• Aucun résultat trouvé

Zeros of the deformed exponential function

N/A
N/A
Protected

Academic year: 2022

Partager "Zeros of the deformed exponential function"

Copied!
38
0
0

Texte intégral

(1)

Contents lists available atScienceDirect

Advances in Mathematics

www.elsevier.com/locate/aim

Zeros of the deformed exponential function

Liuquan Wanga, Cheng Zhangb,∗

aSchoolofMathematicsandStatistics,WuhanUniversity,Wuhan430072,Hubei, People’sRepublicofChina

bDepartmentofMathematics,JohnsHopkinsUniversity,Baltimore,MD21218, UnitedStates

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

Articlehistory:

Received7January2018 Receivedinrevisedform4April 2018

Accepted20April2018 Availableonline26May2018 CommunicatedbyGeorgeAndrews

MSC:

primary30C15,11B83,41A60 secondary11M36,34K06

Keywords:

Deformedexponentialfunction Asymptoticexpansion Eisensteinseries Bernoullinumbers

Letf(x)=

n=0 1

n!qn(n−1)/2xn(0< q <1)bethedeformed exponentialfunction.Itisknownthat thezerosof f(x) are realandforma negativedecreasing sequence(xk) (k 1).

Weinvestigatethecompleteasymptoticexpansionforxkand provethatforanyn1,ask→ ∞,

xk=kq1−k 1 +

n i=1

Ci(q)k−1−i+o(k−1−n) ,

where Ci(q) are some q series which can be determined recursively.WeshowthateachCi(q)Q[A0,A1,A2],where Ai=

m=1miσ(m)qmandσ(m) denotesthesumofpositive divisors of m. When writing Ci as a polynomial inA0,A1

andA2,wefind explicitformulasfor thecoefficients of the linearterms byusing Bernoullinumbers.Moreover,wealso provethat Ci(q) Q[E2,E4,E6],whereE2,E4 andE6 are theclassicalEisensteinseriesofweight2,4and6,respectively.

©2018ElsevierInc.Allrightsreserved.

* Correspondingauthor.

E-mailaddresses:wanglq@whu.edu.cn,mathlqwang@163.com(L. Wang),czhang67@jhu.edu (C. Zhang).

https://doi.org/10.1016/j.aim.2018.05.006 0001-8708/©2018ElsevierInc.Allrightsreserved.

(2)

1. Introduction

Considerthefunction

f(x) :=

n=0

xn

n!qn(n−1)/2 (1.1)

where x,q C, |q| 1. The function f(x) is entire and is called the “deformed ex- ponential function” since it reduces to exp(x) when q = 1. It appears naturally and frequentlyinpureandappliedmathematics.Incombinatorics,thefunctionf(x) relates closelyto thegeneratingfunctionforTuttepolynomialsof thecomplete graphKn [24], theenumerationofacyclicdigraphs[18] andinversionsoftrees[15].Italsorelatestothe Whittaker andGoncharovconstants [2] incomplexanalysis,andthepartitionfunction ofone-sitelatticegaswithfugacityxandtwo-particleBoltzmann weightqinstatistical mechanics [19].Moreover,onecanverifythatthisfunctionistheuniquesolutiontothe functional differentialequation

y(x) =y(qx), y(0) = 1, (1.2)

which is aspecialcase ofthe “pantographequation” [6]. Formoredetaileddiscussions onthis function,wemayrefertothenotesfrom AlanSokal’stalks [20].

Surprisingly, many important propertiesof this functionremain open, e.g.,the dis- tribution of its zeros. In 1952, Nassif [17] studied (on Littlewood’s suggestion) the asymptotic behavioursandthezerosoftheentirefunction

n=0

en2

2πiz2n/n!,

which equalsto f(q12z2) withq=e22πi. Heused thefactthat

2 hasaperiodiccon- tinuedfractionexpansion.Later,Littlewood[11,12] consideredgeneralizationstoTaylor serieswhosecoefficientshavesmoothlyvaryingmoduliandargumentsoftheformen2απi, whereαisaquadraticirrationality.Seealso[4,10,13,23] forthestudiesonthebehaviours of these functions. To ourknowledge,for generalcomplex numberq satisfying|q| 1, the distribution of the zeros of f(x) has not been completely understood up to now.

