Deligne Sur Grothendieck

Quelques id
ees matresses de
luvre de A. Grothendieck
Pierre Deligne
R
esum
e
Cet article tente dexpliquer quatre concepts mathematiques fondamentaux crees par Grothendieck : les schemas, les topos, les
six operations et les motifs.
Abstract
We try to explain four fundamental ideas invented by Grothendieck:
schemes, topos, the six operations and motives.
Dans Recoltes et Semailles (troisième partie), Grothendieck ecrit :

Prenons par exemple la tache de demontrer un theorème qui reste hypothetique (à quoi, pour certains, semblerait se reduire le travail mathematique).
Je vois deux approches extremes pour sy prendre. Lune est celle du marteau
et du burin, quand le problème pose est vu comme une grosse noix, dure et
lisse, dont il sagit datteindre linterieur, la chair nourricière protegee par la
coque. Le principe est simple : on pose le tranchant du burin contre la coque,
et on tape fort. Au besoin, on recommence en plusieurs endroits dierents,
jusqu`
a ce que la coque se casse et on est content. [. . .]
Je pourrais illustrer la deuxième approche, en gardant limage de la noix
quil sagit douvrir. La première parabole qui mest venue a` lesprit tant
ot,
cest quon plonge la noix dans un liquide emollient, de leau simplement
pourquoi pas, de temps en temps on frotte pour quelle penètre mieux, pour
le reste on laisse faire le temps. La coque sassouplit au l des semaines et
des mois quand le temps est m
ur, une pression de la main sut, la coque
AMS 1991 Mathematics Subject Classication: 01A65, 14-03

Institute for Advanced Study, School of Mathematics, Princeton, N.J. 08540, USA
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
12
P. DELIGNE
souvre comme celle dun avocat m

ur a` point ! Ou encore, on laisse m
urir la
noix sous le soleil et sous la pluie et peut-etre aussi sous les gelees de lhiver.
Quand le temps est m
ur cest une pousse delicate sortie de la substantique
chair qui aura perce la coque, comme en se jouant ou pour mieux dire, la
coque se sera ouverte delle-meme, pour lui laisser passage. [. . .]
Le lecteur qui serait tant soit peu familier avec certains de mes travaux
naura aucune diculte à reconnatre lequel de ces deux modes dapproche
est le mien . [1985, p. 552553]
Un peu plus loin [Ibid. p. 554, note ()], Grothendieck met en avant
quatre exemples : Riemann-Roch, structure du 1 premier a` la caracteristique
pour les courbes, rationalite des fonctions L pour les schemas de type ni sur
un corps ni et theorème de reduction semi-stable pour les varietes abeliennes.
Je me rappelle mon earement, en 1965-66 après lexpose de Grothendieck
[SGA 5] prouvant le theorème de changement de base pour Rf! : devissages,
devissages, rien ne semble se passer et pourtant a` la n de lexpose un theorème
clairement non trivial est l`
a.
Bien des idees de Grothendieck nous sont devenues si familières, sont si
parfaitement adequates à leur objet, que nous oublions quelles etaient loin
detre evidentes a` leur naissance, que nous oublions meme leur auteur. Mon
but dans cet article est de decrire quatre de ces idees : schemas, topos, six
operations, motifs.
1.
Sch
emas
Linvention des schemas est la première des idees de Grothendieck à laquelle on pense, peut-etre parce quelle a ete la plus vite acceptee. Lexpose de
Serre a` Stockholm (1962) commence par : Je voudrais exposer ici quelques
uns des developpements recents de la geometrie algebrique. Je dois preciser
que je prends ce dernier terme au sens qui est devenu le sien depuis quelques
annees : celui de la theorie des schemas. Cette acceptation a ete facilitee par
la parution rapide, gr
ace à la collaboration de Dieudonne, des EGA.
Laudace de la denition de Grothendieck est daccepter que tout anneau
commutatif (`
a unite) A denisse un schema ane Spec(A), i.e. de ne pas
chercher à se limiter à une categorie de bons anneaux (intègres, reduits,
noetheriens, . . .). Ceci a un prix. Les points de Spec(A) (ideaux premiers de A)
nont pas un sens geometrique maniable, et le faisceau structural O nest pas
un faisceau de fonctions. Quand on a a` construire un schema, on ne commence
pas en general par construire lensemble de ses points.
Plus important peut-etre : le parti pris de b
atir une theorie relative, dont
` 3
SEMINAIRES
ET CONGRES

QUELQUES IDEES
MAITRESSES DE LUVRE DE A. GROTHENDIECK
13
temoigne lomnipresent
X
f
S
des exposes de Grothendieck. Le cas classique dune variete denie sur un
corps k devient le cas particulier S = Spec(k). Dans une theorie relative, avec
un S-schema X (= schema X sur S), i.e. avec un morphisme de schemas
f : X S, on considère systematiquement le schema X deduit de X par un
changement de base u : S S, i.e. le produit bre X := S S X et sa
projection f sur S :
X
X
S
Dans la categorie des schemas, les produits bres existent toujours : si

permettre que tout anneau commutatif denisse un schema ane donne droit
de cite à des sch
emas bizarres, le permettre fournit une categorie de schemas
ayant de bonnes proprietes.
Une propriete de X sur S sera dite g
eom
etrique si elle a de bonnes proprietes dinvariance par changement de base. Analogue classique : pour une
variete X denie sur un corps k, lensemble des points de X sur une extension
algebrique close de k (par exemple : domaine universel de Weil) est considere
comme geometrique , lensemble des k-points etant arithmetique .
Si X est un schema sur S, et que u : S S est un morphisme de schemas,
un S -point p de X est un morphisme de S-schemas :
X
p
S
On note X(S ) lensemble de S -points de X. Il sidentie a` lensemble des

sections de X S .
Exemple 1.1. Soit X le schema ane sur un corps k deni par des
equations P (X1 , . . . , Xn ) = 0 :
X = Spec(k[X1 , . . . , Xn ] / (P )).
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
14
P. DELIGNE
Soit k une k-algèbre. Lensemble X(k ) := X(Spec(k )) est lensemble des

solutions dans kn des equations P = 0. Pour k une extension algebriquement
close de k, cest lensemble X() que Weil regarde comme sous-jacent à X.
Exemple 1.2. Le schema GLN sur Spec(Z) est tel que pour tout anneau
commutatif A (automatiquement une Z-algèbre : Spec(A) est un schema sur
Spec(Z)), GLN (Spec(A)) est GLN (A).
Dans ces exemples, lintuition geometrique quon a de X, schema sur S,
est bien reetee dans X(S ). Mieux que dans lensemble sous-jacent a` X. Dans
le cas de lexemple 1.2., cet ensemble sous-jacent nest pas un groupe, alors
que chaque GLN (S ) lest. Plus precisement, il est utile dattacher au schema
X sur S le foncteur contravariant des S-schemas dans les ensembles
S ensemble X(S ) des S -points de X.
Dans la categorie de S-schemas, il sagit simplement du foncteur representable
hX : S Hom(S , X)
attache à X. Daprès le lemme de Yoneda, le foncteur X hX est pleinement dèle. Plut
ot que de penser a` un S-schema X comme etant un espace
annele, muni de X S, avec des proprietes convenables, il est souvent commode dy penser comme etant un foncteur
S -points : (Schemas/S)0 (Ens),
qui a la vertu detre representable. Quand on veut denir un espace n de
modules , la première etape est de denir le foncteur correspondant. Typiquement, denir ce foncteur requiert une theorie relative.
Exemple 1.3. Soit X projectif sur un corps k. Question : que signie :
espace de module des sous-schemas fermes de X ? Pour u : S Spec(k),
soit X sur S deduit de X par changement de base. Soit H(S ) lensemble

des sous-schemas fermes Y de X , plats sur S (plats de presentation nie, si
on ne veut pas supposer S noetherien). Reponse : cest un schema Hib(X)
representant le foncteur S H(S ).
Pour etre viable, ce point de vue requiert quon dispose de methodes pour
verier si un foncteur est representable. La plupart des exposes de FGA sont
consacres à ce problème. Une solution denitive, prolongeant ces travaux, a
ete obtenue par M. Artin [1969] tout au moins si on accepte de remplacer la
categorie des schemas par celle, plus naturelle, des espaces algebriques. Plus
naturelle : au meme sens que la topologie etale est plus naturelle que celle de
Zariski.
` 3
SEMINAIRES
ET CONGRES

QUELQUES IDEES
2.
15
Topos
Soit u : S S un morphisme de schemas, le morphisme de changement

de base . La theorie de la descente [Grothendieck FGA, Seminaire Bourbaki
190, 1959-60] considère des problèmes des types suivants. Descente de proprietes : soit X un schema sur S et X sur S deduit de X par changement de
base. Supposons que X /S ait une propriete P . Peut-on conclure que X/S
verie P ? Descente de morphismes : soient X, Y sur S, X , Y deduits par
changement de base et g : X Y un morphisme de S -schemas. Quand g
provient-il par changement de base de g : X Y ? Descente dobjets : soit
X sur S . Quelle est la donnee de descente sur S requise pour construire X
sur S dont X se deduise par changement de base ?
Si S est la somme disjointe des ouverts (Ui )iI dun recouvrement de
S, le changement de base à S est essentiellement la restriction à chaque
Ui , et les problèmes precedents sont des problèmes de localisation sur S et
de recollement. Recoller exige typiquement la consideration des intersections
deux a` deux des Ui Uj , et en termes de u : S S, la somme disjointe des
Ui Uj est simplement le produit bre S S S .
La theorie des topos permet de transposer en theorie de la descente lintuition topologique. Pour u : S S, un morphisme de changement de base dun
type considere en theorie de la descente, par exemple dèlement plat et quasicompact (f pqc), le changement de base de S à S devient une localisation.
Une donnee de descente est lanalogue dune donnee de recollement.
Un antecedent a` la theorie de la descente est la descente galoisienne, correspondant a` S, spectre dun corps k et S , spectre dune extension galoisienne.
Ici, S S S est somme de copies de S indexees par Gal(k /k). Les demonstrations, et en particulier la theorie de Galois, sont toutefois plus simples dans le
cadre plus general de la theorie de la descente. Selon un mot de Cartier : Grothendieck prouve la theorie de Galois, et la descente galoisienne, par descente
galoisienne.
Loutil quest la theorie des topos a permis la construction de la cohomologie etale des schemas, et cest là son succès le mieux connu. Un faisceau sur X
pour la topologie de Zariski est un foncteur contravariant de la categorie des
ouverts de Zariski de X dans celles des ensembles, avec une condition de recollement pour (Ui )iI , un recouvrement ouvert de Zariski de U . En topologie
etale, le site Zariskien, considere plus haut, est remplace par le site etale : la categorie des ouverts de Zariski est remplacee par celle des f : U X etales sur
X, et les recouvrements par les familles couvrantes (Ui )iI : un morphisme surjectif de schema sur X de la somme disjointe des Ui dans U . Un antecedent :
lintroduction par Serre de la notion despace principal homogène isotrivial
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
16
P. DELIGNE
(= localement trivial pour une topologie proche de la topologie etale). Dans

ses articles à Kansas [1955] et au T
ohoku [1957], Grothendieck avait montre
que, une categorie de faisceaux etant donnee, une notion de groupes de cohomologie en resulte. La topologie etale fournit ainsi la cohomologie etale. Que
les groupes H 1 obtenus soient raisonnables nest pas surprenant. Le miracle
est que les H i superieurs soient eux aussi raisonnables.
Pour Grothendieck, limportance de la Theorie des topos depasse de beaucoup le seul cas de la topologie etale. Le titre donne à SGA 4 en est temoin :
Theorie des topos et cohomologie etale des schemas .
Autres applications :
(A) Cohomologie cristalline. La cohomologie cristalline est celle du topos
cristallin, et cette denition rend claire sa fonctorialite. Le topos cristallin est
toutefois dusage delicat et il est souvent necessaire de passer à une interpretation en termes de complexes de de Rham.
(B) Espaces rigides analytiques. Les faisceaux rigides analytiques de Tate
sont des faisceaux coherents sur un topos annele convenable, et leur cohomologie la cohomologie correspondante.
(C) Feuilletages. Une variete X munie dun feuilletage F denit un quotient X/F, qui est un topos localement isomorphe a` celui des faisceaux sur
une variete. Ce point de vue semble avoir ete eclipse par celui de Connes qui
associe plutot a` F une C -algèbre non commutative.
Jaimerais encore mentionner lusage de gros sites (topos) par Grothendieck, notamment pour interpreter des espaces classiants.
3.
Les six op
erations
Le formalisme des six operations presuppose celui des categories derivees.

Pour un historique de ce dernier, je renvoie au texte de Illusie [1990]. Idee de
base : pour toutes sortes de groupes de cohomologie, leur denition fournit non
seulement ces groupes, mais encore un complexe K dont ils sont les groupes
de cohomologie. Typiquement, ce complexe nest pas uniquement determine,
mais il lest a` quasi-isomorphisme près : pour deux variantes K , K de K,
on dispose de K et de morphismes K K K induisant des isomorphismes en cohomologie. Dans la categorie derivee, K en devient unique a`
isomorphisme unique près.
` 3
SEMINAIRES
ET CONGRES

QUELQUES IDEES
17
Les six operations sont des foncteurs entre categories derivees. Il sagit
L
du produit tensoriel derive , du Hom interne RHom (donnant naissance

aux Exti locaux), et pour f : X Y un morphisme de schemas, de deux
foncteurs dimage directe : Rf et Rf! , et de deux foncteurs dimage inverse :
Lf et Rf ! .
Le formalisme de ces operations a dabord ete degage en cohomologie des
faisceaux coherents : Grothendieck [1963]. Ici, pour les images directes, il y a
lieu de ne considerer que des morphismes propres, pour lesquels Rf = Rf! .
Le formalisme fournit une version relative de la dualite de Serre, sous la forme
dune adjonction entre les foncteurs Rf! et Rf !. Il fournit aussi une formule
des points xes plus generale que la Woodshole xed point formula , qui en
est contemporaine (Summer institute on algebraic geometry, Whitney estate,
Woodshole, Massachusetts 1964).
Miraculeusement, le meme formalisme sapplique en cohomologie etale
avec des preuves très dierentes. Il fournit encore formalisme de dualite (de
Poincare) et formules de points xes (`
a la Lefschetz).
4.
Motifs
Soit X une variete algebrique sur k algebriquement clos. Pour chaque

nombre premier premier a` la caracterisque, la topologie etale fournit des
i (X, Z ). Si k est un sous-corps de C, on
groupes de cohomologie -adique Het

dispose disomorphismes de comparaison
H i (X(C), Z) Z
i
Het
(X, Z ).
Pour k de caracteristique > 0, il nexiste pas de cohomologie entière fonci (X, Z ) ont,
torielle donnant lieu a` de tels isomorphismes. Neanmoins, les Het

pour variable, un air de famille . Pour i = 1, et X projective et lisse,
1 (X, Z )
ole de linexistante theorie entière : il redonne les Het
Pic0 (X) joue le r

et est un objet sur Z , en ce que les groupes dhomomorphismes entre
schemas abeliens sont de type ni.
La theorie des motifs est dabord une tentative pour trouver un substitut a` linexistante cohomologie entière, expliquant lair de famille entre les
i (X, Z ), sp
ecialement pour X projectif et lisse. On reconnat la patte du
Het

Matre dans lidee que le problème nest pas de denir ce quest un motif :
le problème est de denir la categorie des motifs, et de degager les structures
quelle porte. Ces structures devraient permettre de prouver la conjecture de
Weil sur le modèle de Serre [1960]. Voir Grothendieck [1969].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
18
P. DELIGNE
Cest à cette occasion que Grothendieck a invente la notion de categorie

tannakienne, developpee dans la thèse de Saavedra [1972]. Il sagissait de formaliser la notion de produit tensoriel de motifs, correspondant par la formule
de K
unneth au produit de varietes.
Meme si la theorie des motifs na pas atteint son but original, son inuence
a ete grande. Je renvoie a` la conference de Seattle sur les motifs [Jannsen et al.,
1994] pour un panorama de ses applications.
Bibliographie
Une bibliographie de Grothendieck est donnee au debut de son Festschrift [Cartier
et al. 1990, p. xiiixx].
Autres sources : lintroduction, et larticle de J. Dieudonne : De lanalyse fonctionnelle aux fondements de la geometrie algebrique , dans ce meme Festschrift
(p. 114), les exposes aux congrès internationaux de mathematiciens de J.-P. Serre
(Stockholm 1962) : Geometrie algebrique (p. 190196), J. Dieudonne (Moscou
1966) : Les travaux de Alexander Grothendieck (p. 2124), et, pour RiemannRoch et les groupes de Grothendieck, celui de H. Cartan (Moscou 1966) : Luvre
de Michael F. Atiyah (p. 914).
Artin (M.)
[1969]
Algebraization of formal moduli I, dans Global Analysis (Papers in honor

of K. Kodaira), Tokyo : Univ. Tokyo Press, 1969, p. 2171.
Cartier (P.), Illusie (L.), Katz (N.M.) et al., ed.

[1990]
The Grothendieck Festschrift, vol. I, vol. 86 de Progress in Mathematics.

Boston : Birkhauser.
Grothendieck (A.)
[1955]
A general theory of bre spaces with structure sheaf. University of Kansas,

1955.
[1957]
Sur quelques points dalgèbre homologique, T

ohoku Math J., 9 (1957),
p. 119221.
[1963]
Residus et dualite, prenotes pour un seminaire Hartshorne , manuscrit.

Voir Hartshorne [1966].
[1969]
Standard conjectures on algebraic cycles, Algebraic Geometry (Coll. Tata

Inst., 1968), Oxford Univ. Press, (1969), p. 193199.
[1985]
Recoltes et Semailles : R
eexions et temoignage sur un pass
e de mathematicien, Montpellier : Univ. Sci. et Tech. Languedoc et CNRS, 1985.
[EGA]
ements de Geometrie Algebrique. Inst. Hautes Etudes
El
Sci. Publ. Math.
4, 8, 11, 17, 20, 24, 28, 32. En collaboration avec J. Dieudonne.
` 3
SEMINAIRES
ET CONGRES

QUELQUES IDEES
19
[FGA]
Fondements de la Geometrie Algebrique. Extraits du seminaire Bourbaki

19571962 : seminaires 149, 182, 190, 195, 212, 221, 232, et 236.
[SGA]
Seminaire de Geometrie Algebrique du BoisMarie. Le seminaire SGA 5 :

Cohomologie -adique et Fonctions L (1965-66) a ete nalement publie :
vol. 589 des Lecture Notes in Math., Berlin-Heidelberg : Springer, 1977.
Hartshorne (R.)
[1966]
Residues and duality, vol. 20 des Lecture Notes in Math., BerlinHeidelberg : Springer, 1966.
Illusie (L.)
[1990]
Categories derivees et dualite, travaux de J.-L. Verdier, Enseign. math.,

36 (1990), p. 369-391.
Jannsen (U.), Kleiman (S.) and Serre (J.-P.), ed.

[1994]
Motives, vol. 55 des Proceedings of Symposia in Pure Mathematics. Providence : Amer. Math. Soc., 1991.
Saavedra (N.)
[1972]
Categories tannakiennes, vol. 265 des Lecture Notes in Math., BerlinHeidelberg : Springer, 1972.
Serre (J.-P.)
[1960]
Analogues k
ahleriens de certaines conjectures de Weil, Ann. of Math., 71
(1960), p. 392-394 ; uvres, vol. II. Berlin-Heidelberg : Springer, 1986,
p. 1-3.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
Naissance des br
es et homotopie
Beno Eckmann
Home is where one starts from. As one grows older

the world becomes stranger, the pattern more complicated
of dead and living
T.S. Eliot
R
esum
e
Il sagit dun episode de lhistoire des mathematiques bien delimite dans son sujet et dans le temps : les origines de la theorie
homotopique des espaces bres, de 1935 a` 1950 environ (les debuts de la theorie des bres vectoriels, avec groupe de structure,
etc. ne sont pas abordes). Durant cette periode, la combinaison
des idees de Hurewicz sur les groupes dhomotopie avec la notion de bre suggeree par les brations de Hopf a livre une
foule de resultats inattendus. Beaucoup de developpements ulterieurs dune importance fondamentale en topologie, en algèbre et
au-del`
a, trouvent leur origine dans cet episode.
Abstract
This is about an episode in the history of mathematics, very much
restricted in content and in time: the origins of the homotopy
theory of bre spaces, roughly from 1935 to 1950 (the beginnings
of the theory of vector bundles bre bundles, structure group,
etc. are not treated). During that period, the combination of
Hurewiczs ideas concerning homotopy groups with the concept
of bre space suggested by the Hopf brations has led to a great
AMS 1991 Mathematics Subject Classication: 01A60, 55-03

Forschungsinstitut f
ur Mathematik, ETH-Zentrum, Ch8092 Z
urich, Suisse
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
22
B. ECKMANN
number of unexpected results. Many later developments of fundamental importance in topology, algebra, and mathematics in
general have their origin in this episode.
Dans cet expose je traite un episode de lhistoire des mathematiques bien

delimite dans son sujet et dans le temps : les origines de la theorie homotopique
des espaces bres, de 1935 à 1950 environ. Je le raconte plus ou moins comme
je lai vecu moi-meme donc de facon assez personnelle.
Deux avertissements :
1. Les debuts de la theorie des bres vectoriels (avec groupe de structure,
bre bundles , sphere bundles ) qui datent a` peu près de la meme
periode, ne seront pas abordes. Je me borne aux resultats et problèmes
lies à lhomotopie.
2. En parlant de nous je pense dun c
ote au groupe des elèves de Heinz
Hopf, de lautre aux trois auteurs ou groupe dauteurs qui ont developpe independamment le sujet, les communications ayant ete interrompues par la guerre : Ehresmann et Feldbau [1941], Hurewicz et Steenrod
[1941], et moi-meme [Eckmann 1941-42a]. Quant aux references jai la
chance de pouvoir utiliser la bibliographie de louvrage monumental de
Dieudonne : A History of Algebraic and Dierential Topology 1900-1960
[Dieudonne 1989].
Les debuts etaient simples. Nous avons realise que lon pouvait combiner
les idees de Hurewicz [1935, 1936] sur les groupes dhomotopie avec la notion
de bre suggeree par les brations de Hopf [1935]. Il en sortait une foule de
resultats nouveaux et de problèmes interessants. On peut dire que beaucoup
de developpements ulterieurs en topologie, algèbre et dans bien dautres disciplines ont leur origine dans cet episode ; ils ont cree un reseau toujours plus
complexe de disciplines et de relations entre elles tout en contribuant a`
lunite des mathematiques.
1.
Fibrations de Hopf et g
en
eralisation
1.1. Le terme bration au sens de cet expose apparat pour la pre

die Abbildungen von Spharen
mière fois en 1935 dans le memoire de Uber
auf Sph
aren niedrigerer Dimension . En annexe on trouve les brations de
` 3
SEMINAIRES
ET CONGRES
ET HOMOTOPIE
NAISSANCE DES FIBRES
23
Hopf de sphères en sphères

p : S 2n+1 CP n , bre S 1
p : S 4n+3 HP n , bre S 3
p : S 15 OP 1 , bre S 7
Elles sobtiennent en representant la sphère par des coordonnees z0 , z1 , ..., zn ,
respectivement nombres complexes, quaternions, et octonions (nombres de
Cayley), et en passant aux coordonnees homogènes z0 : z1 : ... : zn . CP n est
lespace projectif complexe, HP n lespace projectif quaternionien, et OP 1 la
droite projective des octonions dans ce dernier cas, seul n = 1 est possible
puisque les octonions ne sont pas associatifs. Le cas n = 1 donne les brations
S 3 S 2 et S 7 S 4 et S 15 S 8 . La projection p est une application continue, et la preimage p1 (x) dun point x est S 1 , S 3 , S 7 respectivement. Ainsi
les sphères en question sont decomposees de facon très speciale en sphères S 1 ,
S 3 , ou S 7 .
1.2. Sans en avoir donne une denition, Hopf appelait simplement ces
decompositions des brations. Cette expression avait ete utilisee auparavant
par Seifert [1932] dans un cas assez particulier concernant les varietes à 3
dimensions, o`
u interviennent des bres exceptionnelles ; ce concept est
reste interessant jusqu`
a ce jour. Peu après nous avons remarque quil sagissait
dune situation que lon rencontrait en geometrie dans beaucoup dautres cas :
on etait en presence dune application continue p : EB o`
u toutes les
u chaque
preimages p1 (b) = Fb , b B sont homeomorphes entre elles, et o`
1
ace à p, a`
point b B a un voisinage U tel que p (U ) est homeomorphe, gr
U Fb . On dit que E est un espace bre (localement trivial), B est la base,
p la projection, et les Fb sont les bres, homeomorphes à une bre-type F .
1.3. Exemples typiques :
1. E est lespace des vecteurs tangents unites dune variete dierentiable
B de dimension n (munie dune metrique riemannienne), Fb lensemble
des vecteurs tangents unites en b, F = S n1 .
2. E est un groupe de Lie, F un sous-groupe ferme, B lespace homogène
correspondant.
3. E = Vn,m , (m n), lespace des m-repères orthonormes dans Rn , B =
Vn,m1 obtenu en omettant le dernier vecteur, et F = S nm . Analogue
unitaire dans Cn , et dautres obtenus en remplacant m 1 par m k.
4. Cas particulier de 2) et 3). E = U (n), F = U (n 1) et B = S 2n1 , et
de manière analogue pour les groupes orthogonaux.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
24
B. ECKMANN
Dans tous ces cas la projection p est lapplication evidente.
2.
Groupes dhomotopie
En 1935 egalement, puis en 1936, apparaissaient les Notes de Hurewicz
Beitrage zur Topologie der Deformationen . Il fut vite evident quil sagis-
sait dune chose très dierente de ce quon avait fait auparavant en topologie
algebrique (dite combinatoire à lepoque) a` lexception du groupe fondaur,
mental dont les groupes dhomotopie n (X), n = 1, 2, 3, ... etaient, bien s
une generalisation. Dej`

a inventes par Cech
[1932], ces groupes etaient redenis
et utilises par Hurewicz avec des resultats inattendus. Dautre part ils nous
apparurent, un peu plus tard, merveilleusement adaptes aux brations pour
les questions homotopiques.
2.1. Rappelons dabord rapidement les denitions. On considère des espaces X pointes, cest-à-dire munis dun point-base x0 . Les applications ainsi
que les homotopies sont continues et pointees (respectant les point-bases). Les
elements de n (X) sont les classes dhomotopie des applications S n X, le
point-base de S n etant s0 . Soit h une application standard du cube unite I n
dans S n qui est un homeomorphisme de linterieur de I n sur S n s0 et qui envoie le bord In de I n sur s0 ; par lintermediaire de h on peut identier n (X)
a lensemble des classes dhomotopie I n , In X, x0 . En choisissant une direc`
tion distinguee on decompose I n en I 1 I n1 . Loperation de groupe est alors
denie par f +g comme suit : I = {0 t 1} est divise en I1 = {0 t 1/2}
et I2 = {1/2 t 1}. On comprime alors f sur I11 I n1 et g sur I21 I n1
et lon obtient f + g. Pour n = 1 cest bien laddition (non-commutative en
general) du groupe fondamental 1 (X). La denition est compatible avec les
homotopies, et les axiomes de groupe se verient exactement comme pour
1 (X). En particulier, lelement neutre est la classe de lapplication constante
(sur le point-base). De facon generale, quel que soit lespace quon applique
dans X, une application de cette classe est dite homotope a` zero . Pour
simplier la presentation je me permets de ne pas toujours distinguer entre
une application f et sa classe dhomotopie.
Une application h : XY induit, par composition S n XY , un
homomorphisme h : n (X)n (Y ). On voit facilement que pour un revetement XX les groupes dhomotopie n (X) et n (X) sont isomorphes
pour n 2. Par exemple n (S 1 ) = n (R) = 0 pour n 2. Dautres proprietes
elementaires : i (S n ) = 0 pour i < n (par approximation simpliciale on est
dans S n s0 qui est contractile), et i (X Y ) = i (X) i (Y ). Si X est
simplement connexe, alors lhomotopie pointee est equivalente a` lhomotopie
` 3
SEMINAIRES
ET CONGRES
ET HOMOTOPIE
25
libre.
2.2. Si X est un H-espace, c.-à-d. un espace muni dune multiplication
notee x x , continue avec element neutre e à homotopie près, on constate
sans peine que
(f1 + f2 ) (g1 + g2 ) = f1 g1 + f2 g2 ,
toujours a` homotopie près. Do`
u, e etant lapplication constante,
(f + e) (e + g) = f + g = f g,
(e + g) (f + e) = f + g = g f.
Il sensuit que f +g peut etre donne par la multiplication dans X, et que f +g =
g + f . Ainsi pour un groupe topologique G, 1 (G) est abelien ce qui etait
bien connu avant. Mais de facon generale, pour X arbitraire, les applications
I n , In X, x0 peuvent etre identiees aux applications I n1 , In1 X, x0
o`
u X est lespace des applications I, IX,
x0 (lespace des lacets de X en
x0 ). On a donc n (X) = n1 (X) pour n 2. Mais X est un Hespace
par la composition des lacets.
Les groupes dhomotopie n (X) sont donc abeliens pour tout X et pour
tout n 2. On dit que pour cette raison, lorsque ces groupes furent presentes
par Cech
en 1932, on ne croyait pas quils pourraient etre interessants.
2.3. Deux miracles :
1. Avec surprise nous avions constate que lon pouvait donner une demonstration très simple et transparente du fait que
n (S n ) = Z,
lisomorphisme etant donne en associant à f : S n S n son degre. En eet,
par les methodes dapproximation simpliciale on voit sans peine que le degre
est un invariant dhomotopie, et que lon a :
a) lapplication degre : n (S n )Z est un homomorphisme, et
b) n (S n ) est engendre par lidentite (degre = 1).
En dautres termes, on retrouve le theorème de Hopf [1933] qui dit que
deux applications S n S n ayant meme degre sont homotopes.
2. A laide de la suite exacte des brations, dont il sera question dans la
section suivante, on constate que
3 (S 2 ) = Z,
engendre par la bration de Hopf S 3 S 2 . Donc, en particulier, il existe une
innite dapplications non-homotopes S 3 S 2 . Ce fait avait ete etabli en
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
26
B. ECKMANN
1931 par Hopf [1931] dans son celèbre travail o`

u il a introduit l invariant de
Hopf ; la nouvelle demonstration etait très dierente, a` la fois plus simple
et plus precise. Dans la suite exacte
p
...3 (S 1 )3 (S 3 ) 3 (S 2 )2 (S 1 )...
on voit que 3 (S 3 ) = Z est isomorphe à 3 (S 2 ) en vertu de la projection
p : S 3 S 2 qui est la bration de Hopf ; et lidentite de S 3 etant le generateur
de 3 (S 3 ), celui de 3 (S 2 ) est la projection p de lidentite, donc simplement p.
3.
La suite exacte des br

es
Ce que lon appela plus tard la suite exacte dune bration p : EP

avec bre F etait dabord formule sans eches, simplement comme une serie
disomorphismes reliant les groupes dhomotopie de E, de B, et de F . Je
me permets dutiliser ici demblee les suites exactes qui sont beaucoup plus
commodes (Lhistoire des suites exactes est assez complexe et curieuse ; je
nentre pas dans ce sujet.)
3.1. On a des homomorphismes induits
p
n (F )
n (E) n (B),
o`
u p est la bration, et i : F E linclusion dans E de la bre F au-dessus
du point-base de B ; le point-base de E est choisi dans F . Pour denir un
homomorphisme : n (B)n1 (F ), et pour montrer que la longue suite
ainsi obtenue
i
n (E) n (B) n1 (F )...

...n (F )
est exacte (à chaque etape image = noyau) on se sert de la propriete appelee
rel`
evement des homotopies :
Lemme 3.1. Soient f : XE et f = pf : XB. Alors toute homotopie H de f peut se relever en une homotopie H de f telle que H = pH .
En fait les seuls espaces X qui interviennent sont les sphères S n et les
cubes I n . Admettons ce Lemme ; je reviens dans 3.3. sur sa demonstration.
Le Lemme est suggere par le cas classique o`
u il sagit dun revetement (bre
discrète) et o`
u le relèvement existe et est meme unique. Notons une consequence immediate :
` 3
SEMINAIRES
ET CONGRES
ET HOMOTOPIE
27
Si f = pf et si f est homotope `
a g, alors il existe g : XE, homotope

a
`
telle que g = pg .
Autrement dit : Toute application dans B homotope a` une projection
est une projection. En particulier, toute application homotope a` zero est une
projection .
f ,
3.2. Un element de n (B) peut etre represente par une application f de

dans B telle que le bord de I n , qui est une sphère S n1 , est envoye sur
u f
le point-base. Comme I n est contractile, f est une projection f = pf o`
n
n1
applique I dans E et son bord S
dans F . Toutes les classes dapplications

n
f de ce type de I dans E, avec laddition analogue a` celle de n (E), forment
le groupe dhomotopie relatif n (E mod F ) de E modulo F . Il est applique
par p dans n (B), et ce qui vient detre dit montre que p est surjectif ;
linjectivite decoule immediatement du Lemme. On a donc
In
n (E mod F ) = n (B).
Lhomomorphisme est alors deni par le passage de f : In B à f cidessus suivi de la restriction de f à S n1 .
Le fait que la suite longue est exacte se verie facilement dans chacune des
trois etapes. Par exemple, si pour f : S n E on a p f = 0 alors lhomotpie
a zero de pf se relève et f est homotope à une application S n F , donc est
`
une image par i . Je laisse au lecteur le soin dexaminer ce qui ce passe en
petite dimension.
3.3. Tout cela nous paraissait evident, en particulier le Lemme. Mais naturellement il y avait quelque-chose a` demontrer, moyennant une hypothèse
a verier dans les exemples interessants.
`
Lhypothèse que javais choisie etait celle dune r
etraction. On suppose que
tout b B possède un voisinage U (b) tel quil existe une retraction R(x, b) de
ument de b :
p1 (U (b)) sur Fb dependant contin
R(x, b) Fb pour tout x p1 (U (b)), R(x, b) = x si x Fb .
Dans les exemples très concrets mentionnes en 1.3., E, B et F sont des varietes
dierentiables ; on peut les munir dune metrique riemannienne et construire
facilement une telle retraction a` laide des geodesiques orthogonales à chaque
bre.
Si une retraction R(x, b) est donnee, on choisit y Fb et on pose, pour
tout b U (b)
t(b ) = R(y, b ).
Alors t est une application de U (b) dans E qui est un homeomorphisme de
U (b) sur un ensemble V (y) transversal aux bres Fb . Ce relèvement dun
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
28
B. ECKMANN
voisinage autour dun point y arbitraire permet de relever morceau par

morceau une homotopie et de demontrer ainsi le Lemme.
Remarque 3.2. Pour etablir le Lemme et la suite exacte dhomotopie
il sut de considerer une application p : EB, dappeler bres les
images inverses p1 (b), et de supposer lexistence dune retraction R(x, b) ; il
nest pas necessaire quil sagisse dun bre localement trivial au sens de 1.2..
Serre [1951] est alle encore plus loin et a seulement suppose lexistence dun
relèvement pour les applications I n B.
4.
R
esultats
Le Lemme et la suite exacte dhomotopie etablis, les resultats tombaient

du ciel ! Les cas immediats etaient simplement bases sur ce qui etait connu,
de facon elementaire, sur les groupes dhomotopie des sphères : i (S 1 ) = 0
pour i 2, i (S n ) = 0 pour i < n, et n (S n ) = Z pour tout n 1.
4.1. Les brations de Hopf S 3 S 2 avec bre S 1 , S 7 S 4 avec bre S 3 ,
et S 15 S 8 avec bre S 7 , donnent
3 (S 2 ) = Z, 7 (S 4 ) Z, et 15 (S 8 ) Z,
le generateur de Z correspondant a` la projection.
4.2. Les brations S 2k+1 CP k , k 1 avec bre S 1 donnent
2 (CP k ) = Z, et i (CP k ) = i (S 2k+1 ), i > 2.
4.3. Les brations U (n)S 2n1 avec bre U (n 1) donnent
s ((U (n))) = s (U (
s+1
s+1
)), n
pour s = impair,
2
2
s+2
s+2
)), n
pour s = pair.
2
2
Cest ce que lon appela plus tard la stabilite des s ((U n)). Pour les premières valeurs de s les groupes stables sobtenaient facilement : 1 (U (n)) =
1 ((U (1)) = Z, engendre par lidentite S 1 U (1). Pour s = 2 on a
s (U (n)) = s (U (
2 (U (n)) = 2 (U (2)) = 2 (SU (2)) = 2 (S 3 ) = 0,

et pour s = 3 de facon analogue
3 (U (n)) = 3 (S 3 ) = Z, n 2.
` 3
SEMINAIRES
ET CONGRES
ET HOMOTOPIE
29
Pour s > 3 cela saverait bien plus dicile. En examinant de très près
lhomomorphisme dans la suite exacte de la bration en question jarrivais
[Eckmann 1941-42b] a` determiner 4 = 0 et 5 = Z.
4.4. Le Lemme sapplique, naturellement, non seulement aux groupes
dhomotopie, mais egalement à toutes sortes de questions dhomotopie pour
les applications de X dans la base B dune bration EB. Considerons
trois exemples :
a) X = I n . Alors toute application f : I n B est une projection pf . Ce
resultat presque trivial dans le cas de la bration 3) dans 1.3. parat etre la
première consequence du Lemme pour un problème danalyse (theorème de
Wazewski, mentionne dans [Eckmann 1941-42a] : completer une matrice reelle
orthogonale n (m 1), m < n, qui est une fonction continue dans I n , par
une ligne supplementaire).
b) X = B et f = identite. Un relèvement de f sappelle une section
de la bration ; les champs de vecteurs ou de repères tangents a` une variete
en constituent un cas particulier. De la suite exacte on deduit une condition
necessaire pour lexistence dune section : les n (E) se decomposent en somme
directe de n (B) et de n (F ). On arrive ainsi a` etablir des cas interessants
de non-existence de sections. Dans Eckmann [194243a] jai montre, à laide
darguments plus compliques adaptes à ce cas, que les sphères de dimension
4k + 1, k > 0 nadmettent pas deux champs de vecteurs tangents unite et
orthogonaux.
c) On appelle essentielle une application f : XY telle que toute application homotope à f est surjective. Daprès le Lemme on voit que si lidentite
de E est essentielle alors il en est de meme pour la projection EB. Toutes
les brations dans 1.3. en fournissent des exemples, puisque E est une variete
compacte sans bord et que lapplication identique est de degre 1.
Bien s
ur, non seulement les resultats tombaient du ciel, mais aussi les problèmes. Les exemples ci-dessus le montrent clairement. Citons simplement que
la determination des groupes dhomotopie stables des groupes unitaires, pour
s > 5, et du nombre maximum de champs de vecteurs tangents orthonormaux
sur une sphère restait ouverte.
5.
Equivalence
dhomotopie, espaces asph
eriques
5.1. On ne peut pas citer les Notes de Hurewicz [1935] sans parler des espaces aspheriques. Les espaces consideres etaient des complexes cellulaires
(`
a lepoque simpliciaux) ; X est dit aspherique si n (X) = 0 pour tout
n 2. Pour de tels espaces X et Y , Hurewicz montre par induction sur
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
30
B. ECKMANN
les n-squelettes de X que

a) Si f et g : XY induisent les memes homomorphismes
f = g : 1 (X)1 (Y )
des groupes fondamentaux, alors f et g sont homotopes.
b) Pour tout homomorphisme h : 1 (X)1 (Y ) il existe f : XY
telle que f = h.
Do`
u:
c) Si 1 (X) et 1 (Y ) sont isomorphes, alors il existe f : XY et
g : Y X telles que gf et f g sont homotopes a` lidentite de X et de Y
respectivement. On est ainsi amene à la notion dequivalence dhomotopie.
De facon generale, pour des espaces X et Y (non necessairement aspheriques) une application f : XY telle quil existe g : Y X comme dans
c) ci-dessus est dite
equivalence dhomotopie. On dit dans ce cas aussi que X et
Y ont meme type dhomotopie (notation X Y ). Il est clair quil sagit dune
generalisation, très importante, de la notion dhomeomorphisme. Une equivalence dhomotopie f : XY induit des isomorphismes f : n Xn Y
pour tout n 1 ; et de meme pour lhomologie.
Pour un espace aspherique X on a donc les resultats suivants :
Le type dhomotopie de X est complètement d
etermin
e par son groupe
fondamental. En particulier, tous les groupes dhomologie de X sont d
etermines par le groupe fondamental de X. Si le groupe fondamental dun espace
aspherique est trivial, alors X point, c.-`
a-d. X est contractile.
5.2. Des raisonnements analogues sappliquent à des espaces aspheriques
en dimension n > N . Il sut de modier legèrement a) et b) dans 5.1. ci
dessus : il sagit alors dapplications des N squelettes de X et Y ; et dans a)
f nest pas necessairement homotope à g, mais à une application qui concide
avec g sur le (N 1)squelette de X. On en deduit que les groupes dhomologie
Hi (X), i < N sont determines par 1 (X). Il nen est pas ainsi, en general,
de H modulo le sousgroupe des
pour HN (X) ; mais le groupe quotient HN
N
elements spheriques (representes par des cycles images de sphères) de X
est determine par le groupe fondamental.
Plus generalement, si i (X) = 0 pour i < k et k < i < N , alors Hi (X),
(X) sont d
etermines par k (X).
pour i < N et HN
5.3. Le resume ci-dessus ne correspond pas exactement à ce quil y a dans
les Notes de Hurewicz à ce sujet. Dun c
ote elles vont bien plus loin (homomorphisme de Hurewicz, etc.), et de lautre je les ai depassees un peu dans
5.2. mais à lepoque dej`
a il etait clair que les idees sappliquaient de cette
facon plus generale. Je voudrais ainsi non seulement souligner limportance
` 3
SEMINAIRES
ET CONGRES
ET HOMOTOPIE
31
des idees de Hurewicz, mais aussi preparer une application typique et concrète
aux brations.
6.
Fibrations de sph`
eres en tores
6.1. Je considère [Eckmann et al. 1949] une bration S n B dont la

a s dimensions, n > 2 et s 1. On voit facilement que
bre est un tore Ts `
le cas n = 2, s = 1 nest pas possible. La bration est supposee susamment
regulière pour que le Lemme et ses consequences (Section 3) sappliquent. La
suite exacte dhomotopie donne
1 (B) = 0, 2 (B) = 1 (Ts ) = Zs ,
i (B) = 0 pour 2 < i < n.
On desire comparer B à un espace Y connu ayant memes groupes dhomotopie que ci-dessus. On choisit Y egal au produit topologique de s copies
de CP m avec m susamment grand ; il sut (voir 4.2.) dutiliser la suite
dhomotopie de la bration a) dans 1.1. pour constater quon a les i desires,
et meme = 0 un peu au-del`
a de n.
Daprès 5.2. on a alors pour les groupes dhomologie
Hi (B) = Hi (Y ) pour i < n
et
Hn (B) = Hn (Y ) = Hn (Y ).
La dimension de B etant n s, il sensuit que Hn (Y ) = 0, donc que n est

impair (CP m , donc Y a de lhomologie = 0 si et seulement si on est dans une
dimension paire). Comme alors Hn1 (Y ) = 0, s doit etre egal à 1.
ee en tores Ts seulement si n est impair et si
La sphère S n peut etre br
s = 1. Et dans ce cas on a la bration de Hopf.
6.2. On retrouve ainsi le resultat bien connu : si S n est un groupe de Lie,
alors son rang (la dimension des sous-groupes abeliens maximaux) doit etre
egal à 1. Mais la methode geometrique va plus loin [Samelson 1940] :
Si S n , n > 1, est un groupe de Lie, donc de rang 1, les sous-groupes a`
un paramètre, homeomorphes à S 1 sont tous conjugues entre eux. Comme
ils sont determines par la tangente en lelement neutre, leur ensemble peut
etre identie à lespace projectif reel RP n1 ; dautre part il sidentie aux
classes de S n modulo le normalisateur N (S 1 ) dun des sous-groupes S 1 . Ce
normalisateur est forme par un nombre ni de copies de S 1 . On a donc une
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
32
B. ECKMANN
bration S n RP n1 qui donne lieu a` une suite exacte o`

u le i de la base
est egal à i (S n1 ) et le i de la bre egal à i (S 1 ). En particulier, la suite
2 (S 1 )2 (S n )2 (S n1 )1 (S 1 )1 (S n )
u n 1 = 2 et n = 3 :
etant exacte, il sensuit que 2 (S n1 ) = 1 (S 1 ) = Z, do`
Les seules sph`
eres qui puissent
etre des groupes de Lie sont S 1 et S 3 .
7.
Et apr`
es ?
On peut tracer linuence de cet episode sur presque toutes les disciplines
mathematiques jusqu`
a nos jours (cela pourrait probablement se dire de toutes
les idees qui etaient nouvelles il y a longtemps). Suivre les relations mutuelles
entre les dierentes tendances, ecoles et modes serait une tache fascinante
mais très dicile. Ny a-t-il pas des idees et des methodes très à la mode,
importantes pour un cercle de problèmes, qui disparaissent tout a` coup pour
renatre plus tard dans un autre contexte ? The pattern becomes more and
more complicated of dead and living .
Je me borne à mentionner ici une liste de developpements directement lies
a ce que je viens de decrire plus haut, et qui ont eu lieu immediatement après
`
ou meme pendant cet episode. Il ne sagit que dallusions sommaires.
7.1. Groupes dhomotopie des sphères. Les resultats de 4.1. concernent
u il y a un terme Z. Dautres cas
des cas de n (S m ) avec n = 2m 1 o`
semblables etaient connus. Mais en 1950, Serre [1951] a demontre des resultats
sensationnels sur les n (S m ) pour n > m : Ces groupes sont toujours nis a`
lexception du cas m pair et n = 2m 1 o`
u cest une somme directe de Z et
dun groupe ni. Mais autrement le domaine des n (S m ) est trop vaste pour
etre aborde ici de facon plus generale ( suspension de Freudenthal [1937],
groupes dhomotopie stables des sphères).
7.2. Espaces dEilenberg-MacLane K(G, n). Eilenberg et MacLane [1943,
1945b] ont examine des espaces avec i = 0 pour i = n. Daprès 5.2. le type
dhomotopie dun tel espace est determine par n et par n = G, abelien si
n > 1. Ces espaces jouent un r
ole universel pour lhomologie, la cohomologie,
et pour toutes les operations. Lexistence pour un G donne a ete etablie par
Whitehead [1949].
7.3. Homologie et cohomologie des groupes, alg`
ebre homologique. Lhomologie dun espace aspherique etant determinee par son groupe fondamental G
qui peut etre donne arbitrairement des methodes algebriques ont tout de
` 3
SEMINAIRES
ET CONGRES
ET HOMOTOPIE
33
suite ete developpees pour les calculer à partir de G. Cest ainsi que la (co)homologie des groupes est nee, et avec elle lalgèbre homologique beaucoup
plus generale.
7.4. Type dhomotopie. Dès les annees 40, J.H.C. Whitehead a montre
quune application f : XY qui induit des isomorphismes de tous les
groupes dhomotopie est une equivalence dhomotopie. Il a ameliore la demonstration plus tard en creant la notion de CW-complexe [Whitehead Works
III, p. 95-105].
7.5. Cat
egories et foncteurs. Eilenberg et MacLane [1942, 1945a] ont realise que les idees generales derrière les notions dequivalence dhomotopie,
disomorphismes naturels , etc., ont une signication beaucoup plus profonde. Au debut, leur theorie des categories, foncteurs, et equivalences naturelles semblait etre juste un langage precis, mais on en a degage plus tard la
structure mathematique, aussi fondamentale quutile.
7.6. Sphères parall
elisables. Une variete de dimension n, dierentiable (et
munie dune metrique riemannienne) est dite parallelisable si elle admet un
champ continu de n-repères orthonormes tangents. Daprès 4.4. les sphères
S 4k+1 avec k > 0 ne sont certainement pas parallelisables. Kervaire [1958],
Bott et Milnor [1958] ont demontre que S n est parallelisable (si et) seulement
si n = 1, 3, ou 7. Ce resultat se deduira plus tard très simplement du celèbre
theorème dAdams [1960]. Adams [1962] a determine le nombre maximum
exact k tel que S n possède un k-repère tangent.
7.7. Groupes dhomotopie stables des groupes unitaires. Les resultats très
incomplets de 4.3. ont ete peu a` peu ameliores. Le problème se trouvait complètement resolu par Bott [1956] à laide de methodes subtiles de geometrie
dierentielle : n (U (m)) = Z pour n impair, m n+1
2 , et = 0 pour n pair
n+2
et m 2 . Cest la periodicite de Bott (resultat analogue mais plus
complique pour les groupes orthogonaux). Cette solution geometrique a eu
des consequences enormes (K-theorie topologique, foncteurs cohomologiques
generaux).
7.8. Une remarque personnelle a` propos du dernier point : javais considere dès les premiers calculsles applications f lin
eaires de S n dans U (m),
u les Aj sont des matrices m m. Il
c.-à-d. de la forme f (x) =
xj Aj o`
sensuit que les Aj sont des matrices unitaires de Hurwitz-Radon [Hurwitz 1923, Radon 1922, Eckmann 194243b] ; reciproquement tout système de
n + 1 matrices de Hurwitz-Radon donne une application lineaire de S n dans
U (m). Javais conjecture que, dans le domaine stable, chaque classe dhomotopie contient une telle application lineaire, et que si une application lineaire
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
34
B. ECKMANN
est homotope à zero, alors elle lest de facon lineaire, c.-à-d. en vertu dune
matrice de Hurwitz-Radon supplementaire qui fournit une application lineaire
de S n+1 dans U (m). Daprès ce quon sait sur ces matrices cela aurait donne
le theorème de Bott. Mais cest seulement après Bott, en utilisant son resultat sous la forme multiplicative de la K-theorie que jai pu demontrer la
conjecture (voir [Eckmann 1994]). Il reste toutefois le problème dune demonstration directe qui reduirait la periodicite de Bott à la discussion algebrique
des matrices de Hurwitz-Radon.
Bibliographie
Adams (J.F.)
[1960]
On the nonexistence of elements of Hopfinvariant one, Ann. of Math.,

72 (1960), p. 20104.
[1962]
Vector elds on spheres, Ann. of Math., 75 (1962), p. 603632.
Bott (R.)
[1956]
An application of the Morse theory to the topology of Lie groups, Bull.

Soc. Math. France, 94 (1956), p. 251281.
Bott (R.) et Milnor (J.)

[1958]
On the parallelizability of spheres, Bull. Amer. Math. Soc., 64 (1958),

p. 8789.
Cech
(E.)
[1932]
H
oherdimensionale Homotopiegruppen. In Verhandl. des intern. Math.
Kongresses, Z
urich, 1932, vol. 2, p. 203.
Dieudonn
e (J.)
[1989]
A History of Algebraic and Dierential Topology 1900-1960. Boston :

Birkha
user Verlag, 1989.
Eckmann (B.)
[1941-42a] Zur Homotopietheorie gefaserter Raumen, Comment. Math. Helv., 14
(1941-42), p. 141192.
[1941-42b] Uber
die Homotopiegruppen von Gruppenraumen, Comment. Math.
Helv., (1941-42), p. 234256.
[194243a] Systeme von Richtungsfeldern auf Spharen und stetige Losungen komplexer linearer Gleichungen, Comment. Math. Helv., 15 (194243), p. 1
26.
[194243b] Gruppentheoretischer Beweis des Satzes von Hurwitz-Radon u
ber die
Komposition quadratischer Formen, Comment. Math. Helv., 15 (1942
43), p. 358366.
` 3
SEMINAIRES
ET CONGRES
ET HOMOTOPIE
[1994]
35
Hurwitz-Radon matrices revisited : From eective solution of the Hurwitz

matrix equation to Bott periodicity, The Hilton symposium 1993. Topics
in topology theory, CRM Proceedings and Lecture Notes, 6, Providence :
Amer. Math. Soc., 1994, p. 2335.
Eckmann (B.), Samelson (H.) et Whitehead (G.)

[1949]
On bering spheres by toruses, Bull. Amer. Math. Soc., 55 (1949), p. 433

438.
Ehresmann (C.) et Feldbau (J.)

[1941]
Sur les proprietes dhomotopie des espaces bres, C. R. Acad. Sci. Paris,
212 (1941), p. 945948.
Eilenberg (S.) et MacLane (S.)

[1942]
Natural isomorphisms in group theory, Proc. Nat. Acad. Sci. USA, 28

(1942), p. 537543.
[1943]
Relations between homology and homotopy groups, Proc. Nat. Acad. Sci.
USA, 29 (1943), p. 155158.
[1945a]
General theory of natural equivalences, Trans. Amer. Math. Soc., 58

(1945), p. 231294.
Relations between homology and homotopy groups of spaces, I, Ann. of
Math., 46 (1945), p. 480509.
[1945b]
Freudenthal (H.)
[1937]
Uber
die Klassen von Spharenabbieldungen, Compositio Math., 5 (1937),
p. 299314.
Hopf (H.)
[1931]
Uber
die Abbildungen der dreidimensionalen Sphare auf die Kugelache,
Math. Ann., 104 (1931), p. 637665.
[1933]
Die Klassen der Abbildungen der ndimensionalen Polyeder auf die n

dimensionalen Sphare, Comment. Math. Helv., 5 (1933), p. 3954.
Uber
die Abbildungen von Spharen auf Spharen von niedrigerer Dimension, Fund. Math., 25 (1935), p. 427440.
[1935]
Hurewicz (W.)
[1935]
[1936]
Beitr
age zur Topologie der Deformationen I, II, Proc. Konink. Nederl.
Akad. Wetensch., 38 (1935), p. 112119 et 521528.
Beitr
age zur Topologie der Deformationen III, IV, Proc. Konink. Nederl.
Akad. Wetensch., 39 (1936), p. 117126 et 215224.
Hurewicz (W.) et Steenrod (N.)

[1941]
Homotopy relations in bre spaces, Proc. Nat. Acad. Sci. USA, 27 (1941),
p. 6064.
Hurwitz (A.)
Uber
die Komposition quadratischer Formen, Math. Ann., 88 (1923), p. 1
25.
[1923]
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
36
B. ECKMANN
Kervaire (M.)
[1958]
Nonparallelizability of the nsphere for n > 7, Proc. Nat. Acad. Sci. USA,
44 (1958), p. 280283.
Radon (J.)
[1922]
Lineare Scharen orthogonaler Matrizen, Abh. Math. Sem. Univ. Hamburg,

(1922), p. 114.
Samelson (H.)
[1940]
Uber
die Sph
aren, die als Gruppenmannigfaltigkeiten auftreten, Comment. Math. Helv., 13 (1940), p. 144155.
Seifert (H.)
[1932]
Topologie dreidimensionaler geschlossener Raume, Acta Math., 60 (1932),

p. 147238.
Serre (J.-P.)
[1951]
Homologie singulière des espaces bres. Applications, Ann. of Math., 54

(1951), p. 425505.
Whitehead (J.)
[Works III] The Mathematical Works of J.H.C. Whitehead, vol III, London,
New York : Pergamon Press, 1962.
[1949]
On the realizability of homotopy groups, Ann. of Math., 50 (1949), p. 261

263 ; Works III, p. 221223.
` 3
SEMINAIRES
ET CONGRES
Hyperbolic Equations in the

Twentieth Century
Lars G
arding
Abstract
The subject began with Huygenss theory of wave fronts as
envelopes of smoother waves, and subsequent work by Euler,
dAlembert and Riemann. Singularities at the wave fronts were
not understood before Hadamards theory of partie nie at
the beginning of this century. Contributions by Herglotz and
Petrovsky and the theory of distributions created in the forties by
Laurent Schwartz greatly illuminated the study of singularities
of solutions of hyperbolic PDEs. Solutions of Cauchys problem
given by Hadamard, Schauder, Petrovsky, and the author are discussed. More recently, microlocal analysis, initiated by M. Sato
and L. H
ormander led to important advances in understanding
the propagation of singularities. Functional analysis together
with distributions and microlocal analysis are expected to be
useful well into the next century.
R
esum
e
Le sujet debute avec la theorie de Huygens qui considère les
fronts donde comme des enveloppes dondes plus regulières, et
se poursuit par les travaux de Euler, dAlembert et Riemann.
Les singularites des fronts donde nont pas ete comprises avant
la theorie de la partie nie de Hadamard au debut de ce
siècle. Les contributions de Herglotz, Petrovsky et dans les annees quarante, la theorie des distributions de Laurent Schwartz
ont eclaire letude des singularites des solutions des EDP hyperboliques. On passe en revue les solutions au problème de Cauchy
AMS 1991 Mathematics Subject Classication: 01A60, 35-03, 46-03

Dept. of Mathematics, Univ. of Lund, Box 118, S22100 Lund, Sweden
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
38
donnees par Hadamard, Schauder, Petrovsky et lauteur. Plus recemment, lanalyse microlocale de M. Sato et L. Hormander a
permis de grandes avancees dans la comprehension de la propagation des singularites. Lanalyse fonctionnelle, les distributions
et lanalyse microlocale seront certainement des outils importants
du prochain siècle.
1.
Introduction
The rst example of a hyperbolic equation was the wave equation

utt u = 0.
In one space variable n, the solutions describe free movements with velocity
1 in a perfectly elastic medium. A nonlinear version appears in one dimensional hydrodynamics. Riemanns 1860 treatment was later completed by the
Rankine-Hugoniot jump conditions and conditions of entropy. Further examples of hyperbolic equations and systems appeared in the theory of electricity
and magnetism and elasticity.
Originally, the adjective hyperbolic marked the connection between the
wave equation and a hyperbolic conoid. When applied to general partial differential operators or systems the term now indicates that one of the variables
is time t = t(x) and that the solutions of the system describe wave propagation with nite velocity in all directions. More precisely, the solution u of
Cauchys problem with no source function and with data given for t = const.
should have the property that the value of u at a point depends continuously
on the values of the data and their derivatives in a compact set. For an operator P (D) with constant coecients this means that there is a fundamental
solution E(x), i.e. a distribution such that P (D)E(x) = (x), whose support
is contained in a proper, closed cone.
In the rst half of the twentieth century, local existence by classical analysis of solutions to Cauchys problem for hyperbolic equations with smooth
data was the main problem. Soon after, functional analysis and distributions
came into play and the introduction around 1970 of pseudodierential operators and microlocal analysis of distributions was followed by a period of
important results on the propagation of singularities, both free and under reection in a boundary. Later this study was extended to nonlinear equations.
Another question, latent during the period, is the problem of global existence
of solutions for nonlinear equations close to linear ones. It took a new turn
with the study of blow-up times by Fritz John.
` 3
SEMINAIRES
ET CONGRES
HYPERBOLIC EQUATIONS IN THE TWENTIETH CENTURY
39
Only a sample of the main results can be mentioned here. In particular, I

refrain from the various hyperbolic aspects of hydrodynamics and the theory
of scattering in spectral analysis.
The development of the theory of hyperbolic equations from 1900 cannot
be understood without a review of some of the main results from the time
before 1900. It is done here briey under the heading of Prehistory.1
2.
Prehistory
With three space variables the wave equation describes free propagation of
light in physical space with velocity 1. For this equation, Poisson proved what
in modern terms amounts to the fact that the wave operator = t2 has
a fundamental solution
E(t, x) =
1
H(t)(t2 |x|2 )
2
with support on the forward lightcone t = |x|. It was then only too easy to
believe this to be a general phenomenon, for instance that the equations for
the propagation of light in media with double refraction follow the same rule
known under the name of Huygens principle:2 all light from a point-source is
concentrated to the surface given by the rules of geometric optics. Both G.
Lame and Sonya Kovalevski made this mistake till the use of Fourier analysis
proved that the existence of diuse light outside such surfaces is the rule and
the contrary an exception (for a historical review, see [G
arding 1989]).
A fundamental solution of the wave operator for two space variables was
found by Volterra and, at the turn of the century, Tedone tried the general
case, but could only construct what amounts to suciently repeated integrals
with respect to time of purported fundamental solutions. Behind these difculties is the fact that, in contrast to the properties of Laplaces operator,
the fundamental solutions of the wave operator are distributions with singularities outside the pole which get worse as the number n of space variables
increases. Before the theory of distributions, this was a formidable diculty.
3.
Partie nie
The obstacle which stopped Tedone, was surmounted by Hadamard in his

theory of partie nie, found before 1920 and exposed in [Hadamard 1932].
1
2
The remarks and notes of Hadamards book 1932give a fuller account.

Huygenss minor premise according to Hadamard [1932].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
40
His operator is the wave operator with smooth, variable coecients and has
the form

ajk (x)j k + lower terms
(3.1)
L(x, x ) =

where the metric form
ajk j k has Lorentz signature +, .... A direction
for which the inverse metric form is positive, zero or negative is said to be
time-like, light-like and space-like respectively. Surfaces with time-like and
space-like normals are said to be space-like and time-like respectively. The
light rays are the geodesics of length zero. A time function t(x) with t (x)
time-like is given.
The light rays with a positive time direction issued from a point y constitute the forward light cone Cy with its vertex at y. Inside this light cone, the
fundamental solution with its pole at y has the same form as in the elliptic
case
f (x, y)d(x, y)2n
(3.2)
where f is a smooth function and d is the geodesic distance between x and

y. The diculty is that d(x, y) = 0 when x Cy . The partie nie can
be said to be a renormalization procedure which extends this formula for
n odd to a distribution which is also a fundamental solution. For n even,
Hadamard uses what is called the method of descent. In the work by M.
Riesz [1949] the exponent 2 n of (3.2) is replaced by n where is
a complex paramater. At the same time f is made to depend on and a
denominator (/2)(( + 2 n)/2) is introduced. The stage is then set for
an analytical continuation with respect to . In this way and for selfadjoint
operators L, Riesz constructs kernels of the complex powers of L.
In his case, Hadamard could give a complete local solution of Cauchys
problem with data on a space-like surface, but the corresponding mixed problem with reection in a time-like surface presented insurmountable diculties.
4.
Friedrichs-Lewy energy density, existence proofs

by Schauder and Petrovsky
The discovery of Friedrichs and Lewy [1928] that 1 uu with u real is the
divergence of a tensor with a positive energy density on space-like surfaces
produced both uniqueness results and a priori energy estimates, decisive for
the later development.
A great step forward was taken by Schauder [1935, 1936a,b] who proved
local existence of solutions of Cauchys problem and also the mixed problem
` 3
SEMINAIRES
ET CONGRES
41
for quasilinear wave operators. The method is to use approximations starting from the case of analytic coecients and analytic data. The success
of these papers depends on stable energy estimates derived from the energy
tensor and the use of the fact that square integrable functions with square
integrable derivatives up to order n form a ring under multiplication.3
Only a year after Schauder, Petrovsky [1937] extended his results for
Cauchys problem to strongly hyperbolic systems, in the simplest case
(4.1)
ut +
n
Ak (t, x)uk + Bu = v,
uk = u/xk ,
and the corresponding quasilinear versions. Here the coecients are square
matrices of order m and the strong hyperbolicity with respect to the time
variable t means that all m velocities c given by

(4.2)
det(cI +
k Ak (t, x)) = 0
are real and separate for all real = 0. The method is that of Schauder
starting from the analytic case, but Petrovsky had to nd his own energy
estimate. For this he used the Fourier transform, but the essential point is to
be found in thirty rather impenetrable pages. Note that if the system (4.1) is
symmetric, i.e., the matrices Ak are Hermitian symmetric, then (4.2) holds
except that the velocities need not be separate. Moreover,

k (Ak u(t, x), u(t, x)) = O(|u(t, x)|2 + |u(t, x)||v(t, x)|)
t |u(t, x)|2 +
under suitable conditions on the coecients. Hence the proper energy density
on t = const is here simply |u(t, x)|2 dx.
Petrovskys paper was followed by a study [Petrovsky 1938] of conditions
for the continuity of Cauchys problem for operators whose coecients depend
only on time.
5.
Fundamental solutions, Herglotz and

Petrovsky
Herglotz [1926-28] and Petrovsky [1945] used the Fourier transform to construct fundamental solutions E(P, t, x) for constant coecient homogeneous
dierential operators P = P (t , x ) of degree m which are strongly hyperbolic
with respect to t. Every such fundamental solution E is analytic outside a
3
Soon after, Sobolev proved that one gets a ring also when n is replaced by (n + 1)/2
when n is odd and by (n + 2)/2 when n is even.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
42
wave front surface W (P ), which is the real dual of the real surface P (, ) = 0,
and vanishes for t < 0 and outside the outer sheet of W (P ).4 Petrovsky also
found explicit formulas for derivatives of order > mn of a fundamental solution in terms of Abelian integrals, integrated over cycles c(x) of real dimension
n 3 in the complex projective intersection I of P () = 0 and (x, ) = 0. The
cycles depend on the parity5 of n and the component T of C(P )\W where x
is situated. When (x) is homologous to zero in I, the region T is a lacuna,
i.e., the fundamental solution is a polynomial of degree m n in T and hence
vanishes when m < n. The point of the paper is that the vanishing of the
cycle is necessary when the lacuna is stable under small deformations of the
operator.6
The intriguing paper [1937] by Petrovsky became the starting point for the
development after 1950 of a general theory of hyperbolic dierential operators
by Leray and others and the paper [Petrovsky 1945] was generalized and
claried by Atiyah, Bott and G
arding [1970, 1973].
A decisive factor in the further development was the full use of the distributions of Laurent Schwartz and later by pseudodierential operators and
microlocal analysis.
6.
Hyperbolicity for constant coecients
Inspired by Petrovsky [1938], G

arding [1950] gave an intrinsic denition of
the hyperbolicity of dierential operator P (D) with constant coecients and
principal part Pm as follows. The operator is said to be hyperbolic with respect
to a hyperplane (x, N ) = 0 or to be in a class hyp(N ) if
(6.1) all smooth solutions u of P u = v tend to zero locally uniformly in the
halfspace (x, N ) > 0 when all their derivatives tend to zero locally uniformly
in the hyperplane (x, N ) = 0 and all derivatives of v tend to zero locally
uniformly when (x, N ) 0.
It is implicit in this denition that the value of a solution u of P u = 0
at a point only depends on the values of u and its derivatives in a compact
subset of the initial plane.
Applying this condition to exponential solutions ei(x,) with P () = 0
and suitable , an equivalent algebraic condition was found, namely that
Pm (N ) = 0 and that P ( + tN ) = 0 for all real when Im t is large enough
4
The real dual is generated by gradP () when P () = 0 and has m sheets. Its intersection
] sheets.
with t 0 has [ m+1
2
5
When n is even, (x) is just the real intersection.
6
In his work, Petrovsky analysed the homology in middle dimension of a general algebraic
hypersurface.
` 3
SEMINAIRES
ET CONGRES
43
negative.7 It follows easily that Pm belongs to the class Hyp(N ) of homogenous elements in hyp(N ), that Pm ()/Pm (N ) is real for real argument and
that the real, homogeneous hypersurface Pm = 0 consists of of m sheets
meeting the lines = tN + const in m points. When these points are always
separate unless all zero, i.e., the real surface Pm () = 0 is non-singular outside
the origin, P is said to be strongly (strictly) hyperbolic. In this case, Pm + R
belongs to hyp(N ) for any polynomial R of degree < m. In the general case,
Pm + R is hyperbolic if and only if R( + iN )/Pm ( + iN ) is bounded for all
real [Svensson 1969].
The hyperbolicity cone (N ), dened as the connected component of
Pm () = 0 that contains N , is open and convex and has the property that
P hyp() for all .
Every P hyp(N ) has a fundamental solution, the distribution
(6.2)
E(P, N, x) = (2)
ei(x,+i)
d
Rn P ( + i)
cN , c > 0, su. large.
The Fourier-Laplace integral on the right does not depend on the choice of .
As a function of x it is supported in a propagation cone C(P, N ), dual to and
consisting of all x such that (x, ) 0. This cone is proper, closed and convex
and has only the origin in common with all hyperplanes (x, ) = 0, .
The existence of such a fundamental solution is equivalent to the condition
(6.1). Note that a square matrix M (D) of partial dierential operators whose
determinant P (D) belongs to hyp(N ) is itself hyperbolic. In fact, there is
a matrix M (D) such that M (D)M (D) = P (D)I with I a unit matrix and
then M (D) has a fundamental solution M (D)E(P, N, x) with support in the
propagation cone of P .8
7.
The theory of lacunas
Lerays Princeton lectures [1953] and the paper by Atiyah et al. [1970] were
both written in an eort to understand [Petrovsky 1945]. The second one
extends his results to arbitrary P Hyp(N ) which are complete, i. e., not
expressible in fewer than n variables. For this, it is important to consider also
the local hyperbolicity cones (P , N ) (P, N ) where P () Hyp(N ) is
7
It is not dicult to see that hyp(N ) = hyp(N ).

