• Aucun résultat trouvé

irrational Many values of the Riemann zeta function at odd integers are C.R. Acad. Sci. Paris, Ser.I

N/A
N/A
Protected

Academic year: 2023

Partager "irrational Many values of the Riemann zeta function at odd integers are C.R. Acad. Sci. Paris, Ser.I"

Copied!
5
0
0

Texte intégral

(1)

Contents lists available atScienceDirect

C. R. Acad. Sci. Paris, Ser. I

www.sciencedirect.com

Number theory

Many values of the Riemann zeta function at odd integers are irrational

Beaucoup de valeurs aux entiers impairs de la fonction zêta de Riemann sont irrationnelles

Stéphane Fischler

a

, Johannes Sprang

b

, Wadim Zudilin

c

,

d

aLaboratoiredemathématiquesd’Orsay,UniversitéParis-Sud,CNRS,UniversitéParis-Saclay,91405Orsay,France bFakultätfürMathematik,UniversitätRegensburg,93053Regensburg,Germany

cDepartmentofMathematics,IMAPP,RadboudUniversity,POBox9010,6500GLNijmegen,theNetherlands dSchoolofMathematicalandPhysicalSciences,TheUniversityofNewcastle,Callaghan,NSW2308,Australia

a r t i c l e i n f o a b s t r a c t

Articlehistory:

Received29March2018

Acceptedafterrevision17May2018 Availableonline28May2018 PresentedbyJean-PierreSerre

Inthisnote,weannouncethefollowingresult:atleast2(1−ε)log logslogs valuesoftheRiemann zetafunctionatoddintegersbetween3andsareirrational,where

ε

isanypositivereal numberandsislargeenoughintermsof

ε

.Thisimprovesonthelowerbound1+1−εlog 2logs that followsfromthe Ball–Rivoaltheorem. Wegive the mainideas oftheproof, which isbasedonaneliminationprocessbetweenseverallinearformsinoddzetavalueswith relatedcoefficients.

©2018Académiedessciences.PublishedbyElsevierMassonSAS.Thisisanopenaccess articleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/).

r é s um é

Dans cette note, on annonce le résultat suivant : au moins 2(1ε)log loglogss valeurs de la fonction zêtade Riemannauxentiers impairscompris entre3and s sontirrationnelles, où

ε

estunréelstrictementpositifetsunentierimpairsuffisammentgrandenfonction de

ε

.Ceciaméliorelaborne1+1−εlog 2logsquidécouleduthéorèmedeBall–Rivoal.Ondonne lesidéesprincipalesdelapreuve,quiestfondéesurunprocédéd’éliminationentredes formeslinéairesenlesvaleursdezêtaauxentiersimpairsdontlescoefficientssontreliés.

©2018Académiedessciences.PublishedbyElsevierMassonSAS.Thisisanopenaccess articleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Versionfrançaiseabrégée

Danscettenote, onannoncelerésultat suivant(voirlaversion anglaisepourunbrefhistorique).Onnedonnequeles idéesprincipalesdelapreuve(voir[5] pourlapreuvecomplète).

E-mailaddresses:stephane.fi[email protected](S. Fischler),[email protected](J. Sprang),[email protected](W. Zudilin).

https://doi.org/10.1016/j.crma.2018.05.007

1631-073X/©2018Académiedessciences.PublishedbyElsevierMassonSAS.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/).

(2)

Théorème1.Soient

ε >

0ets3unentierimpairsuffisammentgrand(enfonctionde

ε

).Alors,parmilesnombres

ζ (

3

), ζ (

5

), ζ (

7

), . . . , ζ (

s

),

aumoins

2(1ε)log loglogss sontirrationnels.

Soient

ε >

0 et sun entierimpair suffisamment grand(en fonctionde

ε

). Onnote D leproduit de touslesnombres premiersinférieursouégauxà

(

1−2

ε )

logs,detellesortequ’ona Ds1ε.Ensuivantessentiellement[10],onconsidère, pourtoutentiern,lafractionrationnelle

Rn

(

t

) =

D3Dnn

!

s+13D

3Dn

j=0

(

t

n

+

Dj

)

n

j=0

(

t

+

j

)

s+1

et,pourtout j∈ {1

, . . . ,

D},onpose rn,j

=

m=1

Rn

m

+

j

D

.

