety defined o v e r a number f i e l d K .

L~ S be a f i ~ g ~ e s e t of absolute values on K . Let D be a d i v i s o r on
X rational over K and w i t h s i m p l e normal o t o s s i n g s . Let E be ample on
X . Give ~ . Then t h e e exists a prop~ Zariski closed subset
Z(S,D,E,s) = Z(s) such that

m(D,S,P) + hK(P ) <ShE(P ) for P6X(K) - Z(s)

Or in o t h e r words,

E dv lD,v + hK ~ s hE on X(K) - Z(s)

v6S

where K is the canonical class.

 413

EXAMPLES

Example I. Let X : ~I • K = ~ , E = (~) . Let ~ be algebraic,

and let

f(t) = ~ (d ~ - t )
d

where the product is t a k e n over all conjugates d~ of ~ over ~ .

Let D be the divisor of zeros of f . The canonical class K is just
-2(~) . A rational point P corresponds to a rational value t = p/q
with p,q 6~ , q > 0 , and p,q relatively prime. We let S consist
of the absolute value at infinity. If If(p/q) I is small, then p/q
is c l o s e to some root of f If p/q is c l o s e to a , then it h a s to
be far away from the other conjugates of ~ . Consequently Vojta's
inequality yields from the definitions:

-log [~- P/ql - 2h (p/q) 2~ s h (p/q)

with a finite number of exceptional fractions. Exponentiating, this

reads

I~_ ~ >__/__I
= 2+g
q

which is R o t h ' s theorem.

Remark. Some time ago, I conjectured that instead of the qS in R o t h ' s

theorem, one could take a power of log q (even possibly (log q ) 1 + s ) .
Similarly, in V o j t a ' s conjecture, the right hand side should be re-
placed conjecturally by O(log h E ) . If one looks back at t h e Nevanlinna
theory, one then sees that the analogous statement is t r u e , and relies
on an e x t r a analytic argument which is c a l l e d the lemma on logarithmic
derivatives. Cf. Griffiths [Gr].
 414

Example 2. Let X = ~n and let D = L O +...+L be the formal sum

n
of the hyperplane coordinate sections, with L at infinity, and
o
E = L° . Let ~ -m be a rational function such that

(~i) = Li-L °

Let S be a finite set of absolute values. Note that in the case of

~n , the canonical class K contains -(n+1)L O . Consequently, Vojta's
inequality in t h i s case yields

TT
i v6S H(p) n + l + C

for all P outside the closed set Z(e) . This is S c h m i d t ' s theorem,
except that Schmidt arrives at t h e conclusion that the e x c e p t i o n a l set
is a f i n i t e union of h y p e r p l a n e s . In o r d e r to m a k e Vojta's conjecture
imply Schmidt strictly, one would have to r e f i n e it so as to give a
bound on the degrees of the components of the exceptional set, which
should turn out to be I if the original data is linear.

Example 3. Let X be a curve of genus ~ 2 . Take S empty. The

canonical class has degree 2g-2 where g is the genus, and so is
ample. Then Vojta's inequality now reads

hK ~ s hE on X(K) ,

except for a finite set of points. Since K is a m p l e , such an in-

equality holds only if X(K) is f i n i t e , which is F a l t i n g ' s theorem.

Example 4. This is a h i g h e r dimensional version of the preceding

example. Instead of assuming that X is a c u r v e , we let X have any
dimension, but assume that the canonical class is a m p l e . The same
inequality shows that the set of rational points is n o t Zariski dense.
 415

This goes toward an o l d conjecture of m i n e , that if a v a r i e t y is

hyperbolic, then it h a s only a finite number of r a t i o n a l points. The
effect of h y p e r b o l i c i t y should be to e l i m i n a t e the e x c e p t i o n a l Zariski
set in V o j t a ' s conjecture. For progress concerning this conjecture in
the function field case, cf. Noguchi [No], under the related assumption
that the cotangent bundle is a m p l e , and that the r a t i o n a l points are
Zariski dense.

To apply the argument of V o j t a ' s inequality it is n o t necessary to

assume that the canonical invertible sheaf is a m p l e , it s u f f i c e s to be
in a s i t u a t i o n when for a n y a m p l e divisor E , h E = O(h K) . This is
the case for v a r i e t i e s of g e n e r a l type, which means that the rational
m a p of X defined by a sufficiently high multiple o f the c a n o n i c a l
class gives a rational m a p of d i m e n s i o n d = dim X . Then we h a v e
h°(mK)>> md for m sufficiently large, a n d we use the following lemma.

Lemma. Let x be a n o n - s i n g u l a ~ v a r i e t y . Let E be v e r y ample on X , and

let D be a d i v i s o r on X such t h a t h ° (roD)>> m d for m >= m
o
Then t h e r e e x i s t s mI such that h°(mD-E)>>m d , and i n p a r t i c u l a r ,
mD-E is linearly e q u i v a l e n t t o an e f f e c t k v e div~or, for a// m > mI .

Proof. First a remark for any d i v i s o r D . Let E' be a m p l e , and

such that D + E' is ample. T h e n we h a v e an i n c l u s i o n

H°(mD) m H°(mD + mE') ,

which shows that h°(mD) Sh°(mD + mE') = x(m(D + E')) for m large
because the h i g h e r cohomology groups vanish for m large, so
h°(mD)>> md

Now for the lemma, without loss o f g e n e r a l i t y we c a n r e p l a c e E

 416

by any divisor in its class, and thus without loss of generality we

may assume that E is a n irreducible non-singular subvariety of X .
We have the exact sequence

0--> 0 (mD-E) > 0 (roD) .........> 0 (mD)IE > 0

whence the exact cohomology sequence

0 ~ > H°(X,mD-E) > H°(X,mD) > H°(E,(0(D) IE) ®m)

noting that 0(mD) IE : (0(D)I E) e m . Applying the first remark to this

invertible sheaf on E we conclude that the dimension of the term on
the right is < < m d-1 , so h°(X,mD-E)>>m d for m large, and in p a r -
ticular is p o s i t i v e for m large, whence the lemma follows.

For mD - E effective, we get h E ~ hmD + 0(I) as desired.

Example 5. Let A be an a b e l i a n variety, and let D be a very ample

divisor with SNC . Let S be a finite set of absolute values of K
containing the archimedean ones. Let ~1,...,~n be a set of g e n e r a t o r s
for the space of sections of 0(D) . Let 0S be the ring of S-integers
in K (elements of K which are integral at all v { S) . A p o i n t
P 6 A(K) is said to be S-integral relative to t h e s e generators if
~i(P) 6 O S for i = 1,...,n . On the set of such S-integral points, we
have

E d v ~D,v = hD + O(])
v6S

immediately from the definitions. The canonical class is 0 . Then

again Vojta's inequality shows that the set of S - i n t e g r a l points as
above is n o t Zariski dense.

This is in the direction of m y old conjecture that on any affine

open subset of an a b e l i a n variety, the set of S-integral points is
 417

finite. However, in this stronger conjecture, we a g a i n see the dif-

ference between finiteness and the p r o p e r t y of not b e i n g Zariski dense.

E__xample 6. Hall's conjecture Marshall Hall conjectured that if

3 2
x,y are integers, and x -y # 0 then

I
Ix 3 y2 I > max(Ix31 , ly21)

with a finite number of e x c e p t i o n s . Actually, Hall omitted the c ,

but S t a r k and T r o t t e r for p r o b a b i l i s t i c reasons have pointed out that
it is a l m o s t certainly needed, so we put it in.

Vojta has shown that his c o n j e c t u r e implies Hall's. We sketch the

argument. Let

2 2
f : ]PI > ]P2

be the r a t i o n a l map defined on p r o j e c t i v e coordinates by

f(x,y,z) = (x3,y 2z,z3)

Then f is a m o r p h i s m except at (0,1,0) . We have indexed projective

2-space by indices I and 2 to d i s t i n g u i s h the space of d e p a r t u r e
and the space of a r r i v a l . We let L = LI be the h y p e r p l a n e at infinity
2
on ]P12 , and L2 the h y p e r p l a n e at i n f i n i t y on ]P2

Let C be the curve in ]PI2 defined by x 3 -y 2 = 0 . Let ~ be

the r a t i o n a l function defined by

3 2
~(x,y) = X-y
 418

Then the d i v i s o r of ~ is g i v e n by

(~) = C - 3L .

In t e r m s of h e i g h t s , Hall's conjecture can be f o r m u l a t e d in the form

log l~(x,y) I > l h L 2 f ( x , y ) + error term,

or if v denotes the o r d i n a r y absolute value on ~ ,

I
(I) vo<0 (x,y) = - 2~ Uh L
< of(x,y) +error term.

Note that v o ~ = i(~) is a Weil function associated with the d i v i s o r

(~) . Thus Hall's conjecture amounts to an inequality on Weil functions.
By b l o w i n g up the p o i n t of indeterminacy of f and the singularity
of C at (0,0) , one obtains a variety X and a c o r r e s p o n d i n g mor-
phism fl : X > ~2 making the following diagram commutative:

2 > 2
]Pl ]P2 "

The blow ups are c h o s e n so that the e x c e p t i o n a l divisor and C have

simple normal crossings. By t a k i n g D to be their sum t o g e t h e r with
the h y p e r p l a n e at i n f i n i t y , Vojta shows that his c o n j e c t u r e implies
Hall's. By a s i m i l a r technique, Vojta shows that his conjecture implies
several other classical diophantine conjectures. I refer the reader to
his forthcoming paper on the subject.
 419

BIBLIOGRAPHY

[Gr] P. Griffiths, E n t r e holomorphic mappi~s in one and several variabl~:

Hermann Weyl Lectures, Institute for A d v a n c e d Study,
Institute for A d v a n c e d Study, P r i n c e t o n Univ. Press,
P r i n c e t o n NJ, 1976.

[La] S. Lang, Fundamenta~ of Diophantine Geome~y, Springer Verlag,

1984.

[No] Jo N o g u c h i ,A higher dime~ional analogue of Mordell's c o n j e c t ~ e over

function f i e l d , Math. Ann. 258 (1981) pp. 207-212.

[Vo] P. Vojta, Integrol points on v ~ i e t i e s , Thesis, Harvard, 1983.

 A COUNTEREXAMPLE IN 3 - S P A C E TO A C O N J E C T U R E OF H. HOPF

H e n r y C. W e n t e
D e p a r t m e n t of M a t h e m a t i c s
The U n i v e r s i t y of T o l e d o
T o l e d o , O h i o 4 3 6 0 6 , U. S. A.

In this article we produce a counterexample to t h e following conjecture

of H. H o p f . We shall carefully state the theorems involved in t h e c o n -
struction and also provide a geometric description (with s u g g e s t i v e
sketches) of t h e surfaces giving the counterexample. An expanded ver-
sion complete with proofs is t o a p p e a r in a p a p e r of the author [~ .
Conjecture of H e i n z Hopf; If Z is an immersion of an o r i e n t e d closed
hypersurface in Rn with constant mean curvature H ~ 0, t h e n the hyper-
surface is t h e standard embedded (n-l)-sphere.
If the immersed surface is k n o w n to b e e m b e d d e d then a well-known
result of A. D. A l e x a n d r o f f [i] a s s e r t s that the conjecture is true.
H. H o p f himself [4] showed that if Z is an immersion of S2 into R3
with constant mean curvature then the c o n j e c t u r e is s t i l l true. Recently
Wu-Yi Hsiang [5] p r o d u c e d an i m m e r s i o n of S3 into R4 with constant
mean curvature which is n o t isometric to the standard sphere. However,
his construction does not work in t h e classical dimension (=3) and the
conjecture has remained open in t h i s case. We h a v e the following.

Counterexample Theorem: There exist closed immersed surfaces of g e n u s

one in R3 with constant mean curvature. ( In fact, we exhibit a coun-
tably infinite number of i s o m e t r i c a l l y distinct examples.)

We shall exhibit the surface by producing a conformal mapping of the

plane R2 into R3 with constant mean curvature which is d o u b l y per-
iodic with respect to a r e c t a n g l e in t h e p l a n e . Let w = (u,v) = u +iv
represent a typical point in R 2= C while x = (x,y,z) denotes a point
in R3 so t h a t o u r immersion is g i v e n b y a f u n c t i o n x(u,v) We let

d x ' d x = ds 2 = E (du 2 + d v 2) = e 2~ (du 2 + dv 2) (la)

-dx-d[ = Ldu 2 + 2Mdudv + Ndv 2 (Ib)

be the first and second fundamental forms for the surface. We shall set
the m e a n curvature H = ½. The Gauss and Codazzi-Mainardi equations in
this case become (see [4] for details)

Aw + Ke 2m = 0 , K = Gauss curvature = (LN - M 2 ) / E2 (2a)

 422

~[w) = (L - N ) / 2 - iM is a c o m p l e x analytic function. (2b)

NOW suppose that ~(u,v) is a s o l u t i o n to t h e differential equation

6~ + sinh m cosh ~ = 0 (3)

If w e set E = e 2~ , L = e~sinh ~, M = 0, a n d N = e~cosh ~ , then it

follows that the G a u s s and Codazzi-Mainardi equations are satisfied and
by a theorem of B o n n e t the system c a n be i n t e g r a t e d to y i e l d a surface
x(u,v) ,unique u p to a E u c l i d e a n motion in R 3, h a v i n g the g i v e n funda-
mental forms. The equations to b e i n t e g r a t e d are

XUU = ~uXu - ~vXv + L[ (4)

~uv = ~v~u + ~uXv + ~

Xvv = -~u~u + ~vXv + N[

~u = -kl Xu

~v = -k2 ~v
Here k I = L/E = e-~sinh w, k 2 = e-~cosh ~ so w e see t h a t the lines
of curvature correspond to l i n e s parallel t o the coordinate axes in R2°
Furthermore, the surface is f r e e o f u m b i l i c points.
If x(u,v) is to b e a doubly periodic mapping then so must ~(u,v).
However the converse n e e d n o t b e true. Suppose that ~(u,v) is a p o s -
itive solution t o the d i f f e r e n t i a l equation (3) o n a r e c t a n g u l a r domain
~AB lying in t h e first quadrant with t w o of its sides on the c o o r d i n -
ate a x e s a n d t h e v e r t e x opposite the origin at (A,B). Suppose also
that the solution ~(u,v) vanishes on the boundary o f the r e c t a n g l e .
FolLowing the argument used in [3] , one can show that ~ (u,v} satisfies
the following symmetry properties.

a) ~(u,v) is s y m m e t r i c about the lines u = A/2 and v = B/2. (5)

b) For a fixed v, 0 < v < B, ~(u,v) is a n i n c r e a s i n g function
of u, 0 ~ u ~ A/2 . For a fixed u, 0 < u < A, w(u,v) is an
increasing function of v, 0 < v ~ B/2.
c) ~u(U,0) is s t r i c t l y increasing for 0 ~ u < A/2.
Wv(0,v) is s t r i c t l y increasing for 0 < v < B/2.