A theorem of Eremenko cited in[21] considered the case where q lies in any compact set of theopen unit disk D. There are relatively moreworks on themodel casewhere 0< q <1.In1972,Morris et al.[5] usedatheoremof Laguerreto showthatf(x) has infinitelymanyreal zerosandthese zerosareallnegative andsimple.Theyalsoproved thatthere isnootherzerofortheanalyticextension (tothecomplexplane)off(x) by using theso-calledmultipliersequence(amodestgapintheirproofwasfilledbyIserles [8]).Therefore, when0< q <1,thezerosoff(x) formonestrictly decreasingsequence ofnegativenumbers(xk) (k1).Weremarkthatinsomepreviousworks(e.g.,[9,22]),

(3)

thesubscriptsofthesequencestartwith0ratherthan1.Inthispaper,aswellasin[25], thesubscriptsstartwith1fortheelegance ofnotation.

When0 < q < 1,some conjectures on thezeros xk (k 1) have been proposed in [5,8,18]. Forexample,Morris etal.[5] conjecturedthat

k→∞lim xk+1

xk = 1

q. (1.3)

In1973,Robinson[18] alsoderived(1.2) whencountingthelabeledacyclicdigraphs.He speculatedthat

xk=−kq1k+o(q1k). (1.4) These conjectures have been investigated by several authors (see e.g., [7,9,14,22]). In particular,Langley [9] showedthatask→ ∞

xk+1

xk = 1 q

1 + 1 k

+o(k−2). (1.5)

Healso provedthatthere exists apositiveconstantγ, which isindependent ofk, such that

xk=−kq1−k(γ+o(1)). (1.6) Asaconsequence,(1.3) istrue.Recently,oneoftheauthors[25] refinedLangley’swork andconfirmedtheobservation(1.4).Indeed,heshowedthatask→ ∞,

xk =−kq1k 1 +

m=1

σ(m)qmk2+o(k2)

. (1.7)

Hereforany positiveintegern,

σ(n) :=

d|n, d>0

d.

LaterDerfeletal.[3] studiedtheasymptoticbehavioursofthezerosofsolutionsof(1.2) withdifferentinitial conditionsinsteadoftherestrictiony(0)= 1.

Ourresearch onthe zerosof thedeformed exponential functionis motivated bythe conjectures introduced by Sokal [21] in his talk at Institut Henri Poincaréin 2009. In this paper, we obtaina complete asymptotic expansionformula for the zerosxk when 0< q <1.To bemorespecific, wewill approximatexk withremainderterm o(k−n−1) for any n≥1.We alsoestablish the connectionbetween theclassical Eisenstein series and thezerosof the deformedexponential function.To stateour results,we define for i≥0,

(4)

Ai=Ai(q) :=

m=1

miσ(m)qm. (1.8)

Theorem 1.Let0< q <1andn≥1.Thenask→ ∞, xk=−kq1k

1 + n i=1

Ci(q)k1i+o(k1n)

, (1.9)

where each Ci(q) is a multivariate polynomial in A0,A1,...,Ai−1 with rational coeffi- cients.These polynomials canbe determinedrecursively.

Remark 1.For example, C1 = A0, C2 = −A1, C3 = 101A0+ 35A1 + 12A2 1310A20. Therecurrencerelationandthebasicstructureofthesepolynomialswillbepresentedin Sections3and4.Moreover,weconjecturethat(1.9) shouldholdforanycomplexnumber qsatisfying0<|q|<1 (andevenforsomeqsatisfying|q|= 1),ifthecomplexzerosare listedaccordingtotheirmultiplicitiesandorderedbyincreasingmodulus.Furthermore, itisinterestingtoconsiderifonecouldobtainsimilarresultsforthezerosoftherescaled Rogers–Ramanujan function[22]

R(x;˜ y, q) = n=0

xnqn(n1)/2

(1 +y)(1 +y+y2)· · ·(1 +y+· · ·+yn−1),

whichreduces toa“partialthetafunction”wheny= 0, andthe“deformedexponential function”when y= 1.

Whenn≥4,weobservethattheexpressionofCn(q) intermsofA0,A1,. . . ,An−1is not unique. The polynomialgiven by therecurrence relation inTheorem 1is just one candidate.Forexample,wehave

C4= 1

10A114 15A21

6A3+23 5A0A1

= 1

10A111

10A2+23

5A0A16A21+ 4A0A2. (1.10) Thus we continueto study therelations betweenAi’s. Indeed,the followingidentity is established

A3=A2+ 36A2124A0A2.

