ebook img

Asymptotics of Laurent Polynomials of Even Degree Orthogonal with Respect to Varying Exponential Weights PDF

1.2 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Asymptotics of Laurent Polynomials of Even Degree Orthogonal with Respect to Varying Exponential Weights

Asymptotics of Laurent Polynomials of Even Degree Orthogonal with Respect to Varying Exponential Weights 6 K. T.-R. McLaughlin∗ A.H. Vartanian† 0 Departmentof Mathematics Department ofMathematics 0 2 TheUniversity of Arizona University ofCentral Florida n 617N.Santa Rita Ave. P. O. Box161364 a J P. O. Box 210089 Orlando, Florida 32816–1364 3 Tucson, Arizona85721–0089 U. S. A. 1 U.S. A. ] A X. Zhou ‡ C DepartmentofMathematics . h Duke University t a Box 90320 m Durham, North Carolina 27708–0320 [ U.S. A. 1 v 6 11 January2006 0 3 1 0 Abstract 6 0 Let ΛR denote the linear space over R spanned by zk, k Z. Define the real inner product(with ∈ h/ varyingexponentialweights)h···,···iL: ΛR×ΛR→R,(f,g)7→ R f(s)g(s)exp(−NV(s))ds,N∈N,where theexternalfieldVsatisfies:(i)VisrealanalyticonR 0;(Rii)lim (V(x)/ln(x2+1))=+ ;and(iii) at lim (V(x)/ln(x 2+1))=+ .Orthogonalisationofth\e{(o}rdered)|bx|a→s∞e 1,z 1,z,z 2,z2,...,∞z k,zk,... m |x|→0 − ∞ { − − − } with respect to , L yields the even degree and odd degree orthonormal Laurent polynomials iv: {thφem(ezv)}e∞mn=0d:φeg2nr(eze)h=···anξ···id(−2nn)ozd−nd+·d·e·+gξr(ne2ne)zmn,oξn(n2inc)>o0rt,haongdoφn2anl+1L(za)u=reξn−(2tnn−+p11o)zl−ynn−o1+m··i·a+lξsn(:2nπππ+21n)z(nz,)ξ:=−(2nn−(+ξ11)(n2>n)0)−.1Dφe2fin(nze) X and πππ2n+1(z):=(ξ(2nn+11))−1φ2n+1(z). Asymptotics in the double-scaling limit as N,n→∞ such that r N/n=1+o(1)ofπππ− (−z)(intheentirecomplexplane),ξ(2n),φ (z)(intheentirecomplexplane),and a 2n n 2n Hankel determinant ratios associated with the real-valued, bi-infinite, strong moment sequence c = skexp( NV(s))ds areobtainedbyformulatingtheevendegreemonicorthogonalLaurent k R − k Z npolynRomial problem as ao∈matrix Riemann-Hilbert problem on R, and then extracting the large- n behaviour by applying the non-linear steepest-descent method introduced in [1] and further developedin[2,3]. 2000MathematicsSubjectClassification.(Primary)30E20,30E25,42C05,45E05, 47B36:(Secondary)30C15,30C70,30E05,30E10,31A99,41A20,41A21,41A60 AbbreviatedTitle.AsymptoticsofEvenDegreeOrthogonalLaurentPolynomials KeyWords.Asymptotics,equilibriummeasures,Hankeldeterminants,Laurentpolynomials, E-mail: [email protected] ∗ E-mail: [email protected] † E-mail: [email protected] ‡ 1 2 K.T.-R.McLaughlin,A.H.Vartanian,andX.Zhou Laurent-Jacobimatrices,Padéapproximants,parametrices,Riemann-Hilbertproblems,sing- ularintegralequations,strongmomentproblems,variationalproblems AsymptoticsofEvenDegreeOrthogonalLaurentPolynomials 3 1 Introduction and Background ConsidertheclassicalStieltjes(resp.,classicalHamburger)momentproblem(SMP)(resp.,HMP):givena simply-infinite(moment)sequenceofrealnumbers c : { n}∞n=0 (i) findnecessaryandsufficientconditionsfortheexistenceofanon-negativeBorelmeasureµS MP + (resp.,µH )on[0,+ )(resp.,( ,+ )),andwithinfinitesupport,suchthatc = ∞tndµS (t), MP ∞ −∞ ∞ n 0 MP n Z+ := 0 N (resp., c = +∞tndµH (t), n