• Aucun résultat trouvé

The order of vanishing of the Riemann theta function in the KP direction: a geometric proof,

N/A
N/A
Protected

Academic year: 2022

Partager "The order of vanishing of the Riemann theta function in the KP direction: a geometric proof,"

Copied!
8
0
0

Texte intégral

(1)

CHRISTINABIRKENHAKEANDPOLVANHAECKE

Abstrat. Wegiveageometriproofofaformula,duetoSegalandWilson,

whihdesribestheorderof vanishingof theRiemannthetafuntioninthe

diretionwhihorrespondstothediretionofthetangentspaeofaRiemann

surfaeat amarkedpoint. Whilethisformulaappearsinthe workofSegal

andWilsonasaby-produtofsomenon-trivialonstrutionsfromthetheory

of integrable systems(loop groups, innite-dimensionalGrassmannians, tau

funtions,Shurpolynomials,:::)ourproofonlyusesthelassialtheoryof

linearsystemsonRiemannsurfaes.

Contents

1. Introdution 1

2. Preliminaries 2

3. Geometridesriptionoftheorderofvanishing 4

4. Thedivisor

L;p; n

5

5. Formula(s)fortheorder ofvanishing 6

Referenes 7

1. Introdution

Thefundamental paper[SW ℄bySegalandWilsononsoliton equationsleadsto

anexpliitformulafor omputingthevanishing oftheRiemann theta funtionin

adiretionwhihis naturalfrom thegeometripointofview. Inorderto present

this formula, let C be a ompat Riemann surfae (of genus g > 1), let p 2 C

and let denote thetheta divisor Pi g 1

(C). Alsolet # denote Riemann's

theta funtion, for whih (#) =. For apoint L 2 , onsider the embedding

C ! Pi g 1

(C) : q 7! L(q p). The natural diretion alluded to above is the

tangentspae X

p

to this embedded urveat L. Following[SW ℄ the vanishing of

#atL in thediretionofX

p

,denotedord

L (#;X

p

)isobtainedbyonsideringthe

innitesubsetofZ,denedby

S

L

=fs2Zjh 0

(L((s+1)p))=h 0

(L(sp))+1g:

Infat,thesetionsofLoverCnfpgdeneaninnite-dimensionalplaneW,whih

is anelementof the SatoGrassmannian andthe vanishing ofthe taufuntion in

theKP-diretionis,aordingto [SW,Prop.8.6℄,givenbytheodimensionofW,

1991MathematisSubjetClassiation. 14K25,14H40,14H55.

Keywordsandphrases. Thetafuntions,Jaobians,Gapsequenes.

SupportedbyDFG-ontratsHu337/5-1.

(2)

whihis expliitly givenby thenite sum

i0 i s

i

, upon writingS

L

=fs

0

<

s

1

< s

2

< g. The tau funtion oinides, up to an exponential fator, with

the Riemann theta funtion of C ([SW, Th. 9.11℄)and thetangentdiretion X

p

oinides with the KP-diretion (see [S , Lemma 5and Appendix 0℄). Therefore,

theorderofvanishingisgivenby

ord

L (#;X

p )=

X

i0 i s

i :

Thepurposeofthis paperisto giveanalgebrai-geometriproof ofthisresult.

Our proof uses (only) the lassialtheory of linear systems onRiemann surfaes

and it highlights thegeometri meaning of the order of vanishing. As is pointed

outin [SW, footnotep. 51℄ anindependent (analytial)proofof this formulahas

alsobeengivenbyJohnFay(see[F℄),byusingthetheoryofthetafuntions.