DifferentiatingitwithrespecttoqgivesmoresimilaridentitiesonAi’s.Therefore,wefind thatitispossibletoexpressCnasapolynomialinjustA0,A1andA2.Furthermore,the coefficientsofthelineartermsinthatpolynomialcanbegivenexplicitlyusingBernoulli numbers.LetBn bethen-thBernoullinumber.ItiswellknownthatB2m+1= 0 forall m≥1.ThefirstfewvaluesofBi areB0= 1,B1=12, B2= 16 andB4=301.

(5)

Theorem2.Foranyn≥1,Cn canbeexpressedasatrivariatepolynomialinA0,A1and A2 with rational coefficients.Thispolynomial isuniqueandforn≥2,wehave

C2n1=6B2n

n A036B2n

n A1+

1 + 30B2n

n

A2+ higher degree terms, C2n =6B2n

n A1+ 6B2n

n 1

A2+ higher degree terms.

Remark2.Forexample,wefindthat C5=1

21A02

7A1+26

21A2+53

70A20+ 22A2136A0A21

159

35 A0A143

2 A0A2+ 2A1A2+737

210A30+ 24A20A2, C6= 1

21A120

21A274

35A0A11401 35 A212

5A22+705 14 A0A2

101

10 A1A2+1662

5 A0A21321

14 A20A136 5A31

1132

5 A20A2864

5 A20A21+72

5A0A1A2+576 5 A30A2.

TheBernoulli numbersinthelinearterms arefrom Faulhaber’s formulafor thepower sumof the first m positiveintegers(see (2.49)). As showninthe examplesabove, the higherdegreetermslookmuchmorecomplicatedthanthelinearterms,thoughtheycan be explicitlyderivedfrom therecurrencerelation ofCn (see (4.1)),whichis essentially determined by the expressions of the unsigned Stirling numbers of the first kind (see (2.24),(2.46)).

Let

E2=E2(q) := 124 n=1

nqn

1−qn, (1.11)

E4=E4(q) := 1 + 240 n=1

n3qn

1−qn, (1.12)

E6=E6(q) := 1504 n=1

n5qn

1−qn. (1.13)

It is well knownthatE2,E4 and E6 are classicalEisenstein series on thefull modular group

SL(2,Z) =

a b c d

|ad−bc= 1, a, b, c, dZ .

(6)

We willshow thatA0,A1 andA2 canbe representedas polynomialsinE2,E4 and E6

with rationalcoefficientsand viceversa.Namely,

Q[A0, A1, A2] =Q[E2, E4, E6].

Since it is well known that E2, E4 and E6 are algebraically independent over C (see e.g.,[16, Lemma117]),itfollowsthatA0,A1andA2 arealsoalgebraicallyindependent over C. So Theorem 2 implies that Cn(q) are in the ring of quasimodular forms on SL(2,Z).

Corollary 1.Letn≥1.ThenCn(q)Q[E2,E4,E6].

The paper is organized as follows. In Section 2, we study the series expansion of f((k+ak1)q1k) forlarge k,where a=a(k1) is positiveandbounded forlargek.

Hereonecanthinkof(k+ak−1)q1−kasa“prospectiveroot”off.Weobservethatthe first2ktermsofthisseriesdominatetheothers,sothese2ktermsarecarefullyanalyzed in Lemmas 1 and 2. Most of these terms enjoy niceproperties described inLemma 1, while others are more subtle and related to the series expansion of (2.17). Lemma 2 givestheformulaofthecoefficientofeachtermintheseriesexpansionof(2.17).Inthe proof of Lemma 2, the properties of elementary symmetric polynomials and complete homogeneoussymmetricpolynomials playanimportantrole.

InSection3,weproveTheorem1.WefirstdefineCn recursively(see(3.3))byexploit- ingthecoefficientsoftheseriesexpansionof(2.17).Leta=n−1

i=1 Cik−(i−1)+λk−(n−1). Then weuseLemmas1and2toshowthatifλ=Cn,and inadditionthatλ>0 when n= 1,then

(1)k−Cn)f((k+ak−1)q1−k)>0.

This allows usto determine thesigns off(x) attheendpoints of certain intervals. We finish theproof ofTheorem1bytheintermediatevaluetheorem and(1.5).