Z+), where the (improper) inRtegral is to be ∈ 0 { }∪ n MP ∈ 0 understoodintheRiemann-StiRe−l∞tjessense; (ii) when there is a solution of the existence problem, in which case the SMP (resp., HMP) is determinate,findconditionsfortheuniquenessofthesolution;and (iii) when there is more than one solution, in which case the SMP (resp., HMP) is indeterminate, describethefamilyofallsolutions. The SMP was—first—treated in 1894/95 by Stieltjes in the pioneering works [4], and the HMP was introduced and solved in 1920/21 by Hamburger in the landmark works [5]. The subsequent developmentofthetheoryofmomentproblemsbroughtforththeprofoundfactthat,overandabove theindispensableutilityaffordedbytheanalytictheoryofcontinuedfractions,inparticular,S-and realJ-fractions,thetheoryoforthogonalpolynomials[6]playedaseminal,intimateandcentralrôle (see,forexample,[7]). Questionsregardingtwosimply-infinite(moment)sequences cn n Z+ and c n n N ofrealnum- { } ∈ 0 { − } ∈ bers, or, equivalently, doubly- or bi-infinite (moment) sequences cn n Z of real numbers, manifest, in various settings, purely mathematical and/or otherwise, as na{tur}al∈extensions of the foregoing. This generalisation is colloquially refered to as the strong Stieltjes (resp., strong Hamburger) moment problem (SSMP)(resp.,SHMP),namely, given a doubly- or bi-infinite (moment) sequence cn n Z of { } ∈ realnumbers: (1) findnecessaryandsufficientconditionsfortheexistenceofanon-negativemeasureµSS (resp., MP + µSH)on[0,+ )(resp.,( ,+ )),andwithinfinitesupport,suchthatc = ∞tndµSS(t),n Z MP ∞ −∞ ∞ n 0 MP ∈ (resp.,c = +∞tndµSH(t),n Z),wherethe(improper)integralistobeundRerstoodinthesense n MP ∈ ofRiemannR−-S∞tieltjes; (2) whenthereisasolution,inwhichcasetheSSMP(resp.,SHMP)isdeterminate,findconditions fortheuniquenessofthesolution;and (3) whenthereismorethanonesolution,inwhichcasetheSSMP(resp.,SHMP)isindeterminate, describethefamilyofallsolutions. TheSSMP(resp.,SHMP)wasintroducedin1980(resp.,1981)byJonesetal.[8](resp.,Jonesetal.[9]), andstudiedfurtherin[10–14](see,also,thereviewarticle[15]).UnlikethemomenttheoryfortheSMP and the HMP, wherein the theory of orthogonal polynomials, and the analytic theory of continued fractions,enjoyedaprominentrôle,theextensionofthemomenttheorytotheSSMPandtheSHMP introduceda‘rationalgeneralisation’of theorthogonal polynomials, namely,the orthogonalLaurent (or L-) polynomials (as well as the introduction of special kinds of continued fractions commonly referredtoaspositive-Tfractions),whicharediscussedbelow[10–21].(TheSHMPcanalsobesolved usingthespectraltheoryofunboundedself-adjointoperatorsinHilbertspace[22];see,also,[23].) For any pair (p,q) Z Z, with p6q, let ΛC := f: C C; f(z)= q λ zk, λ C, k=p,...,q , ∈ × p,q ∗→ k=p k k∈ (cid:26) (cid:27) AwhfuernectCio∗n:=(Cor\e{l0e}m.Feonrt)anfyΛmC∈iZsc+0a,lsleetdΛaC2mLa:=urΛenC−tm(,mo,rΛL-C2)m+p1ol:=ynΛomC−mi−al1.,m(N,aonPtde:ΛthCbe:=se∪tsmb∈ΛZC+0(ΛanC2md∪ΛΛCC2fmo+r1m). ∈ p,q linearspacesoverthefieldCwithrespecttotheoperationsofadditionandmultiplicationbyascalar.) Bases for each of the spaces ΛC , ΛC , and ΛC, respectively, are z m,...,zm , z m 1,...,zm , and 2m 2m+1 { − } { − − } const.,z 1,z,z 2,z2,...,z k,zk,... (thebasisforΛCcorrespondstothecyclically-repeatedpolesequence − − − {nopole,0, ,0, ,...,0, ,... ).