If the class C in (3.1) is replaced by a smaller Gevrey class, the class Hyp(N ) is the
same, but the class hyp(N ) may permit more lower terms. Actually there is quite a number
of papers dealing with hyperbolicity in Gevrey classes, but they will be disregarded here.
8
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
44
the rst non-vanishing homogeneous term in the Taylor expansion of Pm ( +

). Note that P (N ) is all of Rn when Pm () = 0 and a half-space when
Pm () = 0, gradPm () = 0. The wave front surface W (P, N ) is now dened
to be the union of the local propagation cones C(P , N ) , dual to the local
hyperbolicity cones (P , N ). Modulo constant factors, the resulting formulas
for derivatives x of E of order || are
(x, )q P ()1 ()
x E(P, N, x)
when q = m n || 0 and

(7.1)
x E(P, N, x)
when q < 0. Here

=
tx
(x, )q P ()1 ()

j ...dn
(1)j1 d1 ...d
so that the integrands are rational (n 1)-forms of homogeneity zero on

Z = Cn and hence also closed forms of maximal degree on the n 1dimensional projective space Z . They are holomorphic in Z P and
Z P X respectively where P , X are the complex, projective surfaces
P () = 0 and X : (x, ) = 0 respectively. The forms are integrated over
certain homology classes and tx . Their description is based on the
existence of a continuous map iv() where
v() (P , N ) ReX,
= 0.
The class Hn1 (Z P , X ) is twice the projective image of this map

oriented by (x, )() > 0. The class Hn2 (X X P ) is an
absolute class and tx denotes a tube around it.9
Connected components c of C(P, N ) W (P, N ) where the fundamental
solution E(P, N, x) is a polynomial, necessarily homogeneous of degree m n,
are called Petrovsky lacunas. The formula (7.1) shows c is a Petrovsky lacuna
if the Petrovsky condition = 0 holds for some x c. The main point of
Atiyah et al. [1973] was to prove the converse of this statement by proving
the completeness of the rational cohomology used.10
When possible, residues in the last integral down into X P give integrals over the
original Petrovsky cycles.
10
It has been shown that W (P, N ) may be bigger than the singular support of E(P, N, x)
in C(P, N ) when P is not strongly hyperbolic, but the answer is no for at most double
characteristics [H
ormander 1977].
9
` 3
SEMINAIRES
ET CONGRES
8.
45
Cauchys problem for strongly hyperbolic

operators with variable coecients
In his lectures, Leray [1953] solved Cauchys problem for smooth scalar dierential operators and systems which are strongly hyperbolic in the sense that
the corresponding characteristic polynomials are strongly hyperbolic with respect to some direction. A surface is said to be space-like when the operator
is hyperbolic with respect to its normals.
Assuming uniform hyperbolicity of P (x, D) = D1m + ... with respect to x1
in some band a x1 b, Leray devised a suitable global energy form for
constant coecients which he extended to variable coecients by G
ardings
inequality [1953]. This permitted him to construct solutions of Cauchys
problem with initial data on planes x1 = const. by approximations from the
analytic case. Lerays paper also marks the rst appearance of distributions
in the theory of hyperbolic equations, to be used ever after.
In G
arding [1956, 1958], the energy tensor of Friedrichs and Lewy was
extended to scalar, strongly hyperbolic operators with variable coecients in
the following way, opened up by Leray [1953].
When ||
= m 1, || = m, the product u(x) u(x) with real u is a
divergence k Ck (u, u) where every Ck is a quadratic form in the derivatives
of u of order m 1. It follows that if P (x, D) and Q(x, D) are dierential
operators of degrees m and m 1, then
(8.1)
ImQ(x, D)uP (x, D)u =
k Ck (x, u, u) + C0 (x, u, u)
where all Ck are hermitian forms in the derivatives of u of order at most m1,
C0 containing only derivatives of order m 1.
When Pm (x, D) = D1m + lower terms has constant coecients and is
strongly hyperbolic with respect to x1 , and Q(x, D) = Pm (x, D)/D1 , a
Fourier transform in the variables x = (x2 , ..., xn ) shows that

(8.2)
C1 (u, u)dx c
|D u(x)|2 dx ,
c > 0,
||=m1
when the right side converges. If P (x, D) of order m is uniformly strongly

hyperbolic in a band B : 0 x1 a with time function x1 , if the coecients
are bounded and if the highest coecients satisfy a uniform Lipschitz condition, this formula with an additional term of lower order extends to P (x, D)
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
46
[G
arding 1953]. The result is an inequality for t > 0,11

(8.3)
D
m1
u(t, .) C
P u(x1 , .) dx1
+ Cect Dm1 u(0, .)

for some C, c > 0 where

D k u(t, .) 2 =
(8.4)
|D u(t, x )|2 dx .
||k
The inequality (8.3) also has a local version for lens-shaped subsets of B
bounded from below by space-like surfaces. It follows in particular that solutions of P (x, D)u = 0 which vanish at order m 1 on a space-like surface,
vanish identically.
When the left side of (8.3) is nite, the vector T k u = u(t, .), ..., Dtk u(t, .)
belongs to a certain Hilbert space H k . Let C(H k ), L1 (H k ), L (H k ) denote
functions of t such that, as a function of t, T k u(t, .) is continuous, integrable
and essentially bounded respectively with values in H k .
Associated to (8.3) is the following Cauchys problem
(8.5)
Pu = v
when 0 < t < a,
T m1 u(0, .) = T m1 w(0, .).
Here w C(H m1 ) and v L1 (H 0 ). This problem has a unique solution in

arding [1956, 1958] improved on earlier ones by
C(H m1 ). The proof by G
using only functional analysis and the inequality (8.3).
The existence of a solution can also be expressed as an inequality
(8.6)
u ,0 c sup
v
|(u, P v)|
,
v ,m1
c > 0.
Here all functions are dened on a band 0 t a, u L (H 0 ) with

the corresponding norm and v, equipped with the norm of L (H m1 ), runs
through the space C0 of all smooth compactly supported functions vanishing
close to t = 0. The inequality says in particular that P C0 is dense in L1 (H 0 ).
The analogous inequality
u ,0 c sup
v
|(u, Av)|
,
v ,0
c > 0,
11
It is proved in [Ivrii and Petkov 1974] that this inequality implies that P (x, D) is strongly
hyperbolic when its coecients are suciently dierentiable.
` 3
SEMINAIRES
ET CONGRES
47
where A = D1 + A2 D2 + ... + An Dn + C is strongly hyperbolic as in (4.2),

is a consequence of its scalar counterpart (8.6). In fact, the left side is not
increased if we replace v by Bv where
B = D1 + B2 D2 ... + Bn Dn
has the property that B(x, )A (x, ) = I detA (x, ) where I is the m m
unit matrix and A is the principal part of A. By hypothesis, detA(x, )
is uniformly strongly hyperbolic and hence A(x, D)B(x, D) I detA(x, D)
modulo bounded terms of order < m. Since Bv ,0 Dm1 v ,0 , the
result follows.
Under smoothness assumptions about the coecients, the inequality (8.3)
was extended by H
ormander [1963] to the case when the norm square (8.4) is
replaced by

k,s
2
(8.7)
D u(t, .) =
|D (1 + |D |)s u(t, x )|2 dx
||k
where s is any real number and the right side is dened by the Fourier transform in the varaible x . In this way, also functions with distributional values in
the x direction are taken into account. This inequality permitted Hormander
[1963] to solve the corresponding Cauchys problem very simply by a duality
argument. In particular, when the coecients of P are smooth enough, the
operator P has a fundamental solution E(x, y): P (x, D)E(x, y) = (x y)
which vanishes when x1 < y1 .
Cauchys problem on a manifold. The inequality (8.3) for lens-shaped
regions proves the basic uniqueness theorem for strongly hyperbolic operators
P on a manifold: if P u = 0 in some neighborhood of x0 and the Cauchy data
of u vanish on some smooth space-like surface S : s(x) = s(x0 ), then u = 0
close to x0 .
To deal with more global situations it is convenient to require the existence
of smooth, real time functions t(x) such that P (x, ) hyp(grad t(x)) for all
x.12 The condition grad s(x) (Pm (x, .), tx ) with a xed sign for smooth,
real s(x) denes two opposite classes T of time functions. A region where
some time function is in T+ is positive or negative is called a future and a
past respectively and a surface where some time function is constant is said
to be space-like. The manifold X is said to be complete relative to P if
every compact set is contained in an intersection of a past and a future with
12
By assuming the existence of time functions, Christodoulou and Klainerman [1993] were
able to prove global existence for Einsteins equations with small data.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
48
compact closure.13 The intersection of all futures (pasts) containing a point

x then denes two propagation conoids C (x) issuing from x. Leray [1953]
found suitable Sobolev spaces for the construction of inverses P1 of a strongly
hyperbolic operator P on a manifold, complete with respect to P such that
P1 u vanishes outside the union of the corresponding propagation conoids
issuing from supp u, supposed to be compact.
Nonlinear equations, hyperbolic conservation laws. The control of
lower order derivatives in Cauchys problem for linear, strongly hyperbolic
equations, makes it possible to use successive approximations to prove local
existence for Cauchys problem and quasilinear, and even nonlinear, strongly
hyperbolic equations. The proofs are almost as simple as in the second order
case, but involve a judicious use of Sobolev inequalities. The initial work by
Petrovsky [1937] and Leray [1953] was carried further by Dionne [1962].
Global existence is a problem beset with diculties. Discontinuites may
appear and solutions may cease to exist. This is clear from the much studied
case of nonlinear hyperbolic conservation laws in two variables t, x
ut + f (u)x = 0,
u, f (u) Rn ,
where f is smooth and nonlinear and the matrix f (u)/u has real, separate
eigenvalues. Burgers equation for n = 1, ut + uux = 0 is a model case
exhibiting collisions and rarefaction waves depending on initial data for t =
0. The use of weak solutions [Lax 1957b] motivates jump conditions, the
classical Rankine-Hugoniot jump conditions, and existence proofs have to use
various entropy conditions. The case of arbitrary n has a rened existence
proof for initial data of small bounded variation [Glimm 1965] with a recent
amelioration by Young [1993]. When the initial total variation is not small
and n > 2 blow-up may occur (see Young [1995]). A short text cannot do
justice to the complicated nature and history of hyperbolic conservation laws.
There is ample material in [Smoller 1983].
9.
Mixed boundary problems
Let P (D) be a dierential operator, hyperbolic with respect to the rst

variable x1 , and consider boundary problems for P in a quarter space
x1 0, x2 0 with a source function, Cauchy data C on x1 = 0 and some
other linear data F on a non-characteristic plane x2 = 0. If the problem is
correctly posed, the reduced problem with vanishing source and non-vanishing
13
The full Cauchys problem with data on a space-like surface requires this condition.
` 3
SEMINAIRES
ET CONGRES
49
Cauchy data should also be correctly posed. Hence the data F in the reduced
problem ought to propagate away in the positive x1 and x2 directions [Agmon
1962, Hersh 1963]. In particular, if n = 2 and
m

(D1 + ak D2 ),
P (D) =
ak = 0,
these solutions should be a linear combination of functions of x2 ak x1 with

ak < 0. For n = 2, this principle determines the form of mixed problems
for hyperbolic operators in regions limited by polygons (see [Campbell and
Robinson 1955] and [Thomee 1957]).
In the general case, the principle says that the reduced mixed boundary
problem should not have exponential solutions ei(x,) with P () = 0 which
are exponentially large for x1 > 0, x2 > 0 when the solution is bounded when
x1 = 0, x2 0 and x1 0, x2 = 0. This means that Im 1 > 0, Im 2 > 0.
This criterion is workable since it follows from the hyperbolicity that the
polynomial
2 P (), Im1 > 0, 3 , ... real,
has no real zeros and hence a xed number m+ of zeros with Im 2 > 0. The
remaining, forbidden ones have Im 2 < 0. It is therefore reasonable that
F can only have m+ independent data. Appropriate polynomial boundary
conditions on x2 = 0 have the form
Q1 (D)u = g1 , ...., Qm+ (D)u = gm+
where Q1 , ..., Qm+ should be linearly independent modulo the product of the
permitted factors14 of the polynomial 2 P (). There is a corresponding
determinant, the Lopatinski determinant, which should be hyperbolic in a
certain sense with respect to the rst variable. As shown by Reiko Sakamoto
[1974] and exposed in [Hormander 1983b, pp. 162-179], these conditions are
both necessary and sucient for the mixed problem for strongly hyperbolic
operators to be correctly posed in the C sense. The waves from the Cauchy
data at the boundary x2 = 0 are reected in a way consistent with the boundary condition.
In a wellknown paper by H.-O. Kreiss [1970], the problem above was put
for rst order operators, strongly hyperbolic with respect to the rst variable,
D1 + A2 D2 + ... + An Dn ,
whose coecients are m m matrices. The matrix A2 is supposed to be
diagonal with m+ positive and m m+ negative eigenvalues which gives m+
14
with zeros such that Im 2 > 0 when Im1 > 0.

E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
50
linear boundary operators. It is shown that a strengthening of the condition

above to no solutions with Im1 0 gives L2 bounds of the solution in terms
of similar bounds for the data.
10.
Hyperbolicity for variable coecients
It is proved in [Ivrii and Petkov 1974] that an inequality (8.3) implies that
P (x, D) is strongly hyperbolic when its coecients are suciently dierentiable. The same paper also oers necessary conditions for the hyperbolicity
for operators with variable coecients as dened by an obvious localization
of (6.1) to a neighbourhood N of a point x0 and its intersection I with a
plane (x x0 , ) = 0. It is required that u tends to zero close to x0 when all
the derivatives tend to zero locally uniformly in I and P u tends to zero in
the same way in N . The verication of this property involves existence and
uniqueness of a suitable Cauchys problem.
By the construction of suitable asymptotic solutions it is shown that
P (x0 , D) must be hyperbolic with respect to . The proofs have been simplied by H
ormander [1985a, pp. 400-403]. Earlier proofs by the same method
due to Lax [1957a] for analytic coecients and Mizohata [1961-62] for rst
order systems supposed that is not characteristic.
In the Cauchys problem for the operator D12 x21 D22 + bD2 , studied by
Oleinik [1970], the regularity of the solution requires more and more regularity of the Cauchy data the smaller b is. This is the motivation in [Ivrii and
Petkov 1974] to dene regular hyperbolicity (eective hyperbolicity according to H
ormander [1977]) as hyperbolicity under addition of arbitrary lower
order terms in the operator. The authors then prove the following interesting
result. For an operator P (x, D) to be eectively hyperbolic in an open set it
is necessary that the fundamental matrix (Hamiltonian map)

p
px
, p = Pm (x, ),
(10.1)
pxx px
skewsymmetric in symplectic structure given by dx d, has a pair of nonvanishing real eigenvalues at every point where dp = 0 but d2 p = 0. When this
condition is not satised, there are conditions on the lower terms, exhibited
in [H
ormander 1977]. Finally, it is proved that the condition that
dPm (x, ) = 0
is both necessary and sucient for hyperbolicity with a xed relation between
the regularity of the data and that of the solution independent of lower order
` 3
SEMINAIRES
ET CONGRES
51
terms. This condition implies strong hyperbolicity in an open set and at most
double zeros of Pm (x, ) on a bounding space-like surface. Tricomis operator
D12 x1 (D22 + ... + Dn2 ) in the region x1 0 is here a classical example (see
[H
ormander 1985b, section 23.4]).
In contrast to this situation, the suciency of eective hyperbolicity for
hyperbolicity is a delicate problem. A positive answer is known only for
equations of order two [Iwasaki 1984, Nishitani 1984a,b]) and under a certain
restriction in the general case Ivrii [1978], removed by Melrose [1983]. The
fact that the condition (10.1) is invariant under canonical maps is used by
all these authors to get suitable normal forms of the operators which then
must involve pseudodierential operators. The canonical maps are realized
by Fourier integral operators, a tool created by Hormander [1971] (see below).
Outside of eective hyperbolicity, there are microlocal conditions at multiple characteristics which make the Cauchys problem correctly set in the
sense given above (see [Kajitani and Wakabayashi 1994] and the literature
quoted there).
Systems. Necessary conditions for hyperbolicity with respect to the time
variable x1 for rst order hyperbolic operators
L(x, D) + B(x),
L(x, D) = ED1 + L2 (x)D2 + ... + Ln (x)Dn .
is a much studied subject. The coecients are smooth m m matrices and

E is the unit matrix. It is of course necessary that the determinant h(x, ) =
detL(x, ) be hyperbolic at every x, but this is not enough. A zero of order
r of h(x, ) must give a zero of order r 2 of the cofactor matrix M (x, ) =
(mij (x, )) and if L is eectively hyperbolic in the sense above, then every
h(x, D) + mij (x, D) must be hyperbolic with respect to x1 [Nishitani 1993].
11.
Fundamental solutions by asymptotic series
It was clear from the formulas of Herglotz and Petrovsky that the singularities of the fundamental solutions of homogeneous, strongly hyperbolic operators P (D) Hyp(N ) of degree m lie on the wave front surface, consisting
of [(m + 1)/2] sheets issued from the origin and contained in the dual to
the characteristic surface P () = 0.15 But the abstract existence proofs for
variable coecients did not give this kind of information, nor is it expected
15
The dual of P () = 0 is generated by x = gradP () when P () = 0. It has m sheets

and is invariant under reection in the origin. The wave front surface is the restriction to
(x, N ) 0 and has the number of sheets stated.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
52
unless the coecients are smooth. But for the case of innitely dierentiable
coecients, there are very precise results.
The construction of fundamental solutions of strongly hyperbolic operators by means of oscillating integrals [Lax 1957a, Ludwig 1960] gave the
rst answer.16 The oscillatory integrals used have the following general form
introduced by H
ormander [1971],

(11.1)
u(x) =
a(x, )ei(x,) d.
The amplitude a(x, ) is a smooth function with x in some open subset of Rn

and RN . It is assumed that a(x, ) = O(||m|| ) for large , locally
uniformly in x. The phase function (x, ) is supposed to be smooth and real
and homogeneous of degree 1 in . The assumption that d = 0 makes u a
distribution which is a smooth function of x unless (x, ) = 0 for some . In
practice, the amplitude a(x, ) is often polyhomogeneous, i.e., an asymptotic
sum of terms of decreasing integral homogeneity in for large values of this
variable.
When P (x, D) of order m is strongly hyperbolic with respect to x1 , its
principal symbol p(x, ) is a product of m factors pk (x, ) of homogeneity
1 in . The phase functions used in Laxs paper are
solutions k (x, ) of
xk k , k > 1 when
the equations pk (x, gradk ) = 0 such that k = =
x1 = 0. These functions exist only for small x1 , but permit an extension of
an oscillating integral wk (x2 , ..., xn ) with a polyhomogeneous amplitude and
phase function (and hence singular only at the origin) to an oscillating
integral Wk (x) with polyhomogeneous amplitude and phase function k such
that P (x, D)wk is arbitrarily smooth. By a suitable choice of w1 , ..., wm , the
dierence between a fundamental solution E(x) supported in x1 0 and the
sum W1 + ... + Wm can be made arbitrarily smooth. It follows that E(x) is
regular at x except when the -gradient of some k (x, ) vanishes, in particular
only at the origin when x1 = 0. Since dk invariant under the characteristics
dx/dt = p (x, x ) of the equation p(x, x ) = 0, the fundamental solution is
singular only at the locus of these curves issued from the origin.
For larger times, the locus of characteristics may develop singularities, the
caustics of geometrical optics may occur. Oscillating integrals which represent
the fundamental solution beyond the caustics were constructed in Ludwigs
paper.
16
Both authors treat hyperbolic systems.
` 3
SEMINAIRES
ET CONGRES
12.
53
Microlocal analysis, wave front sets,

pseudodierential operators
All the results above are claried by microlocal analysis which deals with localization in space and frequency of distributions and operators. A beginning
was made by Maslov [1964]. There is also a microlocal analysis for hyperfunctions initiated by Sato [1969] and later developed by his students and others.
However, we shall stick to distributions, following Hormander [1971].17
The setting of microlocal analysis is the cotangent bundle T (X) of a
dierentiable manifold X with local coordinates x, and invariant dierential
form = (dx, ). Let u be a distribution on Rn and let f C0 . Simple
arguments show that the growth at innity of the Fourier transform v() of
f u gets smaller in all directions when f is replaced by a product f g and
g C0 . Hence there is for instance a natural localization H s (x, ) of the
classical space H s at a point x, ( = 0) invariant under multiplication by
smooth functions and consisting of distributions u such that (1 + ||)s v()
belongs to L2 in some conical neighborhood of x, for some f C0 whose
support contains x. Another interesting object is the wave front set WF(u)
of a distribution u, equal to the complement of all x, such that v() has
fast decrease in some conical neighborhood of x, for some f as above. The
wave front set is a closed, conical subset of the cotangent bundle T (X). The
projection of WF(u) on X is the singular support S(u) of u. All these notions
extend to distributions on a manifold.
An important example of wave front set is the following. The wave front
set of the oscillatory integral (11.1) is contained in the set of pairs x, such
that (x, ) = 0. When the phase function is regular, i.e., the dierentials
d are linearly independent, this equation denes a conical Lagrangian manifold, a submanifold of T (Rn ) of maximal dimension were the dierential
form (, dx) vanishes. One important result of Hormander [1971] is that two
oscillatory integrals with regular phase functions with the same Lagrangian
produce the same distributions modulo smooth functions, at least when the
conical support of the amplitudes are small.
When the phase function of (11.1) has the form (x, y, ), x Rn , y
m
R , the integral I(x, y) represents the kernel of what is called a Fourier integral operator [H
ormander 1971]. Generally speaking, the corresponding
operator will map distributions u to distributions v such that
WF(v) C(WF(u))
17
Only the simplest version of microlocal analysis can be given here. For full exposition,
see H
ormanders monumental four volumes [H
ormander 1983a,b, 1985a,b].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
54
where C = (x, , y, ) is a canonical relation such that (x, , y, ) belongs

to the Lagrangian dened by I. This fact makes Fourier integral operators
a powerful tool in microlocal analysis which permits a change of variables in
the cotangent bundle which mixes its two ingredients.
When the phase function above has the form (x y, ) where the three
dimensions are the same, C reduces to the identity and the corresponding
operators are pseudodierential operators in the form developed by Kohn
and Nirenberg [1965]. They were originally given as singular integrals by
Calder
on and Zygmund [1957].
A pseudodierential operator has the form

n
P (x, )
u()d, u
() = ei(x,) u(x)dx.
P (x, D)u(x) = (2)
Here the left side is a denition, u C0 and the symbol P (x, ) of P is a
smooth amplitude with properties as in (11.1), for instance polyhomogeneous.
When P (x, ) is a polynomial in the second variable, P (x, D) is a dierential
operator. The rst non-zero term in the expansion of P is the principal symbol
p(x, ) of P . Pseudodierential operators act on Schwartzs space S and, by
duality also on S . In each case they form an algebra, the map from P to its
principal symbol being a homomorphism. The calculus of pseudodierential
operators extends to distributions on a manifold X. Its symbols are then
dened on the cotangent bundle T (X) with local coordinates (x, ).
One has WF(P u) WF(u) with equality when P (x, D) is elliptic, i.e.,
when CharP = , and then also WF(u) = when P u C . A proper
reduction of singularity may occur at the characteristic set Char(P ) where
p(x, ) = 0 and = 0. One of the uses of pseudodierential operators is the
factorization of the principal parts of hyperbolic operators into a product of
pseudodierential operators of degree 1.
Pseudodierential operators give a short, equivalent denition of WF(u)
for a distribution on a manifold X, namely

CharP.
P uC
This is also the original denition in Hormander [1970].
13.
Propagation of singularities in boundary

problems
A pseudodierential operator P (x, D) is said to be of real principal type

when its principal symbol p(x, ) is real and p(x, ) = 0 in CharP . The
` 3
SEMINAIRES
ET CONGRES
55
operator P has the characteristic equation p(x, x ) = 0 which in turn has the
characteristic curves
(13.1)
xt = p (x, ), t = px (x, ),
p(x, ) = 0,
called (null) bicharacteristics of P . By geometrical optics theory they leave

both CharP and the restriction to CharP of the dierential form invariant.
A basic general result proved by Hormander [1970] gives to the wave front
sets of solutions of P u = 0 a geometrical optics structure when P is a pseudodierential operator of principal type. It says that WF(u)\WF(P u) is
invariant under the bicharacteristic ow (13.1) so that, in other words,
(13.2)
WF(u)\WF(P u) is a union of bicharacteristics.
To prove this result it suces to show that a bicharacteristic interval I outside

WF(P u) is outside WF(u) when its endpoints are. When P has order 1 and
its symbol vanishes outside a neighborhood of I, the proof is not dicult
and the general situation can be reduced to this case. In another version
([Duistermaat and H
ormander 1972], [Hormander 1985b, p. 57]) the condition
that p(x, ) = 0 on Char p is eliminated, there is a radical reduction to the
case P = D1 .
If P is a dierential operator which is strongly hyperbolic with respect to
some , it follows from the general results above that the wave front set W
outside y of the fundamental solution E(P, x, y, ) with pole at y and support
in the halfspace (x y, ) 0 consists of all bicharacteristics issued from y
and directed into this space. The ber of the wave front set over y is Rn \0.
In fact this is the ber over y of (x y) and P E(x, y, ) = (x y). Caustics
appear when the projection of W on x-space is not invertible.
In the Cauchys problem for a hyperbolic operator in a half-space, the
source and the data on the boundary may be distributions and the question
of the singularities of the solution arises. The gross answer is that its wave
front set outside that of the source is generated by null bicharacteristics issuing
into the halfspace from the wave fronts of the source and the data. The precise
answer involves a calculus of pseudodierential operators on a manifold with
boundary introduced by Melrose [1981] and exposed by Hormander [1985a,
pp. 112-141].
The question of singularities of the solution of a mixed problem involve
reections at a time-like boundary. The propagation of singularities in this
case involves some serious microlocal analysis and is the subject of papers by
Chazarain [1973], Melrose [1975], Taylor [1976], Andersson and Melrose [1977],
Eskin [1977], Melrose and Sjostrand [1978, 1982], Ivrii [1980] and others.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
56
To take an example, let u be the solution of a second order equation

P u = f in the interior of a manifold X with boundary X where u = u0 and
consider the wave front set W of u outside the union of the wave front sets of
f and u0 . The simplest case is reection of a bicharacteristic by the law of geometrical optics. In addition, the boundary may contain a glancing set where
the incoming bicharacteristic is tangent to the boundary. The bicharacteristic
may then still be diracted o the boundary, but there may also be gliding
rays which are limits of rays which are reected many times. In all cases,
these bicharacteristics are part of the wave front set. A somewhat nal result
[Melrose and Sj
ostrand 1978, 1982] says roughly that a bicharacteristic in the
wave front set outside that of the source can always be continued except at
points over the boundary where the Hamilton eld is radial.
Propagation of polarization. The notion of characteristic set Char(P )
of a scalar dierential operator extends to a matrix operator P (x, D) of type
M N with principal symbol p(x, ). It is dened as the set of triples (x, , w
C N ) such that p(x, )w = 0, = 0.
The polarization set Wpol (u) of a distribution u(x) with k components is
then dened as the intersection of all Char(P ) with P of type 1 k, for which
P u C . Polarization of electromagnetic waves ts this denition. The
projection of the polarization set is the wave front set WF(u) dened as the
union of the wave front sets of the components of u.
In simple cases, for instance for strongly hyperbolic systems, polarization propagates along certain Hamilton orbits which are unique liftings of
bicharacteristics. The propagation of polarization, not restricted to hyperbolic equations, has been studied in a series of papers by N. Dencker [1982,
1995].
14.
Propagation of singularities for nonlinear

equations
If u H s (x, ) H t (x, ), s > n/2, the properties of the Fourier transform of

u2 show that it may happen that u2 H s+tn/2 (x, ) when is a convex linear
combination of , . It is therefore natural that new and weaker singularities
appear in solutions of equations when nonlinearities are introduced. These
new singularities will then propagate along bicharacteristics which in turn may
meet to give still weaker singularities and so on. According to the number of
steps, this process will be called selfspreading of rst order, second order, etc.
Selfspreading is made explicit in the paper by Rauch and Reed [1982]. It
deals with solutions u = (u1 , ..., um ) of strongly hyperbolic rst order semi
` 3
SEMINAIRES
ET CONGRES
57
linear systems in two variables t, x with right hand sides which are smooth
functions of t, x, u. The initial value u(0, x) is supposed to be of class H s
in an interval I and smooth outside. In the linear case, the singularities lie
on 2m forward characteristics from the endpoints of I which form a net with
crossings. In the semilinear case, new forward characteristics occur from the
lowest crossing points and so on. The end result is an explicit rule giving the
regularity of u in regions bounded by bicharacteristics. Roughly speaking,
the regularity increases with the distance to the origin. For more than two
variables this process of selfspreading of singularities may result in uniformly
distributed singularity. Beals [1983] constructed solutions u(t, .) H s , with
0 < t < 1, s > (n + 1)/2, of the wave equation in 1 + n > 2 variables with a
suitable nonlinear term f (x)u3 and initial data in H s , H s1 , singular only at
the origin. The singular support of one such solution was shown to contain
the part of the forward light cone where t 1 and the solution is regular
there at least of the order 3s n + 1 + 0.
The method of paradierential calculus by Bony [1981] (see also the review
article [Bony 1989]) has given some very general results about the propagation
of singularities of nonlinear strongly hyperbolic equations. The calculus is
based on smooth functions () supported
annulus Ak : 1/k || k
in some
j ) equals 1 () where is
(2
with k > 1 such that the dyadic sum
0
smooth and supported in || < 1. The action of paramultiplication Tu on a
distribution v is dened by the formula

(2j D)(u(2j D)v)
with () as above and (x) smooth and equal to 1 in Ak with support in
a slightly larger annulus. The crucial property of paramultiplication is that
if u H s , v H t , s + t > 0 then uv = Tu v + Tv u + R(u, v) where R maps
H s H t continuously to H s+tn/2 and similar properties for substitution.
Bony proved that a nonlinear dierential operator of order m, F (u) =
F (x, u, Du, ..., Dm u), has a paralinearization L given by

Lu =
TF/ u u
2s2mn/2
when u H s , F (u)
s > m+n/2.
such that Lu belongs to Hloc
= 0 and
The operator L and the ordinary linearization LF = (F/ u) have the

same principal parts.
The preceding result can be applied to the situation when LF is strongly
hyperbolic with some time variable t and F (u) = 0 in some region, u H s .
Outside CharL the solution is locally in H 2smn/2 and regularity one step or
more lower is propagated along bicharacteristics. There are also analogous results about the reection and diraction of bicharacteristics (see [Bony 1989]).
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
58
When the equation F (u) = 0 is semilinear and the interaction between

singularities are taken into account, more precise results have been obtained
by Chemin [1988] using a rened paradierential calculus. Briey, his results
s and s exceeds some s depending on the equation, then
say that if u Hloc
0
singularity of the order at most 3s+O(1) is propagated by a a certain modied
second order selfspreading. Chemin also shows that this result is close to best
possible.
More precise propagation of singularities, closer to the linear case, can
be obtained with special initial data describing simple waves. These are
the conormal distributions. The singular support of such a distribution is
a smooth hypersurface S : s(x) = 0 and the regularity of u does not decrease
under the action of smooth vector elds tangent to S. Example: u = f (s(x))
where f is a homogeneous distribution on the line. For wave equations with
nonlinear lower terms, initial data of this form are propagated close to the
linear case and the rst order selfspreading suces for a precise description.
Problems of this kind, the propagation caused by intersecting hypersurfaces,
by a hypersurface developing a swallowtail, by reexion of a simple wave in
a wall, etc. are the subject of many papers of progressing complexity which
are still being published by, among others, Melrose and collaborators (see e.g.
[Bony 1989, Lebeau 1989, Melrose and Ritter 1985, 1987, Melrose and Barreto
1994, Barreto 1995]). These and many similar papers bear out Bonys remark
that nonlinear singularities require more microlocal analysis, not less, than
linear ones.
15.
Blow-up and global existence for wave

equations
The subject of global solutions of semilinear wave equations got new life in
the eighties. The impetus came from Fritz Johns papers about life-time and
blow-up of semilinear wave equations with small initial data [John Papers,
part. IV]. The best studied equations have the form

(15.1)
ajk (u ) j k u f (u ) = 0,
utt
u = gradu,
j = /xj ,
or
(15.2)
u = g(u, u , u ),
= t2
in 1 + n variables t, x with compactly supported initial data u(0, x) = u0 (x),

ut (0, x) = u1 (x) where > 0 is small. It is also assumed that the co
` 3
SEMINAIRES
ET CONGRES
59
ecients are smooth and that the equations deviate little from u = 0 so that
ajk (u ) jk vanishes of order zero and f (u, u , u ) of order 1 for vanishing
arguments. The lifetime T of the solution is the maximal time below which the
solution is reasonably smooth. The work done with these equations is ample
conrmation of Schauders remark that the solution of nonlinear equations
means getting optimal bounds on solutions of linear equations. In particular,
the improvement below of the blow-up time with increasing n depends on the
increasing dispersion of initial data for the linear equation.
John worked with both equations above but mostly with the case n = 3
where has a fundamental solution 0. One of his many results [John 1979]
says that T 2 for the equation u = u2 . Improving on [John 1976],
John and Klainerman [1984] proved for equations (15.2) that T > ec/ when
n = 3. For n > 4 this was improved by Klainerman [1985a] to existence for
all suciently small > 0 when g does not depend on u. The case n = 3
requires an extra condition on the main term, the null condition, found by
Klainerman [1986] and Christodoulou [1986]. For n = 4 and g = g(u , u ),
H
ormander proved [1991] that T ec/ . The same estimate with 2 was
obtained later by Li Ta-Tsien and Zhou Yi [1995]. Corresponding results
for nonlinear pertubations of the Klein-Gordon equation utt u + u =
F (u, u ), where the linear equation has a better energy density, are less delicate
[Klainerman 1985b].
Recent interest has been focussed on the details of the blow-up. Caarelli
and Friedman [1986] found a space-like smooth blow-up surface for the wave
equation with right side F (u) Aup , A > 0, p > 1. Lindblad [1990a] proved
that the rescaled solution U (t, x) = 4 u(t/2 , x/2 ) of (15.2) in 1+3 variables
and f = u2 has a distribution limit v which solves (15.2) with the right side
v 2 + in some interval 0 < t < T . Here is a measure carried by the
forward lightcone. More precise life times T for two space variables are given
in [Alinhac 1994, 1995] where it also conjectured under certain regularity
assumptions that the quotient 1/(T t) describes the growth of the L2
norm of gradu at the point t close to T . Alinhac suggests that singularities
may appear as folds after a suitable change of variables and proposes better
approximations of the quasilinear equation than just the linear part. Such
methods were also used in [Hormander 1989].
16.
Concluding remarks
The development of the theory of hyperbolic partial dierential equations

in the twentieth century is a continuing eort to master the singularities of
solutions of such equations. In this process new analysis was used and as old
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
60
problems were solved, new ones have appeared.

The diculty that the fundamental solution of a second order hyperbolic
operator has singularities of high order outside the pole was circumvented
by Hadamard in his use of the partie nie. The full use of the Fourier
transform has permitted the construction of fundamental solutions of homogeneous higher order hyperbolic partial dierential operators with constant
coecients, in the beginning with an incomplete treatment of the singularities. The algebraic denition of hyperbolicity has been motivated intrinsically
by the requirements of nite propagation velocity and continuity. New energy densities have made possible existence proofs for Cauchys problem and
mixed problems for linear and nonlinear hyperbolic dierential equations using a passage to the limit from the analytic case. Afterwards, the theory of
distributions gave a better understanding of the nature of singularities and
functional analysis has given simple existence proofs for Cauchys problem
and mixed problems both for smooth and not smooth data. For nonlinear
equations, the control of lower order derivatives make local existence proofs
possible. As shown by the theory of hyperbolic conservation laws, global
existence and uniqueness are much more dicult problems. For quasilinear
equations, the problem of the lifetime of solutions with small initial data has
recently received much attention.
Microlocal analysis is a new tool for the study of the propagation of singularities of solutions of hyperbolic partial dierential equations. For linear
equations and a variety of boundary problems, this study has given almost
denitive results, at least for smooth coecients. Recent eorts are directed
towards the analysis of singularities of solutions of nonlinear hyperbolic equations. Here the nonlinearity itself generates singularities which have been
successfully treated for equations close to linear ones.
Bibliography
Agmon (S.)
[1962]
Problèmes mixtes pour les equations hyperboliques dordre superieur. In

Coll. Int CNRS, vol. 117, Paris , 1962, pp. 1318.
Alinhac (S.)
[1994]
Temps de vie et comportement explosif des solutions dequations dondes

quasilineaires en dimension deux II, Duke Math. J., 73 (1994), pp. 543
560.
[1995]
Temps de vie et comportement explosif des solutions dequations dondes

en dimension deux I, Ann. Sci. Ecole Norm. Sup., (IV) 28 (1995), pp. 225
250.
` 3
SEMINAIRES
ET CONGRES
61
Andersson (K.) and Melrose (R.)

[1977]
The propagation of singularities along gliding rays, Invent. Math., 41

(1977), pp. 197232.
Atiyah (M.), Bott (R.) and G

arding (L.)
[1970]
Lacunas for hyperbolic dierential equations with constant coecients I,

Acta Math., 124 (1970), pp. 109189.
[1973]
Lacunas for hyperbolic dierential equations with constant coecients II,

Acta Math., 131 (1973), pp. 145206.
Barreto (A.)
[1995]
Evolution of semilinear waves with swallowtail singularities. Preprint.
Beals (M.)
[1983]
Self-spreading and strength of singularities, Ann. of Math., 118 (1983),

pp. 187214.
Bony (J.M.)
[1981]
Calcul symbolique et propagation des singularites pour les equations partielles non lineaires, Ann. Sci. Ec. Norm. Sup., (IV) 14 (1981), pp. 209
246.
[1989]
Analyse microlocale des

equations aux derivees partielles non lin
eaires,
vol. 1495 of Lecture Notes in Mathematics, Springer, 1989.
Caffarelli (L.) and Friedman (A.)

[1986]
The blow-up boundary for nonlinear wave equations, Trans. Amer. Math.
Soc., 297 (1986), pp. 223241.
n (A.) and Zygmund (A.)

Caldero
[1957]
Singular integrals and dierential equations, Amer. J. Math., 79 (1957),

pp. 901921.
Campbell (L.) and Robinson (A.)

[1955]
Mixed problems for hyperbolic partial diferential equations, Proc. London Math. Soc., (III) 5 (1955), pp. 129147.
Chazarain (J.)
[1973]
Construction de la parametrix du problème mixte hyperbolique pour

lequation des ondes, C. R. Acad. Sci. Paris, 276 (1973), pp. 12131215.
Chemin (J.Y.)
[1988]
Interaction controlee dans les e.d.p. non lineaires strictement hyperboliques, Bull. Soc. Math. France, 116 (1988), pp. 346383.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
62
Christodoulou (D.)
[1986]
Global solutions of nonlinear hyperbolic equations for small initial data,

Comm. Pure Appl. Math., 39 (1986), pp. 276282.
Christodoulou (D.) and Klainerman (S.)

[1993]
The globasl stability of Minskowski space. Princeton Univ. Press, 1993.
Dencker (N.)
[1982]
The propagation of polarization for systems of real principal type, J.

Funct. Anal., 46 (1982), pp. 351372.
[1995]
The propagation of polarization for systems of transversal type, Ark.

Mat., 33 (1995), pp. 249279.
Dionne (P.A.)
[1962]
Sur les problèmes de Cauchy hyperboliques bien poses, J. Anal. Math.,

10 (1962), pp. 190.
Duistermaat (J.) and H

ormander (L.)
[1972]
Fourier integral operators II, Acta Math., 128 (1972), pp. 183269.
Eskin (G.)
[1977]
Parametrix and propagation of singularities for the interior mixed hyperbolic problem, J. Anal. Math., 40 (1977), pp. 1762.
Friedrichs (K.) and Lewy (H.)
[1928]
Uber
die Eindeutigkeit und das Abhangigkeitsgebiet der Losungen
beim Anfangswertproblem linearer hyperbolischer Dierentialgleichungen, Math. Ann., 98 (1928), pp. 192204.
G
arding (L.)
[1950]
Linear hyperbolic partial dierential operators with constant coecients,

Acta Math., 85 (1950), pp. 162.
[1953]
Dirichlets problem for linear elliptic partial dierential operators, Math.

Scand., 1 (1953), pp. 5572.
[1956]
Solution directe du problème de Cauchy pour les equations hyperboliques,

Coll. Int. CNRS Nancy, 1956, pp. 7189.
[1958]
Cauchys problem for hyperbolic equations, corr. ed., Univ. of Chicago,

1958.
[1989]
History of the mathematics of double refraction, Arch. Hist. Exact Sci.,

40 (1989), pp. 355385.
Glimm (J.)
[1965]
Solutions in the large for nonlinear hyperbolic systems of equations,

` 3
SEMINAIRES
ET CONGRES
63
Hadamard (J.)
[1932]
Le problème de Cauchy et les equations hyperboliques lin

eaires, Paris:
Hermann, 1932.
Herglotz (G.)
[1926-28] Uber
die Integration linearer partieller Dierentialgleichungen mit konstanten Koezienten I-III, Ber. der S
achsischen Akad. der Wiss., 78
(1926), pp. 93126 and 287318; Ibid., 80 (1928), pp. 69114.
Hersh (R.)
[1963]
Mixed problems in several variables, J. Math. Mech., 12 (1963), pp. 317

334.
[1964]
Boundary conditions for equations of evolution, Arch. Rational Mech.

Anal., 16 (1964), pp. 242264.
H
ormander (L.)
[1963]
Linear Partial Dierential Operators, vol. 116 of Grundlehren, BerlinHeidelberg: Springer, 1963.
[1965]
Pseudodierential operators, Comm. Pure Appl. Math., 18 (1965), pp. 89

152.
[1970]
Linear dierential operators, in Actes Congr. Int. Math., Nice 1970, vol. 1,
pp. 121133.
[1971]
Fourier integral operators I, Acta Math., 127 (1971), pp. 79183.
[1977]
The Cauchy problem for dierential equations with double characteristics,

J. Anal. Math., 32 (1977), pp. 118196.
[1983a]
The Analysis of Linear Partial Dierential Operators I. Springer, 1983.
[1983b]
The Analysis of Linear Partial Dierential Operators II. Springer, 1983.
[1985a]
The Analysis of Linear Partial Dierential Operators III. Springer, 1984.
[1985b]
The Analysis of Linear Partial Dierential Operators IV. Springer, 1985.
[1988]
Non-linear hyperbolic dierential equations, Lectures 1986-87, Dept.

Math. Lund Univ., 1988.
[1989]
The fully non-linear Cauchy problem with small data, Bol. Soc. Brasil
Mat., 20 (1989), pp. 127.
[1991]
The fully non-linear Cauchy problem with small data II, in Microlocal Analysis and Nonlinear Waves (edited by M. Beals, R. Melrose and
J. Rauch), vol. 30 of IMA Volumes in Mathematics and its Applications,
Berlin: Springer-Verlag, pp. 5181.
[1993]
Hyperbolic systems with double characteristics, Comm. Pure Appl. Math.,

46 (1993), pp. 261301.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
64
Iagolnitzer (D.)
[1975]
Microlocal essential support of distributions and decomposition theorems

- An introduction, in hyperfunctions and theoretical physics, vol. 449 of
Lecture Notes in Mathematics, Springer, 1975, pp. 121132.
Ivrii (V.)
[1978]
Sucient conditions for regular and completely regular hyperbolicity,

Trans. Moscow Math. Soc., 1 (1978), pp. 165.
[1980]
Wave fronts for solutions of boundary problems for a class of symmetric

hyperbolic systems, Siberian Math. J., 21 (1980), pp. 527534.
Ivrii (V.) and Petkov (M.)

[1974]
Necessary conditions for the correctness of the Cauchy problem, Uspekhi

Mat. Nauk, 29-5 (1974), pp. 370.
Iwasaki (N.)
[1984]
The Cauchy problem for eectively hyperbolic equations (A standard

type), Publ. RIMS, (A) 20 (1984), pp. 543584.
John (F.)
[Papers]
Collected papers, vol. 1. Boston-Basel: Birkhauser, 1985.
[1976]
Delayed singularity formation for solutions of nonlinear partial dierential

equations in higher dimensions, Comm. Pure Appl. Math., 29 (1976),
pp. 649682; Papers I, pp. 498531.
[1979]
Blow-up of solutions of nonlinear wave equations in three space dimensions, Manuscripta Math., 28 (1979), pp. 235268; Papers I, pp. 532565.
[1983]
Lower bounds for the life span of solutions of nonlinear wave equations in
three dimensions, Comm. Pure Appl. Math., 36 (1983), pp. 135; Papers
I, pp. 589623.
John (F.) and Klainerman (S.)

[1984]
Almost global existence to nonlinear wave equations in three space dimensions, Comm. Pure Appl. Math., 37 (1984), pp. 443455.
Kajitani (K.) and Wakabayashi (S.)

[1994]
Microlocal a` priori estimates and the Cauchy problem II, Jap. J. Math.,
20 (1994), pp. 171.
Klainerman (S.)
[1985a]
Uniform decay estimates and the Lorentz invariance of the classical wave
equation, Comm. Pure Appl. Math., 38 (1985), pp. 321332.
[1985b]
Existence of small amplitude solutions to nonlinear Klein-Gordon equations in four space-time dimensions, Comm. Pure Appl. Math., 38 (1985),
pp. 631641.
` 3
SEMINAIRES
ET CONGRES
[1986]
65
The null condition and global existence to nonlinear wave equations, in

Nonlinear systems of partial dierential equations in applied mathematics,
vol. 23-1 of Lectures in Applied Mathematics, AMS, 1986, pp. 293326.
Kohn (J.J.) and Nirenberg (L.)

[1965]
On the algebra of pseudodierential operators, Comm. Pure Appl. Math.,

18 (1965), pp. 269305.
Kreiss (H.O.)
[1970]
Initial boundary value problems for hyperbolic systems, Comm. Pure

Appl. Math., 23 (1970), pp. 277298.
Lax (P.D.)
[1954]
Weak solutions of nonlinear hyperbolic equations and their numerical

computation, Comm. Pure Appl. Math., 7 (1954), pp. 159193.
[1957a]
Asymptotic solutions of initial value problems, Duke Math. J., 24 (1957),

pp. 627246.
[1957b]
Hyperbolic conservation laws II, Comm. Pure Appl. Math., 10 (1957),

pp. 537566.
Lebeau (G.)
[1989]
Equations des ondes semilineaires II, Controle des singularites et caustiques non lineaires, Invent. Math., 95 (1989), pp. 277323.
Leray (J.)
[1933]
Etudes
de diverses equations integrales non-lineaires et de quelques problèmes que pose lHydrodynamique, J. Math. Pures Appl., (IX) 12 (1933),
pp. 182.
[1953]
Hyperbolic Dierential Equations, Princeton: Inst. Adv. Study, 1953.
Li Ta-Tsien and Zhou Yi

[1995]
A note on the life-span of classical solutions to nonlinear wave equations in

four space dimensions, Indiana Univ. Math. J., 44 (1995), pp. 12071248.
Lindblad (H.)
[1990a]
Blow-up for solutions of u = |u|p with small initial data, Comm. Partial
Dierential Equations, 15 (1990), pp. 757821.
[1990b]
On the lifespan of solutions of nonlinear wave equations with small initial

data, Comm. Pure Appl. Math., 43 (1990), pp. 445471.
Ludwig (D.)
[1960]
Exact and asymptotic solutions of Cauchys problem, Comm. Pure Appl.

Math., 13 (1960), pp. 473508.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
66
Maslov (V.P.)
[1964]
Theory of perturbations and asymptotic methods. Moscow, 1964. (in Russian).
Melrose (R.B.)
[1975]
Microlocal parametrix for diractive boundary problems, Duke Math. J.,

42 (1975), pp. 605635.
[1981]
Transformation of boundary problems, Acta Math., 147 (1981), pp. 149

236.
[1983]
The Cauchy problem for eectively hyperbolic operators, Hokkaido Math.

J., 12 (1983), pp. 371391.
Melrose (R.B.) and Barreto (A.)

[1994]
Non-linear interaction of a cusp and a plane. Preprint.
Melrose (R.B.) and Ritter (N.)

[1985]
Interactions of progressive waves for semilinear wave equations I, Ann. of

Math., 121 (1985), pp. 187213.
[1987]
Interactions of progressive waves for semilinear wave equations II, Ark.

Mat., 25 (1987), pp. 91114.
Melrose (R.B.) and Sj

ostrand (J.)
[1978]
Singularities of boundary value problems I, Comm. Pure Appl. Math., 31

(1978), pp. 593617.
[1982]
Singularities of boundary value problems II, Comm. Pure Appl. Math.,

35 (1982), pp. 129168.
Mizohata (S.)
[1961-62]
Some remarks on the Cauchy problem, J. Math. Kyoto Univ., 1 (1961-62),

pp. 109127.
Nishitani (T.)
[1984a]
A necessary and sucient condition for the hyperbolicity of second order equations with two independent variables, J. Math. Kyoto Univ., 24
(1984), pp. 91104.
[1984b]
Local energy integrals for eectively hyperbolic operators I, II, J. Math.

Kyoto Univ., 24 (1984), pp. 623658 and 659698.
[1993]
Necessary conditions for strong hyperbolicity of rst order systems, Journal dAnal. Math., 61 (1993), pp. 181228.
Oleinik (O.A.)
[1970]
On the Cauchy problem for weakly hyperbolic equations, Comm. Pure

Appl. Math., 23 (1970), pp. 569586.
` 3
SEMINAIRES
ET CONGRES
67
Petrovsky (I.G.)
[1937]
Uber
das Cauchysche Problem f
ur Systeme von partiellen Dierentialgleichungen, Mat. Sb., 2-44 (1937), pp. 815870.
[1938]
Uber
das Cauchysce Problem f
ur Systeme linearer Dierentialgleichungen
und nichtanalytische Funktionen, Bull. Moscow Univ. Series math. and
mech., 1-7 (1938), pp. 179.
[1945]
On the diusion of waves and lacunas for hyperbolic equations, Mat. Sb.,
17-59 (1945), pp. 289370.
Rauch (J.) and Reed (M.)

[1982]
Nonlinear microlocal analysis of semilinear hyperbolic systems in one

space dimension, Duke Math. J., 49 (1982), pp. 397475.
Riemann (B.)
[1860]
Uber
die Fortpanzug ebener Luftwellen von endlicher Schwingungsweite,
Abh. Kgl. Ges. Wiss. G
ottingen, 8 (1860); Gesammelte Mathematische
Werke, Springer/Teubner, 1990, pp. 157175.
Riesz (M.)
[1949]
Lintegrale de Riemann-Liouville et le problème de Cauchy, Acta Math.,

81 (1949), pp. 1223.
Sakamoto (R.)
[1974]
E-wellposedness for hyperbolic mixed problems with constant coecients,

J. Math. Kyoto Univ., 14 (1974), pp. 93118.
Sato (M.)
[1969]
Hyperfunctions and partial dierential operators. In Proc. Int. Conf. on

Functional Analysis and Related Topics, Tokyo 1969, pp. 9194.
[1970]
Regularity of hyperfunction solutions of partial dierential equations. In

Actes Congr. Int. Math., Nice 1970, vol. 2, pp. 785794.
Schauder (J.)
[1935]
Das Anfangwertproblem einer hyperbolischen Dierentialgleichung

zweiter Ordnung in beliebiger Anzahl von unabhangigen Veranderlichen,
Fund. Math., 24 (1935), pp. 213246.
[1936a]
Quasilineare Dierentialgleichungen zweiter Ordnung vom hyperbolischen

Typus. Gemischte Randwertaufgabe, Studia Math., 6 (1936), pp. 162189.
(with Krzyzanski).
[1936b]
Gemischte Randwertaufgaben bei partiellen Dierentialgleichungen vom

hyperbolischen Typus, Studia Math., 6 (1936), pp. 190198.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

L. GARDING
68
Smoller (J.A.)
[1983]
Shock waves and reaction-diusion equations, vol. 258 of Gundlehren,

Springer, 1983.
Svensson (L.)
[1969]
Necessary and sucient conditions for hyperbolicity of polynomials with

a hyperbolic principal part, Ark. Mat., 8 (1969), pp. 145162.
Taylor (M.)
[1976]
Grazing rays and reection of singularities of solution to wave equations,

Thom
ee (V.)
[1957]
Estimates of the Friedrichs-Lewy type for mixed problems, Math. Scand.,

5 (1957), pp. 93113.
Young (R.)
[1993]
Sup-norm stability of Glimms scheme, Comm. Pure Appl. Math., 46

(1993), pp. 903948.
[1995]
Exact solutions to degenerate conservation laws. Preprint.
` 3
SEMINAIRES
ET CONGRES
From General Relativity to Group

Representations
The Background to Weyls Papers of 192526
Thomas Hawkins
Abstract
Hermann Weyls papers on the representation of semisimple Lie
groups (1925-26) stand out as landmarks of twentieth century
mathematics. The following essay focuses on how Weyl came to
write these papers. It oers a reconstruction of his intellectual
journey from intense involvement with the mathematics of general
relativity to that of the representation of groups. In particular
it calls attention to a 1924 paper by Weyl on tensor symmetries
that played a pivotal role in redirecting his research interests. The
picture that emerges illustrates how Weyls broad philosophically
inclined interests inspired and informed his creative work in pure
mathematics.
R
esum
e
Les articles de Hermann Weyl sur la representation des groupes
de Lie semi-simples (1925-26) apparaissent comme des etapes majeures des mathematiques du vingtième siècle. En analysant ce
qui a amene Weyl à ecrire ces articles, cet essai presente une reconstruction de sa demarche intellectuelle, depuis les mathematiques de la relativite generale jusquà celles des representations
de groupes. Il attire notamment lattention sur larticle de 1924
sur les symetries tensorielles, pivot de la reorientation de ses domaines de recherche. On voit aussi comment les larges interets et
les motivations philosophiques de Weyl ont inspire et enrichi sa
creativite en mathematiques pures.
AMS 1991 Mathematics Subject Classication: 01A60, 17B10, 22E46

Boston University, Department of Mathematics, Boston, MA 0225, USA.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
70
T. HAWKINS
Dieudonne once wrote that progress in mathematics results, most of the

time, through the imaginative fusion of two or more apparently dierent topics [Dieudonne 1975, p. 537]. One of the most brilliant examples of progress
by fusion is provided by Herman Weyls celebrated papers on the representation of semisimple Lie groups (1925-1926). For in them he fashioned a theory
which embraced I. Schurs recent work (1924) on the invariants and representations of the n-dimensional rotation group, which was conceived within the
conceptual framework of Frobenius theory of group characters and representations, and E. Cartans earlier work (18941913) on semisimple Lie algebras,
which was done within the framework of Lies theory of groups and had been
unknown to Schur. Moreover, in fashioning his theory of semisimple groups,
Weyl drew on a host of ideas from such historically disparate areas as Frobenius theory of nite group characters, Lies theory, tensor algebra, invariant
theory, complex function theory (Riemann surfaces), topology and Hilberts
theory of integral equations. Weyls papers were thus a paradigm of fusion,
and they exerted a considerable inuence on subsequent developments. They
stand out as one of the landmarks of twentieth century mathematics.
It is not my purpose here to describe the rich contents of these remarkable
papers nor to analyze their inuence. This has been done by Chevalley and
Weil [1957], by Dieudonne [1976], and, above all, by Borel [1986]. I wish to
focus instead on how Weyl came to write these remarkable papers. In this
connection Weyl wrote:
for myself I can say that the wish to understand what really is the
mathematical substance behind the formal apparatus of relativity
theory led me to the study of representations and invariants of
groups ...[Weyl 1949, p. 400].
My goal is to attempt to explain what Weyl meant by this remark, that
is, to reconstruct the historical picture of his intellectual journey from his
involvement with the mathematics of general relativity to that of the representation of semisimple Lie groups. In particular, I want to call attention
to a paper by Weyl [1924a], which in my opinion adds a fullness and clarity
to the picture that would otherwise be lacking. The picture that emerges
illustrates how Weyls broad philosophically inclined interests in this instance in theoretical physics inspired and informed his creative work in
pure mathematics.1
1
For another such instance, see [Scholz 1995] where Weyls interest in Fichtes philosophy
is related to his approach to the geometry of manifolds.
` 3
SEMINAIRES
ET CONGRES
FROM GENERAL RELATIVITY TO GROUP REPRESENTATIONS
71
The Space Problem

Weyls involvement with general relativity began in 1916, when, at age 31, he
returned from military service to his position at the Eidgenossische Technische
Hochschule (ETH) in Z
urich. My mathematical mind was as blank as any
veterans, he later recalled, and I did not know what to do. I began to study
algebraic surfaces; but before I had gotten far, Einsteins memoir came into
my hand and set me are.2 By the summer of 1917 Weyl was lecturing on
general relativity at the ETH. These lectures formed the starting point for his
classic book, Raum, Zeit, Materie, which went through four editions during
191823,3 and spawned many collateral publications by Weyl aimed at further
developing the ideas and implications of his lectures. One of the outcomes of
Weyls reections on general relativity was his introduction of what he called
a purely innitesimal geometry.4
Weyl became convinced that Riemannian geometry, including
the quasi
Riemannian geometry of an indenite metric ds2 =
ij gij dxi dxj ,
gij = gij (x1 , . . . , xn ), on which Einsteins theory was based, was not a consistently innitesimal geometry. That is, in Riemannian geometry, a vector
v = (dx1 , . . . , dxn ) in the tangent plane at point P of the manifold could only
be compared with a vector w = (dy1 , . . . , dyn ) in the tangent plane at point
Q in the relative sense of a path-dependent parallel transport from P to Q,
but the lengths of v and w were absolutely comparable in the sense that

|v|
i,j gij (P )dxi dxj
=
.
|w|
i,j gij (Q)dyi dyj
These considerations led Weyl to a generalization of Riemannian geometries in
which the lengths of v and w are not absolutely comparable. As in Riemannian
geometry a nondegenerate quadratic dierential form ds2 of constant signature is postulated but metric relations are determined locally only
up to a pos2
itive calibration (or gauge) factor and so are given by ds = ij gij dxi dxj .
Here varies from point to point in such a way that the comparison of the
lengths of v at P and w at Q is also in general a path-dependent process.5
2
Quoted by S. Sigurdsson [1991, p. 62] from Weyls unpublished Lecture at the Bicentennial Conference (in Princeton).
3
There were actually ve editions, but the second (1919) was simply a reprint of the rst
[Scholz 1994, p. 205n].
4
See Scholz [1994, 1995] for a detailed account of the historical context and evolution of
Weyls ideas on this theory during 191723 .
5
For a complete denition of Weyls geometry see [Scholz 1994, p. 213] and for a contemporary formulation see [Folland 1970]. Weyls geometry represented the rst of a succession
of gauge theories that has continued into present-day physics [Vizgin 1989, p. 310].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
72
T. HAWKINS
Although Weyls geometry was motivated by the above critique of Riemannian geometry, he discovered that he could use its framework to develop
a unied eld theory, that is, a theory embracing both the gravitational and
the electromagnetic eld. Hilbert had been the rst to devise a unied theory within the framework of general relativity in 1915. Weyls theory was
presented in several papers during 191819 and in the third edition (1919)
of Raum, Zeit, Materie. Einstein admired Weyls theory for its mathematical brilliance, but he rejected it as physically impossible. Although Weyl
respected Einsteins profound physical intuition and was accordingly disappointed by the negative reaction to his unied theory, Einsteins arguments
did not convince him that his own approach was wrong. His belief in the correctness of his theory was bolstered by the outcome of his reconsideration, in
publications during 192123, of the space problem rst posed by Helmholtz
in 1866. It was in connection with this problem that Weyl rst began to
appreciate the value of group theory for investigating questions of interest to
him involving the mathematical foundations of physical theories.
In 1866 Helmholtz sought to deduce the geometrical properties of space
from facts about the existence and motion of rigid bodies. He concluded
that the distance between points (x, y, z) and (x + dx, y + dy, z + dz) is

dx2 + dy 2 + dz 2 and that space is indeed Euclidean. He returned to the
matter in 1868, however, after learning from the work of Riemann and Beltrami about geometries of constant curvature. Using the properties of rigid
bodies he had singled out earlier, he argued that Riemanns hypothesis that
metric relations are given locally by a quadratic dierential form is actually
a mathematical consequence of these facts. Later, in 1887, Poincare obtained
Helmoltzs results for two-dimensional space by applying Lies theory of groups
and utilizing, in particular, the consideration of Lie algebras. Lie himself considered the problem in n dimensions by means of the consideration of Lie
groups and algebras in 1892. The Lie-Helmholtz treatment of the space problem, however, was rendered obsolete by the advent of general relativity since,
as Weyl put it:
Now we are ... dealing with a four-dimensional [continuum] with
a metric based not on a positive denite quadratic form but rather
one that is indenite. What is more, we no longer believe in the
metric homogeneity of this medium the very foundation of the
Helmholtzian metric since the metric eld is not something
xed but rather stands in causal dependency on matter [Weyl
1921a, p. 263].
Following the Helmholtz-Lie tradition, Weyl conceived of space (includ
` 3
SEMINAIRES
ET CONGRES
73
ing therewith the possibility of space-time) as an n-dimensional dierentiable

manifold M with metric relations determined by the properties of congruences which are conceived in terms of groups. Thus at each point P M the
rotations at P are assumed to form a continuous group of linear transformations GP , and since the volume of parallelepipeds is assumed to be preserved
by rotations, the GP are taken as subgroups of SL (TP (M)). Metrical relations in a neighborhood U of P are then based on the assumption that all
rotations at P U can be obtained from a single linear congruence transformation A taking P to P by composition with the rotations at P ; that
is, every T GP is of the form T = AT A1 so that GP = AGP A1 .
By passing continuously from P to any point Q of the manifold M, Weyl
was led to the assumption that all the groups GP are congruent to a group
G SL(n) with Lie algebra g sl(n). Thus, whereas in the Lie-Helmholtz
treatment of the space problem, the homogeneity of space entails the identity
of the rotation groups at diverse points, in Weyls formulation the rotation
groups have diering orientations, although they share the same abstract
Lie algebra.
Within this mathematical context Weyl stipulated two postulates: (1) the
nature of space imposes no restriction on the metrical relationship; (2) the
ane connection is uniquely determined by the metrical relationship. His
interesting mathematical interpretation of these two postulates led to the
conclusion that the Lie algebra g must possess the following properties:
a) For all X g, tr X = 0 (i.e., g sl(n, R));
b) dim g = 12 n(n 1);
(k)
c) For any X1 , . . . , Xn g with matrix form Xk = (aij ) with regard to some

basis, if Col i of Xj = Col j of Xi for all i, j = 1, ..., n, then Xi = 0 for all
i = 1, ..., n.
In the fourth edition of Raum, Zeit, Materie, where Weyl rst presented
his analysis of the space problem [Weyl 1921a, 18], he pointed out that the Lie
algebras gQ of all orthogonal groups with respect to a nonsingular quadratic
form Q satisfy (a)(c) and he conjectured as highly probable the following
theorem which he had conrmed for n = 2, 3:
Theorem 1. The only Lie algebras satisfying (a)(c) are the orthogonal
Lie algebras gQ corresponding to a nondegenerate quadratic form Q.
Weyls conjectured theorem thus implied the locally Pythagorean nature of
space. Weyl pointed out that when g does correspond to an orthogonal Lie
algebra, the quadratic form Q is only determined up to a constant of proportionality [Weyl 1921a, p. 146]. Although he did not say it explicitly at
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
74
T. HAWKINS
this point, the truth of Theorem 1 would thus imply that his generalization
of quasi-Riemannian geometry, his purely innitesimal geometry, was also
compatible with the conclusions of his analysis of the space problem.
Within a few months of completing the fourth edition of Raum, Zeit,
Materie, Weyl had obtained a proof of Theorem 1, which he submitted for
publication in April 1921 [Weyl 1921b] and announced in a general talk in
September 1921 [Weyl 1922]. With the proof of Theorem 1 his analysis of the
space problem was complete. Weyl saw it as conrmation of the legitimacy
of his geometrical approach to relativity theory his purely innitesimal
geometry with its concomitant unied eld theory. He was also mindful of
the fact that it had been achieved by utilizing the theory of groups: The
establishment by group theory is hence a new support for my conviction that
this geometry, as geometry of the world, is the basis for the interpretation of
physical eld phenomena, rather than, as with Einstein, the more restrictive
Riemannian [geometry] [Weyl 1922, p. 344]. Indeed, Weyl was so taken
up with Theorem 1 that he even likened the conrmation by logic of the
correctness of his approach to the space problem aorded by Theorem 1 to
the factual conrmation of the correctness of Einsteins relativistic approach
to gravitation aorded by the observed advance of the perihelion of mercury
[Weyl 1921b, p. 269].
During the spring of 1922 Weyl lectured on the space problem in Spain,
and a version of his lectures was then published as a monograph [1923a],
which he regarded as a supplement to Raum, Zeit, Materie since the deeper
conception of the space problem using group theory was only sketched there.
In the eighth lecture, which sketches a proof of Theorem 1, Weyl wrote:
While almost all deeper mathematical theories such as, e.g.,
the wonderful theory of algebraic number elds have little to
signify within the great philosophical continuum of knowledge and
while, on the other hand, what mathematics can contribute to
enlighten the general problem of knowledge mostly stems from
the surface of mathematics, here we have the rare case that a
problem which is fundamental to all knowledge of reality, as is
the space problem, gives rise to deeply penetrating mathematical
questions.[Weyl 1923a, p. 61]
Within the context of the space problem Weyl had discovered group theory
as a powerful tool for dealing with fundamental questions inspired by general relativity and leading to deeply penetrating mathematical questions.
Although he described it as a rare occurrence, as we shall see, this was not
` 3
SEMINAIRES
ET CONGRES
75
the last time that his involvement with the fundamentals of general relativity led to important mathematical questions of a group-theoretic nature and,
ultimately, to his papers on the representation of Lie groups.
Before proceeding to consider these further occurrences, however, there
is one additional, historically important, consequence of Weyls involvement
with the space problem that needs to be mentioned. In 1922 the fourth edition
of Raum, Zeit, Materie was translated into French and read by Elie Cartan,
who, since 1921, had become interested in Einsteins theory. Unaware that
Weyl had already proved the conjectured Theorem 1, Cartan provided a proof
of his own [Cartan 1922]. Strictly speaking, Cartan did not prove Theorem 1.
Instead, he reformulated Weyls somewhat vaguely articulated version of the
space problem in terms of his own approach to geometry based on moving
frames and dierential forms. Cartans approach evolved into the modern
theory of G-structures.6 Within that framework, however, Cartans formulation of the space problem ultimately reduced to the problem of determining
all linear Lie algebras g satisfying Weyls conditions (a) and (b) and, in lieu of
the rather mysterious condition (c), the condition that g leaves no proper subspaces invariant.7 By a theorem Cartan had proved in [Cartan 1909, p. 912]
it followed that g must be semisimple. Since Cartan had already determined
all such linear Lie algebras which leave no vector spaces invariant [Cartan
1913, 1914], it was, as he noted, just a matter of checking which of these Lie
algebras satisfy the dimension condition (b), to arrive at Weyls conclusion
that g must be an orthogonal Lie algebra.8
Expressed in modern terms, what Cartan had done in [Cartan 1913] was to
determine all irreducible representations of a complex semisimple Lie algebra,
and in [Cartan 1914] he did the same for real Lie algebras. However Cartan
did not conceive of his work within the conceptual framework of group representations. He conceived of his work as solving the problem of determining all
groups of projective transformations which leave nothing planar invariant
a problem of importance from the Kleinian view of geometry as the study
6
See in this connection [Scheibe 1988, p. 66] and [Scholz 1994, p. 225]. Scheibe argues
that if what Weyl had in mind is made more precise in accord with what his writings
seem to suggest, then it is not equivalent to Cartans formulation, but the theorem Cartan
proved implies the theorem Scheibe reconstructs from Weyls vague statements [Scheibe
1988, pp. 6869].
7
This property of the satisfying (a)(c) of Weyls Theorem 1 actually follows readily
from propositions Weyl deduced from (a)(c) [Weyl 1923a,c], although he did not expressly
take note of this fact.
8
Both Cartan and Weyl realized that it suces to consider the problem for complex Lie
algebras. In his announcement Cartan indicated that a detailed solution of the problem in
a generalized form was contained in [Cartan 1923].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
76
T. HAWKINS
and classication of groups acting on manifolds. Historically the conceptual

framework of group representations and characters came from Frobenius theory as developed for nite groups during 18961903, and it was Weyl who rst
brought Cartans work within that framework in his papers of 192526.
Weyl learned of Cartans work when the latter sent him his announcement
[Cartan 1922] of a solution to the space problem. In Weyls reply, dated
October 5, 1922, after pointing out that he had already given a proof of
Theorem 1, he wrote:
Untraveled on the paved roads of the general theory of continuous
groups, which have been laid out and constructed thanks to your
masterly skill, I have on my own beat a steep inconvenient footpath
through much underbrush to my goal. I have no doubt that your
method corresponds better to the nature of the matter; still, I
see that you also cannot arrive at the goal without distinguishing
many cases.9
The general consensus seems to be that Weyl, impressed by Cartans papers on Lie algebras, studied them carefully and that this study, combined
with an interest in the theory of invariants piqued by some critical remarks
by the mathematician E. Study (discussed below) led, through the inspiration
provided by a paper on invariants by I. Schur [1924] (also discussed below), to
his celebrated papers of 192526 on the representation of Lie groups. While
there is much truth in such a portrayal of events, it does overlook Weyls deep
seated, philosophically inclined interest in the mathematical foundations of
theoretical physics; in particular, it fails to fully account for Weyls own statement that the wish to understand what really is the mathematical substance
behind the formal apparatus of relativity theory led me to the study of representations and invariants of groups. ... Weyls involvement with the space
problem was certainly an instance of his interest in the mathematical substance underlying relativity theory, and it led him to E. Cartans work. But
the space problem was not the only focal point of this interest. In what follows, I will attempt to give a clearer notion of how other manifestations of
his interest in nding the proper mathematical basis for the mathematical
apparatus of general relativity increased his involvement with the theory of
groups and, in particular, with the theory of their representations and how
this in turn made Cartans work all the more relevant.
9
I am grateful to B.L. van der Waerden, who called these letters to my attention many
years ago and sent me copies after obtaining consent of the holders H. Cartan in the case
of Weyls letters and the ETH in the case of E. Cartans letters.
` 3
SEMINAIRES
ET CONGRES
77
In this connection, it should be kept in mind that Weyls above-quoted reply to Cartan was written when he had only the original proof of his theorem,
which he disliked because it was complicated and lacked an overall unifying
idea [Weyl 1922, p. 344]. He compared it deprecatingly to tightrope dancing [Weyl 1921b, p. 269], and in his popular lectures on the space problem,
including those in Spain in the spring of 1922, he declined for this reason
to sketch the proof. By the spring of 1923, however, when he published a
German version of the lectures in Spain, he included a proof (as Appendix
12) because he had obtained what he felt were far reaching simplications
to his original proof [Weyl 1923a, p. v] so that, although still complicated in
detail due to the need to distinguish many cases, it was now guided by a single
idea, which in fact Weyl pushed further in [Weyl 1923c], where he wrote in
conclusion:
Our game on the chessboard of matrix schemes has been played
to its end. As complicated in details as it may be, it including
the rst part, which was already laid out in ...[Weyl 1923a] ...
Append. 12 rests ... upon a single constructional idea which
determined each step and was tenaciously carried out to the end.
It is interesting to observe that in Weyls presentation of his new proof, he
used another roadway analogy in comparing his and Cartans proofs, but
now with a dierent slant: By contrast with Cartans proof mine does not
take the detour of the investigation of abstract groups. It is based on the classical theory of the individual linear mapping going back to Weierstrass [Weyl
1923a, p. 88]. So now Cartans solution involves a detour because it draws
upon the theory of semisimple Lie algebras, whereas Weyls approach is more
direct and elementary, depending only on the classical theory of the Jordan canonical form of a matrix implicit in Weierstrass theory of elementary
divisors.
That is not to say that Weyl did not appreciate by this time early
1923 the impressive results and deep theory developed by Killing and
Cartan. Indeed, immediately prior to the above quotation, Weyl characterized
Cartans solution to the space problem by writing:
An entirely dierent proof has been given by Cartan ... based on
[his] ... earlier comprehensive and deep investigations ... on the
theory of continuous groups, in which he achieved a far reaching
solution to the problem of determining all abstract groups and
their realization through linear operations .... Now he only needed
to seek out among the groups determined by him those which
satisfy my stipulations.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
78
T. HAWKINS
These words indicate that Weyl certainly understood the gist of what Cartan
had done in his papers and appreciated the profundity of the mathematics.
But as far as the space problem was concerned, the extensive detour required
by Cartans approach was not deemed appropriate by Weyl, who was still
fascinated by his own approach. It is not clear he had found reason enough
to take on the nontrivial task of mastering the details of Cartans papers so
as to put them to his own use. Eventually he did and he was perhaps the
rst mathematician to do so but the motivation to do so seems to have
come not from the space problem but from tensor algebra.
Tensor Algebra and Symmetries

The formal apparatus of relativity theory consisted in large part of the calculus of tensors. This apparatus had evolved out of the work of mathematicians,
notably Christoel, Lipschitz and Ricci, interested in developing the theory
of the transformation of quadratic dierential forms suggested by Riemanns
speculations on the foundations of geometry.10 The principal source of the
resulting theory upon which Einstein and Grossman drew in developing the
mathematical side of general relativity starting in 1913 was the monographic
paper by Ricci and Levi-Civita, Methodes de calcul dierentiel absolu et
leurs applications [1900], which more or less summed up what had been
achieved during 18681900. To this they added the term tensor, the notion
of mixed tensors and (in 1916) Einsteins now-familiar summation convention, but essentially they took over the apparatus of the absolute dierential
calculus of Ricci and Levi-Civita.
In Raum, Zeit, Materie, Weyl also credited the Ricci-Levi-Civita paper
[1900] for the systematic development of tensor calculus,11 but it was he,
who, drawing upon his Gottingen background, recast tensor calculus in its essentially modern form. For one thing, Weyl treated tensor algebra tensor
analysis in a xed tangent plane independently as a preliminary to tensor
analysis, and in developing tensor algebra he did so within the geometrically
avored context of vector spaces, which had grown out of Hilberts work on
integral equations as developed by Erhard Schmidt. It is within the context of
tensor algebra as developed by Weyl in the pages of Raum, Zeit, Materie that
the formal apparatus of relativity theory gave rise to fundamental mathematical questions. As I will attempt to show, Weyls concern with these questions
was a major factor in the considerations that ultimately led to his papers of
10
The history of the tensor calculus from its origins in up to its application to general
relativity is traced in [Reich 1994].
11
See note 4 to p. 53 of the fourth edition [Weyl 1921a].
` 3
SEMINAIRES
ET CONGRES
79
192526. To make these questions intelligible I will rst sketch out the basics
of Weyls approach to tensors.
Let V denote an n-dimensional vector space over the real or complex eld
equipped with a nondegenerate quadratic form Q(v, w), v, w V dening a
scalar product.12 Then if e1 , . . . , en is a basis for V we may express v V in
the form
(1)
v=
n
xi ei .
i=1
The xi are called contravariant coordinates of v since if e1 , . . . , en is another

basis related to the rst by
(2)
ei =
n
mki ek ,
k=1

i ei
then if M denotes
the matrix with (i, k) entry mki , we have v = ni=1 x

n
i
i
k
where x = k=1 mk x
so that, expressed in matrix form (which Weyl did
not use)
1

x.
(3)
x
= MT
The vector v is also uniquely determined by the n values yi = Q(v, ei ), which
are called covariant coordinates of v with respect to the basis e1 , . . . , en since
they transform according to
(4)
y = M y,
and thus with the same coecient matrix as in the basis change (2). Nowadays
the yi would be regarded as coordinates of the element v in the dual space
V dened by v (w) = Q(v, w). That is, the yi are the coordinates of v with
respect to the basis e1 , . . . , en of V dual to e1 , . . . , en .
For Weyl tensors are uniquely determined by multilinear forms. For example, the mixed tensor of rank 3 denoted by Tijk by Einstein and covariant
in the indices i, j and contravariant in the index k is conceived by Weyl as
determined by a multilinear form f = f (u, v, w), where if xi and y j are the
contravariant coordinates of u and v respectively and zk the covariant coordinates of w, then

Tijk xi y j zk .
(5)
f=
i,j,k
12
Weyl does not speak of V as a vector space but rather as an n-dimensional ane space.
Also of course Q is not necessarily positive denite.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
80
T. HAWKINS
In view of the remark following

(4), f may be regarded as a multilinear form on

V V V with w = k yk ek from which (5) follows with Tijk = f (ei , ej , ek ).
Thus Weyl in eect identied the above type tensors with the vector space L
of all such multilinear forms f , which agrees with the present-day formulation
according to which
(6)
L
= V V V.
= (V V V )
In view of (6) the reader may wish to identify the tensor dened by (5) with
the element

(7)
Tijk ei ej ek V V V.
i,j,k
The representation of the tensor determined by f with respect to any basis

is then known by the rules of linear algebra. That is, suppose a basis change
is given by the matrix M dened by (2). Then the representation of f in the
barred variables is

i yj zk ,
Tijk x
(8)
f=
i,j,k
, y = M T y
where the Tijk are obtained from (3)(4) by substituting x = M T x
1
and z = M z in (5). The result is:
=
Tijk mi mj nk ,
(9)
T
i,j,k
where nk denotes the (k, ) entry of M 1 . Before Weyl such a rank three
tensor would have been dened as the totality of a system of functions
Tijk = Tijk (P ), P a point in the underlying manifold, which transform by the
rule laid down in (9), where M =M (P ) is the Jacobian matrix of a variable
change in the underlying manifold.13
The above presentation of the algebra of tensors was novel on Weyls part
but was a reworking of earlier notions. However, Weyl also introduced a
new notion that of a tensor density in his paper [Weyl 1918] and in
14 He was motivated by the consideration of an invariRaum, Zeit, Materie.

ant integral I = W (x)dx where x = (x1 , . . . , xn ). Given a variable change
13
See, e.g., [Einstein and Grossmann 1913]. The same approach is found in [Ricci and
Levi-Civita 1900, 2], although not applied to mixed tensors which were rst introduced by
Einstein and Grossmann [Reich 1994, p. 194].
14
In the fourth edition [Weyl 1921a, see 13]. Pauli [1921, p. 32, n.16] credits Weyl with
this notion and cites Weyls paper [1918] see 5 and the third edition of Raum, Zeit,
Materie; I am grateful to John Stachel for calling this to my attention.
` 3
SEMINAIRES
ET CONGRES
81
where
x = (y), a scalar function W = W (x) transforms from W to W
(y) = W ((x)). In the integral I, however, where W (x) can be regarded as
W
giving
the density of the manifold at x so that I represents its mass, we have
I = W ((y))|x/y|dy, where x/y denotes the Jacobian determinant of
x = (y). Hence the function W , as a scalar density function, transforms
where W
(y) = W ((x))|x/y|. For tensors Weyl inby the rule W W
troduced the analogous notion of a tensor density. Expressed in the tensor
algebra notation presented above, tensor densities are also identied with multilinear forms, such as the form f given in (5), but the rules of transformation
are dierent. To obtain the representation (9) of the tensor density dened
by f in the new coordinate system, instead of using (3) and (4), one uses
(10)
x
= | det(M T )1 |(M T )1 x
on the contravariant variables and

(11)
y = | det M |M y
on the covariant variables. By contrasting tensors and tensor-densities,

Weyl wrote in Raum, Zeit, Materie, it seems to me that we have rigorously
grasped the dierence between quantity and intensity, so far as the dierence
has a physical meaning ... [Weyl 1921a, p. 109]. Weyls notion of tensor
densities is still a part of general relativity today.15
The introduction of the concept of a tensor density seems to have
prompted the following mathematical question. Although it is very Weylian
in nature, it was rst posed by Weyls student at the ETH, Alexander Weinstein.16 Weinstein, who had proof-read the third edition (1919) of Raum,
Zeit, Materie, observed that all of the transformations (3)-(4) and (10)-(11)
underlying Weyls version of tensor algebra involve a matrix M which is
a function of the matrix M of the basis change (2), namely, if we assume
without any real loss of generality that det M > 0, M = (M T )1 in (3),
M = M in (4), M = det((M T )1 )(M T )1 in (10), and M = (det M )M
in (11). In all of these cases, he observed, the law of matrix composition is
preserved, i.e.,
(12)
(M1 M2 ) = M1 M2 .
15
See, e.g., [Misner et al. 1973, p. 501], [Mller 1972, p. 310].

