Scribd Logo
Upload

Welcome to Scribd!

  • (Mathematical Surveys and Monographs 027) Nathan J. Fine - Basic Hypergeometric Series and Applications-American Mathematical Society (1988)

    Uploaded by

    Hiến Nguyễn
    9 views142 pages

    Original Title

    (Mathematical Surveys and Monographs 027) Nathan J. Fine - Basic hypergeometric series and applications-American Mathematical Society (1988)

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    9 views142 pages

    (Mathematical Surveys and Monographs 027) Nathan J. Fine - Basic Hypergeometric Series and Applications-American Mathematical Society (1988)

    Uploaded by

    Hiến Nguyễn

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    of 142
    NATHAN J. FINE CUO N AAR SORVEYS, AND MONOGRAPHS: eT tct tara LeRoi ois by mt eCORAER CIAL MLAs ag Hist RO Scan MATHEMATICAL SURVEYS AND MONOGRAPHS SERIES LIST Volume 1 TA 72 10 u 12 13 14 15 The problem of moments, J. A. Shohat and J. D. Tamarkin The theory of rings, N. Jacobson Geometry of polynomials, M. Marden The theory of valuations, O. F. G. Schilling The kernel function and conformal mapping, S. Bergman Introduction to the theory of algebraic functions of one variable, C. C. Chevalley The algebraic theory of semigroups, Volume I, A. H. Clifford and G. B. Preston The algebraic theory of semigroups, Volume II, A. H. Clifford and G. B. Preston Discontinuous groups and automorphic functions, J. Lehner Linear approximation, Arthur Sard An introduction to the analytic theory of numbers, R. Ayoub Fixed points and topological degree in nonlinear analys J. Cronin Uniform spaces, J. R. Isbell Topics in operator theory, A. Brown, R. G. Douglas, C. Pearcy, D. Sarason, A. L. Shields; C. Pearcy, Editor Geometric asymptotics, V. Guillemin and S. Sternberg, Vector measures, J. Diestel and J.J. Uhl, Jr. 16 17 18 19 20 a 2 23 24 26 Symplectic groups, O. Timothy O'Meara Approximation by polynomials with integral coefficients, Le Baron O, Ferguson Essentials of Brownian motion and diffusion, Frank B. Knight Contributions to the theory of transcendental numbers, Gregory V. Chudnovsky Partially ordered abelian groups with interpolation, Kenneth R. Goodearl The Bieberbach conjecture: Proceedings of the symposium on the occasion of the proof, Albert Baernstein, David Drasin, Peter Duren, and Albert Marden, Editors Noncommutative harmonic analysis, Michael E, Taylor Introduction to various aspects of degree theory in Banach spaces, E. H. Rothe Noetherian rings and their applications, Lance W. Small, Editor Asymptotic behavior of dissipative systems, Jack K, Hale Operator theory and arithmetic in H®™, Hari Bercovici This page intentionally left blank BASIC HYPERGEOMETRIC SERIES AND APPLICATIONS This page intentionally left blank = MATHEMATICAL SURVEYS [| AND MONOGRAPHS |} NUMBER 27 BASIC HYPERGEOMETRIC SERIES AND APPLICATIONS NATHAN J. FINE American Mathematical Society Providence, Rhode Island 1980 Mathematics Subject Classification (1985 Revision). Primary 05A15, 05A17, 05A19, 05A30, 11E25, 11P57, 11P76, 11P80, 3370. LIBRARY OF CONGRESS Library of Congress Cataloging-in-Publication Data Fine, Nathan J. (Nathan Jacob), 1916- Basic hypergeometric series and applications/Nathan J. Fine. . om. — (Mathematical surveys and monographs, ISSN 0076-5976; no. 27) Includes bibliographies. ISBN 0-8218-1524-5 (alk. paper) 1. Hypergeometric series. |. Title. I. Series. QA295.F54 1988 515/.243-—de19 88-6295 Copying and reprinting. individual readers of this publication, and nonprofit libraries acting for them, are permitted to make fair use of the material, such as to copy a chapter for use in teaching or research. Permission is granted to quote brief passages from this publication in reviews, provided the customary acknowledgment of the source is given. Republication, systematic copying, or multiple reproduction of any materia inthis publication (including abstracts) is permitted only under license from the American Mathematical Society. Requests for such permission should be addressed to the Executive Director, American Math- ematical Society, P.O. Box 6248, Providence, Rhode Island 02940. ‘The owner consents to copying beyond that permitted by Sections 107 or 108 of the U.S. Copyright Law, provided that a fee of $1.00 plus $.25 per page for each copy be paid directly to the Copyright Clearance Center, Inc., 21 Congress Street, Salem, Massachusetts 01970. When paying this fee please use the code 0076-5376/88 to refer to this publication. This consent does ‘ot extend to other kinds of copying, such as copying for general distribution, for advertising or promotion purposes, for creating new collective works, or for resale. Copyright ©1988 by the American Mathematical Society. All rights reserved Printed in the United States of America Reprinted 1989, The American Mathematical Society retains all rights ‘except those granted to the United States Government ‘The paper used in this book is acid-free and falls within the guidelines established to ensure permanence and durablity. € Contents Foreword