InSection4,westudyvariousrepresentationsofCn.InSection4.1,wegivemorede- tailsontherecursiveformulaofCn,andprovethatCn canbe writtenasamultivariate polynomial of A0,A1,. . . ,An1, in which the coefficient of An1 is (−1)(nn1)!1 (Proposi- tion5).InSection4.2,wecontinuetodiscussthestructureofthe multivariatepolynomial representation of Cn by those Ai’s, especiallythe linearterms. In particular, we show inProposition6thatthesumofthecoefficientsofthelineartermsA0,A1,. . . ,An1 in Cn equalsto (1)n−1.Furthermore, wegiveexplicitformulasforthecoefficientsof the linear terms A0 and A1 inPropositions 7and 8. InSection 4.3, we first establish the relations between theclassicalEisenstein series and ourAi’s (Proposition9).Then we show inLemma10thateachAn canbewrittenas amultivariate polynomialinA0,A1 and A2.Combiningthesefacts,wecomplete theproof ofTheorem2.

(7)

Remark3.It may be illuminating to consider (1.9) as aformal powerseries. Suppose thatCi(q)=

j=1Cijqj. Using formalpower series, wedenote Fj(t):=

i=1Cijti+1. Thenonemayrewrite(1.9) asaformalpowerseries

xk(q)∼ −kq1k 1 +

j=1

Fj(k1)qj

. (1.14)

Weobservethattheformalpowerseries(1.14) numericallyagreeswiththeexpansionin q ofthe k-th zerogiven in[21,p. 14]. (Note thatin[21] thesequence (xk) starts with subscript 0and foreach j, Fj(k−1) is arational functionin k.) It was conjecturedby Sokal[21, p. 11] that

Fj(k−1)0

forallintegersj,k≥1.Thisconjectureisstill open,evenforfixed k= 1.Inparticular, wewillseethatCi1= (1)i+1 byProposition6,whichimpliesthat

F1(k1) = 1 k(k+ 1).

This verifies thecasej = 1.For j 2,we find thatthedifficulty to obtainthe closed formofFj(k1) liesinthecomplexityofthehigherdegreetermsinTheorem2.

2. Preliminaryresults

Throughout this paper, we fixq with 0< q <1. Weuse thenotation “O(km)”to denote the class of functions of k which is bounded by Ck−m, where C is a constant thatis dependenton the fixed parametersbut independent of k. Let a=a(k−1) be a functioninksatisfyingthata=a0+O(k−1) wherea0>0 isaconstant.Themaingoal ofthissectionisto studythevaluesof f((k+ak1)q1k) forlargek.

Wefirstobservethat

f((k+ak1)q1k) = n=0

(1)nun, (2.1)

where

un =(k+ak−1)n

n! qn(2kn1)/2.

Inorderto proveTheorem1, wewill selectsomespecialfunctionsasa(seeSection3).

For these a and sufficiently large k, we will see that for thesum in (2.1), the first 2k termsdominatetheothers. Therefore,werewrite(2.1) as

(8)

f((k+ak−1)q1−k) =

2k1 n=0

(1)nun+ n=2k

(1)nun =

k1

j=0

(1)j−1vj+ n=2k

(1)nun, (2.2) where wedenote

vj =u2k−j−1−uj (0≤j≤k−1).

Thefollowinglemmashowsthepositivityand“almostmonotonicity”ofthesequencevj. Lemma 1.Thereexistsapositiveinteger K(q)suchthat foranyk≥K(q),

vj >0, 0≤j≤k−1.

Furthermore, there existsa positiveinteger N(q) suchthat forany N ≥N(q) and k≥ q3N,

vj< vj+1, 0≤j≤k−N.

Proof. Sincea=a0+O(k−1) anda0>0,wecanfindapositive integerK(q) suchthat a>0 forany k≥K(q).Nowwe assumethatk≥K(q).BytheAM-GMinequality,we have

2k−1−2j

i=1

(j+i)<

⎜⎜

⎜⎝

2k−1−2j

i=1

(j+i) 2k12j

⎟⎟

⎟⎠

2k−1−2j

=k2k12j

<(k+ak−1)2k−1−2j (0≤j≤k−1). (2.3) This implies

(k+ak1)j

j! < (k+ak1)2k1j

(2k1−j)! (0≤j≤k−1).

So

uj< u2k1j (0≤j≤k−1), whichgivesthefirstinequality.

Note that

vj+1 =q(j+1)(j+2−2k)/2(k+ak1)2kj2

(2k2−j)! (1−wj) (2.4)

(9)

where

wj= (2k2−j)!