Therststepofourapproahonsistsofaninterpretationoftheorderofvanish-

ingas theintersetionmultipliityofthethetadivisorwith aopyofC,properly

embedded (at leastaroundL) in Pi g 1

(C). If wepull bakthetheta divisorus-

ingthis embedding wendadivisorR onC whih isthe sumof theramiation

divisorsofthemaps

' k

L

:C !Grass

k +1 (H

0

(L)

);

whih are the natural generalizations of the morphisms '

L

: C ! P(H 0

(L)

)

denedbythelinearsystemjLj(assumedherebasepointfree). Itfollowsthatthe

orderofvanishingisgivenbythemultipliityofpin R ,leadingto

ord

L (#;X

p )=

n

X

i=1 m

i i:

where fm

1

< < m

n

g is the gap sequene G

p

(L(np)) of L(np) at p. This

formula is independent of n, whih is assumed suÆientlylarge (e.g. n = g will

do). Notiingthatforn=gonehass

i

=g m

g i

(fori=0;:::;g 1)fromwhih

Formula(1)followsatone.

NotiethattheSegal-Wilsonformulaforthevanishingof thetaufuntion may

also be applied in the ase of tau funtions that ome from singular urves. It

would be interestingto adapt ourgeometri argumentsto this ase, leadingto a

formulaforthevanishingofthethetafuntionsfor singularurves,asproposedin

[SW,Remark6.13℄.

Thestruture of this paper isasfollows. InSetion 2wex thenotation and

wereall the notionsof gap numbers for arbitraryline bundles. In Setion 3we

translatetheorderofvanishingofthethetafuntionintermsofintersetiontheory

and we show that this order is given asan inetionary weight. This is used in

Setion 5 to obtain an expliit formula, whih we show to be equivalent to the

formulabySegalandWilson.

2. Preliminaries

Inthissetionweintroduethenotationandolletsomeresultsonurvetheory.

ThroughoutthewholepaperCdenotesaompatRiemannsurfaeofgenusgand

p2C amarkedpoint.

For adivisorD on C we denote byO

C

(D) theorresponding line bundle and

foralinebundleL onC itslinearsystemisdenotedby jLj. Weuse thestandard

abbreviationsh 0

(L)fordimH 0

(C ;L)andL(D)forLO (D),whereLisanyline

(3)

bundleandD is anydivisoronC. WewillusetheRiemann-Rohtheorem inthe

form

h 0

(L)=h 0

(!

C L

1

) g+deg (L)+1;

whereL isanylinebundleonCand !

C

istheanonialbundleofC.

Wenowreallthenotionsofgapnumbersandinetionaryweights. Forproofs

anddetailswereferto[Mi,Set.VII.4℄andto [ACGH,Ch. 1Ex.C℄.

LetLbealine bundleonC ofpositivedegreeandletq2C. Anintegerm1

isalledagapnumber forL atqif

h 0

(L( mq))=h 0

(L( (m 1)q)) 1;

andthe setG

q

(L) ofgap numbersfor Lat qis alled thegapsequene ofL at q;

itsardinalityisr=h 0

(L)andnogapnumberislargerthandegL+1. Writing

G

q

(L)=f1m

1

<m

2

<<m

r

degL+1g;

wehavethatm

1

>1ifandonlyifqisabasepointofLand thatm

r

=degL+1

ifandonlyifL=O

C

(degLq). Forageneralpointq2C thegapsequeneof L

at q is f1;2;:::;h 0

(L)g; apointq for whih thegap sequene of L at q isnot of

this formis alledan inetion pointfor L. Notie that qisan inetionpointif

andonlyifh 0

(L( rq))6=0,wherer=h 0

(L).