Introduction Chapter 1. Fundamental Properties of Basie Hypergeometric Series 1. Definitions 2. Two functional equations ‘The analytic character of F(a, ; t) More transformations ‘The method of iteration Application of iteration (t = tq) Iteration of b+ bg Consequences of §7 Further consequences of §7 10. A product-series identity 11, Iteration of a aq 12, Iteration of (b,t) + (bq, ta) 13, Iteration of (a, t) > (aq,tq) 14, Iteration of (a, ,t) > (ag, bq, ta) 15. A special development 16. The partial-fraction decomposition 17. Jacobi’s triple produet 18. A bilateral series 19. Two product-series identities 20. A general transformation 21. The basic multinomial Notes References ae eenas Chapter 2. Partitions 22. Sums over partitions 23. Partitions with odd parts and with distinet parts 24, Continuation ix xi ea kere 10 12 13 14 15 16 7 18 19 22 24 30 32 35 37 37 44 46 25. CONTENTS ‘The rank of a partition Notes References Chapter 3. Mock Theta-Functions and the Functions L(N), J(N) 26. 27. Mock theta functions The functions L(N) and J(N) Notes References Chapter 4. Other Applications 28. 29. 30. 31. 32. 33. Theorems of Liouville type Another identity and paraphrase Two more identities and their paraphrases Sums of squares Products with multiplicative series More about products Notes References Chapter 5. Modular Equations 34. 35. 37. 38. 39. 40. 41 42. Modular equations, preliminaries A set of functional equations . Application of (35.13) Two modular equations Continuation of §36 Other functional values of H(2) A system of identities Permanent identities Continuation Notes References Bibliography 100 101 104 107 112 117 121 122 123 Foreword In 1948, Nathan Fine published a note in the Proceedings of the National Academy of Sciences announcing several elegant and intriguing partition theo- rems. These results were marked both by their simplicity of statement and (as we shall see in Chapter 2) by the depth of their proof. Fine was at that time engaged in his own special development of q-hypergeometric series, and as the years passed he kept adding to his results and polishing his presentation. Several times, both at Penn and Penn State, he presented courses on this material. I took the course twice, first in 1962-63 at Penn and then in 1968-69 at Penn State. As a graduate student at Penn, I wrote my thesis on mock theta func- tions under Rademacher’s direction. The material that Fine was lecturing about fit in perfectly with my thesis work and introduced me to many aspects of this extensive subject. The course was truly inspiring. As I look back at it, it is hard for me to decide whether the course material or Fine’s exquisite presentation of it impressed me most. Over the years, Fine’s work and the related course notes have greatly assisted me in my work. They were especially helpful in my study of Ramanujan’s “Lost” Notebook which overlaps the present book in significant ways. Research on q-hypergeometric series is significantly more active now than when Fine began his researches. There are now major interactions with Lie algebras, combinatorics, special functions, and number theory. 1 am immensely pleased that Fine has finally decided to publish this mono- graph, and I am grateful to him for allowing me to provide some chapter notes. This book has had a profound impact on my career, and I am glad to see it become available to the mathematical public generally. George E. Andrews This page intentionally left blank Introduction ‘The theory of partitions, founded by Euler, has led in a natural way to series involving factors of the form (1 = aq)( ~ aq?) ---(1~ aq"). These “basic hypergeometric series” or “Eulerian series” were studied system- atically first by Heine [27]. Many early results go back to Euler, Gauss, and Jacobi. A short account is given by Bailey (5], who has made many contribu- tions of his own. Ample references will be found in [3], in Hardy and Wright [25], in MacMahon [30], and in Dickson’s History [11]. Among the later systematic developments of the theory should be mentioned those of Hahn (20, 21, 22], and Sears [39, 40, 41]. For very complete references and expositions, see Andrews (1, 2). ‘The subject of basic hypergeometric series has been developed to such an extent, with such a profusion of powerful and general results, expressed in so compact a notation, as to make it appear quite formidable to the neophyte Nevertheless, the beauty and surprising nearness to the surface of some of the results could easily lead one to embark on an almost uncharted investigation of his own. It was this course that I followed, starting with a modicum of casual information, many, many years ago. By taking this approach, it was inevitable that I should be rediscovering much that was known to even the earliest workers in the field. Still, it was encouraging that many results obtained in this way appeared to be worthwhile and new, and that old ones dropped out as easy by-products. The present work is an outcome of this exploratory journey, and is a welcome opportunity to share with others