RappelonsquelafonctionzêtadeHurwitzestdéfiniepar

ζ (

i

, α )

=

n=0 1

(n+α)i pour

α >

0 eti2.EndéveloppantRn

(

t

)

enélémentssimples,ondémontre,commedans[10],que

rn,j

= ρ

0,j

+

3is iimpair

ρ

i

ζ

i

,

j D

.

Enoutre,ennotantdnlepluspetitcommunmultipledesnpremiersentiers,onmontrequedns++11estunmultiplecommun desdénominateursdesnombresrationnels

ρ

0,jet

ρ

i.

Lepointcrucial,commedans[10],estl’identitésuivante,valablepourtoutdiviseurddeD etpourtoutentieri≥2 :

d

j=1

ζ

i

,

j

D d

D

=

d

j=1

ζ

i

,

j d

=

n=0

d

j=1

di

(

dn

+

j

)

i

=

di

ζ (

i

).

Eneffet,cetterelationmontrequ’enposantrn,d=d j=1rn,jD

d,ona :

rn,d

=

d

j=1

ρ

0,jD

d

+

3is iimpair

ρ

idi

ζ (

i

).

Notons D l’ensemble des diviseursde D. Pour toutdD et pourtout entiern, ona donc unecombinaison linéaire rn,d de1,

ζ (

3

)

,

ζ (

5

)

,. . . ,

ζ (

s

)

.Pour toutentieri impaircompris entre3ets,lecoefficientde

ζ (

i

)

est

ρ

idi (oùlenombre rationnel

ρ

i dépendimplicitementden,maispasded).Sidesentierswd,dD,vérifient

d∈Dwddi=0 pouruncertain i,alorslacombinaisonlinéaire

rn

=

d∈D

wdrn,d (0.1)

ne faitplus apparaître

ζ (

i

)

.Le point central dece procédé d’élimination(mis au point par letroisièmeauteur [12] pour D=2,etgénéraliséparledeuxièmeauteur[10])estquelecoefficientdidépendded,maispasden:onpeutchoisirles entiers wd indépendammentden.

Pour démontrer le théorème 1, on suppose que, parmi

ζ (

3

)

,

ζ (

5

)

, . . . ,

ζ (

s

)

, le nombre d’irrationnels est inférieur à 2(13ε)log loglogss

;ilestalorsinférieurouégalà

δ

1,

δ

=CardDestlenombredediviseursdeD.Ilexistedoncdesindices impairsi1

<

i2

< . . . <

1 comprisentre3etstelsquetoutevaleurirrationnelle

ζ (

i

)

,aveci impaircompris entre3ets, soitl’unedes

ζ (

ij

)

.Onpeutchoisirdesentiersrelatifs wd nontousnuls(pourdD)telsque

d∈D

wddij

=

0 pour toutj

∈ {

1

, . . . , δ

1

}.

(0.2)

(3)

Larelation(0.1) définitalorsunesuite

(

rn

)

n1denombresréelsquisontdescombinaisonslinéairesàcoefficientsrationnels de 1 etdes

ζ (

i

)

pour i impair,3≤is.La propriété cruciale estque les

ζ (

ij

)

nefigurent plus dansces combinaisons linéaires,sibienquetousles

ζ (

i

)

quiyapparaissentavecuncoefficientnonnulsont,parhypothèse,desnombresrationnels.

En multipliant par un dénominateur commun A de ces nombres rationnels, et aussi par dns++11 (quiest un dénominateur commun des coefficients

ρ

i et

ρ

0,j), on obtientune suite d’entiers relatifs.Une étude asymptotique des suites

(

rn,j

)

n1 montrequecettesuited’entiersrelatifstendvers0,doncestidentiquementnulleàpartird’uncertain rang ;précisément, ona,quandn→ ∞:

Adns++11

rn

=

d∈D

wdd

+

o

(

1

)

Adns++11rn,1

,

avec 0

<

lim

n→∞

Adsn++11rn,1

1/n

<

1

o

(

1

)

est une suite qui tend vers 0 quand n tend vers l’infini. Cela impose

d∈Dwdd=0. Autrement dit, tout fa- mille

(

wd

)

d∈D d’entiers relatifs vérifiant (0.2) devrait vérifier aussi

d∈Dwdd=0. Ce n’est pas le cas, car la matrice [dij]d∈D,0jδ1 (dans laquelleon posei0=1) estinversible : sondéterminant estleproduitd’un déterminantde Van- dermondeetd’unpolynômedeSchur,quiestunpolynômeàcoefficientsentiersnaturelsévaluéenlafamilledesdiviseurs de D.