IfthelinearsystemjLjisbasepointfreetheinetionpointshavethefollowing

geometri interpretation. Considerthe morphism '

L

: C ! P(H 0

(L)

)dened

by the linear system jLj. For a generi q 2 C there is a unique k-dimensional

osulatingplaneto'

L

(C)at'

L

(q),yieldingawell-denedmorphism

(1) '

k

L

:C !Grass

k +1 (H

0

(L)

);

alledthek-thassoiatedmap. Thiswayonearrivesat h 0

(L) 1assoiatedmaps

' i 1

L

; i =1;:::;h 0

(L) 1;(' 0

L

= '

L

). In these terms apointq is aninetion

pointifandonlyifqisaramiationpointofoneofthemaps' k

L

. Wedenotethe

ramiationdivisorof' k

L byR

k

(L)andwedene

R (L)= h

0

(L) 1

X

k =1 R

k 1 (L):

The multipliity w

q

(L) of q in R (L) is alled the inetionary weight of q with

respettoL andisgivenby

(2) w

q (L)=

h 0

(L)

X

i=1 (m

i i):

When L is not base point free we dene the inetionary weights w

q

(L) by (2)

and theramiation divisorR (L) byR (L) = P

q w

q

(L)q. Thisdivisor admitsan

alternativedesriptionasthezerodivisorofW =W(z)(dz) n(n 1)

2

wherezisaloal

oordinate,n=h 0

(L)andW(z)=W(f

1

;:::;f

n

)istheWronskianwithrespetto

anybasisf

1

;:::;f

n ofH

0

(L). InpartiularW isaholomorphisetionoftheline

bundleL n

! n(n 1)

2

C .

TakingL=!

C

onereoversthewell-knownnotionofthegapsequeneofq2C,

denoted by G

q

, and the above denition of the inetionary points and weights

redues, by a simple appliation of Riemann-Roh, to the standard denition of

(4)

Finally let us x our onventions about the Jaobian J(C) of C. By deni-

ton J(C) =H 0

(!

C )

=H

1

(C ;Z), sothat the vetorspae H 0

(!

C )

is anonially

identied with the tangent spae of J(C) at every point. By the Abel-Jaobi

theorem there is a anonial isomorphismJ(C) ' Pi 0

(C). Moreoverevery line

bundle L on C of degree g 1 indues an isomorphism J(C) ' Pi 0

(C) !

Pi g 1

(C);P 7!LP. Forourpurposesitisonvenientto workwithPi g 1

(C)

rather than Pi 0

(C). So in the sequel we identify J(C) with Pi g 1

(C) with-

out further notie; theunderlying isomorphism(respetivelyline bundledening

the isomorphism) will always be evident from the ontext. The main advan-

tage working with J(C) = Pi g 1

(C) is, that we have a anonial theta divisor

=fL 2J(C) j h 0

(L)> 0g. ByRiemann-Roh, is invariant with respet to

thenaturalinvolution

(3) :J(C) !J(C);(L)=!

C L

1

:

Morepreiselywehaveh 0

(L)=h 0

((L))foranyL2.

Wedenote by#the Riemanntheta funtion onH 0

(!

C )

forwhih

isthe

zerodivisorof#,where isthenaturalprojetionH 0

(!

C )

!J(C).

ForanyL2J(C)wehaveanembedding

L;p

ofCintoJ(C),givenby

L;p (q)=

L(q p). Clearlyfor dierentL andpthemaps

L;p

onlydierbyatranslation

onJ(C).

3. Geometri desription of the order of vanishing

Let L 2 J(C) and let X be aone-dimensional subvetor spae of the tangent

spaeH 0

(!

C )

at L. Chooseanypointlintheberof overL andonsiderthe

aÆneline l+X whih passesthroughland whih hasdiretion X. Theorder of

vanishing of #j

l+X

at the pointl is independent of the hoie of l 2 1

(L). So

denetheorderofvanishing of #at Linthediretion of X,denotedord

L (#;X),

asord

l

#j

l+X

. Ifdoesnotontainthestraightline

X =(l+X)thenthereexists

asmallneighborhoodU oflinl+X suhthat(U)\=fLgandord

L

(#;X)=

((U))

L

,theintersetion multipliityofwith(U)atL.

LetX

p

denote thetangentspae to

L;p

(C)at L. Notie that,asasubvetor

spae of H 0

(!