}Furthermore, note that, for each 0. f ΛC, there exists a unique l{ Z+ sucht∞hat f∞ΛC.Fo∞rl Z}+ and0.f ΛC,theL-degreeof f,symbolic∈allyLD(f),isdefinedas ∈ 0 ∈ l ∈ 0 ∈ l LD(f):=l. 4 K.T.-R.McLaughlin,A.H.Vartanian,andX.Zhou ForΛC f= λ zj,setC (f):=λ , j Z.Foreachl Z+ and0.f ΛC,definetheleadingcoefficient of f,sym∋bolicaj∈lZlyLjC(f),andj thetrajilin∈gcoefficientof∈f,sy0mbolically∈TCl(f),asfollows: P b b λ , l=2m, LC(f):= m λ , l=2m+1, b−m−1 and b λ , l=2m, TC(f):= −m λ , l=2m+1. bm Thus,forl∈Z+0 and0.f∈ΛCl ,onewrites,for fb:=fl(z):(1)ifl=2m, f (z)=TC(f)z m+ +LC(f)zm; 2m − ··· and(2)ifl=2m+1, f2m+1(z)=LC(f)z−m−1+ +TC(f)zm. ··· Forl Z+,0.f ΛCiscalledmonicifLC(f)=1. ∈ 0 ∈ l Consider the positive measure on R (oriented throughout this work, unless stated otherwise, from to+ )givenby −∞ ∞ dµ(z)=w(z)dz, withvaryingexponentialweightfunctionoftheform e e w(z)=exp( NV(z)), N N, − ∈ wheretheexternalfieldV: R 0 Rsatisfiesthefollowingconditions: e \{ }→ V isrealanalyticon R 0 ; (V1) \{ } lim V(x)/ln(x2+1) =+ ; (V2) x ∞ | |→∞(cid:16) (cid:17) lim V(x)/ln(x−2+1) =+ . (V3) x 0 ∞ | |→ (cid:16) (cid:17) (For example, a rational function of the form V(z) = 2m2 ̺ zk, with ̺ R, k = 2m ,...,2m , m1,2 ∈ N, and ̺−2m1,̺2m2 > 0 would suffice.) Define (uPnki=q−u2eml1y)kthe strongkm∈oment li−near1functiona2l L by its action on the basis elements of ΛC: L: ΛC ΛC, f = λ zk L(f) := λ c , where ck = L(zk) = Rskexp(−NV(s))ds, (k,N) ∈ Z × N→. (Note thaPt,ka∈sZbpker t7→he discussioPnk∈aZbbokvek, c = skexp( NV(sR))ds, N N is a bi-infinite, real-valued, strong moment sequence: c is called k R − ∈ k Z k nthekRthstrongmomentofL.)Assocoia∈tedwiththeabove-definedbi-infinite,real-valued,strongmoment sequence{ck}k∈ZaretheHankeldeterminantsHk(m),(m,k)∈Z×N[10,11,15,17]: cm cm+1 cm+k 2 cm+k 1 ··· − − (cid:12) cm+1 cm+2 cm+k 1 cm+k (cid:12) (cid:12) ··· − (cid:12) H0(m):=1 and Hk(m):=(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) cm...+2 cm...+3 ·.·.·. cm...+k cm+...k+1(cid:12)(cid:12)(cid:12)(cid:12)(cid:12). (1.1) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12)(cid:12)(cid:12)cm+k−1 cm+k ··· cm+2k−3 cm+2k−2(cid:12)(cid:12)(cid:12) (cid:12) (cid:12) Foranypair(p,q) Z Z,withp6q,letΛR :=(cid:12) f: C C; f(z)= q λ zk, λ R,(cid:12)k=p,...,q ,and ∈ × p,q (cid:26) ∗→ k=p k k∈ (cid:27) (dNefiontee:,tahnealsoegtsouΛsRlyaasnadboΛvRe,ffoorrmml∈inZe+0a,rΛsR2pma:c=eΛsR−omv,mer,ΛthR2me+fi1:e=ldΛRR−m−wP1,imt,haenredspΛeRect:=to∪mt∈hZe+0(oΛpR2emra∪tiΛoR2nms+1o)f. p,q additionandmultiplicationbyascalar;furthermore,ΛR ( ΛC)isthelinearspaceoverRspanned byzj, j Z.)Hereafter,weshallbeconcernedonlywith(re⊂al)L-polynomialsinΛR:the—ordered— basefor∈ΛR is 1,z 1,z,z 2,z2,...,z k,zk,... ,correspondingtothecyclically-repeatedpolesequence − − − nopole,0, ,0{, ,...,0, ,... .Definethe}realbilinearform , Lasfollows: , L: ΛR ΛR R, {(f,g)7→hf,g∞iL:=∞L(f(z)g(∞z))=}R f(s)g(s)e−NV(s)ds, N∈N. It ihs···a···ifact[10,11,15h,···17···i] that th×e bili→near form , L thus defined is anRinner product if and only if H(−2m)>0 and H(−2m)>0 m Z+ (see