Furthermore, ~(u,v) can be extended as a s o l u t i o n o f the d i f f e r e n t i a l

equation {3) o n all of R2 by o d d r e f l e c t i o n s across the g r i d lines
u = mA, v = nB (m, n i n t e g e r s ) .
 423

Theorem 2: Suppose e(u,v) is a s o l u t i o n to the d i f f e r e n t i a l equation

(3) on R 2 which is p o s i t i v e o n the fundamental rectangle nAB , van-
ishing on the b o u n d a r y and satisfying the properties (5). The mapping
x(u,v) obtained by integrating the system (4) is a n i m m e r s e d surface
of constant mean curvature H = ½ and satisfying the following symmetry
properties.
(6)
a) T h e curve x((m + ½)A, v) lies in a n o r m a l plane Hm with
as a n o r m a l vector to H . If R is t h e r e f l e c t i o n map about
u m m
in R3 then x((m + ½ ) A + u,v) =
ROx((m + ½ ) A - u,v)
m m
b) The curve x(u, (n + ½)B) lies in a n o r m a l p l a n e nn with
as a n o r m a l v e c t o r to ~ . If R' is the r e f l e c t i o n m a p about
v R3 n n n
in then x ( u (n + ½ ) B + v) = R ' o x ( u , (n + ½ ) B - v) Each ~ is
' n " n
orthogonal to e a c h
m
c) T h e curve x(u,0) is a p l a n a r curve lying in a p l a n e F0 w h i c h
is a t a n g e n t plane t o the s u r f a c e at e a c h point. This curve intersects
each plane Em orthogonally. Xu(U,0) is an e v e n function of u.
This allows us to c o n c l u d e that a l l of t h e p l a n e s Hm are p a r a l l e l .
d) The curve x(0,v) satisfies the condition (x + ~) (0!v) = ~o
a constant vector. Therefore x(0,v) lies on a s p h e r e S(~o,l) with
center ~ and r a d i u s one. Similarly x(kA,v) lies on a s p h e r e
O
S(~ k ,i). The points Ck lie in e v e r y plane ~n
e) x(u + 2A,v) = x(u,v) + 5 where b = ~2 - ~o is a v e c t o r nor~
mal to the p l a n e s Hm carrying Z0 to H 2.
f) x ( u , v + 2B) = 0 x(u,v) where @ is a rotation from n° to n 2
about their line of i n t e r s e c t i o n , i.

The surface will close up if w e can select the rectangle nAB so

that the translation b = 0 (i.e. all the p l a n e s Hm are identical)
a n d so t h a t the r o t a t i o n angle @ is a r a t i o n a l multiple of 2~. W e u s e
a continuity argument to s h o w that this is p o s s i b l e . The p r o c e d u r e is
as follows. Map (via a h o m o t h e t y ) a l l rectangles of similar shape onto
a representative rectangle which we select b y the standard Schwartz-
Christoffel mapping of r e c t a n g l e s onto the unit disk.

v w-plane (A , B ) /~-plane
v

!- J ~u' >u

Figure i: The Fundamental Domain.

 424

We have the following identities satisfied by the v a r i o u s functions

defined on the d o m a i n s pictured in F i g u r e I.
(7)
a) On ~(a,~).
A~ + s i n h ~ cosh ~ = 0
Ao + sinh ~ = 0 where o = 2~.
b) On ~(~)
AW + 2 1 s i n h W cosh W = 0 where W = ~o~ °
AZ + 21 s i n h Z = 0 where z = 2W.
c) On the d i s k D.
A~ + ~If' (z,~) 12(e ~ - e -U) = 0 , where ~ = Zof .

w = f(z,~) = /0z(t 4 + 2(cos 2a)t 2 + i) -½ dt

The proof of the e x i s t e n c e of p o s i t i v e solutions to the s y s t e m

(7c) on D which vanish on the b o u n d a r y (and such that small values
for i correspond to l a r g e solutions ~) is b a s e d on a m e t h o d devel-
o p e d by V.K. Weston [7] and R.L. Moseley [6].

Theorem 3: There exists an o p e n set 0 C (e,l)-plane where for e a c h

el,~2 with 0 < el < ~2 < 7/2 there exists ~ = ~(~i,~2) > 0 so t h a t
[~l,e2] x (0,~]C 0 , and a mapping from 0 to C(D) denoted by ~(z,~,l)
such that
a) Z(w,e,l)= ~(g(w,a),e,l) is a p o s i t i v e solution to (7b) w h i c h
vanishes on the b o u n d a r y .
b) The functions ~' Zu ' Ev depend continuously on (e,l) down
to I = 0 with Z(w,~,0) = Zo(W,e) = 4 log(i/Ig(w,~)I)
c) F o r I > 0 the m a p p i n g (~,l)--~Y(z,~,l) is a c o n t i n o u o s l y
differentiable mapping of O into C(D)

Remark on the proof: One first constructs a good approximate solution

Uo(Z,l) with the c o r r e c t asymtotic limit as I approaches 0 by u s i n g
the L i o u v i l l e f o r m of the e x a c t solution to the d i f f e r e n t i a l equation
AV + le V = 0, n a m e l y le V = IF' (z)12/(1 + IF(z)12) 2 where F(z) is a
complex analytic function with at m o s t simple zeros and poles. Then
one applies a modified Newton iteration scheme, starting with Uo(Z,l)
using the a p p r o p r i a t e integral operator, and shows that the resulting
sequence converges in C(D) to the desired solution.

We w a n t to m e a s u r e the d i s t a n c e between the p a r a l l e l planes ~0

and ~i and w i s h to show that for c e r t a i n (a,l) the d i s t a n c e is zero.
It is b e t t e r to l o o k at the surfaces ~ ( w , e , l ) = ~ o ¢ ( w , ~ , l ) / ! 2/~ d e f i n e d
 425

relative to the fundamental domain ~(e) and to m e a s u r e the distance

b e t w e e n the parallel planes ~' and K{ w h i c h c o r r e s p o n d to the map-
o
ping y . We do this by looking at the curve y(u,0,e,l) , a planar
curve w h i c h cuts through the planes Emi o r t h o g o n a l l y and has the
symmetry i n d i c a t e d in Figure 2.
I
[I ~ , ~ _ , ~ ( 3 A ( ~ ) I 2 , 0 , ~ , k )

S(~,X)

I //~ (A(~),0,~,~)I
r / I

Figure 2 : M e a s u r i n g the D i s t a n c e b e t w e e n the P a r a l l e l Planes ~'O and ~i

The functions y(u.v,e,l) are conformal immersions into R 3 with

constant m e a n curvature H = 2/~ , so that as I approaches 0 the mean
c u r v a t u r e approaches 0 and the mapping tends to a p l a n a r map. The funct-
ions y satisfy a system just like (4) with 0~ r e p l a c e d by W : Z/2,
L is r e p l a c e d by L = /~ L and so on. Since by T h e o r e m 3b the funct-
ion W(u,v,~,l) approaches W(u,v,~,0) = 2 log(i/[g(w,~) I) as I approach-
es 0, the curve y(u,0,~,l) approaches a limit curve y(u,0,~,0) as
I approaches 0. It follows that the d i s t a n c e function S = S(a,X) ,
as i n d i c a t e d in Figure 2, is continuous down to I = 0 and d i f f e r e n t i a -
ble if I is positive. Since W(u,v,~,0) is known e x p l i c i t l y one can
calculate S(~,0), o b t a i n i n g
Z6 (8)
S(~,0) = (cos 20/(2cos 2e - 2cos 26) 9 ) de , 6 = (~/2)-~.

We i m m e d i a t e l y have the following conclusions.

a) S(~,0) is strictly increasing for 0 < ~ < ~/2.
b) S(~,0) approaches -~ as ~ approaches 0.
c) S(~,0) is p o s i t i v e for e greater than ~/4.
It follows that there is e x a c t l y one value ~* , 0 < e* < ~/4, for
which S(~*,0) = 0. We have the following p i c t u r e (see Figure 3).
There is a small r e c t a n g l e L~I,~J x [0,~] with S(el,l ) negative,
S(~2,I) positive, and S(~*,0) = 0. There is a c o n n e c t e d set X includ-
ed in this small r e c t a n g l e on w h i c h S v a n i s h e s and w h i c h separates
the left side of the rectangle from the right side. In p a r t i c u l a r
(e*,0) is in the set X and every line I = constant slices into X.
 426

I s
~i ~*
iso
e2

Figure ~: The Set S=0: All the P l a n e s Hm are Identical.

Now we measure the rotation angle between the p l a n e s ~o and Q1

by looking at the image of the c u r v e y(0,v,~,~), B(~)/2 < v <3B(~)/2.
From Theorem 2d it f o l l o w s that this curve lles on the sphere with
center ~o and r a d i u s (21) -½ , connecting the p l a n e s ~o to ~i
and intersecting them orthogonally. Let T(a,I) be the d i s t a n c e between
these planes as m e a s u r e d on a g r e a t circle of the sphere whose radius
is (21) -½ By r e p e a t i n g the calculation used to c o m p u t e S(~,~),
one finds that for s m a l l l/and ~ less than ~/4, T(~,I) is p o s i t i v e
down to the limit I = 0 with the e x p r e s s i o n for T(e,0) being similar
to t h a t for S(~,0). However, for the angle function @(e,l) we have
the identity @(~,i) = (21) ½ T(~,I) . This gives us the following:

a) @(e,1) is p o s i t i v e for ~ positive.

b) @(~,I) approaches 0 as I approaches 0.

Since X is a c o n n e c t e d set w i t h more than one p o i n t (see F i g u r e 3),

it f o l l o w s by continuity that on the set X the function @(~,~)
takes on a c o n t i n u u m of v a l u e s [0,~ where ~ is p o s i t i v e . Whenever
0(~,~) is a r a t i o n a l multiple of 2~ the surface will close up. This
establishes the e x i s t e n c e of a c o u n t a b l e number of i s o m e t r i c a l l y dis-
tinct immersions of a torus into R3 with constant mean curvature.

A View of the Immersed Tori.

Let ~ = ~AB be a representative rectangle chosen so t h a t the

smallest eigenvalue of the L a p l a c e differential equation

Av + yv = 0 on ~ , v = 0 on boundary ~ (9)

is Y1 = I. This means that 1 = Yl = ~ 2 ( ( I / A 2 ) + ( I / B 2 ) ) and in p a r -

ticular A and B are b o t h greater than ~. We are to s o l v e the d i f f e r -
ential equation

AW + 2 l s i n h W cosh W = 0 on ~ , W = 0 on 6~. (i0)

We h a v e the following facts regarding solutions to the d i f f e r e n t i a l

equation (i0).
 427

a) T h e r e exists a branch of p o s i t i v e solutions to (i0) w h i c h

bifurcate f r o m the zero solution at 21 = Y1 = 1 or I = 1/2.
b) For any p o s i t i v e solution (W,I) we must have 0 < I < 1/2,
and for a n y i in this interval there exists at l e a s t one p o s i t i v e sol-
ution.
c) As I approaches 0 there is a c u r v e of l a r g e positive solutions
(W,I) obtained by applying Theorem 3.
It is t e m p t i n g (but not y e t proven) to c o n j e c t u r e that the b r a n c h
bifurcating f r o m the zero solution at I = 1/2 connects up w i t h the
branch of l a r g e solutions established in T h e o r e m 3. E v e n more tempting
is the following conjecture.
conjecture: Let (WI,II) and (W2,12) be two p o s i t i v e solutions to
the system (i0). If 0 < I 1 < 12 < 1/2 then W 1 is g r e a t e r than W2
at e v e r y point inside ~.
For each solution of the system (i0) w e m a y apply (7) to get a
solution ~(u,v) to the d i f f e r e n t i a l equation (3) and then apply our
recipe to c o n s t r u c t an i m m e r s i o n x(u,v) with constant mean curvature.
In the limit case w h e r e W = 0 the resulting immersion is s i m p l y a
conformal mapping of the p l a n e onto a circular cylinder whose cross
section is a c i r c l e of r a d i u s one.
In the figures that follow we shall sketch the image x(u,v) of
a portion of the fundamental rectangle 2/~ ~AB as i n d i c a t e d in the
first figure and labeled {1,2,3,4,5,6}. A + sign indicates that
~(u,v) is p o s i t i v e and h e n c e the G a u s s curvature of the image surface
K = e-2~sinh ~ cosh e is p o s i t i v e , while a - sign indicates that both
functions are n e g a t i v e . The rest of the surface is o b t a i n e d by r o t a t i n g
the surface 180 ° about the n o r m a l line at the image of 2 f o l l o w e d by
a series of r e f l e c t i o n s about the appropriate planes.
V /~ ( A ,B )
+ + - _
4 5 6
+ + - _
1 i2 3 ____~u

F i.gure 4: The Fundamental Domain /2}, ~(e) = [~[e,1).

1 1 ,Xu 2 3

9 > ~/2

Figure ~: Case I, W=0, A Pure Cylinder.

 428

2 3

Figure 6: Case 2. W is p o s i t i v e on ~(~,l) but not too Large.

1 2
3

/
Figure ~: Case 3. W s o m e w h a t larger, the Planes ~o' ~i s t i l l separated.

If one k e e p s ~ fixed and lets ~ approach 0, then one can easily

show the following.

i) / K dA = area of the Gauss m a p --9 47 as approaches 0.

I~ (~+)

2) ]e 2W d u d v = Area of x(~+) > 4~(2) 2 as ~ approaches 0.

3) /e 2W d u d v = A r e a of ~(~-) "~ 0 as approaches 0.

/<

These calculations suggest that as X approaches 0, x(~+) takes on

the shape of a s p h e r e of r a d i u s 2.
 429

, / ,( /

Figure 8: Case 4. The P a r a l l e l Planes fro' HI are Identical.

If one reflects the sketched Figure 8 about the plane of the

paper (~o = ~i ) you obtain a surface w h i c h resembles a clam shell.
Upon rotating this shall 180 ° about the vertical line Co-(2) one
obtains the other shell. The combined figure is now a clam w i t h the
shells opened a bit.