C )

, X

p

does not depend on L but only on the point p 2 C. We

wish to omputeord

L (#;X

p

)foran arbitraryL 2; ifL2= thenthis order is

triviallyzero. Itanbeshown 1

that forall C, L andpthat

X

p

isnotontained

in ,whih islearlytrueassoonasC, L orpis generi. Forthis theideais to

replae(U)byaaompleteurvewhih,aroundL,looks like(U). Notiethat

ifL2weannotuse

L;p

(C)beausethelatterdoesnotneessarilyinterset

properly. Considerforanyintegern6=0themorphism

(4)

L;p;n

:C !J(C);

L;p;n

(q)=L(nq np):

Notie that

L;p;n

(p) = L and that for a small neighborhood V of p in C the

tangentspaeto

L;p;n

(V)at LispreiselyX

p .

Lemma3.1. ForallL2J(C)and n>0wehave

(1)

L;p;n

(C)andinterset properly ifandonly ifh 0

(L( np))=0;

(2)

L;p; n

(C)andintersetproperly if andonlyif h 0

((L)( np))=0.

1

Thisfollowsfrom[SW , Prop. 8.6 andTh.9.11℄ butageometriproofofthis(geometri!)

(5)

Proof. We prove(1), the proof of (2) is similar. Reall that anirreduible urve

intersetsadivisorproperlypreiselywhentheurveisnotontainedinthesupport

ofthedivisor. So

L;p;n

(C)anddonotintersetproperlyifandonlyifh 0

(L(nq

np))>0forallq2C. Welaimthatthisisequivalenttoh 0

(L( np))>0. Indeed,

byRiemann-Roh

h 0

(L(nq np))=h 0

((L)(np nq)); and

h 0

((L)(np))=h 0

(L( np))+n:

Soh 0

(L(np nq))>0forallq2C ifandonlyifh 0

((L)(np))>n,leadingtoour

laim.

Inpartiular,whenjnjg then

ord

L (#;X

p

)=(

L;p;n (V))

L

;

whereV isasmallneighborhoodofpin C. PullingthisintersetionbaktoC we

getthatforanyjnjg

ord

L (#;X

p

)=mult

p (

L;p;n ):

Thismultipliitywillbeomputedinthenextsetion.

4. The divisor

L;p; n

Theaimofthissetionistoprovethefollowing

Theorem4.1. ForallL2J(C), p2C,and n>0with h 0

((L)( np))=0

L;p; n

=R (L(np)):

For theproofweneedthefollowing

Proposition4.2. ForallL2J(C);p2C andn>0

L;p; n O

J(C)

()=(L(np)) n

! n(n 1)

2

C :

Proof.

StepI:Theasen=1followsexatlyfrom[LB℄Lemma11.3.4withx=0;=

L, and=p.

StepII:Forn1onsiderthedierenemap

Æ n

L :C

2n

!J(C); Æ n

L (p

1

;q

1

;:::;p

n

;q

n )=L(

P

i p

i q

i )

anddenoteby n

i :C

2n

!Cthei-thprojetion. Weshowbyindutiononnthat

foralln1andallL2J(C)

Æ n

L

O

J(C) ()=

n

O

i=1

n

2i 1

(!

C L

1

) n

2i

L

O

C 2n

X

1i<j2n ( 1)

i+j+1

( n

i

; n

j )

; (5)

where denotesthediagonalin C 2

.

Forn=1wehavetoshowthat

Æ 1

L

O

J(C)

()= 1

1

(!

C L

1

) 1

2

LO

C 2

()

forallL2Pi g 1

(C). AordingtotheSeesawPriniple(see[LB℄A.9)itsuÆes

(6)

oinideforallq2C. ButsinetheompositionofÆ 1

L

withthenaturalembedding

C'Cfqg !CC isthemap Æ

!CL 1

;q; 1 and

=wehave,using

StepI,

Æ 1

L

O

J(C)

()jCfqg=

!