h··· ···i 2m 2m+1 ∀ ∈ 0 Equations(1.8)below,andSubsection2.2,theproofofLemma2.2.1);andthisfactisused,withlittle ornofurtherreference,throughoutthiswork(see,also,[24]). AsymptoticsofEvenDegreeOrthogonalLaurentPolynomials 5 Remark1.1. Theselattertwo(Hankeldeterminant)inequalitiesalsoappearwhenthequestionofthe solvabilityof the SHMPisposed(inthis case,the c , k Z,which appearinEquations (1.1)should k be replacedby cSHMP, k Z): indeed, if these two inequa∈lities are true m Z+, then there is a non- k ∈ ∀ ∈ 0 negativemeasureµSH (onR)withthegiven(real)moments.ForthecaseoftheSSMP,therearefour MP (Hankel determinant) inequalities (in this latter case, the c , k Z, which appear in Equations (1.1) k ∈ shouldbereplacedbycSSMP,k Z)whichguaranteetheexistenceofanon-negativemeasureµSS (on k ∈ MP [0,+ ))withthegivenmoments,namely[8](see,also,[10,11]):foreachm Z+,H(−2m)>0,H(−2m)>0, ∞ ∈ 0 2m 2m+1 H(−2m+1)>0, and H(−2m−1)<0. It is interesting to note that the former solvability conditions do not 2m 2m+1 asiumtoilmarastitcaatellmyeimntphlyoltdhsattrtuheefpoorstihtievSeM(rePa(ls)emeothmeelnattste{rcSkfHoMuP}rk∈sZo+0lvdaebtielirtmyicnoenadimtieoanssu).reviatheHMP:(cid:4)a If f ΛR,then ∈ f() L:=( f, f L)1/2 k · k h i is called the norm of f with respect to L: note that f() L>0 f ΛR, and f() L>0 if 0. f ΛR. k · k ∀ ∈ k · k ∈ t{oφ♭nL(zi)f},n∈Zm+0 i,sncaZlle+d: a(real)orthonormalLaurent(orL-)polynomialsequence(ONLPS)withrespect ∀ ∈ 0 (i) φ♭ ΛR,thatis,LD(φ♭):=n; n∈ n n (ii) hφ♭m,φ♭n′iL=0 ∀ m,n′,or,alternatively,hf,φ♭niL=0 ∀ f∈ΛRn−1; (iii) hφ♭m,φ♭miL=:kφ♭m(·)k2L=1. Orthonormalisation of 1,z 1,z,z 2,z2,...,z n,zn,... , correspondingto the cyclically-repeatedpole − − − { } sequence nopole,0, ,0, ,...,0, ,... , with respect to , L via the Gram-Schmidt orthogonal- { ∞ ∞ ∞ } h··· ···i isation method, leads to the ONLPS, or, simply, orthonormal Laurent (or L-) polynomials (OLPs), φm(z) m Z+,which,bysuitablenormalisation,maybewrittenas,form=2n, { } ∈ 0 φ2n(z)=ξ(−2nn)z−n+···+ξn(2n)zn, ξn(2n)>0, (1.2) and,form=2n+1, φ2n+1(z)=ξ(2nn+11)z−n−1+···+ξn(2n+1)zn, ξ(2nn+11)>0. (1.3) − − − − Theφ ’sarenormalisedsothattheyallhaverealcoefficients;inparticular,theleadingcoefficients, n LC(φ2n):=ξ(n2n) andLC(φ2n+1):=ξ(2nn+11), n∈Z+0,arebothpositive, ξ(00)=1,andφ0(z)≡1.Eventhough the leading coefficients, ξ(2n) and−ξ−(2n+1), n Z+, are non-zero (in particular, they are positive), no n n 1 ∈ 0 sFuucrhthreersmtroicreti,onnotaeptphlaiets,btyoctohnesttrrauiclit−niog−nc:oefficients, TC(φ2n):=ξ−(2nn) and TC(φ2n+1):=ξn(2n+1), n∈Z+0. (1) φ2n,zj L=0, j= n,...,n 1; h i − − (2) φ2n+1,zj L=0, j= n,...,n; h i − (3) hφj,φkiL=δjk, j,k∈Z+0,whereδjk istheKroneckerdelta. Moreover,if, for eachm∈Z+0, the orthonormal L-polynomials φ2m(z) and φ2m+1(z), respectively, are suchthatTC(φ2m):=ξ(2mm),0andTC(φ2m+1):=ξ(m2m+1),0,thentherearespecialChristoffel-Darboux formulaefortheOLPs−(see,forexample,[12,17];see,also,[25]): φ2m(ζ)(zφ2m−1(z)−ζφ2m−1(ζ))−ζφ2m−1(ζ)(φ2m(z)−φ2m(ζ))=(z−ζ)ξξ(2−(mm2mm−)1) 2Xmj=−01φj(z)φj(ζ), − ξ(2m+1) 2m φ2m(ζ)(zφ2m+1(z)−ζφ2m+1(ζ))−ζφ2m+1(ζ)(φ2m(z)−φ2m(ζ))=(z−ζ) ξmm(2m) Xj=0 φj(z)φj(ζ), whereφ (z) 0,and(dividingbyz ζandlettingζ z) 1 − ≡ − → φ2m(z)ddz(zφ2m−1(z))−zφ2m−1(z)ddzφ2m(z)=ξξ(2−(mm2mm−)1) 2Xmj=−01(φj(z))2, − 6 K.T.