References:
i. A.D. Alexandroff, Uniquenes s Theorems for Surfaces in the Large,
V. Vestnik, L e n i n g r a d Univ. No. 19 (1958) 5-8: Am. Math. Soc. Transl.
(Series 2) 21, 412-416.
2. L.P. Eisenhart, A T r e a t i s e on the D i f f e r e n t i a l Geometry of Curves
and Surfaces, Dover Reprint (1960)o
3. B. Gidas, W. Ni, L. Nirenberg, S y m m e t r y and Related P r o p e r t i e s via
the M a x i m u m P r i n c i p l e , Comm. Math. Physics 68 (1979) No. 3, 209-243.
4. H. Hopf, D i f f e r e n t i a l Geometry in the L a r g e , ( S e m i n a r Lectures New
York Univ. 1946 and Stanford Univ. 1956) Lecture Notes in M a t h e m a t i c s
No. 1000, S p r i n g e r Verlag, 1983.
5. Wu-Yi Hsiang, G e n e r a l i z e d R o t a t i o n a l H y p e r s u r f a c e s of C o n s t a n t Mean
Curvature in the E u c l i d e a n Space ~, Jour. Diff. Geometry 17(1982)337-356.
6. J.L. Moseley, O__nnA s y m t o t i c Solutions for a D i r i c h l e t P r o b l e m with an
e x p o n e n t i a l Singularity, Rep Amr I, West V i r g i n i a U n i v e r s i t y (1981)
7. V.H. Weston, On the A s y m t o t i c Solution of a Partial D i f f e r e n t i a l
E q u a t i o n w i t h an E x p o n e n t i a l Nonlinearity, SIAM J. Math Anal 9(1978)
1030-1053.
8. H. C. Wente, c o u n t e r e x a m p l e to a c o n j e c t u r e of H. Hopf, (to appear)
Pac. Jour. of Math.
 THE T O P O L O G Y AND G E O M E T R Y OF THE M O D U L I SPACE
OF RIEMANN SURFACES

Scott A. W o l p e r t *
D e p a r t m e n t of M a t h e m a t i c s
U n i v e r s i t y of M a r y l a n d
C o l l e g e Park, MD 20742

I would like to d e s c r i b e a sampling of recent results concerning

the m o d u l i space M of Riemann surfaces. My plan is to p r e s e n t
g
several of the ideas u n d e r l y i n g the recent work of John Harer on the
topol o g y of M and of myself on the H e r m i t i a n and symplectic geom-
g
etry. My purpose is not to give a survey; for i n s t a n c e the reader
is r e f e r r e d to the papers [7, 9] for the recent p r o g r e s s on the
question of w h e t h e r Mg, the m o d u l i space of stable curves, is
unirational.
The d i s c u s s i o n will be d i v i d e d into two parts: the t o p o l o g y of
Mg and Mg, especially t h e homology of t h e mapping c l a s s group Fg
and the g e o m e t r y of the W e i l - P e t e r s s o n metric. As b a c k g r o u n d I shall
start w i t h the basic d e f i n i t i o n s and notation.

i. Definitions and N o t a t i o n .

i.i. Let F be a compact topological surface of genus g with r

boundary components and s distinguished points in F - ~F; set
S = ~F U {points}. I shall always assume that 2g - 2 + s + r > 0
or e q u i v a l e n t l y that F - S admits a complete hyperbolic metric.
Cons i d e r Homeo+(F,S), the group of o r i e n t a t i o n preserving homeo-
morphisms of F restricting to the i d e n t i t y on S and the normal
subgroup I(F,S) of h o m e o m o r p h i s m s isotopic to the i d e n t i t y fixing S.

Definition i.i. Fs = Homeo+(F,S)/I(F,S) is the m a p p i n g class

g,r
grou p for genus g, r boundary components and s punctures.

I shall use the c o n v e n t i o n that an o m i t t e d index is set equal to

zero. For genus g and s punctures the m a p p i n g class group Fs
g
acts p r o p e r l y discontinuously on the T e i c h m u l l e r space T s via
g
biholomorphisms. The q u o t i e n t M s , the classical m o d u l i space of
g
Riemann surfaces, is a c o m p l e x V-manifold. To be more specific
start by c o n s i d e r i n g triples (R,f,P), where f is a h o m e o m o r p h i s m
of the t o p o l o g i c a l surface F to a R i e m a n n surface R with f(S) =P.
An e q u i v a l e n c e relation (the marking) is i n t r o d u c e d by defining:

*Partially supported by the N a t i o n a l Science F o u n d a t i o n , Max P l a n c k

Institute for M a t h e m a t i c s and A l f r e d P. Sloan Foundation.
 432

(R0,f0,P 0) ~ (RI,fI,P I) provided there is a c o n f o r m a l map k with

f0
(F,S) ~ (R0,P 0)

(RI,P 1 )

commutative modulo a homotopy fixing S and P.

Definition 1.2. T s, the T e i c h m u l l e r space for g e n u s g and s

g
punctures, is the set of ~ e q u i v a l e n c e c l a s s e s of t r i p l e s (R,f,P).

Briefly Ts is a c o m p l e x m a n i f o l d and is h o m e o m o r p h i c to
~46g-6+2s. Theg m a p p i n g class group Fs acts n a t u r a l l y on Ts : to
the e q u i v a l e n c e classes {h} 6 F s and g { ( R , f , P ) } 6 Ts assig~ the
g Fs g
class { (R,f o h,P)} 6 T s. The action represents as b i h o l o m o r -
phisms of Ts . g g
g

Definition 1.3. M s = Ts/F s is the m o d u l i space for g e n u s g and s

g g g
punctures.

As an e x a m p l e the r e a d e r will check that for g e n u s 1 and 1

puncthre T1 is the u p p e r half plane H c C and ~i1 is the
elliptic modular group SL(2;Z) acting on H by l i n e a r fractional
transformations.

1.2. NOW I shall review the definition of the complex structure on

T s. For a Riemann surface R with 2g - 2 + s > 0 c o n s i d e r the
g 2
hyperbolic metric ~ = ds , of c o n s t a n t c u r v a t u r e -i. Associated
2
to R are the L (relative to ~) tensor spaces H(R) of h a r m o n i c
Beltrami differentials (tensors of type ~ ~ dz) and Q(R) of
holomorphic quadratic differentials (tensors of type dz ~ dz) . Of
course harmonic is d e f i n e d in terms of the L a p l a c e Beltrami operator
for the h y p e r b o l i c metric. A pairing H(R) × Q(R) -~ C is d e f i n e d by
integration over R : for ~ ~ H(R) and ~ 6 Q(R) define (p,~) =
fR P~" The A h f l o r s and Bers description of the complex structure of

Ts is s u m m a r i z e d in the d i a g r a m
g

T l'0TSg x ( T l ' 0 ) * T S g / i - ~ i

H (R) × Q (R)/ '

 433

where T l'0 is the h o l o m o r p h i c tangent space, (TI'0) * its dual and

these spaces are n a t u r a l l y paired [i, 5]. The h y p e r b o l i c metric
induces a natural inner p r o d u c t on every space of tensors for R.
In p a r t i c u l a r Weil was the first to c o n s i d e r the Hermitian product
on H(R) as a m e t r i c for T s.
g
Definition 1.4. Given ~,~ E H(R) the H e r m i t i a n product <~,~> =
fR ; ~ t - 1 is the W e i l - P e t e r s s o n m e t r i c .
The W e i l - P e t e r s s o n metric is Kahler [2,3,10,21] and its H e r m i t i a n
and symplectic geometry is the subject of C h a p t e r 3.

2. The H o m o l o g y of the M a p p i n g Class Group.

2.1. In this chapter I shall concentrate on three recent exciting

results of John Harer: i) the c o m p u t a t i o n of H2(FS,r),~ in brief
H2(F s ) ~ Z s+l, for g ~ 5, ii) the s t a b i l i t y theorems, in brief
g,r
Hk(F~,
r)~ is i n d e p e n d e n t of g and r when g ~ 3k + i and iii)
the virtual cohomological Fsdimension
is of
d = 4g - 4 + 2 r + s
g,r
for r + s > 0, 4g - 5 for r = s = 0, in p a r t i c u l a r Hk(MS;~)
~ = 0
g
for k > d, [ll, 12, 13]. Of course the reader will c o n s u l t the
references for the c o m p l e t e statements especially for the cases of
punctures and b o u n d a r y components.

2.2. A useful technique for c o m p u t i n g the h o m o l o g y of a g r o u p G is

to c o n s t r u c t a cell complex C on w h i c h G acts cellularly, i.e.
cells are mapped to cells. Then the h o m o l o g y of G may be c o m p u t e d
from the h o m o l o g y of the q u o t i e n t C/G and a d e s c r i p t i o n of the
cell stabilizers. Now Fs is c o m p r i s e d of isotopy classes of
g,r
homeomorphisms; Fs acts on the isotopy invariants of the surface
g,r
F. A n obvious such i n v a r i a n t is the isotopy class of a union of
simple loops. Cell complexes with vertices representing unions of
simple loops, satisfying appropriate hypotheses, appeared previously
in the work of H a t c h e r - T h u r s t o n [15] and H a r v e y [14]. I shall
describe three such c o m p l e x e s (and simple variants) which are at the
center of Harer's considerations.

2.3. The cut system complex CS, [ll], A cut system <Cl,...,Cg>
on F is the isotopy class of a c o l l e & t i o n of d i s j o i n t simple
closed curves Cl,...,Cg such that F - (C 1 U...U Cg) is connected.
A 6imple mov~ of cut systems is the r e p l a c e m e n t of <C.> by <C~>
1 1
where Cj = C[3 for j ~ k and Ck intersects C~ once (all
 434

intersections are positive). I shall use the c o n v e n t i o n below that

any loop omitted from the n o t a t i o n remains unchanged. Now consider
the following sequences of s i m p l e moves (see F i g u r e i).

|>
<Ci> <C i

<
C," >
1

<C.,C.> <C.,C'.>
z 3 i ]
I I I 21
<C~,C.> <C~,C',>
l 3 i ]

<C. ,C.> <C. ,C'.>

± 3 J 3
/ \
<C.,C!> <C.,C>. (R 3)
1 1
\ /J
<C ,C>--<C, C'>
i

Figure 1

The three sequences can be d e s c r i b e d in terms of the r e l e v a n t loops

(see F i g u r e 2)

C~ ;/

¢i
/

C~~C, Cj-
(R3)
~.__J C
Figure 2
 435

Now define a 0-complex CS 0 with one vertex for e a c h cut

system on F; a 1-complex CS 1 by a t t a c h i n g a 1-cell to CS 0 for
each simple move; and a 2-complex CS 2 by a t t a c h i n g a 2-cell to
CS 1 for e a c h occurence of the c y c l e s R I, R 2 and R 3. Hatcher and
Thurston prove that CS 2 is c o n n e c t e d and simply connected. Harer
simplifies the d e s c r i p t i o n of CS 2 and then attaches 3 cells to
obtain a 2-connected 3 complex CS 3, [ii]. The stabilizer of a
cut system (a v e r t e x of CS 3) is e s s e n t i a l l y a braid group and
Hater analyzes the c e l l stabilizers of CS3o Combining this informa-
tion with an a n a l y s i s of t h e h o m o l o g y of the q u o t i e n t CS3/Fg he
obtains the following theorem, [ii].

Theorem 2.1. H2(F ~ ,r ;Z) = Z s+l, g ~ 5.

The reader will find a slight difference between the a b o v e state-
ment and that found in [Ii]. John Harer has a s s u r e d me that the a b o v e
is i n d e e d the correct statement.
The homology group H2(F) also admits an i n t e r p r e t a t i o n as
bordism classes of f i b r e bundles F + W4 ÷ T with T a closed
oriented surface. In p a r t i c u l a r two bundles are b o r d a n t if they
cobound a 5-manifold fibering over a 3-manifold with fibre F. The
bundle F ÷ W4 + T has s canonical sections ~i' • "''°s defined by
the d i s t i n g u i s h e d points of F. In this setting H2 is s p a n n e d by
the s + 1 natural invariants of W: ~(W)/4, ~ the signature and
~. # 0 , j = 1 ..... s the self-intersection numbers of the sections.
3 3

2.4. Certainly a basic question is w h e t h e r or not the h o m o l o g y of

the m a p p i n g class group falls into any p a t t e r n . As an e x a m p l e con-
sider the a n a l o g o u s question forA , the c o a r s e m o d u l i space of
n
principally polarized n dimensional abelian varieties. Borel in a
fundamental paper computed the r a t i o n a l cohomology of A and f o u n d
n
that Hi(An;R) is i n d e p e n d e n t of n for n large r e l a t i v e to i,
[31]. Recently Charney and Lee have extended these results to ~ ,
n
the Satake compactification of A , [8]. H a r e r e s t a b l i s h e s the
n
analogous result for Hk(F2,r)
: ~ the a n s w e r is i n d e p e n d e n t of g and
r provided g ~ 3k + i. In f a c t the r e a d e r will find in T h e o r e m 2.2
that Harer establishes much more but first I would like to m e n t i o n
the w o r k of E. M i l l e r , [18]. Starting with the w o r k of H a r e r as a
basis Miller observes that the b o u n d a r y connected s u m for surfaces

Fg,l # Fh, 1 = F g + h , ! induces the structure on A = lim÷ H , ( F g , I ; ~ )

of a p o l y n o m i a l algebra on e v e n generators with an e x t e r i o r algebra
on o d d g e n e r a t o r s . Furthermore by g e n e r a l i z i n g an e x a m p l e of A t i y a h
 436

for the generator of H2(Pg) Miller is a b l e to find a generator in

each even dimension.
To give a precise statement of Harer's stability theorem it is
necessary to c o n s i d e r the following three maps of surfaces

¢: Fs ÷ Fs , r { 1
g,r g,r+l

~: Fs ÷ Fs r > 2 and
g,r g+l,r-l' -

Z: Es ÷ F s r a 2
g,r g+l,r-2'

where ¢ and ~ are given by sewing on a pair of pants (a c o p y of

0 (see Figure 3)
F0, 3 )

7{

Figure 3
 437

along one boundary for ~, two for Y and E is g i v e n by s e w i n g

together two b o u n d a r y components (see F i g u r e 4)

L--/

Figure 4

Certainly the m a p s induce homomorphisms of m a p p i n g class groups.

The f o l l o w i n g results for the i n d u c e d m a p s on h o m o l o g y w i t h integer
coefficients can be f o u n d in [12].

Theorem 2.2.

~,: Hk(FS ) + Hk(Fg is an i s o m o r p h i s m

g,r ,r+l )

for k > I, g >_ 3k - 2, r h i

~,: H k ( F Sg, r) ÷ Hk s
(Fg+l ,r + l ) is an i s o m o r p h i s m

for k > l, g _> 3k - i, r ~ 2

E,: Hk(Fg,r) + Hk(Fg+l,r_2) is an i s o m o r p h i s m

for g > 3k, r >_ 2.

Corollary 2.3. Hk(F s r ) is i n d e p e n d e n t of g and r for

w,
g >_ 3k + I.

T h e p r o o f of a s t a b i l i t y theorem for ps requires suitable

g,r
 438

F complexes whose connectivity increases with g. Harer starts

with the c o m p l e x X a variant of t h e b a s i c complex Z. In o r d e r to
define X consider the sub cut system of r a n k k i.e. the isotopy
class of k + 1 disjoint simple loops C = { C O .... , C_~ such that
F - { C O , .... Cj_ is c o n n e c t e d . Define the simplicial complex X of
dimension g - 1 by t a k i n g a k-simplex for e v e r y rank k sub cut
system of F and identifying C as a f a c e o f C' if C cC,. The
first theorem is t h a t X has the h o m o t o p y t y p e of a w e d g e of g - 1
dimensional spheres. This is p r o v e n by first enlarging X to Z,
the analogous complex where F - C is n o w a l l o w e d to b e d i s c o n n e c t e d
but each component must have negative Euler characteristic. The
complex Z, the second of t h e t h r e e basic complexes (CS being the
first) , has dimension 3g - 4 + r + s and is 2g - 3 connected.
After studying several additional complexes the proof follows a
standard outline, [12].

2.5. For the s a k e of s i m p l i c i t y I shall only discuss one p~rt of

1
Harer's results on the virtual cohomological dimension: that M
g
has the h o m o t o p y t y p e of a 4g - 3 d i m e n s i o n a l s p i n e . The discussion
starts with the description of a F1 invariant ideal triangulation of ~
g
T I. The triangulation a r i s e s f r o m t h e f o l l o w i n g r e s u l t of S t r e b e l ,
g
[19] .

Theorem 2.4. Given a compact Riemann surface R and a point p

there exists a unique meromorphic quadratic differential ~ on R
such that

i) ~ has exactly o n e pole,

ii) in t e r m s of an appropriate complex coordinate z in a n e i g h -

2
dz
borhood of p ~ = 2 and
z
iii) the real trajectories of ~ are closed.