1. Introduction

When s2 is an eveninteger, thevalue

ζ (

s

)

oftheRiemannzetafunction isa non-zerorationalmultipleofπs and, therefore,atranscendentalnumber.Ontheotherhand,veryfewresultsare knownonthezetavalues

ζ (

s

)

whens3 is odd,thoughweexpectthemalltobetranscendental.

It was onlyin 1978whenApéry astonished themathematicscommunity by hisproof [1] ofthe irrationalityof

ζ (

3

)

, withthenextbreakthroughinthisdirectiontakenin2000byBallandRivoal[2,8],whoprovedthefollowingtheorem.

Theorem1(Ball–Rivoal).Let

ε >

0,ands3beanoddintegersufficientlylargewithrespectto

ε

.Then dimQSpanQ

(

1

, ζ (

3

), ζ (

5

), ζ (

7

), . . . , ζ (

s

))

1

ε

1

+

log 2logs

.

Inspiteof severalrefinements[2,11,4] for smalls,thelower bound inTheorem 1hasnever beenimprovedforlarge valuesofs.TheproofofTheorem1appliesNesterenko’slinearindependencecriterion[7] tocertainlinearcombinationsof oddzetavalues.Toimproveonthisboundusingthesamestrategy,onehastofindlinearcombinationsthatareconsiderably smaller,withnottoolargecoefficients — thiscomesouttobearatherdifficulttask.

Thesituationhasdrasticallychangedwhenthethirdauthorintroduced[12] anewmethod,whichhasbeengeneralized bythesecondauthorin[10].Foragiveninteger D

>

1 andacertainrationalfunction R

(

t

)

,theseries

d

j=1

t=1 R

t

+

j

d

,

whered

|

D

,

givesQ-linearcombinationsof1

,

d3

ζ (

3

),

d5

ζ (

5

), . . . ,

ds

ζ (

s

)

withcoefficientsindependentofd(exceptthatof1).Thismakes itpossibletoeliminate fromtheentirecollectionoftheselinearcombinationsessentiallyasmanyoddzetavaluesasthe numberofdivisorsofD.Forapplicationsofthisidea,see[12,6,10,9].

In thisnote, we sketch the proof of the following result, buildingupon the approach in[12] and[10]. We refer the interestedreadertothefullversion[5] ofthepaperfordetails.

Theorem2.Let

ε >

0,ands3beanoddintegersufficientlylargewithrespectto

ε

.Thenamongthenumbers

ζ (

3

), ζ (

5

), ζ (

7

), . . . , ζ (

s

),

atleast

2(1ε)log loglogss areirrational.

Incomparison,Theorem 1givesonly 1ε

1+log 2logsirrational oddzetavalues,buttheyarelinearlyindependentoverthe rationals,whereasTheorem2endsuponlywiththeirirrationality.

OurproofofTheorem2followstheabove-mentionedstrategyfrom[12] and[10].Themainnewingredient,compared to the proof in [10], is taking D large (about s1ε) and equal to the product ofthe first prime numbers (the so-called primorial) — suchanumberhasasymptoticallythelargestpossiblenumberofdivisorswithrespecttoitssize.

(4)

2. SmalllinearformsinvaluesoftheHurwitzzetafunction

Let sand D be positiveintegers suchthat sisodd ands3D.Letnbe apositive integer,such that Dniseven. We considerthefollowingrationalfunction:

Rn

(

t

) =

D3Dnn

!

s+13D

3Dn

j=0

(

t

n

+

Dj

)

n

j=0

(

t

+

j

)

s+1

which,ofcourse,dependsalsoonsandD,andfor j∈ {1

, . . . ,

D}welet rn,j

=

m=1

Rn

m

+

j

D

.

We write dn for lcm

(

1

,

2

, . . . ,

n

)

.Expanding Rn

(

t

)

into partial fractionsyields thefollowing Q-linearcombinations of valuesoftheHurwitzzetafunction.