-R.McLaughlin,A.H.Vartanian,andX.Zhou d d ξ(2m+1) 2m φ2m(z)dz(zφ2m+1(z))−zφ2m+1(z)dzφ2m(z)= ξmm(2m) Xj=0(φj(z))2. Itisconvenienttointroducethemonic orthogonalLaurent(orL-)polynomials, πππ (z), j Z+:(i) for j=2n,n Z+,withπππ (z) 1, j ∈ 0 ∈ 0 0 ≡ πππ2n(z):=φ2n(z)(ξ(n2n))−1=ν−(2nn)z−n+···+zn, ν−(2nn):=ξ−(2nn)/ξn(2n); (1.4) and(ii)for j=2n+1,n Z+, ∈ 0 πππ2n+1(z):=φ2n+1(z)(ξ(2nn+11))−1=z−n−1+···+νn(2n+1)zn, νn(2n+1):=ξn(2n+1)/ξ(2nn+11). (1.5) − − − − ThemonicorthogonalL-polynomials,πππ (z), j Z+,possessthefollowingproperties: j ∈ 0 (1) πππ2n,zj L=0, j= n,...,n 1; h i − − (2) πππ2n+1,zj L=0, j= n,...,n; h i − (3) hπππ2n,πππ2niL=:kπππ2n(·)k2L=(ξ(n2n))−2,whenceξn(2n)=1/kπππ2n(·)kL(>0); (4) hπππ2n+1,πππ2n+1iL=:kπππ2n+1(·)k2L=(ξ(2nn+11))−2,whenceξ(2nn+11)=1/kπππ2n+1(·)kL(>0). − − − − Furthermore, in terms of the Hankel determinants, H(m), (m,k) Z N, associated with the real- k ∈ × valued, bi-infinite, strong moment sequence c = ske NV(s)ds, N N , the monic orthogonal k R − ∈ k Z L-polynomials,πππ (z), j Z+,arerepresentedviantheRfollowingdeterminaont∈alformulae[10,11,15,17] (see,also,Subsectjion2.∈2,P0roposition2.2.1):form Z+, ∈ 0 c 2m c 2m+1 c 1 z−m (cid:12)c −2m+1 c−2m+2 ··· c−0 z−m+1(cid:12) πππ2m(z)=H2(−1m2m) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) −c... −c... ·.·.·. c ... zm... 1(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12), (1.6) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) c−01 c10 ······ c22mm−−12 zm− (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12) (cid:12) (cid:12) (cid:12) and c c c z m 1 2m 1 2m 1 − − (cid:12) −c 2m− c −2m+1 ··· c−0 z−m (cid:12) πππ2m+1(z)=−H2(−1m2+m1) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) c−... −c... ·.·.·. c ... zm... 1(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12); (1.7) (cid:12)(cid:12)(cid:12)(cid:12) c−01 c01 ······ 2cm2m−1 zm− (cid:12)(cid:12)(cid:12)(cid:12) (cid:12) (cid:12) (cid:12) (cid:12) moreover,itcanbeshownthat(see,forex(cid:12)ample,[15,17]),forn Z+, (cid:12) ∈ 0 ξ(n2n) =kπππ2n1(·)kL!=vutHH22((−−nn22+nn1)), ξ−(2nn−+11) =kπππ2n+11(·)kL!=vutHH2(−n2(−2+nn2+2−n1)2), (1.8) Foreachm Z+,νth(−2nen)m:=onξξi(n(−c22nnno))rt=hoHgH2(o−n2(n−2nn2a+nl)1L),-polynoνmn(2ina+l1)πππ:=(zξξ)−(n(a22nnnn−++d111))th=e−inHdH2(e−n2(x−2+nn2+1m−n1)1a).recallednon-sing(u1l.a9r) ∈ 0 m ν(2n), m=2n, if0,TC(πππm):=ν(−2nn+1), m=2n+1; otherwise,πππm(z)andmaresingular.FromEquations(1.9),itcan beseenthat,foreanchm∈Z+0: (i) πππ2m(z)isnon-singular(resp.,singular)ifH2(−m2m+1),0(resp.,H2(−m2m+1)=0); AsymptoticsofEvenDegreeOrthogonalLaurentPolynomials 7 (ii) πππ2m+1(z)isnon-singular(resp.,singular)ifH2(−m2+m1−1),0(resp.,H2(−m2+m1−1)=0). F[1o0r,e1a1c,1h7m]t∈hZat+0,,folertmµ2mZ:=+:card{z; πππ2m(z)=0}andµ2m+1:=card{z; πππ2m+1(z)=0}.Itisanestablishedfact ∈ 0 (1) the zeros of πππ (z) are real, simple, and non-zero, and µ = 2m (resp., 2m 1) if πππ (z) is 2m 2m 2m − non-singular(resp.,singular); (2) thezerosofπππ2m+1(z)arereal,simple,andnon-zero,andµ2m+1=2m+1(resp.,2m)ifπππ2m+1(z)is non-singular(resp.,singular). For each m Z+, it can be shown that, via a straightforward