(j+ 1)!(k+ak−1)2k−3−2j. Sincewehaveprovedthatvj+1>0,itfollows that

0< wj<1 BytheAM-GMinequality,

wj+1 wj

= (k+ak1)2

(j+ 2)(2k2−j)>1.

By(2.4),wesee thatvj < vj+1is equivalentto qj+1k(1−wj)> k+ak−1

2k−j−1(1−wj−1). (2.5) Usingtherelation

wj−1= (j+ 1)(2k−j−1) (k+ak−1)2 wj, weseethat(2.5) isequivalentto

qk+j+1 j+ 1 k+ak1

wj < qk+j+1 k+ak−1

2k1−j. (2.6) ForsomepositiveintegerN,wedenote

t= 2k1−j (k+N 1≤t≤2k1) and

g(t) = 1

t (2k−t)wkN

(k+ak1)2 −qk−t

1−wkN

k+ak−1

. Directcalculationyields

g(t) =1

t2 + wk−N (k+ak1)2 +

1−wk−N k+ak−1

qk−tlnq (2.7) and

g(t) = 2 t3

1−wk−N k+ak−1

qkt(lnq)2. (2.8)

(10)

Since 0< wj<1,g(t) isdecreasingfort>0.

NotethatwhenNislargeenough(N ≥N1(q)),wehavek≥q3Nmax{K(q),N6}. Hence

1

k+N−1 =k1

1(N1)k1+ (N1)2k2+O(N3k3)

=k−1(N1)k−2+ (N1)2k−3+O(k−7/2). (2.9) Similarlywehave

1

(k+N−1)2 =k22(N1)k3+O(k11/3) (2.10) and

1

(k+N−1)3 =k−3+O(k−23/6). (2.11) Next,

wk−N =

N−2

t=−N+2

k+t k+ak−1

=

N−2

t=−N+2

1 + t

k 1−a0

k2 +O(k3)

=

N−2

t=−N+2

1 + t

k−a0 k2 −a0t

k3 +O(k−3)

= 1−a0(2N3)k−2+

−N+2≤t1<t2≤N−2

t1t2

k2 +O(k−17/6)

= 1−a0(2N3)k2(N2)(N1)(2N3)

6 k2+O(k17/6).

Hence

1−wkN

k+ak1 =1 k

1−a0k−2+O(k−3)

·

a0(2N3)k2+(N2)(N1)(2N3)

6 k2+O(k17/6)

=

a0(2N3) + (N2)(N1)(2N3) 6

k3+O(k23/6). (2.12) Note thatk≥q3N impliesqN =O(k1/3).Nowby(2.11) and(2.12),wededucethat

(11)

g(k+N−1) =

2−cNq1−N(lnq)2

k−3+O(k−7/2) where

cN :=a0(2N3) +1

6(N2)(N1)(2N3).

Similarly,wehave

g(k+N−1) =

2N2 +cNq1−Nlnq

k−3+O(k−7/2).

WhenN issufficientlylarge(N ≥N2(q)≥N1(q)),wewillhave g(k+N−1)<0,

and

g(k+N−1)<0.

Sog(t) andg(t) arealsodecreasingfort≥k+N−1.

Inthesameway,wefindthat g(k+N−1) =

a0(2N1) + N

6(N1)(2N1)−cNq1−N

k−3+O(k−10/3).

WhenN islargeenough(N≥N(q)≥N2(q)),we have g(k+N−1)<0.

SoifN ≥N(q) andk≥q3N,wehave

g(t)≤g(k+N−1)<0, t≥k+N−1.

Therefore, 1

2k1−j (j+ 1)wk−N

(k+ak−1)2 −qk+j+1

1−wk−N k+ak1

<0 (0≤j≤k−N).

Thisimplies

q−k+j+1 j+ 1 k+ak−1

wkN< q−k+j+1 k+ak1

2k1−j (0≤j ≤k−N).

Sincewj< wj+1,we have

qk+j+1 j+ 1 k+ak−1

wj < qk+j+1 k+ak1

2k1−j (0≤j≤k−N).