C L

1

;q; 1 O

J(C)

()=!

C L

1

(q)

= 1

1

(!

C L

1

) 1

2

LO

C

2()jCfqg;

andsimilarlyfortherestritionto fqgC.

Now suppose n > 1 and equation (5) holds for all n 0

< n. Restriting both

sides of equation (5) to C 2n 2

fp;qg and fp

1

;q

1

;:::;p

n 1

;q

n 1 gC

2

for all

p;q;p

1

;q

1

;:::;p

n 1

;q

n 1

2C, andusing the indution hypothesis for n 0

=n 1

andn 0

=1respetively,theSeesawPrinipleimpliesthatalsoequation (5)holds.

StepIII:Considertheembedding |

p

:C !C 2n

; |

p

(q)=(p;q;:::;p;q)and

notiethatÆ n

L Æ|

p

=

L;p; n

,sothat

L;p; n O

J(C) ()=|

p Æ

n

L

O

J(C) ():

Itfollowsthat

L;p; n O

J(C)

()anbeomputedfrom(5). Sine

( n

i

; n

j )Æ|

p (q)=

8

<

:

(p;p) i;j odd

(q;q) i;j even

(p;q)or(q;p) otherwise,

wehavethat

|

p (

n

i

; n

j )

O

C 2()=

8

<

: O

C

i;j odd

! 1

C

i;j even

O

C

(p) otherwise.

It followsthat the pullbak by|

p

of theright hand sideof (5)equalsL n

(n 2

p)

! n(n 1)

2

C

. This ompletestheproof.

Proofof Theorem 4.1. Bythe hoie of n we haveh 0

(L(np)) = n and the urve

L;p; n

(C)intersetsthedivisor properly. Theline bundle(L(np)) n

! n(n 1)

2

C

hastwodistinguisheddivisors,namely

L;p; n

(aordingtoProposition4.2)and

R (L(np))(aordingtoSetion2). Moreoverthesedivisorshavethesamesupport,

sinebydenitionq2

L;p; n

ifandonlyif

L;p; n

(q)2,i.e.,h 0

(L(np nq))>

0,andthisistheaseifandonlyifqisaninetionpointforthelinebundleL(np).

Forgeneri L and pthe line bundle L(np) admits only normal inetion points,

so R (L(np)) = P

n 2

g

i=1 q

i

with pairwise dierent points q

i

, and hene R (L(np)) =

L;p; n

. Thisequality extendsto allL andpfor whih

L;p; n

exists andby

Lemma3.1this isexatlythesetf(L;p)2J(C)Cjh 0

((L)( np))=0g.

Remark 1. Similarly one an show that if

L;p;n

(C) intersets properly, then

L;p;n

=R ((L)(np)).

5. Formula(s)forthe order of vanishing

Inthissetionweprovethefollowingtheorem:

Theorem5.1. Forevery L2

(6) ord

L (#;X

p )=

X

m g+h 0

(L((g m)p));

(7)

wherethe sum runsoverthe g integers msatisfying h 0

(L((g m)p))=h 0

(L((g

m+1)p)) 1.

Wewillobtainitas adiret onsequeneofthefollowingproposition.

Proposition5.2. Withnhosensuhthat

L;p; n

(C)intersetsproperly(e.g.,

n = g), the order of vanishing of # at L in the diretion X

p

is the inetionary

weightof pwithrespettoL(np). Therefore,

(7) ord

L (#;X

p )=

n

X

i=1 m

i i:

wherefm

1

<<m

n

gisthe gapsequeneG

p

(L(np))of L(np)atp.

Proof. AordingtoSetion3andTheorem4.1wehavethat

ord

L (#;X

p

)=mult

p (

L;p; n )=w

p

(L(np))= n

X

i=1 m

i i:

Proofof Theorem 5.1. By denition the sum in equation (6) runs over the set

G

p

(L(gp)) = fm

1

< < m

g

g of gap numbers of L(gp) at p. An immediate

omputationshowsthath 0

(L((g m

i

)p))=g ifori=1;:::;g. Sotheassertion

followsfromProposition5.2withn=g.