The differential ¢ may a l s o be d e s c r i b e d by starting with the

dz 2
differential .....2 on the d i s c D = {Izl ~ i} and identifying arcs
z
on ~D ( l i n e a r l y in r a d i a n m e a s u r e ) to o b t a i n the pair (R,~). To
see t h i s consider the following simple example (see F i g u r e 5) w h e r e
one obtains a surface of g e n u s 2; in g e n e r a l any pattern for a g e n u s
g surface will occur.
The data for a p a i r (R,~) is m e r e l y the combinatorial pattern
for i d e n t i f y i n g arcs on DD, as w e l l as t h e i r lengths. In o r d e r to
record this information for e a c h p a i r of a r c s consider a loop
 439

-#

h '
Figure 5

b a s e d at p, formed by rays from the o r i g i n w i t h endpoints on ~D

identified under the arc pairing and assign to ~ a weight w equal
to the length of the arcs on 3D; to (R,¢) assign the tuple
(Y].,w.).
] Specifically for the loop i n t e r s e c t i n g A1 in the above
example the picture is (see Figure 6)

Figure 6
 440

To i n t e r p r e t this situation the third basic complex A is

introduced. A rank k arc system {d0,...,~k} is t h e isotopy
class of k + 1 simple loops based at p, intersecting only at p,
and representing distinct, nontrivial homotopy classes. A is the
cell complex with a k simplex ~ = <~0,...,~k > for e a c h rank k
arc system and ~ is i d e n t i f i e d as a f a c e of ~' if ~ c ~'
Strebel's theorem provides a F equivariant map T 1 ÷ A. To the
g
pair (R,p) associate the quadratic differential ~ and to
associate the arc system {s 0 ..... ~ k } with weights, choosing one
arc for e a c h pair of segments of ~D occurring in the c o n s t r u c t i o n
of ¢. Simplicial coordinates on A are given simply by the
weights. The map T1 ÷ A is in f a c t a h o m e o m o r p h i s m onto the
g
complement of a subcomplex A ; the ideal triangulation is t h e p u l l -
back of t h e c e l l structure of A.
Recalling the construction of R by identifying segments on ~D,
observe that every pattern must have at l e a s t 4g segments. In
particular the rank k arc systems arising from Strebel's theorem
have rank ~ 2g - i; the 2g - 2 skeleton of A is c o n t a i n e d in
A . Now introduce the d u a l complex Y of A - A . Accordingly Y
will not have cells f o r the 2g - 2 skeleton of A. Thus the
dimension of Y is
d i m T 1 - (2g-l) = 4g - 3. Harer shows that
g
T1 may be F equivariantly retracted onto Y.
g
The c a s e of T1 provides a helpful example. A is the standard
SL(2;Z) tessellation of H, A is t h e rational points of IR = ~H
and Y is S e r r e ' s tree for SL(2;Z).

3. The Weil-Petersson Geometry.

3.1. Ideally the p u r p o s e of introducing an invariant metric o n the

Teichmuller space Ts is to p r o v i d e i n f o r m a t i o n o n the i n t r i n s i c ,
g
i.e. independent of the m e t r i c , p r o p e r t i e s of t h e s p a c e . I shall
try to indicate the extent that the Weil-Petersson metric has success-
fully filled this role. In b r i e f a sketch will be given of the
results in [2, 3, 17, 22-30]. Recently Fischer and Tromba have
independently undertaken an investigation of Teichmuller space and
the W-P metric, substituting the v i e w p o i n t of Riemannian geometry
f o r the classical theory of quasiconformal maps, [20, 21].
As background the reader may check [2, 3, i0, 21] for p r o o f s
that the metric is K ~ h l e r . Recall that the W-P metric is i n v a r i a n t
for the a c t i o n of Fs
Ts on
a n d t h u s d e s c e n d s to t h e m o d u l i s p a c e
of Riemann s u r f a c e s gM s. g
On both Ts and Ms t h e m e t r i c is n o t
g g g
complete, [22]. N o w this r e s u l t is b e s t u n d e r s t o o d in t e r m s o f
 441

the latter results [17, 26] that the metric has an e x t e n s i o n to

Ms, the D e l i g n e - M u m f o r d stable curve c o m p a c t i f i c a t i o n of
Ms.
g g
The study of the W-P g e o m e t r y can be divided into three m a j o r
topics: curvature considerations, e x t e n s i o n of the m e t r i c to the
compactification ~
Mg, and the symplectic g e o m e t r y of the Kihler
form. As a sample of the results I shall start by sketching three
ap_plications. The K i h l e r form ~WP extends to the c o m p a c t i f i c a t i o n
MS; the e x t e n s i o n e--WP defines a c o h o m o l o g y class in H 2 ( ~ ) ,
g 1 g
[26, 27, 28]. In [27] it is found that 2 ~WP is a c t u a l l y the
2~
first Chern class of a known line bundle <I (discussed below) on
--g
M In p a r t i c u l a r 1 - is a rational class and the line bundle
g" 2 2 ~WP
~i is positive. At this point the Kodaira__ t h e o r e m may be quoted to
o b t a i n a purely a n a l y t i c proof that M s is p r o j e c t i v e algebraic. As
g
a second a p p l i c a t i o n c o n s i d e r the N i e l s e n conjecture: every finite
subgroup of F s fixes a point of T s. An i m p o r t a n t i n g r e d i e n t of
g g
K e r c k h o f f ' s proof of the c o n j e c t u r e is that the geodesic length
functions i, are convex along T h u r s t o n ' s e a r t h q u a k e paths, [16].
I have r e c e n t l y found that the length functions i, are strictly
convex along the W-P geodesics [30]. This result p r o v i d e s the
basis of an i n d e p e n d e n t but similar proof of the conjecture. The
proof starts with an o b s e r v a t i o n of F r i c k e - K l e i n : that a suitable
sum S = ~ Z of length functions will be a proper function on T s.
g
Given ~ ~ £g,S
3 a finite subgroup, then the sum S O = Y ~ S(y) is
invariant and is also a sum of length functions. Now SO is
strictly convex along W-P g e o d e s i c s and thus a critical p o i n t is
necessarily a relative minimum (S O is an index 0 M o r s e function).
Since SO is proper it follows that it has a unique m i n i m u m and
finally the ~ i n v a r i a n c e of SO g u a r a n t e e s that the m i n i m u m is
fixed by ~, the d e s i r e d conclusion. A n d finally since the W-P
m e t r i c is Kahler it follows immediately that the length functions
~, are in fact p l u r i s u b h a r m o n i c ; this o b s e r v a t i o n leads to a new
proof that Ts is a Stein manifold, [30].
g

3.2. Ahlfors was the first to c o n s i d e r the c u r v a t u r e of the metric;

he o b t a i n e d singular integral formulas for the Riemann c u r v a t u r e
tensor, [3]. As an a p p l i c a t i o n he found that the Ricci, h o l o m o r p h i c
sectional and scalar c u r v a t u r e s are all negative. Royden later
showed that the h o l o m o r p h i c s e c t i o n a l c u r v a t u r e is b o u n d e d away from
zero and m o r e r e c e n t l y T r o m b a has found that the general sectional
c u r v a t u r e is indeed negative. I shall now present a simple formula
 442

for the c u r v a t u r e tensor, [29]. Recall first that the h o l o m o r p h i c

tangent space T I'0 T s at R ~ T s is i d e n t i f i e d w i t h the space of
g g
harmon i c Beltrami differentials H(R) a n d that dA denotes the
hyperbolic area e l e m e n t on R and D the h y p e r b o l i c Laplacian.
The R i e m a n n curvature is a 4-tensor in p a r t i c u l a r for ~' ~B' ZY'
UB 6 H(R)
r -1
Re,By ~ - -2 ] (D-2) (~a~8) ( ~ y ~ ) dA
R
r -i
= -2 I (m-2) (lJc~ ~) (Py~B)di
JR

where (D-2) -I is the indicated self adjoint operator and observe

for ~, ~ E H that the p r o d u c t ~ is a function. Starting with
the above formula it follows that the m e t r i c has n e g a t i v e sectional
curvature, that the h o l o m o r p h i c s e c t i o n a l and Ricci c u r v a t u r e s are
-i
bounded above by 2~(g-l) and that the scalar c u r v a t u r e is b o u n d e d

above by -3(3g-2)
4~ In fact the a r g u m e n t s show that the c u r v a t u r e s
are g o v e r n e d by the s p e c t r u m of the Laplacian: the n e g a t i v e curva-
ture is a m a n i f e s t a t i o n of the n o n p o s i t i v i t y of the Laplacian. These
last results have also been o b t a i n e d by Royden.
As a further application of the t e c h n i q u e s I wish to c o n s i d e r the
characteristic classes of the T e i c h m u l l e r curve T . T is the
^ g g
natural fibre space over T ; the fibre above R ~ T is a c o m p a c t
g g
submanifold isomorphic to R. If ~ :T ÷ T is the p r o j e c t i o n
g g
then the kernel of the d i f f e r e n t i a l d~ : TI, 0 T + TI,0T defines a
g g
line b u n d l e (v) on T , the v e r t i c a l b u n d l e of the fibration.
g
The r e s t r i c t i o n of (v) to a fibre of z is simply the tangent
bundle of the fibre; by the u n i f o r m i z a t i o n theorem the h y p e r b o l i c
metric induces a metric on (v). I have c o m p u t e d the c u r v a t u r e
2-form for this m e t r i c and found that it is negative, a pointwise
version of A r a k e l o v ' s result that the dual (v)* is
numerically effective, [4]. Once again the c u r v a t u r e is g o v e r n e d
by the s p e c t r u m of the Laplacian. By i n t e g r a t i n g powers of the
Chern class Cl(V) over the fibers of ~ one obtains Fg invariant
characteristic classes <n(P) = f -l(p) Cl(V)n+l defined on Tg.

Mumford has m a n y intriguing results on the b e h a v i o r of these

classes in the c o h o m o l o g y ring: in p a r t i c u l a r the ~ , n S 3g - 3
n
are n o n t r i v i a l and m a n y geometrically interesting cycles may be
written in terms of the < . Mumford guesses that the low d i m e n s i o n a l
n
 443

part of the cohomology ring may actually be p o l y n o m i a l in the K .

1 n
As an example in [29] I find that in fact <i 2 ~WP' eWP the
W-P Kahler form. 2z

3.3. Masur was the first to c o n s i d e r the extension of the W-P

metric from Mg to ~[a' ~g the moduli space of stable curves. In
the paper [17] Masur develops the foundations of this topic. Recall
now that the c o m p a c t i f i c a t i o n locus D = M - M is a divisor with
g g
normal crossings. Briefly Masur shows if D is given locally as
zI = 0 in a c o o r d i n a t e chart z = (z I, .... z n) then

dS~p = dZld~l
lZll2(iog1/l~ll)3+ O(iZli(logi/Izli)l31, [17].
In p a r t i c u l a r the W-P length of a d i f f e r e n t i a b l e curve in ~ is
g
finite and the metric extends to a c o m p l e t e metric on M . Starting
g
from Masur's results it follows that the K'ahler form ~ extends
WP
to a closed, positive, (i,i) current ~I4P on ~ , [26]. In fact
_ g
the above s i n g u l a r i t y is s u f f i c i e n t l y m i l d for ~WP to be w r i t t e n
as ~F, F a continuous function (as an example c o n s i d e r
1
F= log i/[zr),~
~ [28]. Standard a p p r o x i m a t i o n techniques may then be
applied to c o n c l u d e that ~WP is the limit of smooth Kahler forms
in its c o h o m o l o g y class. Thus even though the Kahler form is not
smooth it is suitable for applications in p a r t i c u l a r the Kodaira
theorem may be quoted to c o n c l u d e that there is a p r o j e c t i v e embedding.
N o w the d i s c u s s i o n will continue w i t h the results on d e s c r i b i n g the
c o h o m o l o g y class of
WP"

3.4. The divisor D c M is reducible D = D O U...U P , whore

g [~J
the generic surface r e p r e s e n t e d in Dk has one node separating it
into components of genus k and genus g - k (for k = 0 the node
is nonseparating). C e r t a i n l y the d i v i s o r s ~k define c o h o m o l o g y
classes in H6g_8(Mg) and by Poincar6 duality (over ~, ~g is a
V-manifold) ~WP also defines a class in H6g_8. The first result
is contained in the following, [27].

T h e o r e m 3.1. {D O .... ,D[~], (~Wp)} is a basis for H6g_8(Mg;~).

The sketch of the proof is simple enough. By the result of

Harer on H2(M ) and an a p p l i c a t i o n of M a y e r - V i e t o r i s one v e r i f i e s
g
at the o u t s e t that H2(~g;~) has rank 2 + [~]'z A c a n d i d a t e basis
 444

is then p r e s e n t e d for each of H2 and H6g_8 and the i n t e r s e c t i o n

pairing is evaluated. The pairing is found to be n o n s i n g u l a r and
the proof is complete. The trick for e v a l u a t i n g the integrals of
~WP is to p e r f o r m a single i n t e g r a t i o n [27] and then deduce the
remaining integrals by formal properties.
As an example a 2-cycle E for M is o b t a i n e d by c o n s i d e r i n g
g
the family of (stable) curves given as the one point sum of a fixed
surface SO of genus g - 1 and an elliptic curve E, w h i c h will
vary over all (even degenerate) structures r e p r e s e n t e d in its m o d u l i
space M~ (see Figure 7).

So E

Figure 7

The 2-cycle E is p a r a m e t r i z e d by 1
M I. Now to state the d e s i r e d
formal property of ~WP let s be the W-P K~hler form for Ms .
g g
Then briefly 0
~glE = ~ l1l M--;
~ the r e s t r i c t i o n of __~0 g to E is

n a t u r a l l y identified with the Kahler __f°rm for M~. And so the

I
integral fE ~WP is reduced to the M~± case. This last integral
~2
may be e v a l u a t e d directly; the value is -~, [25]. The formal
properties of the Kahler form will be d i s c u s s e d further in section
3.6 as c o n s e q u e n c e s of the Fen c h e l - N i e l s e n coordinates.
After e v a l u a t i n g the i n t e r s e c t i o n p a i r i n g one finds that

1 -- : 71
272 ~WP

where ~i is the extension of the class K1 d i s c u s s e d in section

3.2. Indeed the above is the g e n e r a l i z a t i o n to M
of the earlier
g
result l-l--2
2 ~WP = <I for Mg. Finally the basic techniques for
 445

c o n s t r u c t i n g cycles and c o m p u t i n g i n t e r s e c t i o n s may be applied to the

higher h o m o l o g y groups. For instance in [27] it is shown that
H2k(Mg) , k < g has rank at least gk I

3.5. The last two sections will be d e v o t e d to the symplectic g e o m e t r y

of the Kahler form ~WP" The symplectic g e o m e t r y of the triple
(ewp,t,,i,), t, the F e n c h e l - N i e l s e n vector fields and ~, the
g e o d e s i c length functions, is dual to the trigonometry, as will be
d e s c r i b e d below, of geodesics in the h y p e r b o l i c metric of a surface.
A c o n s t r u c t i o n of F e n c h e l - N i e l s e n provides for natural flows on
T e i c h m u l l e r space. Fix the free h o m o t o p y class of a n o n t r i v i a l simple
loop {e} on F and an i n c r e m e n t 6 of time. If
R = { (R,f)} ( T g
then {f(~)} is r e p r e s e n t e d on R by a unique g e o d e s i c e~. Cut R
open along ~R' rotate one side of the cut relative to the other (by
a distance of 6) and then glue the sides in their new position. The
hyperbolic structure in the complement of the cut extends n a t u r a l l y
to define a h y p e r b o l i c structure on the new surface. A geodesic
intersecting e~ is d e f o r m e d to a b r o k e n g e o d e s i c Yb with endpoints
separated 6 units along ~. As $ varies a flow, the F-N flow,
is defined on the T e i c h m ~ l l e r space T . The i n f i n i t e s i m a l g e n e r a t o r
g
of the flow is the F-N vector field t . The free h o m o t o p y class
{e} also determines a function Z , the geodesic length function,
on Tg. In brief define Ze(R), R ( Tg to be the length of e~ and
the exterior d e r i v a t i v e di will also be discussed. The basic
formula of the symplectic g e o m e t r y is a d u a l i t y formula

~wp(t , ) = -dZ , [23, 24, 26].