factorisation argument and using ∈ 0 Equations(1.6)and(1.7): (i) ifπππ (z)isnon-singular,uponsetting α(2m), k=1,...,2m := z; πππ (z)=0 , 2m k { 2m } n o 2m α(2m)=ν(2m); k m Yk=1 − (ii) ifπππ2m+1(z)isnon-singular,uponsetting α(k2m+1), k=1,...,2m+1 :={z; πππ2m+1(z)=0}, n o 2m+1 α(2m+1)= ν(2m+1) −1. k − m Yk=1 (cid:16) (cid:17) Unlikeorthogonalpolynomials,whichsatisfyasystemofthree-termrecurrencerelations,monic orthogonal,andorthonormal,L-polynomialsmaysatisfyrecurrencerelationsconsistingofapairof four-termrecurrencerelations[15],apairofsystemsofthree-orfive-termrecurrencerelations(which isguaranteedinthecasewhenthecorrespondingmonicorthogonal,andorthonormal,L-polynomials arenon-singular)[15–17],orasystemconsistingoffourfive-termrecurrencerelations[23]. Remark1.2. Thenon-vanishing of theleadingandtrailingcoefficientsofthe OLPs φ (z) ,that { m }∞m=0 is, ξ(2n), m=2n, LC(φ ):= n m ξ(2n+1), m=2n+1,  −n−1 and  ξ(2n), m=2n, TC(φm):=ξ(−2nn+1), m=2n+1,  n respectively,isofparamountimportance:ifboththeseconditionsarenotsatisfied,thenthe‘length’ oftherecurrencerelationsmaybegreaterthanthree[16](see,also,[24]). (cid:4) It can be shown that (see, for example, [17], and Chapter 11 of [26]), if πππm(z) m Z+, as defined above,isanon-singular, monic orthogonal L-polynomialsequence,thatis, H{(−2n+1)},∈00(m=2n)and 2n H2(−n2+n1−1),0(m=2n+1),then{πππm(z)}m∈Z+0 satisfythepairofthree-termrecurrencerelations z 1 πππ2m+1(z)=β−♮ +β♮2m+1πππ2m(z)+λ♮2m+1πππ2m−1(z),  2mz  πππ2m+2(z)=β♮ +β♮2m+2πππ2m+1(z)+λ♮2m+2πππ2m(z),  2m+1  whereπππ (z) 0, 1 − ≡ β♮ =ν(2m), β♮ =ν(2m+1), 2m m 2m+1 m − λ♮2m+1=−HH2(−m(−2+m21m−)1H)H(−2(−2mm2−m+1+1)2) (,0), λ♮2m+2=−HH2((−−m22+mm2)−H1)(H−22(m−m2−m1)) (,0), 2m 2m 2m+1 2m+1 8 K.T.-R.McLaughlin,A.H.Vartanian,andX.Zhou andλ β /β >0 j N,withλ := c ,leadingtoatri-diagonal-typeLaurent-Jacobimatrix forthe j j 1 j 1 1 − ∀ ∈ − − F ‘mixed’mapping F: ΛR→ΛR, f(z)7→(z−1(⊕∞n=0diag(1,0))+z(⊕∞n=0diag(0,1)))f(z), where diag(1,0):=diag(1,0,...,1,0,...),and diag(0,1):=diag(0,1,...,0,1,...), ⊕∞n=0 ⊕∞n=0 β♮ 1 − 1 λ♮ β♮ 1 − 2 − 2  F = diag(cid:16)β♮0,β♮1,β♮2,...(cid:17) −λ♮3 −−λβ♮3♮4 −−1λβ♮4♮5 −−1λβ♮5♮6 −.1.β.♮6 −λ.♮21.m.+1 −−λβ.♮2♮2.mm.++12 −β.♮21.m.+2 .1.. ... , withzerosoutsidetheindicateddiagonals(intermsof φm(z) m Z+,thepairofthree-termrecurrence { } ∈ 0 relationsreads[16]: φ2m+1(z)=(z−1+g2m+1)φ2m(z)+f2m+1φ2m 1(z), − φ2m+2(z)=(1+g2m+2z)φ2m+1(z)+f2m+2φ2m(z), pwahireroefffi2mv+e1-,tef2rmm+2r,ec0u,rmren∈cZe+0re,lφa−ti1o(nz)s≡[107,]a,nwdithφ0πππ(z)(≡z)1);0o,tjh=e1rw,2i,se, {πππm(z)}m∈Z+0 satisfythe following j − ≡ πππ2m+2(z)=γ♭2m+2,2m−2πππ2m−2(z)+γ♭2m+2,2m−1πππ2m−1(z)+(z+γ♭2m+2,2m)πππ2m(z) +γ♭2m+2,2m+1πππ2m+1(z), πππ2m+3(z)=γ♭2m+3,2m 1πππ2m 1(z)+γ♭2m+3,2mπππ2m(z)+(z−1+γ♭2m+3,2m+1)πππ2m+1(z) − − +γ♭2m+3,2m+2πππ2m+2(z), where γ = 0, k < 0, l > 2, leading to a penta-diagonal-type Laurent-Jacobi matrix for the ‘mixed’ l,k G mapping G: ΛR→ΛR, g(z)7→(z(⊕∞n=0diag(1,0))+z−1(⊕∞n=0diag(0,1)))g(z), γ♭ γ♭ 1 − 2,0 − 2,1 γ♭ γ♭ γ♭ 1 − 3,0 − 3,1 − 3,2  G=−γ♭4,0 −−γγ♭4♭5,,11 −−−γγγ♭4♭5♭6,,,222 −−−.γγγ.♭4♭5♭6.,,,333 −γ♭2−−m.γγ+1.2♭5♭6.,,,244m−2 −−γγ♭2♭2−mm.γ++1.♭632.,,,522mm−−11 −−γγ♭2♭2..mm1..++..32,,22mm −−γγ♭2♭2mm..++..32..,,22mm++11 −γ♭2m.