I owe my awareness of Weinsteins paper to A. Borel [1986, p. 54]. Weinstein was one
of Weyls few students, and one he regarded highly. He went on to distinguish himself as
an analyst. See in this connection the biographical sketch by Diaz in Weinstein Selecta, see
[Weinstein 1923].
16
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
82
T. HAWKINS
Just as Weyl asked: what is the mathematical basis for the locally pythagorean
nature of space in relativity theory, so now Weinstein asked: what is the
mathematical basis for the transformation rules (3)-(4), (10)-(11) of tensor
algebra? That is: Are there other rules M M satisfying (12) and hence
other sorts of tensors? Weinstein proved the answer is no in the sense that
(3)-(4), (10)-(11) are the only elementary rules; all others are composed out
of these. Hence there are no essentially new types of tensors to consider. He
called his result the fundamental theorem of tensor calculus.
As with the space problem, so here too Weinsteins question involved a
group, this time the group of all matrices of positive determinant. At the
advice of Weyl, Weinstein proved his result by working on the Lie algebra
level. By virtue of (12), of course, today we would say that Weinstein was
studying degree n representations of this group, but Weinstein made no reference to such a theory. That is not surprising. Frobenius had developed a
representation theory for nite groups in 18961904, but nothing comparable
in scope had been done for continuous groups. Some things had been done
which, in retrospect, can be seen as contributions to such a theory, although
it is quite conceivable that neither Weyl nor Weinstein were aware of this
fact at the time Weinstein worked on his dissertation, which was submitted
for publication on February 22, 1922.17 In addition to the above-mentioned
work of E. Cartan, which, as we have seen, Weyl seems to have rst learned
about in October 1922, there was the doctoral dissertation of Frobenius student Issai Schur [1901] devoted to the study of the type of representation of
GL(n, C) that occurs in the theory of invariants. Schurs dissertation will
be discussed further on. It was probably not known to Weyl until 1924. In
any case, Weyl discovered a completely dierent, conceptually simpler way to
connect representations of GL(n, C) with those of the symmetric group than
that developed by Schur. As we shall see, this discovery was a by-product
of his own interest in the mathematical underpinnings of tensor algebra and
ultimately led him to his own fundamental theorem about tensors and to
the involvement with the representation of continuous groups that culminated
in his papers of 192526.
The aspect of tensor algebra that proved signicant in this connection
had to do with the symmetry properties of tensors. In relativistic physics and
in geometry the tensors that arose were not totally general; they came with
specic symmetry properties. Thus in the pages of Raum, Zeit, Materie [Weyl
1921a], the stress tensor Sik is seen to be a symmetric tensor of rank 2 (8),
and the four-dimensional relativistic electromagnetic intensity vector Fik of
17
The paper was published in Mathematische Zeitschrift [Weinstein 1923] and also separately as Weinsteins doctoral dissertation at the ETH.
` 3
SEMINAIRES
ET CONGRES
83
20 is a skew-symmetrical tensor of rank 2. The symmetry properties of the

Riemann curvature tensor Rijkl (17) are more complex, being given by
(i)
Rjikl = Rijlk = Rijkl ;

(iii)
(ii)
Rklij = Rijkl ;
Rijkl + Riklj + Riljk = 0.
With such examples in mind, Weyl wrote emphatically at the beginning of 7

on Symmetrical Properties of Tensors, that: the character of a quantity is
not in general described fully, if it is stated to be a tensor of such and such
an order [i.e., rank], but symmetrical characteristics have to be added [Weyl
1921a, p. 54].
Weyl realized that permutations could be used to characterize symmetry
properties in general. Consider, for example, a covariant tensor of rank 3,
Tijk , which following Weyl, we regard as a multilinear function

Tijk xi y j z k .
(13)
f = f (x, y, z) =
i,j,k
If S is some permutation of the x, y and z variables, let fS denote the

form that arises from f by permuting the variable series according to S.
Then f is symmetric if fS = f for all S and skew-symmetric if fS = (sgn S)f ,
where as usual the sign of S is 1 according to whether S is an even or odd
permutation. Weyl concluded his discussion of tensor symmetry by observing
that the most general form of a symmetry condition is expressible by one or
more equations of the form

eS fS = 0,
(14)
S
where the eS are numbers and S runs over all possible permutations of the
variables.
Weyls emphasis on the symmetry properties of tensors and the manner in
which he conceived of them, i.e., in terms of permutations and (14) naturally
suggest questions about the mathematical basis of tensor symmetry. Here are
some questions suggested by the above presentation, and eventually posed by
Weyl. Suppose C is a symmetry class of tensors determined by one or more
symmetry relations of the form (14). What are the possibilities for C? That
is, what is the mathematical basis for understanding the possibilities for C?
Also, is there an analog for C of the following properties P , P which hold,
respectively, for symmetric and skew symmetric tensors:
Property
P . If f is an arbitrary covariant tensor of rank , then the tensor
1
f = ( ! ) S fS is symmetric. Furthermore, all symmetrictensors of rank
1
) S fS .
are so expressible since if f is such a tensor then f = ( !
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
84
T. HAWKINS
Property
P . If f is an arbitrary covariant tensor of rank , then the tensor
1
f = ( ! ) S (sgn S)fS is skew symmetric. Furthermore all skew symmetric
tensors
of rank are so expressible since if f is such a tensor, then f =
1
) S (sgn S)fS .
( !
Although Weyl did not explicitly mention properties P , P in Raum, Zeit,
Materie, it is doubtful they escaped his notice. Indeed,
he used the fact that
1
) S (sgn S)fS to show
any skew symmetric tensor f is expressible as f = ( !
that (for = 3) every such f is expressible as a linear combination of the special skew symmetric volume tensors (dened by Weyl using determinants)
which have become the standard basis for the subspace of skew symmetric
tensors [Weyl 1921a, p. 55].
As we shall see, Weyl posed and answered the above questions in a paper
submitted in January 1924 [Weyl 1924a]. I suspect he may have had them in
mind much earlier, but his resolution of them or at least his publication
of these results may have been prompted by an episode involving the
mathematician Eduard Study (1862-1930) which occurred in 1923.
Response to Study
Study was an idiosyncratic, somewhat cantankerous mathematician whose
primary mathematical research interest was in the theory of invariants and
its geometrical applications. For a while in the late 1880s and early 1890s, he
became a part of Lies school, being charged by Lie with the task of relating
his theory of transformation groups to the theory of invariants. During this
period his work on ternary invariants even led him to conjecture, in a letter
to Lie, what amounts to the complete reducibility theorem for semisimple Lie
groups the theorem Weyl rst succeeded in proving in his papers of 1925-26.
But Study nally abandoned his eorts to deal with groups on the abstract
level of Lies theory and concentrated instead on more concrete problems,
including the study of the invariants of groups other than the general linear
group. In particular he studied the invariants of the orthogonal group in
[Study 1897].
At the beginning of 1923 Study published a book on the theory of invariants [Study 1923], and in the lengthy introduction he chastised those working
on relativity theory for their neglect of the tools of the theory of invariants
in favor of tensor calculus. He pointed out that for over fty years a highly
developed theory of invariants with respect to the general linear group had
been in existence and, citing his own work on orthogonal invariants, he noted
that an invariant theory of other groups had also been indicated. But with
` 3
SEMINAIRES
ET CONGRES
85
the majority of authors there is nothing to indicate that they live in an age
in which the theory of groups is in full bloom (p. 3). In short, Study
continued (p.4), they are behind the times, and not just a little. Even with
an otherwise knowledgeable writer one can read for example the following:
Many will be appalled at the deluge of formulas and indices with which the
leading ideas are inundated. It is certainly regrettable that we have to enter
into the purely formal aspect in such detail and to give it so much space but,
nevertheless, it cannot be avoided. That quotation, although not identied
as such, was drawn from Weyls book, Raum, Zeit, Materie.18
Study went on to criticize Weyl for accepting the formalism of the tensor
calculus as an unavoidable, necessary evil. That is not to say that Study
was against formalism. Quite the contrary! He believed the formal aspects
of mathematics were important, but the formalism must be of the right kind:
Mathematics is neither the art of calculation nor the art of avoiding calculations. To mathematics, however, belongs the art of avoiding superuous
calculations and carrying out the necessary ones adroitly. In this regard, one
can learn from the older authors (p. 4). What Study had principally in
mind was the symbolical method of the theory of invariants which went back
to Aronhold and Clebsch. This method reduced the problem of determining
invariants to the far easier problem of determining symbolical or vector invariants. In sum (to use Studys own analogy): mathematicians had thought
that in the tensor calculus they were borrowing from the garden of their
neighbor the physicist the seeds of the golden apples of the Hesperides but
were contenting themselves with a harvest of potatoes! The neglected theory
of invariants and in particular the symbolical method, Study implied, would
prove to be far more valuable.
It will be helpful for what is to follow to briey indicate the nature
of the theory of invariants in Studys time and the gist of the symbolical
method. Let G denote a group of nonsingular linear transformations of variables x = (x1 , . . . , xn ), y = (y1 , . . . , yn ), . . . In the classical theory G was
GL(n, C), but by Studys time other groups such as the orthogonal group
O(n, C) were also considered, thanks largely to Studys eorts. Invariants
are dened with respect to one or more base forms (Grundformen), which
are homogeneous polynomials of specic type in one or more variables series
x, y, . . . with unspecied coecients.
Consider, for example, as base form the

bilinear form f (a; x, y) = ni,j=1 aij xi yj . Then each T G induces a linear
18
I am grateful to Erhard Scholz for informing me that Study was using the rst edition
of 1918 or its 1919 reprint as second edition. The quotation is from p. 111. In subsequent
editions published before 1923 the passage was changed and is not as vulnerable to Studys
criticism. See p. 123 of the third edition and p. 137 of the fourth edition [Weyl 1921a].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
86
T. HAWKINS
transformation M (T ) of a = (a11 , ..., ann ) as follows. The variable change

x = T x , y = T y , transforms f (a; x, y) into f (a , x , y ) = f (a, T x , T y ) and
the relation between the coecients aij and aij is given by a nonsingular linear
transformation: a = M (T )a. An invariant of the form f (a; x, y) is any homogeneous polynomial I(a) = I(a11 , . . . , ann ) for which I(a ) = (det T ) I(a)
for all a = M (T )a, i.e., for all T G. Here, in the traditional presentation
T M (T ) is not quite a representation of G since M (T1 T2 ) = M (T2 )M (T1 ),
but this can be remedied by considering T M (T 1 ). In eect this is
the representation determined by the action of G on the vector space of
all bilinear forms. The symbolical method uses the polarization process of
Aronhold to transform each invariant I(a) into a symbolical or vector invariant i(, , . . .), i.e., a homogeneous polynomial in vectors = (1 , . . . , n ),
= (1 , . . . , n ),. . . , such that i(T , T , . . .) = (det T ) i(, , . . .).19 Since
the original invariant can be recaptured from the vector invariant, the problem of determining the invariants of G with respect to the form f (a; x, y)
reduces to the simpler problem of nding vector invariants. In [1897] Study
determined all vector invariants of the orthogonal group, thereby in principle
solving the problem of determining all the invariants of the orthogonal group
with respect to a set of base forms.
Weyl replied to Studys criticism in two papers. The rst reply was explicit
and was contained in a paper submitted at the end of October 1923 [Weyl
1923c]. This paper was intended as the rst of a series of papers in which Weyl
proposed to deal with mathematical topics of interest to all mathematicians
and to emphasize clarifying known results rather than presenting new ones.
One such topic Weyl dealt with was that of determining the invariants, in
the sense of the symbolical method of determining vector invariants, for the
classical groups, the symplectic group being treated here for the rst time.
Thus he tacitly accepted invariant theory and the symbolical method as a
part of basic mathematics, but in a footnote referring to Studys criticism,
he rejected Studys suggestion that invariant theory and, in particular, the
symbolical method belonged in a treatise on relativity theory. Even if he
possessed Studys great command of the theory of invariants, Weyl declared:
I would not apply the symbolical method in my book Space, Time, Matter
and not a single word would have been said about the completeness theorems
of invariant theory. Everything in its proper place!
Weyls paper [1923c] is sometimes seen as revealing an awakening interest
in the theory of invariants, which in turn encouraged his work on the representation of Lie groups. However, this does not quite agree with Weyls
19
In his book [Weyl 1946, pp. 56, 243245] Weyl gives a clear exposition of the polarization process and its role in the symbolical method.
` 3
SEMINAIRES
ET CONGRES
87
own words quoted at the beginning that his study of both the representation
and invariants of groups was motivated by his interest in the mathematical
substance behind the formal apparatus of relativity theory. In my opinion, in
order to understand Weyls move towards an interest in group representation
theory, it is more enlightening to consider what I regard as his second reply
to Studys criticism.
Weyls second reply was implicit Study is nowhere mentioned by name
and came about six weeks later in a paper submitted in January 1924 On
the symmetry of tensors and the scope of the symbolical method in the theory
of invariants[Weyl 1924a].20 The paper has two parts. In part one, on tensor
symmetries, Weyl answered the questions on tensor symmetries formulated
above. In part two, he applied these results to a question in the theory of
invariants that may well have been prompted by his encounter with Study.
Let me explain.
In part two Weyl considered the kind of invariant theoretic question that
would be of interest to a relativist. As we have seen, a typical problem considered in the theory of invariants would be that of determining the invariants
of the general linear group with the base form being the general covariant
tensor f of rank = 3 given in (13). In modern terms, this is the study
of the polynomial invariants of the representation of the general linear group
induced by its action on the 3-fold tensor product V V V . Formulated
as such, this would be a standard invariant-theoretic problem. But, as was
pointed out when discussing Weyls treatment of tensor symmetries in Raum,
Zeit, Materie, he stressed the fact that in physics and geometry tensors come
endowed with specic symmetry properties. Echoing this sentiment, Weyl
wrote in the rst part of [Weyl 1924a, p. 472]: For every tensor which arises,
a category characterized by symmetry relations must be specied a priori,
inside of which the tensor is to be thought of as freely variable.
So suppose that we consider instead of the general tensor of rank 3, the
tensors of that rank with prescribed symmetry properties as given by equations of the form (14). Then such tensors transform among themselves by
variable changes of, say, elements in the general linear group. As in the standard situations of invariant theory, the transformation of the coecients of
these tensors is linear and we may consider the invariants with respect to
these linear transformations. In other words, if W V V V consists of
the tensors satisfying some symmetry relations of the form (14), then W is a
a representation module in its own right, and we may consider the invariant
20
In Weyls Gesammelte Abhandlungen II, this paper is misleadingly placed (with the date
of submission omitted) after Weyls two notes of November 1924 [Weyl 1924b,c] announcing
his principal results on the representation of semisimple Lie groups.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
88
T. HAWKINS
polynomials of the associated representation. This is what Weyl proposed to

consider, albeit expressed in the older terminology. These are the type of invariants that might arise within relativistic physics, where physical quantities
are given by tensors with specic symmetry properties.
Now it turns out, as Weyl observed, that the symbolical method breaks
down in this case. That is, because f is not the general rank 3 covariant
tensor but is restricted by symmetry conditions (14), the link between its
invariants and the symbolical ones that is fundamental to the symbolical
method is severed. As a simple illustration of this fact, consider the general
skew symmetric rank two tensor f = a12 x1 y 2 a12 x2 y 1 , which has the linear
invariant I(a) = a12 . The symbolical method associates with I the expression
i(, ) = 1 2 , where = (1 , 2 ), = (1 , 2 ). But i is not a vector
invariant. Thus an ordinary invariant need not give rise in the usual manner
to a symbolical one, and so direct application of the method fails.
Weyl, however, perceived a way to salvage the symbolical method. Suppose, for example, that f belongs to the class of skew symmetric tensors.
Then by virtue of property P , f is obtained from a completely general tensor f of the same rank, and by virtue of this fact, Weyl could see how
to push through the symbolical method by utilizing f . To illustrate this
point we consider again the above skew symmetric tensor f of rank 2. By
property P , f can be obtained from the completely general rank two tensor
f = a11 x1 y 1 + a12 x1 y 2 + a21 x2 y 1 + a22 x2 y 2 :
1
1
1
1
= (a12 a21 )x1 y 2 (a12 a21 )x2 y 1 a
12 x1 y 2 a
12 x2 y 1 .
f = f f(12)
2
2
2
2
Thus f is expressible in terms of the coecients of the completely general f .
As a consequence the linear invariant I = a
12 = 12 (a12 a21 ) is expressible
in terms of the coecients of f and is an invariant with respect to f as
base form. Thus the symbolical method, which requires that the coecients
of the base form be completely unconstrained, may now be applied to I =
1
2 (a12 a21 ) to obtain the (skew symmetric) vector invariant

1 1 2
1
.
i(, ) = 1 2 2 1 =
2
2 1 2
Weyl could see how to do the same sort of thing for any symmetry class of
tensors, because he could generalize properties P , P to any such class. This
constitutes the rst part of his paper. There he proved the following result:
Theorem 2. Let f (x, y, z, ) =
of rank . Then:
` 3
SEMINAIRES
ET CONGRES
i j k
ijk Tijk x y z
denote a tensor
89
a) If W is the class of such tensors f which satisfy several symmetry relations

of the form (14), there is a single such relation which characterizes W.
b) Given any symmetry class
W, there exist constants cS , S S , such that
if f is arbitrary, then f = SS cS fS W. Moreover
every f W is

obtained in this manner since if f W, then f = SS cS fS W.
How did Weyl obtain these results? As he tells us: By applying the
representation theory of Frobenius to the group of all permutations one easily obtains a complete insight into the possible symmetry characteristics of
tensors [Weyl 1924a, pp. 4689]. Exactly how Weyl hit upon Frobenius
theory is not known, but the left
hand side of the general symmetry relation
(14) of Raum, Zeit, Materie, SS eS fS , certainly suggests looking at the
group algebra of the symmetric group S , and the structure of this algebra
was known to be related to the representations of S . In [1924a] Weyl wrote
(14) in the equivalent form

kS fS 1 = 0
(15)
and observed that one can associate to the form f an element f in the group
algebra H of the symmetric group S , namely

fS S.
(16)
f=
Since the f given in (16) depends upon the values of the variables x, y, z,
dening f , (16) actually denes a family of elements in H. Weyl glossed over
this point, but his results go through nonetheless.21 Direct
calculation then
shows that (15) is equivalent to kf = 0, where k =
SS kS S, and, for
example, part (b) of Theorem 2 can be deduced from the following result
about the group algebra H:

Theorem
3.
Given
k
=
SS kS S H, there is an element
c =
SS cS S H such that kf = 0 if and only if f = cf for some
f in H. Moreover, cf = f for all f satisfying kf = 0.

Theorem 3 was proved using the fact that the group algebra of the symmetric group decomposes into a sum of complete matrix algebras22 the
group algebra version of Frobenius complete reducibility theorem for nite
21
Weyl later touched on this point in his exposition of tensor symmetries and the group
algebra H in his book on group theory and quantum mechanics [Weyl 1931, p. 283].
22
That is, the linear associative algebra of all m m matrices for some m Z+.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
90
T. HAWKINS
groups. By virtue of this decomposition into complete matrix algebras the

proof of Theorem 3 is reduced to basic linear algebra, as Weyl showed. The
same is true of part (a) of Theorem 2.
So Weyl had discovered the value of Frobenius theory of group characters
and representations for answering questions from tensor algebra as well as
questions about the scope of the symbolical method. If this paper is viewed
as a response to Study, then the message would seem to be: I have discovered
in the representation theory of groups something of far greater importance
than the essentially formal symbolical method. Not only does it enable me
to gain insight into questions about the foundations of tensor algebra that
interest me, it also provides insight into the scope of the symbolical method
itself.
Once again, Weyl had discovered in the theory of groups the means to
answer questions motivated by relativity theory. In the case of the space
problem, however, he had utilized only the basic elements of Lies theory
primarily the fact that one could deal with certain group theoretic questions
more easily by working with the innitesimal group or Lie algebra. The same
can be said for Weinsteins work on his Fundamental Theorem of Tensor
Calculus. Now for the rst time, Weyl had gone beyond the elements of
group theory to achieve his goal. He had discovered the power of the theory
of representations for answering the questions that intrigued him.
The GroupTheoretic Foundation of Tensor

Calculus
The paper by Frobenius that was especially relevant here was his paper on the
primitive idempotents of the group algebra of the symmetric group [Frobenius 1903]. It was written after he learned of the work published by Alfred
Young in 1901 and 1902. Youngs work dealt with the theory of invariants,
but Frobenius could see that it related to the group algebra of the symmetric group and that Young had in eect obtained a formula for the primitive
idempotents. Since the primitive idempotents determine the irreducible representations and their characters, Frobenius deemed it worthwhile to rederive
Youngs formulae and to relate them to his theory of group characters. I
should mention that Frobenius himself preferred to present his work on group
representations without the explicit use of the hypercomplex numbersof the
group algebra and this was true of the above mentioned paper as well. As he
explained there: It is less signicant that I abstain from the use of hypercomplex numbers, since, as convenient as they occasionally are, they do not
` 3
SEMINAIRES
ET CONGRES
91
always serve to make the presentation more transparent [Frobenius 1903, p.

266]. Weyl, however, could read between the lines and interpret Frobenius
theory in terms of the group algebra, which was precisely the point of view
that was relevant to his study of tensor symmetries.
Weyl realized that the element c of the group algebra in Theorem 3 is an
idempotent. He also realized that if c is an idempotent,
then the totality W
of all tensors obtained from the operation f f = SS cS fS of part (b)
of Theorem 2 is a symmetry class with symmetry equation (15) given by
k = 1 c, 1 being the identity element of the symmetric group. Since the
primitive idempotents determine the irreducible representations of the symmetric group S , it would be natural to wonder whether if c is a primitive
idempotent, and hence given by Youngs formula, the representation of the
special linear group which it generates by means of the representation module
W would be one of the irreducible representations classied by Cartan according to highest dominant weights in [Cartan 1913]. Indeed, although Cartan
did not couch his results in terms of tensors, to anyone with a background in
tensor algebra, it would have been clear that the representation modules he
constructed could be regarded as consisting of tensors.
Weyl eventually discovered how to associate with a given dominant weight
a Young tableau such that the primitive idempotent c it denes generates,
in the above sense, the symmetry class W of tensors which is an irreducible
representation module for that weight. Consider, for example, G =SL(4, C)
and the irreducible representation module of highest weight = 3i=1 pi i ,
where the i are the fundamental dominant weights and pi 0 in accordance
with Cartans theory. If we set m1 = p1 + p2 + p3 , m2 = p2 + p3 , m3 = p3 ,
then m1 m2 m3 , and the mi dene the shape of a Young tableau, with
third of length m3 , which
rst row of length m1 , second row of length m2 and
corresponds to the symmetric group S with = 3i=1 mi . For example, if
= 21 + 2 + 33 , so m1 = 6, m2 = 4, m3 = 3 and = 13, then
T
1 2 3 4 5 6
7 8 9 10
11 12 13
is a Young tableau of the given shape. If R denotes the subgroup of elements

of S13 which permute the numbers in the rows of T among themselves and
if C is dened analogously with respect to the columns of T , then the corresponding Young-Frobenius
primitive idempotent c turns out to be a constant

multiple of e = P S (P )P , where (P ) = sgn C if P = RC with R R
and C C and (P ) = 0 otherwise.23 Since a constant multiple does not
23
The constant multiple is
f
,
!
where f is the degree of the irreducible representation of

E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
92
T. HAWKINS
change the symmetry class dened by part (b) of Theorem 2, an irreducible

module of weight = 21 + 2
+ 33 consists of all the tensors obtained
from the operation f f =

SS13 (S)fS applied to the general ten
sor
f . In other words, the tensors characterized by the symmetry relation
S kS fS 1 = 0, where k = 1 c, form an irreducible module of hightest
weight 21 + 2 + 33 . Noting that Cartan had not indicated this connection
between symmetry classes of tensors and irreducible representations, Weyl
expressed the conviction that it was through this connection that the entire
matter is rst placed in the right light.24
The fact that tensors with specic symmetry characteristics are the basis
of all the irreducible representations of SL(n, C) had an intriguing implication
for Weyl: If the complete reducibility theorem of Frobenius theory were true
for SL(n, C), it would mean that tensors with specic symmetry properties are the building blocks for all representations of SL(n, C). Weyl
perhaps realized that Lie, prompted by Studys above-mentioned conjectures,
had conjectured the truth of what amounts to the complete reducibility theorem for SL(n, C) in the third and nal volume of his Theory of Transformation
Groups [Lie 1893, pp. 7856]; but, in any case, it was a paper by Frobenius
student, Issai Schur, that put Weyl in a position to prove it. On January 10,
1924, and hence at about the same time as Weyl submitted his paper [1924a]
on tensor symmetries, Schur presented a paper [1924] to the Berlin Academy
in which he pointed out how Frobenius theory of group representations, including the complete reducibility theorem, could be extended to the rotation
group of n-dimensional space, SO(n, R). He also referred to his dissertation
[Schur 1901] in which he had studied representations A R(A) of GL(n, C)
for which the entries of the matrix R(A) are polynomials in the entries of
the matrix A, and this may have been how Weyl learned of it.25 These are
the sort of representations that come up implicitly in the classical theory of
invariants and that is why Schur was interested in them. Making critical
use of the polynomial nature of the entries of R(A), he showed how to associate with each irreducible representation R(A) an irreducible representation
of a symmetric group. As we have seen, such a connection can be given by
13 associated to c. Frobenius [1903, p. 265] and Weyl [1925, p. 573] gave dierent ways
of dening f directly in terms of .
24
Weyl made this comment in his rst communication on these matters [Weyl 1924b,
p. 462]; they are discussed in somewhat more detail in [Weyl 1925, pp. 5589, 5713].
Complete details, in the context of GL(n, C ), are given in his lectures [Weyl 1934, pp.
21.].
25
Although Schurs dissertation was a brilliant piece of work, it was only published as a
separate pamphlet (as was required of dissertations). Many dissertations (e.g., Weinsteins)
were also published in journals.
` 3
SEMINAIRES
ET CONGRES
93
the above primitive idempotent c of Young and Frobenius, but Schur, whose
work predates that of Young and Frobenius, made the connection in a different, more complicated way.26 However he showed how it could be used to
obtain many beautiful theorems, including a complete reducibility theorem
for the polynomial representations R(A).
In neither his dissertation nor his paper [1924] was Schurs primary goal
an extension of Frobenius theory to continuous groups. In [Schur 1924] he
was concerned with a counting problem in the theory of invariants that had
been solved in the classical case of the in- and covariants of binary forms
by Cayley in 1856 and for the invariants of nite groups by T. Molien in
1897.27 Molien had used the representation theory of nite groups to solve
his problem, and Schur realized he could do the same for the invariants of
the rotation group by extending Frobenius theory to this group. His method
of extension was based upon a technique introduced by Adolph Hurwitz and
involved replacing summation over a nite group with invariant integration
over a compact Lie group. Hurwitz had used the technique to extend Hilberts
basis theorem to orthogonal invariants a new result but he also used it
to give another proof for invariants with respect to SL(n, C). The application
of the technique to SL(n, C), which is not compact, involved an idea which
Weyl later dubbed the unitarian trick. Weyl saw how to use the same sort
of trick to establish the complete reducibility theorem for SL(n, C), thereby
showing that tensors (with prescribed symmetry conditions) are the building
blocks for all possible representations.
The paper Weyl presented to the Gottingen Academy of Sciences in
November 1924 [Weyl 1924b] announcing this discovery (as well as others),
was entitled Das gruppentheoretische Fundament der Tensorrechnung, and
in it he opined that the true group theoretic foundation of the tensor calculus
was to be found in the above-mentioned consequence of the complete reducibility theorem for SL(n, C). In the rst part of his famous series of papers
on the representations of semisimple Lie groups [Weyl 1925, pp. 5456], which
bore the same title as the Gottingen paper, Weyl put the matter as follows.
Tensors, he explained, are examples of what he called linear quantities.
Consider for example the mixed tensors of rank three Tijk discussed earlier at
(5). Each such Tijk may be regarded as an N -tuple (with N = n3 ). The change
of basis (2) corresponding to a matrix M SL(n, C) brings with it the varii , y j yj , zk zk which leads to an expression (8) for f
able changes xi x
26
After Schur learned of Weyls approach, he returned to the subject of his dissertation and
developed another way to make the connection that was simpler than his original approach
[Schur 1927, pp. 70, 72.].
27
The history of this counting problem is treated in my paper [Hawkins 1986].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
94
T. HAWKINS
in terms of the barred variables, and implicitly denes the linear transformation R(M ) : Tijk Tijk . It is easily seen that R(M2 M1 ) = R(M2 )R(M1 ).
For this reason the Tijk constitute a linear quantity. In general, according
to Weyl, a linear quantity is an N -tuple (a1 , . . . , aN ) which transforms
by a linear transformation R(M ) : (a1 , . . . , aN ) (
a1 , . . . , a
N ) such that
R(M2 M1 ) = R(M2 )R(M1 ). Of course linear quantities are just representations of SL(n, C), but Weyl recast the notion here in a form that was more
congenial to the mathematical context from which he was coming the
mathematics of relativity theory. Thus the complete reducibility theorem for
SL(n, C) became in this language the theorem that the only linear quantities
are the tensors. It was in this form that it was particularly meaningful for
him.
Weyl regarded this theorem as the proper group theoretic justication
of the tensor calculus [Weyl 1925, p. 546]. In other words, he had obtained
through the theory of groups, and in particular through the theory of group
representations as augmented by his own contributions what he felt
was a proper mathematical understanding of tensors, tensor symmetries, and
the reason they represent the source of all linear quantities that might arise
in mathematics or physics. Once again, he had come to appreciate the importance of the theory of groups and now especially the theory of group
representations for gaining insight into mathematical questions suggested
by relativity theory. Unlike his work on the space problem or Weinsteins
work on the fundamental theorem of the tensor calculus, however, Weyl now
found himself drawing upon far more than the rudiments of group theory.
His study of tensor symmetries had drawn upon Frobenius theory of group
representations and his own fundamental theorem for tensors had involved
him with the continuous analog of Frobenius theory. And of course Cartan
had showed that the space problem could also be resolved with the aid of
results about representations. In short, the representation theory of groups
had proved itself to be a powerful tool for answering the sort of mathematical
questions that grew out of Weyls involvement with relativity theory.
Frobenius had more or less developed all the essentials of the theory of representations for nite groups, but that was not at all the case for continuous
groups, notwithstanding the important contributions contained in the work
of Cartan and Schur. Their work certainly suggested to Weyl the potential
richness of a continuous analog of Frobenius theory, but it did not constitute
a coherent theory. Schur, who was unaware of Cartans work, had concentrated on two specic groups and had emphasized the role of group characters, whereas Cartan dealt with all semisimple groups but on the innitesimal
level using his theory of weights and without any complete reducibility the
` 3
SEMINAIRES
ET CONGRES
95
orem. Having become convinced of the importance of group representation

theory, Weyl went on to extend his results about SL(n, C) so as to create
in his brilliant papers of 192526 a coherent theory of the representation of
all semisimple groups within which the approaches of Schur and Cartan were
linked together for the rst time.
Epilogue
Two additional points need to be made.
1. Weyl was not the only mathematician interested in tensor algebra who
saw a connection with the representation theory of the symmetric group. As
early as 1919 J. A. Schouten (18831971) studied the problem of expressing
a tensor as a sum of irreducible tensors with symmetry properties. To this
end he utilized the group algebra of the symmetric group and Frobenius theory of group characters (in a formula for the principal idempotents). He was,
however, unaware of Frobenius paper [1903] or Youngs work and independently developed notions akin to that involving Young tableaux. Schoutens
work was expounded (with complete references to his earlier publications)
in his book [Schouten 1924, VII], which Weyl cites in [Weyl 1924b, p. 462,
n.2]. There is no evidence that Weyl knew of Schoutens work earlier and was
inuenced by it. Schouten actually submitted a note [1923] illustrating his
method (on an example suggested by Study!) to the Rendiconti del Circolo
Matematico di Palermo a year before Weyls paper [1924a] on tensor symmetries was submitted to the same journal. Had Weyl known of this note he
most certainly would have cited it in his own.
2. I have suggested that Weyl wrote his papers of 192526 on representation theory with the conviction that the theory was a powerful instrument
for answering questions suggested by theoretical physics. Weyl acted on that
conviction shortly after he completed the above papers. This time however
group representations were utilized to deal with questions arising from the
new quantum mechanics initiated by the work of Heisenberg and Schrodinger
in 192425. By 1927 their work had led to further developments by theoretical
physicists such as Born, Pauli, Jordan and Dirac and, from the mathematicians side, by von Neumann. Weyl seems to have assimilated and mastered
these developments as rapidly as he had mastered relativity theory a decade
earlier. Thus in a paper, Quantenmechanik und Gruppentheorie, we nd
him posing the question: How do I arrive at the matrix, the Hermitean
form, which represents a given quantity in a physical system of known constitution?[Weyl 1927, p. 90] To explain, precisely, what Weyl meant by this
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
96
T. HAWKINS
would take us too far aeld. But his further words should have a familiar ring
to them by now: Here with the help of group theory I believe I have succeeded in arriving at a deeper insight into the true nature of things (p. 91).
By group theory Weyl meant representation theory. This time it was to
the study of unitary projective (or ray) representations of the abelian Lie
group R2f that he turned for the deeper insight. Once again the mathematics
generated by the question went on to have a fruitful life of its own.28
Bibliography
Borel (A.)
[1986]
Hermann Weyl and Lie Groups, in Hermann Weyl, 18851985 (edited

by K. Chandrasekharan), Berlin-Heidelberg: Springer, pp. 5382.
Cartan (E.)
[uvres]
uvres compl`
etes, 6 vols., Paris: Gauthier-Villars, 19521955.
[1909]
Les groupes de transformations continus, innis, simples, Ann. Sci. Ecole

Norm. Sup., (III) 26 (1909), pp. 93161; uvres, II-2, (pp. 857925).
[1913]
Les groupes projectifs qui ne laissent invariante aucune multiplicite plane,

Bull. Sci. Math., 41 (1913), pp. 5396; uvres, I-1, pp. 255398.
[1914]
Les groupes projectifs continus reels qui ne laissent invariante aucune

multiplicite plane, J. Math. Pures Appl., 10 (1914), pp. 149186; uvres,
I-1, pp. 493530.
[1922]
Sur un theorème fondamental de M. H. Weyl dans la theorie de lespace

metrique, C. R. Acad. Sci. Paris, 175 (1922), pp. 8285; uvres, III-1,
pp. 629632.
[1923]
Sur un theorème fondamental de M. H. Weyl, J. Math. Pures Appl., (IX)

2 (1923), pp. 167192; uvres, III-1, pp. 633658.
Chevalley (A.) and Weil (A.)

[1957]
Hermann Weyl, Enseign. Math., 3 (1957), pp. 5382. Reprinted in H.

Weyl, Ges. Abh., 4, pp. 655685.
Dieudonn
e (J.)
[1975]
Introductory remarks on algebra, topology and analysis, Hist. Math., 2

(1975), pp. 537548.
[1976]
Weyl, Hermann, in Dictionary of Scientic Biography, 16, 1976, pp. 280

285.
28
See Mackey [1988, pp. 140.] for an interesting account of Weyls question and its historical background as well as a discussion of Weyls answer in terms of group representations
and the subsequent mathematics it generated.
` 3
SEMINAIRES
ET CONGRES
97
Einstein (A.) and Grossmann (M.)

[1913]
Entwurf einer verallgemeinerten Relativitats-theorie und die Theorie der

Gravitation. I. physikalischer Teil von A. Einstein. II. mathematischer
Teil von M. Grossmann, Zeit. Math. Phys., 62 (1913), pp. 225261.
Folland (G.)
[1970]
Weyl manifolds, J. Dierential Geom., 4 (1970), pp. 145153.
Frobenius (G.)
[1903]
Uber
die charakteristischen Einheiten der symmetrischen Gruppe,
Sitzungsberichte der Akademie der Wiss. zu Berlin, (1903), pp. 328358.
Ges. Abh 3, Springer, 1968, pp. 244274.
Hawkins (T.)
[1986]
Cayleys counting problem and the representation of Lie algebras, in Proceedings of International Congress of Mathematicians, Berkeley, 1986,
vol. 2, pp. 16421656.
Lie (S.)
[1893]
Theorie der Transformationsgruppen, vol. 3 Leipzig, 1893.
Mackey (G.W.)
[1988]
Hermann Weyl and the Application of Group Theory to Quantum

Mechanics, in Exakte Wissenschaften und ihre philosophische Grundlegung. Vortr
age des Internationalen Hermann-Weyl Kongresses, Kiel 1985
(edited by W. Deppert et al.), Frankfurt am Main: Peter Lang, 1988,
pp. 131159.
Misner (C.), Thorne (K.) and Wheeler (J.)

[1973]
Gravitation, New York: W. H. Freeman, 1973.
Mller (C.)
[1972]
The Theory of Relativity, Oxford: Clarendon Press, 2nd ed., 1972.
Pauli (W.)
[1921]
Relativit
atstheorie, Leipzig : Teubner, 1921. English translation by
G. Field, New York: Pergamon Press, 1958.
Reich (K.)
[1994]
Die Entwicklung des Tensorkalk

uls. Vom absoluten Dierentialkalk
ul zur
Relativit
atstheorie. Basel: Birkhauser, 1994.
Ricci (G.) and Levi-Civita (T.)

[1900]
Methodes de calcul dierentiel absolu et leurs applications, Math. Ann.,

54 (1900), pp. 125201.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
98
T. HAWKINS
Scheibe (E.)
[1988]
Hermann Weyl and the nature of spacetime, in Exakte Wissenschaften

und ihre philosophische Grundlegung. Vortr
age des Internationalen
Hermann-Weyl Kongresses, Kiel 1985 (edited by W. Deppert et al.),
Frankfurt am Main: Peter Lang, 1988.
Scholz (E.)
[1994]
Hermann Weyls contribution to geometry, 19171923, in The Intersection of History and Mathematics (edited by C. Sasaki, M. Sugiura and
J.W. Dauben), Basel: Birkhauser, 1995, pp. 203230.
[1995]
Hermann Weyls Purely innitesimal geometry, in Proceedings of the

international congress of mathematicians, Z
urich 1994 (edited by S.D.
Chatterji), vol. II, Z
urich: Birkhauser, pp. 15921603.
Schouten (J.A.)
[1923]
Uber
die Anwendung der allgemeinen Reihenentwicklungen auf eine
bestimmte quaternare Form sechsten Hauptgrades, Rend. Circ. Mat.
Palermo, 47 (1923), pp. 409425. [Dated Delft, Januar 1923].
[1924]
Der Ricci-Kalk
ul. Eine Einf
uhrung in die neueren Methoden und Probleme der mehrdimensionalen Dierentialgeometrie. Berlin: Springer,
1924.
Schur (I.)
[Ges. Abh.] Gesammelte Abhandlungen, 3 vols., Berlin-Heidelberg: Springer, 1973.
[1901]
Uber
eine Klasse von Matrizen, die sich einer gegebenen Matrix zuordnen
lassen. Berlin, 1901; Ges. Abh., 1, pp. 172.
[1924]
Neue Anwendungen der Integralrechnung auf Probleme der Invariantentheorie, Sitzungsberichte der Akademie der Wiss. zu Berlin, (1924),
pp. 189208; Ges. Abh., 2, pp. 440459.
[1927]
Uber
die rationalen Darstellungen der allgemeinen linearen Gruppe,
Sitzungsberichte der Akademie der Wiss. zu Berlin, (1927), pp. 5875;
Ges. Abh., 3, pp. 6885.
Sigurdsson (S.)
[1991]
Hermann Weyl, Mathematics and Physics, 19001927, Ph.D. thesis,

Harvard University, 1991.
Study (E.)
[1897]
Uber
die Invarianten der projectiven Gruppe einer quadratischen Mannigfaltigkeit von nicht verschwindenden Discriminante, Ber. Verh. S
achs.
Akad. Wiss., Math.-Phys. Kl., 49 (1897), pp. 443461.
` 3
SEMINAIRES
ET CONGRES
[1923]
99
Einleitung in die Theorie der Invarianten linearer Transformationen auf

Grund der Vektorenrechnung, Braunschweig: Vieweg, 1923.
Vizgin (V.P.)
[1989]
Einstein, Hilbert, and Weyl: The genesis of the geometrical unied eld
theory program, in Einstein and the History of General Relativity (edited
by D. Howard and J. Stachel), Boston: Birkhauser, 1989, pp. 300314.
Weinstein (A.)
[1923]
Fundamentalsatz der Tensorrechnung, Math. Z., 16 (1923), pp. 78

91. Published separately as Weinsteins Inaugural Dissertation, Berlin,
1922; Alexander Weinstein Selecta (J.B. Diaz ed.) London-San Francisco:
Pitman, 1978, pp. 115.
Weyl (H.)
[Ges. Abh.] Gesammelte Abhandlungen, 4 vols., Springer, 1968.
[1918]
Reine Innitesimalgeometrie, Math. Z., 2 (1918), pp. 384411. [Received

8 June 1918]; Ges. Abh., 2, pp. 128.
[1921a]
Raum, Zeit, Materie, 4th. ed., Berlin: Springer, 1921. Reprinted by Dover
Publications, New York, 1952. Translated into French by Gustave Juvet
and Robert Leroy as Temps, espace, mati`
ere, Paris: Blanchard, 1922.
[1921b]
Die Einzigartigkeit der Pythagoreischen Massbestimmung, Math. Z., 12

(1921), pp. 114146. [Received 24 April 1921]; Ges. Abh., 2, pp. 263295.
[1922]
Das Raumproblem. (Vortrag, gehalten auf der Jenenser Tagung),

September 1921, Jahresber. Deutsch. Math.-Ver., 31 (1922), pp. 205221.
[Received 1 February 1922]; Ges. Abh., 2, pp. 328344.
[1923a]
Mathematische Analyse des Raumproblems. Vorlesungen gehalten in

Barcelona und Madrid, Berlin: Springer, 1923. [Preface dated Z
urich,
im April 1923].
[1923b]
Zur Charakterisierung der Drehungsgruppe, Math. Z., 17 (1923), pp. 293

320. [Received 17 February 1923]29; Ges. Abh., 2, pp. 345372.
[1923c]
Randbemerkungen zu Hauptproblemen der Mathematik, Math. Z., 20

(1924), pp. 131150 [Received 25 October 1923]; Ges. Abh., 2, pp. 433
452.
Uber
die Symmetrie der Tensoren und die Tragweite der symbolischen
Methode in der Invariantentheorie, Rend. Circ. Mat. Palermo, 48 (1924),
pp. 2936 [Dated Z
urich, Januar 1924]; Ges. Abh., 2, pp. 468475.
[1924a]
29
The date of receipt of this paper is given as 17 February 1922, but it would seem to
be a typographical error. For example, throughout the paper Weyl cites (in text as well
as in footnotes) the German version [1923a] of his lectures in Spain (originally presented
in French and Castilian). The lectures themselves did not occur until March 1922, and the
preface to [1923a] is dated April, 1923. I therefore assume the correct date of receipt was 17
February 1923.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
100
T. HAWKINS
[1924b]
Das gruppentheoretische Fundament der Tensorrechnung, Nachr. Ges.

Wiss. G
ottingen, (1924), pp. 218224 [Presented 21 November 1924];
Ges. Abh., 2, pp. 461467.
[1924c]
Zur Theorie der Darstellung der einfachen kontinuierlichen Gruppen.

(Aus einem Schreiben an Herrn I. Schur), Sitzungsber. Akad. Wiss. Berlin,
(1924), pp. 338345 [Letter dated 28 November 1924]; Ges. Abh., 2,
pp. 453460.
[1925]
Theorie der Darstellung kontinuierlicher halbeinfacher Gruppen durch

lineare Transformationen, IIII, Math. Z., 23 (1925), pp. 271301 [received
January 21, 1925]; Ibid., 24 (1926), pp. 328376 [received February 11,
1925]; Ibid., 24 (1926), pp. 377395 [received April 23, 1925]; Ges. Abh.,
2, pp. 544645.
[1927]
Quantenmechanik und Gruppentheorie, Zeit. Phys., 46 (1927), pp. 146;

Ges. Abh., 3, pp. 90135.
[1931]
The Theory of Groups and Quantum Mechanics, London: Methuen, 1931,

translated by H. P. Robertson from the second edition (1930) of Gruppentheorie und Quantenmechanik. Reprinted by Dover, New York, 1950.
[1934]
The Structure and Representation of Continuous Groups, lithographed

notes by R. Brauer of Weyls lectures at the Institute of Advanced Study,
Princeton, 19341935.
[1946]
The Classical Groups. Their Invariants and Representations, 2nd ed.,

Princeton University Press, 1946.
[1949]
Relativity theory as a stimulus in mathematical research, Proc. Amer.

Philos. Soc., 93 (1949), pp. 535541; Ges. Abh., 4, pp. 394400.
` 3
SEMINAIRES
ET CONGRES
Histoire de la th
eorie des faisceaux
Christian Houzel
R
esum
e
La notion de faisceau a ete introduite par Jean Leray juste après
la guerre, dans le prolongement de travaux entrepris durant sa
captivite en Autriche. Leray a deni des groupes de cohomologie
pour les applications continues, et relie la cohomologie dune application a` celle de sa source grace à la suite spectrale, introduite a`
ce propos. Henri Cartan a reformule la theorie des faisceaux dans
son Seminaire et, avec Jean-Pierre Serre, il en donna des applications spectaculaires a` la theorie des espaces analytiques. Par
la suite, Serre a etendu à la geometrie algebrique ces methodes
que Grothendieck a largement renovees et generalisees. Enn,
Sato a exploite les methodes de Grothendieck dans le cadre des
D-modules, fondant ainsi lanalyse microlocale.
Abstract
Sheaf theory was introduced by Jean Leray just after the Second World War, as a continuation of his work while he was a
prisoner in Austria. Leray dened cohomology groups for continuous maps, and related them to the cohomology of the source
space by means of the spectral sequence he introduced for this
purpose. Henri Cartan reformulated sheaf theory in his seminar
and, together with Jean-Pierre Serre, gave spectacular applications to the theory of analytic spaces. Subsequently Serre extended these methods to algebraic geometry, when Grothendieck
enlarged and generalized them enormously. Finally Sato applied
Grothendiecks methods to D-modules, creating microlocal analysis.
AMS 1991 Mathematics Subject Classication: 01A65, 55-03, 14-03, 35A27

I.U.F.M. Paris 10, rue Molitor, 75016 Paris.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
102
1.
C. HOUZEL
Introduction
La notion de faisceau, introduite pour la première fois par J. Leray en 1946,

avec la theorie cohomologique correspondante et la notion de suite spectrale,
est une de celles qui a renouvele le plus profondement les methodes de la geometrie. Leray avait en vue une reconstruction de la topologie algebrique, mais
la theorie des faisceaux na pas tarde à donner, entre les mains de H. Cartan
et de J.-P. Serre, les outils necessaires à la theorie des espaces analytiques
et à la geometrie algebrique. A. Grothendieck a developpe lalgèbre homologique dans un cadre assez large pour contenir la cohomologie a` valeurs dans
un faisceau et ses travaux de geometrie algebrique lont amene à reformuler
lalgèbre homologique en termes de categories derivees et à etendre les notions
despace topologique et de faisceau en denissant les topos. Les categories derivees donnent le bon cadre pour denir les operations fondamentales sur les
faisceaux (images directes ou images reciproques, produits tensoriels, objets
Hom). Ce cadre a ete systematiquement exploite par M. Sato pour elaborer la theorie des D-modules (ou systèmes dequations aux derivees partielles
lineaires) sur des varietes analytiques reelles ; à cote des six operations de Grothendieck, Sato considère aussi deux nouvelles operations, la specialisation et
la microlocalisation le long dune sous-variete, qui conduisent a` des faisceaux
sur le bre tangent et sur le bre cotangent respectivement. Dans cet expose,
nous indiquerons comment Leray a invente la notion de faisceau et comment
Cartan la transformee ; laissant de cote les travaux de Grothendieck exposes
par P. Deligne, nous terminerons en indiquant la denition des foncteurs de
specialisation et de microlocalisation de Sato.
2.
Le cours de captivit
e de Leray
Prisonnier en Autriche pendant la guerre, Leray a participe à une universite de captivite dans lOag XVII ; il avait prefere traiter un sujet plus
loin des applications que sa specialite (lhydrodynamique) de peur detre requis pour travailler a` leort de guerre allemand et il avait choisi de faire un
cours de topologie algebrique. Dans ce cours, quil a publie en 1945 dans le
Journal de math
ematiques pures et appliquees [Leray 1945a,b,c], il cherchait a`
se debarrasser des hypothèses inutiles et à associer aux espaces topologiques
des invariants algebriques sans passer par des constructions intermediaires.
Les invariants quil considerait etaient les groupes de cohomologie plut
ot que
les groupes dhomologie ; la distinction entre les deux theories datait de 1935
[Alexander 1935, Kolmogorov 1936], et la cohomologie presente lavantage
davoir toujours une structure multiplicative dont celle de lhomologie derive
` 3
SEMINAIRES
ET CONGRES

HISTOIRE DE LA THEORIE
DES FAISCEAUX
103
dans le cas o`
u on dispose de la dualite de Poincare. Leray appelle homologie
la cohomologie et il parle de groupes de Betti pour signier lhomologie.
Le procede de Leray est inspire de lhomologie de Cech

mais, pour eviter
les constructions intermediaires (nerf dun recouvrement, etc.), il remplace les
recouvrements utilises par Cech

par des objets qui portent dej`
a une structure
algebrique : les couvertures . Il appelle complexe abstrait une suite de
groupes commutatifs libres de type ni, chacun muni dune base (X p ) ,

avec une suite de (co)bords appliquant le groupe de degre (Leray dit plut
ot
dimension ) p dans le groupe de degre p + 1. Les cobords sont lineaires et
denis par leur action sur les elements de base :
X p X p ;
on impose que le cobord dun cobord soit nul. Un complexe abstrait est rendu
concret en associant à chaque element de base X p un support |X p |

qui est une partie non vide de lespace topologique E dont on veut denir la
cohomologie ; on impose que |X q | soit contenu dans |X p | chaque fois que
a X p , cest-à-dire quil existe une suite delements de

X q est adherent `
base commencant par X p et aboutissant à X q et dont chacun intervient

dans le cobord du precedent. Un tel complexe concret K est une couverture
sil verie les axiomes suivants :
les supports sont fermes ;
pour tout point x de E, le sous-complexe xK engendre par les elements
de base dont le support contient x est un simplexe , cest-à-dire que
sa cohomologie est triviale ;
la somme K 0 des elements de degre 0 est un (co)cycle, le cocycle unite.
Leray denit alors les formes dune couverture K à coecients dans
un anneau A : en degre p, ce sont les combinaisons lineaires Lp des X p à

coecients dans A. Le cobord dune forme est deni a` partir de celui de K
et on sait donc denir les formes qui sont des cocycles et celles qui sont des
cobords. La cohomologie H p (E, A) est denie comme celle des formes dune
couverture quelconque de E. Pour cela, si K et K sont deux couvertures, il
convient didentier une forme Lp de la couverture K avec la forme Lp .K
de la couverture intersection K.K ; celle-ci est denie comme un quotient du
produit tensoriel K K pour lequel on pose
|X p X q | = |X p | |X q |
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
104
C. HOUZEL
et on annule les elements de support vide. Leray demontre que, lorsque lespace E est normal, il sut de considerer une famille de couvertures stable
par intersection et admettant des supports arbitrairement petits ; lorsque E
est compact, une seule couverture sut, à condition que ses supports soient
simples (cest-à-dire a` cohomologie triviale).
3.
Les faisceaux et la suite spectrale
Letape suivante dans la construction de Leray consiste a` associer une theorie cohomologique à toute application continue fermee dun espace normal E
dans un autre E ; elle est exposee dans une suite de notes aux Comptes rendus de lAcademie des sciences en 1946. Lidee vient probablement de letude
de la topologie dune variete en considerant sa projection dans une variete de
dimension inferieure et les proprietes des bres de cette projection ; Picard
avait traite la topologie des surfaces algebriques de cette manière et cette
methode avait ete etendue par Lefschetz en dimension plus grande. Leray
se refère explicitement au travail de N. Steenrod sur lhomologie des espaces
bres [Steenrod 1943].
Le problème est que la cohomologie des bres varie. Picard et Lefschetz,
dans le cas o`
u les bres sont des courbes algebriques, se servaient de lequation dierentielle veriee par les periodes des integrales abeliennes sur ces
courbes (connexion de Gauss-Manin) et de la monodromie de cette equation.
Steenrod avait introduit la notion de système local de coecients dans
le cas dun bre ; les bres sont homeomorphes, mais il faut tenir compte de
loperation du groupe fondamental de la base dans leur homologie (analogue
a la monodromie de Picard-Lefschetz). Dans le cas general quil considère,
`
Leray introduit la notion de faisceau pour relier entre elles les cohomologies
des bres : au lieu de considerer seulement les bres 1 (x ) (x E ) et
leur cohomologie, il considère les fermes F de E , leurs images reciproques
1 (F ) et la cohomologie de ces images reciproques.
Cela le conduit à denir un faisceau B de modules (ou danneaux) sur un
espace topologique E comme une fonction associant a` chaque ferme F de E
un module (ou un anneau) BF de manière que B = 0 ; pour chaque couple
de fermes f, F tels que f F , on se donne de plus un homomorphisme de
restriction bF bF .f de BF dans Bf et on impose la condition de transitivite
(bF .f ).f = bF .f
chaque fois que
f f F.
` 3
SEMINAIRES
ET CONGRES

DES FAISCEAUX
105
Un tel faisceau est dit normal si tout element bF dun BF est la restriction
dun bV BV o`
u V est un voisinage ferme de F et si, de plus, la condition
bF .f = 0 (avec f F ) implique lexistence dun voisinage ferme v de f
contenu dans F et tel que bF .v = 0. Ces conditions signient que BF est la
u V parcourt la famille des voisinages fermes de F .
limite inductive des BV o`
Lexemple typique dun faisceau sur E est donne par
F H p (F, A)
o`
u A est un anneau xe ; ce faisceau est normal si lespace E est normal.
Leray denit alors la cohomologie de E relative a` un faisceau B en considerant les formes dune couverture
à coecients dans B ; en degre q, ce
de E
u X q, parcourt la base du groupe
sont des combinaisons lineaires
b X q, o`
de degre q de la couverture et, pour chaque , b B|X q, | . Pour avoir de
bonnes proprietes, Leray suppose E et B normaux. Lorsque E admet une
couverture C dont les supports sont simples relativement à B, la cohomologie
peut se calculer en utilisant uniquement cette couverture.
Si maintenant : E E est une application continue fermee entre deux
espaces topologiques normaux (Leray dit une representation fermee ) et si
B est un faisceau normal de modules sur E, Leray denit le faisceau image
(B) sur E en posant
(B)F = B1 (F )
pour tout ferme F de E ; les restrictions de (B) sont induites par celles
de B et on voit que (B) est un faisceau normal. Lanneau de coecients A
etant choisi, on considère le faisceau
B p : F H p (F, A)
sur E et le module H q (E , (B p )) est le (p, q)-ième module de (co)homologie
de relatif a` A.
Dans sa deuxième note, Leray montre comment la cohomologie de
contient une information sur la cohomologie de E : cest la première apparition de la suite spectrale. Lidee vient de lanalyse du lemme qui servait,
dans le cours de captivite, à etablir que la cohomologie dun espace normal
peut se calculer à laide dune famille de couvertures stable par intersection
et avec des supports arbitrairement petits ; Leray demontrait que, si K est
une couverture et C est un complexe tel que K e soit un simplexe pour
tout support e de C , les cohomologies de C et de K C sont identiques. Il
note maintenant P1p,q le (p, q)-ième module de cohomologie de et il arme
que ce module contient des sous-modules
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
106
C. HOUZEL
p,q
p,q
p,q
p,q
p,q
p,q
0 = Qp,q
1 = Q0 Q2 . . . Qq1 Pp+1 . . . P2 P1
tels que, pour chaque indice r [1, p], on ait un homorphisme surjectif
pr,q+r+1
r : Prp,q Qrpr,q+r+1 /Qr1
p,q
de noyau Pr+1
; de plus le p-ième module de cohomologie E p,0 de E relatif a`
lanneau A contient des sous-modules
0 = E 1,p+1 E 0,p E 1,p1 . . . E p1,1 E p,0

et il existe un homomorphisme surjectif
p,q
E p,q /E p1,q+1
: Pp+1
de noyau Qp,q
q1 . Leray donne la description de r (induit par le cobord de la
cohomologie de E ) et de mais pas celle des sous-modules en jeu, qui doit se
deduire de r et de . On voit que la structure de la cohomologie de permet
de calculer le gradue associe à la cohomologie de E ltree par les E pr,r . Ceci
permet à Leray dobtenir un certain nombre de resultats : par exemple, si E
est compact et que toutes les bres 1 (x ) sont simples , 1 induit un
isomorphisme de la cohomologie de E sur celle de E. Leray donne aussi des
applications a` la cohomologie dun espace bre de base simplement connexe
dans le cas o`
u A est un corps, a` la cohomologie de lespace homogène quotient
E dun groupe compact simplement connexe E par un sous-groupe ferme
dans le cas o`
u A = Q et il retrouve les resultats de Gysin [1941] sur les bres
en sphères et ceux de Samelson [1941] sur les groupes compacts operant sur
des sphères.
Leray a developpe sa theorie dans des cours au Collège de France en 194750, publies en 1950 dans le Journal de math
ematiques pures et appliquees
[Leray 1950a,b]. Il note, cette fois, B(F ) la valeur dun faisceau (B) sur un
ferme F et F1 b la restriction a` F1 F dun element b de B(F ) ; il considère
seulement des espaces X localement compacts. Un faisceau B sur X est dit
continu si la limite inductive des B(W ) pour les voisinages fermes W de
est nulle et que, pour tout ferme F , B(F ) est la limite inductive des B(V )
pour les voisinages fermes V de F ; si ces conditions sont veriees et
que, de plus, pour tout compact K, B(K) est la limite inductive des B(V ), V
voisinage ferme de K, on dit que B est propre.
Dans cette nouvelle presentation, Leray nimpose plus aux complexes detre
libres et il nest donc plus question de base privilegiee ; mais il suppose quils
sont munis dune structure multiplicative : ce sont des anneaux dierentiels. Un complexe abstrait K devient concret lorsquon attribue a` chacun de
` 3
SEMINAIRES
ET CONGRES

DES FAISCEAUX
107
ses elements k un support S(k) X (espace localement compact considere),

avec un comportement convenable du support relativement aux operations
algebriques (addition, multiplication) ; on peut alors lui associer un faisceau
F F K quotient de K par lideal des k dont le support ne rencontre pas F .
On dit que K est une couverture sil est sans torsion, gradue en degres positifs
avec un cobord de degre 1 et sil possède un element unite u de support X
tel que, pour tout x X, la cohomologie de xK soit reduite aux multiples de
xu.
Si B est un faisceau sur X, la cohomologie de X relative a` B est calculee
a laide dune couverture ne X : cest une couverture telle que, pour tout
`
recouvrement ouvert ni (V ) du compactie X , on puisse decomposer
lautomorphisme identique de X en une somme dendomorphismes tels que
S( k) V S(k)
pour tout k X et tout indice (partition de lunite). Pour les espaces
localement compacts de dimension nie, il existe des couvertures nes : on
les construit par le procede de Cech

ou celui dAlexander ; elles sont nulles en
degre strictement superieur a` la dimension de X. On considère alors le produit
tensoriel X B engendre par les k b avec b B(F ) et S(k) F (ferme de
X) et son quotient
X B par le sous-module des elements de
support vide (le

support de
k b etant lensemble des x X tels que x k b = 0) ;
ce quotient est un complexe dont la cohomologie, independante du choix de
la couverture ne X , est notee H (X B).
Dans la redaction de ses cours, Leray a adopte la presentation algebrique
` une ltration dun
de la suite spectrale elaboree par J.-L. Koszul [1947a,b]. A
anneau dierentiel ltre A :
A A(1) . . . A(p) A(p+1) . . .
on associe une ltration de la cohomologie HA et un calcul du gradue associe par une succession dapproximations de plus en plus precises. Au niveau
dapproximation r, on remplace le groupe C p des cocycles de ltration p
par un groupe plus grand
Crp = {a A(p) | a A(p+r) }
et le groupe D p des cobords de ltration p par le groupe plus petit
Drp = {a | a Crp+r };
on pose alors
p
p+1
+ Cr1
).
Hrp A = Crp /(Dr1
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
108
C. HOUZEL
Le cobord de A induit un cobord Hrp A Hrp+r A faisant de Hr A un

complexe et on verie que la cohomologie de ce complexe sidentie à Hr+1 A
(approximation du niveau suivant). Lorsque la ltration de A est bornee superieurement, le gradue associe à HA, cest-à-dire C p /(C p+1 + D p ) en degre p,
sidentie a` la limite inductive H A des Hr A ; si de plus Hl+1 A est concentre
en degre 0 pour un certain l, la suite (Hr A) est stationnaire pour r > l et sa
valeur est HA.
Si B est un faisceau dierentiel ltre propre sur un espace localement
compact X, Leray lui associe de meme un faisceau spectral (Fr B). Si X est
une couverture ne de X, lanneau spectral Hr (X B) ne depend pas du
choix de X et on le note Hr (X B) ; on a H2 (X B) H(X F1 B).
Dans le cas o`
u B est gradue avec une graduation bornee inferieurement et un
cobord de degre > 0, lhypothèse que B(x) est concentre en degre 0 pour tout
x X implique alors que H(X B) H(X F1 B) o`
u FB est le faisceau
de cohomologie de B. On peut calculer H(X B) a` laide du complexe de
Cech
K associe à un recouvrement ferme ni (F ) si on sait que, pour toute
intersection F des F , H(F B) HB(F).
Considerons maintenant une application continue : X Y , o`
u X et
Y sont localement compacts ; si B est un faisceau dierentiel ltre propre sur
X, on introduit des couvertures nes X de X et Y de Y pour faire les calculs
de cohomologie. La couverture 1 Y de X est denie comme le quotient de
Y par lideal des y tels que 1 (S(y)) soit vide, les supports etant les images
reciproques par des supports de Y ; la cohomologie H( 1 Y X B)
sidentie a` H(X B) muni dune ltration qui ne depend que de et non
du choix de X et de Y. Sous des hypothèses convenables de dimension nie,
le gradue associe sidentie a` la limite inductive des Hr ( 1 Y X B) ; on
a H2 ( 1 Y X B) H(Y F1 (X B)).
Leray applique en particulier ces resultats au cas o`
u est une bration de
bre F et o`
u le faisceau B est un anneau constant A ; alors F(X A) est
localement isomorphe à H(F A) et Leray retrouve les resultats de G. Hirsch
[1948], de Gysin [1941], de Chern et Spanier [1950], ainsi que ceux de Wang
[1949] pour le cas o`
u Y est une sphère dhomologie.
4.
Impact des id
ees de Leray et travaux de
H. Cartan
En 1947, A. Weil a communique à H. Cartan ses idees de demonstration

des theorèmes de de Rham [Weil 1947] ; la demonstration complète (maintenant classique) na ete publiee quen 1952 dans les Commentarii Mathematici
` 3
SEMINAIRES
ET CONGRES

DES FAISCEAUX
109
Helvetici [Weil 1952]. Daprès le commentaire quen fait Weil lui-meme dans
ses uvres, lidee de base lui avait ete suggeree par une conversation quil
avait eue a` Paris avec Leray en 1945. La meme annee 1947 sest tenu à Paris
un colloque international de topologie algebrique ; Leray et Cartan y ont pris
part, mais la publication du colloque par le CNRS en 1949 contient une forme
remaniee de leurs contributions ([Leray 1949], [Cartan 1949]). Cartan sest
donc interesse tout de suite a` la notion de faisceau ; il avait dailleurs dej`
a
rencontre quelque chose danalogue a` propos de certains problèmes de passage du local au global. Par exemple, Cartan [1945] sinteresse à lhomologie
Hn (U, T) des ouverts U dun espace localement compact de dimension n et à
la manière dont elle depend de U ; pour chaque inclusion V U douverts,
il y a un morphisme naturel de restriction Hn (U, T) Hn (V, T) et Cartan
etablit les proprietes de recollement qui sexprimeraient maintenant en disant
que U Hn (U, T) est un faisceau. Lautre problème de passage du local au
global concerne la theorie des fonctions de plusieurs variables complexes et
Cartan lavait considere dès 1934 ; là encore, les donnees locales sont relatives
a des ouverts. En 1950, K. Oka a publie dans le Bulletin de la Soci
`
ete mathematique de France un article dans lequel il introduit la notion dideal de
domaine indetermine, dont il attribue lintention a` H. Cartan : il sagit dun
ensemble (I) de couples (f, ) o`
u est un ouvert de Cn (ou un revetement
dun tel ouvert) et f est une fonction holomorphe dans et on suppose que
1. pour (f, ) (I) et holomorphe dans , (f, ) (I) ;
2. si (f, ) et (f , ) appartiennent a` (I), il en est de meme de
(f + f , ) ;
3. si (j ) est une suite croissante de domaines et que (f, j ) (I) pour
tout j, alors (f, j ) (I).
Il etait donc naturel pour Cartan de denir les faisceaux sur les ouverts
plut
ot que sur les fermes. Cest precisement ce quil fait dans son Seminaire
consacre à la topologie algebrique dans les annees 1948-51 ; la partie sur les
faisceaux de la première annee (1948-49) nest pas publiee. Le seminaire de
1950-51 [Cartan 1950-51] contient une nouvelle presentation de la theorie o`
u
les faisceaux sont denis en termes despaces etales suivant une idee de M.
Lazard : un faisceau sur un espace topologique X est un espace topologique F
muni dune application p : F X qui est un homeomorphisme local ; de plus
u
les bres Fx = p1 (x), (x X ) sont munies de structures de K-modules, o`
K est un anneau commutatif xe et on suppose que les lois de composition de
` chaque ouvert
ces structures sont continues au sens de la topologie de F . A
X de X , on associe le module (F, X) des sections de F au-dessus de X,
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
110
C. HOUZEL
cest-à-dire des applications s : X F qui, composees avec p donnent

lidentite de X ; si X Y (X, Y ouverts de X ), on a un homomorphisme de
restriction (F, Y ) (F, X) et ces homomorphismes sont transitifs. La
limite inductive des (F, X) pour X voisinage ouvert dun point x sidentie
a la bre Fx et inversement, on peut denir un faisceau en associant à tout
`
ouvert X un module FX et à toute inclusion X Y un homomorphisme de
restriction FY FX avec la condition de transitivite ; pour tout x X on
denit la bre Fx comme la limite inductive des FX (X voisinage ouvert de
x) et lespace etale F est la somme disjointe des Fx munie dune topologie
convenable. Bien entendu, lhomomorphisme canonique FX (F, X) nest
en general ni injectif ni surjectif. La denition des faisceaux comme espaces
etales devait sembler preferable car elle se faisait en termes de structure sur
un ensemble plut
ot quen termes de foncteur sur la categorie des ouverts.
Cartan (exp. 15) introduit des familles de supports generalisant les cas
consideres par Leray de la famille de tous les fermes ou de la famille des
compacts ; les elements de sont des fermes paracompacts, est hereditaire et stable par reunion nie et tout element de a un voisinage appartenant a` . Si une famille de supports est donnee, on considère, pour
tout faisceau F , le module (F ) des sections de F à supports dans ; si
f : F G est un homomorphisme de faisceaux, il denit un homomorphisme f : (F ) (G), mais la surjectivite de f nentrane pas en
general celle de f . Cependant, si Ker f est n, cest-à-dire si, pour tout recouvrement ouvert localement ni de X , il existe une partition de lautomorphisme identique de Ker f subordonnee à ce recouvrement, on peut conclure
a partir de celle de f . Comme exemples de faisceaux
a la surjectivite de f `
`
ns, Cartan donne le faisceau des cochanes dAlexander-Spanier (dej`
a utilise
par Leray) ou celui des cochanes singulières ; sur une variete dierentiable,
on peut encore considerer le faisceau des formes dierentielles (dun degre
donne).
La cohomologie à supports dans une famille est denie dune manière
axiomatique et on etablit son existence et son unicite (exp. 16) ; lanneau de
` chaque faisceau F , la cohomologie associe,
base K est suppose principal. A
pour tout entier q, un module Hq (X , F ) dependant fonctoriellement de F et
nul pour q < 0 ; on pose que H0 = et que à toute suite exacte courte de
faisceaux
0 F F F 0
sont associes (fonctoriellement) des homomorphismes de connexion
q : Hq Hq+1 permettant dobtenir une suite exacte longue de cohomologie. Après avoir etabli lunicite dune telle theorie par recurrence sur q,
en utilisant le plongement dun faisceau dans un faisceau n, Cartan en de
` 3
SEMINAIRES
ET CONGRES

DES FAISCEAUX
111
montre lexistence en utilisant une r

esolution ne ( faisceau fondamental )
0 K C0 C1 . . .
du faisceau constant de valeur lanneau de base K ; une telle resolution se
construit a` laide des cochanes dAlexander-Spanier ou, sur une variete differentiable, à laide des formes dierentielles et la cohomologie du faisceau
u designe le produit tensoriel des
F est denie comme H q ( (C F )) o`
faisceaux.
5.
Applications aux espaces analytiques et a

` la
g
eom
etrie alg
ebrique
Dans un article de 1950, Cartan a utilise un cas particulier de faisceaux en

theorie des fonctions analytiques de plusieurs variables complexes : il sagit
des faisceaux sur Cn qui sont des sous-faisceaux de Om . Ils sont denis en
m des
associant à chaque ouvert X de Cn un sous-module FX du module OX
m-uples de fonctions holomorphes dans X de manière que, pour X Y , le
m engendr
e par les restrictions à X des elements de FY
sous-module de OX
soit contenu dans FX ; pour une partie quelconque A de Cn , Cartan denit
FA comme la limite inductive des FX pour X voisinage ouvert de A et il
sinteresse en particulier au cas o`
u, pour tout ouvert X, FX est lensemble
m
des f OX dont le germe en chaque point x de X appartient a` Fx (il revient
au meme de dire que si le germe de f en un point x appartient a` Fx , pour
tout point y assez voisin de x, le germe de f en y appartient a` Fy ). La notion
importante introduite dans cet article est celle de faisceau coh
erent ; Cartan
dit que F est coherent en un point a sil existe un voisinage ouvert X de a
tel que FX engendre Fx pour tout x assez voisin de a. Il reprend ici, dans le
langage des faisceaux, une notion quil avait consideree dès 1944 sous le nom
de système coherent de modules.
En 1951-52, le Seminaire Cartan a ete consacre à la theorie des espaces
analytiques complexes. Les faisceaux y sont denis en termes despaces etales ;
un sous-faisceau F de Oq est dit coherent en un point x (exp. 15) sil existe
un voisinage ouvert U de x et un système ni delements ui de OUq dont
les germes engendrent Fy pour tout y assez voisin de x. Cartan etablit les
theorèmes classiques dOka dans le langage des faisceaux coherents : le faisceau
des relations entre un nombre ni de sections locales de Oq est coherent ; le
faisceau dideaux des fonctions nulles sur un sous-ensemble analytique est
coherent. La notion de coherence est etendue par la suite a` des faisceaux plus
generaux : un faisceau F de Omodules est dit coherent sil est localement
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
112
C. HOUZEL
isomorphe au quotient dun Op par un sous-faisceau coherent. Ceci permet à

Cartan de formuler et detablir les fameux theorèmes A et B sur les varietes
de Stein : si X est une telle variete et si F est un faisceau coherent dessus, les
bres de F sont engendrees par les sections globales et la cohomologie de F
est nulle en degres 1.
Au colloque belge sur les fonctions analytiques de plusieurs variables complexes en 1953, ces resultats sont repris avec une denition legèrement differente de la coherence : un faisceau est dit coherent sil est localement isomorphe au conoyau dun morphisme Op Oq [Cartan 1953]. Autrement
dit, les faisceaux coherents sont ceux qui sont localement de presentation nie ; on sait que cela equivaut bien a` la coherence, du fait que O lui-meme est
coherent.
De la meme annee 1953 datent le theorème de nitude de Cartan-Serre
pour la cohomologie dune variete analytique complexe compacte à coecients
dans un faisceau coherent [Cartan et Serre 1953], les theorèmes de Serre sur la
cohomologie dune variete projective a` coecients dans un faisceau coherent
[Serre 1953] (analogues aux theorèmes A et B en tordant le faisceau par un
O(n), n susamment grand ; cf. Seminaire Cartan [1953-54, exp. 19]), et le
theorème de dualite de Serre pour les faisceaux analytiques localement libres
(publie seulement en 1955, [Serre 1955b], dans les Comment. Math. Helv.)
Cest aussi lepoque o`
u la theorie des faisceaux a commence à etre utilisee
en dehors de lecole francaise, par exemple dans les travaux de K. Kodaira et
D. Spencer et dans ceux de F. Hirzebruch.
En 1955, J.-P. Serre a publie son article fondamental Faisceaux algebriques coherents dans lequel il applique la theorie des faisceaux a` la geometrie algebrique abstraite (sur un corps de base K algebriquement clos)
[Serre 1955a]. Les faisceaux y sont denis en termes despaces etales, mais
Serre donne les conditions necessaires et susantes pour quun prefaisceau
U FU etant donne, les homomorphismes canoniques FU (U, F)
soient des isomorphismes (F designe le faisceau engendre ; le terme prefaisceau , introduit plus tard par Grothendieck, manque encore). Les faisceaux
coherents sur un faisceau danneaux A sont denis comme des faisceaux de
A-modules localement de type ni et tels que le module des relations entre un
nombre ni de sections locales soit localement de type ni. Les varietes algebriques anes sont munies de la topologie de Zariski (dont les fermes sont les
sous-ensembles algebriques) et du faisceau des fonctions regulières à valeurs
dans K (fonctions denies par des fractions rationnelles). Une variete algebrique (au sens general) est un espace topologique X muni dun sous-faisceau
OX du faisceau des fonctions a` valeurs dans K et qui est localement isomorphe
(pour ces structures) à une variete ane ; Serre impose de plus une condition
` 3
SEMINAIRES
ET CONGRES

DES FAISCEAUX
113
de separation, la diagonale doit etre fermee dans X X, mais il ne suppose

pas ses varietes irreductibles. Comme la topologie de Zariski nest pas separee,
et encore moins paracompacte, on ne peut pas denir la cohomologie des faisceaux à laide de resolutions nes comme le faisait Cartan ; Serre revient au
procede de Cech
en commencant par etablir les proprietes cohomologiques des
varietes anes, analogues aux theorèmes A et B de Cartan, qui lui permettent
de justier ce procede pour la cohomologie des faisceaux coherents. Le point
essentiel est quon peut trouver des recouvrements arbitrairement ns dune
u les Qi sont des
variete ane X par des ouverts en nombre ni du type XQi o`
fonctions regulières sur X ; il resulte alors du theorème des z
eros que, pour un

entier N , il existe des fonctions regulières Ri telles que 1 = Ri QN
e
i , identit
qui remplace les partitions de lunite du cas paracompact.
Les developpements ulterieurs de la theorie des faisceaux et de ses applications à la geometrie algebrique sont surtout dus a` A. Grothendieck. Renvoyant
a larticle de P. Deligne pour plus de details, contentons-nous dindiquer que
`
larticle de Grothendieck en 1957 au T
ohoku Math. J. propose un cadre dalgèbre homologique assez large pour contenir la cohomologie des faisceaux ;
les groupes de cohomologie sont les foncteurs derives du foncteur (sections
globales) et on les calcule au moyen de resolutions injectives dont Grothendieck demontre lexistence en toute generalite. Après les notes dun cours
dA. Borel a` lETH de Z
urich [Borel 1951] o`
u la theorie des faisceaux est presentee daprès Leray, le premier livre entièrement consacre à la theorie des
faisceaux est celui de R. Godement [1958] ; dans ce livre, elabore daprès les
notes dun cours a` lUniversite dIllinois (1954-55), Godement introduit de
nouvelles classes de faisceaux acycliques très commodes pour le calcul de la
cohomologie : les faisceaux asques et les faisceaux mous. Signalons aussi la
denition de lhomologie a` valeur dans un faisceau par Borel et Moore [1960],
en vue de la dualite de Poincare.
6.
Applications a
` lanalyse microlocale
Dès 1959, M. Sato avait deni les hyperfonctions sur une variete analytique
reelle M au moyen de la cohomologie du faisceau des fonctions holomorphes
sur un voisinage complexe de M (cohomologie relative au complementaire de
M , par la suite remplacee par la cohomologie à supports dans M ) ; le faisceau
des hyperfonctions est asque. Letude des operateurs pseudo-dierentiels
analytiques a ensuite conduit Sato a` denir les microfonctions comme sections dun faisceau sur le bre en sphères cotangent à la variete [Sato 1969].
Par la suite le bre en sphères a ete remplace par le bre cotangent lui-meme
et les constructions de Sato se sont eclairees par lutilisation du langage des
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
114
C. HOUZEL
categories derivees de Grothendieck.