(12)

This proves(2.6) andhencethefactthat

vj< vj+1 (0≤j≤k−N). 2

However,thesequencevj maynotbemonotonewhenk−N < j≤k−1.Soweneed moredelicateanalysisonthese vj’s, whichis thecruxoftheproblem.For1≤j≤N,

vkj=uk+j1−ukj =

(k+ak1)k+j1

(k+j−1)! (k+ak1)kj (k−j)!

q−(k+j−1)(k−j)/2

=

j1

i=1

k+ak1 k+i

0 i=1j

k+i k+ak−1

⎠(k+ak−1)k 1

k!q−(k+j−1)(k−j)/2

=

j1

i=0

1 +ak2 1 +ik−1

−1 i=1j

1 +ik1 1 +ak−2

⎠1 +ak2 k−2

·qj(j1)/2·(k+ak1)k21

k!qk(k1)/2. (2.13)

Wedefine forj≥1,

G(α;x) :=

j1 i=0

1 +αx2

1 +ix , (2.14)

H(α;x) :=

1

i=1−j

1 +ix

1 +αx2. (2.15)

In particular, when j = 1 we haveG(α;x)= 1+αx2 and H(α;x)= 1 since we agree thattheemptyproductequals 1.Itisthen clearfrom(2.13) that

vk−j =

G(a;k−1)−H(a;k−1)1 +ak−2

k2 ·qj(j−1)/2·(k+ak−1)k−21

k!q−k(k−1)/2. (2.16) Now weanalyzetheseries expansionoftheproduct

G(α;x)−H(α;x) x2

1 +αx2

(2.17) forαbeing apowerseriesofx.

Lemma 2.Let

α(x) :=

i=0

aixi. (2.18)

(13)

Forn≥1,thecoefficient ofxn1 intheexpansionof G(α(x);x)−H(α(x);x)

x2 (1 +α(x)x2) has theform

μ(an1+S0(n) +S1(n)ν+S2(n)ν2+...+Sn(n)νn), (2.19) whereμ= 2j1,ν=j(j−1),andeachSi(n)isapolynomialof a0,a1,...,an2,which is independentof j and has rational coefficients.In particular,S0(1) =S0(2)= 0 and S1(1)= 16.

TherestofthissectionwillbedevotedtogivingaproofofLemma2.First,wecompute thecoefficientsinthe expansionsG(α;x) and H(α;x) aspowerseries inxregardingα asaparameter.Thatis,

G(α;x) = N=0

GNxN, H(α;x) = N=0

HNxN,

whereGN and HN arepolynomials ofαwithcoefficientsdependingonj.Forexample, G0=H0= 1,

G1=H1=12j(j−1),

G2=241(j1)j(j+ 1)(3j2) +jα, H2=241(j2)(j1)j(3j1) + (1−j)α, G3=481(j1)2j2(j+ 1)(j+ 2)12(j1)j2α, H3=481(j3)(j2)(j1)2j2+12(j1)2jα.

TorepresentGN andHN,wedefineforj 1 andi≥0 that

qi(j) :=ei(1,2,· · ·, j−1), (2.20) Qi(j) := (1)ihi(1,2,· · ·, j−1) (2.21) where

ei(X1, X2,· · ·, Xn) =

1≤n1<n2<···<ni≤n

Xn1Xn2· · ·Xni (2.22) and

hi(X1, X2,· · ·, Xn) =

1n1n2≤···≤nin

Xn1Xn2· · ·Xni (2.23)

(14)

are elementary symmetric polynomials and complete homogeneous symmetric polyno- mials respectively.Note thathereweagreethat

e0(X1, X2,· · ·, Xn) =h0(X1, X2,· · ·, Xn) = 1

so that q0(j) = Q0(j) = 1. By definition it is clear that qi(j) = Qi(j) = 0 if i j.

Moreover, weremarkthatqi(j) isjusttheunsignedStirlingnumbersofthefirstkind

qi(j) =c(j, j−i). (2.24)

It isnotdifficultto seethat k=0

ek(X1, X2,· · ·, Xn)tk = n i=1

(1 +Xit), (2.25)

k=0

hk(X1, X2,· · ·, Xn)tk= n i=1

1

1−Xit. (2.26)

These identitiesimplythefollowing well-knownfundamentalrelation: form≥1, m

i=0

(1)iei(X1,· · ·, Xn)hm−i(X1, X2,· · ·, Xn) = 0. (2.27) Therefore, wehavethefollowing recurrencerelationforQk(j):

Qk(j) = k i=1

qi(j)Qk−i(j), k1. (2.28) Recall thegeneralizedbinomialcoefficientfork∈N,z∈C

z k

= z(z−1)· · ·(z−k+ 1)

k! .