Lemma1.Foreach j∈ {1

, . . . ,

D},wehave

rn,j

= ρ

0,j

+

3is i odd

ρ

i

ζ

i

,

j D

with

dns+1i

ρ

i

∈ Z

for i

∈ {

3

,

5

, . . . ,

s

}

and dsn++11

ρ

0,j

∈ Z

for any j

∈ {

1

, . . . ,

D

}.

Wepointoutthatthecoefficient

ρ

iof

ζ (

i

,

Dj

)

inthelinearformrn,j doesnotdependon j.Theexpansionofrn,j isvery classical; onlyoddzetavaluesappearbecauseofthesymmetryphenomenonfrom[2].Theproofthatdsn+1i

ρ

i∈Zforodd ifollowsthatof[3,Lemma 4.5].Thelastassertion,namelydns++11

ρ

0,j∈Z,isprovedasin[10,Lemma 1.4].

An important feature ofthe linear formsrn,j is that they are simultaneouslyvery small: even when multipliedby a commondenominatoroftherationalcoefficients,theystilltendto0.

Lemma2.AssumethatDlogs D largerthansomeeffectivelycomputableabsoluteconstant.Thenwehave

nlim→∞rn1/,nj

=

g

(

x0

) <

3−(s+1) and lim

n→∞

rn,j

rn,j

=

1 for any j

,

j

∈ {

1

, . . . ,

D

},

where

g

(

x

) =

D3D

(

x

+

3

)

3D

(

x

+

1

)

s+1

(

x

+

2

)

2(s+1)

andx0istheuniquepositiverootofthepolynomial

(

X+3

)

D

(

X+1

)

s+1XD

(

X+2

)

s+1.

The followingelementarylemmaaboutthenon-vanishingofgeneralizedVandermondedeterminantsisusedtogetrid ofunwantedirrationaloddzetavalues(in§3below).

Lemma3.Fort1,letx1

, . . . ,

xtbepairwisedistinctpositiverealnumbersand

α

1

, . . . , α

tpairwisedistinctnon-negativeintegers.

Thenthematrix[xαji]1i,jthasnon-zerodeterminant.

3. Eliminationofoddzetavalues

Take0

< ε <

13,andletsbeoddandsufficientlylargewithrespectto

ε

.WeletD betheproductofallprimeslessthan orequalto

(

1−2

ε )

logs,sothat Ds1ε.Noticethat D has

δ

=2π((12ε)logs)divisors,withlog

δ

(

1−3

ε )(

log 2

)

log loglogss. Assume that thenumberofirrational oddzetavaluesbetween

ζ (

3

)

and

ζ (

s

)

islessthanorequal to

δ

1.Let3=i1

<

i2

< . . . <

iδ−1sbe oddintegers such that if

ζ (

i

)

/

Q andi isodd, 3is, then i=ij for some j.Moreover, we let i0=1,andconsiderthesetDofalldivisorsofD,sothat CardD=

δ

.Lemma3showsthat thematrix[dij]d∈D,0j≤δ−1 is invertible.Therefore,thereexistintegers wd∈Z,wheredD,suchthat

d∈D

wddij

=

0 for any j

∈ {

1

, . . . , δ

1

}

and

d∈D

wdd

=

0

.

(3.1)

(5)

Thecrucialpoint(asin[10,§3])isthatforanydDandanyi2,

d

j=1

ζ

i

,

j

D d

D

=

d

j=1

ζ

i

,

j d

=

n=0

d

j=1 di

(

dn

+

j

)

i

=

di

ζ (

i

).

Lemma1impliesthatthequantitiesrn,d=d j=1rn,jD

d arelinearformsintheoddzetavalues:

rn,d

=

d

j=1

ρ

0,jD

d

+

3is iodd

ρ

idi

ζ (

i

),

(3.2)

whileLemma2leadstothefollowingasymptotics:

rn,d

= (

d

+

o

(

1

))

rn,1 with lim

n→∞r1n,/1n

=

g

(

x0

).

(3.3)

Weshallusenowtheintegers wd toeliminate

δ

1 oddzetavalues,includingallirrationalones, asin[12] and[10].