Indiquons comment on peut denir la specialisation et la microlocalisation
dun faisceau sur une variete analytique reelle M en suivant la presentation
donnee par Kashiwara et Schapira [1990]. On suppose que M est plongee dans
une variete analytique reelle X comme sous-variete de codimension l et on
M
denit la deformation normale de X le long de M comme une variete X
munie dune projection p dans X et dune application t dans R ; on impose
que (p, t) denisse un isomorphisme de p1 (X M ) sur (X M ) R et un
isomorphisme de = t1 (R ) sur X R tandis que t1 (0) sidentie au bre
TM X normal a` M dans X. La construction se fait localement et on peut la
decrire en supposant que M = Rnl et que X = Rl Rnl = Rn , un point x
de X secrivant (x , x ) avec x Rl et x Rnl tandis que les points de M
M = Rn R, p(x , x , y) = (tx , x )
sont de la forme (0, x ) ; on prend alors X

et t(x , x , y) = y. On a alors un diagramme de varietes analytiques
M
X
TM X
o`
u les èches horizontales sont des plongements ; on pose p = p j. Considerons un objet F de la categorie derivee Db (X) des complexes de faisceaux à
cohomologie bornee sur X. On lui associe le specialise le long de M ,
M (F ) = s1 Rj p1 F s! j! p! F,
objet de la categorie derivee Db (TM X) ; cest un objet conique au sens quil
est invariant par les homotheties positives du bre normal et son support
CM (SuppF ) est le cone normal au support de F le long de M , cest-à-dire lintersection de p1 (SuppF ) avec TM X. On interprete la cohomologie de M (F )
de la manière suivante : si V est un ouvert conique de TM X, H j (V, M (F ))
u U est un ouvert variable de X tel que
est la limite inductive des H j (U, F ) o`
V CM (X U ) = ; la bre H j (M (F )) du faisceau de cohomologie en
u U parcourt les
un point de TM X est la limite inductive des H j (U, F ) o`
ouverts tels que
/ CM (X U ). Le microlocalise M (F ) de F le long de M
est le transforme de Fourier-Sato de M (F ), deni au moyen du diagramme
p2
X
TM X X TM
p1
TM X
` 3
SEMINAIRES
ET CONGRES
TM
X

DES FAISCEAUX
115
X est le br
o`
u TM
e conormal à M dans X ; on a
1
M (F ) = Rp2 ! ((p1
1 M (F ))|P ) Rp2 RP (p1 M (F ))
X. Ainsi
en notant P la partie positive et P la partie negative de TM X X TM
X) ; si V est un c
one ouvert convexe
M (F ) est un objet conique de Db (TM
j
j
(U, F ) o`
u U est
dans TM X, H (V, M (F )) est la limite inductive des HZU
un ouvert de X tel que U M = (V ) et Z est un ferme tel que CM (Z) soit
contenu dans le polaire V de V et la bre H j (M (F ))p en un point p de
X est la limite inductive des H j (F )
e tel que CM (Z)(p)
TM
(p) pour Z ferm
Z
soit contenu dans lensemble des vecteurs normaux v pour lesquels #v, p$ > 0.
On peut appliquer ces constructions au cas o`
u X est une variete analytique
complexe munie dune fonction holomorphe f et o`
u M = Y est le lieu des
zeros de f . Soit
C C
p:C
le revetement universel de C , deni par p(z) = e2iz ; on construit un diagramme cartesien
X
C
p
a laide duquel on denit le foncteur f des cycles voisins et le foncteur f des

`
cycles evanescents de Grothendieck : pour un objet F de la categorie derivee
D b (AX ) des complexes de Amodules a` cohomologie bornee sur X,
p p1 (F ) i1 RHom(f 1 p! AC , F )
f (F ) = i1 R
et
f (F ) = i1 RHom(f 1 K, F )
o`
u i : Y X est linjection canonique et o`
u K est le complexe sur C
0 p! AC AC 0
Lorsque F est faiblement Cconstructible, on a f (F ) s1 Y (F ) et f (F )
u s : Y TY X est la section dimage f1 (1) (f est denie par
s1 Y (F ) o`
la dierentielle df et s : Y TY X est la section denie par df ).
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
116
C. HOUZEL
Bibliographie
Alexander (J.W.)
[1935]
On the chains of a complex and their duals, Proc. Nat. Acad. Sci. USA,
21 (1935), p. 509511.
Borel (A.)
[1951]
Cohomologie des espaces localement compacts daprès J. Leray, dans Sem.
Top. Alg. Ecole

Polyt. Fed., Z
urich ; 2e ed. 1957 ; 3e ed. dans vol. 2 des
Lecture Notes in Math., Springer, 1964.
Borel (A.) et Moore (J.C.)

[1960]
Homology theory for locally compact spaces, Michigan Math. J., 7 (1960),
p. 137159.
Cartan (H.)
[1944]
Ideaux de fonctions analytiques de n variables complexes, Ann. Sci. Ecole

Norm. Sup., III, 61 (1944), p. 149197.
[1945]
Methodes modernes en topologie algebrique, Comment. Math. Helv., 18

(1945), p. 115.
[1949]
Sur la notion de carapace en topologie algebrique, dans Topologie algebrique, Paris 1947, Colloques Internat. Centre Nat. Rech. Sci. 12, 1949,
p. 12.
[1950]
Ideaux et modules de fonctions analytiques de variables complexes, Bull.

Soc. Math. de France, 78 (1950), p. 2964.
[1950-51]
Seminaire Cohomologie des groupes, suite spectrale, faisceaux . Paris.
[1951-52]
Seminaire Theorie des fonctions analytiques de plusieurs variables complexes . Paris.
[1953]
Varietes analytiques complexes et cohomologie, in Colloque sur les fonctions de plusieurs variables, Bruxelles 1953, Centre belge de Rech. math.
p. 4155.
[1953-54]
Seminaire Theorie des fonctions automorphes et des espaces analytiques , Paris.
Cartan (H.) et Serre (J.-P.)

[1953]
Un theorème de nitude concernant les varietes analytiques compactes,

C. R. Acad. Sci. Paris, 237 (1953), p. 128130.
Chern (S.S.) et Spanier (E.)

[1950]
The homology structure of bre bundles, Proc. Nat. Acad. Sci. USA, 36
(1950), p. 248255.
` 3
SEMINAIRES
ET CONGRES

DES FAISCEAUX
117
Dieudonn
e (J.)
[1989]
A History of Algebraic and Dierential Topology 1900-1960. Boston :

Birkh
auser, 1989.
Godement (R.)
[1958]
Topologie algebrique et theorie des faisceaux. Paris : Hermann, 1958.
Gray (J.W.)
[1979]
Fragments of the history of sheaf theory, dans Applications of Sheaves,

vol. 753 des Lecture Notes in Math., Springer, 1979.
Grothendieck (A.)
[1957]
Sur quelques points dalgèbre homologique, T

ohoku Math. J., 9 (1957),
p. 119221.
[1973]
Groupes de monodromie en geometrie algebrique, dans Seminaire de

Geom. alg. du Bois-Marie 1967-69, SGA 7 II, vol. 340 des Lecture Notes
in Math., Springer, 1973.
Gysin (W.)
[1941]
Zur Homologietheorie der Abbildungen und Faserungen der Mannigfaltigkeiten, Comment. Math. Helv., 14 (1941), p. 61122.
Hirsch (G.)
[1948]
Un isomorphisme attache aux structures brees, C. R. Acad. Sci. Paris,

227 (1948), p. 13281330.
Houzel (C.)
[1990]
Les debuts de la theorie des faisceaux, dans [Kashiwara et Schapira 1990,

p. 722].
Kashiwara (M.) et Schapira (P.)

[1990]
Sheaves on Manifolds. Berlin : Springer, 1990.
Kolmogorov (A.)
[1936]
Uber
die Dualitat im Aufbau der kombinatorischen Topologie, Mat.
Sborn., 1 (1936), p. 701705.
Koszul (J.L.)
[1947a]
Sur les operateurs de derivation dans un anneau, C. R. Acad. Sci. Paris,

224 (1947), p. 217219.
[1947b]
Sur lhomologie des espaces homogènes, C. R. Acad. Sci. Paris, 224

(1947), p. 477479.
Leray (J.)
[1945a]
Sur la forme des espaces topologiques et sur les points xes des representations, J. Math. Pures Appl., (IX) 24 (1945), p. 95167.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
118
C. HOUZEL
[1945b]
Sur la position dun ensemble derme de points dun espace topologique,

J. Math. Pures Appl., (IX) 24 (1945), p. 169199.
[1945c]
Sur les equations et les transformations, J. Math. Pures Appl., (IX) 24

(1945), p. 201248.
[1946a]
Lanneau dhomologie dune representation, C. R. Acad. Sci. Paris, 222

(1946), p. 13661368.
[1946b]
Structure de lanneau dhomologie dune representation, C. R. Acad. Sci.

Paris, 222 (1946), p. 14191422.
[1946c]
Proprietes de lanneau dhomologie de la projection dun espace bre sur

sa base, C. R. Acad. Sci. Paris, 223 (1946), p. 395397.
[1946d]
Sur lanneau dhomologie de lespace homogène, quotient dun groupe clos

par un sous-groupe abelien, connexe, maximum, C. R. Acad. Sci. Paris,
223 (1946), p. 412415.
[1949]
Lhomologie ltree, dans Topologie algebrique, Paris 1947, Colloques Internat. Centre Nat. Rech. Sci. 12, 1949, p. 6182.
[1950a]
Lanneau spectral et lanneau ltre dhomologie dun espace localement

compact et dune application continue, J. Math. Pures Appl., (IX) 29
(1950), p. 1139.
[1950b]
Lhomologie dun espace bre dont la bre est connexe, J. Math. Pures
Appl., (IX), 29 (1950), p. 169213.
Oka (K.)
[1950]
Sur quelques notions arithmetiques, Bull. Soc. Math. France, 78 (1950),

p. 127.
[1951]
Lemme fondamental, J. Math. Soc. Japan, 3 (1951), p. 204214 et 259

278.
Samelson (H.)
[1941]
Beitr
age zur Topologie der Gruppenmannigfaltigkeiten, Ann. of Math.,
45 (1941), p. 10911137.
Sato (M.)
[1959]
Theory of hyperfunctions, I et II, J. Univ. Tokyo Sect. IA, 8, (1959),

p. 139193 et 398437.
[1969]
Hyperfunctions and partial dierential equations, dans Proc. Intern.

Conf. on Functional Analysis and Related Topics, Tokyo 1969, p. 9194.
Sato (M.), Kawai (T.) et Kashiwara (M.)

[1973]
Microfunctions and pseudodierential equations, in Hyperfunctions and

Pseudodierential Equations, vol. 287 des Lecture Notes in Math., Springer, 1979, p. 265529.
` 3
SEMINAIRES
ET CONGRES

DES FAISCEAUX
119
Serre (J.-P.)
[1953]
Quelques problèmes globaux relatifs aux varietes de Stein, dans Colloque

sur les fonctions de plusieurs variables, Bruxelles 1953, Centre belge de
Rech. math., p. 5768.
[1955a]
Faisceaux algebriques coherents, Ann. of Math., 61 (1955), p. 197278.
[1955b]
Un theorème de dualite, Comment. Math. Helv., 29 (1955), p. 926.
Steenrod (N.)
[1943]
Homology with local coecients, Ann. of Math., 44 (1943), p. 610627.
Wang (H.)
[1949]
The homology groups of the ber bundles over a sphere, Duke Math. J.,
16 (1949), p. 3338.
Weil (A.)
[uvres]
uvres scientiques, 3 vol. New York : Springer, 1979.
[1947]
Lettre a` H. Cartan, uvres II, p. 4547.
[1952]
Sur les theorèmes de de Rham, Comment. Math. Helv., 26 (1952), p. 119

145 ; uvres II, p. 1743.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
Le mouvement brownien
Un essai sur les origines de la th
eorie math
ematique
Jean-Pierre Kahane
R
esum
e
Vue de loin, lhistoire du mouvement brownien se divise en deux
periodes : entre 1900 et 1950, une evolution lente, ponctuee par
les travaux dEinstein, Wiener et Levy ; depuis 1950, une eorescence indescriptible. En lexaminant de près, on repère dierents
thèmes et sources presents dès lorigine. On sattache ici surtout
a la première periode, en degageant cinq sources principales, et
`
en survolant les thèmes correspondants au cours de la seconde
periode : 1) Einstein, Wiener et le processus de Wiener 2) Langevin, Doob et les equations dierentielles stochastiques 3) Borel, Steinhaus et les series de fonctions aleatoires 4) Bachelier,
Kolmogorov, les processus et les diusions 5) Pearson, Polya et
les marches au hasard.
Abstract
The paper is a historical survey of the mathematical theory of
Brownian motion, with a particular emphasis on the period 1900
1950, and only short allusions to recent developments. It is organized along ve lines: 1) Einstein, Wiener, and the Wiener process 2) Langevin, Doob, and stochastic dierential equations
3) Borel, Steinhaus, and random series of functions 4) Bachelier, Kolmogorov, processes and diusions 5) Pearson, P
olya,
and random walks.
Si un mathematicien regarde de loin lhistoire du mouvement brownien au

cours de ce siècle, il y verra sans doute deux periodes : entre 1900 et 1950,
AMS 1991 Mathematics Subject Classication: 01A60, 60J65

Math
ematiques; B
at. 425, Universite Paris-Sud, 91405 Orsay Cedex.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
124
J.-P. KAHANE
une evolution lente et lineaire, reperable par les pères fondateurs que furent
Albert Einstein, Norbert Wiener, Paul Levy ; et depuis 1950, une eorescence
dicile à matriser, avec la poursuite des proprietes nes qui font du mouvement brownien lun des prototypes de la fractalite, le mouvement brownien
sur les varietes, le mouvement brownien à plusieurs paramètres, le mouvement
brownien a` la source ou au carrefour des etudes sur les processus gaussiens,
les processus à accroissements independants, les processus de Markov avec
leur lien a` la theorie du potentiel, les martingales, les equations dierentielles
stochastiques, les integrales de chemins, les superprocessus qui decrivent des
particules qui se scindent au cours du temps, etc... La litterature sur le mouvement brownien est facile a` inventorier et meme à lire dans la première periode,
et dicile à matriser dans la seconde ; Daniel Revuz et Marc Yor, dans leur
livre Continuous martingales and Brownian motion [Revuz et Yor 1991] font
etat dune litterature enorme, dont la bibliographie quils donnent, avec 500
titres, ne fournit quune faible idee. Il y a heureusement, sur dierents aspects,
beaucoup de bons livres qui permettent dacceder dans cette foret.
La theorie mathematique du mouvement brownien, mise en place par Norbert Wiener, est à la fois si simple au depart, si belle et si riche quelle a conquis
une large audience chez les mathematiciens et aussi chez les physiciens. Mais
il faut preciser dès maintenant que ce nest quune des idealisations mathematiques du mouvement reel de particules en suspension dans un liquide, tel
quil fut observe et decrit par le botaniste anglais Richard Brown en 1828, et,
a sa suite, par plusieurs physiciens experimentateurs au XIXe siècle. Ce nest
`
meme pas la meilleure idealisation pour lapplication de la theorie dEinstein
a la determination du nombre dAvogadro. Wiener, dailleurs, fut toujours
`
prudent a` cet egard.
Jai choisi de parler surtout de la première periode. Elle est beaucoup
moins lineaire quil y parat dabord. Einstein nest pas la source unique, ni
Wiener le seul canal. Il y a des auents divers, dont on retrouve parfois la
trace dans leorescence contemporaine. Jai identie cinq cheminements, que
je meorcerai de suivre en reperant les croisements et les prolongements dans
la periode contemporaine. Schematiquement, chaque voie est signalee par un
initiateur, un formalisateur et un sujet. Le plan de lexpose est donc ainsi fait :
1. Einstein, Wiener et le processus de Wiener
1 bis. Denitions et commentaires
2. Langevin, Doob et les equations dierentielles stochastiques
3. Borel, Steinhaus et les series de fonctions aleatoires
4. Bachelier, Kolmogorov, les processus et les diusions
5. Pearson, P
olya et les marches au hasard
Un appendice contiendra mes excuses pour tout ce que je naurai pas dit.
` 3
SEMINAIRES
ET CONGRES
LE MOUVEMENT BROWNIEN
1.
125
Einstein, Wiener et le processus de Wiener
En meme temps que Smoluchowski, sur lequel je reviendrai, Einstein publie dans Annalen der Physik trois articles fondateurs pour la theorie du
mouvement brownien, en 1905 et 1906. Je rappelle que les Annalen der Physik de 1905 contiennent egalement les articles dEinstein sur la relativite et
sur leet photoelectrique.
Voici le titre et la conclusion du premier article :
Ueber die von der molekularkinetischen Theorie der Warme gefordete Bewegung von in ruhenden Fl
ussigkeiten suspendierten
Teilchen
(Sur le mouvement, exige par la theorie cinetique moleculaire de la chaleur,
de particules en suspension dans un liquide au repos)
Moge es bald einem Forscher gelingen, die hier aufgeworfene, fur

die Theorie der Warme wichtige Frage zu entscheiden !
(Souhaitons que bient
ot un chercheur parvienne à trancher la question ici
posee, si importante pour la theorie de la chaleur !)
La question posee est celle de lexistence du mouvement indique par le titre.
Einstein en fait la theorie sous lhypothèse de lagitation thermique moleculaire, sans connatre les observations faites sur le mouvement brownien.
Entre le premier et le second article il prend connaissance du mouvement
brownien. La question de lexistence est donc reglee. Il montre alors comment
des mesures faites sur le mouvement des particules peuvent conduire à une
nouvelle determination des dimensions moleculaires, ce que dit bien le titre
uldimensionen .
du second article : Eine neue Bestimmung der Molek
Enn le troisième article, Zur Theorie der Brownschen Bewegung ,
donne une theorie generale, tenant compte de la gravite, et incluant le mouvement brownien de rotation. Einstein a donne ensuite un expose synthetique
sous la forme dun petit livre, publie en 1922, et disponible depuis 1926 en
traduction anglaise.
La formule principale est
(x)2 =
RT 1
N 3a
o`
u R est la constante des gaz parfaits, T la temperature absolue, N le nombre
dAvogadro ( 6 1023 ), la viscosite, a le rayon de la particule, supposee
spherique, et le temps correspondant au deplacement x. Quant a` (x)2 ,
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
126
J.-P. KAHANE
cest une moyenne, pour un intervalle de temps donne , du carre des deplacements, dans une direction donnee, dun grand nombre de particules. Mais
cest aussi la moyenne, pour une suite dintervalles de temps consecutifs, du
carre des deplacements dans une direction dune particule individualisee.
Le programme dEinstein fut realise par Jean Perrin, et publie en 1909
dans les Annales de Chimie et de Physique sous le titre : Mouvement brownien et realite moleculaire. Perrin obtient par ce moyen N 7 1023 , ce
qui conrme les estimations obtenues par dautres procedes. Ce travail devait
lui valoir le prix Nobel en 1926. Dans son article de 1909 et dans son livre de
1912 Les atomes, Perrin decrit eloquemment lextreme irregularite des trajectoires et le fait quapparemment elles nont de tangente en aucun point. Sa
description, dans Les atomes, se conclut par une phrase que Wiener se plaisait
a citer :
`
Cest un cas où il est vraiment naturel de penser a` ces fonctions

continues sans derivees que les mathematiciens ont imaginees, et
que lon regardait a` tort comme de simples curiosites mathematiques, puisque lexperience peut les suggerer.
Dans la theorie dEinstein il apparat aussi que, pour un donne, les
deplacements x ont une distribution gaussienne, et que cette distribution,
fonction du temps et de lespace, satisfait à lequation de diusion de la chaleur. Les mathematiciens, Wiener le premier, ont retenu de lequation dEinstein la proportionnalite de (x)2 et de = t, et le fait que les x sont
gaussiens. Lequation, normalisee, secrit alors
(x)2 = t
ou, en explicitant les valeurs du temps, et en utilisant le symbole E() de
lesperance au lieu de surligner pour la valeur moyenne,

E (Xt Xs )2 = |t s|.
Aujourdhui (disons, depuis Wiener, Steinhaus et Kolmogorov) lesperance
nous apparat comme une integrale sur un espace de probabilite (, A, P ), Xt
signie Xt () ( ), et lequation dEinstein signie que, dans lespace
L2 (), le point Xt decrit ce que I. Schoenberg a appele plus tard une helice,
cest-à-dire une courbe qui glisse isometriquement sur elle-meme quand on
translate le temps. Cest une très jolie helice, o`
u le carre de la distance de
deux points est la distance des paramètres et o`
u par consequent, daprès le
theorème de Pythagore, trois points quelconques sont toujours les sommets
` 3
SEMINAIRES
ET CONGRES
127
dun triangle rectangle. Selon Einstein, le processus est gaussien et centre,

cest-à-dire que lhelice se trouve dans un sous-espace H de L2 ()
e de
constitu
u2
variables gaussiennes centrees (ce qui signie E(exp uX) = exp 2 EX 2 ), et
dans un tel espace lorthogonalite equivaut a` lindependance. Conformement
au sens physique, les accroissements de Xt sur des intervalles disjoints de
temps sont des v. a. independantes ; on dit que t Xt est un processus à
accroissements independants. Au stade o`
u nous laisse Einstein, seule la realite
physique du mouvement brownien fonde lexistence de lhelice brownienne que
je viens de decrire.
Wiener entre en scène bien plus tard, en 1923, avec un article fondamental intitule Dierential space . Il connat la theorie dEinstein depuis sa
visite à Cambridge en 1913 ; il etait venu, a` 19 ans, etudier la logique avec
Bertrand Russell, mais Russell lui avait suggere daller ecouter Hardy et de
lire Einstein. Dans lintervalle, il a etudie lintegrale de Daniell, et sest interesse à lintegration dans des espaces de fonctions. Son idee de base est de
construire sur lespace des fonctions continues reelles sur R+ , C(R+ ), une mesure de probabilite telle que les accroissements sur des intervalles de temps
disjoints aient la distribution gaussienne prevue par la theorie dEinstein. Ces
accroissements sont des dierences, do`
u le titre de Dierential space.
La mesure ainsi construite sappelle justement mesure de Wiener et lintegrale par rapport a` cette mesure moyenne de Wiener. Une fois eectuee la
construction, Wiener intègre des fonctionnelles diverses. Il verie quen tout
point t donne la probabilite de derivabilite en t est nulle (ce qui, contrairement
a certains commentaires, est encore loin de prouver que la non-derivabilite
`
partout est presque s
ure), puis il etablit que la probabilite de verier une
condition de H
older dordre 12 sur un intervalle donne est egale à 1. Ainsi,
la mesure de Wiener est concentree sur des fonctions holderiennes. Enn,
Wiener donne la loi des coecients de Fourier. Sur lintervalle (0, 2), cela
permet de developper une fonction nulle en 0 sous la forme
n (1 cos nt) + sin nt

t
n
Xt = 0 +
n
2
1
o`
u 0 , 1 , 1 , est une suite de variables gaussiennes normalisees (E = 0,
E 2 = 1) et independantes, ce que jappellerai un echantillon normal.
Cest la serie de Fourier-Wiener, encore implicite en 1923, explicitee à
loccasion de la collaboration avec Paley et Zygmund en 1933.
Dans une etude ecrite en 1964 sur Wiener et lintegration dans les espaces
fonctionnels, Marc Kac met en evidence la profonde originalite de Wiener
et, en contre partie, la diculte queurent les mathematiciens de lepoque a`
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
128
J.-P. KAHANE
comprendre sa demarche. Only Paul Levy in France, who had himself been
thinking along similar lines, fully appreciated their signicance.
Letape suivante est en eet luvre de Paul Levy sur le mouvement brownien, qui setend de 1934 a` 1966 et comprend le premier grand ouvrage sur
la question, Processus stochastiques et mouvement brownien, en 1948. Nous
allons a` plusieurs reprises rencontrer des theorèmes de Paul Levy, mais je
serai loin de rendre compte de toute la richesse de son apport.
1.bis D
enitions et commentaires
Jai parle au depart de la simplicite du processus de Wiener. Cette simplicite nest apparue quau cours du temps. La presentation quen fait Wiener
dans son livre avec Paley, Fourier transforms in the complex domain [Paley
et Wiener 1934] est bien plus accessible que celle de Dierential space , et
jy reviendrai. Voici comment se presente la chose aujourdhui.
Il sagit dabord de construire une helice ayant la geometrie voulue dans
un espace de Hilbert ; dans L2 (R),
(t R)
t 1[0,t]
fait laaire ([0, t] designant lintervalle joignant 0 et t). Soit W une isometrie
lineaire de L2 (R) sur un espace de Hilbert gaussien, H, et
Xt = W (1[0,t] ).
Alors Xt decrit une helice brownienne. Soit (un ) une base de L2 (R), et (n )
son image dans H par W ; cest un echantillon normal. Dans L2 (R) on peut
decomposer 1[0,t] suivant la base (un ) :
1[0,t] =
donc
Xt =
an (t)un
an (t)n
dans L2 (R)
dans H.
Cela donne une version explicite du processus de Wiener sous la forme

Xt () =
an (t)n ()
et reduit son etude a` celle dune certaine serie de fonctions aleatoires. Pratiquement, on se borne souvent à etudier le mouvement brownien sur R+
` 3
SEMINAIRES
ET CONGRES
129
ou sur lintervalle [0, 1], en supposant toujours X0 = 0. Alors (un ) designe

unebase dansL2 (R+ ) ou L2 ([0, 1]). En choisissant la base trigonometrique
(1, 2 cos 2t, 2 sin 2t, ) dans L2 ([0, 1]), on obtient la serie de FourierWiener. En choisissant la base de Haar, les an (t) sont des fonctions triangles
portees par des intervalles dyadiques et letude de la serie est assez facile.
Loperateur W peut se denir sur L2 (E) pour tout espace mesure E. Cest
la denition mathematique du bruit blanc sur E [Kakutani 1961]. Etant donne
f L2 (E), on peut ecrire

W (f ) = f dW.
Cest lintegrale de Wiener sur E. Avec cette notation, il est naturel decrire
Wt pour le processus de Wiener sur R. Mais il est aussi interessant de consideu est une mesure temperee ; en se restreignant
rer le bruit blanc sur (Rd , ), o`
aux f D(Rd ), le bruit blanc sidentie alors a` une distribution temperee
aleatoire sur Rd . Si est une probabilite, on peut se representer W comme
une limite de mesures discrètes aleatoires de la forme
N
1
Xn
N 1
o`
u les Xn sont des v. a. dans Rd independantes ayant toutes pour distribution, et les sont choisis au hasard selon la probabilite naturelle, ou aussi
bien de mesures de la forme
N
1
(Xn ).
N 1
La rapidite de la convergence de ces series vers W, testee sur une classe convenable de fonctions f, par exemple les fonctions indicatrices de produits dintervalles, est lun des problèmes actuels de la statistique.
Il y a dautres helices dans H que lhelice brownienne, et elles ont toutes des
interpretations probabilistes interessantes. Voici quelques exemples, introduits
par I. Schoenberg et J. von Neumann vers 1940. Il sera commode de prendre
ici pour L2 et H des espaces de Hilbert complexes.

Soit une mesure positive sur R telle que |sin ut|2 (du) < . Alors
t eiut 1 denit une helice dans L2 (R, (du)) ; en eet,
(eiut 1) (eius 1) 2L2 (R,) = (t s)

avec
(t) =
|eiut 1|2 (du).

E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
130
J.-P. KAHANE
Cest la forme generale des fonctions dhelice sur R ; on les appelle aussi,
suivant A. Beurling, fonctions denies negatives ou fonctions de type negatif.
En choisissant convenablement, on obtient (t) = |t| , 0 < < 2. Par
isometrie de L2 (R, ) sur H, ici complexie, on a un processus (Xt ) tel que
X0 = 0 et
Xt Xs 2H = |t s| .
Pour = 1, on retrouve le mouvement brownien. Pour les autres , on dit
que (Xt ) est un mouvement brownien fractionnaire.
Au lieu de R comme espace des paramètres, on peut partir de nimporte
quel groupe abelien localement compact G ; une helice est denie par une
application de G dans un espace de Hilbert telle que le carre de la distance de
deux points de lhelice soit une fonction (la fonction dhelice) de la dierence
des paramètres.
2
n ) pour espace de Hilbert, o`
u n est une suite
Prenons G = T
et (Z,
2
positive telle que
n sin nt < pour tout t. Alors t e2int denit une
helice dont la fonction dhelice est

(t) =
n |e2int 1|2 .
Par isometrie de 2 (Z, n ) dans H, on obtient la forme generale dun processus
gaussien stationnaire 1-periodique, et son expression sous forme de serie de
Fourier aleatoire

n n ()e2int
Xt =
(n ) etant ici, par commodite, un echantillon normal complexe.
Prenons enn G = Rd , L2 (Rd , ) comme espace de Hilbert, et

t eiut 1 L2 (Rd , (du))
pour parametrage dhelice (ut = u1 t1 + u2 t2 + + ud td ) ; la fonction dhelice
est, comme plus haut,

(t) = |eiut 1|2 (du).
On peut choisir (t) = |t|, distance euclidienne : on obtient le brownien a` d
paramètres de Paul Levy. Pour (t) = |t| , 0 < < 2, cest le brownien
fractionnaire dindice `
a d paramètres.
Tous ces processus ont ete bien etudies et sont faciles à simuler. Les derniers, par exemple, ont ete utilises graphiquement par Benot Mandelbrot
pour produire des reliefs articiels.
` 3
SEMINAIRES
ET CONGRES
131
Un problème theorique important est de determiner, en fonction de la

geometrie de lhelice, sil existe ou non des versions du processus continues
presque s
urement. Il a ete resolu par R. Dudley [1967] et X. Fernique [1975],
et leur solution a servi de base aux travaux ulterieurs de Michael Marcus et
de Gilles Pisier, dont je parlerai ensuite.
Plus generalement, etant donne un ensemble E dans H, espace de Hilbert
gaussien, existe-t-il ou non une version p. s. continue du processus correspondant ? Le parametrage de E na plus dimportance, seule sa geometrie
intervient. Le problème est attribue à Kolmogorov, et il semblait inaccessible
dans sa generalite quand Michel Talagrand, en 1990, lui a donne une solution
facile a` formuler et immediatement celèbre.
Le processus de Wiener a deux faces. Dun c
ote cest un processus gaussien ; cest la face que nous venons de parcourir a` grandes enjambees. De lautre
cest un processus a` accroissements independants et stationnaires. Cela veut
dire que la loi des accroissements Xt Xs ne depend que de t s, et que
les accroissements Xt1 Xs1 , Xt2 Xs2 , Xt3 Xs3 , etc... sont independants
quand les intervalles ouverts (s1 , t1 ), (s2 , t2 ), (s3 , t3 ) etc... sont disjoints. En
supposant X0 = 0, la loi du processus est donc bien denie par celle des Xt
(t R+ ), donc par leur fonction caracteristique, quon voit facilement etre de
la forme

E eiuXt = et(u) .
2
Le processus de Wiener correspond à (u) = u2 . La fonction la plus generale

est donnee par la formule de Levy-Khintchine

u2
(u) = au + b + (eiuy 1)d1 (y) + (eiuy 1 iuy)d2 (y)
2
dont les quatre termes font apparatre la derive deterministe (drift), le processus de Wiener, des processus de Poisson et des processus de Poisson avec
derive. Le livre de Paul Levy sur laddition des variables aleatoires [Levy
1937] en fait une presentation très parlante. Curieusement, dans le cas reel,
les fonctions qui apparaissent ici sont les memes que dans le cas gaussien
(o`
u (t) = t2 correspond a` un processus degenere) ; ainsi, elles ont ete introduites et decouvertes independamment, pour des sujets dierents, par Levy et
Khintchine, par von Neumann et Schoenberg [1941], et par Beurling et Deny
[1959].

u
|u| , 0 < < 2, a beaucoup danalogies et de liens
Le cas (u) = C |u|
avec le processus de Wiener ; les processus correspondants ont ete soigneusement etudies par Levy et on les appelle processus de Levy stables dindice
; leurs versions sont p. s. discontinues. Lorsque la mesure 1 est portee par
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
132
J.-P. KAHANE
R+ et que a = b = 0, la mesure 2 etant nulle, le processus est croissant. Le

processus de Levy croissant dindice 12 joue un r
ole essentiel dans la theorie
du mouvement brownien ; ladherence de lensemble de ses valeurs a meme
loi que lensemble des zeros du brownien. Dans son livre de 1948, Paul Levy
etablit ce fait et part de l`
a pour donner une nouvelle et très interessante
construction du mouvement brownien : on met en place lensemble des zeros,
puis, sur chaque intervalle contigu, une excursion brownienne qui part de 0 au
debut de lintervalle et y revient a` la n ; si on les normalise en les ramenant à
un intervalle de temps unite, ces excursions ont toutes la meme loi, elles sont
independantes entre elles, et independantes de lensemble des zeros.
2.
Langevin, Doob et les

equations di
erentielles
stochastiques
A le meme epoque quEinstein, en 1906, Marian Smoluchowski publiait,

dans le Bulletin de lAcademie des Sciences de Cracovie et dans Annalen
der Physik, une theorie du mouvement brownien tout a` fait analogue [Smoluchowski 1906a,b]. La seule dierence etait un facteur 64
27 dans lexpression
2
de (x) . En meme temps que Jean Perrin en France exploitait la formule
dEinstein, le physicien suedois Theodor Svedberg, futur prix Nobel lui aussi
(1926), partait de la formule de Smoluchowski et croyait la valider [Svedberg
1907].
En 1908, Paul Langevin publie une courte note aux Comptes rendus de
lAcademie des sciences. Il dit dabord que lapproche de Smoluchowski, rectication faite, mène à la formule dEinstein. Puis, en tout petits caractères, il
expose lumineusement, suivant sa propre approche, lensemble de la theorie.
Comme peu de mathematiciens connaissent cette presentation, la voici, en
detail.
Dabord, contrairement a` lidealisation que nous venons de voir, les particules browniennes ont une vitesse, soit, dans la direction 0x,
u=
dx
.
dt
Si leur masse est m, leur energie moyenne est 12 mu2 . Selon lhypothèse fondamentale de la mecanique statistique cette energie moyenne est egale à lenergie
cinetique moyenne dune molecule, soit RT
2N :
mu2 =
` 3
SEMINAIRES
ET CONGRES
RT
.
N
133
Pour une particule spherique de rayon a et masse m, dans un liquide de

viscosite , lequation du mouvement est a priori
m
du
= 6au ;
dt
la resistance visqueuse arrete le mouvement en un temps très court. Mais il y a

des uctuations dans les chocs moleculaires, qui entretiennent le mouvement.
Lequation que propose Langevin est donc
(L)
du
= 6au + X,
dt
en ajoutant que sur la force complementaire X nous savons quelle est

indieremment positive et negative, et sa grandeur est telle quelle maintient
lagitation de la particule.
De lequation (L) Langevin tire
m
d(xu)
= mu2 6axu + xX
dt
puis, en prenant les valeurs moyennes et en admettant que xX = 0,

m
RT
dxu
=
6axu
dt
N
ce qui donne

dx2
2xu =
dt

RT 1
6a
+ C exp
t
=
N 3a
m
et, compte tenu des valeurs numeriques, la dernière exponentielle est negligeable pour t 108 sec. ; au bout de ce temps, le regime est pratiquement
permanent. Donc, pour 108 sec.,
x2t+ x2t =
RT 1
.
N 3a
Sous lune des deux hypothèses, daccroissements orthogonaux ou daccroissements stationnaires, on obtient
(xt+ xt )2 =
RT 1
,
N 3a
la formule dEinstein.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
134
J.-P. KAHANE
Physiquement, la vitesse des particules existe, mais, pour les observables,

le mouvement correspond au modèle dEinstein.
Mathematiquement, il restait a` denir une fonction u(t, ) et un X()
veriant (L).
Dautres physiciens que Langevin se sont interesses après lui, mais independamment semble-t-il, à la distribution des vitesses du mouvement brownien : cest le cas de Fokker en 1914 et Planck en 1917, puis de G. E. Uhlenbeck
et L. S. Ornstein en 1930, qui ont donne respectivement leurs nom à lequation de Fokker-Planck et au processus dOrnstein-Uhlenbeck (cf. le livre de
N. Wax [1954]). Dautres travaux de physique theorique ont ete menes par
Francis Perrin, le ls de Jean Perrin, sur le mouvement brownien de rotation [Perrin 1928] et, en suivant lapproche de Langevin, sur le mouvement
brownien dun ellipsode [Perrin 1934-36].
La première theorie mathematique de lequation de Langevin (L) a ete
donnee par J. L. Doob [1942]. Cest lun des actes de naissance des equations
dierentielles stochastiques (lautre, la meme annee, vient de K. Ito [1942]).
Voici le programme de Doob : a stochastic dierential equation will be introduced in a rigorous way to give a precise meaning to the Langevin dierential
equation for the velocity function dx(s)
ds .
Lidee de Doob est de prendre pour X le bruit blanc. Lequation de Langevin secrit alors
dU = U + dW,
W etant le processus de Wiener, soit

t
U (t)e
es dW (s)
et lintegrale de Wiener au second membre donne un autre processus de Wiener avec changement de temps, soit
U (t)et =
En choisissant =
1
2
1
W1 (e2t ) + C te .
2
et la constante nulle, on a
U (t) = e 2 W1 (et )
t
o`
u W1 est un processus de Wiener. On voit facilement que U (t) est un processus gaussien stationnaire ; les mathematiciens lutilisent souvent sous le nom
de processus dOrnstein-Uhlenbeck comme une sorte de cousin du mouvement
brownien. Par exemple, le fait que le changement de t en t conserve la loi
` 3
SEMINAIRES
ET CONGRES
135
du processus
signie, pour le processus de Wiener, que
dOrnstein-Uhlenbeck
1
1
W
(t)
et
t
W
(
)
ont
la
m
e
me
loi,
facon de transporter a` lorigine ce quon
t
t
sait du comportement a` linni. Mais lorigine physique est bien dierente :
W (t) idealise la trajectoire dune particule brownienne et U (t) sa vitesse, et
les deux idealisations sont incompatibles.
Il y aurait pour X bien dautres choix. Si par exemple on prend pour X
un processus dOrnstein-Uhlenbeck U0 , lequation de Langevin prend la forme
dU1
= U1 + U0
dt
et elle admet une solution U1 de classe C 1 . En iterant, on peut obtenir des
modèles de mouvement brownien de classe C k et meme de classe C .
On pourrait aussi renoncer a` un modèle gaussien, en prenant pour X
une distribution (au sens de Schwartz) aleatoire soumise à la seule condition
E(xX) = 0. Un modèle poissonien serait sans doute plus près de la realite
physique.
Des equations du type de Langevin se trouvent maintenant un peu partout
en physique et en mathematiques. Voici, par exemple, la dynamique de Langevin telle quelle apparat dans les methodes de determination de minimum
absolu pour une fonction dierentiable g : Rk R (on doit se representer k
comme très grand, et cest une question danalyse numerique importante et
dicile). Une methode de descente ordinaire ferait tomber dans un trou local. La methode, empruntee aux chimistes et dite de recuit simule (simulated
annealing), est une descente bruitee [Geman et Hwang 1986], suivant le processus x(t) a` valeurs dans Rk , solution de lequation dierentielle stochastique
dx(t) = g(x(t))dt + (t)dW
o`
u dW est le bruit blanc sur Rk muni de la mesure de Lebesgue, et (t) une
fonction quon souhaite
faire tendre vers 0 aussi vite que possible. On peut

1
choisir (t) = O log t (R. Azencott [1992]).
Les problèmes physiques continuent a` alimenter les recherches actuelles sur
le mouvement brownien, et lintuition des physiciens conduit a` des problèmes
mathematiques serieux. Pour en avoir une idee, je renvoie à letude de Hans
F
ollmer [1984] et a` lexpose de B. Duplantier [1989] a` la Journee SMF de
1989.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
136
3.
J.-P. KAHANE
Borel, Steinhaus et les s

eries de fonctions
al
eatoires
En 1896, alors que le prolongement analytique des series de Taylor etait
un sujet a` lordre du jour, Emile

Borel publia une note aux Comptes rendus
de lAcademie des sciences au contenu provocateur : en general pour une serie
de Taylor, dit-il, le cercle de convergence est une coupure [Borel 1896]. En
1897, a` linvitation de Mittag-Leer, il donna une ebauche de demonstration
dans Acta Mathematica. De facon remarquable, cette ebauche contient dejà
le theorème de Borel-Cantelli. Mais les concepts fondamentaux qui permettraient de transformer les intuitions en enonces et lebauche en demonstration,
cest-à-dire la probabilite totalement additive et lindependance, font encore
defaut. Cest en 1898 que Borel introduit la mesure totalement additive, et
` lire la notice de Borel
en 1909 ce quil appelle probabilites denombrables. A
de 1912, on a limpression que cet enonce provocateur sur les series de Taylor a ete le declencheur de ses recherches ulterieures en probabilites. Mais ce
nest quen 1929 quun sens clair lui a ete donne, par Hugo Steinhaus [1930],
avec une vraie demonstration. Avant den venir l`
a, bien des etapes furent
necessaires. Voici les principales.
En 1902, cest lapparition, dans sa thèse, de la mesure de Lebesgue et
en 1906, dans ses Lecons sur les series trigonometriques, la denition et les
proprietes de lintegrale comme on la connat aujourdhui ; il sagit uniquement
de fonctions reelles dune variable reelle, denies sur un intervalle.
En 1922, Rademacher introduit son système de fonctions à valeurs 1,
denies sur lintervalle [0, 1] a` lexception des points dyadiques par
rn (t) = (1)tn
si t = 0, t1 t2
en ecriture dyadique. Cest un système orthonormal dans L2 ([0, 1]), et aussi

le prototype dun syst`
eme de fonctions
independantes. Rademacher montre

2
an rn (t) converge presque
que, sous lhypothèse
an < , la serie
partout.
En 1925,
Khintchine
et
Kolmogorov
donnent
la
r
e
ciproque
:
si
a2n = ,

la serie
an rn (t) diverge p. p.. Cest le debut des travaux sur les series de
variables aleatoires independantes.
Dans lintervalle, en 1923, Steinhaus reduit la theorie des probabilites denombrables a` celle de la mesure de Lebesgue sur [0, 1]. Les variables aleatoires
sinterprètent comme fonctions mesurables sur cet intervalle.
En 1929, Steinhaus donne son modèle dune suite de variables aleatoires
independantes. Au point t = 0, t1 t2 t3 il associe la suite
1 = 0, t1 t3 t5
` 3
SEMINAIRES
ET CONGRES
137
2 = 0, t2 t6 t10
3 = 0, t4 t12 t20
Ce sont des v. a. independantes, toutes distribuees sur [0, 1] selon la mesure

de Lebesgue. Dans ce cadre il etablit ce qui sera plus tard la loi du zero-un de
Kolmogorov : une propriete asymptotique de la suite (n ) a necessairement
pour probabilite 0 ou 1. Enn il montre que pour une serie de Taylor aleatoire
de la forme

cn e2in z n
lim c1/n
=
1
n
1
le cercle de convergence |z| = 1 est une coupure ; cest la première fois quest
mise en forme lintuition de Borel de 1896.
Linteret de lapproche de Steinhaus va bien au-del`
a. En fait, un modèle
universel pour une suite de variables aleatoires independantes est constitue
u les fn sont des fonctions mesurables denies sur [0, 1]. Avant
par (fn (n )) , o`
laxiomatique de Kolmogorov, il y a l`
a un fondement solide pour toute la
theorie des probabilites.
Après Steinhaus, letude des series de fonctions aleatoires fait un nouveau
bond avec Paley et Zygmund [1930, 1932]. Paley et Zygmund se posent la
question detendre le theorème de Borel-Steinhaus aux series de Taylor

cn rn (t)z n
cn z n
=
1

et,
plus generalement, detudier des series de fonctions de la forme
fn
2i
n
ou
e
fn . A cote des series de Taylor, ils etudient des series de Fourier
aleatoires du type de Rademacher ou de Steinhaus :
(R) :
cn eint ,
(S) :
cn e2in eint
(jecris des exponentielles imaginaires pour abreger

lecriture ; ils prefèrent
ecrire des developpements en cosinus et sinus). Si
|cn |2 = , il savère
quep.s. (R) et (S) ne sont pas des series de Fourier-Lebesgue. Au contraire,
si
|cn |2 < , elles representent des fonctions aleatoires qui appartiennent
p. s. a` Lp ([0, 2]) pour tout p < . A quelle condition sur la suite (cn )
representent-elles p. s. des fonctions bornees, continues ? Paley et Zygmund
donnent des conditions necessaires et des conditions susantes, et leurs recherches seront poursuivies par Salem et Zygmund, Kahane, et Pierre Billard ;
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
138
J.-P. KAHANE
letat de la question en 1968 se trouve dans mon livre Some Random Series
of Functions (paraphrase du titre de la serie darticles de Paley et Zygmund).
En 1933 les trajectoires de Wiener et de Paley et Zygmund se rencontrent.
On reconnat la main de Wiener dans lintroduction a` leur article commun. Il
explique comment, independamment de la consideration des series (R) et (S)
et de leur usage pour des contre-exemples, il a ete amene par le mouvement
brownien a` des series trigonometriques gaussiennes, que jecrirai
(G) :
cn n eint
o`
u (n ) est un echantillon normal complexe. Il pose la question dun traitement commun aux series (R), (S) et (G), et larticle etend en eet aux series
(G) des theorèmes obtenus par Paley et Zygmund pour les series (R) et (S).
Le sujet devait etre reconsidere par Kahane et par Billard dans les annees
1950 et 1960. Mais ce nest quen 1978, a` la suite des travaux de Dudley et de
Fernique sur les processus gaussiens, quil allait etre complètement elucide par
Marcus et Pisier : chaque propriete telle que convergence ou sommabilite dans
un Lp , convergence uniforme, convergence ponctuelle en tout point, convergence ponctuelle presque partout, appartenance a` un Lp ou continuite de la
fonction representee, a la meme probabilite, 0 ou 1, pour les series (R), (S) et
(G) ayant les memes coecients cn . Connaissant (par Dudley-Fernique) des
conditions sur (G) pour la continuite, on a donc des conditions necessaires et
susantes pour la continuite des fonctions representees par (R) ou (G). Les
travaux de Marcus et Pisier, exposes dans leur livre de 1981, concluent donc
ce programme de Wiener. Mais ils sont prolonges de facon vigoureuse par les
etudes actuelles sur probabilites et espaces de Banach, quon trouve dans le
livre de Michel Ledoux et Michel Talagrand [1991].
De sa frequentation de Paley et Zygmund, Wiener retient la simplicite de
lapproche probabiliste de Steinhaus. Jusqualors, il a deduit la loi des coecients de la serie de Fourier-Wiener de sa construction de la mesure de Wiener
sur lespace des fonctions continues dune variable reelle. A partir de 1933, il
part comme Steinhaus de lintervalle [0, 1] muni de la mesure de Lebesgue, il
transporte cette mesure sur [0, 1]N au moyen de la transformation de Steinhaus, il denit des echantillons normaux comme fonctions des coordonnees
n , et il transporte enn la mesure sur les fonctions continues au moyen de
la serie de Fourier-Wiener. Le point de vue est complètement inverse. Dans
le livre de 1934 de Paley et Wiener les deux derniers chapitres sont consacres au processus de Wiener introduit de cette manière, et designe comme
fundamental random function.
` 3
SEMINAIRES
ET CONGRES
139
Or, tandis que Wiener adoptait le point de vue de Steinhaus, Kolmogorov

etait en train dadopter et de systematiser le point de vue de Wiener. Ainsi,
au moment meme o`
u Wiener etait acquis au modèle probabiliste constitue
par lintervalle reel [0, 1] muni de la mesure de Lebesgue, Kolmogorov donnait son axiomatique et etablissait ses theorèmes de prolongement pour la
construction despaces de probabilites adaptes aux processus à etudier. A cet
egard, les Grundbegrie ont enterre Paley-Wiener, mais Paley-Wiener reste
une reference cle pour le mouvement brownien.
Dans larticle de PaleyWienerZygmund [1933] se trouve la première demonstration de la non-derivabilite partout du processus de Wiener, et elle se
trouve reproduite, longue citation de Jean Perrin a` lappui, dans le livre de
Paley et Wiener [1934]. De plus, en 1934, Wiener sait que p. s. le processus
X() verie pour tout > 0
1
(t [0, a], h 0)
uniformement X(t + h) X(t) = O |h| 2
1
(h 0)
pour tout t X(t + h) X(t) = |h| 2 +
( signie le contraire de o).
Voici ce quon sait de plus aujourdhui
en fait de proprietes presque s
ures :

1
|h|log |h| , et la nonderivabilite forte en
le module de continuite est en O

( |h|). Plus precisement, on a
|X(t + h) X(t)|

1
1
h0
2|h|log |h|
|X(t + h) X(t)|

c>0
lim sup
|h|
h0
uniformement
lim sup
pour tout t
[Levy 1937]
[Dvoretzky 1963].
De plus, on a en chaque point t xe, donc (Fubini) presque partout

X(t + h) X(t)
=1;
lim sup
h 0
2 h log log h1
cest la loi du logarithme itere de Khintchine 1927 pour h , traduite
par P. Levy au voisinage de t en 1940. Ces resultats sont essentiellement
inameliorables. Il existe des points rapides, o`
u
lim sup
h0
|X(t + h) X(t)|

>0
1
2|h|log |h|
[Orey et Taylor 1974]
et des points lents, o`

u
lim sup
h0
|X(t + h) X(t)|

<
|h|
[Kahane 1974].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
140
J.-P. KAHANE
Naturellement, les points rapides et les points lents constituent un ensemble

de mesure nulle, puisque presque partout on a la loi du logarithme itere à
gauche et à droite.
Ce comportement erratique constitue lun des charmes des trajectoires
du mouvement brownien. Lhelice brownienne, en regard, a une majestueuse
regularite : son equation de denition,
Xt+h Xt 2H =
|h|
signie quelle est uniformement holderienne dordre 12 . Lhelice brownienne

donne donc une première idee, grossière, du comportement (presque s
ur) des
trajectoires, et letude ne fait apparatre des dierences. Il en est de meme
avec la dimension de Hausdor : lhelice a pour dimension 2, et la mesure
2-dimensionnelle sur lhelice concide avec le temps quon y passe. Dans un
espace de dimension 2, le mouvement brownien (dont les composantes sont
par denition des mouvements browniens lineaires independants) a des trajectoires de dimension 2 mais de mesure 2-dimensionnelle nulle [Levy 1948], et
cest un exercice de virtuosite que de decouvrir la fonction determinante, h(x),
selon laquelle la mesure de Hausdor des trajectoires browniennes planes est
nie et non nulle [Taylor 1964] ; on trouve h(x) = x2 log x1 log log log x1 .
Il en est encore de meme avec la variation quadratique. On a identiquement

Xt Xt 2 =
|ti+1 ti | = b a
i+1
i H
pour toute decomposition dun intervalle [a, b] au moyen des points ti (a =
t0 < t1 < < tn = b), donc la variation quadratique sur lhelice concide
avec la variation du paramètre. Pour les trajectoires, il est egalement vrai que,
pour une suite de partages emboites de [a, b] dont le pas tend vers 0, on a p.s.
lim
|X(ti+1 ) X(ti )|2 = b a
(Wiener [1923] ; Paul Levy [1948]), mais il est faux que cela ait lieu pour une
trajectoire donnee ; en fait, la limite superieure est innie. On peut chercher
pour quelles fonctions h(x) il est presque s
ur que, pour toute suite de partages
de [a, b] dont le pas tend vers 0, on ait
lim sup
h (|X(ti+1 ) X(ti )|) < ;
la meilleure fonction h(x) est x2 /log log x1 [Taylor 1972].
` 3
SEMINAIRES
ET CONGRES
141
Dans les espaces euclidiens Rd il ny a pas dhelice brownienne mais seulement des quasi-helices veriant
1 d Xt+h Xt 2 /|h| 1 + d .
Elles sont faciles à construire par un procede automatique, et leurs projections
planes ressemblent au mouvement brownien plan. Cest un des très nombreux
cas o`
u un automate convenable imite le hasard ([Assouad 1980], [Kahane
1981]).
La methode de Paul Levy pour evaluer le module de continuite du brownien est au moins

aussi importante que le resultat. Elle consiste a` determiner
h
la loi de X t + 2 lorsque X(t) et X(t + h) sont connus. La methode lui a
servi pour dautres processus a` accroissements independants, et cest la cle
pour certaines simulations numeriques [Bouleau et Lepingle 1994].
Je reviens à Borel pour terminer cette section. Son idee est quun phenomène inhabituel, tel que le non-prolongement dune serie de Taylor, peut
apparatre comme general dans un cadre convenable. Cette idee a ete exploitee dans des cadres varies : ensembliste, et à cet egard Cantor, sur les
nombres transcendants, avait montre la voie ; topologique, avec lutilisation
de la theorie de Baire, et cela a ete une specialite polonaise des annees 1920 ;
et, naturellement, probabiliste comme y pensait Borel. Il peut etre très dicile
de construire certains objets, tandis quune mesure de probabilite convenable
les fait apparatre en masse. Paley et Zygmund ont ete les initiateurs de cette
methode, et mes livres de 1968 et 1985 en donnent beaucoup daspects ; le
prototype est la reciproque du theorème de Riesz-Fischer : si la suite des amplitudes rn nest
pas de carre sommable, peut-on choisir les phases n de sorte
que la serie
rn cos(nt + n ) ne soit pas une serie de Fourier-Lebesgue, et
comment ? Reponse : oui, au hasard. Pour certains usages, il simpose de penser au mouvement brownien. Je me borne à un exemple. D. Mensov a montre
que toute fonction continue sur T est egale, sauf sur un ensemble de mesure
arbitrairement petite, à une fonction de la classe U (T), cest-à-dire dont la
serie de Fourier est uniformement convergente. Peut-on remplacer U (T) par
A(T), la classe des fonctions dont la serie de Fourier est absolument convergente ? La reponse est negative, comme la montre Y. Katznelson, meme si on
impose à la fonction donnee detre holderienne (A. M. Olevskii). Le meilleur
contre-exemple connu est le mouvement brownien : sur aucun ensemble de
mesure positive la fonction de Wiener nest egale à une fonction de la classe
A(T) [Hruscev et al. 1981].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
142
4.
J.-P. KAHANE
Bachelier, Kolmogorov, les processus et les

diusions
En 1900, Louis Bachelier soutint une thèse de doctorat sur la theorie de

la speculation o`
u lon retrouve aujourdhui les ingredients principaux de la
theorie du mouvement brownien. Bachelier, dans le monde mathematique,
etait un outsider. Orphelin a` 15 ans, commis de commerce, petit operateur
a la Bourse de Paris, il avait fait des etudes dans des conditions diciles. Il
`
avait suivi, a` la Sorbonne, les lecons dHenri Poincare, alors charge du cours de
probabilites. Il reconnat dans sa thèse le caractère aleatoire de la uctuation
des cours en Bourse et se propose den faire une theorie en vue des previsions
a operer pour les marches à terme.
`
Bachelier manipule avec audace des probabilites conditionnelles (quil appelle connexes) et des probabilites composees. Après avoir centre son processus
de uctuation, il designe par px,tdx la probabilite pour que, partant de 0, il
se trouve au temps t dans lintervalle (x, x + dx), et il etablit (p. 19) que

pz,t1 +t2 = px,t1 pzx,t2 dx,
ce qui est la formule de base des processus de Markov (dont lintroduction
par Markov pour des chanes, cest-à-dire des processus a` temps discret, date
de 1907). Puis il montre que
x2
1
p = e t
t
est une solution, et quelle satisfait lequation de la chaleur. Cela lautorise à

parler du rayonnement ou de la diusion de la probabilite (p. 30). Il en tire
des règles pratiques : la loi des ecarts
33), et la loi de la fonction
de prime (p.
que Paul Levy designera par M (t) = sup X(s) , ainsi formulee :
st
la probabilite pour quun cours soit atteint ou depasse à lepoque

t est la moitie de la probabilite pour que ce cours soit atteint ou
depasse dans lintervalle de temps t. (p. 59)
Cependant Bachelier fut meconnu. Son histoire est excellemment racontee
par Benot Mandelbrot dans The Fractal Geometry of Nature. Cest Kolmogorov en 1931, dans un article des Mathematische Annalen, qui sortit son nom
de loubli. Larticle est consacre aux processus de Markov a` temps continu,
dont le processus du mouvement brownien est un exemple. La loi dans lavenir
` 3
SEMINAIRES
ET CONGRES
143
ne depend que de letat present, donc la loi du processus est donnee par les
probabilites de transition que, plus explicitement que Bachelier, Kolmogorov
designe par P (t0 , x, t, y) (on part de x au temps t0 , on se trouve en y au temps
t). Kolmogorov fait une etude analytique de la loi du processus, en etudiant
les equations dierentielles ou integrodierentielles veriees par les probabilites de transition ; comme il le remarque lui-meme [Kolmogorov, Papers II,
p. 521, note], la theorie est developpee en termes classiques et sans utiliser les
espaces de trajectoires. Il se refère à Bachelier comme initiateur : Bachelier,
dit-il, fut le premier a` faire une etude systematique du cas o`
u la probabilite
P (t0 , x, t, y) depend contin
ument de t.
Comme Bachelier lavait reconnu dès le depart avec le mouvement brownien, les processus de Markov ont un rapport etroit avec des equations de
diusion. Le sujet est très actuel en geometrie des varietes riemanniennes. Le
laplacien, lequation de la chaleur et le mouvement brownien sur les varietes
sont des sujets intimement lies ; jy reviendrai rapidement dans la dernière
section.
Larticle de Kolmogorov de 1931 est un travail danalyse. En 1933, avec
les Grundbegrie, il donne le cadre conceptuel universellement adopte depuis
lors pour les espaces de trajectoires. Son fameux theorème de prolongement
dune probabilite simplement additive sur une algèbre à la -algèbre engendree est essentiellement un outil pour construire un espace de probabilite
correspondant a` la donnee dun processus. Comme je lai dej`
a dit cest une
vaste systematisation de la première approche de Wiener. Cependant Kolmogorov ne reprend pas letude des processus de Markov a` partir du point de vue
probabiliste quil a degage. Cette etude probabiliste des processus de Markov
allait etre luvre de jeunesse de Kiyosi It
o.
Le deuxième article dIto dans ses Papers (p. 42) est la traduction en
anglais dun article en japonais de 1942. Le contenu de cet article na pas
attendu cette traduction tardive pour etre connu. Mais sa lecture, après un
demi-siècle, est un regal. On y trouve a` letat naissant et parfaitement exposes,
dans le cadre dune etude probabiliste des processus de Markov : 1) la notion
de dierentielle dun processus de Markov et son application au mouvement
brownien ; 2) la denition de lintegrale dIt
o ; 3) a` laide de cette integrale,
des theorèmes dexistence de solutions pour des equations dierentielles stochastiques.
Je me bornerai à evoquer lintegrale dIt
o. Lintegrale de Wiener, nous
lavons vu, est lintegrale dune fonction deterministe par rapport au bruit
blanc ; le temps ny joue pas de r
ole. Dans lintegrale dIt
o, le temps est
essentiel, et on peut integrer des fonctions aleatoires. It
o considère un mouvement brownien Xt sur R+ et un processus Yt qui nest fonction que des X ,
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
144
J.-P. KAHANE
0 t. Il denit
Y dX
0
comme une limite de sommes de Riemann o`

u les Y se trouvent decales à
gauche des accroissements de X :
n

Yi1 Xti Xti1
(0 = t0 = 0 < 1 < t1 < 2 < tn = t).
Lesperance de ces sommes est toujours nulle, donc aussi lesperance de lintegrale. Cela explique des formules en apparence paradoxales, et très utiles :
t
1
1
X dX = Xt2 t
2
2
0
Xt
t
t
1
a (x )d
a(X )dX =
a()d
0
0 2
0

t
1 t
A (X )dX +
A (X )d.
A(Xt ) = A(0) +
2 0
0
Il obtient alors une foule de processus comme solutions dequations dierentielles stochastiques du type
t
t
a(, Y )d +
b(, Y )dX .
Yt = c +
0
Parmi les applications actuelles de lintegrale dIt

o, et les motivations pour
etudier certaines fonctionnelles du mouvement brownien, il faut mentionner
les mathematiques nancières. Le point de depart est un modèle de Black
et Scholes [1973] pour la xation des prix doptions dachats ou de ventes a`
des dates et prix donnes, et pour les protocoles de gestion des avoirs nanciers par les banques qui correspondent a` ces produits derives. Les methodes
probabilistes de la nance ont fait lobjet dun expose de Hans Follmer au
Congrès International de Zurich en 1994 ; malheureusement le texte de cet
expose ne gure pas dans les Proceedings. Un seminaire Bachelier, dirige par
Nicole El Karoui, et consacre aux mathematiques nancières, se tient à lInstitut Henri Poincare, et des cours sont disponibles sur le sujet. Ma reference
preferee est à la fois un cours et un pamphlet, par Nicolas Bouleau en 1995,
intitule La nance, ingenierie ou hysterie ? Il est serieux et clair comme expose, et percutant dans sa partie critique ; cest le germe dun livre, à paratre
chez Odile Jacob [Bouleau 1998].
Ainsi lintegrale et les processus dIto, lointains descendants de la theorie de la speculation de Bachelier, retournent a` la speculation nancière. Ils
meritent a` tous egards detre integres dans la culture generale des mathematiciens.
` 3
SEMINAIRES
ET CONGRES
5.
145
Pearson, P
olya et les marches au hasard
Les marches au hasard remontent au debut du calcul des probabilites ;

levolution du gain dun joueur au jeu de pile ou face en donne une image. Pour
les marches au hasard isotropes dans le plan, au cours dune suite de temps
discrets, on se refère aux random ights de Pearson [1905], meme si lorigine
semble appartenir a` Lord Rayleigh, dans un article de 1880 dont le titre est
très clair : Composition of n isoperiodic vibrations of unit amplitude and
of phases distributed at random . Lhistoire du sujet au debut du siècle est
decrite dans larticle de S. Chandrasekhar en 1943, Stochastic problems in
physics and astronomy , dont lessentiel est consacre au mouvement brownien
(cf. [Wax 1954]).
Chandrasekhar ne cite pas Georges Polya. Pourtant, pour les mathematiciens, Polya est le premier theoricien des marches au hasard. Cest lui qui,
en 1921, pose et traite pour les marches au hasard sur Zd les questions essentielles : la recurrence, la transience, lexistence de points multiples. On connat
sa boutade : dans le plan, tous les chemins mènent a` Rome , cest-à-dire
que la promenade sur Z2 est recurrente, tous les points sont de multiplicite
innie. Au contraire la promenade dans Z3 est transiente, et tous les points
sont de multiplicite nie.
Les marches au hasard ont des avatars multiples et une forte relation a`
la theorie du potentiel ; jen dirai un mot tout a` lheure. Regardons dabord
comment se presentent recurrence, transience et points multiples pour le mouvement brownien dans Rd .
Sur R, le mouvement brownien passe une innite de fois non denombrable
en chaque point, et le temps local mesure le temps passe en chaque point. On
peut denir le temps local, pour le brownien comme pour dautres processus,
par sa derivee, la mesure aleatoire (X a) ; cest le procede utilise dans
mes livres. Sa loi est simple et remarquable, et Paul Levy a montre comment
construire le mouvement brownien X(t) en partant de lensemble X 1 (0),
dont on connat bien la loi, et en mettant en place sur les intervalles contigus
ce quon appelle des excursions browniennes ; la reference de base est son livre
de 1948.
Dans R2 le mouvement brownien revient sans cesse au voisinage de chaque
point, ce qui est une forme de recurrence, mais, comme je lai dej`
a dit, il ne
visite que les points dune aire nulle ; il ne revient jamais exactement à son
point de depart [Levy 1948]. Par contre, il y a des points quil visite une innite
non denombrable de fois [Dvoretzky et al. 1958]. Paul Levy considerait ce
resultat comme lun des plus surprenants theorèmes de lanalyse moderne
(2e edition de Processus stochastiques et mouvement brownien [1965, p. 325]).
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
146
J.-P. KAHANE
Dans R3 le mouvement brownien sen va a` linni : il est transient. Cependant il a des points doubles [Dvoretzky et al. 1950], et il na pas de point
triple [Dvoretzky et al. 1957].
Dans R4 le mouvement brownien est transient et na pas de point double,
et naturellement il en est de meme en dimension superieure [Dvoretzky et al.
1950].
Tous ces resultats, sur les points multiples du mouvement brownien dans
Rd , dependent dun fait très important etabli par Kakutani dès 1944 : la
capacite newtonienne dun borelien E de Rd (d 2) est nulle si et seulement
sil est presque s
ur que le mouvement brownien partant dun point a` distance
positive de E ne rencontre jamais E [Kakutani 1944a]. Il est facile de voir que
limage par le brownien dun intervalle reel I est de capacite positive dans R2
et R3 et de capacite nulle dans R4 , et il en resulte que pour deux intervalles
rationnels disjoints I et J les images X(I) et X(J) sont presque s
urement
4
disjointes dans R , et se rencontrent avec probabilite positive dans R3 et
R2 . Cela etablit les resultats sur les points doubles [Dvoretzky et al. 1950].
Le cas des points triples dans R3 se traite de facon analogue. Les points de
multiplicite c necessitent plus de travail, et on y est revenu recemment.
Le role des marches au hasard et du mouvement brownien en theorie du
potentiel avait ete reconnu, avant Kakutani, par Courant, Friedrichs et Lewy
dans un article de 1928. En eet, pour un ouvert dans Rd , le mesure harmonique dune portion de frontière par rapport a` un point interieur sinterprète comme la probabilite pour que le mouvement brownien issu de ce point
sechappe de louvert a` travers cette portion de frontière. Le resultat analogue sur des graphes est encore plus facile à voir et faisait peut-etre partie
du folklore. Mais Kakutani faisait de cette observation simple un usage merveilleux ; je lai vu, au tableau noir devant un certain ensemble de Cantor, se
demander sil etait de capacite positive ou non en sidentiant mentalement
a une particule brownienne et en cherchant a` passer au travers.
`
Dans la tradition de Kakutani, le plus beau resultat de ces dernières annees
est sans doute le theorème de N. G. Makarov [1989] : pour un domaine de
Jordan dans le plan, la mesure harmonique est concentree sur un borelien de
dimension 1. Donc, si la dimension de la courbe frontière est proche de 2, une
trajectoire brownienne issue de linterieur sarrete presque s
urement sur une
très petite partie de cette courbe, de dimension 1.
On imagine bien comment le mouvement brownien sur des varietes riemanniennes permet dinterpreter des objets abstraits comme la frontière de
Martin. Letude du comportement asymptotique du mouvement brownien a
commence avec Dynkin dans les annees 1950 et cest un sujet très etudie aujourdhui. Pour les surfaces a` courbure negative constante, complètes et sans
` 3
SEMINAIRES
ET CONGRES
147
bord, les trajectoires convergent vers des points frontière en un temps inni ;
si la courbure crot rapidement, la convergence a lieu en un temps ni ; la discussion a ete faite par Azencott dans les annees 1970. Toute la litterature sur
les frontières de Furstenberg, les espaces de Poisson, les fonctions harmoniques
bornees, est plus ou moins liee au mouvement brownien.
La recurrence et la transience de marches sur des groupes selon dautres
probabilites de transition que lequiprobabilite des points les plus voisins sont
encore très liees aux diusions et au potentiel. Je me contente devoquer
les travaux dYves Guivarch et de Nicolas Varopoulos. Voici un resultat de
Varopoulos [1985] qui complète bien les theorèmes de Polya : un groupe denombrable sans torsion ne porte une marche recurrente que si cest Z ou Z2 .
Enn, le mouvement brownien est present implicitement dans beaucoup de
travaux contemporains danalyse ou de geometrie sur des groupes. Un exemple
remarquable est le livre de Varopoulos, Salo-Coste et Coulhon, qui est une
monographie de recherche sans rapport apparent au mouvement brownien ;
mais on decouvre à la n du livre, p. 141, que the problem of transience
and recurrence of Brownian motion on covering manifolds was the original
motivation for the theory developed in this book.
Les marches au hasard sur les graphes et les groupes constituent un domaine de recherche en plein essor. Larticle dexposition de Woess en 1994,
qui pretend se borner au survol de sujets choisis dans ce domaine, comporte
264 references !
Appendice
Ayant parcouru les quelques chemins que javais denis au depart je crois
avoir balaye de facon `
a peu près correcte la periode 1900-1950 et montre
quelques-unes des ramications qui sensuivent. Cependant je nai pas touche aux plus grands sujets des annees 1980 et 90 : martingales et mouvement
brownien, mouvement brownien plan, lois explicites de processus lies au brownien. Cela naurait pas ete très dicile à partir dune sixième source, qui se
situe au milieu du siècle : Paul Levy. Comme il nest pas question dallonger
cet article, je me contente de presenter au lecteur mes excuses et quelques
suggestions de lecture.
La Bible est le livre de Paul Levy de 1948 et sa reedition de 1965. Il est dejà
question du mouvement brownien dans son livre de 1937. Le volume V de ses
uvres contient le reste de ce quil a ecrit sur le mouvement brownien. On peut
lire comme un roman le fascicule 126 du Memorial des sciences mathematiques
(1954) et les conferences de Rome (1962).
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
148
J.-P. KAHANE
Le livre de Nelson [1967] est riche dinformations historiques et de vues

personnelles. On peut en dire autant du livre recent de Chung [1995].
Jai cite dès le debut le livre de Revuz et Yor [1991], qui traite des martingales continues et du mouvement brownien. Les martingales ont un rapport
etroit avec lanalyse classique, en particulier avec letude des fonctions analytiques dans le disque unite. Le livre de R. Durrett, sur le mouvement brownien et les martingales en analyse, fait excellemment le point en 1984 ; il faut
le completer aujourdhui par louvrage de Richard Bass [1995] sur les techniques probabilistes de lanalyse. Le mouvement brownien plan a ete etudie
par Levy dès 1940, et son theorème sur linvariance des trajectoires par representation conforme a ete publie en 1947. Le sujet est en plein developpement.
La meilleure reference est le cours de Jean-Francois Le Gall a` Saint-Flour en
1990, a` completer par les travaux de son elève Wendelin Werner [1995]. La
geometrie de la courbe brownienne fait lobjet dun expose de Gerard Ben
Arous [1990] au seminaire Bourbaki. On trouve une liste de problèmes et de
references dans larticle de B. Duplantier, G. F. Lawler, J.-F. Le Gall et T. J.
Lyons du Bulletin des sciences mathematiques [1993].
Les lois explicites de processus lies au mouvement brownien commencent
aussi avec Paul Levy. Cest une specialite de Marc Yor, et son livre de 1992
est sans doute le meilleur moyen de sinitier au sujet avec en complement son
cours de Caracas 1995. Il serait imprudent de ma part de citer dautres noms,
mais cest à regret ; les recherches de lois explicites se mènent activement en
plusieurs parties du monde et en plusieurs villes en France, et pas seulement
a Paris.
`
Reste qu`
a Paris se tient regulièrement, depuis des annees, un seminaire sur
letude ne du mouvement brownien. Les annales de ce seminaire constituent
une pièce matresse pour lhistoire contemporaine du mouvement brownien.
Si le lecteur de cet article est mis en appetit, ou sil se sent frustre, ce
qui peut etre la meme chose, je lui recommande de changer complètement de
point de vue, et de se plonger dans lhistoire contemporaine du mouvement
brownien. Marc Yor redige actuellement (juin 1996) une etude sur ce sujet,
qui sera bient
ot disponible, et paratra dans louvrage collectif sur le developpement des mathematiques depuis 1950 dont linitiateur est Jean-Paul Pier.
Dans laller et retour entre le passe et le present, qui est le propre de toute
histoire, leorescence et la qualite des travaux actuels sur le mouvement
brownien sont les stimulants principaux pour la recherche des sources et des
cheminements. Larticle à venir de Marc Yor est donc, pour une bonne part,
la justication de celui-ci.
` 3
SEMINAIRES
ET CONGRES
149
Bibliographie
Assouad (P.)
[1980]
Plongements lipschitziens dans Rn et courbes de von Koch, C. R. Acad.