+1.3.,2m+2 .1.. ... , withzerosoutsidetheindicateddiagonals.Thegeneralformofthese(systemof)recurrencerelations isapairofthree-andfive-termrecurrencerelations[23]:forn Z+, ∈ 0 zφ2n+1(z)=b♯2n+1φ2n(z)+a♯2n+1φ2n+1(z)+b♯2n+2φ2n+2(z), zφ2n(z)=c♯2nφ2n−2(z)+b♯2nφ2n−1(z)+a♯2nφ2n(z)+b♯2n+1φ2n+1(z)+c♯2n+2φ2n+2(z), whereallthecoefficientsarereal,c♯=b♯=0,andc♯ >0,k N,and 0 0 2k ∈ z−1φ2n(z)=β♯2nφ2n−1(z)+α♯2nφ2n(z)+β♯2n+1φ2n+1(z), AsymptoticsofEvenDegreeOrthogonalLaurentPolynomials 9 z−1φ2n+1(z)=γ♯2n+1φ2n−1(z)+β♯2n+1φ2n(z)+α♯2n+1φ2n+1(z)+β♯2n+2φ2n+2(z)+γ♯2n+3φ2n+3(z), whereallthecoefficients arereal,β♯=γ♯=0,β♯>0,andγ♯ >0,l N,leading, respectively,tothe 0 1 1 2l+1 ∈ real-symmetric,tri-penta-diagonal-typeLaurent-Jacobimatrices, and ,forthemappings J K J: ΛR ΛR, j(z) zj(z) and K: ΛR ΛR, k(z) z−1k(z), → 7→ → 7→ a♯ b♯ c♯ 0 1 2 b♯ a♯ b♯  J =c♯21 b1♯2 bac♯42♯2♯3 bba♯3♯4♯3 bbacc♯6♯4♯4♯5♯4 bba♯5♯5♯6 babcc♯6♯8♯6♯6♯7 bba♯7♯7♯8 bc♯2♯2.backk.♯8♯8♯8++.21 ab♯2♯2.bkk.♯9++.12 bac♯2♯2..kk♯1..++0..22 b♯2.k.+.3 c♯2.k.+.4 , and α♯ β♯ 0 1 β♯ α♯ β♯ γ♯  K = 1 γβ♯21♯3 αβ2♯3♯2 αγββ♯3♯43♯3♯5 αββ♯4♯5♯4 γαγββ♯5♯6♯5♯7♯5 α.βββ.♯8♯7♯6♯6. ααγ.β.♯7♯7♯8♯7. γ♯2.ββk.♯9♯8+.1 β♯2γk♯9+1 αβ♯2.♯2kk.++.21 αβ♯2.♯2kk.++.22 γβ♯2.♯2kk.++.33 , withzerosoutsidetheindicateddiagonals;moreover,asshownin[23], and areformalinverses, J K thatis, = =diag(1,...,1,...)(see,also,[27–31]). JK KJ Itisconvenientatthispointtodiscuss,ifonlysuccinctly,afewofthemultitudinousapplications ofL-polynomials(completedetailsmaybefoundintheindicatedreferences): (1) as stated at the beginning of the Introduction, L-polynomials are intimately related with the solution of the SSMP and the SHMP. It is important to note [14] that the classical and strong moment problems (SMP, HMP, SSMP, and SHMP) are specialcases of a more general theory, wheremomentscorrespondingtoanarbitrary,countablesequenceof(fixed)pointsareinvolved (in the classical and strong moment cases, respectively, the points are repeated and 0, ∞ ∞ cyclicallyrepeated),andwhereorthogonalrationalfunctions[26,32,33]playtherôleoforthogonal polynomialsandorthogonalLaurent(orL-)polynomials;furthermore,sinceL-polynomialsare rationalfunctionswith(fixed)polesattheoriginandatthepointatinfinity, thesteptowards amoregeneraltheorywherepolesareatarbitrary,butfixed,positions/locations inC is ∪{∞} natural,withapplicationsto,say,multi-pointPadé,andPadé-type,approximants[24,34–38]; 10 K.T.-R.McLaughlin,A.H.Vartanian,andX.Zhou b (2) innumericalanalysis,thecomputationofintegralsoftheform f(s)dµ(s),whereµisapositive a measure on [a,b], and 6a<b6+ , is an important probRlem. The most familiar quadra- −∞ ∞ tureformulaearetheso-calledGauss-Christoffelformulae,thatis,approximatingtheintegral b f(s)dµ(s) via a weighted-sum-of-products of function values of the form n A f(x ), a j=1 j,n j,n nR ∈ N, where one chooses for the nodes {xj,n}nj=1 the zeros/roots of ϕn(z), thePpolynomial of b degreenorthogonalwithrespecttotheinnerproduct f,g = f(s)g(s)dµ(s),andforthe(posi- h i a tive)weights{Aj,n}nj=1theso-calledChristoffelnumbers[35].WRhenconsideringthecomputation ofintegralsoftheform π g(eiθ)dµ(θ),wheregisacomplex-valuedfunctionontheunitcircle π D:= z