Forthat,introducern=

d∈Dwdrn,d.Eqns. (3.1)–(3.3) implythat

rn

=

d∈D wd

d

j=1

ρ

0,jD d

+

iI

ρ

i d∈D

wddi

ζ (

i

) =

d∈D

wdd

+

o

(

1

)

rn,1

,

where I= {3

,

5

,

7

, . . . ,

s}\ {i1

, . . . ,

iδ−1}.In particular,no irrationalzetavalue

ζ (

i

)

,where3≤is,appearsin thislinear combination, and limn→∞|rn|1/n=g

(

x0

) <

3−(s+1) since

d∈Dwdd=0. Denoting by A a common denominator ofthe (rational)numbers

ζ (

i

)

,iI,wededucefromLemma1that Adsn++11rnisaninteger.Nowtheprimenumbertheoremimplies thatdn1/+n1easn→ ∞,hencefromLemma2weconcludethatthesequenceofintegerssatisfies

0

<

lim

n→∞

|

Adsn++11

rn

|

1/n

=

es+1g

(

x0

) <

e 3

s+1

<

1

.

ThiscontradictionimpliesthetruthofTheorem2.

Acknowledgements

WethankMichelWaldschmidtforhisadvice,andOleWarnaarforhiscommentsonanearlierdraftofthepaper.

References

[1]R.Apéry,Irrationalitédeζ (2)etζ (3),in:JournéesArithmétiques,Luminy,1978,in:Astérisque,vol. 61,1979,pp. 11–13.

[2]K.Ball,T.Rivoal,Irrationalitéd’uneinfinitédevaleursdelafonctionzêtaauxentiersimpairs,Invent.Math.146 (1)(2001)193–207.

[3]S.Fischler,Shidlovsky’smultiplicityestimateandirrationalityofzetavalues,preprint,arXiv:1609.09770 [math.NT],J.Aust.Math.Soc.(2018),inpress.

[4]S.Fischler,W.Zudilin,ArefinementofNesterenko’slinearindependencecriterionwithapplicationstozetavalues,Math.Ann.347(2010)739–763.

[5]S.Fischler,J.Sprang,W.Zudilin,Manyoddzetavaluesareirrational,preprint,arXiv:1803.08905 [math.NT],2018.

[6]C.Krattenthaler,W.Zudilin,Hypergeometryinspiredbyirrationalityquestions,preprint,arXiv:1802.08856 [math.NT],2018.

[7]Y.Nesterenko,Onthelinearindependenceofnumbers,Vestn.Mosk.Univ.,Ser.Filos.(Mosc.Univ.Math.Bull.)40 (1)(1985)46–49(69–74).

[8]T.Rivoal,LafonctionzêtadeRiemannprenduneinfinitédevaleursirrationnellesauxentiersimpairs,C.R.Acad.Sci.Paris,Ser.I331 (4)(2000) 267–270.

[9]T.Rivoal,W.Zudilin,Anoteonoddzetavalues,preprint,arXiv:1803.03160 [math.NT],2018.

[10]J.Sprang,Infinitelymanyoddzetavaluesareirrational.Byelementarymeans,preprint,arXiv:1802.09410 [math.NT],2018.

[11]W.Zudilin,IrrationalityofvaluesoftheRiemannzetafunction,Izv.Akad.NaukSSSR,Ser.Mat.(Izv.Math.)66 (3)(2002)49–102(489–542).

[12]W.Zudilin,Oneoftheoddzetavaluesfromζ (5)toζ (25)isirrational.Byelementarymeans,SIGMA14 (028)(2018)(8pages).

Références

Documents relatifs

Consequently the Multiple Zeta Values satisfy many linear relations with rational coefficients.. A complete description of these relations would in principle settle the problem of

Ap´ ery of the irrationality of ζ(3) in 1976, a number of articles have been devoted to the study of Diophantine properties of values of the Riemann zeta function at positive

By Poincaré duality (loc. cit.) the pairing in the second line is nondegenerate, and therefore the pairing in the first line is too... PROOF: The first equality is

Schur [11] has shown that every nonconstant polynomial f has an infinite number of prime divisors... The values and the

NESTERENKO, Arithmetic properties of values of the Riemann zeta function and generalized hypergeometric functions. NIKISHIN, On irrationality of values of functions

In Section 4, we will give an original explicit expression of the constant C (F, β). To this end, we recall in Section 3 results of Dwork [5], André [1] and of the author [10] on

E, LITTLEWOOD, Proc.. Actually two groups of new terms emerge, the number of terms in each group depending on T, so that the result falls into the category of

This allows us to obtain a new proof of the Ball-Rivoal theorem on irrational- ity of infinitely many values of Riemann zeta function at odd integers, inspired by the proof of