Sci. Paris, 290 (1980), p. 591594.
Azencott (R.)
[1973]
Diusion sur les varietes dierentiables, C. R. Acad. Sci. Paris, 276

(1973), p. 363365.
[1992]
Sequential simulated annealing : speed of convergence and acceleration

techniques, dans Simulated Annealing (ed. par R. Azencott), New York :
Wiley, 1992, p. 110.
Bachelier (L.)
[1900]
Theorie de la speculation (Thèse), Ann. Sci. Ecole

Norm. Sup., (III) 17
(1900), p. 21-86.
Bass (R.)
[1995]
Probabilistic techniques in analysis, Basel-Heidelberg-New York : Springer, 1995.
Ben Arous (G.)

[1990]
Geometrie de la courbe brownienne, Sem. Bourbaki, exp. no 730 (1990).
Beurling (A.) et Deny (J.)

[1959]
Dirichlet spaces, Proc. Nat. Acad. Sci. USA, 45 (1959), p. 208215.
Billard (P.)
[1963]
Series de Fourier aleatoirement bornees, continues, uniformement conver

gentes, Studia Math., 22 (1963), p. 309329 ; Ann. Sci. Ecole
Norm. Sup.,
(III) 82 (1963), p. 131-179.
Black (F.) et Scholes (M.)

[1973]
The pricing of options and corporate liabilities, J. of Political Economy,

81 (1973), p. 635654.
Borel (E.)
[uvres]
uvres, tomes I à IV, CNRS, 1972.
[1896]
Sur les series de Taylor, C. R. Acad. Sci. Paris, 123 (1896), p. 10511052 ;
uvres II, p. 647648.
Bouleau (N.)
[1998]
Martingales et march
es nanciers, Paris : Odile Jacob, 1998.
Bouleau (N.) et Lepingle (D.)

[1994]
Numerical methods for stochastic processes, New York : Wiley, 1994.

E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
150
J.-P. KAHANE
Chandrasekar (S.)
[1943]
Stochastic problems in physics and astronomy, Rev. Modern Phys., 15

(1943), p. 1-89 ; reimpr. dans [Wax 1954, p. 391].
Chung (K.L.)
[1995]
Green, Brown and probability. Singapour : World Scientic, 1995.
Courant (R.), Frieedrichs (K.) et Lewy (K.)

[1928]
Ueber die partiellen Dierenzengleichungen der mathematischen Physik,

Math. Ann., 1000 (1928), p. 3774.
Doob (J.L.)
[1942]
The Brownian motion and stochastic equations, Ann. of Math., 43 (1942),

p. 351369 ; reimpr. dans [Wax 1954, p. 319337].
Dudley (R.)
[1967]
The sizes of compact subsets of Hilbert space and continuity of Gaussian

processes, J. Funct. Anal., 1 (1967), p. 290330.
Duplantier (B.)
[1989]
Le mouvement brownien en physique, les polymères et leur relation avec

les phenomènes critiques, dans Journees de la Societe Mathematique
de France, 28 janvier 1989, trois expos
es sur le mouvement brownien,
(P. Le Gall : introduction, Ben Arous : grandes deviations et noyau de la
chaleur, Duplantier : supra).
Duplantier (B.), Lawler (G.F.), Le Gall (J.F.) et Lyons (T.J.)

[1993]
The geometry of the Brownian curve, Bull. Sci. Math., 117 (1993), p. 91
106.
Dvoretzky (A.)
[1963]
On the oscillation of the Brownian motion process, Israel J. Math., 1

(1963), p. 212214.
Dvoretzky (A.), Erd

os (P.) et Kakutani (S.)
[1950]
Double points of paths of Brownian motion in n-space, Acta Sci. Math.

(Szeged), 12 (1950), p. 7581 ; Papers II de Kakutani, p. 102108.
[1958]
Points of multiplicity c of plane Brownian motion paths, Bull. Res. Council Israel, sect. F 7 (1958), p. 175180 ; Papers II de Kakutani, p. 124129.
[1961]
Non increase everywhere of the Brownian motion process, dans Proc. 4th
Berkeley Symp., 1960, vol. 2, Univ. Calif. Press, 1961, p. 103116 ; Papers
II de Kakutani, p. 130143.
Dvoretzky (A.), Erd

os (P.), Kakutani (S.) et Taylor (S.J.)
[1957]
Triple points of Brownian paths in 3-space, Proc. Cambr. Phil. Soc., 53

(1957), p. 856862 ; Papers II de Kakutani, p. 117123.
` 3
SEMINAIRES
ET CONGRES
151
Dynkin (E.B.)
[1960]
Markov processes and analysis problems connected with them, Uspehi

Mat. Nauk, 15 (1960), p. 324.
[1963]
Processus de Markov et problèmes danalyse (en russe), dans Proc. Int.

Congr. Math. Stockholm 1962, Uppsala, 1963, p. 3658.
Einstein (A.)
[1905]
Ueber die von der molekularkinetischen Theorie der Warme gefordete

Bewegung..., Annalen der Physik, 17 (1905), p. 549560.
[1906a]
Eine neue Bestimmung der Molek

uldimensionen, Ann. Physik, 19 (1906),
p. 289306.
[1906b]
Zur Theorie der Brownschen Bewegung, Ann. Physik, 19 (1906), p. 371

381.
[1922]
Untersuchungen u
ber die Theorie der Brownschen Bewegung , vol. 199
of Ostwalds Klassiker der exakten Wissenschaften, Leipzig Akadem. Verlagsgesellschaft, 1922. Traduction anglaise : Investigations on the theory
of Brownian movement, New York : Dutton 1926, reimprime par Dover,
1956.
Fernique (X.)
[1975]
Regularite des trajectoires des fonctions aleatoires gaussiennes, dans Ecole

dete de probabilites de Saint-Flour IV, 1974, Lecture Notes in Math.,
vol. 480, Springer, 1975, p. 196.
F
ollmer (H.)
[1984]
Von der brownschen Bewegung zum brownschen Blatt, dans Perspectives

in mathematics (ed. par J. Jager, J. Moser et R. Remmert), Basel : Birkh
auser, 1984.
Geman (S.) et Hwang (C.)

[1986]
Diusions for global optimization, SIAM J. Control Optim., 24 (1986),

p. 10311043.
Guivarch (Y.)
[1980]
Quelques proprietes asymptotiques des produits de matrices aleatoires,
dans Ecole
dete de probabilites de Saint-Flour VIII (1978) (ed. par P.L.
Hennequin), vol. 774 des Lecture Notes in Math., Berlin-Heidelberg :
Springer, 1980, p. 177250.
ev (S.V.), Kahane (J.P.) et Katznelson (Y.)

Hru
sc
[1981]
Mouvement brownien et series de Fourier absolument convergentes,

C. R. Acad. Sci. Paris, 292 (1981), p. 389391.
(K.)
Ito
[Papers]
Selected papers, New York : Springer, 1987.

E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
152
[1942]
J.-P. KAHANE
Dierential equations determining a Markov process (translation from the

Japanese) ; Papers, p. 42-75.
Kac (M.)
[1964]
Wiener and integration in function spaces, Bull. Amer. Math. Soc., 72

(1964), p. 5268.
Kahane (J.P.)
[1968]
Some random series of functions. Heath et Cambridge University Press,

2e ed., 1968.
[1974]
Sur lirregularite locale du mouvement brownien, C. R. Acad. Sc. Paris,

278 (1974), p. 331333.
[1981]
Helices et quasi-helices, Adv. Math., 78 (1981), p. 417433, [volume

L. Schwartz].
Kakutani (S.)
[Papers]
Selected papers, 2 vol., Boston-Basel-Stuttgart : Birkhauser, 1986.
[1944a]
On Brownian motions in n-space, Proc. Imp. Acad. Japan, 20 (1944),

p. 648652 ; Papers II, p. 8892.
[1944b]
Two dimensional Brownian motion and harmonic functions, Proc. Imp.

Acad. Japan, 20 (1944), p. 706714 ; Papers II, p. 93101.
[1961]
Spectral analysis of stationary Gaussian processes, dans Proc. 4th Berkeley Symp. 1960, Univ. Calif. Press, 1961, p. 239247 ; Papers II, p. 7785.
Khintchine (A.)
[1927]
Uber
das Gesetz der grossen Zahlen, Math. Ann., 96 (1927), p. 152168.
Khintchine (A.) et Kolmogorov (A.)
[1925]
Ueber Konvergenz von Reihen deren Glieder durch den Zufall bestimmt
werden, Mat. Sbornik, 32 (1925), p. 668677.
Kolmogorov (A.)
[Papers]
Selected papers, Birkhauser, 1986.
[1931]
Uber
die analytischen Methoden in der Wahrscheinlichkeitsrechnung,
Math. Ann., 104 (1931), p. 415458 ; Papers II, p. 62108.
[1933]
Grundbegrie der Wahrscheinlichkeitsrechnung. Berlin : Springer, 1933.
Langevin (P.)
[1908]
Sur la theorie du mouvement brownien, C. R. Acad. Sci. Paris, 146

(1908), p. 530532.
Le Gall (J.F.)
[1992]
Some properties of planar Brownian motion, dans Ecole

dete de probabilites de Saint-Flour XX, 1990 (ed. par P.L. Hennequin), vol. 1527 des
Lecture Notes in Math., Berlin-Heidelberg : Springer, 1992, p. 111236.
` 3
SEMINAIRES
ET CONGRES
153
Lebesgue (H.)
[uvres]
uvres scientiques, vol. I a` V. Genève : LEnseignement mathematique,

1972-1973.
[1902]
Integrale, longueur, aire, thèse ; uvres I, p. 203331.
[1906]
Lecons sur les s

eries trigonom
etriques. Paris : Gauthier-Villars, 1906.
Ledoux (M.) et Talagrand (M.)

[1991]
Probability in Banach space, Berlin-Heidelberg : Springer, 1991.
L
evy (P.)
[uvres]
uvres, vol. I a` VI. Paris : Gauthier-Villars, 1973.
[1937]
Theorie de laddition des variables al

eatoires. Paris : Gauthier-Villars,
1937. 2e edition 1954.
[1948]
Processus stochastiques et mouvement brownien. Paris : Gauthier-Villars,

1948. 2e edition, 1965.
Makarov (N.)
[1989]
Probability methods in conformal mappings, Leningrad Math. J., 1

(1990), p. 156 ; traduction de Algebra i Analiz, 1 (1989), p. 359.
Mandelbrot (B.)
[1982]
The fractal geometry of nature, San Francisco : Freeman, ed. rev. 1982.
Marcus (M.) et Pisier (G.)

[1981]
Random Fourier series with applications to harmonic analysis. Princeton

University Press, 1981.
Nelson (E.)
[1967]
Dynamical theories of Brownian motion. Princeton University Press,

1967.
von Neumann (J.) et Schoenberg (I.J.)

[1941]
Fourier integrals and metric geometry, Trans. Amer. Math. Soc., 50

(1941), p. 226251.
Orey (S.) et Taylor (S.J.)

[1974]
How often on a Brownian path does the law of iterated logarithm fail ?,
Proc. London Math. Soc., 28-3 (1974), p. 174192.
Paley (R.E.A.C.) et Wiener (N.)

[1934]
Fourier transforms in the complex domain. Amer. Math. Soc. Colloq.

Publ., 19, New York, 1934.
Paley (R.E.A.C.), Wiener (N.) et Zygmund (A.)

[1933]
Notes on random functions, Math. Z., 37 (1933), p. 647668.

E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
154
J.-P. KAHANE
Paley (R.E.A.C.) et Zygmund (A.)

[1930]
On some series of functions, Proc. Cambridge Philos. Soc., 26 (1930),

p. 337357, 458474.
[1932]
On some series of functions, Proc. Cambridge Philos. Soc., 28 (1932),

p. 190205.
Perrin (F.)
[1928]
Etude mathematique du mouvement brownien de rotation, thèse, Paris.
[1934-36]
Mouvement brownien dun ellipsode I, II., J. Phys. Radium, 5 (1934),

p. 497511 ; Ibid., 7 (1936), p. 111.
Perrin (J.)
[1909]
Mouvement brownien et realite moleculaire, Ann. Chim. Phys., 18 (1909),

p. 1114.
[1912]
Les atomes, Paris : Felix Alcan, 1912.
Rademacher (H.)
[1922]
Einige S
atze u
ber Reihen von allgemeinen Orthogonalfunktionen, Math.
Ann., 87 (1922), p. 112138.
Revuz (D.) et Yor (M.)

[1991]
Continuous martingales and Brownian motion. Berlin-Heidelberg : Springer, 1991 ; 2e edition 1994.
Salem (R.) et Zygmund (A.)

[1954]
Some properties of trigonometric series whose terms have random signs,

Acta Math., 91 (1954), p. 245301.
Schoenberg (I.J.)
[1938]
Metric spaces and positive denite functions, Trans. Amer. Math. Soc.,
44 (1938), p. 522536.
Smoluchowski (M.)
[1906a]
Essai dune theorie du mouvement brownien et de milieux troubles, Bull.

Acad. Sci. Cracovie, (1906), p. 577602.
[1906b]
Zur kinetischen Theorie der Brownschen Molekularbewegung und der Suspensionen, Ann. Physik, 21 (1906), p. 756780.
Sredniawa (B.)
[1991]
Collaboration of Marian Smoluckowski and Theodor Svedberg in the investigations of Brownian motion and density uctuations (Krak
ov, preprint).
Steinhaus (H.)
[1923]
Les probabilites denombrables et leur rapport a` la theorie de la mesure,

Fund. Math., 4 (1923), p. 286310.
` 3
SEMINAIRES
ET CONGRES
[1930]
155
Ueber die Wahrscheinlickeit daf

ur, dass der Konvergenzkreis einer Potenzreihe ihre nat
urlich Grenze ist, Math. Z., 31 (1930), p. 408416.
Svedberg (T.)
[1907]
Studien zur Lehre von den kolloidalen Losungen, Regiae Societatis Scientiarum Upsaliensis, IV (1907), p. 1160.
Taylor (S.J.)
[1964]
The exact Hausdor measure of the sample path for planar Brownian
motion, Proc. Cambridge Philos. Soc., 60 (1964), p. 253258.
[1972]
Exact asymptotic estimates of Brownian path variation, Duke Math. J.,

39 (1972), p. 219241.
Varopoulos (N.)
[1985]
Random walks on groups. Applications to Fuchsian groups, Ark. Mat.,

23 (1985), p. 171176.
Varopoulos (N.T.), Saloff-Coste (L.) et Coulhon (T.)

[1992]
Analysis and geometry on groups. Cambridge Univ. Press, 1992.
Wax (N.)
[1954]
Selected papers on noise and stochastic processes, New York : Dover,

1954. Contains papers of Chandrasekhar, Uhlenbeck et Ornstein, Wang
et Uhlenbeck, Rice, Kac, Doob.
Werner (W.)
[1995]
Sur lensemble des points autour desquels le mouvement brownien tourne

beaucoup, Probab. Theory Related Fields, 99 (1995), p. 11142.
Wiener (N.)
[Papers]
Selected papers, Cambridge, Mass. : M.I.T. Press, 1964.
[Works]
Collected works, vol. I a` IV. Cambridge, Mass. : M.I.T. Press, 1985.
[1923]
Dierential space, J. Math. Phys., 2 (1923), p. 131174.
Woess (W.)
[1994]
Random walks on innite graphs and groups - a survey of selected topics,

Bull. London Math. Soc., 26 (1994), p. 160.
Yor (M.)
[1992]
Some aspects of Brownian motion. Part I : Some special functionals,

Lectures in mathematics, ETH Z
urich, Basel : Birkhauser, 1992.
[1995]
Local times and excursions for Brownian motion, a concise introduction,

Lecciones en Matematicas. Facultad de Ciencias, Universidad Central de
Venezuela, 1995.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
Interrelations between
Mathematics and Physics
Yu. I. Manin
Abstract
After briey describing the mathematical structure of modern
physics, this paper analyzes the divergence between the development of physics and of mathematics in the rst half of the 20th
century, with emphasis on the role in each discipline of rigorous
denitions and proofs, of algebraic calculations and of intuitive
ideas.
R
esum
e
Après avoir decrit la structure mathematique de la physique moderne, cet article analyse la divergence entre mathematiques et
physique dans la première moitie du XXe siècle, en etudiant,
pour chacune des disciplines, le role respectif des denitions et
demonstrations rigoureuses, des calculs algebriques et des idees
intuitives.
1.
Foreword
I would like to start with an explicit description of the conceptual framework

of this study.
To render it concisely, it is useful to look at the case of comparative
linguistics. The history of a language is not a history of all, or even of the
most important, utterances (oral or written) in this language. Rather, it is a
history of evolution of the language as a system. Hence we need a preliminary
description of the system(s) whose genesis we are studying.
AMS 1991 Mathematics Subject Classication: 00A30, 01A60, 01A65

MaxPlanckInstitut f
ur Mathematik, Bonn, Deutschland.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
158
YU. I. MANIN
An application of this Saussurian scheme to the history of mathematics

(which, incidentally, I do not consider to be a mere language) was probably
particularly appealing to Jean Dieudonne who, as an active member of the
Bourbaki group, participated in the creation of a systematic picture of modern
mathematics.1 In this talk I follow his example, on a much humbler scale.
Needless to say that restrictions of time, space, and competence, force me to
choose a thin chain of connected ideas and present them in a highly selective
way.
Thus I refuse (somewhat reluctantly) to discuss the history with Rankean
insistence on wie es eigentlich gewesen ist. One reason for this refusal is that
the history of contemporary mathematics tends to degenerate into credit and
priority assignments, lacking pathetically the dramatic appeal with which the
history of struggles for real power is charged. A more personal and compelling
motive is succinctly put by Joseph Brodsky in his autobiographical essay Less
Than One: The little I remember becomes even more diminished by being
recollected in English.
A last word of warning and apology is due. Any system is, of course, a
theoretical construct. As such, it is at best relative and culture dependent, at
worst subjective. It is precisely in this function that it can serve as material
for the history of mathematics of the 20th century.
2.
2.1.
Mathematical Physics as a System

Physics
Physics describes the external world, and in its domain of competence, does
this in two complementary modes: classical and quantum.
In the classical mode, events occur to the matter and elds which reside
and evolve in the spacetime. Physical laws directly constrain observables.
They are basically deterministic and expressed by the dierential equations
which (sometimes demonstrably, sometimes hypothetically) satisfy appropriate uniqueness and existence theorems.
A statistical submode of the classical mode of description deals with probabilities and averages which (sometimes demonstrably, sometimes presumably)
can be deduced from an ideal deterministic description. The need for a statis1
Jean Dieudonne, as I remember him, had a strong voice, strong hands, and strong
opinions. In particular, he insisted on using tensor products and commutative diagrams
instead of classical subscripts and superscripts in calculations involving tensors. I used to
believe his judgement that this was a chalksaving device, until one day I had to calculate
with tensors myself. Then I found out that subscripts were much more economical.
` 3
SEMINAIRES
ET CONGRES
INTERRELATIONS BETWEEN MATHEMATICS AND PHYSICS
159
tical treatment arises from two basic premises: too many degrees of freedom
and/or instability. (Metaphorically speaking, instability means that each consecutive decimal digit is a new degree of freedom.)
A fundamental physical abstraction is that of an isolated system which
evolves in oblivion of the rest of the world, and of interaction between potentially isolated systems, or one isolated system and the rest of the world.
In one of the most remarkable ights of fancy of classical physics, space
time itself appears as such an isolated system governed by Einsteins equations
of general relativity (perhaps, with an energymomentum tensor summarily
responsible for everything which is not pure spacetime).
In the quantum mode of theoretical description, the observable world is
inherently probabilistic. Moreover, and more signicantly, the basic laws
which are in a sense deterministic govern an unobservable entity, the
probability amplitude, which is a complex valued function on a quantum path
space. Roughly speaking, the amplitude of a composite event is the product
of the amplitudes of its constituents, whereas the amplitude of an event which
is a sum of alternatives is the sum of the amplitudes of these alternatives.
The probability of an event is the modulus squared of its amplitude. Physical observables are the appropriate averages, even if one speaks about an
elementary act of scattering of an individual particle. The observable wave
behavior of, say, light is only an imperfect reection of the inherent wave
behavior of the amplitudes (wave functions) of an indeterminate number of
photons described by the Fock space of the quantized electromagnetic eld.
Partly as a result of historical development, many quantum models contain as an intermediate stage a classical model which is then quantized. The
word quantization rather indiscriminately refers to a wide variety of procedures of which two of the most important are operator, or Hamiltonian,
quantization, and path integral quantization. The rst is more algebraic and
usually has a rmer mathematical background. The second possesses an enormous heuristic and aesthetic potential. I haven chosen the latter for my more
detailed subsequent discussion.
If I had included the rst one, the picture of the divergence of Mathematics
and Physics in the rst half of this century sketched below in Sec. IV would
appear less pronounced. Nevertheless, the main results of my analysis would
survive.
One more subject matter deserving a separate historical and structural
study is the duality between these two approaches. It started with classical mechanics, Lagrange, and Hamilton, and continued via Heisenberg
Schr
odinger wave mechanics to the path integral/scattering matrix controversy. On the fringes of physics it contains such recent mathematical gems as
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
160
YU. I. MANIN
Virasoro algebra representations on the moduli spaces of curves.
2.2.
Mathematics
If there is one most important notion of mathematical physics, it is that of

action functional. It encompasses the classical ideas of energy and work,
its
density in a domain of spacetime is the Lagrangian, and multiplied by 1
and exponentiated, it furnishes the basic probability amplitude. Action is
measured in absolute Planck units, and therefore can be thought of as a real
number. More precisely, we will consider the following scheme of description
central for both modes of physical description referred to above.
The modeling of a physical system starts with the specication of its
kinematics. This includes a space P of virtual classical paths of the system and
an action functional S : P R. For example, P may consist of parametrized
curves in a classical phase space of a mechanical system, or of Riemannian
metrics on a given smooth manifold (spacetime), or of triples (a connection
on a given vector bundle, a metric on it, a section of it) etc. The
value of the
action functional at a point p P is usually given in the form p L, that is a
volume form integrated over one of the spaces guring in the description of p.
Classical equations of motion specify a subspace Pcl P. This subset
consists of the solutions of the variational equations (S) = 0, i. e., of the
stationary points of the action functional.
If the classical description is the statistical one, then exp(S) is the probability density.
In the quantum description, we choose physically motivated subsets
B P, typically determined by boundary conditions, and dene the average
of an observable O in B by a path integral of the type

R
O(p) e i p L Dp.
(2.1)
O B :=
B
These are our main actors. In the following, I present some musings about
the history of this picture as seen through the eyes of physicists and mathematicians.
I will be most interested in the idea of the integral and its nal incarnation,
in the form of the path integral.
3.
The Integral
The notion of an integral is one of the central and recurring themes in the
history of mathematics for the last two millennia. The ardent problem solving
` 3
SEMINAIRES
ET CONGRES
161
is periodically followed by the anxious denition seeking, only to be replaced

by new nonrigorous but amazingly ecient heuristics leaving a logically
minded fundamentalist in each of us baed.
Richard Feynman who created the hierogram (2.1) (still lacking a precise mathematical meaning exactly in those cases when it is most needed by
physicists2 ) used to boast that (2.1) allowed the calculation of the anomalous
magnetic momentum of the electron, which coincided with its experimental
value up to ten digits:
As of 1983, the theoretical number was 1.00115965246, with an uncertainty of about 20 in the last two digits; the experimental number was
1.00115965221, with an uncertainty of about 4 in the last digit. This accuracy is equivalent to measuring the distance from Los Angeles to New York, a
distance of over 3000 miles, to within the width of a human hair. [Feynman
1988, p. 118]
This feat was recently matched by physical calculations (even called predictions, cf. [Candelas et al. 1991]) of various interesting numbers in algebraic
geometry, such as the number Nd of rational curves of degree d on a generic
threedimensional quintic (e. g. 70428 81649 78454 68611 34882 49750 for
d = 10, a theoretical(?) number still unchecked in an experiment(?) involving a mathematical denition of Nd and a computer.) The ideology of path
integration played an essential role in these calculations, leading to an interpretation of an instance of (2.1) as a sum over instantons in a sigmamodel,
which in this particular case are rational curves on a quintic.
The intuitive physical picture of an integral is the quantity of something
in a domain. If the rst calculations of this something are later interpreted
as, say, the volume of a pyramid, one can hardly doubt that they were used
for estimating the actual quantity of stone (and slaves labor) needed for the
building of an Egyptian pharaohs tomb. Keplers Stereometria Doliorum
mentions wine casks in its title. The domain in question acquired a temporal
dimension when the length of a path was calculated as an integral of velocity,
and the notion of energy was gradually replaced by that of action. In the
twentieth century, topology became one of the substances the quantity of which
could be measured by integration of closed dierential forms (De Rham theory
of periods anticipated by Poincare). Probability turned out to be another
such substance, and Wieners treatment of Brownian motion as a measure in
a space of continuous paths paved the way both for Kolmogorovs axiomatic
2
For a more positive view, see [Glimm and Jae 1981], a remarkable book which inuenced the structure of this essay. On page 313 however the authors say: ... it is a
theoretical puzzle whether a theory of electrodynamics exists in the sense of a mathematical
framework ...
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
162
YU. I. MANIN
treatment of probability and our present reluctant acceptance of Feynmans

integral. (This is at least partially supported by the successes of constructive
eld theory and stochastic integration. However, the random surfaces inherent
in string path integrals present considerable diculties.)
Mathematically, any calculation (or denition) of an integral is based upon
two physically intuitive principles: additivity with respect to domains and
integrands, and a form of limiting procedure. There are at least two archetypal
forms of passing to a limit.
One is represented by Cavalieris indivisibles, Riemann sums, etc. It is
connected with the topological structure of the domain of integration, specically with the idea of boundary and thin layers of (d + 1)dimensional objects
surrounding a ddimensional object. The Stokes formula in all its modications belongs to this circle of ideas, while the De Rham complex is its linear
dual form.
Another form of limiting procedure is measuretheoretical rather than a
topological one. There are basic domains lled with well measured quantities of the substance of interest (volume, action, probability ...). We try to
approximate other distributions by using mosaic portraits of them and allowing the size of local discrepancies to tend to zero. However, locality is not
topological anymore, and the image of boundary becomes useless or irrelevant. Instead, we have to deal with measurable sets which must only form an
algebra with respect to intersections and unions. Innitedimensional constructions are usually of this type. The well known eect volume in high
dimensions tends to concentrate near the boundary prevents using the image
of indivisibles eectively. Even in nite dimensions, the boundary can fail to
serve the role of Cavalieris indivisible if it is very rough (fractal). The subtle
measure theoretic studies of the beginning of this century had much to say
about it.
There
are two integrals in (2.1), of quite dierent nature. The action

S = p L is usually a classical entity, L being a local Lagrangian. A beautiful
recent idea due to a collaboration of physicists and mathematicians (E. Witten [1989] and M. F. Atiyah [1989] playing leading roles, A. S. Schwarz having
supplied a crucial rst example) consisted in considering those path integrals
in which the action is a topological invariant of p. Locally this means that
classical equations of motion (S) = 0 are identically satised. An example
of such an action functional is the ChernSimons invariant dened on the
space of connections on a vector bundle over a threedimensional manifold.
The quantum observables (whose choice and name was motivated by the theory of strong interactions) are Wilson loops: averaged traces of monodromy
representations along closed curves in the base.
` 3
SEMINAIRES
ET CONGRES
163
In this context,
the algebraic properties of the path integral reected in the

additivity of p L and resulting multiplicativity of the whole of (2.1) become
so strong that they can be used to dene a suciently rigid mathematical
structure of Topological Quantum Field Theory which can then be studied
by precise mathematical means. This was done by G. Segal and M. F. Atiyah.
See [Reshetikhin and Turaev 1991] and [Blanchet et al. 1995] for some recent
mathematical developments in this area.
The history of the integral seen from our vantage point can be conceived
in terms of a Toynbeean challenge/response scheme. Challenges come from
physics broadly construed, including geometry. It can be convincingly argued
that even Euclidean geometry is in fact just the kinematics of rigid bodies
in the absence of a gravitational eld (curved the spacetime), and both
the invention and the development of the rst nonEuclidean geometries (of
constant curvature) was inextricably connected with physics. Gauss wanted
to know what was the actual geometry of interstellar space. Hilberts return
to axiomatics was a mathematical response to the challenge of the discovery
of multiple possible geometries of the physical world.
4.
The Schism
In this section of my talk I argue that the main event in the relationship
between mathematics and physics in the rst half of this century was their
estrangement, after several centuries of close alliance.
The divergence started in the last two decades of the last century and was
connected with the deepening understanding of two microworlds: a mathematical one embodied in the idea of the classical continuum of real numbers,
and a physical one open to experiment.
Roughly speaking, around the turn of the century Peano, Jordan, Cantor,
Borel, Stieltjes, and Lebesgue discovered and displayed with great subtlety the
new properties of continuum, continuity and measurability. They have given
a series of denitions of integration of increasing generality, and invented
constructions and existence proofs for many strange mathematical objects
which did not belong to the world of classical geometry and analysis but had
to be accepted as a consequence of classical ways of mathematical reasoning
stretched, as it seemed, to their limit.
The growing reaction against many counterintuitive discoveries led mathematicians to selfanalysis centered around several basic problems: What is a
mathematical proof? What meaning can be given to a statement about existence of a mathematical object? What is the status of mathematical innity?
The outcome of this is well known. Fifty years of introspection were
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
164
YU. I. MANIN
quite fruitful from the mathematical viewpoint: they produced mature mathematical logic, including theory of proof, theory of computability, and a clear
picture of the hierarchy of expanding languages and axiom systems that mathematicians have had to adopt consecutively in their quest for mathematical
truth.
In the meantime, physicists were engaged in a totally dierent quest.
Plancks discovery of a quantum of action announced on December 14, 1900,
initiated the quantum age. Physics needed sophisticated mathematics to formulate newly discovered nonclassical laws, but new mathematics was of no
help. Whatever was needed was hastily invented or reinvented: matrix algebra, spinors, Fock space, the delta function, the representation theory of
Lorentz group. None of the pioneers (Bohr, Einstein, Pauli, Schrodinger,
Dirac) needed the Lebesgue integral, or was interested in the cardinality of
continuum. Logic interested them even less.
This does not mean that physicists had no philosophical preoccupations;
in fact they had. But if mathematicians discussed the relationships between
language and thought, physicists were troubled by the relation of language to
reality. The basic problem confronted by the critics of classical mathematics
was the inexpressibility of innity, related to the inherently nitary syntactic
structure of language. The basic problem confronted in the BohrEinstein
controversy was the inexpressibility of quantum indeterminacy, related to
the inherently classical semantics of language. Philosophy of mathematics
and philosophy of physics almost completely lost contact with each other.
Such ardent critics of the alleged inadequacies of contemporary research as
Brouwer in mathematics and Pauli in physics shared not a single common
idea. Mathematical criticism tended to become deeply autistic, while physical
criticism strived to nd better ways to express complex reality.3
A gap formed in traditional professional interactions as well. From the
rst successes of the quantum electrodynamics in the thirties until the renewed interaction in the sixties, mathematicians contributed almost nothing
to the main physics research program of this century: Quantum Field Theory. Similarly, physicists payed no attention not only to mathematical logic
(understandably) or analytical number theory (traditionally), but also to the
emerging algebraic topology. Thirty years later, topology was to become
the new common ground for the two communities. Somewhat paradoxically,
mathematics gained from this renewed interaction more than physics: new invariants of three and fourdimensional manifolds, quantum groups, quantum
cohomology were its fruits.
3
It is characteristic that G. H. Hardys Rouse Ball Lecture [Hardy 1929] on Mathematical
Proof delivered in 1928 does not even mention existence of quantum physics.
` 3
SEMINAIRES
ET CONGRES
165
The following well known empirical observation ts well into the picture
we have sketched. Whenever a fresh mathematical tool for understanding
physics is needed, physicists are very quick at inventing new or transforming
already existing algebraic formalism to deal with it. We have already mentioned Heisenberg algebra, spinors and Dirac delta function. One can add
the SchwingerDyson equation (for an otherwise undened path integral),
the Berezin integral on supermanifolds and Wittens topological invariants
expressed as path integrals of a topological QFT. All this constitutes only a
small sample of inventions which are by now thoroughly absorbed and transformed into honest mathematics.
It is only when one has to deal with innitary constructions, that is,
limits of various kinds, that mathematicians do their job unassisted. According to Bourbakis chapters on integration [Bourbaki 1974], mathematicians
contributed to the theory of integral in the last century exclusively careful
analysis of limits.
After the creation of the modern notion of a topological space and the
discovery of limiting procedures basic to measure theory, the next major package of startlingly new innitary constructions was introduced by Alexander
Grothendieck with his treatment of homological algebra, derived categories
and functors, topos and sites. But this is another story.
5.
Discussion
Direct contact between mathematical and physical modes of thought more

often than not creates a tension. The basic values are dierent, the accepted
types of social behavior clash, time scales for a problem to keep attention of the
public tend to be incommensurable.4 In a remarkable piece of introspection,
Dyson [1972] has shown how impenetrable the walls between mathematics
and physics can be in one and the same mind. We would be much more
tolerant to each other if we could discern in ourselves the two personalities so
convincingly displayed by Dyson. A recent discussion (cf. [Jae and Quinn
1993, 1994]) shows the vulnerability of our community, when in a period of
renewed fruitful interaction we try to harmonize our attitudes to what is and
what is not a proof, what may and what may not be published, and who
4
The relevant psychological diculties are not often expressed in print. For an interesting
recent reaction see S. Mac Lanes contribution in [Jae and Quinn 1994] of which we cite
only one sentence: Thus, when I attended a conference to understand the use of a small
result of mine, I heard lectures about topological quantum eld theory, without a slightest
whi of a denition; I was told that the notion had cropped up at some prior conference, so
that Everybody knew it.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
166
YU. I. MANIN
should be credited for what.

All of this is fortunately restricted to our social life. It seems that deep
insights survive however we mess them up, and it is precisely the complementarity of mathematical and physical thinking that makes their interaction
creative.
The crucial distinction between the ways we present our ideas in the last
half of this century lies not so much in our attitudes towards a rigorous proof
as towards exact denitions.
Mathematicians have developed a very precise common language for saying whatever they want to say. This precision is embodied rst of all in the
denitions of the objects they work with, stated usually in the framework
of a more or less axiomatic set (or category) theory, and in the skillful use
of metalanguage (which our natural languages provide) to qualify the statements. All the other vehicles of mathematical rigor are secondary, even that
of rigorous proof. In fact, barring direct mistakes, the most crucial diculty
with checking a proof lies usually in the insuciency of denitions (or lack
thereof). In plain words, we are more deeply troubled when we wonder what
the author wants to say than when we do not quite see whether what he or
she is saying is correct. The aws in the argument in a strictly dened environment are quite detectable. Good mathematics might well be written down
at a stage when proofs are incomplete or missing, but informed guesses can
already form a fascinating system: outstanding instances are A. Weils conjectures and Langlandss program, but there are many examples on a lesser
scale.
The etymology of the term denitio shows that its primary function is
to set strict limits. In the course of a given study, we agree to consider only
locally compact topological spaces satisfying the countability condition, only
nitedimensional Lie algebras, only coarse moduli spaces of stable algebraic
curves and so on. If we fail to mention a relevant restriction in the course of
presenting a professional seminar, we will be politely reminded about it. If we
claim to having done anything serious, our work will be carefully scrutinized
for all the necessary caveats.
Of course, our denitions are far from being arbitrary. One function of a
good denition is to be a carrier of analogies between various situations, and
to this end the cage of a denition must be of optimal size. For example, one
can convincingly argue that by far the most important result of the group
theory is exactly the denition of an abstract group and its action on a set,
because it describes a structure reappearing again and again in geometry,
number theory, probability, the theory of spacetime, theory of elementary
particles, and so on. The whole ideology of Bourbakis treatise consists in
` 3
SEMINAIRES
ET CONGRES
167
representation of mathematics as a building supported by a strict system of

good denitions (axioms of basic structures). And since a good denition is
sometimes the work of generations of good mathematicians, the temptation
to believe that we already know them all can be great.
To the contrary, an inexperienced reader of the most interesting physical papers is often left in a vacuum about the precise meaning of the most
common terms. Physicists are undoubtedly constrained by their own rules,
but these rules are not ours. What is a current algebra, a supersymmetry
transformation, a topological eld theory, a path integral, nally? They are
very open concepts, and it is precisely their openness that makes them so
interesting.
Here is what the history of our two metiers teaches: we cannot live without
each other. At least for some of us, life becomes dull if it goes on for too long
without contacts with good physics.
It is the interaction with a wildly diering set of values that counts most.
As a perceptive study by Hardy Grant [1995] shows, in terms of cultural
history of Isaiah Berlins variety, mathematics is a very classical endeavour.
In fact, it is based upon a commonly accepted idea of truth and ways to
achieve it, forming a stable system. The Romantic Revolution of a century
and a half ago did not really inuence mathematics mainly because there was
little place in it for personal whims.
In this century romantics comes from physics: the vast expanses of the
Universe, the wonderfully erratic behavior of the microworld, the observers
subjectivism and the power of the unobservable, the Big Bang, the Anthropic
Principle, our in turn humble and megalomaniacal attempts to cope with
irreverent Nature.
Mathematics supplies hygienic habits and headaches.
Bibliography
Atiyah (M.F.)
[1989]
Topological quantum eld theories, Inst. Hautes Etudes

Sci. Publ. Math.,
68 (1989), pp. 175186.
Blanchet (C.), Habegger (N.), Massbaum (G.) and Vogel (P.)

[1995]
Topological quantum eld theories derived from the Kaufman bracket,

Topology, 34-4 (1995), pp. 883927.
Bourbaki (N.)
[1974]
Elements
dhistoire des math
ematiques, ed. rev. Paris : Hermann 1974.
English transl. Heidelberg: Springer, 1994.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
168
YU. I. MANIN
Candelas (P.), de la Ossa (X.), Green (P.) and Parkes (L.)

[1991]
A pair of CalabiYau manifolds as an exactly soluble superconformal

theory, Nuclear Phys., 359 (1991), pp. 2174.
Dyson (F.)
[1972]
Missed opportunities, Bull. Amer. Math. Soc., 78 (1972), pp. 635652.
Feynman (R.P.)
[1948]
The space-time approach to non-relativistic quantum mechanics, Rev.

Mod. Phys., 20 (1948), pp. 367387.
[1988]
QED. The Strange Theory of Light and Matter. Princeton Univ. Press,
1988.
Glimm (J.) and Jaffe (A.)

[1981]
Quantum physics. A functional integral point of view. New York - Heidelberg - Berlin: Springer, 1981.
Grant (H.)
[1995]
What is modern about modern mathematics?, Math. Intelligencer, 17-3

(1995), pp. 6266.
Hardy (G.H.)
[1929]
Mathematical proof, Mind, XXXVIII-149 (1929), pp. 125.
Jaffe (A.) and Quinn (F.)

[1993]
Theoretical mathematics: toward a cultural synthesis of mathematics and

theoretical physics, Bull. Amer. Math. Soc., (N.S.) 29 (1993), pp. 113.
[1994]
Response to comments on theoretical mathematics, Bull. Amer. Math.

Soc., (N.S.) 30 (1994), pp. 208211.
Reshetikhin (N.) and Turaev (V.)

[1991]
Invariants of 3-manifolds via link polynomials and quantum groups,

Invent. Math., 103 (1991), pp. 547597.
Witten (E.)
[1989]
Quantum eld theory and the Jones polynomial., Comm. Math. Phys.,
121 (1989), pp. 351399.
` 3
SEMINAIRES
ET CONGRES
Probl`
emes de similarit
e pour les
op
erateurs sur lespace de Hilbert
Gilles Pisier
R
esum
e
Deux operateurs T1 et T2 sur un espace de Hilbert H sont dits
semblables sil existe un operateur inversible S sur H tel que
T1 = S 1 T2 S. Un problème classique (le problème de Halmos ) demande de caracteriser les operateurs T semblables à une
contraction. Deux autres variantes de ce problème sont apparues :
lune, proposee par Dixmier en 1950 concerne la caracterisation
des groupes moyennables, lautre, due a` Kadison en 1955, les representations de C -algèbres. Cet article tente de faire le point
sur ces problèmes.
Abstract
Two operators T1 , T2 on a Hilbert space are called similar if there
exists an invertible operator S on H such that T1 = S 1 T2 S.
A classical problem (the Halmos problem) asks for a characterization of the operators T which are similar to a contraction. Two important variants have appeared. One proposed by
Dixmier in 1950 asks whether the similarity property for all uniformly bounded representations on a group G is equivalent to the
amenability of G. The other one, proposed by Kadison in 1995,
is about representations of a C -algebra. This report attempts
to describe the genesis and the present state of these problems.
Cet expose est consacre à plusieurs problèmes de similarite pour les operateurs lineaires bornes T : H H sur un espace de Hilbert H. On note
AMS 1991 Mathematics Subject Classication: 01A60, 17B10, 22E46
Equipe dAnalyse - URA 754-CNRS Universite Pierre et Marie Curie - Paris 6, Bote 186,
4 place Jussieu 75252 Paris Cedex 5
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
170
G. PISIER
B(H) lalgèbre formee de ces operateurs munie de la norme usuelle et GL(H)

le sous-groupe des elements inversibles.
D
enition. Deux operateurs T1 , T2 dans B(H) sont dits semblables
sil existe S inversible dans B(H) tel que S 1 T1 S = T2 .
Comme pour les matrices, cela revient à dire quil existe un changement
de base (oblique, pas orthonormale) qui transforme T1 en T2 . Plus generalement, si (T1i )iI et (T2i )iI sont deux familles dans B(H), on dit quelle sont
semblables sil existe S inversible dans B(H) tel que S 1 T1i S = T2i pour tout
i dans I.
En theorie des operateurs sur lespace de Hilbert, on appelle contraction
tout operateur T tel que
T 1.
Le principal problème qui nous interesse ici est de reconnatre les operateurs
T qui sont semblables à une contraction ou bien les familles doperateurs qui
sont semblables à une famille de contractions.
Ce problème est apparu sous diverses formes à partir de 1946 dabord
pour des groupes doperateurs ou bien pour des semi-groupes, le cas dun seul
operateur T correspondant aux groupes (ou semi-groupes) engendres par un
unique generateur. Parallèlement, sont apparues des versions linearisees
de ces questions pour les sous-algèbres de B(H). Nous proposons ici une presentation retrospective de ces problèmes (pour la plupart toujours ouverts) en
nous attachant a` souligner les relations existant entre eux et leur connections
avec dautres theories. De plus, nous introduisons le concept dapplication
complètement bornee , qui est lun des principaux nouveaux outils que les
recherches sur ces questions ont contribue à degager.
La theorie des operateurs sur lespace de Hilbert est un sujet très vaste,
surtout si lon y ajoute la theorie des algèbres doperateurs (qui, bien que
proche de la precedente, correspond en fait a` un groupe de chercheurs très
dierent). Nous ne ferons queeurer ces theories. Pour la première, nous
renvoyons le lecteur au livre classique [Sz.-Nagy et Foias 1970] que lon peut
considerer comme fondateur de la theorie contemporaine des operateurs,
celle que lon trouve developpee dans le Journal of Operator Theory, cree en
Roumanie en 1979. Voir aussi la cinquantaine de volumes de la serie Operator
Theory: Advances and applications et la revue Integral equations and operator theory (depuis 1979) publies par Birkhauser et diriges par I. Gohberg
a Tel Aviv. Pour les algèbres doperateurs, les references classiques sont les
`
ebres et algèbres de von Neumann)
livres bien connus de J. Dixmier (C -alg`
eux-memes fortement inuences par la serie darticles (On rings of operators)
publies par Murray et von Neumann dans Annals of Mathematics entre 1936
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
171
et 1943. Pour une reference generale plus recente, voir [Kadison et Ringrose
1986] et aussi, dans une autre direction, [Connes 1990].
Le problème de similarite pour les groupes remonte a` Sz.-Nagy [1946] et
Dixmier [1950]. Pour les operateurs individuels les references initiales sont
[Sz.-Nagy 1959] et [Halmos 1970]. Enn pour les C -algèbres, le problème
remonte à Kadison [1955].
Dans le texte qui va suivre, nous traitons dabord le cas des groupes doperateurs plonges dans B(H) (1), puis le cas dun operateur individuel ou dun
semi-groupe (2) et enn nous considerons un problème analogue pour les
homomorphismes sur une C -algèbre ( 3). Enn, nous montrons au 4 comment la theorie recente des applications complètement bornees permet
dunier ces dierents problèmes.
Nous renvoyons le lecteur a` notre livre recent [Pisier 1995] pour plus de
details.
1.
Groupes
Le resultat qui suit est a` la racine de toutes les etudes ulterieures sur les
operateurs semblables.
Th
eor`
eme 1.1. [Sz.-Nagy 1946] Soit T un operateur inversible dans B(H).
Les assertions suivantes sont equivalentes :
` une contraction.
i. T et T 1 sont tous deux semblables a
ii. T est semblable a
` un operateur unitaire.
iii. sup T n < .
Remarque 1.2. Il est important de noter que si T est inversible, dire

que T et T 1 sont tous deux des contractions revient a` dire que T est unitaire
(= inversible isometrique).
Preuve. Les implications (ii) (i) (iii) sont triviales. Reste à montrer
(iii) (ii). Supposons (iii). Soit C = sup{T n | n Z}. Sur H on denit
une nouvelle norme hilbertienne comme suit
N
1 k 2
T h
h H |||h||| = lim
u N
2
k=1
o`
u u est un ultraltre non trivial sur N (Sz.-Nagy utilisait ici une limite de
Banach plus classique !). On a evidemment C1 h |||h||| Ch et de
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
172
G. PISIER
plus |||T h||| |||h||| pour tout h. (Noter que cest la condition sur les inverses
de T qui garantit la nondegenerescence, cest-à-dire que h C|||h|||. Sans
cette condition, on risquerait de trouver |||h||| 0 !)
La generalisation suivante est due a` Dixmier (et aussi, independamment
a Day [1950]) qui sest rendu compte quil sagissait dun phenomène d
`
u a` la
moyennabilite de Z (voir la denition plus loin).
Th
eor`
eme 1.3. [Dixmier 1950] Soit G un groupe localement compact.
Soit : G GL(H) une repr
esentation continue (pour la topologie forte des
operateurs sur H) et uniform
ement born
ee, i.e. telle que
sup (t)B(H) < .
tG
On pose
|| = sup (t)B(H) .
tG
Alors si G est moyennable, il existe S : H H inversible avec S S 1

esentation unitaire. On dit alors que
||2 tel que t S 1 (t)S soit une repr
est unitarisable .
La notion de groupe moyennable remonte semble-t-il a` von Neumann
[1929] dont le nom apparat de facon fondamentale dans tous les aspects des
problèmes consideres dans cet expose.
Un groupe localement compact G est dit moyennable sil existe une forme
lineaire positive : L (G) C avec = (1) = 1 qui est invariante
a gauche, i.e. telle que (b) = (t b) pour tout t dans G. Cela inclut
`
a la fois tous les groupes commutatifs (en particulier Z) et tous les groupes
`
compacts. Si G est discret, cela veut dire quil admet un analogue des limites
de Banach . Voir [Pier 1984] pour plus dinformations.
Larticle de Dixmier se termine par les deux questions suivantes :
Probl`
eme 1.4. [Dixmier 1950] Existe-t-il des groupes G admettant une
repr
esentation uniformement born
ee : G GL(H) mais non unitarisable ?
Probl`
eme 1.5. [Dixmier 1950] Si oui, la classe de tels groupes est-elle
exactement la classe des groupes non moyennables ? (En dautres termes, le
Th
eor`
eme 1.3. admet-il une r
eciproque ?)
Un exemple typique de groupe discret non moyennable est le groupe libre
a N 2 generateurs. Plus generalement, tout groupe discret contenant
FN `
F2 comme sous-groupe nest pas moyennable.
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
173
Le problème 1.4. a ete resolu par Ehrenpreis et Mautner [1955] qui ont
montre que G = SL2 (R) est un contre-exemple. Leur travail fut ensuite clarie et considerablement amplie par Kunze et Stein [1960]. Ces derniers
construisent en fait toute une famille analytique {z | 0 < Re z < 1} de representations uniformement bornees qui ne sont unitarisables que si Re z = 12
ou bien si z R. Pour des extensions ulterieures, voir [Sally 1968].
Larticle Kunze et Stein [1960] contient aussi des applications remarquables aux convoluteurs bornes de Lp (G) pour G = SL2 (R). Cest ce quon
appelle le phenomène de Kunze-Stein : pour f Lp (G), g L2 (G) on a
f g L2 (G) si 1 p < 2. Ce dernier resultat nest pas valable pour G = R.
Pour des resultats plus recents sur ce phenomène, voir [Cowling 1978, 1982].
Bien entendu, puisque SL2 (R) est un contre-exemple au problème 1.4., a
fortiori SL2 (R) muni de la topologie discrète en est aussi un, et comme tout
groupe est un quotient dun groupe libre, il en resulte que les groupes libres
(et plus precisement FN pour tout N 2) fournissent des contre-exemples.
Il etait donc naturel de chercher des constructions explicites, aussi simples
que possibles, de representations uniformement bornees et non unitarisables
sur FN (N 2). Cest lobjet de la serie de publications : Mantero et Zappa
[1983], Fig`
a-Talamanca et Picardello [1983], Bozejko et Fendler [1991], Pytlik
et Szwarc [1986], Bozejko [1987b], Szwarc [1988], Fendler [1990], Wysoczanski
[1993].
Pour une autre approche, voir [Valette 1990b,a]. Lanalyse harmonique sur
le groupe libre est developpee dans [Cartier 1973], puis [Figà-Talamanca et
Picardello 1983].
Signalons aussi que Jean Dieudonne lui-meme sest interesse aux groupes
moyennables, voir Dieudonne [1960].
Le problème 1.5. reste ouvert en toute generalite. Neanmoins daprès un
argument classique d induction des representations (qui setend aisement
au cas uniformement borne), tout groupe G contenant F2 comme sous-groupe
admet une representation uniformement bornee non unitarisable. On a longtemps cru (cetait un problème fameux pose par von Neumann dès 1929) que
tout groupe non moyennable contenait F2 comme sous-groupe (ce qui aurait
resolu armativement le problème 1.5.). Cest vrai pour les groupes lineaires,
daprès un theorème classique de Tits [1972], mais faux en general : en effet Olshanskii [1980] a construit un groupe non moyennable dans lequel tout
sous-groupe non trivial est une copie de Z. Son travail est très proche des
contre-exemples de Novikov et Adian [1968] sur le problème de Burnside (cf.
[Adian 1979]).
Le moins quon puisse dire est que ces exemples sont diciles à comprendre
(voir [Paterson 1988] pour une description esquissee).
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
174
G. PISIER
Plus recemment, Gromov [1987] a construit par une methode dierente des
groupes innis possedant la propriete T de Kazhdan (donc non moyennables
en un sens extreme), mais sans sous-groupe libre.
Remarque 1.6. Daprès Vasilescu et Zsido [1974], toute representation
uniformement bornee : G B(H) sur un groupe arbitraire, dont limage
est une algèbre de von Neumann nie est unitarisable.
Le resultat suivant d
u a` lauteur (non encore publie) fournit une reponse
partielle au problème 1.5..
Th
eor`
eme 1.7. Soit G un groupe discret. Les assertions suivantes sont
equivalentes.
i.
Le groupe G est moyennable.
ii. Pour tout H , pour toute repr

ee : G
1
B(H) il existe S : H H inversible avec SS ||2 tel que
S 1 (.)S est une representation unitaire.
iii. Il existe une constante K et < 3 tels que pour tout H et toute repr
ee : G B(H) il existe S inversible
avec S 1 S K|| tel que S 1 (.)S est unitaire.
Remarque 1.8. Pour resoudre complètement le problème 1.5. (dans le
cas discret) il surait de savoir montrer que (iii) (i) en remplacant dans
(iii) la fonction t Kt par une fonction quelconque.
On peut formuler une variante du problème 1.5. à laide de lespace des
coecients des representations : soit G un groupe localement compact, on
note B(G) lespace des coecients des representations unitaires (continues)
de G, cest-à-dire lensemble des fonctions f : G C pour lesquelles il existe
une representation unitaire continue : G B(H) et , dans H tel que
(1.1)
t G
f (t) = (t), .
Noter que si = , on obtient une fonction continue de type positif sur G.

Lespace B(G) concide donc (par polarisation) avec lensemble des combinaisons lineaires de telles fonctions, en dautres termes f B(G) si et seulement
si f peut secrire f = f1 f2 + i(f3 f4 ) avec f1 , . . . , f4 continues de type
positif sur G.
On munit lespace B(G) de la norme
f B(G) = inf{ }
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
175
o`
u linmum porte sur toutes les decompositions possibles de la forme (1.1).
Comme il est bien connu, B(G) (muni de cette norme) est une algèbre de Banach pour le produit ponctuel des fonctions sur G (qui correspond au produit
tensoriel des representations). Dans le cas de groupes compacts non commutatifs, la thèse dEymard [1964] contient une etude detaillee de cet espace, vu
comme le dual de la C -algèbre du groupe G (i.e. celle qui est engendree par
la representation unitaire universelle de G).
Par analogie, on peut introduire lespace Bub (G) des coecients des representations uniformement bornees de G , i.e. lensemble des f de la forme (1.1)
avec representation uniformement bornee. Cet espace est encore une algèbre
mais na pas de norme naturelle. En revanche, si lon xe c 1, lensemble
Bc (G) forme des fonctions f de la forme (1.1) avec || c est un espace
de Banach pour la norme f = inf{}. Mais alors on a seulement :
f Bc (G), g Bd (G) f.g Bcd (G). Si toutes les representations uniformement bornees sont unitarisables, alors B(G) = Bub (G). Le problème
suivant, ouvert en toute generalite, est donc a priori plus dicile que le problème 1.5..
Probl`
eme 1.9. Si B(G) = Bub (G), est-ce que G est moyennable ?
Remarque 1.10. La demonstration de (iii) (i) dans le theorème 1.7.
montre en fait que G est moyennable sil existe K et < 3 tels que pour tout
c > 1 assez grand on a linegalite f B(G) K C f Bc (G) pour tout f dans
Bc (G).
La diculte de ce problème tient au manque dune description maniable
des elements de Bub (G) (ou bien de Bc (G) pour c > 1). En revanche, lanalyse
harmonique fournit plusieurs classes naturelles de fonctions sur G, notons une
telle classe F (G), telles que F (G) Bub (G) et telles que F (G) = B(G) si et
seulement si G est moyennable. Un exemple dune telle classe est donne par
lensemble M0 (G) des multiplicateurs dits de Herz-Schur de G, cest-à-dire
lensemble des fonctions f : G C pour lesquelles il existe un Hilbert H et
des fonctions bornees x : G H et y : G H telles que
(1.2)
s, t G f (st) = xs , yt .
On le munit de la norme
f M0 (G) = inf{sup xs H sup yt H }
s
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
176
G. PISIER
o`
u linmum porte sur toutes les decompositions possibles de f comme en (1.2).
On a evidemment Bub (G) M0 (G) B(G). On peut montrer que, si
G est discret, legalite B(G) = M0 (G) caracterise les groupes moyennables
(cf. [Bozejko 1985]). Signalons toutefois que, daprès Haagerup [1985], on a
M0 (G) = Bub (G) si G = FN avec N 2.
Pour dautres illustrations voir [Nebbia 1982] et [Losert 1984]. Bien entendu, l`
a encore, si G contient F2 , on sait que le problème 1.9. a une reponse
armative. On peut donner (suivant Pytlik et Szwarc [1986]) un exemple particulièrement simple pour G = F (groupe libre a` une innite denombrable
de generateurs). Pour x G, soit |x| la longueur dun mot (reduit) element
de F . Soit Wn = {x F | |x| = n}. Soit z C avec |z| < 1. On pose
z (x) = z |x| ,
alors z Bub (G), mais z
/ B(G) si z
/ R. Noter que lon a

z n 1Wn ,
z =
n0
do`
u un exemple encore plus simple
1W1 Bub (G)
mais
1W1
/ B(G),
et de meme pour Wn pour tout n 2. En fait, toutes les fonctions spheriques sont dans Bub (G) quand G = Fn ; voir [Mantero et Zappa 1983] et
[Szwarc 1988].
2.
Op
erateurs individuels. Alg`
ebres uniformes
A la suite du theoreme 1.1., Sz.-Nagy a cherche une caracterisation des

operateurs T B(H) semblables à une contraction. Une condition evidemment necessaire est que T soit à puissances bornees i.e. que lon ait
supn1 T n < . Ce qui la conduit [Sz.-Nagy 1959] a` demander si cette
condition est aussi susante. La reponse est armative si T est compact
[Sz.-Nagy 1959] ou bien si r(T ) < 1 [Rota 1960], mais on sait depuis Foguel [1964] quen general la reponse est negative. En eet, ce dernier a donne
le premier exemple doperateur a` puissances bornees non semblable a` une
contraction. (Pour dautres exemples voir [Peller 1982] et [Bozejko 1987a]).
Larticle de Rota [1960] contient aussi une jolie formule pour le rayon spectral
r(T ) dun operateur arbitraire T sur un Hilbert H, comme suit
r(T ) = inf{S 1 T S | S GL(H)}.
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
177
Il resulte de cette formule que, pour tout * > 0, tout operateur a` puissances
bornees (donc tel que r(T ) 1) est semblable à un operateur de norme
< 1 + *. En revanche, la question suivante est apparemment toujours ouverte
(cf. [Peller 1982]) : soit T B(H) a` puissances bornees et soit * > 0, T est-il
semblable à un operateur a` puissances bornees par 1 + * ?
Immediatement après le contre-exemple de Foguel, on sest rendu compte
quon pouvait remplacer etre à puissances bornees par une condition necessaire plus forte faisant intervenir des polynomes P (T ) = a0 IH + a1 T + +
an T n (avec a0 , a1 , . . . , an C) en T et non plus seulement des monomes. En
eet, daprès une celèbre inegalite de von Neumann [1951], toute contraction
T B(H) verie
(vN)
P polyn
ome P (T ) P = sup{|P (z)| | z D},
o`
u D = {z C | |z| < 1}. Si T est seulement semblable à une contracu si
tion, i.e. si T = S 1 TS avec T 1, alors P (T ) = S 1 P (T)S do`
1
C = S S
P polyn
ome P (T ) CP .
(2.1)
On appelle polynomialement borne tout operateur T veriant (2.1)

pour une constante C. Comme on vient de le voir, semblable à une contraction implique polynomialement borne . On sait depuis Lebow [1968] que
lexemple de Foguel nest pas polynomialement borne, do`
u la question suivante posee par Halmos [1970] : tout operateur polynomialement borne est-il
semblable à une contraction ? Ce problème est reste ouvert jusqu`
a très recemment. Curieusement, ce nest que quelques heures avant ma conference à Nice
que je me suis convaincu que ma dernière tentative etait concluante : pour
tout C > 1, il existe un operateur T veriant (2.1) mais non semblable à une
contraction (voir [Pisier 1997]). Plus precisement, ce dernier article contient
aussi une version ni-dimensionnelle qui senonce comme suit. Il existe une
constante > 0 pour laquelle la propriete suivante a lieu : pour tout n et tout
* > 0, il existe Tn B(.2n ) (i.e. Tn est si lon veut une matrice n x n agissant
sur lespace de Hilbert de dimension n, note ici .2n ) veriant
P polyn
ome P (Tn ) (1 + *)P
mais tel que
*
log(n + 1) inf{SS 1 | S 1 T S 1}.

E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
178
G. PISIER
Cette minoration est motivee par la majoration suivante due a` Bourgain [1986]
qui est valable pour tout n, pour tout T dans B(.2n )
inf{S 1 S ; S 1 T S 1} K(1 + *)4 log(n + 1)
o`
u K est une constante numerique (independante de n ou de T ). Bien entendu,
en prenant la somme directe sur n des operateurs Tn , on obtient une reponse
negative a` la question precedente de Halmos. En revanche, la question suivante
(analogue `
a celle de Peller dej`
a mentionnee) reste ouverte : soit * > 0, un
operateur polynomialement borne de constante arbitraire est-il semblable a`
un operateur polynomialement borne de constante 1 + * ?
Linegalite de von Neumann (vN) a suscite de nombreux travaux. On
ignore toujours si sa validite caracterise les algèbres de Banach à unite plongeables (comme sous-algèbres de Banach) dans B(H). Voir a` ce sujet [Dixon
1995] et les references qui sy trouvent.
Dans une autre direction, il est naturel de chercher a` etendre (vN) au cas de
polyn
omes à plusieurs variables, do`
u la question suivante : soit n 2 et soient
a-deux (i.e. Ti Tj = Tj Ti i = j),
T1 , . . . , Tn , n contractions commutant deux-`
est-il vrai que pour tout polyn
ome P (z1 , . . . , zn ) on a
(2.2)
P (T1 , . . . , Tn ) sup{|P (z1 , . . . , zn )| | z1 , . . . , zn D} ?
Daprès Ando [1963], la reponse est oui si n = 2, le cas n = 3 est

reste ouvert jusqu`
a Varopoulos [1974] (suivi de Crabb et Davie [1975]) qui
montre que cest non pour n > 2. Dans le contre-exemple de lappendice
de [Varopoulos 1974] le polyn
ome est homogène de degre 2 et dim(H) = 5 ;
dans celui de [Crabb et Davie 1975] (particulièrement simple) le degre est 3
et dim(H) = 8. Signalons un problème fort embarrassant qui reste ouvert :
existe-t-il une constante Cn telle que dans la situation de (2.2) on ait pour
tout polyn
ome P `
a n variables
P (T1 , . . . , Tn ) Cn sup{|P (z1 , . . . , zn )| | zi D} ?
Varopoulos [1974] a etabli que C3 > 1, mais on ignore en fait si C3 <
ou bien si Cn < pour un n > 2 ! En revanche, il resulte de C3 > 1 que
Cn quand n , car on a Cn+m Cn Cm pour tous n, m.
En fait, linegalite de von Neumann peut etre vue comme consequence
dun theorème fondamental de dilatation d
u a` Sz.-Nagy [1953] :

Th
eor`
eme 2.1. Pour toute contraction T sur H, il existe un Hilbert H
contenant H comme sous-espace et un operateur unitaire
H

U :H
` 3
SEMINAIRES
ET CONGRES
tel que
T n = PH U n |H
PROBLEMES
DE SIMILARITE
179
pour tout n 1, donc aussi P (T ) = PH P (U )|H pour tout polyn

ome P . On
dit alors que U est une dilatation de T .
Noter que (vN) en est un corollaire immediat, puisque la theorie spectrale
(noter que U est unitaire donc normal) nous assure que
P (U = sup{|P (z)| | z (U )} P .
On a donc P (T ) P (U ) P . Ando [1963] demontre une extension
H
de ce theorème : pour deux contractions qui commutent, il construit H
et qui sont des dilatations
et deux unitaires U1 , U2 qui commutent sur H
a-encore, (2.2) pour n = 2 en
respectivement de T1 et T2 . Bien entendu, l`
resulte. Les contre-exemples pour n > 2 montrent que le theorème de dilatation dAndo nest plus valable pour n contractions qui commutent si
n > 2. Plus precisement, Parrott [1970] a construit explicitement trois contracegalit
e de
tions (T1 , T2 , T3 ) dans B(H), commutant entre elles et veriant lin
von Neumann, mais neanmoins nadmettant pas de dilatation en trois unitaires qui commutent. Cet exemple a joue un r
ole important dans la theorie
par la suite.
Ces theorèmes de dilatation nous rapprochent dun autre thème classique
en theorie des operateurs, celui des sous-espaces invariants. En general, dans
nest pas
les resultats de dilatation comme ci-dessus, le sous-espace H H
invariant pour loperateur U (ou pour U1 , U2 ), mais neanmoins la correspondance P PH P (U )|H est un homomorphisme dalgèbre. Il est donc naturel
qui possèdent cette prode se demander quels sont les sous-espaces H H
priete. Le resultat suivant de Sarason [1965] donne une reponse très complète
et très elegante a` cette interrogation.
Th
eor`
eme 2.2. Soit H un Hilbert. Soit A une algèbre unif`
ere et soit
: A B(H) un homomorphisme dalg`
ebre. Les propri
etes suivantes dun
sous-espace ferme H H sont
equivalentes.
i. Lapplication a PH (a)|H est un homomorphisme de A dans B(H).
ii. Il existe deux sous-espaces H2 H1 H invariants par (A) (i.e. tels
que (a)Hi Hi a A) tels que H = H1 H2 (i.e. H = H1 (H2 ) ).
Les sous-espaces H veriant les conditions precedentes sont appeles semiinvariants . Il est important de souligner que lenonce precedent applique
dans la situation du Theorème 2.1. (avec pour A lalgèbre des polyn
omes)
nous dit que toute contraction T B(H) peut secrire PH1 H2 U|H1 H2 avec
H1 , H2 invariants pour U . Par consequent, les sous-espaces invariants de T
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
180
G. PISIER
sont tous de la forme H3 H2 avec H3 U -invariant tel que H2 H3 H1 .