An immediate c o n s e q u e n c e is that the symplectic form is invariant

under the F-N flows on Tg, in p a r t i c u l a r the flows are W-P
volume preserving. There are also formulas for the Lie d e r i v a t i v e s
t~B and t t ~ 7, [24].

~(t ,t B) = t Z = [ cos @ (3.1)

8 p(~#8 P
 446

41 Z2
t t~ = ~ e +e sin @ sin @ (3.2)
P q
Y (P'q) E~#Yx~#7 2(e ~-i)

mI m2
_[ e +e
sin ~ sin 8 .
r s
(r,s)(~#Bx~#Y 2 (e~_l)

The right hand side of (3.1) e v a l u a t e d at R E T is the sum of

g
cosines of the angles at the i n t e r s e c t i o n s of the g e o d e s i c s ~ and
B~. Similarly the right hand side of (3.2) is a sum of t r i g o n o m e t r i c
invariants for pairs of intersections; ~i and ~'2 are the lengths
of the segments of y defined by p, q and likewise, for mI and
m2 relative to 8. R e c e n t l y W. Goldman has g e n e r a l i z e d these formulas
to the r e p r e s e n t a t i o n space Hom(zI(F),G)/G, G a Lie group w i t h
n o n d e g e n e r a t e symmetric b i l i n e a r form on its Lie algebra, [i0]. Con-
sequences of the above formulas are c o n s i d e r e d in [25, 26]. In
p a r t i c u l a r if ~ # ~ @ ~ then t t ZB > 0 and (3.2) represents a
q u a n t i t a t i v e version of K e r c k h o f f ' s o b s e r v a t i o n that the geodesic
length functions are convex along e a r t h q u a k e paths, [16]. Finally
note that the i n f i n i t e s i m a l g e n e r a t o r s of T h u r s t o n ' s e a r t h q u a k e flows
form the c o m p l e t i o n (in the c o m p a c t - o p e n topology) of the F-N vector
fields.

3.6. Introducing coordinates on T e i c h m u l l e r space is a q u e s t i o n of

p a r a m e t r i z i n g Riemann surfaces. Fenchel-Nielsen suggested a particu-
larly simple solution to this problem. It starts w i t h the o b s e r v a t i o n
that the lengths of a l t e r n a t i n g sides of a right hexagon in the
h y p e r b o l i c plane may be chosen arbitrarily. Given such a hexagon,
form its metric double across the r e m a i n i n g sides to obtain the basic
object P, a pair of pants (see Figure 8).

23
Figure 8
 447

Topologically P is the c o m p l e m e n t of three d i s j o i n t discs in $2;

metrically P is a h y p e r b o l i c surface w i t h geodesic boundary. The
key o b s e r v a t i o n is that pants P and P' may be m e t r i c a l l y summed
along their b o u n d a r i e s p r o v i d e d merely that the b o u n d a r i e s are of
equal length. Now fixing a c o m b i n a t o r i a l pattern, then summing the
pants Pl,...,P2g_2, one obtains the general genus g surface
(see Figure 9).

Figure 9

The coordinates for T

are simply the free p a r a m e t e r s for this
g
construction. There are exactly two p a r a m e t e r s at each summing locus.
Of course the first is simply the length Z of the locus, this
varies freely in IR+. The second, the twist p a r a m e t e r T, measures
the net d i s p l a c e m e n t between the boundaries. The p a r a m e t e r T is
d e f i n e d to be the h y p e r b o l i c d i s t a n c e b e t w e e n the feet of p e r p e n d i c u l a r s
d r o p p e d from a p p r o p r i a t e b o u n d a r i e s (see Figure i0).

Figure i0
 448

After an initial choice T is d e t e r m i n e d by analytic c o n t i n u a t i o n

and varies freely in IR. In this way F e n c h e l - N i e l s e n e s t a b l i s h e d
the following result, [i, 26].

T h e o r e m 3.2. The map Tg ÷ (~+× IR) 3g-3 given by R ÷ 3g-3

(£j,Tj)j= 1
is a homeomorphism.

In p a r t i c u l a r T is a cell. F u r t h e r m o r e the D e l i g n e - M u m f o r d
_ g
compactification M can be c o n s t r u c t e d from M by simply allowing
g g
the length p a r a m e t e r s Z, to vanish, [1,26]. The d i s c u s s i o n of the
p r e v i o u s section already suggests a r e l a t i o n s h i p between the Kahler
form WWp and the F-N coordinates. Recall that ~WP is i n v a r i a n t
under the F-N vector fields and observe that the coordinate vector
fields ~ are indeed F-N vector fields. C o n s e q u e n t l y the coeffi-
3
cients of ~WP in F-N coordinates (Tj,Zj) must be i n d e p e n d e n t of
the twist variables T.. In fact much more is true, [26].
3

T h e o r e m 3.3. ~WP = [
j d Z .3 A d T .3
.

Briefly (~,T.,Z.) is a c o m p l e t e l y integrable H a m i l t o n i a n system.

3 3
The Kahler form ~WP is F i n v a r i a n t in p a r t i c u l a r the 2-form
g
dZ. ^ dT. is i n d e p e n d e n t of the c o m b i n a t o r i a l pattern for c o m b i n i n g
j 3 3
pants. Finally the d i s c u s s i o n will be concluded w i t h two applications
of the above formula. Set a length p a r a m e t e r Zk equal to zero to
o b t a i n a d e g e n e r a t e surface S (see Figure ii)

Figure ii
 449

the sum at p u n c t u r e s of s u r f a c e s S1 and S 2. It follows i m m e d i a t e l y

from the above f o r m u l a that ~S converges to the sum+ ~S as~S
1 2
Z k + 0. Briefly stated the K a h l e r form for a sum of s u r f a c e s is the
sum of the c o m p o n e n t forms, [27]. To d e m o n s t r a t e the s e c o n d formal
property, behavior for an u n r a m i f i e d c o v e r i n g R ÷ S of surfaces,
c o n s i d e r the f o l l o w i n g e x a m p l e (see F i g u r e 12).

Figure 12

R is a 4 punctured torus, S is a one p u n c t u r e d torus and the

covering transformation is the order 4 r o t a t i o n about the axis of
the hole. Introducing F-N coordinates r e l a t i v e to the i n d i c a t e d
loops the r e a d e r w i l l e a s i l y check that ~R = 4eS; the K a h l e r form
multiplies under u n r a m i f i e d c o v e r i n g s , [27].
 450

REFERENCES

1 W. Abikoff, Topics in the Real A n a l y t i c T h e o r y of T e i c h m u l l e r

Space, L.N.M. 820 S p r i n g e r - V e r l a g , N e w York, 1980.
2 L . V . Ahlfors, Some remarks on T e i c h m u l l e r ' s space of Riemann
surfaces, Ann. of Math., 74 (1961), 171-191.
3 L . V . Ahlfors, Curvature properties of T e i c h m u l l e r space, J.
Analyse Math., 9 (1961), 161-176.
4 S. Arakelov, F a m i l i e s of a l g e b r a i c curves w i t h fixed degeneracies,
Izv. Akad. Nauk., 35 (1971).
5 L. Bers, On b o u n d a r i e s of T e i c h m u l l e r spaces and on K l e i n i a n
groups, I, Ann. of Math., 91 (1970), 570-600.
6 L. Bers, Fibre spaces over T e i c h m u l l e r spaces, Acta. Math., 130
(1973), 89-126.
7. M. C. Chang and Z. Ran, U n i r a t i o n a l i t y of the m o d u l i spaces of
curves of genus ii, 13 (and 12), Invent. Math., to appear.
8 R. C h a r n e y and R. Lee, C o h o m o l o g y of the Satake compactification,
Topology, 22 (1983), 389-423.
9 D. E i s e n b u d and J. Harris, Limit linear series, the i r r a t i o n a l i t y
of M and other applications, Bull. A.M.S., i0 (1984), 277-280.
g
10 W. M. Goldman, The s y m p l e c t i c n a t u r e of f u n d a m e n t a l groups of
surfaces, preprint.
ll J. Hater, The second h o m o l o g y group of the m a p p i n g class g r o u p of
an o r i e n t a b l e surface, Invent. Math., 72 (1983), 221-239.
12 J. Hater, S t a b i l i t y of the h o m o l o g y of the m a p p i n g class groups
of o r i e n t a b l e surfaces, Ann. of Math., to appear.
13 J. Harer, The virtual c o h o m o l o g i c a l d i m e n s i o n of the m a p p i n g class
groups of o r i e n t a b l e surfaces, preprint.
14 W. Harvey, B o u n d a r y structure for the m o d u l a r group, Ann. of Math.
Studies, 97 (1978), 245-251.
15 A. Hatcher and W. Thurston, A p r e s e n t a t i o n for the m a p p i n g class
group of a closed o r i e n t a b l e surface, Topology, 19 (1980), 221-
237.
16. S. P. Kerckhoff, The N i e l s e n realization problem, Ann. of Math.,
117 (1983), 235-265.
17. H. Masur, The e x t e n s i o n of the W e i l - P e t e r s s o n m e t r i c to the b o u n d a r y
of T e i c h m u l l e r space, Duke Math. J., 43 (1976), 623-635.
18. E. Miller, The h o m o l o g y of the moduli space and the m a p p i n g class
group, preprint.
19. K. Strebel, On q u a d r a t i c d i f f e r e n t i a l s with c l o s e d t r a j e c t o r i e s
and second order poles, J. A n a l y s e Math., 19 (1967), 373-382.
20. A. E. F i s c h e r and A. J. Tromba, On a p u r e l y "Riemannian" proof of
the structure and d i m e n s i o n of the u n r a m i f i e d m o d u l i space of a
compact Riemann surface, Math. Ann., 267 (1984), 311-345.
21. A. E. F i s c h e r and A. J. Tromba, On the W e i l - P e t e r s s o n metric on
T e i c h m ~ l l e r space, preprint.
22. S. A. Wolpert, N o n c o m p l e t e n e s s of the W e i l - P e t e r s s o n m e t r i c for
T e i c h m u l l e r space, Pac. J. Math., 61 (1975), 573-577.
23. S. A. Wolpert, The F e n c h e l - N i e l s e n deformation, Ann. of Math.,
115 (1982), 501-528.
 451

24 S. A. Wolpert, On the s y m p l e c t i c g e o m e t r y of d e f o r m a t i o n s of a
h y p e r b o l i c surface, Ann. of Math., 117 (1983), 207-234.
25 S. A. Wolpert, On the K a h l e r form of the m o d u l i space of o n c e
p u n c t u r e d tori, Comment. Math. Helv., 58 (1983), 246-256.
26 S. A. W o l p e r t , On the W e i l - P e t e r s s o n g e o m e t r y of the m o d u l i
space of curves, Amer. J. Math., to appear.
27 S. A. W o l p e r t , On the h o m o l o g y of the m o d u l i space of stable
curves, Ann. of Math., 118 (1983), 491-523.
28 S. A. W o l p e r t , On o b t a i n i n g a p o s i t i v e line b u n d l e from the W e i l -
P e t e r s s o n class, Amer. J. Math., to appear.
29 S. A. W o l p e r t , C h e r n forms and the R i e m a n n t e n s o r for the m o d u l i
space of curves, p r e p r i n t .
30 S. A. W o l p e r t , Geodesic length functions and the N i e l s e n problem,
preprint.
31 A. Borel, S t a b l e real c o h o m o l o g y of a r i t h m e t i c groups, Ann. Sci.
Ecole Norm. Sup. 4e(7) (1974), 235-272.
Lecture Notes in Mathematics Vol. 1111
ISBN 978-3-540-15195-1 
c Springer-Verlag Berlin Heidelberg 2008

F. Hirzebruch

Arbeitstagungen

Addendum
Lecture Notes Vol. 1111 (Springer) includes the programmes of all 25
Arbeitstagungen held between 1957 and 1984. The archives for AT 57 and
AT 58 were very incomplete. However N. Kuiper had made a note for himself
of the topics of the talks and communicated these to me in a letter of June
22nd 1991.

Arbeitstagung 1957

13.–20. Juli 1957

Atiyah: 1) Non-projective algebraic surfaces

2) Chern classes and connections
Grauert: Real cross-sections which are analytic
Grothendieck: Kohärente
¨ Garben und verallgemeinerte
Riemann-Roch-Hirzebruch-Formel auf algebraischen
Mannigfaltigkeiten. Vier lange Vorträge
(4 long lectures)
Kuiper: A generalisation of convexity
Tits: Geometrie von Ausnahmegruppen
∧
Hirzebruch: cn (η ) = (n − 1)! generator H 2n (S 2n , Z)
a-2 Addendum

Arbeitstagung 1958

9.–16. Juli 1958

Bott: Homotopy groups of classical groups

Grauert: Levi’s problem
Grothendieck: Rational sections in fibre spaces
Kervaire: Parallelisability of spheres
Milnor: Computations on cobordism
D. Puppe: Non-additive functors and cross-effects
Remmert: Plurisubharmonic functions
Serre: Intersection theory
Stein: Komplexe Basen zu holomorphen Abbildungen
Thom: Combinatorics and C ∞ -manifolds
Hirzebruch: Applications of Milnor’s results
 Anhang

Programme

der

25 ArbeZJ~stagungen von 1957 - 1984

 454

I. M a t h e m ~ c h e ~be~%ta~!u~~ 1957

Die A k t e n enthalten kein o f f i z i e l l e s Programm. Aus der K o r r e s p o n d e n z

konnten folgende Vortragsthemen festgestellt werden.

I) Vortragsserie A. Grothendieck: Koh~rente Garben und v e r a l l g e -

meinerte Riemann-Roch-Hirzebruch-Formel auf
algebraischen Mannigfaltigkeiten

2) M.F. Atiyah sprach fur e i n e n g r @ S e r e n Kreis ~ber das T h e m a

"Some examples of complex manifolds"

Dieser Vortrag von A t i y a h wurde in den Bonner Mathematischen

Schriften Nr. 6 (1958) ver~ffentlicht.