Lemma 3.Wehave

GN =

[N/2]

m=0

G(N, m)αm, N = 0,1,2, ...

where

G(N, m) = j

m

QN−2m(j). (2.29)

Moreover,

(15)

HN =

[N/2]

m=0

H(N, m)αm, N = 0,1,2, ...

where

H(N, m) = (−1)N 1−j

m

qN2m(j). (2.30)

Proof. Since

j−1

i=0

1 1 +ix =

i=0

Qk(j)xk, (2.31)

wehave

G(α;x) = j m=0

j m

αmx2m

k=0

Qk(j)xk = N=0

xN

[N/2]

m=0

j m

QN2m(j)αm. Similarly,

H(α;x) = m=0

1−j m

αmx2m

k=0

qk(j)(−x)k

= N=0

xN

[N/2]

m=0

(1)NqN−2m(j) 1−j

m

αm. 2

Forfixedn,both qn(j) andQn(j) arepolynomialsinj.Hencetheycanbenaturally extendedtobe twofunctionsdefinedonthewholereal line.Thefollowing lemmagives arelationbetweenthese twofunctions.

Lemma4.Forn≥0,wehaveQn(1−t)= (1)nqn(t),t∈R. Proof. Wedenoteforj≥1,

φj(x) :=

j1 i=0

1 1 +ix =

n=0

Qn(j)xn, (2.32)

ψj(x) :=

j1 i=1

(1−ix) = n=0

(1)nqn(j)xn. (2.33) Inparticular,φ1(x)=ψ1(x)= 1.Note that

φj(x) = (1 +jx)φj+1(x) = n=0

Qn(j+ 1) +jQn−1(j+ 1)

xn (2.34)

(16)

where wesetQ−1(t)= 0.Comparingthecoefficientofxn onbothsides,wededucethat forany n≥0,

Qn(j) =Qn(j+ 1) +jQn1(j+ 1). (2.35) Similarly,observingthatψj+1(x)=ψj(x)(1−jx),wededucethatforanyn≥0,

qn(j+ 1) =qn(j) +jqn−1(j) (2.36) where we setq−1(t)= 0.Since(2.35) and(2.36) hold forallj 1 andboth Qn(t) and qn(t) arepolynomialsint,weconcludethatforany t∈R

Qn(t) =Qn(t+ 1) +tQn1(t+ 1), (2.37) qn(t+ 1) =qn(t) +tqn−1(t). (2.38) Now weletQn(t)=Qn(1−t).Then(2.37) implies

Qn(t+ 1) =Qn(t)−tQn−1(t). (2.39) Comparing(2.39) with(2.38),weseethatthepolynomialsQn(t) and(1)nqn(t) satisfy thesamerecurrencerelation.Next,bydirectcomputation,wefindthat

Q0(j) =q0(j) = 1, Q1(j) =−j(j−1)

2 , q1(j) = j(j−1)

2 . (2.40)

Thus

Q1(t) =−q1(t) =−t(t−1) 2 .

Now suppose thatQn−1(t)= (1)n−1qn−1(t) for somen≥2.By(2.38) and(2.39) we deduce that

Qn(t+ 1)−Qn(t) = (1)nqn(t+ 1)(1)nqn(t).

Summingovert from1toj−1,weobtain

Qn(j)−Qn(1) = (1)nqn(j)(1)nqn(1). (2.41) By definition, we have Qn(1) = (1)nqn(1) = 0 for n 1. Therefore, (2.41) implies that Qn(j)= (1)nqn(j) for any j 1.This impliesthat Qn(t)= (1)nqn(t) for any t∈R. 2

(17)

SinceG0=H0,G1=H1,weget G(α(x);x)−H(α(x);x)

x2 =

N=0

(GN+2−HN+2)xN. Hence

G(α(x);x)−H(α(x);x)

x2 (1 +α(x)x2) = N=0

(GN+2−HN+2+ (GN−HN)α(x))xN. (2.42) ToproveLemma2, weneedto computethecoefficientofxn−1 in(2.42).ForN 2m, wedefine

Δ(N, m) :=G(N, m)−H(N, m) = j

m

QN2m(j)(1)N 1−j

m

qN2m(j).