Cela donne une connexion remarquable avec le problème classique des sousespaces invariants, qui est bien s
ur toujours ouvert : tout operateur T de
B(H) possède-t-il un sous-espace ferme invariant non trivial ? En eet, ce
problème se ramène evidemment au cas des contractions. Par ailleurs, le cas
des operateurs unitaires ou normaux ne pose aucun problème, puisque la
theorie spectrale classique sapplique. On sait donc bien s
ur que tout unitaire
U dans B(H) admet des sous-espaces invariants non triviaux, mais on ignore
si etant donnes deux sous-espaces (fermes) invariants H2 H1 avec H1 = H2 ,
H3 H1 . En eet, cela impliquerait
il en existe un troisième H3 veriant H2 =
=
(daprès les theorèmes 2.1. et 2.2.) que
T = PH1 H2 U|H1 H2
(donc une contraction arbitraire) admet un sous-espace invariant non trivial.
Remarque 2.3. Von Neumann [1951] a introduit les ensembles spectraux
pour un operateur T B(H). Il sagit des compacts K C contenant le
spectre de T et tels que pour toute fonction rationnelle R à p
oles hors de K
on a
R(T ) sup |R(z)|.
zK
Plusieurs questions considerees ci-dessus setendent dans ce cadre. Voir

[Sz.-Nagy 1958], [Paulsen 1986] pour plus dinformations.
Remarque 2.4. La notion de dilatation (ou bien son inverse la
compression ) setend naturellement aux homomorphismes dalgèbres : Soit
une algèbre arbitraire et A A
une sous-algèbre. Considerons un homoA
B(H)
est
morphisme : A B(H). On dit quun homomorphisme
:A
tel
une dilatation de sil existe un plongement (isometrique) de H dans H
que
(a)|H .
a A (a) = PH
On dit aussi alors que est la compression de
de sur H. Daprès ce
qui precède H est necessairement semi-invariant pour
(A).
La plupart des questions precedentes peuvent etre etudiees dans le cadre
des algèbres uniformes . On appelle ainsi toute sous-algèbre unitale fermee
de lalgèbre C(K) des fonctions continues sur un compact K. En general on
suppose que A separe les points de K, on dit alors que A est propre si
A = C(K). Par exemple, lalgèbre du disque est une algèbre uniforme propre
dans C(D).
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
181
Les questions que nous venons daborder conduisent naturellement a` envisager pour une algèbre uniforme A C(K) la propriete suivante : pour tout
homomorphisme unital borne u : A B(H), il existe S dans GL(H) tel que
a S 1 u(a)S est de norme 1. Cette propriete est veriee si A = C(K) par
exemple parce que A est une C -algèbre nucleaire (voir le paragraphe suivant).
Curieusement, jusqu`
a très recemment, on ne connaissait aucun exemple dalgèbre uniforme propre soit possedant cette propriete, soit ne la possedant pas !
La reponse negative a` la question de Halmos permet maintenant darmer
que lalgèbre du disque ne possède pas cette propriete, il en est donc de meme
pour le polydisque et il devient concevable quaucune algèbre uniforme propre
ne la possède.
Signalons que les theorèmes de dilatation mettent en evidence la sousclasse des algèbres uniformes A C(K) possedant la propriete suivante : tout
homomorphisme contractant u : A B(H) admet une dilatation
(cf. Remarque 2.4.) qui est encore un homomorphisme
u
: C(K) B(H)
contractant.
Daprès ce qui precède, A(D) et A(D 2 ) possèdent cette propriete mais pas
A(D n ) pour n > 2. Les domaines du plan complexe Cn , pour lesquels
lalgèbre uniforme A() (formee des fonctions analytiques continues sur )
possède cette propriete, ne sont pas bien compris. Nous renvoyons au livre
[Douglas et Paulsen 1989] pour une etude approfondie de ce genre de questions, en liaison avec la topologie algebrique.
3.
C -alg`
ebres
Rappelons que toute C -algèbre A peut etre realisee comme une sousalgèbre involutive fermee de B(H). Si A est unitale, on peut realiser A avec
IH A. Les morphismes adaptes aux C -algèbres sont appeles des representations (C -representations si lon veut preciser). Une representation : A1 A2
entre deux C -algèbres est donc une application lineaire telle que
x, y A (xy) = (x)(y) et (x ) = (x) .
Une particularite des C -algèbres est que la structure dalgèbre determine la
norme. Si A est une C -algèbre, elle nadmet quune seule norme pour laquelle
elle est une C -algèbre. Autrement dit : toute representation injective (entre
C -algèbres) est automatiquement isometrique et, plus generalement, toute
representation : A1 A2 a necessairement une norme au plus egale à
1. On peut aussi decrire les C -algèbres en termes de groupes. Soit U (H) (resp.
U (A)) lensemble des elements unitaires de B(H) (resp. A). Soit G U (H)
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
182
G. PISIER
un sous-groupe, alors le sous-espace ferme engendre par G dans B(H) est une
C -algèbre à unite et toute sous-C -algèbre de B(H) contenant lunite est de
cette forme. De plus, daprès un theorème de Russo et Dye [1966], si A est
unitale, la boule unite fermee de A est lenveloppe convexe fermee de U(A).
Supposons A realisee dans B(H) et posons G = U (A) U (H). Alors,
les representations : A B(H) sont exactement les representations unitaires de G dans B(H) qui setendent en une application sur le sous-espace
lineairement engendre par le groupe G dans B(H). De plus, le theorème de
Russo-Dye nous permet decrire
= ||G |.
En resume, se donner une representation de C -algèbre A revient a` se donner
une representation unitaire de U (A) respectant les relations lineaires.
Soient A1 , A2 deux C -algèbres. Convenons dappeler homomorphisme
toute application lineaire u : A1 A2 telle que
u(xy) = u(x)u(y)
x, y A1
(autrement dit, il sagit des morphismes dalgèbres). Le problème de similarite

pour les C -algèbres a ete pose dès 1955 par Kadison. Il sagit du problème
suivant :
ebre a
` unite. Soit u : A B(H) un
Probl`
eme 3.1. Soit A une C -alg`
homomorphisme borne tel que u(1) = 1. Existe-t-il S : H H inversible tel
esentation ?
que lhomomorphisme u
(a) = S 1 u(a)S soit une C -repr
Sil existe un tel S, nous dirons suivant le terminologie de Kadison que u est
orthogonalisable .
Remarques 3.2.
i. Dire que u est orthogonalisable revient a` dire que la restriction de u
a U (A) est unitarisable au sens du 1.
`
ii. Dautre part, comme les unitaires sont exactement les contractions inversibles dinverse contractant, un homomorphisme u : A B(H) tel
que u(1) = 1 est une representation (resp. est orthogonalisable) si et
seulement si la famille {u(a) | a A a 1} est formee de contractions (resp. est semblable à une famille de contractions).
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
183
iii. Tout homomorphisme borne u : A B(H) setend avec la meme norme

en un homomorphisme borne (et continu pour la topologie *-faible) sur
lalgèbre de von Neumann engendree par A (donc aussi sur le bidual
de A). On peut donc se restreindre dans le problème 3.1. au cas des
algèbres de von Neumann.
Le problème 3.1. est toujours ouvert en toute generalite mais on a de nombreux resultats partiels dus principalement a` E. Christensen et U. Haagerup,
comme suit :
i. La reponse est oui si A est nucleaire ( nucleaire est lequivalent C algebrique de moyennable . On dit que A est nucleaire si pour toute
C -algèbre B, il existe une unique norme de C -algèbre sur A B).
ii. La reponse est oui si A = B(H) (mais, daprès [Wassermann 1976],
B(H) nest pas nucleaire) ou plus generalement si A ne possède aucun
etat tracial.
iii. La reponse est oui pour tout A mais à condition que u admette un
vecteur cyclique, i.e. un vecteur H tel que u(A) = H. (Il sut en
fait que A admette un nombre ni de vecteurs 1 , . . . , n tels que
u(A)1 + . . . + u(A)n = H.)
(i) est d
u independamment a` J. Bunce [1981] et E. Christensen [1981], (ii) et
(iii) sont dus à Haagerup [1983], mais [Christensen 1981] contient un resultat
très voisin de (iii) demontre independamment. Dans ces derniers resultats,
linegalite dite de Grothendieck non-commutative due a` lauteur [Pisier
1978] joue un r
ole important.
Le problème 3.1. semble etre un point nevralgique , en eet de très
nombreux problèmes souleves dans des contextes dierents se sont reveles en
fait equivalents au problème 3.1. Nous illustrons cela par plusieurs exemples
dans le reste de cette section.
On peut formuler un problème très proche du problème 3.1. en termes de
derivations. Soit A B(H) une sous-algèbre. Rappelons quune derivation
: A B(H) est une application lineaire telle que (ab) = (a)b + a(b) pour
tous a, b dans A. Par exemple, pour T B(H), la formule T (a) = aT T a
denit une derivation de A dans B(H). Les derivations de cette forme sont
appelees interieures.
erivation
Probl`
eme 3.3. Soit A B(H) une sous C algèbre. Toute d
(continue) : A B(H) est-elle interieure ?
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
184
G. PISIER
Ce problème reste lui-aussi ouvert en toute generalite. Noter toutefois que

la reponse est armative pour les derivations qui prennent leurs valeurs dans
A (i.e. pour : A A) (cf. [Kadison 1966]). Signalons aussi que la continuite
des derivations est automatique .
Les references classiques sur ce problème sont [Kaplansky 1953], [Kadison
et Ringrose 1971], [Kadison 1966], [Sakai 1966] et les travaux plus recents
dE. Christensen [1978, 1982a,b].
Il est important dobserver que : A B(H) est une derivation (bornee)
si et seulement si u : A B(H H) deni par

a (a)
u (a) =
0
a
est un homomorphisme (borne).
Cette observation permet de montrer que si la reponse est oui pour
A dans le problème 3.1., il est de meme dans le problème 3.3. Mais en fait
Kirchberg [1996] a donne recemment une demonstration (assez simple) du fait
que ces deux problèmes sont essentiellement equivalents.
Plus precisement, etant donnee une C -algèbre A, tout homomorphisme
borne u : A B(H) (H Hilbert arbitraire) est orthogonalisable si et
seulement si pour toute representation : A B(H) (H arbitraire) toute
-derivation bornee : A B(H) (i.e. (ab) = (a)(b)+ (a)(b) a, b A)
est interieure (i.e. T B(H) tel que (a) = (a)T T (a).
Le problème 3.3. peut aussi etre formule dans le langage de la cohomologie
de Hochschild adaptee aux algèbres de Banach. La theorie de Hochschild,
publiee dans les annees 1945-1947 est à lorigine purement algebrique (voir
[Hochschild 1945]) mais les modications à faire dans un contexte topologique
sont assez evidentes. Nous nous bornerons a` un très bref apercu. Soit A une
algèbre de Banach et soit X un A-bimodule de Banach. En fait pour le present
expose lexemple principal est celui dune sous-algèbre fermee A de lespace
B(H). Alors X = B(H) lui-meme peut-etre vu comme A-bimodule.
On notera Ln (A, X ) (ou simplement Ln ) lespace des formes n-lineaires
bornees f : An X , ce sont les n-cochanes . On denit un cobord
: Ln Ln+1 par la formule suivante, pour f dans Ln et a0 , . . . , an dans A
f (a0 , . . . , an ) = a0 f (a1 , . . . , an )
n1

(1)j+1 f (a0 , . . . , aj aj+1 , aj+2 , . . . , an )
+
j=0
+
(T B(H), x A).
` 3
SEMINAIRES
ET CONGRES
(1)n+1 f (a0 , . . . , an1 )an
PROBLEMES
DE SIMILARITE
185
Pour n = 0, on pose L0 = X et T (a) = aT T a (T X , a A).

On verie aisement que lon a une cohomologie, i.e. que = 0. Pour
n = 1, 2, . . . , limage de : Ln1 Ln est notee B n (A, X ) (ou simplement
B n ). Ce sont les n-cobords . On note Z n (A, X ) (ou Z n ) le noyau de lapplication : Ln Ln+1 . Cest lensemble des n-cocycles . Enn, on note
H n (A, X ) le quotient B n /Z n . Cest le n-ième groupe de cohomologie , qui
en fait ici est un espace de Banach. Il est facile de verier que Z 1 est forme
des derivations bornees de A dans X et B 1 des derivations interieures, i.e. de
la forme f (a) = xa ax pour un x X . Le problème 3.3. peut alors etre
reformule comme suit
ebre dans B(H). A-t-on
Probl`
eme 3.4. Soit A B(H) une C -alg`
H 1 (A, B(H)) = {0} ?
Lorigine de toutes ces questions sur les derivations se trouve semble-t-il
dans [Kaplansky 1953]. Après de premiers travaux dus a` Kamowitz [1962]
et Guichardet [1966], ce sont principalement B.E. Johnson, R. Kadison et J.
Ringrose qui ont developpe la cohomologie des algèbres de Banach de 1968 à
1972. Voir en particulier le memoire de Barry Johnson [1972] et le livre [Pier
1988] sur la moyennabilite des algèbres de Banach.
Entre temps, Sakai [1966] (completant le travail de Kadison [1966]) a montre que pour toute algèbre de von Neumann M B(H) on a
u il resulte aussi que (pour toute C -algèbre A) toute
H 1 (M, M ) = {0}, do`
derivation : A B(H) prenant ses valeurs dans A est interieure. Lextension du theorème de Sakai aux dimensions superieures a resiste depuis lors :
on ignore si H n (M, M ) = {0} pour n 2. En revanche, la version complèteace
ment bornee de cette conjecture (due a` Kadison-Ringrose) est connue gr
aux nombreux travaux de E. Christensen et A. Sinclair dans les annees 80,
pour lesquels nous renvoyons au livre de A. Sinclair et R. Smith [1995]. Voir
aussi le survey tout recent de Ringrose [1996]. Signalons en passant que le
langage cohomologique sadapte aussi aux problèmes discutes dans le premier
paragraphe sur les representations des groupes. Pour une plus vaste presentation des theories cohomologiques, voir aussi le livre de Helemskii [1989].
Parmi les problèmes en vogue à une certaine epoque (voir par exemple
[Foias 1972], [Fillmore et Williams 1971] et [Voiculescu 1972]) et qui se sont
reveles equivalents au problème 3.1., on peut citer le problème des images
invariantes , attribue souvent à Dixmier, dont la thèse [Dixmier 1949] est
consacree à un sujet voisin. En reponse à une demande dinformation de ma
part, Dixmier ma ecrit quil avait pose ce problème, pour A algèbre de von
Neumann, dans une lettre adressee à von Neumann vers 1950. Ce problème
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
186
G. PISIER
se pose comme suit : soit H un Hilbert, on dit quun sous-espace vectoriel

V H est une image doperateurs sil est limage dun operateur borne sur
H, i.e. il existe T dans B(H) tel que T (H) = V (Foias [1972] les appelle
parafermes ). On peut alors enoncer le problème des images invariantes
(toujours ouvert) comme suit.
Probl`
eme 3.5. Soit V H une image doperateurs et soit A B(H)
une C -sous-algèbre unitale. Supposons que V est invariant par A, i.e. que
aV V pour tout a dans A. Existe-t-il un op
erateur T dans A (= le commutant de A) tel que T(H) = V ?
Remarque 3.6. Bien s
ur, si T A et V = T(H) alors V est invariant
par A, la question est donc de savoir si la reciproque est vraie. Soulignons que
V nest pas suppose ferme (sinon cest evident car le projecteur orthogonal
PV est dans A si V est un invariant par A).
Suivant une idee de Foias [1972] nous allons voir quune reponse armative
au problème 3.1. en entrane une aussi pour le problème 3.5. En eet, soit
T B(H) tel que T (H) = V . Posons K = Ker(T ) . Noter que T (K) = V.
Soit a A et k K.
Puisque aT (K) T (K), il existe un unique
k dans K tel que aT k = T
k.
Le theorème du graphe ferme montre que la correspondance k
k denit un operateur lineaire borne que lon note u(a). On verie aisement que
a u(a) B(K) est un homomorphisme unital et une nouvelle application
du theorème du graphe ferme montre que a u(a) est borne. On notera que
lon a (soulignons que T 1 est non borne) a A u(a) = T 1 aT.
Supposons que u est semblable a` une representation, i.e. il existe S :
K K inversible borne tel que S 1 u(.)S soit une representation. On a
donc T 1 aT = S 1 (a)S, donc si x = T S 1 : K H on a x(a) = ax
pour tout a dans A. Mais comme A est une C -algèbre et (a ) = (a) ,
u en prenant les adjoint (a)x = x a soit
on a x(a) = x(a ) = a x do`
nalement x(a)x = xx a mais aussi x(a)x = (x(a))x = (ax)x = axx .

Loperateur T = xx : H H est donc dans le commutant A et il est clair
que T(H) = x(H) = T (H). On a donc montre que oui au problème 3.1.
entrane oui au problème 3.5. En realite linverse est aussi vrai et les deux
problèmes sont en fait equivalents (voir Mathes [1989] pour plus de details et
dautres references).
En eet, supposons inversement que lon sache repondre oui au problème 3.5. pour une C -algèbre A. Soit alors u : A B(H) un homomorphisme borne de la forme u(a) = T 1 aT avec T borne injectif dimage
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
187
V = T (H). Soit alors T A comme dans le problème 3.5. Comme la projection sur (Ker T) est dans A , on peut se ramener a` T injectif. On a alors
u(a) = T 1 aT = T 1 (TaT1 )T mais T1 T : H H et T 1 T : H H sont
bornes par le theorème du graphe ferme, donc u est orthogonalisable .
Reste à montrer que lon peut eectivement se restreindre a` des homomorphismes u de cette forme dans le problème 3.1. Montronsle.
Soit u : A B(H) un homomorphisme unital borne. Supposons H separable pour simplier, soit (n ) une suite dense, et soit Hn H deni par
Hn = u(A)1 + . . . + u(A)n .
Lhomomorphisme un : A B(Hn ) deni par un (a) = u(a)|Hn admet evidemment (1 , . . . , n ) comme sous-ensemble ni cyclique, donc daprès Haau Sn : Hn Hn borne inversible
gerup [1983], un est orthogonalisable. Do`
et une representation n : A B(Hn ) tels que un (.) = Sn1 n (.)Sn . On
la somme directe H1 H2 . . . et soient
peut supposer Sn = 1. Soit H

denis par S = Sn ,
(a) =
S : H H,
: A B(H) et u
: A B(H)

(a) = un (a). On a S = 1, S est injectif et dimage dense
n (a) et u
Daprès ce qui precède, u
(.)S.
est orthogonalisable. On en
et u
(.) = S1
deduit alors que tout ultraproduit des un et donc nalement u lui-meme est
orthogonalisable. On peut aussi utiliser le theorème 4.2. ci-dessous, il sut
alors dobserver que
ucb < .
ucb = sup un cb =
On notera que lon peut supposer sans restreindre la generalite dans le
problème 3.5. que A est une algèbre de von Neumann, puisque cetait dej`
a le
cas pour le problème 3.1., par passage au bidual. Voir Ong [1981] pour plus
de details sur ce point.
Une autre version du problème 3.1. est apparue dans les travaux de
W. Arveson sur les algèbres à chanes . Il sagit dune classe de sousalgèbres non autoadjointes de B(H), dont lexemple le plus simple est
lalgèbre des matrices triangulaires superieures dans lalgèbre Mn des matrices n n ou bien dans B(.2 ). Plus precisement, on appelle chane sur
un Hilbert H une famille C de sous-espaces fermes de H totalement ordonnee
pour linclusion et telle que pour toute sous-famille (Ei )iI de C on a

Ei C
et
Ei C.
iI
iI
Etant donnee une telle chane , on lui associe la sous-algèbre A(C)

B(H) formee des operateurs T de B(H) tels que
i I
T (Ei ) Ei .
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
188
G. PISIER
Ces algèbres seront appelees ici algèbres à chanes (il ny a apparemment pas de terme francais dans la litterature, les anglo-saxons disent nest
algebras ). Par exemple, lalgèbre des matrices triangulaires superieures dans
Mn correspond a` la chane formee des sous-espaces Ei engendres par les i
premiers vecteurs de la base canonique (i = 1, . . . , n) dans lespace .2n . Ces
algèbres ont ete introduites par Ringrose [1961] a` la suite des travaux de Kadison et Singer [1960] sur les algèbres triangulaires . Voir Davidson [1988]
pour un expose complet de ce sujet. Le resultat le plus frappant de la theorie
est une formule de distance due a` Arveson [1975], comme suit : soit A B(H)
une sous-algèbre, on note Lat(A) lensemble des sous-espaces fermes E H
qui sont invariants par A (i.e. aE E a A). Soit alors T dans B(H). On
note d(T, A) la distance (normique) de T à A. Il est facile de voir que pour
tout sous-espace A-invariant E on a
(1 PE )T PE d(T, A),
do`
u lon tire
T B(H)
sup{(1 PE )T PE | E Lat(A)} d(T, A).
Le fondement de la theorie dArveson [1975] est le fait que si A est une

algèbre à chanes, on a
egalit
e dans la formule precedente.
Cela a conduit a` etudier cette formule de distance pour des classes plus
generales dalgèbres. On appelle reexive une sous-algèbre A B(H) telle
que
(T (E) E E Lat(A)) T A.
On appelle hyper-reexive une sous-algèbre A B(H) pour laquelle il
existe une constante K telle que
T B(H) d(T, A) K sup{(1 PE )T PE |
E Lat(A)}.
Comme on vient de le voir, ceci est vrai avec K = 1 pour toute algèbre
` chane. Bien que cela nait pas ete clair dès le depart, il y a des algèbres
a
reexives qui ne sont pas hyper-reexives (voir [Kraus et Larson 1985]). Le
theorème du bicommutant de von Neumann (qui assure que M = (M ) ) nous
garantit evidemment que toute algèbre de von Neumann M est reexive. Mais
est-elle hyper-reexive ?
Cest un problème ouvert, equivalent au problème 3.1. Explicitement :
Probl`
eme 3.7. Toute alg`
ebre de von Neumann M est-elle hyper-r
eexive ? Plus pr
ecisement, supposons M B(H), existe-t-il une constante K telle
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
189
que, pour tout T dans B(H) on a

sup{(1 P )T P
P M } K d(T, M )
Essentiellement, la reponse est armative (daprès les travaux de Christensen [1978, 1982a,b]) pour la meme classe dalgèbres que pour le problème 3.3.
Les problèmes 3.1. et 3.3. sont aussi intimement lies à des problèmes de
perturbation des algèbres de von Neumann poses dans [Kadison et Kastler
1972] : soient M, N deux algèbres de von Neumann, notons d(M, N ) leur
distance de Hausdor et supposons d(M, N ) < * (cela signie que tout point
de la boule unite de M est à distance < * dun point de la boule unite de N ,
et vice versa). Si * > 0 est est assez petit, existe-t-il un unitaire U dans B(H)
tel que U I est o(*) et tel que U M U = N ? Une reponse positive
aux problèmes 3.1. et 3.3. permettrait (daprès les travaux de Christensen
[1977a,b, 1980]) de progresser sur ce problème qui reste lui-aussi ouvert.
4.
Applications compl`
etement born
ees
Comme nous allons le voir, la theorie recente des applications complètement bornees permet de traiter dans un cadre commun tous les problèmes
precedents.
Soit A B(H) un sous-espace vectoriel et soit u : A B(H) une application lineaire. On note Mn (B(H)) (resp. Mn (A)) lensemble des matrices n n
a coecients dans B(H) (resp. dans A), muni de la norme induite par lespace
`
B(H . . . H) (resp. B(H . . . H)). On note un : Mn (A) Mn (B(H)),
lapplication denie par
un ((aij )) = (u(aij )).
(Si lon pose Mn (A) = Mn A, alors un = IMn u).
D
enition. Dans la situation precedente, on dit que u : A B(H) est
complètement bornee (c.b. en abrege) si
sup {un ; un : Mn (A) Mn (B(H))} < ,
n1
et lon pose
ucb = sup un .
n1
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
190
G. PISIER
Par exemple, toute C -representation : A B(H) est c.b. et

cb = 1. En general, on a u ucb et il existe des applications bornees mais pas c.b. (par exemple la transposition des matrices de
B(.2 ) dans lui-meme est isometrique mais pas c.b.).
Dans la situation precedente, on dit que u est complètement positive (c.p.
en abrege) si un : A B(H) est complètement positive pour tout n, cest-àdire si
a Mn (A) Mn (B(H))+
entrane
un (a) Mn (B(H))+
(rappelons que pour une C -algèbre M , M+ = {x M |x = x et x 0}).

Bien entendu, cette notion nest utile que si Mn (A) Mn (B(H))+ nest pas
trop petit, par exemple si A est une sous-C -algèbre, ou bien si A est un
système doperateurs (ce qui veut dire I A et a A a A).
La notion dapplication c.p. sur une C -algèbre remonte à [Stinespring
1955]. Elle est devenue omnipresente dans la theorie des algèbres doperateurs à la suite des travaux fondamentaux dArveson [1969] sur les systèmes
doperateurs. La notion dapplication c.b. est apparue dej`
a dans [Arveson
1969] mais ce dernier ne considère pratiquement que des contractions complètes (on appelle ainsi toute application u veriant ||u||cb 1). Cette notion
sest rapidement developpee à partir de 1980 après la decouverte dun theorème remarquable de factorisation d
u independamment a` Wittstock [1981],
Haagerup [1980], Paulsen [1982], qui senonce comme suit :
Th
eor`
eme 4.1. Soit A B(H) un sous-espace vectoriel de B(H) et
soit u : A B(H) une application lin
eaire. Fixons une constante C 0. Les
assertions suivantes sont
equivalentes :
i. Lapplication u est c.b et ucb C.
et des operateurs V1 :
ii. Il existe une repr
esentation : B(H) B(H)

H H et V2 : H H avec V1 V2 C tels que
a A
u(a) = V1 (a)V2 .
Voir les livres Paulsen [1986] ou Pisier [1995] pour plus de details. Cest a`
travers le resultat suivant d
u a` Paulsen [1984b] (et a` Haagerup [1983] pour une
C -algèbre A) que les applications complètement bornees permettent dunier

les problèmes de similarite qui precèdent.
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
191
Th
eor`
eme 4.2. Soit A B(H) une sous-algèbre ferm
ee contenant
lunit
e de B(H). Soit u : A B(H) un homomorphisme avec u(1) = 1
et soit C une constante. Les assertions suivantes sont
equivalentes.
i. Lapplication u est c.b. et ucb C.
ii. Il existe S : H H inversible satisfaisant S 1 S C tel que
erie
lhomomorphisme u
deni par u
(a) = S 1 u(a)S a A v

ucb 1.
une repr
un sousiii. Il existe un Hilbert H,
esentation : B(H) B(H),

espace E H et un isomorphisme S : H E v
eriant SS 1 C
et tel que
(4.1)
u(a) = S 1 [PE (a)|E ]S.
D
emonstration esquiss
ee. Les implications (iii) (ii) (i) sont evidentes.
Supposons (i). Le theorème 4.1. permet decrire
u(a) = V1 (a)V2
avec V1 V2 C
et
V1 V2 = 1.
le sous-espace ferme engendre par (A)V2 (H) et soit E2

Soit alors E1 H
E1 le noyau V1 restreint a` E1 . On pose E = E1 E2 et Sh = PE V2 h pour

tout h dans H. On peut alors verier que S est inversible et S 1 x = V1 x si
x E, do`
u S S 1 V2 V1 C. Enn on verie aisement (4.1).
Remarque 4.3. Le sous-espace E apparaissant dans (iii) ci-dessus est
necessairement semi-invariant pour (A), daprès le theorème 2.2.
Le theorème 4.2. peut sappliquer dans chacune des situations precedentes.
Revenons tout dabord au 1.
Soit G un groupe discret. Notons C[G] lalgèbre du groupe et C (G) la
C -algèbre (pleine) du groupe G , cest-à-dire la C -algèbre engendree par la

representation universelle de G . On a une inclusion naturelle C[G] C (G)
(`
a image dense). On a alors :
Corollaire 4.4. Une representation uniformement born
ee : G B(H)
est unitarisable si et seulement si elle setend en une application compl`
etement
born
ee sur C (G).
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
192
G. PISIER
Considerons maintenant le 2. Soit T B(H) un operateur polynomialement borne. Alors daprès le theorème 4.2., T est semblable a` une contraction
si et seulement si lhomomorphisme T : P P (T ) est complètement borne
sur lalgèbre du disque A(D) (i.e. la completion des polyn
omes pour la norme
). Donnons-en un enonce plus precis :
Corollaire 4.5. Soit T B(H) et soit C une constante. Les assertions
suivantes sont
equivalentes :
i. Il existe un op
erateur S inversible sur H tel que
S 1 T S 1
et
S 1 S C.
ii. T cb C.
iii. Pour tout n et tout polyn
ome P (z) =
N
ah z h `
a coecients dans Mn
on a
N
ah T h Mn (B(H)) C sup P (z)Mn .

|z|=1
On peut aussi donner des enonces analogues pour des k-uplets doperateurs (T1 , . . . , Tk ) commutant entre eux ou bien pour des algèbres uniformes
A() avec domaine de Cn à la place de A(D).
Passons maintenant aux problèmes consideres au 3. Applique à une C algèbre, le theorème 4.2. nous donne :
Corollaire 4.6. Le probl`
eme 3.1. est
equivalent au suivant : est-il vrai
ebre a
`
que pour tout homomorphisme unital u : A B(H) sur une C -alg`
etement born
e .
unit
e, borne implique automatiquement compl`
Quant aux derivations, Christensen [1977b] montre quune derivation
: A B(H) est complètement bornee si et seulement si elle est interieure.
Le problème 3.3. revient donc à montrer que, pour une derivation, on a
encore : borne c.b. Plus precisement, si A est une algèbre de von Neumann
on peut ecrire (cf. [Christensen 1977b])
cb = 2 inf{x | x B(H),
(a) = ax xa a A}.
Si (a) = ax xa, linegalite

cb 2x
` 3
SEMINAIRES
ET CONGRES
(donc cb 2 inf x)
PROBLEMES
DE SIMILARITE
193
est evidente, cest linverse qui est non trivial.

Dans le langage cohomologique, on peut bien s
ur introduire lanalogue c.b.
1
1
de H (A, B(H)) que lon note Hcb (A, B(H)), mais les remarques precedentes
montrent que
1
(A, B(H)) = {0}
Hcb
pour toute C -algèbre A. Signalons quil existe une notion dapplication multilin
eaire c.b. (voir [Christensen et Sinclair 1989]) qui permet de denir aussi
n (A, B(H)), et [Christensen et al. 1987] ont montr
e que ce groupe est touHcb
jours reduit a` {0}. Voir [Sinclair et Smith 1995] pour un expose detaille.
Le problème des images invariantes de Dixmier (problème 3.5.) se prete
aussi à une reformulation en termes c.b. En eet, soit V H une image
= H H . . .
doperateur invariante par une C -algèbre A B(H). Posons H

(somme directe innie) et soit A B(H) lampliation de A, cest-à-dire avec
lidentication

= {a I | a A}.
=H
.2 , que A
H
2
Supposons V
Supposons V = T (H) avec T B(H) et posons V = (T I)(H).

invariant par A. Alors il existe T dans A tel que T (H) = T (H) si et seulement
B(H 2 .2 )
si V est un sous-espace vectoriel invariant pour la C -algèbre A
engendree par les operateurs de la forme a t avec a A et t compact de .2
dans .2 . Pour ce dernier enonce, voir [Paulsen 1982].
Bibliographie
Adian (S.I.)
[1979]
The Burnside problem and identities in groups, Springer, 1979.
Ando (T.)
[1963]
On a pair of commutative contractions, Acta Sci. Math. (Szeged), 24

(1963), p. 8890.
Arveson (W.)
[1969]
Subalgebras of C -algebras, Acta Math., 123 (1969), p. 141224 ; Part II,

Acta Math., 128 (1972) p. 271308.
[1975]
Interpolation problems in nest algebras, J. Funct. Anal., 20 (1975),

p. 208233.
Bourgain (J.)
[1986]
On the similarity problem for polynomially bounded operators on Hilbert

space, V, Israel J. Math., 54 (1986), p. 227241.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
194
G. PISIER
Boz ejko (M.)

[1985]
Positive denite bounded matrices and a characterization of amenable

groups, Proc. Amer. Math. Soc., 95 (1985), p. 357360.
[1987a]
Littlewood functions, Hankel multipliers and power bounded operators

on a Hilbert space, Colloq. Math., 51 (1987), p. 3542.
[1987b]
Uniformly bounded representations of free groups, J. Reine Angew.

Math., 377 (1987), p. 170186.
Boz ejko (M.) et Fendler (G.)

[1991]
Herz-Schur multipliers and uniformly bounded representations of discrete

groups, Arch. Math., 57 (1991), p. 290298.
Bunce (J.W.)
[1981]
The similarity problem for representations of C -algebras, Proc. Amer.

Math. Soc., 81 (1981), p. 409414.
Cartier (P.)
[1973]
Harmonic analysis on trees, dans Harmonic analysis on homogeneous

spaces, Providence : AMS (Proc. Sympos. Pure Math. vol. 26), 1973,
p. 419424.
Christensen (E.)
[1977a]
Perturbation of operator algebras, Invent. Math., 43 (1977), p. 113.
[1977b]
Perturbation of operator algebras, II, Indiana Univ. Math. J., 26 (1977),

p. 891904.
[1978]
Extensions of derivations, J. Funct. Anal., 27 (1978), p. 234247.
[1980]
Near inclusions of C -algebras, Acta Math., 144 (1980), p. 249265.
[1981]
On non self adjoint representations of operator algebras, Amer. J. Math.,

103 (1981), p. 817834.
[1982a]
Extensions of derivations, II., Math. Scand., 50 (1982), p. 111122.
[1982b]
Derivations and their relation to perturbation of operator algebras, dans

Operator algebras and applications, Providence : AMS (Proc. Sympos.
Pure Math. vol. 38-2), 1982, p. 261273.
Christensen (E.), Effros (E.) et Sinclair (A.)

[1987]
Completely bounded multilinear maps and C -algebraic cohomology,

Invent. Math., 90 (1987), p. 279296.
Christensen (E.) et Sinclair (A.)

[1989]
A survey of completely bounded operators, Bull. London Math. Soc., 21

(1989), p. 417448.
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
195
Connes (A.)
[1990]
Geometrie non commutative. Paris : Inter Editions, 1990. Trad. angl. San
Diego : Academic Press, 1994.
Conway (J.)
[1991]
The theory of subnormal operators, Providence : AMS (Math. Surveys,

vol. 26), 1991.
Cowling (M.)
[1978]
The Kunze-Stein phenomenon, Ann. of Math., 107 (1978), p. 209234.
[1982]
Unitary and uniformly bounded representations of some simple Lie

groups, CIME Course 1980, Napoli : Liguori, (1982), p. 49128.
Crabb (M.) et Davie (A.)

[1975]
von Neumanns inequality for Hilbert space operators, Bull. London Math.
Soc., 7 (1975), p. 4950.
Davidson (K.)
[1988]
Nest Algebras, Harlow : Longman (Pitman Research Notes in Math.,

vol. 191), 1988.
Day (M.)
[1950]
Means for the bounded functions and ergodicity of the bounded representations of semi-groups, Trans. Amer. Math. Soc., 69 (1950), p. 276291.
Dieudonn
e, (J.)
[1960]
Sur le produit de composition (II), J. Math. Pures Appl., 39 (1960),

p. 275292.
Dixmier (J.)
[1949]
Etude sur les varietes et les operateurs de Julia, Bul. Soc. Math. France,
77 (1949), p. 11101.
[1950]
Les moyennes invariantes dans les semi-groupes et leurs applications, Acta

Sci. Math. (Szeged), 12 (1950), p. 213227.
Dixon (P.)
[1995]
Banach algebras satisfying the non-unital von Neumann inequality, Bull.

London Math. Soc., 27 (1995), p. 359262.
Douglas (V.) et Paulsen (V.)

[1989]
Hilbert modules over function algebras, Harlow : Longman (Pitman Research Notes in Math., vol. 217), 1989.
Ehrenpreis (L.) et Mautner (F.)

[1955]
Uniformly bounded representations of groups, Proc. Nat. Acad. Sci. USA,

41 (1955), p. 231233.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
196
G. PISIER
Eymard (P.)
[1964]
Lalgèbre de Fourier dun groupe localement compact, Bull. Soc. Math.

France, 92 (1964), p. 181236.
Fendler (G.)
[1990]
A uniformly bounded representation associated to a free set in a discrete

group, Colloq. Math., 59 (1990), p. 223229.
` -Talamanca (A.) et Picardello (M.)

Figa
[1983]
Harmonic analysis on free groups. New York : Marcel Dekker, 1983.
Fillmore (P.A.) et Williams (J.P.)

[1971]
On operator ranges, Adv. Math., 7 (1971), p. 254281.
Foguel (S.)
[1964]
A counterexample to a problem of Sz.-Nagy, Proc. Amer. Math. Soc., 15

(1964), p. 788790.
Foias (C.)
[1972]
Invariant paraclosed subspaces, Indiana Univ. Math. J., 21 (1972), p. 887

906.
Gromov (M.)
[1987]
Hyperbolic groups, dans Essays in group theory (ed. par S. Gersten),

Springer, p. 75263.
Guichardet (A.)
[1966]
Sur lhomologie et la cohomologie des algèbres de Banach, C. R. Acad.

Sci. Paris, 262 (1966), p. 3841.
Haagerup (U.)
[1980]
Decomposition of completely bounded maps on operator algebras, manuscript.
[1983]
Solution of the similarity problem for cyclic representations of C algebras, Ann. of Math., 118 (1983), p. 215240.
[1985]
M0 A(G) functions which are not coecients of uniformly bounded representations, manuscript.
Halmos (P.)
[1970]
Ten problems in Hilbert space, Bull. Amer. Math. Soc., 76 (1970), p. 887
933.
Helemskii (A.Y.)
[1989]
The homology of Banach and topological algebras, Dordrecht : Kluwer,

1989.
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
197
Hochschild (G.)
[1945]
On the cohomology groups of an associative algebra, Ann. of Math., 46

(1945), p. 5867.
Johnson (B.E.)
[1972]
Cohomology in Banach algebras, Mem. Amer. Math. Soc., vol. 127, Providence : AMS, 1972.
Kadison (R.)
[1955]
On the orthogonalization of operator representations, Amer. J. Math., 77

(1955), p. 600620.
[1966]
Derivations of operator algebras, Ann. of Math., 83 (1966), p. 280293.
Kadison (R.) et Kastler (D.)

[1972]
Perturbations of von Neumann algebras, Amer. J. Math., 94 (1972), p. 38

54.
Kadison (R.) et Ringrose (J.)

[1971]
Cohomology of operator algebras, Acta Math., 126 (1971), p. 227243.
[1986]
Fundamentals of the theory of operators algebras, Vol. II Advanced

Theory, Orlando : Academic Press, 1986.
Kadison (R.) et Singer (I.)

[1960]
Triangular operator algebras, Amer. J. Math., 82 (1960), p. 227259.
Kamowitz (H.)
[1962]
Cohomology groups of commutative Banach algebras, Trans. Amer.

Math. Soc., 102 (1962), p. 352372.
Kaplansky (I.)
[1953]
Modules over operator algebras, Amer. J. Math., 75 (1953), p. 839853.
Kirchberg (E.)
[1996]
The derivation and the similarity problem are equivalent, J. Operator

Theory, 36 (1996), p. 5962.
Kraus (J.) et Larson (D.)

[1985]
Some applications of a technique for constructing reexive operator algebras, J. Operator Theory, 13 (1985), p. 227236.
Kunze (R.A.) et Stein (E.)

[1960]
Uniformly bounded representations and harmonic analysis of the 22 real

unimodular group, Amer. J. Math., 82 (1960), p. 162 ; Part II, Amer. J.
Math., 83 (1961), p. 723786.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
198
G. PISIER
Lebow (A.)
[1968]
A power bounded operator which is not polynomially bounded, Michigan

Math. J., 15 (1968), p. 397399.
Losert (V.)
[1984]
Properties of the Fourier algebra that are equivalent to amenability, Proc.

Amer. Math. Soc., 92 (1984), p. 347354.
Mantero (A.M.) et Zappa (A.)

[1983]
The Poisson transform on free groups and uniformly bounded representations, J. Funct. Anal., 51 (1983), p. 372399.
Mathes (B.)
[1989]
Operator ranges and completely bounded homomorphisms, Proc. Amer.

Math. Soc., 107 (1989), p. 155164.
Nebbia (C.)
[1982]
Multipliers and asymptotic behaviour of the Fourier algebra of nonamenable groups, Proc. Amer. Math. Soc., 84 (1982), p. 549554.
von Neumann (J.)

[1929]
Zur allgemeinen Theorie des Masses, Fund. Math., 13 (1929), p. 73116.
[1951]
Eine Spektraltheorie f
ur allgemeine Operatoren eines unitaren Raumes,
Math. Nachr., 4 (1951), p. 49131.
Novikov (P.S.) et Adian (S.I.)

[1968]
Innite periodic groups I, II, III., Izvest. Akad. Nauk SSSR Ser. Mat., 32
(1968), p. 212244, 251524, 709731.
Olshanskii (A.Y.)
[1980]
On the problem of an invariant mean on a group, Russian Math. Surveys,

35 (1980), p. 180181.
Ong (S.C.)
[1981]
Operator topologies and invariant operator ranges, Canad. Math. Bull.,

24 (1981), p. 181186.
Parrott (S.)
[1970]
Unitary dilations for commuting contractions, Pacic J. Math., 34 (1970),

p. 481490.
Paterson (A.)
[1988]
Amenability, Providence : A.M.S. (Math. Surveys, vol. 29), 1988.
Paulsen (V.)
[1982]
Completely bounded maps on C -algebras and invariant operator ranges,

Proc. Amer. Math. Soc., 86 (1982), p. 9196.
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
199
[1984a]
Every completely polynomially bounded operator is similar to a contraction, J. Funct. Anal., 55 (1984), p. 117.
[1984b]
Completely bounded homomorphisms of operator algebras, Proc. Amer.

Math. Soc., 92 (1984), p. 225228.
[1986]
Completely bounded maps and dilations, Harlow : Longman (Pitman Research Notes in Math. vol. 146), 1986.
Peller (V.)
[1982]
Estimates of functions of power bounded operators on Hilbert space, J.

Operator Theory, 7 (1982), p. 341372.
Phillips (J.)
[1982]
Invariant of C algebras stable under perturbations, dans Operator algebras and applications, Providence : AMS (Proc. Sympos. Pure Math.
vol. 38-2), 1982, p. 275279.
Pier (J. P.)

[1984]
Amenable locally compact groups, New York : Wiley, 1984.
[1988]
Amenable Banach algebras, Harlow : Longman (Pitman Research Notes

in Math. vol. 172), 1988.
Pisier (G.)
[1978]
Grothendiecks theorem for noncommutative C -algebras with an appendix on Grothendiecks constants, J. Funct. Anal., 29 (1978), p. 397415.
[1995]
Similarity problems and completely bounded maps, vol. 1680 des Lecture
Notes, Springer, 1995.
[1997]
A polynomially bounded operator on Hilbert space which is not similar

to a contraction, J. Amer. Math. Soc., 10 (1997), p. 351369.
Pytlik (T.) et Szwarc (R.)

[1986]
An analytic family of uniformly bounded representations of free groups,

Acta Math., 157 (1986), p. 287309.
Ringrose (J.)
[1961]
Algebraic isomorphisms between ordered bases, Amer. J. Math., 83

(1961), p. 463478.
[1982]
Cohomology theory for operator algebras, dans Operator algebras and

applications, Providence : AMS (Proc. Sympos. Pure Math. vol. 38-2),
1982, p. 229252.
[1996]
The cohomology of operator algebras : a survey, Bull. London Math. Soc.,

28 (1996), p. 225241.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
200
G. PISIER
Rota (G.)
[1960]
On models for linear operators, Comm. Pure Appl. Math., 13 (1960),

p. 468472.
Russo (B.) et Dye (H.)

[1966]
A note on unitary operators in C -algebras, Duke Math. J., 33 (1966),

p. 413416.
Sakai (S.)
[1966]
Derivations of W -algebras, Ann. of Math., 83 (1966), p. 273279.
Sally (P.J.)
[1968]
Unitary and uniformly bounded representations of the two by two unimodular group over local elds, Amer. J. Math., 90 (1968), p. 406443.
Sarason (D.)
[1965]
On spectral sets having connected complement, Acta Sci. Math. (Szeged),

26 (1965), p. 289299.
[1967]
Generalized interpolation in H , Trans. Amer. Math. Soc., 127 (1967),

p. 179203.
Sinclair (A.) et Smith (R.)

[1995]
Hochschild cohomology of von Neumann algebras. LMS Lecture Note

Series, vol. 203, Cambridge Univ. Press, 1995.
Stinespring (W.)
[1955]
Positive functions on C -algebras, Proc. Amer. Math.Soc., 6 (1955),

p. 211216.
Sz.-Nagy (B.)
[1946]
On uniformly bounded linear transformations on Hilbert space, Acta Sci.

Math. (Szeged), 11 (1946), p. 152157.
[1953]
Sur les contractions de lespace de Hilbert, Acta Sci. Math. (Szeged), 15

(1953), p. 8792.
[1958]
Spectral sets and normal dilations of operators, dans Proc. Intern. Congr.
Math., Edinburgh, 1958, Cambridge Univ. Press, p. 412422.
[1959]
Completely continuous operators with uniformly bounded iterates, Publ.

Math. Inst. Hungarian Acad. Sci., 4 (1959), p. 8992.
Sz.-Nagy (B.) et Foias (C.)

[1970]
Harmonic analysis of operators on Hilbert space. Budapest : Akademiai

Kiad
o, 1970.
Szwarc
[1988]
An analytic family of irreducible representations of the free group, Ann.

Inst. Fourier, 38 (1988), p. 87100.
` 3
SEMINAIRES
ET CONGRES
PROBLEMES
DE SIMILARITE
201
Tits (J.)
[1972]
Free subgroups in linear groups, J. Algebra, 20 (1972), p. 250272.
Valette (A.)
[1990a]
Cocycles darbres et representations uniformement bornees, C. R. Acad.

Sci. Paris, 310 (1990), p. 703708.
[1990b]
Les representations uniformement bornees associees à un arbre reel,

Bull. Soc. Math. Belg., 42 (1990), p. 747760.
Varopoulos (N.)
[1974]
On an inequality of von Neumann and an application of the metric theory

of tensor products to operators theory, J. Funct. Anal., 16 (1974), p. 83
100.
Vasilescu (F.H.) et Zsido (L.)

[1974]
Uniformly bounded groups in nite W -algebras, Acta Sci. Math. (Szeged), 36 (1974), p. 189192.
Voiculescu (D.)
[1972]
Sur les sous-espaces parafermes invariants dune algèbre de von Neumann,

Bull. Sci. Math., 96 (1972), p. 161168.
Wassermann (S.)
[1976]
On tensor products of certain group C -algebras, J. Funct. Anal., 23

(1976), p. 239245.
Wittstock (G.)
[1981]
Ein operatorwertiger Hahn-Banach Satz, J. Funct. Anal., 40 (1981),

p. 127150.
Wysoczanski (J.)
[1993]
An analytic family of uniformly bounded representations of a free product

of discrete groups, Pacic J. Math., 157 (1993), p. 373385.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
From Riemann Surfaces

to Complex Spaces
Reinhold Remmert
We must always have old

memories and young hopes
Abstract
This paper analyzes the development of the theory of Riemann
surfaces and complex spaces, with emphasis on the work of Riemann, Klein and Poincare in the nineteenth century and on the
work of Behnke-Stein and Cartan-Serre in the middle of this century.
R
esum
e
Cet article analyse le developpement de la theorie des surfaces
de Riemann et des espaces analytiques complexes, en etudiant
notamment les travaux de Riemann, Klein et Poincare au XIXe
siècle et ceux de Behnke-Stein et Cartan-Serre au milieu de ce
siècle.
Table of Contents
1.
Riemann surfaces from 1851 to 1912
1.1.
1.1 .
1.2.
1.3.
Georg Friedrich Bernhard Riemann and the covering principle

Riemanns doctorate
Christian Felix Klein and the atlas principle
Karl Theodor Wilhelm Weierstrass and analytic congurations
AMS 1991 Mathematics Subject Classication: 01A55, 01A60, 30-03, 32-03

Westf
alische WilhelmsUniversit
at, Mathematisches Institut, D48149 M
unster,
Deutschland
This expose is an enlarged version of my lecture given in Nice. Gratias ago to J.-P. Serre
for critical comments. A detailed exposition of sections 1 and 2 will appear elsewhere.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
204
R. REMMERT
1.4.
1.5.
1.6.
1.7.
1.8.
1.9.
The feud between Gottingen and Berlin

Jules Henri Poincare and automorphic functions
The competition between Klein and Poincare
Georg Ferdinand Ludwig Philipp Cantor and countability of the
topology
Karl Hermann Amandus Schwarz and universal covering surfaces
The general uniformization theorem
2.
Riemann surfaces from 1913 onwards
2.1.
2.2.
2.3.
2.4.
2.5.
Claus Hugo Hermann Weyl and the sheaf principle

The impact of Weyls book on twentieth century mathematics
Tibor Rad
o and triangulation
Heinrich Adolph Louis Behnke, Karl Stein and non-compact Riemann
surfaces
Analytic congurations and domains of meromorphy
3.
Towards complex manifolds, 1919-1953
3.1.
3.2.
3.5.
Global complex analysis until 1950

Non-univalent domains over C n , 1931-1951, Henri Cartan
and Peter Thullen
Dierentiable manifolds, 1919-1936: Robert Konig, Elie Cartan,
Oswald Veblen and John Henry Constantine Whitehead,
Hassler Whitney
Complex manifolds, 1944-1948: Constantin Caratheodory, Oswald
Teichm
uller, Shiing Shen Chern, Andre Weil and Heinz Hopf
The French Revolution, 1950-53: Henri Cartan and Jean-Pierre Serre
3.6.
Stein manifolds
4.
Complex spaces, 1951-1960
4.1.
4.2.
4.3.
Normal complex spaces, 1951

Reduced complex spaces, 1955
Complex spaces with nilpotent holomorphic functions, 1960
3.3.
3.4.
Epilogue
1.
1.1.
Riemann surfaces from 1851 to 1912

Georg Friedrich Bernhard Riemann and the covering
principle
The theory of Riemann surfaces came into existence about the middle of
the nineteenth century somewhat like Minerva: a grown-up virgin, mailed in
` 3
SEMINAIRES
ET CONGRES
FROM RIEMANN SURFACES TO COMPLEX SPACES
205
the shining armor of analysis, topology and algebra, she sprang forth from
Riemanns Jovian head (cf. H. Weyl, [Ges. Abh. III, p. 670]). Indeed on
November 14, 1851, Riemann submitted a thesis Grundlagen f
ur eine allgemeine Theorie der Functionen einer ver
anderlichen complexen Gr
osse (Foundations of a general theory of functions of one complex variable) to the faculty of philosophy of the University of Gottingen to earn the degree of doctor
philosophiae. Richard Dedekind states in Bernhard Riemanns Lebenslauf,
that Riemann had probably conceived the decisive ideas in the autumn holidays of 1847, [Dedekind 1876, p. 544]. Here is Riemanns denition of his
surfaces as given in [Riemann 1851, p. 7]:
Wir beschr
anken die Veranderlichkeit der Grossen x, y auf ein endliches
Gebiet, indem wir als Ort des Punktes O nicht mehr die Ebene A selbst,
sondern eine u
ber dieselbe ausgebreitete Flache T betrachten. . . . Wir lassen
die M
oglichkeit oen, dass der Ort des Punktes O u
ber denselben Theil der
Ebene sich mehrfach erstrecke, setzen jedoch f
ur einen solchen Fall voraus,
dass die auf einander liegenden Flachentheile nicht langs einer Linie zusammenh
angen, so dass eine Umfaltung der Flache, oder eine Spaltung in auf
einander liegende Theile nicht vorkommt.
(We restrict the variables x, y to a nite domain by considering as the locus
of the point O no longer the plane A itself but a surface T spread over the
plane. We admit the possibility . . . that the locus of the point O is covering
the same part of the plane several times. However in such a case we assume
that those parts of the surface lying on top of one another are not connected
along a line. Thus a fold or a splitting of parts of the surface cannot occur).
Here the plane A is the complex plane C, which Riemann introduces on
page 5. Later, on page 39, he also admits die ganze unendliche Ebene A,
:= C {}. It is not clear what is meant by mehrfach
i.e., the sphere C
erstrecke. Does he allow only nitely or also innitely many points over a
point of A? The last lines in Riemanns denition are vague: his intention
is to describe local branching topologically. For algebraic functions this had
already been done in an analytic manner by V. Puiseux [1850]. A careful
discussion of the notion of Windungspunkt (m 1) Ordnung (winding point
of order m 1) is given by Riemann on page 8.
Riemanns denition is based on the covering principle: let z : T C

Then T is called a
be a continuous map of a topological surface T into C.
(concrete) Riemann surface over C (with respect to z) if the map z is locally

nite1 and a local homeomorphism outside of a locally nite subset S of T .
In this case there exists around every point x X a local coordinate t with
1
This means that to every point x T there exist open neighborhoods U , resp. V , of x,
resp. z(x), such that z induces a nite map U V .
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
206
R. REMMERT
t(x) = 0. If z(x) = z0 , resp. z(x) = , the map z is given by zz0 = tm , resp.

z = tm , with m N \ {0} and m = 1 whenever x
S. A unique complex
is a meromorphic
structure (cf. section 1.2.) on T such that z : T C
function is obtained by lifting the structure from C; the winding points are
contained in S.
can be weakened. According to Simion
The requirements for the map z : T C
Stoilow it suces to assume that z is continuous and open and that no z-ber contains
a continuum [Stoilow 1938, chap. V].
Riemanns thesis is merely the sketch of a vast programme. He gives no

examples, Aquila non captat muscas (Eagles dont catch ies). The breathtaking generality was at rst a hindrance for future developments. Contrary
to the Zeitgeist, holomorphic functions are dened by the Cauchy-Riemann
dierential equations. Explicit representations by power series or integrals are
of no interest. Formulae are powerful but blind. On page 40 Riemann states
his famous mapping theorem. His proof is based on Dirichlets principle.
Six years later, in his masterpiece Theorie der Abelschen Funktionen,
Riemann [1857] explains the intricate connections between algebraic functions and their integrals on compact surfaces from a birds-eye view of (not
yet existing) analysis situs. The number p, derived topologically from the
number 2p + 1 of connectivity and called Geschlecht (genus) by Clebsch
in [Clebsch 1865, p. 43], makes its appearance on p.104 and radiates like
wild yeast through all meditations. The famous inequality d m p + 1
for the dimension of the C-vector space of meromorphic functions having at
most poles of rst order at m given points occurs on pages 107-108; Gustav
Rochs renement in [Roch 1865] became the immortal Riemann-Roch theorem. The equation w = 2n + 2p 2 connecting genus and branching, which
was later generalized by Hurwitz to the Riemann-Hurwitz formula, [Hurwitz
1891, p. 376; 1893, pp. 392 and 404], is derived by analytic means on page
114.
Riemann and many other great men share the fate that at their time
there was no appropriate language to give their bold way of thinking a concise
form. In 1894 Felix Klein wrote, [1894, p. 490]: Die Riemannschen Methoden waren damals noch eine Art Arcanum seiner direkten Sch
uler und wurden
von den u
brigen
Mathematikern
fast
mit
Mitrauen
betrachtet
(Riemanns
methods were kind of a secret method for his students and were regarded almost with distrust by other mathematicians). M. A. Stern, Riemanns teacher
of calculus in G
ottingen, once said to F. Klein [1926, p. 249]: Riemann sang
damals schon wie ein Kanarienvogel (Already at that time Riemann sang like
a canary).
` 3
SEMINAIRES
ET CONGRES
207
Poincare wrote to Klein on March 30, 1882: Cetait un de ces genies qui
renouvellent si bien la face de la Science quils impriment leur cachet, non
seulement sur les uvres de leurs elèves immediats, mais sur celles de tous
leurs successeurs pendant une longue suite dannees. Riemann a cree une
theorie nouvelle des fonctions [Poincare 1882b, p. 107]. Indeed Riemanns
writings are full of almost cryptic messages to the future. The spirit
of Riemann will move future generations as it has moved us [Ahlfors 1953,
pp. 493, 501].
1.1 . Riemanns doctorate With his request of November 14, 1851, for admission to a doctorate, Riemann submits his vita. Of course this is in Latin as the
university laws demanded. On the following day, the Dean informs the faculty:
It is my duty to present to my distinguished colleagues the work of a
new candidate for our doctorate, Mr. B. Riemann from Breselenz; and
entreat Mr. Privy Councillor Gauss for an opinion on the latter and, if
it proves to be satisfactory, for an appropriate indication of the day and
the hour when the oral examination could be held. The candidate wants
to be examined in mathematics and physics. The Latin in the request
and the vita is clumsy and scarcely endurable: however, outside the
philological sciences, one can hardly expect at present anything better,
even from those who like this candidate are striving for a career at the
university.
15 Nov., 51.
Respectfully,
Ewald
Gauss complies with the Deans request shortly thereafter (undated, but certainly still in November 1851); the great man writes in pre-S
utterlin calligraphy the
following referees report:
The paper submitted by Mr. Riemann bears conclusive evidence of the
profound and penetrating studies of the author in the area to which the
topic dealt with belongs, of a diligent, genuinely mathematical spirit of
research, and of a laudable and productive independence. The work is
concise and, in part, even elegant: yet the majority of readers might
well wish in some parts a still greater transparency of presentation. The
whole is a worthy and valuable work, not only meeting the requisite
standards which are commonly expected from doctoral dissertations, but
surpassing them by far.
I shall take on the examination in mathematics. Among weekdays Saturday or Friday or, if need be, also Wednesday is most convenient to
me and, if a time in the afternoon is chosen, at 5 or 5:30 p.m. But I
also would have nothing to say against the forenoon hour 11 a.m. I am,
incidentally, assuming that the examination will not be held before next
week.
Gauss
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
208
R. REMMERT
It seems appropriate to add some comments. The Dean of the Faculty was the
well-remembered Protestant theologian Georg Heinrich August Ewald (1803-1875).
He was, as was the physicist Wilhelm Eduard Weber (1804-1891), one of the famous
G
ottinger Sieben who in 1837 protested against the revocation of the liberal constitution of the kingdom of Hannover by King Ernst August and lost their positions.
Knowing that Ewald was an expert in classical languages, in particular Hebrew grammar, one may understand his complaints about Riemanns poor handling of Latin.
It is to be regretted that Gauss says nothing about the mathematics as such in
Riemanns dissertation which - in part - had been familiar to him for many years.
Indeed, Riemann, when paying his formal visit to Gauss for the rigorosum, was
informed that for a long time he [Gauss] has been preparing a paper dealing with
the same topic but certainly not restricted to it; [Dedekind 1876, p. 545]. The paper
referred to here is Gausss article (Bestimmung der) Convergenz der Reihen, in
welche die periodischen Functionen einer veranderlichen Grosse entwickelt werden,
Gausss Werke X-1, pp. 400-419; cf. also Werke X-2, p. 209. The reader is unable to
learn from Gausss report even what topic is dealt with in the dissertation (geometry
or number theory or ...). Gauss is famous for his sparing praise and, of course, his
short report must be rated as a strong appraisal. For further details see [Remmert
1993b].
It is interesting to compare the evaluation with the one Gauss wrote in 1852
of Dedekinds dissertation. Here he simply writes (File 135 of the Philosophische
Fakult
at of the University of Gottingen): The paper submitted by Mr. Dedekind
[published in Dedekinds Werke I, pp. 1-26] deals with problems in calculus which
are by no means commonplace. The author not only shows very good knowledge in
this eld but also an independence which indicates favorable promise for his future
achievements. As paper for admission to the examination this text is fully sucient.
1.2.
Christian Felix Klein and the atlas principle
The rst to attempt to explain Riemanns conceptual methods to a broader

audience was Carl Neumann, [1865]. However, his Vorlesungen u
ber Riemanns Theorie der Abelschen Integrale from 1865 were beyond the scope of
the mathematical community. In the mid 1870s Felix Klein began to study
and grasp the richness of the revolutionary new ideas and became Riemanns
true interpreter. Later R. Courant called him the most passionate apostle of
Riemanns spirit [Courant 1926, p. 202]. Klein did away with the idea that
Riemann surfaces are lying a priori over the plane. He reports that in 1874
he learned from Friedrich Emil Prym that Riemann himself realized that his
He writes:
surfaces are not necessarily lying multiply sheeted over C.
Er [Prym] erz
ahlte mir, da die Riemannschen Flachen urspr
unglich
durchaus nicht notwendig mehrblattrige Flachen u
ber der Ebene sind, da
man vielmehr auf beliebig gegebenen krummen Flachen ganz ebenso komplexe Funktionen des Ortes studieren kann, wie auf den Flachen u
ber der
` 3
SEMINAIRES
ET CONGRES
209
Ebene., [Klein 1882a, pp. 502]

(He [Prym] told me that Riemann surfaces are as such primarily not necessarily multi-sheeted surfaces over the plane; that one rather can study complex
functions on arbitrarily given curved surfaces as on surfaces over the plane).
However in 1923 Klein revokes this and states that in 1882 Prym said
that he does not remember his conversation with Klein and that he never had
indicated anything of this kind [Klein 1923a, p. 479]. Here we have maybe a
case where a great idea springs from a remark the speaker does not remember
and which the listener misunderstood.
Kleins new approach to Riemann surfaces is by means of dierential geometry. On every real-analytic surface in R3 , if provided with the Riemannian
metric ds2 = Edp2 + 2F dpdq + Gdq 2 induced from ambient Euclidean space
R3 , there does exist, at least locally, a potential theory and hence a function
theory. One can argue as follows: according to Gauss [1822] locally there
always exist isothermal parameters x, y such that ds2 = (x, y)(dx2 + dy 2 )
holds. The map (x, y) x + iy is locally a conformal bijection of the surface onto a domain in C. Hence harmonic and holomorphic functions can be
locally dened in an invariant way.
Kleins arguments are heuristic and based on his interpretation of holomorphic functions in terms of electric elds. He used this method already
in [Klein 1882a]. Ten years later, in [Klein 1891-92], he states his ideas
rather clearly. He replaces (page 22) the surface in R3 by a zweidimensionale geschlossene Mannigfaltigkeit, auf welcher irgendein deniter Dierentialausdruck ds2 vorgegeben ist. Ob diese Mannigfaltigkeit in einem Raume
von 3 oder mehr Dimensionen gelegen ist oder auch unabhangig von jedem
ausseren Raum gedacht ist, das ist nun dabei ganz gleichg
ultig (two dimen
sional closed [=compact] manifold carrying an arbitrary ds2 metric. It does
not matter at all whether this manifold is lying in a space of 3 or more dimensions or whether it is thought of independently from any ambient space).
And then Klein, realizing that a conformal structure is needed only locally,
takes the decisive step from local to global by saying, [loc. cit., p. 26]:
Eine zweidimensionale, geschlossene, mit einem Bogenelement ds2 ausgestattete Mannigfaltigkeit (welche keine Doppelmannigfaltigkeit ist) ist jedenfalls dann als Riemannsche Mannigfaltigkeit [=Flache] zu brauchen, wenn
man sie mit einer endlichen Zahl von Bereichen dachziegelartig u
berdecken
kann, deren jedes eindeutig und konform auf eine schlichte Kreisscheibe abgebildet werden kann.
(A two dimensional closed orientable manifold with an element of arclength ds2 can always be used as a Riemann surface, if there exists a tile-like
covering by nitely many regions each of which permits a bijective conformal
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
210
R. REMMERT
mapping onto a disk).

Since the composition of conformal maps is eo ipso conformal, Klein needs
no compatibility conditions for his maps. Klein hesitates to allow atlases
with innitely many charts, cf. [loc. cit., p. 27]. For him, Riemann surfaces
are always compact. Non-compact surfaces and arbitrary atlases are rst
admitted in the work of Paul Koebe, [1908, p. 339]. However he does not yet
dare to call such objects Riemann surfaces.
In those days Riemann surfaces were only a helpful means to represent
multivalued functions. Klein was the rst to express the opposite opinion,
cf. [Klein 1882a, p. 555]: Die Riemannsche Flache veranschaulicht nicht nur
die in Betracht kommenden Funktionen, sondern sie deniert dieselben (The
Riemann surface is not just an illustration of the functions in question, rather
it denes them). Klein also forged an intimate alliance between Riemanns
ideas and invariant theory, algebra, number theory and - above all - group
theory: Verschmelzung von Riemann und Galois (fusion of Riemann and
Galois) was one of his aims.
Kleins tile-like coverings are nowadays called complex atlases with the tilemaps as charts. His procedure is the atlas principle which can be formulated
in todays language as follows. Consider a Hausdor space X and refer to a
topological map : U V of an open set U X onto an open set V C as
a chart on X. A family {Ui , i } of charts on X is called a complex atlas on
X if the sets Ui cover X and if each map
j 1
i : i (Ui Uj ) j (Ui Uj )
is biholomorphic. A maximal complex atlas is called a complex structure
on X. An (abstract) Riemann surface is a Hausdor space provided with a
complex structure. Every concrete Riemann surface is an abstract Riemann
surface. The converse is a deep theorem which requires the construction of
non-constant meromorphic functions, cf. section 2.4.
1.3.
Karl Theodor Wilhelm Weierstrass and analytic

congurations
The principle of analytic continuation was formulated by Weierstrass in [1842,

pp. 83-84] (published only in 1894); Riemann, [1857, p. 89], likewise describes
this method. For Weierstrass an analytic function is the set of all germs
of convergent Laurent series with nite principal part (which he just calls
In
power series) obtained from a given germ by analytic continuation in C.
todays language this is just a connected component of the sheaf space M of
meromorphic functions, where M carries its canonical topology.
` 3
SEMINAIRES
ET CONGRES
211
Analytic congurations [analytische Gebilde] arise from analytic functions

by attaching to them new points as follows: Consider the set M of all germs
Fc of convergent Puiseux series of the form

n>
an (z c)n/k if c C , resp.
an z n/k if c = ,
n>
where k is an arbitrary positive integer (for k = 1 we have Fc M). The

Fc c and the evaluation map : M C,
center map z : M C,
Fc Fc (c) are dened in an obvious way and, equipping M with its canonical topology (in the same way as is done for M), one readily proves the
following:
M is a topological surface (Hausdor ), M is open in M and its complement M M is locally nite in M . The maps z and are continuous.
In addition z is locally nite and, at points of M, a local homeomorphism.
Thus, by Riemanns denition (section 1.1.), the space M is a concrete
with respect to z, the functions z and are meromorRiemann surface over C
phic on M and M M is the set of winding points of z. Every connected

component X of M is an analytic conguration with X M as underlying
analytic function. The set X X M consists of all irregular germs of X.
Weierstrass analytic congurations (X, z, ) are (sophisticated) examples
of connected concrete Riemann surfaces, see also [Heins 1980]. Conversely, it
is a fundamental existence theorem that every connected concrete Riemann
surface is an analytic conguration. For compact surfaces this was shown
in [Riemann 1857] and [Weyl 1913] by using Dirichlets principle. For noncompact surfaces there seems to be no proof in the classical literature (see
section 2.4. for further details).
1.4.
The feud between G

ottingen and Berlin
Already Cauchy had the sound denition of holomorphic functions by differentiability rather than by analytic expressions. Riemann shared this view
whole heartedly. Everywhere in [Riemann 1851] he advocates studying holomorphic functions independently of their analytic expressions, e.g. he writes
on pages 70-71: Zu dem allgemeinen Begrie einer Function einer veranderlichen complexen Grosse werden nur die zur Bestimmung der Function
nothwendigen Merkmale hinzugef
ugt, und dann erst gehe man zu den verschiedenen Ausdr
ucken u
ber deren die Function fahig ist. (To the general
notion of a function of one complex variable one just adds those properties
necessary to determine the function [i.e., complex dierentiability], and only
then one passes to the dierent [analytic] expressions which the function is
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
212
R. REMMERT
capable of taking on). His convincing examples, on page 71, are meromorphic
functions on compact surfaces. They are algebraic functions and vice versa.
Riemanns credo is in sharp contrast to Weierstrass confession of faith
which he stated on October 3, 1875, in a letter to Schwarz:
[Ich bin der festen] Uberzeugung,

dass die Functionentheorie auf dem
Fundamente algebraischer Wahrheiten aufgebaut werden muss, und dass es
deshalb nicht der richtige Weg ist, wenn umgekehrt zur Begr
undung einfacher
und fundamentaler algebraischer Satze das Tranzendente, um mich kurz auszudr
ucken, in Anspruch genommen wird, so bestechend auch auf den ersten
Anblick z.B. die Betrachtungen sein mogen, durch welche Riemann so viele
der wichtigsten Eigenschaften der algebraischen Funktionen entdeckt hat
[Weierstrass 1875, p. 235].
(I am deeply convinced that the theory of functions must be founded
on algebraic truths, and that, conversely, it is not correct if, in order to
establish simple fundamental algebraic propositions, one has to recourse to
the transcendental (to put it briey), no matter how impressive at rst glance
the reections look like by means of which Riemann discovered so many of
the most important properties of algebraic functions).
It was Weierstrass dogma that function theory is the theory of convergent
Laurent series (he already studied such series in [Weierstrass 1841] and just
called them power series). Integrals are not permitted. The nal aim is always the representation of functions. Riemanns geometric yoga with paths,
cross-cuts, etc., on surfaces is excluded, because it is inaccessible to algorithmization. By pointing out in [Weierstrass 1870] the defects of Riemanns main
tool, the Dirichlet principle, Weierstrass won the rst round. Weierstrass criticism should have come as a shock, but it did not. People felt relieved of the
duty to learn and accept Riemanns methods. The approach by dierentiation
and integration was discredited. It is with regret that A. Brill and M. Noether
wrote: In solcher Allgemeinheit lat der [Cauchy-Riemannsche] Funktionsbegri, unfabar und sich ver
uchtigend, controlierbare Schl
usse nicht mehr zu
[Brill and Noether 1894, p. 265]. (In such generality the notion of a function
is incomprehensible and amorphous and not suited for veriable conclusions).
The denition of holomorphic functions by power series prevailed through
the rest of the 19th century. But already in 1903, W. F. Osgood ridiculed
the pride of the Weierstrass school to be able to base the theory on one limit
process only. He writes with respect to the unwillingness to give a rigorous
proof of the monodromy principle: For a school to take this stand, who for
puristic reasons are not willing to admit the process of integration into the
theory of functions of a complex variable, appears to be straining at a gnat
and swallowing a camel [Osgood 1903-04, p. 295].
` 3
SEMINAIRES
ET CONGRES
213
In spite of all opposition the advance of Riemanns way of thinking could

not be stopped. In 1897, in his Zahlbericht, Hilbert attempts to realize Riemanns principle of carrying out proofs merely by thought instead of by computation [Hilbert 1897, p. 67].
1.5.
Jules Henri Poincar

e and automorphic functions
In the early eighties Poincare contributed new and epoch-making ideas to

the theory of Riemann surfaces. In his CR-note [Poincare 1881] of February
14, 1881, he outlines his program: Study of (nitely generated) discontinuous groups G of biholomorphic automorphisms of the unit disc D and of
G-invariant meromorphic functions. He calls such groups, resp. functions,
groupes fuchsiens, resp. fonctions fuchsiennes. Non-constant fuchsian functions are constructed as quotients of -series
m

dg
ai z + bi
(z) =
H(gz)
=
H
(ci z + di )2m ,
dz
ci z + di
gG
i=1
where H is a rational function without poles on D and m 2 is an integer.