Das P r o g r a m m der Tagung sah u n g e f ~ h r wie folgt aus:

Samstag, den 13.7.:

17.30 Uhr A. Grothendieck

Sonntag, 14.7.:
10.15 Uhr Ausflug mit D a m p f e r f a h r t

Montag, 15.~±:
9.30 Uhr A. G r o t h e n d i e c k
15.15 Uhr M.F. Atiyah
17.15 Uhr N. Kuiper

Dienstag, 16.7.:
11.30 Uhr A. Grothendieck
15.15 Uhr F. H i r z e b r u c h

Mittwoch, 17.7.:
11.30 Uhr A. Grothendieck
15.15 Uhr M.F. At iyah

Freitag, 19.7.: J. Tits

Samstag, 20.7.: M.F. A t i y a h

H. Grauert
 455

2. Mc~hemat~che ArbeZ~@~taguv~ 1958

Die Akten enthalten kein gedrucktes Programm. Die Tagung fand

vom 9. bis 16. Juli 1958 start. Nach privaten Aufzeichnungen haben
die folgenden Herren vorgetragen:

S. A b h y a n k a r
R. Bott
H. Grauert
A. Grothendieck
M. Kervaire
J. Milnor
D. Puppe
R. Remmert
J.-P. Serre
K. Stein
R. Thom
 456

3. Mathemat~che Arbe~staguung~ 1959

Die Akten enthalten nur folgende Angaben:

Samstag, den 11.7.:

17.30 - 18.30 Uhr A. Borel: Something on transformation groups

Sonntag, den 12.7.:

10.30 - 11.30 Uhr J.-P. Serre: Fundamental groups
12.00 - 13.00 Uhr D. Puppe: Semisimplicial monoid complexes

Montag, den 13.7.:

9.30 - 10.30 Uhr S. L a n g : " ....
11.00 - 12.00 Uhr J. Tits: " ....

Dienstag, den 14.7.:

10.00 - 11.00 Uhr K. Stein: On proper maps
11.30 - 12.30 Uhr J.-P. Serre: Proalgebraic groups
17.00 - 17.30 Uhr Pictures
17.30 - 18.30 Uhr M.F. Atiyah: Fibre homotopy type

Mittwoch, den 15.7.:

I0.00 - 11.00 Uhr A. Borel: On torsion in L i e - g r o u p s
11.30 - 12.30 Uhr H. Grauert: How to blow down
17.30 - 18.30 Uhr J. Milnor: Scissor and paste arguments

Nach privaten Aufzeichnungen haben auBerdem die Herren J.F. Adams,

A. Dold und N. Kuiper vorgetragen.
 457

4. Mathemat~che A r b e ~ s t a g u n g 1960 (11. - 17. Juni]

M.F. Atiyah : Sheaves and vector bundles

J. Milnor: Spherical modifications

S. Smale: Poincar6's conjecture in higher dimensions (I)
M. Kervaire: A manifold without d i f f e r e n t i a b l e structures (I)

J. Stallings: Combinatorial homotopy-spheres

S. Smale: Poincar6's conjecture in higher dimensions (II)
M. Kervaire: A manifold without differentiable structure (II

H. Grauert: Some p r o b l e m s concerning non-compact K~hler manifolds

S. Lang: Cross-sections on algebraic families of curves
Pictures
A. Dold: Representable functors and t e n s o r - p r o d u c t s

R. Thom: Singular homology and sheaf theory

R. Remmert: Rigid complex manifolds

J.F. Adams: Old-fashioned topology

A. Borel: Non-singular homology and sheaf theory
J.P. Serre: L-series

N. Kuiper: Curvature of index

R. Bott: Cauchy formula
M. Hirsch: Embedding theorems
 458

5. Mathemati~che Arbeitsta@ung 1961 (16.- 23. Ju~_ )

A. Borel: Arithmetic s u b g r o u p s of Lie g r o u p s I

J. Milnor: The h a n d l e body t h e o r e m I

M.F. Atiyah: Finiteness t h e o r e m for e o m a p c t Lie g r o u p s
J. Stallings: A 5-dimensional example a g a i n s t the H a u p t v e r m u t u n g

A. Borel: Arithmetic subgroups of Lie g r o u p s II

J. Milnor: The h a n d l e b o d y t h e o r e m II
J. Milnor: Handle body t h e o r e m and results of Wall

A. A n d r e o t t i : Vanishing theorems
M. Kneser: Approximation in a l g e b r a i c g r o u p s
M. Kneser: S i e g e l ' s t h e o r e m and T a m a g a w a n u m b e r s
A. G r o t h e n d i e c k : Schemes of m o d u l i

B. Eckmann: Lusternik-Schnirelmann category

J. Tits: Flags and B r u h a t ' s theorem
R. Palais: C o n j u g a c y of c o m p a c t d i f f e o m o r p h i s m g r o u p s

J. Kohn: P o t e n t i a l theory on non c o m p a c t c o m p l e x m a n i f o l d s

A. G r o t h e n d i e c k : Duality theorems in a l g e b r a i c g e o m e t r y
R. Remmert: On h o m o g e n e o u s c o m p a c t K~hler m a n i f o l d s

N. Kuiper: Manifolds admitting functions with few critical p o i n t s

A. Shapiro: Graded Clifford modules

F. Peterson: Squaring operations in a sphere bundle

I. Porteous: H o m o m o r p h i s m s of v e c t o r b u n d l e s
 459

6. Mathematische Arbeit~ta~ung 1962 [13. - 20. Jubi)

S. Lang: On the Nash e m b e d d i n g theorem ~ la Moser

S. Smale: Stable manifolds of a d i f f e o m o r p h i s m

H. Hironaka: Resolution of s i n g u l a r i t i e s
R. Swan: The G r o t h e n d i e c k ring of a finite group

M.F. Atiyah: Harmonic spinors

C.T.C. Wall: Classification problems in d i f f e r e n t i a l topology
R. Remmert: On h o m o g e n e o u s compact complex manifolds

J. Stallings: P i e c e w i s e linear approximation of stable

homeomorphisms
M. Kervaire: 2-spheres in 4-manifolds
Pictures
K. Stein: Extension of m e r o m o r p h i c mappings

A. Haefliger: Links
R. Abraham: Transversality of m a p p i n g s
H. Grauert: Rigid singularities

J. Eells -
Harmonic maps
J.H. Sampson:

W. Browder: Homotopy type of d i f f e r e n t i a b l e manifolds

M.F. Atiyah: Explanations of my p r e c e d i n g lecture
N. Kuiper: Smoothing problems

A. Kosinski: Piecewise linear functions on c o m b i n a t o r i a l manifolds

J.C. Moore: Hopf algebras
 460

7. Mathematische Arbeitstagung 1963 (14. - 21. Juni)

S. Lang: T r a n s c e n d e n t a l numbers

M.F. Atiyah: Elliptic operators

J. Eells: D e f o r m a t i o n s of maps
M. Kervaire: Higher dimensional knots

F. Oort: On a u t o m o r p h i s m s of v a r i e t i e s
H. Grauert: On S u p e r - C o c y c l e s
S. Lang: On Manin's theorem

A. Haefliger: Combinatorial manifolds

J. Cerf: F4 = 0
Pictures
R. Palais: I n t e g r o - d i f f e r e n t i a l operators

F. Hirzebruch: E l e m e n t a r y proof of Bott's p e r i o d i c i t y theorem

S. Abhyankar: Jungian s i n g u l a r i t i e s
A. Dold: O b s t r u c t i o n theory for cohomology functors

R. Thom: On generic s i n g u l a r i t i e s of envelopes (with slides)

R. Palais: Integro-differential operators (If)

M.F. Atiyah: B o u n d a r y value problems
N. Kuiper: Smoothing of c o m b i n a t o r i a l manifolds

R. Palais: Morse theory

V. Poenaru: T h i c k e n i n g and u n k n o t t i n g
i. Cs~sz[r: Allgemeine A p p r o x i m a t i o n s s ~ t z e
 461

8. M a t h e m ~ c h e Arbeit~ta~u~{:g 1964 (15. - 19. J u ~ ]

A. Borel: I n t r o d u c t i o n to a u t o m o r p h i c forms

N. Kuiper: The u n i t a r y g r o u p of H i l b e r t space is k - c o n n e c t e d

R. Wood: G e n e r a l i z a t i o n of B o t t ' s p e r i o d i c i t y theorem
S.S. Chern: Holomorphic vector bundles

W. Haken: Poincar~'s conjecture in d i m e n s i o n three

M. Kneser: G a l o i s c o h o m o l o g y of p - a d i c linear g r o u p s
W. Browder: Introduction to c o b o r d i s m t h e o r y

D. Gromoll: Exotic spheres and m e t r i c s of p o s i t i v e curvature

J. Eells: Variational t h e o r y on m a n i f o l d s
Pictures
R. S a c k s t e d e r : Foliated manifolds

K. Jinich: (X,Fredholm) = K(X)

E. Thomas: E n u m e r a t i o n of v e c t o r b u n d l e s
J.F. Adams: Some a p p l i c a t i o n s of K - t h e o r y to h o m o t o p y t h e o r y
D. Anderson: S e v e r a l a s p e c t s of K - t h e o r y

R. Wood: Pre-Palais
R. Palais: Index t h e o r e m for e l l i p t i c b o u n d a r y p r o b l e m s
A. Borel: Pseudo-concavity for a r i t h m e t i c groups

A. Kosinski: Is the H a u p t v e r m u t u n g true for m a n i f o l d s ?

Shih Weishu: Characteristic classes in K - t h e o r y

R. Palais: The h o m o t o p y type of some infinite dimensional
manifolds
 462

9. M a t h e m ~ c h e A~beZl~stagun~ 1965 (19. - 25. Juni)

M.F. Atiyah: Elliptic operators on m a n i f o l d s and g e n e r a l i z e d

Lefschetz fixed p o i n t theorem I

A. Douady: An hour of c o u n t e r e x a m p l e s
M.F. Atiyah: Elliptic operators on m a n i f o l d s and g e n e r a l i z e d
Lefschetz fixed point theorem II
S. Lang: Division points on curves

W. K]ingenberg: Closed geodesics

R. Thom: Topological models for m o r p h o g e n e s i s in b i o l o g y
H. Hironaka: Projectiveness criterions of K l e i m a n

A. Grothendieck: Riemann-Roch I
J.P. Serre: Formal groups

A. Grothendieck: Riemann-Roch II
J. Milnor: Projective class groups in t o p o l o g y
N. Kuiper: Piecewise linear m i c r o b u n d l e s are bundles

G. Harder: Galois cohomology of semi-simple groups

I.G. Macdonald: Spherical functions on p-adic groups
D. Epstein: Duality theorems for abelian schemes

A. Van de Ven: Almost complex manifolds

D. Husemoller: Cohomology theory
G. Segal: Equivariant K-theory
R. Palais: Symplectic manifolds (im Rahmen des M a t h e m a t i s c h e n
Kolloquiums)
 463

10. Mathemati~che Arbei;Lsta~un~ 1966 (16. - 22. Juni)

M.F. Atiyah: G l o b a l a s p e c t s of e l l i p t i c operators

H. Grauert: Non-archimedean analysis

J. Eells: Deformations of s t r u c t u r e s
M. Kervaire: Congruence subgroups after H. Bass

J.P. Serre: 1-adic Galois groups

L. Siebenmann: Applications of W a l l ' s invariant
S. Smale: On the structure of d i f f e o m o r p h i s m s

M.F. Atiyah: Index t h e o r e m (I)

J. Tate: p-adic Galois representations
R. Bott: Vector fields and c h a r a c t e r i s t i c numbers

M.F. Atiyah: Index t h e o r e m (II)

F. H i r z e b r u c h : Exotic spheres and s i n g u l a r i t i e s

S. Smale: Group c o n s t r u c t i o n s in the theory of d i f f e o m o r p h i s m s

A. Borel: R i g i d i t y of a r i t h m e t i c groups
J. Milnor: S i n g u l a r i t i e s of h y p e r s u r f a c e s

J.F. Adams: Mahowald's r e s u l t on the J - h o m o m o r p h i s m

A. Van de Ven: Chern n u m b e r s of (almost-) complex surfaces
D. Sullivan: M a n i f o l d s with singularities
 464

11. Mathematische A r b e ~ t a g u n 9 1967 (76. - 22. Juni)

M.F. Atiyah: Hyperbolic equations and algebraic geometry

J.P. Serre: Congruence subgroups and Coxeter hyperbolic groups

v v
I. Safarevic: Simple Lie algebras in finite characteristics
D.V. Anosov: Dynamical systems

N. Kuiper: Algebraic equations for c o m b i n a t o r i a l 8-manifolds

J.W.S. Cassels: Definite functions as sums of squares
M. Postnikov: K-theory for infinite complexes

v v
I. Safarevic: Algebraic analogue of u n i f o r m i s a t i o n
D.V. Anosov: Asymptotic theory of some partial differential
equations
M. Karoubi: Real and complex K - t h e o r y

H. Levine: Extending immersions of the circle in the plane

B. Venkov: Cohomology of some groups

J.F. Adams: Complex cobordism

A. Brumer: p-adic L-functions
J.I. Manin: On rational surfaces

R. Abraham: Generic properties of H a m i l t o n i a n systems

T. Matsumoto: Congruence subgroups and central group e x t e n s i o n s
K. Jinich: Report on part of the Tulane C o n f e r e n c e
 465

12. M a t h e m a t i s c h e A r b e i t ~ t a q u n q 1969 (13. - 20..lug,,),,,

N. Kuiper: Stable homeomorphisms and the annulus conjecture,

Kirby's results

P.A. Griffiths: Algebraic cycles

J.F. Adams: Quillen's work on c o b o r d i s m and formal groups
R. Bott: Topological obstructions for foliations

W.-C. Hsiang: Manifolds with fundamental group k

W. $chmid: Langlands ~ conjecture
P. Deligne: Hodge theory of singular varieties

A. Borel: Picard-Lefschetz transformations and a r i t h m e t i c

groups
F. Hirzebruch: sl-actions on m a n i f o l d s

L. Siebenmann: Topological manifolds and r e l a t e d examples

J. Mather: Stratification of a g e n e r i c mapping
D. Quillen: Homotopy theory of schemes

J. Tate: K 2 of global fields

R. Thom: Topological linguistics
D. Gromoll: Periodic geodesics
R. Gardner: G e o m e t r i c solution of the Gauchy P r o b l e m and a
g e n e r a l i z a t i o n of c h a r a c t e r i s t i c s

C.T.C. Wall: Integrable almost piecewise linear structures

J. Eells: Topology of B a n a c h m a n i f o l d s
G. Harder: A remark on T a m a g a w a numbers

A. Bak: Quadratic modules and unitary K-theory

A. Weinstein: Symplectic manifolds
W.C. Hsiang: Falsity of the s-cobordism theorem for lower
dimensional manifolds
 466

13. MathematNmche A r b e ~ t a g u n £ 1970 (12. - 19. Juni)

C.T.C. Wall: A s u r v e y of free a c t i o n s on spheres

S. Lang: Transcendental mappings

R. Kirby: T r i a n g u l a t i o n of m a n i f o l d s
M. Kervaire: Projective class g r o u p s and class field t h e o r y

S. H i l d e b r a n d t : Boundary values in c a p i l l a r i t i e s
D. B u r g h e l e a : H o m o t o p y g r o u p s of spaces of d i f f e o m o r p h i s m s
T. t o m Dieck: B o r d i s m of c o m m u t i n g involutions