(2.43) Forexample,

Δ(0,0) = Δ(1,0) = 0,

Δ(2,0) = 16j(j−1)(2j1), Δ(2,1) = 2j1,

Δ(3,0) =121(j1)2j2(2j1), Δ(3,1) =12(j1)j(2j1), Δ(4,0) = 2401 (1 +j)j(−1 + 2j)(412j+ 17j210j3+ 5j4),

Δ(4,1) = 241(1 +j)j(−1 + 2j)(23j+ 3j2), Δ(4,2) = 0, Δ(5,0) =14401 (1 +j)2j2(1 + 2j)(1256j+ 61j210j3+ 5j4),

Δ(5,1) =481(1 +j)2j2(1 + 2j)(6−j+j2), Δ(5,2) = 0.

Proposition 1.Let μ = 2j 1 and ν = j(j 1). Then for N 2m the polynomial Δ(N,m)can bewritten as

Δ(N, m) =μ(s0+s1ν+· · ·+skνk), s0, s1, . . . , sk Q, (2.44) wherek≤2N−3m−1

2

.Moreover, if (N,m)= (2,1),then s0= 0.

Inordertoprovethisproposition,weneedthefollowinglemmas.

Lemma 5.Any polynomial of j can be written into a polynomial of μ = 2j 1 and ν = j(j−1), in which the degree of μ in each term is at most 1. Moreover, such a representationisunique.Inorderwords,anypolynomialϕ(j)canbeuniquelywrittenas

(18)

ϕ(j) =s0,0+s0,1ν+...+s0,n0νn0+μ(s1,0+s1,1ν+...+s1,n1νn1).

Furthermore, ifallthecoefficients ofϕ(j)are rationalnumbers, theneach si,lQ. Proof. Notethatμ2= 4ν+ 1.Bythebinomialtheorem wehave

jn=

μ+ 1 2

n

= 1 2n

n k=0

n k

μk

= 2n

[n/2]

k=0

n 2k

μ2k+

[(n1)/2]

k=0

n 2k+ 1

μ2k+1

= 2−n

[n/2]

k=0

n 2k

(4ν+ 1)k+ 2−nμ

[(n−1)/2]

k=0

n 2k+ 1

(4ν+ 1)k. (2.45) This provestheassertionsforthepolynomialjn.

Sinceanypolynomialϕ(j) isalinearcombinationsofjn(n= 0,1,· · ·),weknowthat ϕ(j) can be written inthe desired form. If ϕ(j) has rational coefficients, then clearly eachsi,lQ.

Theuniquenessisclear,sinceotherwisewewillhavearelation ϕ1(ν) +μϕ2(ν) = 0

for some nonzero polynomials ϕ1(j) and ϕ2(j), which is impossible because μ2 = 4ν+ 1. 2

Remark4.Weseefrom(2.45) thatj2n=νn+n2μνn1+ lowerdegreetermsandj2n+1=

1

2μνn+ lowerdegreeterms.

Lemma 6.Fork≥1, qk(j) = 1

2kk!j2k 2k+ 1

3·2k(k1)!j2k1+O(j2k2), (2.46) Qk(j) = (1)k

2kk! j2k+(1)k(2k5)

3·2k(k1)!j2k1+O(j2k2). (2.47) Proof. Wedenote

pm(j) :=

j−1 k=1

km. (2.48)

It isknownthat

Références

Documents relatifs

From the limited evidence presented above, we might conjecture that the perfect numbers have density zero, while the abundant and deficient numbers each make up a positive proportion

There is a period domain D n which classifies polarized Hodge structures of the type correspond- ing to smooth cubic hypersurfaces of dimension n... The period map for cubic surfaces

(Main Theorem) Any asymptotically flat, maximal initial data set (AFB) of the form given in definition 2 in the Introduc- tion satisfying the global smallness assumption B stated in

The set of simple solutions to the unperturbed problem is given by circles of latitude of radius (1 + k 2 0 ) −1/2 with respect to an arbitrarily chosen north pole. Together with

Instead of writing down equations for the Hurwitz space Bel d (X), we enumerate all Bely˘ı maps and effectively compute equations to check for isomorphism between curves.. We saw

The alias modes proposed by Noble et al. Here we only describe how the modes have been interpreted for the purpose of Figure 7. The rep mode denotes a reference belonging to an

the emphasis has been on the investigation of the geometric position and asymptotic distribution of the zeros of functions defined by exponential polynomials of the

Applications to the multiplicity of linear recurring sequences In this section we want to give effective upper bounds for the solutions of the equation Gn = 0 and Gn