Thus one obtains new Riemann surfaces D/G with lots of non-constant meromorphic functions. In subsequent CR-notes Poincare sketches his theory, e.g.
the fundamental fact that, for a given group G, two fuchsian functions are
always algebraically dependent and that there exist two fuchsian functions
u, v such that every other fuchsian function is a polynomial in u and v. (The
eld of fuchsian functions is isomorphic to a nite extension of the rational
function eld C(X).)
In 1882 Poincare gives a detailed exposition of his result in two papers
[Poincare 1882c] in the just founded journal Acta mathematica. In the rst
paper he shows, by using for the rst time the non-euclidean geometry of the
upper half plane H, that there is a correspondence between fuchsian groups
and certain tilings of H by non-euclidean polygons. In the second paper he
gives two proofs for the normal convergence of his -series (p. 170-182).
Poincare does not use the methods of Riemann. In fact he was probably
not aware of them at that time. Dieudonne writes in [Dieudonne 1975, p. 53]:
Poincares ignorance of the mathematical literature, when he started his
researches, is almost unbelievable. He hardly knew anything on the subject
beyond Hermites work on the modular functions; he certainly had never
read Riemann, and by his own account had not even heard of the Dirichlet
principle.
Soon Poincare realized the uniformizing power of his functions. In his
CR-note [Poincare 1882a] of April 10, 1882, he announces the theorem that
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
214
R. REMMERT
for every algebraic curve (X, Y ) = 0 (of genus 2) there exist two nonconstant fuchsian functions F (z) and F1 (z) such that (F (z), F1 (z)) 0.
His proof is based (as Kleins proof, cf. section 1.6.) on a m
ethode de continuit
e [Poincare 1884, pp. 329]: equivalence classes of fuchsian groups, resp.
algebraic curves, are considered as points of varieties S, resp. S . There is a
canonical map S S and this turns out to be a bijection. The method had
to remain vague at a time when no general topological notions and theorems
were available. However, on June 14, 1882, Weierstrass wrote prophetically to
Sonia Kowalevskaja: Die Theoreme u
ber algebraische Gleichungen zwischen
zwei Ver
anderlichen . . ., welche er [Poincare] in den Comptes rendus gegeben
hat, sind wahrhaft imponierend; sie eronen der Analysis neue Wege, welche
zu unerwarteten Resultaten f
uhren werden [Mittag-Leer 1923, p. 183].
(The theorems about algebraic equations between two variables . . ., which
he gave in the Comptes rendus, are truly impressive, they open new roads to
analysis and shall lead to unexpected results.)
The notation fonction fuchsienne did not prevail. From the very beginning, Klein, who was in a state of feud with Fuchs, protested strongly
against this term in his letters to Poincare, cf. [Klein 1881-82]. But Poincare
remained unmoved, cf. [Poincare 1882b]. On April 4, 1882, he wrote conclusively: Il serait ridicule dailleurs, de nous disputer plus longtemps pour un
nom, Name ist Schall und Rauch et après tout, ca mest egal, faites comme
vous voudrez, je ferai comme je voudrai de mon cote.[Klein 1881-82, p. 611]
In the end, as far as functions are concerned, Klein was successful: in
[Klein 1890, p. 549], he suggested the neutral notation automorphic instead
of fuchsian, which has been used ever since. However, the terminology
groupe fuchsien has persevered.
1.6.
The competition between Klein and Poincar

e
Much has been said about the genesis of the theory of uniformization for algebraic Riemann surfaces and the competition between Klein and Poincare.
However there was never any real competition. Poincare, in 1881, had the
-series and hence was far ahead of Klein; as late as May 7, 1882, Klein asks
Poincare how he proves the convergence of his series [Klein 1881-82, p. 612].
It is true that Klein, unlike Poincare, was aware of most papers on special discontinuous groups, in particular those by Riemann, Schwarz, Fuchs, Dedekind
and Schottky, cf. [Klein 1923b]. At that time he was interested in those Riemann surfaces Xn , which are compactications of the quotient surfaces H/n ,
where n is the congruence subgroup of SL2 (Z) modulo n. For n = 7 this is
Kleins curve of genus 3 with 168 automorphisms; in [Klein 1879, p. 126],
` 3
SEMINAIRES
ET CONGRES
215
he constructs a beautiful symmetric 14-gon as a fundamental domain. But

Klein restricted himself to the consideration of fundamental domains which
can be generated by reection according to the principle of symmetry [Klein
1926, p. 376]. Of course he was aware of the connections between fundamental
domains and non-Euclidean geometry, but it seems that he never thought of
attaching a fundamental domain to an arbitrarily given discontinuous group.
According to Dieudonne [1975], Klein set out to prove the Grenzkreistheorem only after realizing that Poincare was looking for a theorem that would
give a parametric representation by meromorphic functions of all algebraic
curves. Klein succeeded in sketching a proof independently of Poincare, [Klein
1882b]. He used similar methods (suering from the same lack of rigor).
1.7.
Georg Ferdinand Ludwig Philipp Cantor and countability

of the topology
At a very early time the following question was already being asked: How
are obtained by anmany germs of meromorphic functions at a point a C
of a given germ at a? In other words: What is the
alytic continuation in C
cardinality of the bers of an analytic conguration? Clearly all cardinalities
0 are possible. In 1835 C. G. J. Jacobi knew that on a surface of genus
2 the set of complex values at a point a obtained by analytic continuation
of a germ of an Abelian integral can be dense in C [Jacobi 1835, 8]. In
1888 G. Vivanti conjectured that only cardinalities 0 can occur. Cantor
informed him that this is correct and that, already several years before, he
had communicated this to Weierstrass, cf. [Ullrich 1995].
In 1888 Poincare and Vito Volterra published proofs in [Poincare 1888],
resp. [Volterra 1888]. Their result can be stated as follows: Every connected
concrete Riemann surface X has countable topology ( i.e., a countable base of
open sets). At the bottom of this is a purely topological fact, cf. [Bourbaki
1961, Chap. 1, 11.7]. The Poincare-Volterra theorem implies at once that
an analytic conguration diers from its analytic function only by at most
countably many irregular germs.
1.8.
Karl Hermann Amandus Schwarz and universal covering

surfaces
The idea of constructing a universal covering surface originated with Schwarz

in 1882. On May 14, 1882, Klein writes to Poincare:
Schwarz denkt sich die Riemannsche Flache in geeigneter Weise zerschnit
ten, sodann unendlichfach u
berdeckt und die verschiedenen Uberdeckungen
in den Querschnitten so zusammengef
ugt, da eine Gesamtache entsteht,
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
216
R. REMMERT
welche der Gesamtheit der in der Ebene nebeneinander zu legenden Polygonen entspricht. Diese Gesamtache ist . . . einfach zusammenh
angend und
einfach berandet, und es handelt sich also nur darum, einzusehen, da man
auch eine solche einfach zusammenhangende, einfach berandete Flache in der
bekannten Weise auf das Innere eines Kreises abbilden kann [Klein 1881-82,
p. 616].
(Schwarz regards the Riemann surface as being dissected in a suitable
way, then innitely often covered and now these dierent coverings glued
together along the cross sections in such a way that there arises a total surface
corresponding to all polygons lying side by side in the plane. This total surface
is . . . simply connected and has only one boundary component. Thus it is only
necessary to verify that such a simply connected surface can be mapped in
the well known way onto the interior of a disc.)
Poincare immediately realized the depth of this idea. He writes back to
Klein on May 18, 1882: Les idees de M. Schwarz ont une portee bien plus
grande.
1.9.
The general uniformization theorem
Already in [1883] Poincare states and attempts to prove the general theorem of uniformization: Soit y une fonction analytique quelconque de x, non
uniforme. On peut toujours trouver une variable z telle que x et y soient
fonctions uniformes de z. In his Analyse [Poincare 1921], written in 1901,
he writes that he succeeded in triompher des dicultes qui provenaient de
la grande generalite du theorème à demontrer. Here he uses the universal
covering surface. In his Paris talk, when discussing his twenty-second problem Uniformization of analytic relations by automorphic functions, Hilbert
[1900, p. 323] points out, however, that there are some inconsistencies in
Poincares arguments. A satisfactory solution of the problem of uniformization was given in 1907 by Koebe and Poincare in [Koebe 1907] and [Poincare
1907a].
2.
Riemann surfaces from 1913 onwards
Classical access to Riemann surfaces is by Schere und Kleister (cut and

paste). It was not until 1913 that H. Weyl, in his seminal work Die Idee der
Riemannschen Fl
ache [1913], gave rigorous denitions and proofs. In 1922
T. Rad
o proved that the existence of a complex structure implies that the
surface can be triangulated. In 1943 H. Behnke and K. Stein constructed
non-constant holomorphic functions on every non-compact Riemann surface.
` 3
SEMINAIRES
ET CONGRES
217
Their results easily imply that all such connected surfaces are analytic congurations.
2.1.
Claus Hugo Herman Weyl and the sheaf principle
Inuenced by Hilberts denition of a (topological) plane in [Hilbert 1902],

Weyl rst introduces 2-dimensional connected manifolds which are locally
discs in R2 . However he does not postulate the existence of enough neighborhoods: his manifolds are not necessarily Hausdor. The separation axiom, cf.
[Hausdor 1914, pp. 211, 457], is still missing in 1923 in the second edition
of his book. In his encomium to Hilbert, Weyl [1944, p. 156] calls the paper
[Hilbert 1902] one of the earliest documents of set-theoretic topology. Furthermore he writes: When I gave a course on Riemann surfaces at Gottingen
in 1912, I consulted Hilberts paper . . .. The ensuing denition was given
its nal touch by F. Hausdor. This last sentence hardly gives full justice
to Hausdor. It is not known whether Hausdor pointed out to Weyl the
shortcomings of his denition.
Weyl assumes the existence of a triangulation in order to have exhaustions
by compact domains; 2-dimensional connected manifolds which can be triangulated he calls surfaces. He shows that countably many triangles suce,
hence the topology of his surfaces is countable.
In order to carry out function theory on a surface X along the same lines
as in the plane, the notion analytic function on the surface has to be introduced in such a way da sich alle Satze u
ber analytische Funktionen in der
Ebene, die im Kleinen g
ultig sind, auf diesen allgemeinen Begri u
bertragen (that all statements about analytic functions in the plane which are valid
locally carry over to this more general notion), cf. [Weyl 1913, p. 35]. Thus
the further procedure is nearly canonical. Weyl writes (almost verbatim):
For every point x X and every complex-valued function f in an arbitrary
neighborhood of x it must be explained when f is to be called holomorphic
at x and this denition must satisfy the conditions of compatibility. Clearly
Weyl comes near to the notion of the canonical presheaf of the structure sheaf
OX . His nal denition in todays language is:
A Riemann surface is a connected topological surface X with a triangulation and with a complex structure sheaf O.
Weyl immediately shows that analytic congurations are topological surfaces (the diculty is to triangulate them). He shares Kleins belief that
surfaces come rst and functions second. He writes, loc. cit., p. IV/V: Die
Riemannsche Fl
ache ... mu durchaus als das prius betrachtet werden, als
der Mutterboden, auf dem die Funktionen allererst wachsen und gedeihen
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
218
R. REMMERT
k
onnen (The Riemann surface must be considered as the prius, as the virgin
soil, where upon the functions foremost can grow and prosper).
Weyl covers all of classical function theory in his kleine Buch (booklet) of
only 167 pages. The topics, everyone for itself a monumentum aere perennius,
are:
existence theorems for potential functions and meromorphic functions,
analytic congurations are Riemann surfaces,
compact surfaces are algebraic congurations,
theorems of Riemann-Roch and Abel,
Grenzkreistheorem and theory of uniformization.
At bottom of all arguments is Dirichlets principle, which Hilbert [1904], had
awakened from a dead sleep.
Contrary to what has often been said, the book does not give a complete symbiosis of the concepts of Riemann and Weierstrass: The question
whether every connected non-compact Riemann surface is isomorphic to a
Weierstrassian analytic conguration, is not dealt with. In fact no convincing
proof was known in those days (see also paragraph 5 below).
2.2.
The impact of Weyls book on twentieth century

mathematics
Die Idee der Riemannschen Fl

ache was well ahead of its time. Not only did
it place the creations of Riemann and Klein on a rm footing, but, with its
wealth of ideas, it also foreshadowed coming events. Concepts like covering
surface, group of deck transformations, simply connected, genus and R
uckkehrschnittpaare (as priviledged bases of the rst homology group) occur as
a matter of course. In 1913 no one could surmise the impact Weyls work
would bring to bear on the mode of mathematical thinking in the twentieth
century.
An immediate enthusiastic review came from Bieberbach. He wrote (almost verbatim, cf. [Bieberbach 1913]):
Die Riemannsche Funktionentheorie hatte bisher ein eigent
umliches
Gespr
age, in dem die einem schon die Anzeichen des nahen Todes und den
Sieg der extrem Weierstraischen Richtung in der Funktionentheorie erhoten
oder bef
urchteten je nach der Gem
utsstimmung; Anzeichen jedoch, die in den
Augen der anderen der Theorie keinen Abbruch taten, da man u
berzeugt war,
` 3
SEMINAIRES
ET CONGRES
219
das werde sich alles noch in die Reihe bringen lassen, wenn die Zeit erst
erf
ullet sei. Und so ist es denn: Herr Weyl hat alles in die Reihe gebracht.
(Till now Riemanns function theory had a curious aura which some people
hopefully or fearfully saw, according to their mood, as a sign of approaching
death and victory of the extremely Weierstrassian route. Others did however
not see this as a sign that would do damage to the theory, because they were
convinced that everything could be put in order in due time. And so it is:
Mr. Weyl did put everything in order2 .)
The book was a real eye-opener and had a long lasting inuence. Kunihiko
Kodaira, in his famous Annals paper, writes: Our whole theory may be
regarded as a generalization of the classical potential theory. The famous
book of H. Weyl Die Idee der Riemannschen Flache has always served us as
a precious guide [Kodaira 1949, p. 588]. And Jean Dieudonne, calls the book
a classic that inspired all later developments of the theory of dierentiable
and complex manifolds [Dieudonne 1976, p. 283].
A reprint of the rst edition with corrections and addenda appeared in
1923. This second edition was reproduced in 1947 by the Chelsea Publishing
Company. A third completely revised edition appeared in 1955. The fourth
and fth edition followed in 1964 and 1974. The rst edition of Die Idee der
Riemannschen Fl
ache was never translated into a foreign language. A translation The concept of a Riemann surface of the third edition by G. R. MacLane
was published in 1964 by Addison-Wesley. There are no longer triangulations
and Weyl gives hints to the new notion of cohomology.
Weyl died soon after the third edition appeared. One cannot write a
better swan song. C. Chevalley and A. Weil wrote in their obituary: Qui de
nous ne serait satisfait de voir sa carrière scientique se terminer de meme ?
[Chevalley and Weil 1957, p. 668].
An annotated reissue of the book from 1913 was published in 1997 by
Teubner Verlag Leipzig where the rst edition was also printed.
2.3.
Tibor Rad
o and triangulation
In 1922 Rad
o realized that the existence of a complex structure on a connected
topological surface implies the countability of the topology and hence (in a
not trivial way which he underestimated) the existence of a triangulation.
2
Five years later the neophyte Ludwig Georg Elias Moses Bieberbach had turned into
an apostate. In [Bieberbach 1918, p. 314], he writes in words alluding to coming dark years
of German history: Bis jetzt sind die topologischen Betrachtungen noch nicht ausgeschaltet. Und damit frit noch immer ein Erz
ubel am Marke der Funktionentheorie (Till now
topological considerations are not exterminated. And thereby a pest is still gorging at the
marrow of function theory).
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
220
R. REMMERT
However he only gave a sketch of proof since Heinz Pr

ufer had told him that
every connected topological surface admits a triangulation [Rad
o 1923]. Soon
Pr
ufer found real analytic counterexamples. Then, in 1925, Rad
o published
his theorem using the main theorem of uniformization, cf. [Rad
o 1925]3 .
It should be mentioned in passing that, already in 1915, Hausdor knew
the existence of the long line (=a one dimensional connected topological
manifold with non-countable topology). He discussed this explicitly in his
private notes [Hausdor 1915].
For the denition of a complex structure Rad
o uses the atlas principle.
Thus Rad
o was the rst to introduce Riemann surfaces in the way which
has been used ever since: A Riemann surface is a topological surface with a
complex structure.
2.4.
Heinrich Adolph Louis Behnke, Karl Stein and

non-compact Riemann surfaces
As Riemann and Klein knew and as was proved rigorously by Weyl, there exist
many non-constant meromorphic functions on every abstract connected Riemann surface and the compact ones are even algebraic congurations. A natural question is: Are there non-constant holomorphic functions on every abstract non-compact connected Riemann surface? In the thirties Caratheodory
strongly propagated this problem. Classical approaches by forming quotients
of dierential forms, resp. Poincare-series, fail due to possible zeros in the
denominators. Only in 1943 Behnke and Stein were able to give a positive
answer in their paper [Behnke and Stein 1947-49] (publication was delayed
due to the war). They developed a Runge approximation theory for holomorphic functions on non-compact surfaces and reaped a rich harvest. There are
lots of holomorphic functions. In fact they proved the following fundamental
theorem (Hilfssatz C at the end of [Behnke and Stein 1947-49]).
Let A be a locally nite set in an abstract non-compact Riemann surface
X. Assume that to every point a A there is attached (with respect to a local
na

ca ta , na 0. Then
coordinate ta at a) a nite Laurent series ha =
>
there exists in X \ A a holomorphic function f having at each point a A a

ca ta .
Laurent series of the form ha +
>na
Today there exist simpler proofs: Take a compact disc U in the surface X and construct
(e.g. by solving a Dirichlet problem on U by means of the Perron-principle) a non-constant
harmonic function on X U . Then the universal covering of X U has non-constant
holomorphic functions and hence, by the theorem of Poincare and Volterra, a countable
topology. Now it follows directly that X U and therefore X itself has a countable toplogy.
` 3
SEMINAIRES
ET CONGRES
221
In particular this implies:

Every non-compact abstract Riemann surface X is a concrete Riemann
surface z : X C over the complex plane C.
In addition R. C. Gunning and R. Narasimhan showed in [1967] that the
function z can be chosen in such a way that its dierential dz has no zeros.
Hence X can even be spread over C without branching points (domaine
etal
e).
The theorem of Behnke and Stein has consequences in abundance. Let us
mention just two of them.
Every non-compact Riemann surface X is a Stein manifold (cf. 3.6.).
Every divisor on a non-compact Riemann surface X is a principal divisor.
2.5.
Analytic congurations and domains of meromorphy
Every meromorphic function f on a connected concrete Riemann surface

determines an analytic conguration: Choose a schlicht point
z : X C
p X and consider the analytic conguration (Xf , z , f ) containing the
germ (f z 1 )z(p) which arises by pulling down the germ fp to z(p) by means
This conguration is independent of the choice of p and there
of z : X C.
is a natural holomorphic map : X Xf such that z = z and f = f .
The map is injective if z(X) contains a dense set A such that f separates
every z-ber over A. If is bijective, we identify Xf with X, z with z and
f with f and then call (X, z, f ) the analytic conguration of the function f
and X the domain of meromorphy of f (with respect to z).
Theorem Every non-compact connected concrete Riemann surface
is the domain of meromorphy of a function f holomorphic on X.
z:X C
Such a function f is obtained in the following way. The above theorem of
Behnke and Stein implies the existence of a function g O(X), g
= 0, with a
zero set which has every boundary point of X as a point of accumulation.
This last statement can be made precise by using a method developed by
H. Cartan and P. Thullen [1932] to handle corresponding problems in several
variables. Multiplication of g with a suitably chosen function h O(X) yields
a holomorphic function f on X which vanishes at the zeros of g (and may be
elsewhere) and which in addition separates enough z-bers to show that X is
a domain of meromorphy.
The theorem completes the symbiosis of Riemannian and Weierstrassian
function theory. It was rst stated (with a meromorphic function f ) by Koebe
in his CR-Note [Koebe 1909]; twenty years later Stoilow deals with Koebes
realization theorem in his book [Stoilow 1938, chap. II]. In 1948 Herta
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
222
R. REMMERT
Florack, a student of Behnke and Stein, proved the theorem along the lines
indicated above [Florack 1948].
3.
Towards complex manifolds, 1919-1953
Riemann surfaces are one dimensional complex manifolds. The general notion
of a complex manifold came up surprisingly late in the theory of functions of
several complex variables. Of course higher-dimensional complex tori had
already been implicitly studied in the days of Abel, Jacobi and Riemann: the
periods of integrals of Abelian dierentials on a compact Riemann surface
of genus g immediately assign a g-dimensional complex torus to the surface.
And non-univalent domains over C n were in common use since 1931 through
the work of H. Cartan and P. Thullen. Nevertheless, the need to give a general
denition was only felt by complex analysts in the forties of this century. At
that time the notion of a general manifold was already well understood by
topologists and dierential geometers.
3.1.
Global complex analysis until 1950
The theory of functions of several complex variables has its roots in papers by
P. Cousin, H. Poincare and F. Hartogs written at the end of the nineteenth
century. The points of departure were the Weierstrass product theorem and
the Mittag-Leer theorem. The fact that zeros and poles are no longer isolated caused diculties. These problems were studied for more than 50 years
in domains of C n only. In the thirties and forties of this century the theory
of functions of several complex variables was a dormant theory. There were
only two books. A so-called Lehrbuch [1929] by W. F. Osgood (Harvard) at
Teubner, and an Ergebnissebericht by H. Behnke and P. Thullen (M
unster)
at Springer [Behnke and Thullen 1934]. In addition there were some original
papers in German and French by Behnke, Caratheodory, Cartan, Hartogs,
Kneser, Oka and Stein. Osgood, however, even then thought that the theory
was so complicated that one could only write about it in German. And it
is said that Cartan asked his students who wanted to learn several complex
variables: Can you read German? If answered in the negative, his advice was
to look for a dierent eld.
Among the main topics of complex analysis in the thirties and forties were
the following, cf. [Behnke and Thullen 1934]:
analytic continuation of functions (Kontinuit
atssatz) and distribution of
singularities,
` 3
SEMINAIRES
ET CONGRES
223
the Levi problem,

the Cousin problems,
domains and hulls of holomorphy,
automorphisms of circular domains (Cartans mapping theorem).
In the beginning Riemanns classical mapping theorem was a catalyst. But
already in 1907 Poincare knew that bounded domains in C 2 of the topological type of a ball are not always (biholomorphically) isomorphic to a ball,
[Poincare 1907b]. Karl Reinhardt [1921] proved that polydiscs and balls in
C 2 are not isomorphic. In 1931, H. Cartan classied all bounded domains in
C 2 which have innitely many automorphisms with a xed point (domaines
cercl
es) [Cartan 1931a]. In 1933 Elie and Henri Cartan showed that every
bounded homogeneous domain in C 2 is isomorphic to a ball or a polydisc
[Cartan 1933, p. 462]. For further details see [Ullrich 1996].
In Germany, Riemanns mapping theorem served as a misguiding compass
for rather a long time; Ernst Peschl (Bonn) once told the author that in his
youth - under the spell of Caratheodory - he wasted many hours with hopeless
mapping problems.
The state of the art in those decades is reected by four quotations:
a) Malgre le progrès de la theorie des fonctions analytiques de plusieurs variables complexes, diverses choses importantes restent plus ou moins obscures
[Oka 1936].
b) Trotz der Bem
uhungen ausgezeichneter Mathematiker bendet sich die
Theorie der analytischen Funktionen mehrerer Variablen noch in einem recht
unbefriedigendem Zustand [Siegel 1939]. (In spite of the eorts of distinguished mathematicians the theory of analytic functions of several variables
is still in a rather unsatisfactory state.)
c) Letude generale des varietes analytiques, et des fonctions holomorphes
sur ces varietes, est encore très peu avancee [Cartan 1950, p. 655].
d) The theory of analytic functions of several complex variables, in spite of
a number of deep results, is still in its infancy [Weyl 1951].
3.2.
Non-univalent domains over C n , 1931-1951:

Henri Cartan and Peter Thullen.
In disguise complex manifolds made their rst appearance in function theory

of several complex variables in 1931 as non-univalent domains over C 2 in a
paper of H. Cartan. In [Cartan 1931b] he draws attention to Hartogs domains
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
224
R. REMMERT
in C 2 which are homeomorphic to a ball and have, in todays language, a nonunivalent hull of holomorphy. One year later, when writing their paper, Cartan and Thullen [1932] made virtue out of necessity. They study domains over
C n , i.e. complex manifolds with a projection into C n . They wisely restrict
themselves to the unramied case, where the projection is everywhere a local
isomorphism. Their denition is that used in the Ergebnisbericht [Behnke and
Thullen 1934, p. 6].
3.3.
Dierentiable manifolds, 1919-1936: Robert K

onig, Elie
Cartan, Oswald Veblen and John Henry Constantine
Whitehead, Hassler Whitney.
Abstract Riemann surfaces were already well understood when abstract

dierentiable surfaces were not yet even dened. In higher dimensions it
was the other way around: abstract dierentiable manifolds came rst and
were extensively studied by topologists and dierential geometers. Complex
manifolds were just a by-product. Everything sprang forth from Riemanns
Habilitationsschrift [Riemann 1854] Ueber die Hypothesen, welche der Geometrie zu Grunde liegen (On the hypotheses which are the basis of geometry).
The philosophical concept of n fach ausgedehnte Grosse (n-fold extended
quantity) guides Riemann to n-dimensional manifolds with a Riemannian
metric. Coming generations tried and nally succeeded to give a precise
meaning to these visions.
The concept of a global dierential manifold was already roughly dened
in [1919] by R. K
onig and later used by E. Cartan, [1928, 50, 51]. However
the rst to attempt a rigorous and precise denition were O. Veblen and J. H.
C. Whitehead in 1931-32, cf. [Veblen and Whitehead 1931] and their Cambridge Tract [Veblen and Whitehead 1932]. Their axioms seem rather clumsy
today, but they did serve the purpose of putting the subject on a rm foundation, cf. [Milnor 1962]. Their work had a lasting inuence, e.g. H. Whitney
refers to it in his profound paper [Whitney 1936] lapidarily entitled Dierentiable manifolds. Here, by using approximation techniques, Whitney shows
that abstract manifolds always have realizations in real number spaces. More
precisely every connected n-dimensional dierentiable manifold with countable
topology is dieomorphic to a closed real analytic submanifold of R2n+1 . He
poses the problem of whether any real analytic manifold can be analytically
embedded into a Euclidean space and says that this is probably true. The
positive answer was given in 1958 by H. Grauert using his solution of the Levi
problem and the fact that Stein manifolds can be embedded into complex
number spaces [Grauert 1958b].
General dierentiable manifolds already appeared in 1935 in the textbook
` 3
SEMINAIRES
ET CONGRES
225
by P. Alexandrov and H. Hopf where they devote the last pages 548-552 to
vector elds on such manifolds.
3.4.
Complex manifolds, 1944-1948:

Constantin Carath
eodory, Oswald Teichmu
ller, Shiing
Shen Chern, Andr
e Weil and Heinz Hopf
From the very beginning it was felt that Riemanns approach to complex
analysis should also bear fruits in higher dimension. But only in 1932, at the
International Congress in Z
urich, did Caratheodory in [Caratheodory 1932]
strongly advocate studying four dimensional abstract Riemann surfaces (as
he called them) for their own sake. However, due to his rather cumbersome
approach, there was no response by his contemporaries.
Only after dierentiable and real analytic manifolds had already been
studied intensively, and with great success, was time ripe for complex manifolds. It seems dicult to locate the rst paper where complex manifolds
explicitly occur. In 1944 they appear in Teichm
ullers work on Veranderliche
Riemannsche Fl
achen, [Teichm
uller 1944, p. 714]; here we nd for the rst
time the German expression komplexe analytische Mannigfaltigkeit. The
English complex manifold occurs in 1946 in Cherns work [1946, p. 103];
he recalls the denition (by an atlas) just in passing. And in 1947 we nd
variete analytique complexe in the title of Weils paper [1947]. Overnight
complex manifolds blossomed everywhere. Let us just call attention to Hopfs
papers [1948] and [1951]. The rst one contains, among others, the result
that the spheres S 4 and S 8 with their usual dierentiable structures cannot
be provided with a complex structure. The second one is a beautifully written
survey reecting the state of the theory at that time.
In 1953 Borel and Serre showed, that a sphere S 2n , n 4, carrying an
arbitrary dierentiable structure, never admits an almost complex structure
[Borel and Serre 1953, p. 287].
3.5.
The French Revolution, 1950-1953: Henri Cartan and

Jean-Pierre Serre
I remember from my student days a lecture by H. Cartan in M

unster in
December 1949 (his rst lecture at a German university after the war). He
was proselytizing in those days for the great, new ideas of ber bundles on
complex manifolds. From that time on the development was breath taking.
It was only three years after Cartans lament at the Cambridge congress,
at a colloquium in Brussels, that he and his student Serre presented to a
dumbfounded audience their theory of Stein manifolds. This culminated with
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
226
R. REMMERT
two theorems on cohomology groups with coecients in coherent analytic

sheaves ([Cartan 1953], [Serre 1953], see also next paragraph). A German
participant commented tersely: We have bows and arrows, the French have
tanks.
Whoever wants to recapture the struggle for mastery of the new ideas
should read Serres letters to his matre, Les petits cousins [Serre 1952].
The fundamental new concept was the notion of a coherent analytic sheaf.
Overnight sheaves appeared everywhere in complex analysis. Il faut faisceautiser (we must sheafy), was the motto of this French revolution. In
1953, these Sturm und Drang years were already history. It took time to
become accustomed to the new way of thinking. But there is the force of
habit. One remembers C. G. J. Jacobi who once remarked:
Da es n
amlich in der Mathematik darauf ankommt, Schl
usse auf Schl
usse
zu h
aufen, wird es gut sein, so viele Schl
usse als moglich in ein Zeichen zusammenzuh
aufen. Denn hat man dann ein f
ur alle Mal den Sinn der Operation
ergr
undet, so wird der sinnliche Anblick des Zeichens das ganze Rasonnement
ersetzen, das man fr
uher bei jeder Gelegenheit wieder von vorn anfangen
mute. (As in mathematics it is important to accumulate conclusion after conclusion, so it will be good to gather together as many conclusions as
possible in one symbol. For, if the meaning of the operation has been established once and for all, then the sensory perception of the symbol will replace
the whole line of reasoning that previously had to be each time started from
scratch.) For analytic sheaf theory this symbol may well be H q (X, S).
3.6.
Stein manifolds
In his memorable work [Stein 1951], Karl Stein introduced complex manifolds
which share basic properties with non-compact Riemann surfaces and domains of holomorphy in C n . These manifolds were baptized Stein manifolds
by Cartan4 . Following the original denition, a complex manifold X with
countable topology is called a Stein manifold if the following three axioms are
satised:
Separation axiom: Given two dierent points p, p in X there exists a
function f holomorphic on X which takes dierent values at p and p .
Local coordinates axiom: For every point p X there exist functions
f1 , . . . , fn holomorphic on X which give local coordinates on X at p.
4
In the fties Cartan liked to tease Stein at meetings in Oberwolfach: Cher ami, avez
vous aujourdhui une vari
et
e de vous dans votre poche? When Stein lectured about his
manifolds he circumvented the notation by varying a well known phrase of Montel: ... les
variet
es dont jai lhonneur de porter le nom.
` 3
SEMINAIRES
ET CONGRES
227
Convexity axiom: For every innite, locally nite set M in X there exists
a function f holomorphic on X which is unbounded on M .
A domain in C n is a Stein manifold if and only if it is a domain of holomorphy; every non-compact Riemann surface is a Stein manifold. Many theorems
about domains of holomorphy can be extended to Stein manifolds. Cartan
obtained the
Main Theorem [Cartan 1951-52] For every coherent analytic sheaf S
over a Stein manifold X the following two statements hold:
1) The global sections of S generate every OX -module Sx , x X.
2) All cohomology groups H q (X, S), q 1, vanish.
This theorem was rst proved in [Cartan 1951-52]. It contains, among
many others, the classical results pertaining to the Cousin problems (cf. [Cartan 1953], [Serre 1953]).
In [1955], Grauert showed that X is already a Stein manifold if the rst
two axioms are replaced by the following:
Weak separation axiom: For every point p X there exists a holomorphic
map f : X C n such that p is an isolated point in its ber f 1 (f (p)).
Moreover Grauert proved that every connected complex manifold satisfying this weak separation axiom has eo ipso a countable topology. (Note that
E. Calabi and M. Rosenlicht in [1953], constructed 2-dimensional connected
complex manifolds without countable bases of open sets.)
4.
Complex spaces, 1951-1960
Complex spaces are complex manifolds with singularities. Singularities were,

of course, already known in Riemanns days: for him singularities were mainly
double points [Riemann 1857, 6]. A systematic study of singularities was
started by Alfred Clebsch, Max Noether and Italian geometers in the last
century. In Weyls book, singularities are not discussed.
When complex manifolds came into life it was clear from the very beginning that they were not general enough. The singularity of w2 z1 z2 = 0
at the origin shows that one has to admit spaces which locally are not even
homeomorphic to an open set in Rn . However singular points were not considered for a long time. When studying non-univalent domains over Cn in
the thirties and forties, mathematicians excluded possible branching, because
they were well aware of the mysteries lying hidden in the ramication points.
Still in 1951 Kiyoshi Oka complains: On ne sait presque rien sur les domaines
interieurement ramies, [Oka 1951, p. 128].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
228
R. REMMERT
4.1.
Normal complex spaces, 1951
In order to include singularities one needs a category of local model spaces

larger than the category of open sets in C n . In 1951-52 two suggestions were
made: H. Behnke and K. Stein, in their paper [1951], chose nite, analytically
ramied coverings of domains in C n as local models; H. Cartan, in his seminar
[1951-52], used special analytic sets in domains of C n as local representatives
(cf. exp. 13, p. 3).
A characteristic feature of both denitions is that the complex spaces
which are obtained from these categories by local patching are locally irreducible and that their holomorphic functions are exactly those continuous functions which are holomorphic in the classical sense at all smooth
points. For Behnke-Stein spaces the powerful Riemann extension theorem
for bounded holomorphic functions is valid. For Cartan spaces the structure
sheaf is normal: every stalk is a normal ring, i.e. an integral domain which is
integrally closed in its quotient eld; it is for this reason that Cartans spaces
are called normal complex spaces.
While Behnke and Stein proceed in the geometric spirit of Riemanns
covering principle, Cartans approach is in the algebraic spirit of Weierstrass
and Dedekind. Indeed he has immediately at his disposal the local Weierstrass
theory of convergent power series (preparation theorem, etc.), whereas Behnke
and Stein cannot even be sure that there are locally enough holomorphic
functions to separate nearby points.
Using local Weierstrass theory it is a matter of routine to show that every
normal complex space is a Behnke-Stein space. The converse is not at all
obvious; it comes down to proving the following
Theorem Every nite, analytically ramied covering of a complex manifold is a normal complex space.
This was carried out in [Grauert and Remmert 1958].
So nally, one hundred years after Riemanns creation, at the same time,
in dierent places, higher dimensional Riemann surfaces were born. One is
reminded of a owery line in a letter of Farkas Bolyai to his son J
anos from
spring 1825: [Manche Dinge] haben gleichsam eine Epoche, wo sie dann
an mehreren Orten aufgefunden werden, gleichwie im Fr
uhjahr die Veilchen
mehrw
arts ans Licht kommen.(Certain things just have their epoque, when
they are found at dierent places, just as in spring when violets come into
light everywhere).5
5
Bolyai encourages his son to publish the discovery of non-Euclidean geometry straight
` 3
SEMINAIRES
ET CONGRES
4.2.
229
Reduced complex spaces, 1955
In 1954 Cartan called his spaces espaces analytiques generaux [Cartan 195354, exp. 6, p. 9]. But they were not general enough: soon it became clear
that spaces having reducible points with local components not necessarily of
the same dimension also had to be admitted. In 1955, Serre, in his GAGA
paper [Serre 1956], allowed all analytic sets in domains of C n as local models.
Holomorphic functions now are exactly those continuous functions which are
locally restrictions of functions holomorphic in ambient C n . The complex
spaces belonging to this category are called reduced since all stalks in their
structure sheaves are reduced rings, i.e., without non-zero nilpotent elements.
There may be however zero divisors
= 0 (for example if the space consists of
two dierent lines through a point of C 2 ).
Important properties of local function theory in C n remain true for reduced complex spaces. In particular the convergence theorem of Weierstrass
holds: the limit of a locally uniformly convergent sequence of holomorphic
functions is holomorphic, cf. [Grauert and Remmert 1958, p. 290]. Furthermore Hartogs theorem remains true: a complex-valued function f on a
cartesian product X Y of reduced complex spaces X, Y is holomorphic on
X Y , if for every pair of points x X, y Y the restrictions f |x Y resp.
f |Xy are holomorphic on Y resp. X, [loc. cit. p. 292, p. 56].
4.3.
Complex spaces with nilpotent holomorphic functions,

1960
Serres denition of a complex space seemed to be the end of the journey.

However the study of bers of holomorphic maps shows that reduced complex
spaces do not yet t all purposes. For example the 2-fold covering C
C, z z 2 , has the origin 0 as winding point and it is natural to attach
to the ber over 0 the 2-dimensional C-algebra O0 /O0 z 2 which has non-zero
nilpotent elements. This leads to the following category of local models: A
pair (A, OA ) is called a complex model space, if there exists a domain D in
C n , 1 n < , and a coherent sheaf of ideals J OD such that A is the
zero set of J in D and OA is the restriction of the sheaf OD /J to A. Reduced
spaces arise if J is its own radical. The structure sheaf of an arbitrary complex
space is no longer a subsheaf of the sheaf of continuous functions, i.e., there
may be non-zero nilpotent holomorphic functions which are invisible to the
geometric eye.
away; cf. P. St
ackel: Die Entdeckung der nichteuklidischen Geometrie durch J. Bolyai,
Math. Naturw. Ber. Ungarn, vol. XVII, 1-19 (1901), p. 14.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
230
R. REMMERT
The complex spaces obtained by local patching of such complex model

spaces were introduced by Grauert [1960]. The way to this concept had
been paved before in Algebraic Geometry by Alexandre Grothendieck. At a
noteworthy meeting in Strasbourg in May 1957, Grauert and Grothendieck
exchanged their ideas, cf. [Remmert 1993a]. The new way of thinking caused
diculties even for well educated mathematicians: in his lectures at Harvard
in 1958, Grothendieck always carried a small card in his breast pocket inscribed by John Tate that he pulled out during the discussion periods: There
may be nilpotent elements. (Source: The Unreal Life of Oscar Zariski, by
Carol Parikh, Acad. Press 1990, p.155). By the early sixties everything was
settled and the new spaces were simply called complex spaces.
Such an extension of the concept of a complex space was necessary. Indeed,
in the late fties Grauert was struggling with the proof of his famous
Theorem Mapping Theorem [Grauert 1960, p. 287] Let X, Y be complex

spaces and let f : X Y be a proper holomorphic mapping. Then for every
coherent analytic sheaf S over X all the higher direct images fn (S), n N,
are coherent over Y .
He was compelled to allow nilpotent elements in order to be able to use
the full force of power series expansions (innitesimal neighborhoods). The
implications of the mapping theorem are tremendous: the niteness theorem
of Cartan and Serre [1953], is a corollary (just let Y be a point). Furthermore
it is obvious that the image set f (X) = support of f0 (OX ) is an analytic set
in Y .
There are further generalizations of the notion of a complex space. In his thèse,
Adrien Douady [1966] introduced innite-dimensional complex spaces (espaces analytiques banachiques). Here the local models are analytic sets in Banach spaces.
Douady needs this remarkable category for the proof that the set H(X) of all compact analytic subspaces of a given complex space X carries, in a canonical way, a
complex structure; only in the end H(X) turns out to be of nite dimension. Another
generalization is that of a relative complex space. For instance, one may consider
maps X Y , where Y is a real dierentiable manifold, and where complex structures which are compatible with , are given on the bers. Such spaces occur in a
natural way in the deformation theory of complex structures. Still another generalization, with a parallel theory, is that of p-adic rigid spaces (and non rigid ones as
well).
` 3
SEMINAIRES
ET CONGRES
231
Epilogue
The notion of a complex space with nilpotent functions in its structure sheaf
is a beautiful example of how algebraic notions nally overgrow an analyticgeometric theory. According to Felix Klein geometers have the peculiar joy
of seeing what they are thinking. Algebraic presentations are abstract, mercilessly abstract (E. Artin, Collected Papers, p. 538). Algebraization of local
function theory started with Weierstrass, the rst real breakthrough coming
in 1933 with the paper by Walter R
uckert Zum Eliminationsproblem der
Potenzreihenideale [1933], written in 1931. He proved that the ring of convergent power series in n variables is noetherian and factorial. Furthermore he
obtained the Nullstellensatz (only the henselian property is missing). R
uckert wrote his paper in Freiburg (Krull) under the spell of Emmy Noether and
proudly writes that he only needs formal methods and no function theory: In
dieser Arbeit wird gezeigt, da eine sachgemae Behandlung des Eliminationsproblems ... nur formale Methoden, also keine funktionentheoretischen Hilfsmittel ben
otigt. Als solche Methoden erweisen sich die allgemeine Idealtheorie
und die allgemeine K
orpertheorie. [R
uckert 1933, p. 260] (In this paper it
is shown that a proper treatment of elimination theory only requires formal
methods and no aid from function theory. Such methods are the general theory of ideals and of elds.) R
uckerts statement is not quite true: in addition
he uses the full power of the preparation theorem. Complex analysts did not
pick up R
uckerts new way of thinking in the thirties and R
uckerts paper fell
into oblivion.
The true algebraization of local function theory took place only in the
fties in Cartans seminaire [1960-61] in four exposes written by Christian
Houzel called Geometrie analytique locale. This approach was not welcomed
everywhere with pleasure; some people felt that this was a Kings road to
chaos. The question was: Is algebra helping geometry or is it perhaps the
other way round? In his lecture entitled The Fundamental Ideas of Abstract
Algebraic Geometry at the International Congress 1950 in Cambridge, Oscar
Zariski found a wise answer: In helping geometry modern algebra is helping
itself above all. (Coll. Pap. III, p. 375). Already in 1939 Hermann Weyl
had prophetically written the timeless lines [Weyl Ges. Abh. III, p. 681]:
In these days the angel of topology and the devil of abstract
algebra ght for the soul of each individual mathematical domain.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
232
R. REMMERT
Bibliography
Ahlfors (L.V.)
[1953]
Development of the theory of conformal mapping and Riemann surfaces

through a century, in Contributions to the theory of Riemann surfaces,
Ann. of Math. Stud., vol. 30, Princeton Univ. Press, 1953, pp. 313. See
also Collected Papers I, Boston-Basel: Birkhauser, 1982, pp. 493-501.
Alexandrov (P.) and Hopf (H.)
[1935]
Topologie. Berlin: Springer, 1935.
Behnke (H.) and Stein (K.)

[1947-49] Entwicklung analytischer Funktionen
Math. Ann., 120 (1947-49), pp. 430461.
[1951]
auf
Riemannschen
Flachen,
Modikation komplexer Mannigfaltigkeiten und Riemannsche Gebiete,

Math. Ann., 124 (1951), pp. 116.
Behnke (H.) and Thullen (P.)

[1934]
Theorie der Funktionen mehrerer komplexer Ver

anderlichen, Berlin:
Springer, 1934. Second enlarged edition 1970.
Bieberbach (L.)
[1913]
Review of Die Idee der Riemannschen Fl

ache, Jahresber. Deutsch.
Math.-Verein., 22 (1913), pp. 216219.
[1918]
Uber
die Einordnung des Hauptsatzes der Uniformisierung in die Weierstraische Funktionentheorie, Math. Ann., 78 (1918), pp. 312331.
Biermann (K.R.)
[1976]
Weierstrass, Karl Theodor Wilhelm, Dictionary of Scientic Biography,

vol. 14, 1976, pp. 219224.
Borel (A.) and Serre (J.-P.)

[1953]
Groupes de Lie et puissances reduites de Steenrod, Amer. J. Math., 75

(1953), pp. 409448. See also Borel, uvres I, Springer, 1983, pp. 262-301;
not in Serres uvres.
Bourbaki (N.)
[1961]
Elements de mathematiques, Livre III: Topologie g

enerale, Paris: Hermann,
3rd ed, 1961.
Brill (A.) and Noether (M.)

[1894]
Die Entwicklung der Theorie der algebraischen Functionen in alterer und

neuerer Zeit, Jahresber. Deutsch. Math.-Verein., 3 (1894), pp. 107566.
Burau (W.) and Schoeneberg (B.)
[1973]
Klein, Christian Felix, Dictionary of Scientic Biography, vol. 7, 1973,

pp. 396400.
` 3
SEMINAIRES
ET CONGRES
233
Calabi (E.) and Rosenlicht (M.)

[1953]
Complex analytic manifolds without countable base, Proc. Amer. Math.

Soc., 4 (1953), pp. 335340.
odory (C.)
Carathe
[1932]
Uber
die analytischen Abbildungen von mehrdimensionalen Raumen, in
Verh. Int. Math. Kongress Z
urich, 1932, vol. I, pp. 93101. See also Ges.
Math. Schriften, IV, M
unchen: Beck, 1956, pp. 234-246.
Cartan (E.)
[1928]
Lecons sur les espaces de Riemann. Paris: Gauthier-Villars, 1928.
Cartan (H.)
[uvres] uvres, 3 vols., Berlin: Springer, 1979.
[1931a]
Les fonctions de deux variables complexes et le problème de la representation analytique, J. Math. Pures Appl., (IX) 10 (1931), pp. 1114; uvres
I, pp. 141-254.
[1931b] Sur une classe remarquable de domaines, C. R. Acad. Sci. Paris, 192 (1931),
pp. 10771079; uvres I, pp. 370372.
[1933]
Sur les groupes de transformations pseudo-conformes, C. R. Acad. Sci.

Paris, 196 (1933), pp. 993995; uvres I, pp. 460-462.
[1950]
Problèmes globaux dans la theorie des fonctions analytiques de plusieurs

variables complexes, in Proc. Int. Congr. Math., Cambridge, 1950, vol. I,
pp. 152164; uvres II, pp. 654-666.
[1951-52] Fonctions analytiques de plusieurs variables complexes, Paris: Seminaire

E.N.S., 1951-52.
[1953]
Varietes analytiques complexes et cohomologie, in Colloque sur les fonctions

de plusieurs variables, Bruxelles, 1953, Centre Belge Rech. math., pp. 4155;
uvres II, pp. 669-683.
[1953-54] Fonctions automorphes et espaces analytiques, Paris: Seminaire E.N.S.,

1953-54.
[1960-61] Familles despaces complexes et fondements de la g
eometrie analytique,
Paris: Seminaire E.N.S., 1960-61.
Cartan (H.) and Serre (J.-P.)
[1953]
Un theorème de nitude concernant les varietes analytiques compactes,

C. R. Acad. Sci. Paris, 237 (1953), pp. 128130; see also H. Cartan, uvres
II, pp. 684-686, and Serre, uvres I, pp. 271-273.
Cartan (H.) and Thullen (P.)

[1932]
Zur Theorie der Singularitaten der Funktionen mehrerer komplexen Veranderlichen. Regularitats- und Konvergenzbereiche, Math. Ann., 106 (1932),
pp. 617647 ; see also H. Cartan, uvres I, pp. 376-406.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
234
R. REMMERT
Chern (S.S.)
[1946]
Characteristic classes of Hermitian manifolds, Ann. of Math., 46 (1946),

pp. 85121 ; Selected Papers I, New York: Springer, 1978, pp. 101-137.
Chevalley (C.) and Weil (A.)

[1957]
Hermann Weyl (1885-1955), Enseign. Math., (II) 3, (1957), pp. 157187;

see also H. Weyl, Ges. Abh. IV, pp. 655-685.
Clebsch (A.)
[1865]
Ueber diejenigen ebenen Curven, deren Coordinaten rationale Functionen

eines Parameters sind, J. Reine Angew. Math, 64 (1865), pp. 4365.
Courant (R.)
[1926]
Felix Klein, Jahresber. Deutsch. Math.-Verein., 34 (1926), pp. 197213.
Dedekind (R.)
[Werke] Gesammelte mathematische Werke, 3 vols, Braunschweig, 1930-1932.
[1876]
Bernhard Riemanns Lebenslauf, in Riemanns Werke, pp. 539558.
Dieudonn
e (J.)
[1975]
Poincare, Jules Henri, Dictionary of Scientic Biography, vol. 11, 1975,

pp. 5161.
[1976]
Weyl, Hermann, Dictionary of Scientic Biography, vol. 14, 1976, pp. 281
285.
Douady (A.)
[1966]
Le problème des modules pour les sous-espaces analytiques compacts dun

espace analytique donne, Ann. Inst. Fourier, 16 (1966), pp. 195.
Florack (H.)
[1948]
Regul
are und meromorphe Funktionen auf nicht geschlossenen Riemannschen Fl
achen, Schriftenreihe Math. Inst. Univ. M
unster, 1 (1948),
34 pp.
Freudenthal (H.)
[1975]
Riemann, Georg Friedrich Bernhard, Dictionary of Scientic Biography,

vol. 11 (1975), pp. 447456.
Gauss (C.F.)
[Werke] Werke, 12 vols, Gottingen, 1863-1929.
[1822]
Allgemeine Au
osung der Aufgabe die Theile einer gegebenen Flache auf
einer andern gegebenen Flache so abzubilden dass die Abbildung dem Abgebildeten in den kleinsten Theilen ahnlich wird, Copenhagener Preisschrift,
1822, in Werke, vol. IV, Gottingen, 1873, pp. 189216.
` 3
SEMINAIRES
ET CONGRES
235
Grauert (H.)
[Papers] Selected papers, 2 vols, Berlin: Springer, 1994.
[1955]
Charakterisierung der holomorph vollstandigen komplexen Raume, Math.

Ann., 129 (1955), pp. 233259; Papers I, pp. 165-191.
[1958a]
Die Riemannschen Flachen der Funktionentheorie mehrerer Veranderlichen,

in Proc. Int. Congr. Math., Edinburgh 1958, pp. 362375.
[1958b] On Levis problem and the embedding of real-analytic manifolds, Ann. of

Math., 68 (1958), pp. 460472; Papers I, pp. 192-204.
[1960]
Ein Theorem der analytischen Garbentheorie und die Modulraume kom

plexer Strukturen, Inst. Hautes Etudes
Sci. Publ. Math., 5 (1960), pp. 233
292; Papers II, pp. 512-571.
Grauert (H.) and Remmert (R.)

[1957]
Espaces analytiquement complets, C. R. Acad. Sci. Paris, 245 (1957),

pp. 882885; see also Grauert, Papers I, pp. 267-270.
[1958]
Komplexe R
aume, Math. Ann., 136 (1958), pp. 245318; see also Grauert,
Papers I, pp. 9-82.
Gunning (R.C.) and Narasimhan (R.)

[1967]
Immersion of open Riemann surfaces, Math. Ann., 174 (1967), pp. 103108.
Hausdorff (F.)
[1914]
Grundz
uge der Mengenlehre, Leipzig, 1914. Reprint New York: Chelsea,
1949.
[1915]
Private notes (1915). Fasc. 121, p. 1-16, in box 31, Univ. Bibl. Bonn.
Heins (M.)
[1980]
Weierstrassian global complex analysis from the Riemannian point of view,

in Aspects of contemporary complex analysis, Acad. Press 1980, pp. 225249.
Hilbert (D.)
[Ges. Abh.] Gesammelte Abhandlungen,
Springer, 2nd ed. 1970.
vols.,
Berlin-Heidelberg-New York:
[1897]
Die Theorie der algebraischen Zahlenkorper, Jahresber. Deutsch. Math.Verein., 4 (1897), pp. 175546; Ges. Abh. I, pp. 63-363.
[1900]
Mathematische Probleme, Nachr. K. Ges. Wiss. G

ottingen, (1900), pp. 253
297; Ges. Abh. III, pp. 290-329.
[1902]
Uber
die Grundlagen der Geometrie, Nachr. K. Ges. Wiss. G
ottingen,
(1902), pp. 233241. Enlarged version in Math. Ann. 56 (1903), pp. 381422;
not in Hilberts Ges. Abh.; but in Grundlagen der Geometrie, Anhang IV,
pp. 178-230, Leipzig: Teubner, 1903.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
236
[1904]
R. REMMERT
Uber
das Dirichletsche Prinzip, Math. Ann., 59 (1904), pp. 161186;
Ges. Abh. III, pp. 1537.
Hopf (H.)
[1948]
Zur Topologie der komplexen Mannigfaltigkeiten, in Studies and Essays presented to R. Courant, New York: Interscience, 1948, pp. 167185.
[1951]
Uber
komplex-analytische Mannigfaltigkeiten, Rend. Mat. Univ. Roma, 10
(1951), pp. 169182.
Houzel (C.)
[1978]
Fonctions elliptiques et integrales abeliennes, in Abrege dhistoire des

mathematiques 1700-1900 (edited by J. Dieudonne), vol. 2, Paris: Hermann
1978, pp. 1113.
Hurwitz (A.)
[Werke] Mathematische Werke, 2 vols., Basel: Birkhauser, 1932-1933.
[1891]
Uber
Riemannsche Flachen mit gegebenen Verzweigungspunkten, Math.
Ann., 39 (1891), pp. 161; Werke I, pp. 321-383.
[1893]
Uber
algebraische Gebilde mit eindeutigen Transformationen in sich,
Math. Ann., 41 (1893), pp. 403442. See also Werke I, pp. 391-430.
Jacobi (C.G.J.)
[1835]
De functionibus duarum variabilium quadrupliciter periodicis, quibus theoria transcendentium Abelianarum innititur, J. Reine Angew. Math, 13
(1835), pp. 5578; Gesammelte Werke 2, Berlin, 1882 pp. 23-50.
Klein (F.)
[Ges. Abh.] Gesammelte Mathematische Abhandlungen, 3 vols. Berlin: Springer, 19211923.
[1879]
Uber
die Transformation siebenter Ordnung der elliptischen Funktionen,
Math. Ann., 14 (1879), pp. 428471; Ges. Abh. 3, pp. 90-136.
[1881-82] Briefwechsel zwischen F. Klein und H. Poincare in den Jahren 1881-82,

Ges. Abh., 3. pp. 587621.
[1882a]
Uber
Riemanns Theorie der algebraischen Funktionen und ihrer Integrale,
Leipzig, 1882; Ges. Abh. 3, pp. 499-573.
[1882b] Uber
eindeutige Funktionen mit linearen Transformationen in sich. Zweite
Mitteilung. [Das Grenzkreistheorem], Math. Ann., 20 (1882), pp. 4951;
Ges. Abh. 3, pp. 627-629.
[1890]
Zur Theorie der allgemeinen Lameschen Funktionen, Nachr. K. Ges. Wiss.

G
ottingen, 4 (1890), Ges. Abh., 2 (1890), pp. 540549.
[1891-92] Riemannsche Fl
achen. Vorlesungen, gehalten in G
ottingen 1891/92.
Reprinted by Springer-Teubner, Leipzig, 1986.
` 3
SEMINAIRES
ET CONGRES
237
[1894-95] Riemann und seine Bedeutung f

ur die Entwicklung der modernen
Mathematik, Jahresber. Deutsch. Math.-Verein., 4 (1894-95), pp. 7187;
Ges. Abh. 3, pp. 482-497.
[1923a]
Vorbemerkungen zu den Arbeiten u

ber Riemannsche Funktionentheorie,
1923, Ges. Abh., 3, pp. 477481.
[1923b] Zur Vorgeschichte der automorphen Funktionen, 1923, Ges. Abh., 3, pp. 577
586.
[1926]
Vorlesungen u
ber die Entwicklung der Mathematik im 19. Jahrhundert,
vol. 1, Berlin: Springer, 1926.
Kodaira (K.)
[1949]
Harmonic elds in Riemannian manifolds (generalized potential theory),

Ann. of Math., 50 (1949), pp. 587665. See also Collected Works, vol. I,
Priceton Univ. Press, 1975, pp. 172-250.
Koebe (P.)
[1907]
Uber
die Uniformisierung beliebiger analytischer Kurven, Nachr. K. Ges.
Wiss. G
ottingen, (1907), pp. 191210, 633669.
[1908]
Uber
die Uniformisierung beliebiger analytischer Kurven, Dritte Mitteilung,
Nachr. K. Ges. Wiss. G
ottingen, (1908), pp. 337360.
[1909]
Fonction potentielle et fonction analytique ayant un domaine dexistence

donne à un nombre quelconque (ni ou inni) de feuillets, C. R. Acad. Sci.
Paris, 148 (1909), pp. 14461448.
K
onig (R.)
[1919]
Beitr
age zu einer allgemeinen Mannigfaltigkeitslehre, Jahresber. Deutsch.
Math.-Verein., 28 (1919), pp. 213228.
Laugwitz (D.)
[1996]
Bernhard Riemann 1826-1866. Basel: Birkhauser, 1996.
Milnor (J.W.)
[1962]
The work of J. H. C. Whitehead, in J. Whitehead Works I, pp. xxi-xxxiii.
Mittag-Leffler (G.)
[1923]
Weierstrass et Sonja Kowalewsky, Acta Math., 39 (1923), pp. 133198.
Neumann (C.)
[1865]
Vorlesungen u
ber Riemanns Theorie der Abelschen Integrale, Leipzig, 1865;
2nd ed., 1884.
Oka (K.)
[1936]
Sur les fonctions analytiques de plusieurs variables, I. Domaines convexes

par rapport aux fonctions rationnelles, J. Sci. Hiroshima Univ., 6 (1936),
pp. 245255.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
238
[1951]
R. REMMERT
Lemme fondamental, J. Math. Soc. Japan, 3 (1951), pp. 204214, 259278.

See also Sur les fonctions analytiques de plusieurs variables, Tokyo: Iwanami
Shoten, 1961, pp. 127-137; English translation in Okas Collected Papers,
pp. 109-132, Springer, 1984.
Osgood (W.F.)
[1903-04] On a gap in the ordinary presentation of Weierstrass theory of functions,
Bull. Amer. Math. Soc., III 10 (1903-04), pp. 294301.
[1929]
Lehrbuch der Funktionentheorie, vol. II, 2nd ed., Teubner, 1929.
Poincar
e (H.)
[uvres] uvres de Henri Poincar
e, 11 vols. Paris: Gauthier-Villars, 1928-1956.
[1881]
Sur les fonctions fuchsiennes, C. R. Acad. Sci. Paris, 92 (1881), pp. 333335;
uvres, II, pp. 1-4.
[1882a]
Sur les fonctions fuchsiennes, C. R. Acad. Sci. Paris, 94 (1882), pp. 1038
1040; uvres II, pp. 41-43.
[1882b] Extrait dune lettre a` M. F. Klein, Math. Ann., 20 (1882), pp. 5253; uvres; II, pp. 106-107; not in [Klein 1881-82].
[1882c]
Theorie des groupes fuchsiens, et Sur les fonctions fuchsiennes, Acta Math.,
1 (1882), pp. 162, 193294; uvres, II, pp. 108-168, 169-257.
[1883]
Sur un theorème de la theorie generale des fonctions, Bull. Soc. Math.

France, 11 (1883), pp. 112125; uvres, IV, pp. 57-69.
[1884]
Sur les groupes des equations lineaires, Acta Math., 4 (1884), pp. 201311;
uvres, II, pp. 300-401.
[1888]
Sur une propriete des fonctions analytiques, Rend. Circ. Mat. Palermo, 2
(1888), pp. 197200; uvres, IV, pp. 11-13.
[1907a]
Sur luniformisation des fonctions analytiques, Acta Math., 31 (1907), pp. 1

63. See also uvres, IV, pp. 70-139.
[1907b] Les fonctions analytiques de deux variables complexes et la representation

conforme, Rend. Circ. Mat. Palermo, 23 (1907), pp. 185220; uvres IV,
pp. 244-289.
[1921]
Analyse de ses travaux sur la theorie generale des fonctions dune variable
(written in 1901), Acta Math., 38 (1921), pp. 6570; uvres, IV, pp. 1-8.
Puiseux (V.)
[1850-51] Recherches sur les fonctions algebriques, J. Math. Pur. Appl., 15 (1850),
pp. 365480; 16, (1851), pp. 228240.
` 3
SEMINAIRES
ET CONGRES
239
(T.)
Rado
[1923]
Bemerkung zur Arbeit des Herrn Bieberbach: Uber

die Einordnung des
Hauptsatzes der Uniformisierung in die Weierstrasche Funktionentheorie
(Math. Ann. 78), Math. Ann., 90 (1923), pp. 3037.
[1925]
Uber
den Begri der Riemannschen Flache, Acta Sci. Math. (Szeged), 2
(1925), pp. 101121.
Reinhardt (K.)
[1921]
Uber
die Abbildungen durch analytische Funktionen zweier Veranderlicher,
Math. Ann., 83 (1921), pp. 211255.
Remmert (R.)
[1956]
Sur les espaces analytiques holomorphiquement separables et holomorphiquement convexes, C. R. Acad. Sci. Paris, 243 (1956), pp. 118121.
[1993a]
Komplexe Analysis in Sturm und Drang, Mitt. DMV, 1/93 (1993), pp. 513.
English translation in Math. Intelligencer, 17-2, (1995), pp. 411.
[1993b] The Riemann-le Nr. 135 of the Philosophische Fakultat of the Georgia
Augusta at G
ottingen, Math. Intelligencer, 15-3 (1993), pp. 4448.
Remmert (R.) and Schneider (M.)
[1997]
Analysis situs und Flachentheorie, in re-edition of H. Weyl: Die Idee der

Riemannschen Fl
ache, Teubner 1997, pp. 183-195.
Riemann (B.)
[Werke] Gesammelte Mathematische Werke (H. Weber and R. Dedekind, eds.),
Leipzig, 1876.
[1851]
Grundlagen f
ur eine allgemeine Theorie der Functionen einer ver
anderlichen complexen Gr
oe. Inauguraldissertation, Gottingen 1851; Werke,
pp. 3-48.
[1854]
Ueber die Hypothesen, welche der Geometrie zu Grunde liegen, 1854; Werke,
pp. 272287.
[1857]
Theorie der Abelschen Functionen, J. Reine Angew. Math., 54 (1857),

pp. 115155; Werke, pp. 88-142.
Roch (G.)
[1865]
Uber
die Anzahl der willk
urlichen Constanten in algebraischen Functionen,
J. Reine Angew. Math., 64 (1865), pp. 372376.
R
uckert (W.)
[1933]
Zum Eliminationsproblem der Potenzreihenideale, Math. Ann., 107 (1933),

pp. 259281.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
240
R. REMMERT
Schumacher (G.)
[1996]
Uber
die Entwicklung der Komplexen Analysis in Deutschland vom Ausgang des 19. Jahrhunderts bis zum Anfang der siebziger Jahre, Jahresber.
Deutsch. Math.-Verein., 98 (1996), pp. 41133.
Serre (J.-P.)
[uvres] uvres, 3 vols. Berlin-Heidelberg: Springer, 1986.
[1952]
Les petits cousins, in Miscellanea mathematica (edited by P. Hilton,

F. Hirzebruch and R. Remmert), Springer, 1991, pp. 277291; not in Serres
uvres.
[1953]
Quelques problèmes globaux relatifs aux varietes de Stein, in Colloque sur

les fonctions de plusieurs variables, Bruxelles, 1953, Centre Belge Rech.
math., pp. 5768; uvres I, pp. 259-270.
[1956]
Geometrie algebrique et geometrie analytique, Ann. Inst. Fourier, 6 (1956),

pp. 142; uvres I, pp. 402-443.
Siegel (C.)
[1939]
Einfuehrung in die Theorie der Modulfunktionen n-ten Grades, Math. Ann.,

116 (1939), pp. 617657.
Stein (K.)
[1951]
Analytische Funktionen mehrerer komplexer Veranderlichen zu vorgegebenen Periodizit

atsmoduln und das zweite Cousinsche Problem, Math. Ann.,
123 (1951), pp. 201222.
Stoilow (S.)
[1938]
Lecons sur les principes topologiques de la theorie des fonctions analytiques,

Paris: Gauthier-Villars, 1938.
Teichm
uller (O.)
[1944]
Ver
anderliche Riemannsche Flachen, Deutsche Math., 7 (1944), pp. 344359;
Gesammelte Abhandlungen, Berlin-Heidelberg: Springer, 1982, pp. 712-727.
Ullrich (P.)
[1995]
Georg Cantor, Giulio Vivanti und der Satz von Poincare-Volterra, in
Tagungsband des IV. Osterreichischen

Symposions zur Geschichte der
Mathematik (edited by C. Binder), Wien: Osterr.

Ges. Wiss. Geschichte,
1995, pp. 101107.
[1996]
The Riemann mapping problem, Rend. Circ. Mat. Palermo, (II) Suppl. 44
(1996), pp. 942.
Veblen (O.) and Whitehead (J.H.C.)

[1931]
A set of axioms for dierential geometry, Proc. Nat. Acad. Sci. USA, 17
(1931), pp. 551561. See also J. Whitehead Works I, pp. 93-103.
` 3
SEMINAIRES
ET CONGRES
[1932]
241
The Foundations of Dierential Geometry, Cambridge Univ. Press, 1932.