C; z=1 andµiRs−,say,apositivemeasureon[ π,π],inparticular,whengiscontinuous { ∈ | | } − onD,keepinginmindthatafunctioncontinuousonDcanbeuniformlyapproximatedbyL- polynomials,itisnaturaltoconsider,insteadoforthogonalpolynomials,Laurentpolynomials, whicharealsorelatedtotheassociatedtrigonometricmomentproblem[35,39](see,also,[40]); (3) for V: R 0 R as described by conditions (V1)–(V3), consider the function g(z)= (1+ \{ }→ R sz) 1dµ(s), where dµ(s) =exp( NV(s))ds, N N, which is holomorphic for z C R,Rwith − − ∈ ∈ \ associatedasymptoticexpansions e e ∞ ∞ g(z) = ( 1)mcmzm=:L0(z) and g(z) = ( 1)mc mz−m=:L∞(z), C\R∋z→0 Xm=0 − C\R∋z→∞−Xm=1 − − wherec = sle NV(s)ds,l Z,withrespecttothe(unbounded)domain z C; ε6 Arg(z) 6π ε , l R − ∈ { ∈ | | − } whereArgR( )denotestheprincipalargumentof ,andε>0issufficientlysmall.Giventhepair ∗ ∗ offormalpowerseries(L0(z),L (z)),therationalfunctionP (z)/Q (z),whereP (z)belongs ∞ k,n k,n k,n to the space of all polynomials of degree at most n 1, and Q (z) is a polynomial of degree k,n − exactly n with Q (0),0, is said to be a [k/n](z) two-point Padé approximant to (L0(z),L (z)), k,n ∞ k 0,1,...,2n ,ifthefollowingconditionsaresatisfied: ∈{ } L0(z) Pk,n(z)(Qk,n(z))−1 = (zk), − z 0O → L∞(z) Pk,n(z)(Qk,n(z))−1 = (z−1)2n−k+1 . − z O →∞ (cid:16) (cid:17) The ‘balanced’ situation corresponds to the case when k = n, in which case, the two-point Padéapproximantsaredenoted,simply,as[n/n](z).Animportant,relatedproblemofcomplex approximationtheoryistostudytheconvergenceofsequencesoftwo-pointPadéapproximants constructedfromthe—formal—pair(ofpowerseries)(L0(z),L (z))tothefunctiong(z)onC R; ∞ \ in particular, denoting by E (z) the ‘error term’ for the [n/n](z) approximant, that is, E (z):= n n g(z) [n/n](z),itcanbeshownthat,following[41], − E (z)= φ ( 1/z) −1 φn(s)e−NV(s) ds, z C R, (TPA1) n (cid:16) n − (cid:17) ZR 1+sz ∈ \ where φm(z) m Z+ aretheorthonormalL-polynomialsdefinedinEquations(1.2)and(1.3).The { } ∈ 0 main question regarding the convergence of two-point Padé approximants for this class of functions is with which rate it takes place, that is, the so-called quantitative result [42]: this necessitates obtaining results for the asymptotic behaviour (as n ) of the orthonormal L- →∞ polynomials φ (z) in the entire complex plane. The theory of orthogonal L-polynomials is a n natural framework for developing the theory of two-point Padé approximants, for both the scalarandmatrixcases[24,41–44]; (4) itturnsoutthat,unlikethe(finite)non-relativisticTodalattice,whosedirectandinversespectral transform was constructed by Moser [45], and which is based on the theory of orthogonal polynomials and tri-diagonal Jacobi matrices, the direct and inverse scattering transform for the (finite) relativistic Toda lattice, introduced by Ruijsenaars [46], is based on the theory of orthogonalL-polynomialsandpairsofbi-diagonalmatrices[47](see,also,[48]);and (5) for afinite,countable or uncountableindex setK, let ςp, p K C+:= z C; Im(z)>0 ,with { ∈ } ⊂ { ∈ } ς ,ς p,q K,and ̟ , p K Cbegivenpointsets.AfunctionF(z)whichisanalyticfor p q p ∀ ∈ { ∈ } ⊂

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.