S. Smale: R e p o r t on work of M o u l t o n (celestial mechanics)

D.V. Anosov: N e w e x a m p l e s of smooth e r g o d i c systems

R. Kiehl: Satz von G r a u e r t (direkte B i l d e r von Garben)

T.A. Springer: Discrete series of finite C h e v a l l e y g r o u p s
E. B r i e s k o r n : Singular elements in simple Lie g r o u p s

R. Takens: Partially hyperbolic fixed p o i n t s

H. Bass: The M i l n o r ring of a field
J. Eells: Fredholm structures and W i e n e r integrals

J.-L. Verdier: Serre's duality theorem

H.V. Pittie : The r e p r e s e n t a t i o n ring of c o m p a c t Lie g r o u p s
F. Waldhausen: Attempt on h i g h e r W i t h e h e a d g r o u p s

R. Gardner: R i g i d i t y and u n i q u e n e s s of c o n v e x h y p e r s u r f a c e s
U. Koschorke: Pseudo-kompakte Teilmengen unendlich dimensionaler
Mannigfaltigkeiten
 467

14. Mathemat~che A r b e ~ t a q u n g 1971 (10. - 14. Juni)

M.F. Atiyah: The R i e m a n n - R o c h theorem for m u l t i h o m o g e n e o u s

varieties

W. Schmid: The singularities of Griffith's period mapping

I.G. Macdonald: Affine root systems and t h e t a - i d e n t i t i e s
P. Deligne: The Weil conjecture for surfaces of degree 4 in P3

S. Lang: Frobenius automorphisms of modular function fields

G. Lusztig: The N o v i k o v higher signatures and families of
elliptic operators
F. Waldhausen: Applications of infinite simple homotopy types

G. Segal: Algebraic K-theory

C.T.C. Wall: Quadratic forms
H. Grauert: Deformation of s i n g u l a r i t i e s

D. Mumford: Degeneration of curves and n o n - a r c h i m e d e a n

uniformisation
R. Bott: The c l a s s i f y i n g space of foliations

R. Thom: The four v e r t e x theorem

W.-Ch. Hsiang: A reduction theorem of d i f f e r e n t i a b l e actions
J. Tits: Free groups in linear groups

M. Ar t i n : L~roth's theorem
E. Bombier i : On p l u r i c a n o n i c a l models of a l g e b r a i c surfaces
A.I. Kostr ikin : Deformation of Lie algebras

R. Langlands: On a r i t h m e t i c a l l y equivalent representations

J. Simons: Geometric invariants related to c h a r a c t e r i s t i c classes
K. Shiohama: On the d i f f e r e n t i a b l e "pinching problem"
 468

15. M ~ h e m a t i s c h e Arbeit~ta~ung 1972 (09. - 15. Juni)

W. Schmid: Degeneration of a l g e b r a i c manifolds

M. Karoubi: Hermitian K-theory

T.A. Springer: Steinberg functions on a finite Lie algebra
J. Cheeger: Manifolds with n o n - n e g a t i v e curvature

T. Petrie: Real a l g e b r a i c actions on p r o j e c t i v e spaces

th
D.I. Liebermann: n order de Rham theory
E. Win k e l n k e m p e r : Open book decomposition of m a n i f o l d s

A. Bak: Computation of surgery obstruction groups

A. Michenko: On infinite dimensional representations of

discrete groups
M.F. Atiyah: Invariants of odd d i m e n s i o n a l manifolds
S. Lichtenbaum: Values of z e t a - f u n c t i o n s at negative integers

F.J. Almgren: Geometric measure theory and e l l i p t i c variational

problems
M. Platonov: Conjectures of Artin and K n e s e r - T i t s
J.W. Robbin: Topological classification of linear e n d o m o r p h i s m s
of Rn
K. Ueno: Classifications of a l g e b r a i c varieties

M. Miranda: Hypersurfaces in Rn of p r e s c r i b e d mean curvature

M. Shub: Dynamical systems, filtrations and entropy
J.A. Shaneson: Codimension -2 p r o b l e m s and h o m o l o g y equivalent
manifolds
 469

16. Mathematische A r b e i t s t a g u n g 1974 (12. - 18. l u n i ]

I.M. Singer: ~ - i n v a r i a n t and its r e l a t i o n to re]l q u a d r a t i c fields

H. Bass: Russian progress on S e r r e ' s p r o b l e m

N.A'Campo: Resolution and d e f o r m a t i o n of plane curve singularities
J.-P. Serre: Modular forms and Galois r e p r e s e n t a t i o n s

J. Eells: Introduction to s t o c h a s t i c R i e m a n n i a n g e o m e t r y

R. Howe: The Weyl r e p r e s e n t a t i o n o v e r finite fields

G. Lusztig: Some discrete series r e p r e s e n t a t i o n s of finite
classical groups
S. K o b a y a s h i -
Hyperbolic geometry and d i o p h a n t i n e p r o b l e m s
S. Lang:

D. Epstein: Foliations with compact leaves

D. Quillen: F i n i t e g e n e r a t i o n of K - g r o u p s in the function
field case
D. Zagier: Modular forms in one and two v a r i a b l e s

G. Harder: Betti n u m b e r s of m o d u l a r spaces of v e c t o r b u n d l e s

W. Schmid: On the d i s c r e t e series of s e m i - s i m p l e Lie groups
E. Ruh: Equivariant pinching problems
K. Ueno: C a n o n i c a l bundle formula for c e r t a i n fibre spaces
and a l g e b r a i c v a r i e t i e s of p a r a b o l i c type

M.F. Atiyah: A s y m p t o t i c p r o p e r t i e s of e i g e n v a l u e s in R i e m a n n i a n
geometry
C.T.C. Wall: N o r m s of units in g r o u p rings
S. Lang: F e r m a t curves and units in the m o d u l a r function
field
 470

17. M a t h e m a ~ c h e Arb~tagung 19~5 21. - 2~. J u n i )

M.F. Atiyah: A l g e b r a s of o p e r a t o r s ±n H i l b e r t space and

K-theory

J. Moser: Isospectral deformations

G. Lusztig: M a c d o n a l d ' s c o n j e c t u r e on discrete series of
finite C h e v a l l e y g r o u p s
B. Kostant: The ~]-function formula of M a c d o n a l d

R. M a c P h e r s o n : Gelfand's formula for the first P o n t r j a g i n class

W. Ziller: C l o s e d g e o d e s i c s and h o m o t o p y symmetric spaces
J.-P. Serre: Lower b o u n d s of d i s c r i m i m a n t s of n u m b e r fields

T . t o m Dieck: B u r n s i d e ring of a c o m p a c t Lie g r o u p

J.C. Jantzen: Modular representations of semi simple g r o u p s

W. Schmid: B l a t t n e r ' s c o n j e c t u r e on the d i s c r e t e series of
semi simple real Lie g r o u p s
A.N. V a r c h e n k o : Newton diagrams of s i n g u l a r i t i e s

A.I. Kostrikin: T a n n a k a - A r t i n ' s c o n j e c t u r e on the m u l t i p l i c a t i v e

g r o u p of d i v i s i o n algebras
B. Mazur: R a t i o n a l p o i n t s on m o d u l a r curves
W. C a s s e l m a n : The n - c o h o m o l o g y of r e p r e s e n t a t i o n s of semi simple
Lie g r o u p s

A.N. Parshin: Residues and symbols

J.H.M. Steenbrink: M i x e d Hodge structure on v a n i s h i n g cycles
E. Calabi: Nearly flat t r i a n g u l a t i o n s of R i e m a n n i a n m a n i f o l d s
 471

18. M a t h e m ~ c h e A~bei.~tagung 1977 (21. - 27. J u n i ]

M.F. Atiyah: The classical geometry of Yang-Mills Fields (I)

J.-P. Serre: Function field analogue of SL2(Z)

H.W. Lenstra: Euclidean number fields
A. Tromba: Recent progress in P l a t e a u ' s problem

M.F. Atiyah: The classical geometry of Y a n g - M i l l s fields (II)

M. Gromov: Hyperbolicity in d y n a m i c a l systems
P. G r i f f i t h s : Application of r e s i d u e s to g e o m e t r y

M. Berger: Wiedersehensmannigfaltigkeiten
( C o n j e c t u r e of B l a s c h k e )
A. V a n de Ven: Inequalities for C h e r n numbers of surfaces
G. Zuckerman: Representations of semi simple lie g r o u p s

Ch. Thomas: Space form problems

R. Finn: Surface tension phenomena and geometry
J.-P. B o u r g u i g n o n
Solution of the C a l a b i conjecture
C.L. T e r n g :

C. P r o c e s i : Ideals of d e t e r m i n a n t s and Young diagrams

F. Sakai: Kondaira dimension of o p e n complex manifolds
A. A n d r e o t t i : D o m a i n of r e g u l a r i t y of s o l u t i o n s of p a r t i a l
differential equations
 472

19. M a t h e m a t ~ c h e ~ b e ~ t a g u n 9 1978 (16. - 23. Juni]

M.F. At iyah : Yang-Mills instantons and algebraic geometry

E. Calabi: SU(n)- and S p ( n ) - m a n i f o l d s

H. Jacquet: From GL(2) to GL(3)
J.-P. Bourguignon: Differential-geometry of the Yang-Mills equation

J. Eells: Holomorphic and h a r mo n i c maps of surfaces

J. Steenbrink: Non-rationality of the quartic threefold
B. Gross: The C h o w l a - S e l b e r g formula

B. Mazur: R a t i o n a l points on e l l i p t i c curves and

c o n g r u e n c e s of L - s e r i e s
D. Burghelea: C o m p u t a t i o n of h o m o to p y groups of d i f f e o m o r p h i s m
groups of compact m a n i f o l d s
P. Schweitzer: Residues of real foliation singularities

T. Banchoff: The fourth dimension and computer animated geometry

J. Milnor: Volume of h y p e r b o l i c manifolds

K. Ueno: Birational geometry of fibre spaces

F. Adams: Finite H - s p a c e s and algebras over the
Steenrod A l g e b r a
N. Hitchin: Twistor spaces

A. Todorov: Surfaces with pg = I and K2 : I

P. Baum: K-homology and R i e m a n n - R o c h
J. Brining: R e p r e s e n t a t i o n s of compact Lie groups and
e l l i p t i c operators

I. Piatetsk~Sbmp~o: Automorphic forms on the m e t a p l e c t i c group

F. Waldhausen: Algebraic K-theory of t o p o l o g i c a l spaces
 473

20. Mathematische Arbeit~tagung 1979 (06. - 16. f u j i )

J. Tits: On L e e c h ' s lattice and sporadic groups

F. Adams: G. Segal's Burnside ring c o n j e c t u r e

F. Bogomolov: Converse Galois problems for some C h e v a l l e y groups
Wang Yuan: Goldbach problem

D. Vogan: Size of r e p r e s e n t a t i o n s
L. Berard-Bergery: A n e w example of E i n s t e i n manifolds

V. Kac: Infinite d i m e n s i o n al Lie algebras

G. Mostow: New negatively curved surfaces
G. Lusztig: Representations of Hecke algebras

B. Gross: Conjectures of Stark and Tate

Wu-ch u n g Hsiang: Topological space form p r o b l e m s
E. Looijenga: Singularities and g e n e r a l i z e d root systems
M.F. Vigneras: Isospectral but not isometric Riemannian surfaces

A.N. Parshin: Zeta functions and K - t h e o r y

Min-Oo: Curvature deformations relating to the
Y a n g - M i l l s fields
G. Harder: Cohomology and v a l u e s of L - f u n c t i o n s

A. Todorov: Moduli of K ~ h l e r i a n K3-surfaces

R.P. Langlands: On orbital integrals for real groups
J.-P. Serre: The monster game
 474

21. M a t h e m ~ c h e Arbe~taqung 1980 (15. - 19. Juni}

M.F. Atiyah: V e c t o r B u n d l e s on R i e m a n n S u r f a c e s

A. Katok: Counting closed geodesics on s u r f a c e s

R. Bott: E q u i v a r i a n t Morse theory
K. Ribet: Mazur and Wiles (cyclotomic fields)

A. Borel : L 2 - c o h o m o l o g y of a r i t h m e t i c g r o u p s
I. Bake iman" Topological methods in the t h e o r y of
Monge-Amp@re equations
H. King : T o p o l o g y of real a l g e b r a i c v a r i e t i e s

J. M i l n o r : Groups of p o l y n o m i a l g r o w t h (Gromov's work)

Y.-t. Siu: Andreotti-Fraenkel conjecture

M. Ar%in: Mori's work
B. Gross: L-series of e l l i p t i c c u r v e s

F. Takens: Turbulence and strange attractors

W. Ziller: Periodic motions in H a m i l t o n i a n systems
P. Slodowy: S i m p l e groups over ~((t)) and s i m p l e - e l l i p t i c
singularities

S. Kudla: Geodesic cycles and the Weil r e p r e s e n t a t i o n

D. Epstein: A t h e o r e m of T h u r s t o n w i t h a p p l i c a t i o n s to
group actions and f o l i a t i o n s
F. Adams: R e c e n t work on h o m o t o p y t h e o r y
 475

22. M a t h e m ~ c h e A~bei~taqun~ 1981 (12. - 18. ] u n i )

M.F: Atiyah: Convexity and c o m m u t i n g Hamiltonians

B. Mazur: Abelian extensions of

D. De Turck: "Manifold" of Ricci curvatures
B. Malgrange: Vanishing cohomology and B e r n s t e i n polynomials

D. Mostow: Complex reflection groups

W. Fulton: Complex projective geometry (varieties of small
codimension)
R. MacPherson: Intersection homology and n i l p o t e n t orbits

J. Tate: Stark's conjecture about L-series at s = 0

W. Meyer: Gromov's work on Betti numbers
K. Diederich: Complete K~hler domains

W.D. Neumann: Thurston's work

A. Derdzinski: Einstein metrics

S. Zucker: L2-cohomology of a r i t h m e t i c groups
A. Wiles: Explicit constructions of class fields

J. Duistermaat: Asymptotics of spherical functions

M-F. Vigneras: W o r k s of W a l d s p u r g e r (automorphic forms of
h a l f i n t e g r a l weight)
R. Schultz: Topological similarity of r e p r e s e n t a t i o n s
 476

25. M ~ h c m ~ c h e ArbeJ~tagung 1982 (15. - 21. duni)

M.F. Atiyah: The Yang-Mills equations and the structure

of 4 - m a n i f o l d s

D. Quillen: Determinants of Y - o p e r a t o r s
J. Coates: Heights on elliptic curves
D. V o g a n : Representations with cohomology

R.S. Palais: Hamilton's work on positively curved 3-manifolds

R. H a r t s h o r n e : Space curves
M. Berger: Gromov's filling of R i e m a n n i a n manifolds

J. B e r n s t e i n : Beilinson-Bernstein construction

S.T. Yau: Manifolds with positive scalar curvature

S. Mori: Rational curves in 3 - f o l d s and applications
G. H a r d e r : Tate conjecture for H i l b e r t modular surfaces

L. Siebenmann: M. F r e e d m a n ' s work on 4-dimensional manifolds

S.J. Patterson: Limit s e t s of K l e i n i a n groups
D. E p s t e i n : On C h a p t e r I of T h u r s t o n