See also J. Whitehead Works I, pp. 105-205.
Volterra (V.)
[1888]
Sulle funzioni analitiche polidrome, Rend. Accad. Lincei, IV-2 (1888),

pp. 355361; Opere Matematiche 1, Roma: Accad. Nazion. Lincei, 1954,
pp. 356-362.
Weierstrass (K.)
[Werke] Mathematische Werke, 7 vols. 1894-1927.
[1841]
Zur Theorie der Potenzreihen, (1841), Werke I, pp. 6774.
[1842]
Denition analytischer Funktionen vermittelst algebraischer Dierentialgleichungen, (1842), Werke I, pp. 7584.
[1870]
Uber
das sogenannte Dirichletsche Prinzip, (1870), Werke II, pp. 4954.
[1875]
Aus einem bisher noch nicht veroentlichten Briefe an Herrn Professor

Schwarz, vom 3. October 1875, Werke II, pp. 235244.
Weil (A.)
[1947]
Sur la theorie des formes dierentielles attachees à une variete analytique

complexe, Comment. Math. Helv., 20 (1947), pp. 110116; uvres I,
New York-Heidelberg: Springer, 1979, pp. 374-380.
Weyl (H.)
[Ges. Abh.] Gesammelte Abhandlungen, 4 vols., Berlin-Heidelberg: Springer, 1968.
[1913]
Die Idee der Riemannschen Fl

ache. Teubner, Leipzig, 1913. Annotated reedition, 1997.
[1944]
David Hilbert and his mathematical work, Bull. Amer. Math. Soc., 50
(1944), pp. 612654; Ges. Abh. 4, pp. 130-172.
[1951]
A half-century of mathematics, Amer. Math. Monthly, 58 (1951), pp. 523

553.
Whitehead (J.)
[Works] The Mathematical Works of J.H.C. Whitehead 3 vols., Oxford-London:
Pergamon Press, 1962.
Whitney (H.)
[1936]
Dierentiable manifolds, Ann. of Math., 37 (1936), pp. 645680.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
On the History of Hilberts

Twelfth Problem
A Comedy of Errors
Norbert Schappacher
Abstract
Hilberts 12th problem conjectures that one might be able to
generate all abelian extensions of a given algebraic number eld
in a way that would generalize the so-called theorem of Kronecker and Weber (all abelian extensions of Q can be generated
by roots of unity) and the extensions of imaginary quadratic elds
(which may be generated from values of modular and elliptic functions related to elliptic curves with complex multiplication). The
rst part of the lecture is devoted to the false conjecture that
Hilbert made for imaginary quadratic elds. This is discussed
both from a historical point of view (in that Hilberts authority
prevented this error from being corrected for 14 years) and in
mathematical terms, analyzing the algebro-geometric interpretations of the dierent statements and their respective traditions.
After this, higher-dimensional analogues are discussed. Recent
developments in this eld (motives, etc., also Heegner points) are
mentioned at the end.
R
esum
e
Le douzième problème de Hilbert propose une facon conjecturale
dengendrer les extensions abeliennes dun corps de nombres, en
generalisant le theorème dit de Kronecker et Weber (toutes les
extensions abeliennes de Q sont engendrees par des racines de
AMS 1991 Mathematics Subject Classication: 01A60, 20-03, 11G15, 11R37

Universite Louis Pasteur, I.R.M.A., 7 rue Rene Descartes, 67084 Strasbourg Cedex.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
244
N. SCHAPPACHER
lunite) ainsi que les extensions des corps quadratiques imaginaires (qui sont engendrees par des valeurs de fonctions modulaires et elliptiques liees aux courbes elliptiques a` multiplication
complexe). La première partie de lexpose est centree autour de la
conjecture incorrecte de Hilbert dans le cas du corps quadratique
imaginaire. Elle est dicutee aussi bien du point de vue historique
(pendant quatorze ans, lautorite de Hilbert empecha la decouverte de cette erreur), que du point de vue mathematique, en analysant les interpretations algebro-geometriques des enonces dierents relatifs a` ce cas et de leurs traditions. On discute ensuite des
analogues en dimension superieure. Les developpements recents
(motifs, etc., aussi points de Heegner) sont mentionnes à la n.
A good problem should be

well motivated by already established theories or results,
challenging by its scope and diculty,
suciently open or vague, to be able to fuel creative research for a long
time to come, maybe for a whole century.
David Hilbert tried to follow these precepts in his celebrated lecture Mathematische Probleme at the Paris International Congress of Mathematicians in
1900.1 He did not have time to actually present in his speech all 23 problems
which appear in the published texts.2 In particular, the 12th problem on the
generalization of the Kronecker-Weber Theorem by the theory of Complex
Multiplication did not make it into the talk. This may be due to the slight
technicality of the statements involved. But Hilbert held this 12th problem
in very high esteem. In fact, according to Olga Tausskys recollection, when
he introduced Fueters lecture Idealtheorie und Funktionentheorie at the
1932 International Congress at Z
urich, Hilbert said that the theory of complex multiplication (of elliptic modular functions) which forms a powerful link
between number theory and analysis, is not only the most beautiful part of
mathematics but also of all science.3
1
[ICM 1900, pp. 58-114] (French translation by L. Laugel of an original German version),
[Hilbert 1901] (denite German text), cf. [Alexandrov 1979].
2
[Reid 1970, p. 81f]. See also Enseign. Math., 2 (1900), pp. 349-355.
3
Obituary Notice for Hilbert in Nature, 152 (1943), p. 183. I am grateful to J. Milne for
giving me this reference. In [ICM 1932, p. 37], one reads about Hilbert presiding over this
rst general talk of the Z
urich congress: Der Kongress ehrt ihn, indem die Anwesenden
sich von ihren Sitzen erheben.
` 3
SEMINAIRES
ET CONGRES
ON THE HISTORY OF HILBERTS TWELFTH PROBLEM
245
The present article covers in detail a period where a number of initial mistakes by most mathematicians working on the problem were nally straightened out. At the end of the 1920s the explicit class eld theory of imaginary
quadratic elds was established and understood essentially the way we still see
it today. However, the higher dimensional theory of singular values of Hilbert
modular forms remained obscure. Later developments are briey indicated in
the nal section of the paper.
What I describe here in detail is a comedy for us who look back. It is
genuinely amusing to see quite a distinguished list of mathematicians pepper
their contributions to Hilberts research programmme with mistakes of all
sorts, thus delaying considerably the destruction of Hilberts original conjecture which happened to be not quite right. The comedy is at the same time
a lesson on how, also in mathematics, personal authority inuences the way
research progresses or is slowed down. It concerns the condition of the
small group of researchers who worked on Hilberts 12th problem. The errors
made are either careless slips or delusions brought about by wishful thinking
which was apparently guided by Hilberts claim. The authors were just not
careful enough when they set up a formalism which they controlled quite well
in principle (a weakness in the formalism may, however, be behind the big
error in Webers false proof of the KroneckerWeber Theorem see section 2 below). Meanwhile Hilbert was conspicuously absent from the scene
after 1900.4 This is also not atypical for the comedy where the characters
are mostly left to themselves when it comes to sorting out their complicated
situation:
Say, is your tardy master now at hand? ...
Ay, Ay, he told his mind upon mine ear.
Beshrew his hand, I scarce could understand it.
Spake he so doubtfully, thou couldst not feel his meaning?
Nay, he struck so plainly, I could too well feel his blows; and
withal so doubtfully, that I could scarce understand them.
(Shakespeare, The comedy of errors, II-1)
The history of complex multiplication has already received a certain attention
in the literature see in particular the well-researched book [Vladut 1991].
Apart from newly introducing a few details into the story, my main dierence
4
Hilbert did intervene indirectly, as thesis advisor. As such he should have been better
placed han anybody else to see, for example, that Takagis thesis of 1901 produced extensions
that provided counterexamples to Fueters thesis of 1903... See section 3 below.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
246
N. SCHAPPACHER
with existing publications is the emphasis that I put on Hilberts peculiar

perspective of his problem, which is not only very much dierent from our
current viewpoint, but seems also to be the very reason which led him to the
slightly wrong conjecture for imaginary-quadratic base elds in the rst place.
As for the style of exposition, I try to blend a general text which carries
the overall story, with some more mathematical passages that should be understandable to any reader who knows the theories involved in their modern
presentation.
I take the opportunity to thank the organizers of the Colloquium in
honour of Jean Dieudonne, Materiaux pour lhistoire des math
ematiques au
ecle, at Nice in January 1996, for inviting me to contribute a talk. I
XXe si`
also thank all those heartily who reacted to earlier versions of this article and
made helpful remarks, in particular Jean-Pierre Serre and David Rowe.
1.
Hilberts statement of the Twelfth Problem
Coming back to the features of a good problem stated at the beginning, let us
look at the motivation which Hilbert chose for his 12th problem. He quoted
two results.
First, a statement going back to Kronecker, as Hilbert says, and which
is known today as the Theorem of Kronecker and Weber. It says that every
Galois extension of Q with abelian Galois group is contained in a suitable
cyclotomic eld, i.e., a eld obtained from Q by adjoining suitable roots of
unity. This was indeed a theorem at the time of the Paris Congressalthough
not proved by the person Hilbert quoted. . . We will briey review the history
of this result in section 2 below.
Second, passing to Abelian extensions of an imaginary quadratic eld,
Hilbert recalled the Theory of Complex Multiplication. As Hilbert puts it:
Kronecker himself has made the assertion that the Abelian equations
in the domain of an imaginary quadratic eld are given by the transformation equations of the elliptic functions [sic! ] with singular moduli so that,
according to this, the elliptic function [sic! ] takes on the role of the exponential function in the case considered before.5 The slight incoherence of
this sentence, which goes from certain elliptic functions (pluralas in Kro5
Kronecker selbst hat die Behauptung ausgesprochen, da die Abelschen Gleichungen

im Bereiche eines imagin
aren quadratischen K
orpers durch die Transformationsgleichungen
der elliptischen Funktionen mit singul
aren Moduln gegeben werden, so da hiernach die
elliptische Funktion die Rolle der Exponentialfunktion im vorigen Falle u
bernimmt. [Hilbert
1901, p. 311].
` 3
SEMINAIRES
ET CONGRES
247
neckers6 standard usage in this context) to the elliptic function (denite

singular), is not a slip.7 In fact, it gives the key to Hilberts interpretation
of Kronecker, and to his way of thinking of the 12th problem. What Hilbert
actually means here becomes crystal clear in the nal sentence on the 12th
problem, because there he expands the singular the elliptic function into
the elliptic modular function.8 So Hilbert was prepared, at least on this
occasion, to use the term elliptic function also to refer to (elliptic) modular
functions, i.e., to (holomorphic, or meromorphic) functions f : H C,
where H = { C | ( ) > 0} denotes the complex upper half plane, such
that

a + b
a b
) = f ( ),
for all H,
SL2 (Z).
f(
c d
c + d
And Hilberts denite singular, the elliptic (modular) function, refers undoubtedly to the distinguished holomorphic modular function j : H C
which extends to a meromorphic function j : H {i} C with a simple
pole at i, where it is given (up to possible renormalization by some rational
factor, in the case of some authors) by the well-known Fourier development
in q = e2i :
1
j(q) = + 744q + 196884q + 21493760q 2 + . . .
q
See for instance [Weber 1891, 41] who calls this function simply die Invariante, and cf. [Fueter 1905, p. 197], a publication on this problem which
arose from a thesis under Hilberts guidance.
To be sure, this was and is not at all the standard usage of the term elliptic
function. Rather, following Jacobidespite original criticism from Legendre
who had used the term to denote what we call today elliptic integralsit was
customary as of the middle of the 19th century to call elliptic functions the
functions that result from the inversion of elliptic integrals, i.e., the (meromorphic) doubly periodic functions with respect to some lattice. If one takes
the lattice to be of the form Z + Z , for H, then a typical example of such
an elliptic function is Weierstrasss well-known -function

1
1
1
,
(z, ) = 2 +
z
(z m n)2 (m + n)2
m,n
For instance [Kronecker 1877, p. 70], [Kronecker 1880, p. 453]. Cf. section 4 below.
Laugel missed this in his French translation of the text [ICM 1900, p. 88f], and thereby
blurred the meaning of the sentence.
8
. . . diejenigen Funktionen . . . , die f
ur einen beliebigen algebraischen Zahlk
orper die
entsprechende Rolle spielen, wie die Exponentialfunktion f
ur den K
orper der rationalen
Zahlen und die elliptische Modulfunktion f
ur den imagin
aren quadratischen Zahlk
orper.
[Hilbert 1901, 313].
7
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
248
N. SCHAPPACHER
where the prime restricts the summation to pairs (m, n) = (0, 0).
Also Kronecker seems to have reserved the term elliptic function for
these doubly periodic functions which depend on two parameters: the lattice
(or the modulus, in a terminology going back to Legendre) and a complex
number z modulo the lattice. His frame of reference for the theory of these
functions was Jacobis formalism, not Weierstrasss, but since the translation
back and forth between these two formalisms was routine by the end of the
19th century, we do not elaborate on this here.
However, when Kronecker speaks of transformation equations of elliptic
functions as he does in the very passage that Hilbert picked up, this may
be ambiguous in that the transformations aect in general both parameters.
So as an extreme case these transformation equations might describe functions which no longer depend on the point variable z at all, and behave with
respect to the lattice-variable like a modular function. As a matter of fact,
in another key passage where Kronecker states his Jugendtraum, he mentions
two dierent sorts of algebraic numbers to be used to generate the Abelian
extensions of an imaginary quadratic eld: the singular moduli of elliptic
functions, and those values of elliptic functions with a singular modulus
where the complex argument (i.e., z, in our notation) is rationally related to
the periods.9
Today, one calls singular moduli the values j( ) for those H which
satisfy a (necessarily imaginary) quadratic equation over Q. In Kronecker,
modulus has to be understood as alluding to the quantity k or in Legendres normal form of the elliptic integrals, or in Jacobis formalism. Once
the Weierstrass formalism is set up, j( ) may be rationally expressed in k2 .
Regardless of the formalism, the term singular modulus always characterizes
the cases with an imaginary quadratic ratio between the basic periods.
We will review in section 4 below the arguments about what Kronecker
actually conjectured concerning the explicit generation of all Abelian extensions of an imaginary quadratic number eld. For the time being, we continue
to discuss Hilberts presentation of his 12th problem.
A comparison between both cases that Hilbert chose as motivation brings
out very clearly the picture he had in mindand which he also attributed to
Kronecker:
If the ground eld is Q, there is the analytic function x eix
which has the property that, if we substitute elements x of the
9
. . . Gleichungen . . . , deren Wurzeln singul

are Moduln von elliptischen Functionen oder
elliptische Functionen selbst sind, deren Moduln singul
ar und deren Argumente in rationalem Verh
altnis zu den Perioden stehen. [Kronecker 1877, p. 70].
` 3
SEMINAIRES
ET CONGRES
249
given eld Q into it, the values eix generate all Abelian extensions
of Q.
If the ground eld K is imaginary quadratic, then there is the
analytic function j( ) which has the property that, if we
substitute elements of the given eld K into it, the values j( )
generate all Abelian extensions of K.
The rst statement is the Kronecker-Weber theorem. The second statement
is false. First of all, it is false for the trivial reason that roots of unity generate
Abelian extensions of K which cannot in general be obtained from singular
j-values. Since Hilberts prose is not very formal, and since roots of unity
were already brought into the game in the rst step, to generate the Abelian
extensions of Q, we may naturally correct the second statement to mean that
all Abelian extensions of K can be generated by roots of unity and singular
values j( ), K. This is how Hilberts claim was understood by those
who worked on the problem: Fueter, Weber, Hecke, Takagi, Hasse. But this
statement is still wrong, as we know today: one does need other functions, for
instance, suitable values (z, ), for K and rational z, to get all Abelian
extensions of K.
We will discuss Hilberts wrong conjecture and its inuence on the work
in the area in section 3 below. We will review the argument against Hilberts
historic claim (to the eect that Kronecker had had the same conjecture in
mind) in section 4. For now, let us just try to understand the beautifully
simple image that Hilbert is trying to convey to usnever mind that it is
mathematically incorrect and probably also not what Kronecker conjectured.
If what Hilbert claims were true, this would indicate a marvellous economy
of nature, which provided just one function for all imaginary quadratic elds
at once, giving all Abelian extensions by simply evaluating it at the elements
of the base eld in question.
Hilbert assumed that what he saw as Kroneckers conjecture would be
proved without much trouble by a slight renement of the already existing
elements of class eld theory.10 It is with this optimistic picture in mind that
he then formulated the general problem (cf. [Fueter 1905, p. 197]): Given
a eld K of nite degree over Q, to nd analytic functions whose values at
suitable algebraic numbers generate all Abelian extensions of K. Here Hilbert
had actually more up his sleeves than one can guess from the rather general
10
Der Beweis der Kroneckerschen Vermutung ist bisher noch nicht erbracht worden; doch
glaube ich, da derselbe auf Grund der von H. Weber entwickelten Theorie der komplexen
Multiplikation unter Hinzuziehung der von mir aufgestellten rein arithmetischen S
atze u
ber
Klassenk
orper ohne erhebliche Schwierigkeiten gelingen mu. [Hilbert 1901, p. 311f].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
250
N. SCHAPPACHER
discussion of the analogies between function theory and algebraic number

theory which he inserts into the text of the 12th problem. We will briey
discuss his research programme in section 5 below.
Even today, as we are approaching the centenary of Hilberts lecture, we
are still waiting to see these analytic functions and their special values in
general. Meanwhile, it seems clear that generalizing the theory of complex
multiplication is not going to do this job for us.
2.
The Theorem of Kronecker and Weber
In [Kronecker 1853, p. 10] we read:

. . . We obtain the remarkable result: that the root of every
Abelian equation with integer coecients can be represented as
a rational function of roots of unity. . . 11
Thus Kronecker seems to claim that he has established the theorem which
today goes by the name of Kronecker and Weber. But in fact, in 1853, his terminology of Abelian equations only referred to equations with cyclic Galois
group. This is of course the crucial case of the theorem, and the reduction
to it of the general case is indicated for instance in [Kronecker 1877, p. 69].
Another problem with the above quote is that in [Kronecker 1853, p. 8] he
indicates that he has not been able to deal with the case of cyclic extensions
of degree 2 , with at least 3.
Kroneckers contemporaries apparently did not think he had a valid proof
of the result. Hilbert for instance, in [Hilbert 1896, p. 53], distinguishes
between Kronecker who stated (aufgestellt) the theorem, and Weber who
gave a complete and general proof of it. I happily go along with Olaf
Neumann saying: Nowadays it is hard to estimate to what extent Kronecker
really could prove his theorem.12 Still, it is conceivable that new light might
be shed on this and other questions by a perusal of the handwritten notes of
Kroneckers Berlin courses of which a remarkably rich collection, from between
1872 and 1891, is one of the historical treasures of the library of the Strasbourg
Mathematical Institute.13
11
. . . ergiebt n
amlich das bemerkenswerthe . . . Resultat: da die Wurzel jeder Abel schen
Gleichung mit ganzzahligen Co
ezienten als rationale Function von Wurzeln der Einheit
dargestellt werden kann. . .
12
[Neumann 1981, p. 120]. Much of the present section owes to this careful article.
13
There are 27 bound volumes of handwritten notes. They belonged to Kurt Hensel.
After Hensels death, in the Summer of 1942, several hundred items of his personal mathematical library were sold by his daughter-in-law to the (Nazi) Reichs-Universit
at Straburg.
` 3
SEMINAIRES
ET CONGRES
251
Kronecker was very pleased with the theorem.14 He proudly emphasized

[Kronecker 1856, p. 37] the novelty that it does not reduce certain algebraic
numbers to others of smaller degree, but rather elucidates their nature by
linking them with cyclotomy.
It is astonishing how comparatively little attention Heinrich Weber (5 March
1842,15 17 May 1913) and his work have received so far among historians
of mathematics and among mathematicians.16 He is remembered for having
been the nineteenth century German mathematician who acccepted the greatest number of job oers from dierent universities. Thus he held positions
at Heidelberg, Z
urich, Konigsberg, Berlin, Marburg, and Gottingen (chair of
Gauss - Dirichlet - Riemann - Clebsch - Fuchs - Schwarz),... before he nally
moved from there to Strasbourg in 1895. David Hilbert was Webers successor
in G
ottingen; he had been Webers student back in Konigsberg, along with
Hermann Minkowski.
Weber moved from mathematical physics to algebra and number theory.
His achievements that are remembered include the following.
The fundamental paper [Dedekind and Weber 1882] where the notion
of point on an abstract algebraic curve is dened for the rst time
in history, thus taking a decisive step towards the creation of modern
algebraic geometry. Looking up H. Weber in the index of [Bourbaki
1984] leads one only to numerous allusions to this one article.
His Lehrbuch der Algebra in three volumes: [Weber 1894, 1896, 1908].
Suce it to say here that this work marks the transition from the late
19th century treatment of algebra17 to the modern algebra whose rst
full-edged textbook treatment was going to be van der Waerdens wellknown treatise of 193031.18 The third volume [Weber 1908] would not
M. Kneser kindly found out the correspondence between Hasse and Marieluise Hensel concerning this transaction in NSUG, Nachla Hasse, 24, p. 3.
14
See for instance [Kronecker 1877, p. 69], where he adds the comment: Dieser Satz giebt,
wie mir scheint, einen werthvollen Einblick in die Theorie der algebraischen Zahlen; denn er
enth
alt einen ersten Fortschritt in Beziehung auf die naturgem
asse Classication derselben,
welcher u
uhrt.
ber die bisher allein beachtete Zusammenfassung in Gattungen hinausf
15
In [Voss 1914] the 5th of May is given as the day of birth. This mistake is repeated
quite often in the literature.
16
Published exceptions are [Frei 1989, 1995], cf. also [Katsuya 1995]. For Webers administrative role in Straburg, see [Manegold 1970, p. 195] and [Craig 1984, pp. 141-145]. Cf.
[Wollmersh
auser 1981]. See also the preprint [Schappacher and Volkert 1998].
17
As represented for instance by the famous book by Camille Jordan, Trait
e des substi
tutions et des
equations alg
ebriques, recently re-edited by Editions
Jacques Gabay, Paris,
1989. Webers Algebra resembles Jordans treatise in many respects.
18
Recently re-edited as Algebra, Springer, 1993.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
252
N. SCHAPPACHER
be called algebra today. It is in fact the second, thoroughly reworked

edition of [Weber 1891], and contains a classical treatment of elliptic
functions, especially their arithmetic theory, along with parts of algebraic number theory and class eld theory, as well as a small chapter on
dierentials of curves in the higher rank case including Riemann-Roch.
Generalizing slightly from a lecture of Dedekinds of 1856/57, Weber
was the rst to dene our abstract notion of group in print: [Weber
1893]. This made it into the Lehrbuch der Algebra, see the beginning
of [Weber 1896]. See also [Franci 1992, p. 263] for a few details and
relevant references.
Weber had a leading role in the edition of Riemanns Collected Papers which is particularly remarkable for making important parts of
Riemanns Nachla available as well.
Weber developed a notion of class eld in [Weber 1897-98]; see also [Weber 1908, p. 164]. Cf. [Frei 1989], [Katsuya 1995, 1.3]. He emphasized
the decomposition behaviour, as opposed to Hilberts chief interest in
the unramiedness of the (Hilbert) class eld. More precisely, we read
in [Weber 1908, p. 164]: Denition of the class eld. The prime ideals
pi of degree one in the principal class A1 , and only these, are to split in
the eld K(A) again into factors of degree 1.19 This denition enables
the argument (which follows our quote) that was to remain the essence
of the analytic part of class eld theory for almost half a century: the
deduction of the inequality n h from the analysis near s = 1 of
partial zeta-functions of the ground eld and the class eld.20
Webers numerous contributions to elementary mathematics (partly in joint
work with Wellstein) are all but forgotten, and so are many of his widespread
interests, which are however well reected in the Festschrift for his 70th birthday.21 Klein portrayed Weber as a particularly exible mind.22
19
Denition des Klassenk
orpers. Die Primideale pi ersten Grades der Hauptklasse A1 ,
und nur diese, sollen im K
orper K(A) wieder in Primideale ersten Grades zerfallen.
20
The terminology of ray classes etc., if not the corresponding concepts, seem to be
due to Fueter; see [Fueter 1903, 1905]. Fueter appears to give insucient credit to [Weber
1897-98]. Fueters works are not mentioned in [Frei 1989].
21
Festschrift Heinrich Weber zu seinem siebzigsten Geburtstag am 5. M
arz 1912 gewidmet
von Feunden und Sch
ulern, mit dem Bildnis von H. Weber in Heliograv
ure und Figuren im
Text, Leipzig und Berlin: Teubner, 1912.
22
H. Weber ist 1842 in Heidelberg geboren, wo er auch seine Studien beginnt und bei
Helmholtz und Kirchho h
ort. Von 187383 wirkt er in K
onigsberg, 189295 ist er Ordinar-
` 3
SEMINAIRES
ET CONGRES
253
Given this somewhat eclectic appreciation of Webers achievements today

it is maybe not surprising that, in spite of some similar criticism by Frobenius
of Webers proof of the Kronecker-Weber theorem in [Weber 1909],23 it seems
to have gone unnoticed until 1979 that the proofs of the Kronecker-Weber
theorem proposed in [Weber 1886, 1896], and [Weber 1908] were also not
valid, due to a basic miscalculation of the Galois action on certain complicated
Lagrange resolvents at the very beginning of the argument.24 For the details
we refer to the concluding comments in [Neumann 1981, pp. 124125]. So
it was in fact Hilbert himself who gave the rst valid proof of the result, in
[Hilbert 1896]. Weber published his rst correct proof at age 69, two years
before his death, in [Weber 1911]. As Olaf Neumann suggests, it would be
tting to refer to the result as the theorem of Kronecker-Weber-Hilbert.
One may speculate [Neumann 1981, p. 124] that Weber was in fact misled
by Kroneckers composition of Abelian equations. If so, this would provide a
beginning of an explanation of this error within the historical context. Such
an explanation seems desirable because otherwise it is all too uncanny to see
the author of the Lehrbuch der Algebra deceiving himself at an essential place
about the Galois action in the composite of two normal extensions.
Today it is common to deduce the theorem from the existence theorem of
class eld theory. But there are also a number of direct proofs in the litera
ture: [Speiser 1919], [Cebotarev

1924], [Safarevi
c 1951], [Zassenhaus 1968-69],
[Greenberg 1974-75] and [Washington 1982, chap. 14].
ius in G
ottingen; dann geht er nach Straburg, wo er 1913 stirbt. Er ist eine schmiegsame
und doch wieder energische Natur und besitzt eine wunderbare F
ahigkeit, leicht in ihm
zun
achst fremde Auassungen einzudringen, so z.B. in die Riemannsche Funktionentheorie
und die Dedekindsche Zahlentheorie. Diese seine Anpassungsf
ahigkeit hat es ihm erm
oglicht,
auf fast allen Gebieten unserer Wissenschaft in den letzten Dezennien mitzuarbeiten und
die umfassenden Lehrb
ucher, den Weber-Wellstein, den Riemann-Weber, die Algebra zu
schaen, die wir alle kennen und benutzt haben. Seiner Mitwirkung an der Herausgabe von
Riemanns Werken 1876 wurde bereits gedacht; die zweite Auage 1892 hat Weber allein
besorgt. [Klein 1926, p. 275].
23
See the excerpt [Frobenius 1911] from the letter of Frobenius to Weber, 19 June 1909,
in NSUG 8 Cod. Ms. philos. 205, which corrects some aws in Webers preceding proof
[Weber 1909], and suggests the simpler arguments for the following paper [Weber 1911]. Cf.
the surrounding letters by Frobenius in NSUG, loc. cit.
24
See for instance [Weber 1896, p. 209, formula (7)]. This formula is incorrect as soon as
the radicals and the roots of unity entering into the resolvent form extensions of Q which
are not linearly disjoint. Personally, I hit upon this problem when I proposed to Mlle A.
Rauch a m
emoire de matrise with a view to rewriting Webers proof in modern notation.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
254
3.
N. SCHAPPACHER
Work on Hilberts claim for imaginary quadratic

elds
Around the turn of the century a number of Hilberts students were involved
in a research programme one of the centres of which was Hilberts 12th problem. For the more arithmetic development of class eld theory, one has to
mention in particular Ph. Furtwangler and F. Bernsteinsee the 1903 volume
of the G
ottinger Nachrichten. On what was then seen as the function theoretic side of the problem, there was O. Blumenthal, and later E. Heckesee
section 5 below. But it was the Swiss mathematician Rudolf Fueter who attacked the 12th problem head on in [Fueter 1903, 1905], adopting the following
philosophy which, one may assume, was inspired by Hilbert.
Suppose that, for a given number eld K say, Galois over Q, as Fueter
always assumes , analytic functions have been constructed certain singular
values of which generate a lot of Abelian extensions of K. We would then
like to have a general class eld theoretic method to prove that these values
suce to generate all Abelian extensions of K. The method proposed by
Fueter comes down to the observation that we are done if we can show that
all ray class elds are contained in what the special values give us. Indeed,
it would follow from the Hauptsatz of chapter IV [Fueter 1905, p. 232] that
every Abelian extension of K is contained in a suitable ray class eld. The
execution of this strategy in [Fueter 1905] is, however, invalidated in the case
of Abelian extensions of even degree by a group theoretical mistake in the
reduction steps of the rst chapter [Fueter 1905, p. 207].25
Still, Fueters strategy could have very well led to a timely destruction of
Hilberts overly optimistic claim. For the convenience of the reader, let us
explain this in the classical ideal theoretic language of class eld theory, say,
like in [Hasse 1926a]. A comparison with [Fueter 1905], and in particular with
[Weber 1908] shows that such a refutation of Hilberts claim would have been
well within the reach of these authors at the beginning of the century.26
Let K be an imaginary quadratic number eld, and oK its ring of integers.
The values j( ), K H, are precisely the j-invariants of lattices a C
such that the ring of multipliers of the lattice, oa : = { C | a a}, is
an order in K, i.e., is of the form oa = of = Z + f oK , for some integer
f 1. Now, given such an order of , the extension Kf = K(j(a)) does not
depend on the lattice a such that oa = of . In fact, all of these values j(a)
are conjugate over K, and their number equals the class number of proper
See [Fueter 1914, p. 177f, note ].
A modern, extremely concise justication of the claims which we will use can be obtained
from [Serre 1967].
25
26
` 3
SEMINAIRES
ET CONGRES
255
of -ideals. The eld thus obtained is an Abelian extension of K which Weber

called Ordnungsk
orper (for the conductor f , which Weber calls Q), and which
he recognized as the class eld associated with the group of ideals prime to
f , modulo principal ideals generated by elements K satisfying
r (modf ),
gcd(, f ) = 1
for some rational number r depending on see [Weber 1908, 124]. Today
this eld is called the ring class eld of K modulo f , a terminology going back
to Hilbert.
Since roots of unity generate the ray class elds of Q, the Abelian extension
of K generated by Kf and by the f -th roots of unity corresponds to the group
of principal ideals generated by elements
r (modf ),
r 2 1 (modf ),
gcd(, f ) = 1
for some rational number r depending on . These conditions do not in

general imply that 1 (modf ). But it is this latter condition that
describes the ray class eld of conductor f of K, because K being totally
imaginary there is no real place to distinguish between the two units 1.27
The essential gap between the two conditions is that one may have dierent
signs at dierent prime divisors of f . Thus, if we call K the union of the elds
Kf , for all f , and K the union of all ray class elds of K, then Gal(K /K )
is an innite product of groups of order 2. Therefore, even independently of
the existence theorem of class eld theory, which says that K = K ab , the
eld K proposed by Hilbert in his 12th problem is not big enough to contain
all Abelian extensions of K.
On the 4th of July, 1903, Heinrich Weber wrote to his former student
and friend David Hilbert to tell him that now, after the end of the teaching
term, he felt free to embark again on some serious work, and asked him for
information about works of Hilberts students on Complex Multiplication. He
explained that he had been out of touch with this theory for a while and had
to start by learning the new developments. He mentioned that he had just
received Fueters thesis [Fueter 1903] which looks very promising, judging
from its title and the table of contents.28
27
As Takagi points out nicely in [Takagi 1920, p. 103], the ray class elds of K are
analogous to the maximal totally real subelds of the cyclotomic elds. He had himself
overlooked this point in his work on extensions of Q(i), see [Takagi 1903, p. 28]; cf. footnote
34 below.
28
NSUG, 8 Cod. Ms. philos. 205, sheets 3940. Unfortunately the letters from Hilbert
to Weber do not seem to have survived. . .
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
256
N. SCHAPPACHER
What he did not mention in this letter was the work of his own student
Daniel Bauer at Strasbourg who submitted his dissertation [Bauer 1903] that
same year. There Bauer studies the following conjecture which Weber had
made in a vague formin agreement with Hilberts conjecture, although Weber probably wrote this down before Hilberts lecture at the Paris ICM in
his encyclopedia article [Weber 1900, end of 11, p. 731]. Let a C be as
quadratic
eld K.
above a lattice such that oa is an order of the imaginary
(Bauers thesis excludes the cases where K = Q( 3), Q( 4), i.e., where
oK has extra units besides 1). Let m be any oK -ideal prime to the conductor of oa . Dene the m-th Teilungsk
orper Tm to be the extension of K(j(a))
generated by the m-division points of Webers -function associated to the
lattice a. In the cases without extra units (the only ones that Bauer considers), this is just a weight zero variant of the Weierstrass -function: up to a
a)g3 (a)
(z; a). Today we may say that Tm is the
rational factor, (z) equals g2 ((
a)
eld generated over K(j(a)) by the x-coordinates of the points annihilated by
all elements of m, on a model dened over K(j(a)) of the elliptic curve C/a.
Tm is certainly Abelian over K(j(a)). Weber suggests [loc. cit.] that these
Teilungsk
orper are always contained in suitable composites of ring class elds
of K and cyclotomic elds.
Bauer purports to prove that, if m = p oK , for an odd prime number p,
then the eld generated over K by Kp and the p-th roots of unity coincides
with Tm [Bauer 1903, p. 4 and p. 32f]. This cannot be quite right in the
case where p splits into the product of two prime ideals in oK , because then
we may choose, in the class eld theoretic analysis of the elds in question,
dierent signs at the prime divisors of p. I have not traced down Bauers
arguments. They are coached in terms of Jacobis elliptic function sn rather
than Webers .
In the third volume of his Lehrbuch der Algebra, Weber [1908] discusses
elds called Teilungsk
orper at various places, rst in 154. There he considers
the elds Tm dened above, under the additional assumption that oa = oK ,
so that K(j(a)) is the Hilbert class eld K1 of K. Taking division values
of the -function, rather than the eld generated by both coordinates of the
m-torsion points of an elliptic curve isomorphic to C/oK dened over K1 ,
can be seen today to be the geometric analogue of the fact that we cannot
distinguish between 1 in the ray condition. Note in passing that adjoining all
the coordinates of torsion points does not in general give Abelian extensions
of K.29
29
This is related to a condition introduced by Shimura into the theory of Abelian varieties
with complex multiplication. For the case of elliptic curves, see for instance [Schappacher
1982].
` 3
SEMINAIRES
ET CONGRES
257
Hasse in his particularly tidy work [Hasse 1927] showed how to construct
the ray class elds of K directly from these Teilungsk
orper Tm . Weber however, for technical reasons, was led, in the third part of [Weber 1897-98]
as well as in [Weber 1908], to work with more complicated elds, replacing
the -function by certain quotients of theta series. These elds he still calls
Teilungsk
orper , and denotes them by the same symbol Tm [Weber 1908, 158,
end]. As Hasse points out in [Hasse 1926a, p. 55], Weber even gets caught up
in a confusion between the two sorts of elds in [Weber 1908, 167, (5)]. Let
us gloss over this additional problem here. Then Weber nally derives for his
Teilungsk
orper Tm in [Weber 1908, 167] a class eld theoretic description
which in our language pins them down as the ray class elds of K, modulo
given ideals m of oK .30
Then he sets out in [Weber 1908, 169] to show that the ray class elds
can be indeed generated over K by singular moduli and roots of unity. If m
is an ideal of oK dividing the rational integer f , Weber wants to conclude
the congruence 1 (mod m) from the conditions r (modf ), r 2
1 (modf ). Now, this is alright if m is the power of a prime ideal of oK not
dividing 2. But Weber thinks he can always reduce to this case without loss of
generality. In fact, at the end of [Weber 1908, 158], he had claimed that any
orper Tn with n equal
Teilungsk
orper Tm was the composite of various Teilungsk
to powers of prime ideals. This were true if he had adjoined all the coordinates
of torsion points, not just division values of particular functions. Translating
back to the characterization by ray class groups, Weber overlooked precisely
the possibility of choosing dierent signs in 1 modulo dierent prime factors
of m.
This is how Weber missed his chance to disprove Hilberts claim in the
third volume of his Lehrbuch der Algebra [Weber 1908, 169].31
As late as 1912 Erich Hecke, another thesis student of Hilberts, assures us
in the preface to his thesis [Hecke 1912] that Fueter has proved Hilberts claim
in [Fueter 1905, 1907]. He is careful to add, however, a footnote to the eect
that Fueter will ll a few gaps in his proof in a book soon to be published. As a
matter of fact, this book was to appear only in 1924, more than 20 years after
Fueter had begun working on the problem under Hilberts guidance (and then
it was promptly mauled by Hasse in his merciless review [Hasse 1926b]. . . ).
Ten years before the book, one year after Heinrich Webers death, the general
agreement on Hilberts claim had nally come to an end in [Fueter 1914].
30
For details see [Hasse 1926a, p. 43f]. Even though this is not at all recalled in the later
sections of Viertes Buch of [Weber 1908], it seems that Weber actually restricts attention
to ideals m prime to 2 all along.
31
One more incorrectness in this part of [Weber 1908] is mentioned in [Hasse 1926a, p. 55].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
258
N. SCHAPPACHER
This long article shows a rather hapless Fueter. He

now has a counterexample to Hilberts claim: for K = Q(i), the eld K( 4 (1 + 2i)) cannot be
generated by singular moduli and roots of unity. He has also understood the
group theory mistake he had made in [Fueter 1903]. Furthermore, he guesses
what the correct picture is going to be: the Teilungsk
orper will do the job,
and in general they are strictly bigger than the elds considered by Hilbert.
He formulates this as the Hauptsatz [Fueter 1914, p. 253] and claims it explicitly (Dagegen gilt der Hauptsatz. . . ). Then he talks about what one has to
do to prove this. His problem is precisely the one that Hasse solved in [Hasse
1927]: to work with Webers original denition of the Teilungskorper and see
its relation to the ray class elds. Since he does not know how to do this,
he explains that the investigation necessitates a discussion of the function
theoretic side of the problem. I have not yet executed these considerations,
and they would have actually led too far astray. I will cover this problem in
its full context in a Teubner textbook. But I do believe that I have made
sucient progress on the number theoretic side.32
It was Teiji Takagi who got there rst. In the nal chapter V of his
momentous paper [Takagi 1920] which he wrote up when the end of the
War and the upcoming rst postwar ICM (Strasbourg 1920) promised the
renewal of contact with European colleagues [Iyanaga 1990, p. 360f] the
author does what Weber should have done in the third volume of his Lehrbuch
der Algebra. In fact, Takagi follows Weber as closely as he can, working with
the modied, more complicated Teilungsk
orper , but getting things right. To
be sure, the crucial thing that Weber could not have done easily 15 years
before Takagi is the proof of the fact that every Abelian extension of K is
contained in a suitable ray class eld. Takagi, in [Takagi 1920, p. 90, Satz
28], deduces this in complete generality as the key result of his tremendous
development of general class eld theory, which occupies the bulk of the article
[Takagi 1920] and which in turn was made possible also by prior work of the
Hilbert school, in particular Ph. Furtwangler. Cf. [Katsuya 1995, 3].
Believing his own account [Iyanaga 1990, p. 360], one concludes that Takagi had started his own serious investigations on class elds in 1914 when
World War I began . . . because he could not expect the ow of academic
32
Ist dagegen die K
orperklassenzahl von 1 verschieden, so verlangt die Untersuchung ein
Eingehen auf die funktionentheoretische Seite des Problems. Diese Betrachtungen habe ich
noch nicht durchgef
uhrt, sie w
urden auch zu weit abseits f
uhren. Ich werde dieses Problem
in einem Teubnerschen Lehrbuche im Zusammenhange darstellen. Doch glaube ich, da
die zahlentheoretische Seite durch meine Entwicklungen ausreichend gef
ordert ist. [Fueter
1914, p. 255].
` 3
SEMINAIRES
ET CONGRES
259
books and journals from Germany anymore. [Katsuya 1995, p. 116] But at
least in some ways Takagis ne article of 1920 was the culmination of almost
20 years of work and calls for a ashback. In fact, Takagi had been, so to say,
a member of the club all alongyet remained an outsider at the same time.
He had come to Germany in 1898 to study, rst with Frobenius in Berlin, and
as of Spring 1900 with Hilbert in Gottingen. It was Hilbert who supervised
his thesis [Takagi 1903] which Takagi nished writing in the Spring of 1901
and submitted to the Imperial University of Tokyo.
Even if Takagis anecdotal account diminishes Hilberts direct guidance
of the thesis [Iyanaga 1990, p. 357], the inuence of the master is evident
throughout the thesis: The short introduction, which the author (humbly?)
calls almost superuous33, uses close reformulations of sentences from
Hilberts text on the twelfth problem. In particular, Takagi also states
Kroneckers conjecture quoting the ambiguous transformation equations of
the elliptic functions with singular moduli. He does not elaborate at all
on the meaning of this. What he does in his dissertation is actually quite
dierent in spirit from Hilberts version of Kroneckers conjecture, although
inspired by another work of Hilberts in the area:
Fixing the base eld K = Q(i), Takagi shows that all Abelian extensions
of K are contained in the extensions of K generated by division values of
the lemniscatic elliptic function, i.e., essentially of the Weierstrass -function
associated to the elliptic curve y 2 = x3 x. The method is to transfer Hilberts
proof of the Kronecker-Weber theorem [Hilbert 1896] to the lemniscatic case.34
So from his very rst exposure to the problem Takagi was oriented towards division elds rather than general ring class elds. This orientation
can be clearly traced through his subsequent publications on complex multiplication.35 His decisive contribution [Takagi 1920] is therefore also the fruit
33
Diese fast u
ussigen Einleitungsworte schliesse ich mit dem Ausdruck herzlichsten
ber
Dankes an den Herrn Prof. Hilbert in G
ottingen, dessen Anregung diese Erstlingsarbeit ihr
Entstehen verdankt [Takagi 1903, p. 13]. This sentence seems to contradict the abovementioned anecdote according to which Takagi simply told Hilbert what he was working on
and Hilbert accepted. . . It is presumably because he did not get his doctorate in G
ottingen
that Takagi is missing from the Verzeichnis der bei Hilbert angefertigten Dissertationen
in the third volume of Hilberts Gesammelte Abhandlungen, 1970, pp. 431433.
34
Takagi himself points out in [Takagi 1920, p. 145, footnote 3] a mistake in [Takagi 1903,
p. 28]. Cf. our footnote 27 above. Another mistake, concerning [Takagi 1903, p. 29, H
ulfssatz
1], is noted and briey discussed by Iwasawa in [1990, p. 343, footnote 2]. Note that the
lemniscatic analogue of the Kronecker-Weber theorem is already claimed, at least vaguely,
in [Kronecker 1853, p. 11]. The article [Masahito 1994] (which is not always easy to follow,
but certainly insists on the importance of the lemniscatic case for the prehistory of complex
multiplication in the 19th century) does not mention Takagis thesis.
35
See Nos 7, 9, and 10 of Teiji Takagi, [Papers, pp. 342351].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
260
N. SCHAPPACHER
of a line of thought independent of the main intention of Hilberts twelfth

problem, yet still suggested by Hilbert, in the very special and concrete case
of lemniscatomy.
4.
Kroneckers Jugendtraum
Kroneckers letter to Dedekind dated 15 March 1880 begins:

Thank you very much for your kind lines of the 12th. I believe
they are to give me a welcome occasion to let you know that I
believe to have overcome today the last of many diculties that
were still withstanding the completion of an investigation which
I had taken up again more intensely in the last few months. It
concerns the dearest dream of my youth, to wit, the proof that the
Abelian equations with square roots of rational numbers are exhausted by the transformation equations of elliptic functions with
singular moduli exactly in the same way as the rational integral
Abelian equations by the cyclotomic equations.36
In section 1 above we have discussed the possible ambiguity of these transformation equations of elliptic functions with singular moduli. We quoted
a passage from [Kronecker 1877, p. 70] (footnote 9 above), where Kronecker
mentions in a row equations the roots of which are singular modules of elliptic functions or elliptic functions themselves the modules of which are singular
and the arguments of which have a rational ratio with the periods. In that
same passage Kronecker goes on to conjecture that all equations Abelian over
quadratic elds are exhausted by those which come from the theory of elliptic
functions.
Mentioning both kinds of functions and special values at the same time
makes good sense for many reasons. Helmut Hasse, in his painstaking discussion of what Kroneckers Jugendtraum really consisted in, noted that the
orientation of Kroneckers research in this area actually moved from singular
36
Meinen besten Dank f
ur Ihre freundlichen Zeilen vom 12.c.! Ich glaube darin einen
willkommenen Anlass nden zu sollen, Ihnen mitzutheilen, dass ich heute die letzte von vielen Schwierigkeiten besiegt zu haben glaube, die dem Abschlusse einer Untersuchung, mit der
ich mich in den letzten Monaten wieder eingehender besch
aftigt habe, noch entgegenstanden.
Es handelt sich um meinen liebsten Jugendtraum, n
amlich um den Nachweis, dass die
Abel schen Gleichungen mit Quadratwurzeln rationaler Zahlen durch die TransformationsGleichungen elliptischer Functionen mit singul
aren Moduln grade so ersch
opft werden, wie
die ganzzahligen Abel schen Gleichungen durch die Kreistheilungsgleichungen. [Kronecker
1880, p. 453].
` 3
SEMINAIRES
ET CONGRES
261
moduli to division values [Hasse 1930, p. 514] which is another major argument to show that Hilberts interpretation of the Jugendtraum was not
that intended by Kronecker.
A mathematical reason for coupling both kinds of functions, which is very
close to the way we view things today, is that division values make (geometric)
sense only over a eld of denition of the corresponding (geometric) object,
which in the case at hand is the eld generated by the corresponding singular
modulus. It seems hard to decide how much of this geometric perspective
may have been present already in Kronecker or Weber.37 It yields an understanding of the analogy between the Kronecker-Weber theorem and the
Jugendtraum which is completely dierent from Hilberts point of view in his
12th problem. See section 6 below.
Hasse [1930] wrote his thorough philological analysis as a kind of penitence. For he had never cared before to check Hilberts historical claim
(repeated in particular by Fueter, see for instance [Fueter 1905]) that Kroneckers Jugendtraum was precisely what Hilbert expected: the generation
of all Abelian extensions of an imaginary-quadratic eld by singular moduli
and roots of unitythis is what is called interpretation (a) of the Jugendtraum in [Hasse 1930]. Thus in [Hasse 1926a, p. 41], he had still written that
Kroneckers conjecture . . . turns out to be only partially correct. Now, in
[Hasse 1930, p. 515], he went so far as to conclude that if Kronecker had any
precise formulation of his Jugendtraum-theorem in mind at all, then it can
only be what is called interpretation (b) in [Hasse 1930], i.e., the generation
of all Abelian extensions of an imaginary-quadratic eld by singular moduli
and division values.
I nd little to add to Hasses study of this historical issue, if one accepts
the question the way he poses it. In particular, Hasse shows convincingly by
quoting from other places in Kronecker why the term transformation equations appearing in the Jugendtraum quote in [Kronecker 1880] introduces an
ambiguity of meaning, and he argues carefully to show that Kronecker was
indeed envisaging to use both kinds of algebraic quantities to generate all
Abelian extensions of imaginary-quadratic elds: singular moduli as well as
division values of corresponding elliptic functions.
On the other hand, it seems only fair to say that a casual reading of [Kronecker 1880], especially from the middle of page 456 on where Kronecker mentions only singular moduli explicitly, can easily create the impression that
Kronecker did want to do without the division values, which would amount
to Hilberts claim. Adding to this Hilberts optimistic conviction that this
37
For the same reason we do not think that Vladuts remark [1991, p. 79, last paragraph]
concerning interpretation (c), of the Jugendtraum in [Hasse 1930] is historically sensible.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
262
N. SCHAPPACHER
claim was correct, and t into a beautiful general picture, Hilberts double
errormathematical and historicalreduces to a minor slip. What we have
shown is how long this double slip could survive, carried as it were by Hilberts
tremendous authority.
But when we look at this story, we have to be careful not to forget how
dierently we are programmed today in these matters: For us, moduli tend
to be points on a moduli scheme and thus represent algebro-geometric objects
as such, whereas division values suggest Galois representations, which will be
Abelian in the presence of complex multiplicationsee section 6 below. Such
a conceptual separation of the two kinds of singular values that Kronecker
brought into play did not exist at the turn of the century. For instance, the
chapter Multiplication und Theilung der elliptischen Funktionen in [Weber 1891] culminates in a 68 about Reduction of the division equation to
transformation equations. And Kronecker himself once stated this continuity
very forcefully that he saw between the two notions in the case of complex
multiplication.38
5.
Hilbert Modular Forms
In the introduction to Otto Blumenthals Habilitationsschrift [Blumenthal

1903b] (submitted at Gottingen in 1901) we read: In the years 189394
Herr Hilbert investigated a way to generalize modular functions to several
independent variables. . . . Herr Hilbert has most kindly given me these notes
for elaboration.39 I do not know whether Hilberts original notes on what
was to become the theory of Hilbert Modular Forms still exist.
Blumenthal was the rst student to whom Hilbert gave an aspect of this
research programme. He was to develop the analytic theory, relative to an
arbitrary totally real eldsee [Blumenthal 1903b,a, 1904b,a,c]. Today it
is part of the folklore of this subject40 that Blumenthals works contain in
particular the mistake that he thinks he needs only one cusp to compactify
the fundamental domain for the full Hilbert modular group, whereas h are
38
W
ahrend f
ur die Kreisfunctionen nur Multiplication, f
ur die allgemeinen elliptischen
Functionen aber Multiplication und Transformation stattndet, verliert die Transformation
bei jener besonderen Gattung elliptischer Functionen [sc. f
ur welche complexe Multiplication stattndet] zum Theil ihren eigenth
umlichen Charakter und wird selbst eine Art
von Multiplication, indem sie gewissermaaen die Multiplication mit idealen Zahlen
darstellt. . . [Kronecker 1857, p. 181].
39
In den Jahren 189394 besch
aftigte sich Herr Hilbert mit einer Verallgemeinerung der
Modulfunktionen auf mehrere unabh
angige Variable. . . . . . . Herr Hilbert hat mir diese
Notizen zur Ausarbeitung freundlichst u
berlassen.
40
Cf. Schoenebergs notes to [Hecke 1912] in his edition of Heckes Mathematische Werke.
` 3
SEMINAIRES
ET CONGRES
263
needed (h the class number of the eld in question). This error was passed on
to the second student that Hilbert sent into this eld, Erich Hecke. He was to
explore the application of Hilbert modular forms to the 12th problem in the
case of a real quadratic eld in his thesis [Hecke 1912]. Exploiting a relation
with theta functions which was found by Hilbert,41 Hecke has at his disposal
a Hilbert modular function analogous to the j-function of the elliptic case
(but not holomorphic in the fundamental domain), and he wants to generate
interesting Abelian extensions of a totally imaginary quadratic extension of
the given real quadratic eld by suitable special (singular) values of this
Hilbert modular function. He does obtain a statement in this direction in
his dissertation [Hecke 1912, p. 57], but the result is far from satisfactory, as
Hecke is the rst to point out.
In his Habilitationsschrift [Hecke 1913], he then tries to go further by
taking a Hilbert modular function which is regular everywhere in the fundamental domain. Since such a function has to be constant, this work is strictly
speaking empty. To get some impression of what Hecke does manage to understand in spite of his impossible function, one may take a modern point
of view, and say that he is developing part of the theory of Abelian surfaces
with complex multiplication. In this language, one of the surprising features
of the theory that Hecke discovers is the fact that CM-eld and reex eld
are in general dierentsee for instance [Hecke 1913, p. 70].
It is with reference to this that Andre Weil speaks of Heckes audace
stup
eante to tackle a theory for which the time was clearly not yet ripe [Weil
uvres II, art. 1955 c, d]. This critical compliment should be transferred
at least partly to Hilbert who had become convinced, with his tremendous
mathematical optimism, of the sweeping perspective which he wrote into his
12th problem.
6.
Outlook on later developments, and another

historical tradition
The focus of this paper was on the comedy of errors which arose from
Hilberts formulation of Kroneckers Jugendraum. This story may leave the
41
Die interessanteste Analogie mit den Modulfunktionen aber bezieht sich auf den
Zusammenhang der neuen Funktionen mit dem Transformationsproblem der -Funktionen
mehrerer Ver
anderlicher. Herr Hilbert zeigt hier, da seine Funktionen bei diesem Problem eine ganz
ahnliche Rolle spielen, wie die Modulfunktionen in Bezug auf die elliptischen
Funktionen. Er leitet insbesondere eine Formel ab, aus der sich schlieen l
at, da man
zu Funktionen des Fundamentalbereichs gelangen kann, indem man Quotienten von ThetaNullwerten bildet. [Blumenthal 1903b, p. 510]; see also [Blumenthal 1904b].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
264
N. SCHAPPACHER
somewhat stale aftertaste of being a series of unnecessary mistakes bearing no

serious relation with the mathematical substance involved. Considering more
recent developments around Hilberts 12th problem reveals quite a dierent
aspect. Roughly from the end of the twenties or the beginning thirties on, the
point of view of Arithmetic Algebraic Geometry began to set in and dominate
more and more the domain of complex multiplication.
Arithmetic Algebraic Geometry was explicitly initiated by Poincare in his
seminal research programme [Poincare 1901] on the arithmetic of algebraic
curves. Still, its connection with the theory of complex multiplication had to
wait for about half a century, until several background theories had reached
the necessary maturity. In particular, the reduction of elliptic curves modulo
primes, the L-function of a curve over a nite eld (Kongruenzfunktionenk
orper , in the German school), the global L-function of a curve over a number
eld, . . . all these notions that began to crystallize in the twenties and thirties, nally come together in the beginning fties to shape what is still today
our basic understanding of the arithmetic theory of Complex Multiplication.
So talking about Hilberts 12th problem from this point of view is similar
to Bourbakis approach to history in his El
ements dhistoire des mathematiques: we place ourselves in todays mathematical context and try to recognize what we know, in documents which cannot be said to really possess this
knowledge. Thus the Kronecker-Weber theorem, looked at from the point of
view of arithmetic algebraic geometry, provides an example of the generation
of Abelian extensions of a eld of denition K from one-dimensional !-adic
representations of some group variety dened over K (or, more generally, of
a motive of rank 1). More precisely, the Abelian extensions of Q are generated by the torsion points of the multiplicative group Gm over Q. Similarly,
departing from Hilberts narrow (and probably incorrect) interpretation of
Kroneckers Jugendtraum, the coordinates of the torsion points of an elliptic
curve with complex multiplication by K, which is dened over the Hilbert
class eld K1 of K, do suce to generate (over K1 ) all Abelian extensions of
K.
In this perspective, the plethora of singular j-values which Hilbert proposed are really uncalled for. They have no analogue at all in the KroneckerWeber theorem because Gm is already dened over Q, and they should, seen
from this new vantage point, enter into the theory only as generators of elds
of denition for the given objects of arithmetic algebraic geometry, i.e., for a
given elliptic curve with complex multiplication.
This analysis motivates the generalization of both classical results: the
Kronecker-Weber Theorem and CM elliptic curves, in the arithmetic theory
of CM Abelian varieties of any dimension. And it is in this interpretation that
` 3
SEMINAIRES
ET CONGRES
265
Heckes dissertation and Habilitationsschrift do appear as a rst step in this

direction, i.e., as an attempt at a theory of Abelian surfaces with complex
multiplication.
The thirties and forties were characterized by a mutual fertilization of
the theory of complex multiplication with other developments in the domain
of arithmetic geometry (Hasse and his school, the Weil conjectures, Hecke
characters). These developments all express the general tendency to place
individual geometric objects, and the study of their arithmetic properties, at
the centre of the theories. Deurings theory of the L-function of a CM elliptic
curve [Deuring 1953-57] is one of the most visible consequences of this trend
within the traditional domain of the one-dimensional theory.
The higher dimensional theory was developed very quickly in the early
fties by Shimura, Taniyama, and Weil see [Shimura and Taniyama 1961].
It turned out to be quite a bit more complicated than the one-dimensional
case. New features include the distinction between CM-eld and reex eld,
the non reducibility of the class equation in general, the problem (solved
explicitly only in 1980, by Tate and Deligne) of describing the action of all of
Aut(C) on a CM Abelian variety, etc. And what is more relevant to Hilberts
12th problem: in higher dimensions the theory systematically fails to provide
enough elements to generate all Abelian extensions of the reex eld.
This point of view is modern in that the objects dealt withelliptic curves,
group varietiesdo not show up as such in the arithmetic investigations of the
19th and early 20th century related to our subject. There one mainly talked
about special values of modular or elliptic functions. But the modern point of
view also has its own historical roots. In fact, there is a strong tradition which
goes back to Gausss remark at the end of the introduction of Chapter VII
on cyclotomy of his Disquisitiones Arithmeticae,42 where he suggests that it
is possible to complement the cyclotomic theory which he is about to develop
in the book by an analogous lemniscatic theory. This clue was taken up in
particular by Eisenstein in [1850], and from there it entered into Kroneckers
seminal papers of the 1880ies, and further into Webers work.43 A relatively
modern version, but presented with a view to simplifying certain formulas in
42
Ceterum principia theori, quam exponere aggredimur, multo latius patent, quam hic
extenduntur. Namque non solum ad functiones circulares, sed pari successu
multas alias
R ad
dx
functiones transcendentes applicari possunt, e.g. ad eas qu ab integrali (1x
4 ) pendent,
prtereaque etiam ad varia congruentiarum genera. . . ; see [Schappacher 1997].
43
See [Schappacher 1997], cf. [Vl
adut 1991, chap. 3 and chap. 4, in particular pp. 74
76]. Note that Eisensteins special case can be conveniently used to settle the normalizing
property needed in the identication of the Taniyama group, and the simultaneous proof of
the generalization due to Deligne and Tate of the Shimura-Taniyama reciprocity law; see
[Schappacher 1994, 4.4.4].
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
266
N. SCHAPPACHER
Webers Algebra can be found in [Deuring 1954].

The key result established in this tradition is a prototype of what is known
today as the Shimura-Taniyama congruence relation, and thus of one of the
central theorems of CM arithmetic, and the key for the computation of the
Hasse-Weil L-function of the curve. Hilbert was surely aware of this tradition
and its potential arithmetic relevance, in particular to higher reciprocity laws.
He does not, however, make an explicit connection between his 12th problem
and the ninth on general higher reciprocity laws. Only his comments in the
middle passage of the 12th problem, on the analogies between the theory of
algebraic functions of one variable and number theory, might conceivably be
understood as hints in this direction. Still, it is striking that Hilbert does
not seem to want to build this aspect of Kroneckers work into the research
programme he proposes. He rather appears to have had a denite project in
mind, and rewrote the history of complex multiplication in the 19th century
accordingly.
The higher-dimensional theory was linearized in the book [Serre 1968]
and succesfully integrated into a theory motives by Deligne, with Langlandss
M
archen leading the waycf. [Schappacher 1994] and the literature cited
there.
But the history of ideas is not a one-way street, and the tradition of looking
at the theory of complex multiplication mainly as a source of singular values
of modular forms or functions and a tool for working with them, not only kept
very much alive thoughout our centuryfrom Hilberts 12th problem, to the
thesis under Emil Artins supervision [Sohngen 1935], all the way to Shimuras
reciprocity law for singular values of Hilbert modular forms, but was nally
turned even to diophantine problems following Kurt Heegners seminal ideas
and their development by Birch.
With motives and Heegner pointswhich both can only be fully appreciated today against the background of the theory of Shimura varietieswe
have reached active current research. I hope to be able to come back on a
later occasion to a more detailed historic analysis of these developments of
the past fty years.
Bibliography
Unpublished sources
NSUG = Nieders
achsische Staats- und Universitatsbibliothek Gottingen, Abteilung
f
ur Handschriften und seltene Drucke.
` 3
SEMINAIRES
ET CONGRES
267
NSUG Cod. Ms. D. Hilbert 422, 35 letters from Heinrich Weber to David Hilbert,
dated between 1885 and 1912; in the Hilbert Nachla.
NSUG 8 NSUG 8 Cod. Ms. philos. 205, 37 letters from various authors, most of
them to Heinrich Weber; obituary notice of Heinrich Weber with handwritten notes
by David Hilbert. The documents were in Hilberts possession and were given to
NSUG after his death.
Published sources
Alexandrov (P.S.) (ed.)
[1979]
Die Hilbertschen Probleme, erl

autert von einem Autorenkollektiv unter
der Redaktion von P.S. Alexandrov. Leipzig, 1979.
Bauer (D.)
[1903]
Uber
den Teilungsk
orper der elliptischen Funktionen mit singul
arem
Modul, diss. Strasbourg.
Blumenthal (O.)
[1903a]
Abelsche Funktionen und Modulfunktionen mehrerer Veranderlicher,

Verh. Naturf. Ges. Cassel, 2-1 (1903), p. 210.
[1903b]
Uber
Modulfunktionen von mehreren Veranderlichen, Math. Ann., 56
(1903), pp. 509548.
[1904a]
Bemerkungen zur Theorie der automorphen Funktionen, Nachr. K. Ges.

Wiss. G
ottingen, (1904), pp. 9297.
[1904b]
Uber
Thetafunktionen und Modulfunktionen mehrerer Veranderlicher,
Jahresber. Deutsch. Math.-Verein., 13 (1904), pp. 120132.
[1904c]
Uber
Modulfunktionen von mehreren Veranderlichen, II, Math. Ann., 58
(1904), pp. 497527.
Bourbaki (N.)
ements dhistoire des math
[1984]
El
ematiques. Paris: Masson, 1984.
Cebotarev
(N.)
[1924]
Dokzatelstvo teoremy Kroneckera-Webera otnositelno abelevych

oblastej, Mat. Sb., 31 (1924), pp. 302309.
Craig (J.E.)
[1984]
Scholarship and Nation Building. The Universities of Strasbourg and

Alsatian Society 1870-1939, Chicago: University of Chicago Press, 1984.
Dedekind (R.) and Weber (H.)

[1882]
Theorie der algebraischen Funktionen einer Veranderlichen, J. Reine

Angew. Math., 42 (1882), pp. 181290.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
268
N. SCHAPPACHER
Deuring (M.)
[1953-57]
Die Zetafunktion einer algebraischen Kurve vom Geschlechte Eins, Nachr.

K. Ges. Wiss. G
ottingen. (1953), pp. 8594; Zweite Mitteilung, Ibid.
(1955), pp. 1342; Dritte Mitteilung, Ibid. (1956), pp. 3776; Vierte
Mitteilung, Ibid. (1957), pp. 5580.
[1954]
Zur Transformationstheorie der elliptischen Funktionen Akad. Wiss. &

Lit. Mainz, Abh. Math.-Nat. Kl. (1954), pp. 95104.
Eisenstein (G.)
[1850]
Uber
die Irreductibilitat und einige andere Eigenschaften der Gleichung,
von welcher die Theilung der ganzen Lemniscate abhangt...; J. Reine
Angew. Math., 39 (1850) pp. 160179, 224287.
Franci (R.)
[1992]
On the axiomatization of group theory by American mathematicians:

19021905, in Amphora, Festschrift f
ur Hans Wuing zu seinem 65.
Geburtstag, herausgegeben von S.S. Demidov (edited by M. Folkerts,
D. Rowe and C. Scriba), Basel-Boston: Birkhauser, 1992, pp. 261277.
Frei (G.)
[1989]
Heinrich Weber and the emergence of class eld theory, in The History of
Modern Mathematics, vol. I: Ideas and their Reception (edited by Rowe
and McCleary), San Diego: Academic Press, pp. 424450.
[1995]
Heinrich Weber (18421913), in Die Albertus-Universit

at zu K
onigsberg
und ihre Professoren, aus Anla der Gr
undung der Albertus-Universit
at
vor 450 Jahren herausgegeben von Dietrich Rauschning und Donata v.
Neree, Berlin: Duncker & Humblot, pp. 509520.
Frobenius (G.)
[1911]
Gegenseitige Reduktion algebraischer Korper (Auszug aus einem

Schreiben an H. Weber), Math. Ann., 70 (1911), pp. 457458; Gesammelte
Abhandlungen, III, Berlin - Heidelberg: Springer, 1968, pp. 491492.
Fueter (R.)
[1903]
Der Klassenk
orper der quadratischen K
orper und die komplexe Multiplikation. Ph.D. thesis, Gottingen.
[1905]
Die Theorie der Zahlstrahlen, J. Reine Angew. Math., 130 (1905),

pp. 197237.
[1907]
Die Theorie der Zahlstrahlen, II, J. Reine Angew. Math., 132 (1907),
pp. 255269.
[1914]
Abelsche Gleichungen in quadratisch-imaginaren Zahlkorpern, Math.

Ann., 75 (1914), pp. 177255.
` 3
SEMINAIRES
ET CONGRES
[1924]
269
Vorlesungen u
aren Moduln und die komplexe Multiplikation
ber die singul
der elliptischen Funktionen, I, Leipzig-Berlin: Teubner, 1924.
Greenberg (M.)
[1974-75]
An elementary proof of the Kronecker-Weber theorem, Amer. Math.

Monthly, 81 (1974), pp. 601607; 82 (1975), p. 803.
Hasse (H.)
[1926a]
Bericht u
ber neuere Untersuchungen und Probleme aus der Theorie der
algebraischen Zahlkorper, Jahresber. Deutsch. Math.-Verein., 35 (1926),
pp. 155. I quote from the book edition, W
urzburg-Wien: Physica, 1970.
[1926b]
Book review of [Fueter 1924], Jahresber. Deutsch. Math.-Verein., 35

(1926), pp. 5562.
[1927]
Neue Begr
undung der komplexen Multiplikation, J. Reine Angew. Math.,
157 (1927), pp. 115139; see also Mathematische Abhandlungen 2, BerlinNew York: de Gruyter, 1975, pp. 327.
[1930]
Leopold Kroneckers Werke, in Kronecker Werke V, pp. 510515.
Hecke (E.)
[1912]
H
ohere Modulfunktionen und ihre Anwendung auf die Zahlentheorie,
Math. Ann., 71 (1912), pp. 137; see also Mathematische Werke, Gottingen: Vandenhoeck 1959, pp. 2158.
[1913]
Uber
die Konstruktion relativ-abelscher Zahlkorper duch Modulfunktionen von zwei Variablen, Math. Ann., 74 (1913), pp. 465510; Math.
Werke, pp. 69114.
Hilbert (D.)
[Ges. Abh.] Gesammelte Abhandlungen,
Springer 1970.
vols.,
Berlin-Heidelberg-New York:
[1896]
Ein neuer Beweis des Kroneckerschen Fundamentalsatzes u

ber Abelsche
Zahlk
orper, Nachr. K. Ges. Wiss. G
ottingen, (1896), pp. 2939; Ges.
Abh., pp. 5362.
[1901]
Mathematische Probleme, Arch. Math. Phys., 1 (1901), pp. 4463, 213

237; Ges. Abh. III, pp. 290329. [First German publication of this text in
Nachr. K. Ges. Wiss. G
ottingen 1900. Hilbert modied it for the French
translation in ICM 1900, in problems 13, 14, and 23.].
ICM
[1900]
Compte rendu du deuxi`

eme Congrès International des Mathematiciens
tenu `
a Paris du 6 au 12 ao
ut 1900, procèsverbaux et communications,
Paris: Gauthier-Villars, 1902.
[1932]
Verhandlungen des Internationalen Mathematikerkongresses Z

urich 1932,
Z
urich-Leipzig: Orell F
ussli, 1933.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
270
N. SCHAPPACHER
Iwasawa (K.)
[1990]
On papers of Takagi in number theory, in Takagi Papers, pp. 342351.
Iyanaga (S.)
[1990]
On the life and works of Teiji Takagi, in Takagi Papers, pp. 354371.
Katsuya (M.)
[1995]
The establishment of the TakagiArtin class eld theory, in The intersection of history and mathematics (edited by C. Sasaki, M. Sugiura and
J.W. Dauben), vol. 15 of Science networks, Historical Studies, BaselBoston: Birkh
auser, 1995, pp. 109128.
Klein (F.)
[1926]
Vorlesungen u
ber die Entwicklung der Mathematik im 19. Jahrhundert,
vol I, Berlin: Springer, 1926.
Kronecker (L.)
[Werke]
Leopold Kroneckers Werke, 5 vols, Berlin-Leipzig: Teubner, 1895-1931.
[1853]
Uber
die algebraisch auosbaren Gleichungen, Monatsber. Akad. Berlin
(20.6.1853), (1853), pp. 365374; Werke IV, pp. 311.
[1856]
Uber
die algebraisch auosbaren Gleichungen, Monatsber. Akad. Berlin
(14.4.1856), (1856), pp. 203215; Werke IV, pp. 2537.
[1857]
Uber
die elliptischen Funktionen, f
ur welche komplexe Multiplikation
stattndet, Monatsber. Akad. Berlin (29.10.1857), (1857), pp. 455460;
Werke IV, pp. 179183.
[1877]
Uber
Abelsche Gleichungen, Monatsber. Akad. Berlin (16.4.1877), (1877),
pp. 845851; Werke IV, pp. 6371.
[1880]
Letter to R. Dedekind, dated Berlin, 15 March 1880. Partly published in

Werke V, pp. 453457, (three pages, numbered with gaps).
Manegold (K.)
[1970]
Universit
at, technische Hochschule und Industrie. Ein Beitrag zur
Emanzipation der Technik im 19. Jhd. unter besonderer Ber
ucksichtigung der Bestrebungen Felix Kleins, Schriften zur Wirtschafts- und
Sozialgeschichte, 16 (1970), Berlin.
Masahito (T.)
[1994]
Three aspects of complex multiplication, in The intersection of history

and mathematics (edited by C. Sasaki, M. Sugiura and J.W. Dauben),
vol. 15 of Science Networks, Historical Studies, Basel-Boston: Birkhauser,
1994, pp. 91108.
` 3
SEMINAIRES
ET CONGRES
271
Neumann (O.)
[1981]
Two proofs of the Kronecker-Weber theorem according to Kronecker,

and Weber, J. Reine Angew. Math., 323 (1981), pp. 105126.
Poincar
e (H.)
[1901]
Sur les proprietes arithmetiques des courbes algebriques, J. Math. Pures

Appl., (V) 7 (1901), pp. 161-233.
Reid (C.)
[1970]
Hilbert, Berlin-Heidelberg: Springer, 1970.
Schappacher (N.)
[1982]
Une classe de courbes elliptiques à multiplication complexe, in Seminaire de Theorie des Nombres, Paris 1980-81, vol. 22 of Progress in
Mathematics, Birkhauser, pp. 273279.
[1994]
CM motives and the Taniyama group, in Motives (Jannsen, Kleiman,

Serre, eds.), Proceedings of Symposia in Pure Mathematics, vol. 55-1,
Providence: Amer. Math. Soc., 1994, pp. 485508.
[1997]
Some Milestones of Lemniscatomy, in Algebraic Geometry (S. Sertoz, ed.),

Proceedings of Bilkent Summer School, Ankara 1995, Lecture Notes in
Pure and Applied Mathematics Series, vol. 193, 1997, New York: Dekker,
pp. 257290.
Schappacher (N.) and Volkert (K.)

[1998]
Heinrich Weber: un mathematicien à Strasbourg, 18951913, in Sciences

et cultures Les trois universit
es de Strasbourg 18721945, to appear.
Serre (J.P.)
[1967]
Complex multiplication, in Algebraic Number Theory, Proc. Instructional

Conf., Brighton 1965 (edited by J. Cassels and A. Frolich), LondonNew York: Academic Press, pp. 292296; uvres, II, Berlin-Heidelberg:
Springer, 1986, p. 455-459.
[1968]
Abelian !-adic Representations and Elliptic Curves, Benjamin 1968; reissued by Addison-Wesley, 1989.
(I.)
Safarevi
c
[1951]
A new proof of the Kronecker-Weber theorem, Trudy Mat. Inst. Steklov,

38 (1951) pp. 382387; see also Collected Mathematical Papers, BerlinHeidelberg: Springer, 1989, pp. 5458.
Shimura (G.) and Taniyama (Y.)

[1961]
Complex multiplication of Abelian varieties and its applications to number

theory, Tokyo: The Mathematical Society of Japan,1961.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998
272
N. SCHAPPACHER
S
ohngen (H.)
[1935]
Zur komplexen Multiplikation, Math. Annalen, 111 (1935), pp. 302328.
Speiser (A.)
[1919]
Die Zerlegungsgruppe, J. Reine Angew. Math., 149 (1919), pp. 174188.
Takagi (T.)
[Papers]
Collected Papers, Tokyo-Berlin: Springer, 1990.
[1903]
Uber
die im Bereiche der rationalen complexen Zahlen Abelschen Korper,
J. Col. Sci. Imp. Univ. Tokyo, 19-5 (1903), pp. 142; Papers, pp. 13-39.
[1920]
Uber
eine Theorie des relativ Abelschen Zahlkorpers, J. Col. Sci. Imp.
Univ. Tokyo, 41-9 (1920), pp. 1133; Papers, pp. 73167.
du
Vla
t (S.G.)
[1991]
Kroneckers Jugendtraum and modular functions, vol. 2 of Studies in the

Development of Modern Mathematics, New York: Gordon & Breach, 1991.
Voss (A.)
[1914]
Heinrich Weber, Jahresber. Deutsch. Math.-Verein., 23 (1914), pp. 431

444.
Washington (L.C.)
[1982]
Introduction to Cyclotomic Fields, vol. 83 of Graduate Texts in Mathematics, New York: Springer, 1982.
Weber (H.)
[1886]
Theorie der Abelschen Zahlkorper, Acta Math., 8 (1886), pp. 193263.
[1891]
Elliptische Funktionen und algebraische Zahlen, Braunschweig, 1891.
[1893]
Die Grundlagen der Galoisschen Gleichungs-Theorie, Math. Ann., 43

(1893), pp. 521524.
[1894]
Lehrbuch der Algebra, I, Braunschweig, 1894, second edition 1898. I usually quote the Chelsea reprint of the second edition which contains minor
changes.
[1896]
Lehrbuch der Algebra, II, Braunschweig, 1896, second edition 1899. I usually quote the Chelsea reprint of the second edition which contains minor
changes.
[1897-98]
Uber
Zahlengruppen in algebraischen Zahlkorpern, I, II, III, Math. Ann.,
48 (1897), pp. 433473; 49 (1897), pp. 83100; 50 (1898), pp. 126.
[1900]
Komplexe Multiplikation, in Encyclop

adie der mathematischen Wissenschaften, vol. I.2.C, Zahlentheorie, Leipzig: Teubner, 1900, pp. 716
732.
` 3
SEMINAIRES
ET CONGRES
273
[1907]
Uber
zyklische Zahlkorper, J. Reine Angew. Math., 132 (1907), pp. 167
188.
[1908]
Lehrbuch der Algebra, III, Braunschweig: Vieweg, 1908. I usually quote

the Chelsea reprint of this only edition of the third volume which contains
minor changes.
[1909]
Zur Theorie der zyklischen Zahlkorper, Math. Ann., 67 (1909), pp. 3260.
[1911]
Zur Theorie der zyklischen Zahlkorper, II, Math. Ann., 70 (1911), pp. 459
470.
Weil (A.)
[uvres]
uvres scientiques, 3 vols, New York-Heidelberg: Springer, 1979. References to papers or to commentaries follow the abbreviations introduced
in this edition.
Wollmersh
auser (F.)
[1981]
Das Mathematische Seminar der Universitat Straburg 18721900, in

E.B. Christoel, The Inuence of his Work on Mathematics and the Physical Sciences (edited by F. Butzer). Basel: Birkhauser, 1981.
Zassenhaus (H.)
[1968-69]
On a theorem of Kronecker, Delta, 1 (1968-69), pp. 114.
E
MATHEMATIQUE
SOCIET
DE FRANCE 1998

Deligne Sur Grothendieck

Transféré par

Informations du document

Description :

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Description :

Droits d'auteur :

Formats disponibles

Deligne Sur Grothendieck

Titre original :

Transféré par

Description :

Droits d'auteur :

Formats disponibles

Quelques id

Dans Recoltes et Semailles (troisi`eme partie), Grothendieck ecrit :

AMS 1991 Mathematics Subject Classication: 01A65, 14-03

souvre comme celle dun avocat m

Dans la categorie des schemas, les produits bres existent toujours : si

On note X(S ) lensemble de S -points de X. Il sidentie a` lensemble des

Soit k une k-alg`ebre. Lensemble X(k ) := X(Spec(k )) est lensemble des

 espace de module des sous-schemas fermes de X  ? Pour u : S Spec(k),

soit X sur S deduit de X par changement de base. Soit H(S ) lensemble

Soit u : S S un morphisme de schemas, le  morphisme de changement

(= localement trivial pour une topologie proche de la topologie etale). Dans

Le formalisme des six operations presuppose celui des categories derivees.

du produit tensoriel derive , du Hom interne RHom (donnant naissance

Soit X une variete algebrique sur k algebriquement clos. Pour chaque

Cest `a cette occasion que Grothendieck a invente la notion de categorie

Algebraization of formal moduli I, dans Global Analysis (Papers in honor

Cartier (P.), Illusie (L.), Katz (N.M.) et al., ed.

The Grothendieck Festschrift, vol. I, vol. 86 de Progress in Mathematics.

A general theory of bre spaces with structure sheaf. University of Kansas,

Sur quelques points dalg`ebre homologique, T

Residus et dualite, prenotes pour un  seminaire Hartshorne , manuscrit.

Standard conjectures on algebraic cycles, Algebraic Geometry (Coll. Tata

ements de Geometrie Algebrique. Inst. Hautes Etudes

Fondements de la Geometrie Algebrique. Extraits du seminaire Bourbaki

Seminaire de Geometrie Algebrique du BoisMarie. Le seminaire SGA 5 :

Categories derivees et dualite, travaux de J.-L. Verdier, Enseign. math.,

Jannsen (U.), Kleiman (S.) and Serre (J.-P.), ed.

Home is where one starts from. As one grows older

AMS 1991 Mathematics Subject Classication: 01A60, 55-03

Dans cet expose je traite un episode de lhistoire des mathematiques bien

1.1. Le terme  bration  au sens de cet expose apparat pour la pre

Hopf  de sph`eres en sph`eres

Dans tous ces cas la projection p est lapplication evidente.

une generalisation. Dej`

1931 par Hopf [1931] dans son cel`ebre travail o`

La suite exacte des br

Ce que lon appela plus tard la suite exacte dune bration p : EP

n (E) n (B) n1 (F )...

3.2. Un element de n (B) peut etre represente par une application f de

voisinage autour dun point y arbitraire  permet de relever morceau par

Le Lemme et la suite exacte dhomotopie etablis, les resultats tombaient

2 (U (n)) = 2 (U (2)) = 2 (SU (2)) = 2 (S 3 ) = 0,

les n-squelettes de X que

6.1. Je consid`ere [Eckmann et al. 1949] une bration S n B dont la

Hn (B) = Hn (Y ) = Hn (Y ).

La dimension de B etant n s, il sensuit que Hn (Y ) = 0, donc que n est

bration S n RP n1 qui donne lieu a` une suite exacte o`

On the nonexistence of elements of Hopfinvariant one, Ann. of Math.,

Vector elds on spheres, Ann. of Math., 75 (1962), p. 603632.

An application of the Morse theory to the topology of Lie groups, Bull.

Bott (R.) et Milnor (J.)

On the parallelizability of spheres, Bull. Amer. Math. Soc., 64 (1958),

A History of Algebraic and Dierential Topology 1900-1960. Boston :

Hurwitz-Radon matrices revisited : From eective solution of the Hurwitz

Eckmann (B.), Samelson (H.) et Whitehead (G.)

On bering spheres by toruses, Bull. Amer. Math. Soc., 55 (1949), p. 433

Ehresmann (C.) et Feldbau (J.)

Eilenberg (S.) et MacLane (S.)

Natural isomorphisms in group theory, Proc. Nat. Acad. Sci. USA, 28

espace de module des sous-schemas fermes de X ? Pour u : S Spec(k),

Soit u : S S un morphisme de schemas, le morphisme de changement

Residus et dualite, prenotes pour un seminaire Hartshorne , manuscrit.

1.1. Le terme bration au sens de cet expose apparat pour la pre

Hopf de sph`eres en sph`eres

voisinage autour dun point y arbitraire permet de relever morceau par

+ Cect Dm1 u(0, .)