We now relate Theorem 5.1 to the Formula (1), given by Segal and Wilson.

Reallfromtheintrodutiontheinniteset

S

L

=fs2Zjh 0

(L((s+1)p))=h 0

(L(sp))+1g:

Proposition5.3. DenoteS

L

=fs

0

<s

1

<s

2

<g. Then

ord

L (#;X

p )=

X

i0 i s

i :

Proof. Note rstthat s

0

degL 1= g. For g sg 1wehavethat

s2S

L

ifandonlyifg s2G

p

(L(gp))=fm

1

<<m

g

g,sothats

i

=g m

g i

fori =0;:::;g 1. Onthe otherhand n2 S

L

for any ng so that s

n

=nfor

anyng. Summingupwend

X

i0 i s

i

= g 1

X

i=0

i g+m

g i

= g

X

i=1 m

i i:

HeneFormula(1)followsfrom Proposition5.2.

Referenes

[ACGH℄ Arbarello,E.,Cornalba, M.,GriÆths,P.A.,Harris,J.,GeometryofAlgebraiCurves,

VolumeI,SpringerGrundlehren,Bd.267(1984)

[F℄ Fay,J.,On theeven-ordervanishing of Jaobian thetafuntions,DukeMath. J.,51,

109{132(1984)

[LB℄ Lange,H.,Birkenhake,Ch.,ComplexAbelianVarieties,SpringerGrundlehren,Bd.302

(1992)

[Mi℄ Miranda,R.,AlgebraiCurvesandRiemannSurfaes,AMS,GraduateStudiesinMath-

ematis,Vol5(1995)

[S℄ Shiota,T.,CharaterizationofJaobianvarietiesintermsofsolitonequations,Invent.

(8)

[SW℄ Segal,G.,WilsonG.,Loopgroupsandequationsof KdVtype,Publ.I.H.E.S.,61,5{65

(1985)

Christina Birkenhake, Mathematishes Institut, Bismarkstrae11/2,D-91054Er-

langen

E-mailaddress: birkenmi.uni-erlangen .de

URL:http://www.mi.uni-erlang en. de/~ bir ken/

PolVanhaeke,

Universit

edePoitiers, D

epartementdePoitiers, Tel

eport2,Boule-

vardMarieetPierreCurie,BP30179,F-86962FuturosopeChasseneuilCedex

E-mailaddress: Pol.Vanhaekemathlabo .un iv-p oit iers .fr

URL:http://wwwmathlabo.sp2mi .un iv-p oit iers .fr /~va nha ek/

Références

Documents relatifs

Yet the linear order is important in a language like English or French where declensions have disappeared, except for personal pronouns 1 , and where the syntactic function of a

Petites valeurs de la fonction d’Euler et hypothèse de Riemann, Séminaire de Théorie des nombres D.P.P., Paris, 1981–82, 207–218, Progress in Mathematics, vol.. Grandes valeurs de

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des

12: Polar contour map of the phase of the complex variable s, the radius representing the magnitude of the zeta function, the azimuthal angle representing the phase angle PhZ of

In his 1859 paper[1], Riemann went further and extended the zeta function ζ(s), by analytical continuation, to an absolutely convergent function in the half plane ℜ (s) &gt; 0, minus

We present a proof of the Generalized Riemann hypothesis (GRH) based on asymptotic expansions and operations on series.. The advan- tage of our method is that it only uses

If a minimum experimentation design was required to validate the hypothesis (proof of the known), additional design activity had to be performed to enable the proof of the

The roots f 1 , f 2 of the derivative polynomial are situated in the interior of the triangle abc and they have an interesting geometric property: f 1 and f 2 are the focal points