S.S. Chern: Web geometry

B. M a z u r : -extensions and heights
P
F. A d a m s : C a r i s s o n ' s p r o v e of S e g a l ' s Burnside ring conjecture
 477

24. M ~ h e m a t ~ c h e A ~ b ~ s t a g u n g 1983 (16. - 22. J u ~ )

M.F. Atiyah: Instantons, monopoles and rational maps

G. Faltings: The c o n j e c t u r e s of Tare and Mordell I

B. Gross: Heights and L-series I (On the conjecture of
B i r c h and Swinnerton-Dyer)
S.W. Donaldson: Stable holomorphic bundles and curvature

C. Procesi: The solution of the S c h o t t k y p r o b l e m

(Characterization of J a c o b i a n varieties)
D. Quillen: Cyclic homology and H o c h s c h i l d - h o m o l o g y
N. Hingston: Equivariant Morse theory

D. Zagier: Heights and L - s e r i e s II (and a p p l i c a t i o n s to the

class number problem)
D. McDuff: The Arnold conjecture on symplectic fixed points
(after Conley and Zehnder)
G. Faltings: The conjectures of Tate and Mordell II
(Moduli spaces of a b e l i a n varieties)

J. Milnor: Monotonicity for the e n t r o p y of quadratic maps

E. Friedlander- On the c o n j e c t u r e s of L i c h t e n b a u m and Quillen
(after Suslin and others)
D. Quillen: Arithmetic surfaces and analytic torsion
B. Moonen: Polar m u l t i p l i c i t i e s and curvature integrals

D. Eisenbud: Special d i v i s o r s on curves and K o d a i r a d i m e n s i o n

of the moduli space (mostly after M u m f o r d and
Harris and Gieseker)
F. Kirwan: Cohomology of q u o t i e n t s in algebraic and
symplectic geometry
W. Tutschke: G e n e r a l i z a t i o n s of the C a u c h y - K o w a l e w s k i and
H o l m g r e n t h e o r e m s to the case of g e n e r a l i z e d
analytic functions
G. W~stholz: Group varieties and t r a n s c e n d e n c e
 478

25. Ma%h(~ma2i.~ch(~ Arb(:~t_~<a_g}Sglq j 9 8 4 {15. - 22. 3u~<)

J. Tits: Groups and g r o u p functors attached to Kac-Moody data

M. Atiyah: The e i g e n v a l u e s of the Dirac operator

A. Connes: K-theory, cyclic cohomo]ogy and o p e r a t o r algebras
G. Segal: Loop groups

G. Harder: Special v a l u e s of Hecke L-functions and a b e l i a n

integrals
H. Wente: A counterexample in 3-space to a c o n j e c t u r e of
H. Hopf
G. Faltings: Compactification of Ag/Z
C.T.C. Wall: Geometric structures and algebraic surfaces

J. Harris: Recent work on Hodge structures

Y.T. Siu: Some recent r e s u l t s in c o m p l e x m a n i f o l d theory

related to v a n i s h i n g t h e o r e m s for the s e m i p o s i t i v e
case
W. Schmid: Recent p r o g r e s s in r e p r e s e n t a t i o n theory
W. Ballmann: Manifolds of n o n - p o s i t i v e curvature
B. M a z u r -
Ch. Soul~: Conjectures of B e i l i n s o n on L - f u n c t i o n s and K - t h e o r y

H.-O. Peitgen: Morphology of J u l i a sets

S.S. Chern: Some applications of the m e t h o d of m o v i n g frames
S. Lang: Vojta's conjecture on h e i g h t s and G r e e n ' s function
S. Donaldson: 4-manifolds with indefinite intersection form

D. Zagier: M o d u l a r points, m o d u l a r curves, modular surfaces

and m o d u l a r forms
G. van der Geer: Schottky's problem
R. Bryant: G 2 and S p i n ( 7 ) - h o l o n o m y
S. Wolpert: Homology of T e i c h m ~ l l e r spaces

J.-P. Serre: l-adic representations

M.F. Atiyah: On M a n i n ' s manuscript "New d i m e n s i o n s in geometry"
 Teilnehmer der A r b e ~ t a g u n g 1984

K. Abe (U. of Connecticut) S. Donkin (London)

U. A b r e s c h (MPI/SFB) A. Douady (ENS Paris)
J.F. Adams (Cambridge) W. E b e l i n g (Bonn)
S. A k b u l u t (MPI/SFB) P. E b e r l e i n (MPI/SFB)
H. A n d e r s e n (Aarhus) J. Eells (Warwick)
Y. Andr6 (Inst. Poincar6, Paris) H. Eggers (Regensburg)
G. A n d r z e j c z a k {MPI/SFB) J. Ehlers (MPI Phys. & Astrophys.)
V. A n o s o v (Steklov Inst., Moskau) P.E. E h r l i c h (U. of Missouri)
M. Artin (MIT) M. Eichler(Basel)
M.F. Atiyah (Oxford) H. E l i a s s o n (MPI/SFB)
H. Azad (Bochum) O. Endler (Bonn und IMPA)
A. Back (SUNY) D. Erle (Dortmund)
A. Bak (Bielefeld) H. Esnault (MPI/SFB)
A. Baker (Cambridge) G. Faltings (Wuppertal)
W. Bal l m a n n (Bonn) G. Fischer (D~sseldorf)
T. B an c h o f f (Brown U.) H. Flenner (G~ttingen)
B. B a n i e q b a l (Manchester) E. F r i e d l a n d e r (North W e s t e r n U.)
W. Barth (Erlangen) S. F r i e d l a n d e r (U. of Chicago)
G. Barthel (Konstanz) S. Gallot (Paris VII)
H.J. Baues (Bonn) G. van der Geer (Amsterdam)
E. Becker (Dortmund) E.U. Gekeler (Bonn)
K. Behnke (MPI/SFB) H. Grauert (G~ttingen)
R. Berndt (Hamburg) R. Greene (UCLA)
B. Birch (Oxford) G.M. Greuel (Kaiserslautern)
S. B6c h e r e r (Freiburg) B. Gross (Brown U.)
C.-F. B ~ d i g h e i m e r (Heidelberg) F. G r u n e w a l d (Bonn)
A. B o j a n o w s k a (ETH ZUrich) R. Gupta (Brown U.)
M. B~kstedt (Bielefeld) F. v. Haeseler (Bremen)
W. Borho (Wuppertal) H. Hamm (MUnster)
R. Bott (Harvard) G. Harder (Bonn)
J.P. B o u r g u i g n o n (Ec. Polyt. Palaiseau) J. Harris (Brown U.)
G. B r a t t s t r ~ m (Harvard) F. H e g e n b a r t h (Dortmund)
E. B r i e s k o r n (Bonn) E. Heintze (M~nster)
E. Brown (Brandeis U.) J. Heinze (Springer Verlag)
R. B r u g g e m a n (Utrecht) G. H e l m i n c k (Amsterdam)
J. BrUning (Augsburg) R. Henderiks (Rotterdam)
R.L. Bryant (Rice U.) M. H e r r m a n n (K~In)
D. B u r g h e l e a (Ohio State U.) S. H i l d e b r a n d t (Bonn)
E. Calabi (U. of Pennsylvania) U. Hirsch (Bielefeld)
S.S. Chern (MSRI Berkeley) F. H i r z e b r u c h (~IPI und U. Bonn)
J. Coates (Orsay) T. H~fer (MPI/SFB)
R. C o n n e l l y (Cornell U.) J. H u e b s c h m a n n (Heidelberg)
A. Connes (IHES) K. Hulek (Erlangen)
M. Crabb (Wuppertal) J. H u r r e l b r i n k (Louisiana State U.)
C.B. Croke (U. of Pennsylvania) D. H u s e m o l l e r (Haverford College)
R. Cushman (Utrecht) H.-C. Im Hof (Basel)
W. Dan i e l e w s k i (Bochum) K. Ivinskis (Bonn)
W. Decker (Kaiserslautern) S. J a c k o w s k i (ETH ZUrich)
J. Deprez (Leuven) J.C. Jantzen (Bonn)
A. Derdzinski (MPI/SFB) B. Julia (E.N.S. Paris)
T. tom Dieck (G~ttingen) H. J~rgens (Bremen)
K. D i e d e r i c h (Wuppertal) W. van der K a l l e n (Utrecht)
A. Dold (Heidelberg) W. Kamber (U. of Illinois)
P. D o m b r o w s k i (K~in) H. Karcher (Bonn)
S.K. D o n a l d s o n (Oxford) M. Karoubi (Sceaux)
 480

U. Karras (Dortmund) B.Z. Moroz (MPI/SFB)

J. Kazdan (U. of Pennsylvania) D. M o r r i s o n (Princeton)
I. Kersten (Regensburg) G.D. M o s t o w (Yale U.)
M. Kervaire (Genf) P. Mrozik (Siegen)
F. Kirwan (Harvard) Y. N a m i k a w a (MPI/SFB)
N. Klingen (K~In) H.J. Nastold (MUnster)
W. K l i n g e n b e r g (Bonn) W.D. N e u m a n n (U. of Maryland)
P. K l u i t m a n n (MPI/SFB) W.-M. Ni (U. of Minnisota)
K. Knapp (Wuppertal) T. N i s h i m o r i (Hokkaido U.)
H. KnSrrer (Bonn) K. Nomizu (Brown U.)
F. Koll (Bonn) E. O e l j e k l a u s (MUnster)
J. Konarski (Bochum) A. Ogg (Orsay)
K. K o p f e r m a n n (Hannover) R. O]ivier (Bonn)
S. K o s a r e w (Regensburg) F. Oort (Utrecht)
U. K o s c h o r k e (Siegen) E. Ossa (Wuppertal)
M. K o s s o w s k i (Rice U.) J.-P. Otal (MPI/SFB)
B. K o s t a n t (MIT) M. Otte (Bielefeld)
H. Kraft (Basel) A. Parsh~n (Steklov Inst., Moskau)
G. Kramarz (Bonn) G. Pa t r i z i o (MPI/SFB)
J. Kramer (Basel) H.-O. Peitgen (Bremen)
M. Kr~mer (Bayreuth) A. Pereira do Valle (Bonn)
M. Kreck (Mainz) Th. P e t e r n e l l (M~nster)
F.-V. K u h l m a n n (Heidelberg) Ch. Peters (Leiden)
K. K U h n e - H a u s m a n n (Bonn) V.Q. Phong (Bremen)
W. Kdhnel (TU Berlin) A. Pickl (Regensburg)
N. Kuiper (IHES) B. Pickl (Regensburg)
R.S. K u l k a r n i (Indiana U.) R. Piene (Oslo)
B.A. K u p e r s h m i d t (U. of Tennessee) U. Pinkall (MPI/SFB)
N. K u z n e t s o v (Khabarovsk) P. P l a t o n o v (Minsk)
C. L a c k s c h e w i t z (Berlin) H. Popp (Heidelberg)
K.Y. Lam (U. of British Columbia) V. Poupko (Wien)
R. Lamotke (K~in) A. Prestel (Konstanz)
S. Lang (Yale U.) M. PrUfer (Bremen)
J. Langer (MPI/SFB) V. Puppe (Konstanz)
M. Laska (MPI/SFB) R.M. Range (SUNY, Albany)
K. Lebe!t (Essen) T. Ratiu (UC, Berkeley)
R. Lee (Yale U.) B. Reinhart (U. of Maryland)
L. Lemaire (BrUssel) R. Remmert (MUnster)
M. Levine (U. of Pennsylvania) D. Repovs (Ljubljana)
I. Lieb (Bonn) K. Ribet (UC, Berkeley)
J. Little (Holy Cross) D. Richter (Bremen)
M. Lorenz (MPI/SFB) O. R i e m e n s c h n e i d e r (Hamburg)
J. M a r s d e n (UC, Berkeley) J. Rohlfs (Eichst~tt)
S. M a u r m a n n (Bonn) G. Roland (Bonn)
K.-H. M a y e r (Dortmund) M. Rost (Regensburg)
B. Mazur (Harvard) E. Rub (Bonn)
J. M c C a r t h y (MPI/SFB) J.M. Sampson (Baltimore)
J. M e s i r o v (AMS) E. Sato (MPI/SFB)
W. Meyer (MPI/SFB) D. Saupe (Bremen)
W.T. Meyer (M~nster) N. S c h a p p a c h e r (MPI/SFB)
H. Miller (U. of Washington) A. Scharf (Bonn)
T. M i l l e r (Bonn) R. S c h a r l a u (Bielefeld)
P. M i l m a n (Toronto) W. Schelter (U. of Texas)
M. M i n - O o (Bonn) K.D. Schewe (Bonn)
G. M i s l i n (ZUrich) U. S c h m i c k l e r - H i r z e b r u c h (Vieweg Vlg.)
V. M o n c r i e f (Yale U.) W. Schmid (Harvard)
B. Moonen (K~in) C.-G. Schmidt (IHES)
R. Moore (Canberra) G. S c h n e i d e r (Kaiserslautern)
 481

P. Schneider (Heidelberg) D. Zagier (MPI und U. of Maryland)

R. Sch~n (Heidelberg) G. Zeidler (Leipzig)
R. School (U. of Maryland)
F. Schreyer (Kaiserslautern)
M. Schroeder (G@ttingen)
G. Schumacher(M~nster)
J. Schwermer (Bonn)
R. Sczech (MPI/SFB)
D. Segal (Manchester)
G. Segal (Oxford)
W. Seiler (Mannheim)
J.-P. Serre (Coll~ge de France)
D. Siersma (Utrecht)
Y.-T. Siu (Harvard)
B. Smyth (Notre Dame)
Ch. Soul6 (Paris VII)
E. Spanier (MPI/SFB)
B. Speh (Cornell U.)
C. S p ~ n e m a n n (Hannover)
J. S t e e n b r i n k (Leiden)
J. S t i e n s t r a (Utrecht)
R. Strebel (Heidelberg)
J. Strooker (Utrecht)
U. Stuhler (Wuppertal)
S. Suter (MPI/SFB)
L. Szpiro (ENS Paris)
Ch. Thomas (Cambridge)
K. T i m m e r s c h e i d t (Essen)
J. T i t s (Colldge de France)
A. Todorov (Sofia)
P. Tondeur (U. of Illinois)
G. T r a u t m a n n (Kaiserslautern)
G. T r i a n t a f i l l o u (U. of Minnesota)
A. Tromba (UC, Santa Cruz)
G. Tsagas (Thessaloniki)
A. Tschimmel (Wuppertal)
T. Tsuboi (Tokyo)
A. Van de Ven (Leiden)
P. V e r h e y e n (Leuven)
E. Viehweg (Essen)
M.-F. V i g n e r a s (ENSJF Paris)
W. Vogel (Halle)
T. Vorst (Rotterdam)
F. W a l d h a u s e n (Bielefeld)
C.T.C. Wall ¢Liverpool)
N. Walter (Mannheim)
M. Wang (McMaster U.)
L. W a s h i n g t o n (U. of Maryland)
H.C. Wente (U. of Toledo)
J. Werner (MPI/SFB)
A. Wiles (Princeton)
K. W i r t h m ~ l l e r (Regensburg)
J. Wolfart (Frankfurt)
S. Wolpert (U. of Maryland)
P.-M. Wong (Notre Dame)
J. Wood (Chicago)
G. W~stholz (MPI/SFB)
M. Yoshida (MPI/SFB)