ny joys of discovery. At least at the start, I have restricted myself deliberately to the study of a power series in ¢ with coefficients having one Bulerian factor in numerator and denominator. This funetion, (1~aq), | (L=ag)(1 — aq?) 12 F(a,b;t) = 1+ ———t + ~-——_—_ t+. (eS rig)! * = ba) = 09?) is a special case of the Heine series. It satisfies first-order linear difference- equations in each of the three arguments, such as (1-OF(a,djt (1b) + (0 — atq) F(a, 6; 19). xii INTRODUC ON These are given in §2 and §4 of Chapter 1, and in §5 there is presented an elementary method for using them to obtain transformations of F. This method is exploited in §§6-14; the next three sections contain special developments to be used later. In §18, the series for F(a, 6; ) is extended to the left, and the resulting bilateral series is summed as an infinite product. [ater found that this elegant and fruitful result goes back to Ramanujan. The exploitation of a special case permits one to evaluate the coefficients of several classes of infinite products in terms of divisor sums (§19) (see also §10). ‘The latter are generally of the form Ey.s,...(N:p). denoting the excess of the number of divisors of N congruent to r,s,... (mod p) over the number congruent to —r,—s,.... §§20 and 21 contain a small start in the direction of a general transformation theory for basic series in many variable: The treatment is still quite elementary, but some of the results seem to be new and interesting. The basic multinomial (§21) is a generalization of a finite sum given in two special cases by Gauss, who used them to evaluate the Gaussian sums. ‘The developments of Chapter 1 lead naturally, by paraphrase, into the arith- metic domains of partition theory (Chapter 2), theorems of Liouville type (§§28, 29, 30), and sums of squares (§31). Contact is also made with the mock theta- functions of Ramanujan (Chapter 3); these are linked with the rank of partitions, introduced by Dyson [13] and treated by Atkin and Swinnerton-Dyer [3]. In §32 I give a number of examples of modular functions A + BC(N)q™ in which the coefficients C(V) are multiplicative. These are, of course, related to the Hecke theory [26]. In all our cases, the modular functions are products of Dedekind 7-functions, of a class defined and studied by Ramanujan (see [23, 24)). As one example, which I regard as very beautiful, I quote (32.5): gta = oP) g ; Il oe! = 14D Bi szai(N:24)0". net Chapter 5 contains the beginnings of an elementary constructive approach to the field of modular equations. §§40, 41, 42 deal with a system of curious identities related to theta-functions. Over the years I have been assisted greatly by many persons and institutions. Among them, I wish to acknowledge the Office of Naval Research during the year 1949-50, under Contract N ONR 90,000, the National Science Foundation when I was a NSF Post-doctoral Fellow in 1953-54, the Guggenheim Foundation when I was a Fellow in 1958-59, the University of Pennsylvania and the Pennsyl- vania State University when I was on their faculties. Finally, T am tremendously indebted to George Andrews, without whose assistance and forceful encourage- ment this work would not have seen the light of day. In addition, he has greatly enhanced it with his scholarly Chapter Notes. The references in square brackets in the Notes are to his list of References following each Chapter. Those in the text refer to the Bibliography. INTRODUCTION iti Some of the material was presented by me as a Hedrick Memorial Lecturer in 1966. At that time I was invited to expand it as a Survey and accepted, but was unable to fulfill my commitment because of ill health. The American Mathematical Society was kind enough to revitalize that invitation more recently. For this I am truly grateful. This page intentionally left blank To Ruth, Naomi, Emily, and Isabel This page intentionally left blank CHAPTER 1 Fundamental Properties of Basic Hypergeometric Series 1. Definitions. We propose to study the series (= ag)(1~ ag?) += (1 ag") 1d F(a,b;t sq) =14+ Yo ON Tn a (osbstsa) = 1+ So Goria?) 5) In general, q will denote a fixed complex number of absolute value less than 1, so that we may write q = exp(nir), where I(r) > 0. By g° we shall always mean exp(crir). For n > 1, we define net (12) (@)n = (a5 4)n = (1 a)(1— ag) ---(1~ ag"), and for n= 0, (1.21) (a)o =1. We shall also use (1.22) (@)o0 = (2;4)o0 = [] (1 = a9"). n>0 In general, where there is no danger of ambiguity, the parameter q will be omitted from the notation. Thus (1.1) may be written (29)m pn Lon nB0 (uv) F(a,b;t) We shall also have occasion to use the Gauss polynomials: (mn (13) [7] =) Damn OS"S™ " 0 (0 0). Multiply (2.3) by ¢"+? and sum for n > 0; then SD Angst 1 = BO Ansa (ta)! = tO Ant” = ata SO An(ta)”, nB0 nB0 nB0 nB0 which leads easily to the required 1=b | b-atq — F(a,b; tq). Toe} Tae Mahia) Equation (2.2) allows us to advance the parameters (a, 5) to (ag, bg), and (2.4) advances t to tq. We shall refer to these as the transformations (a,6) — (ag, bq), t + tq, and so on. (2.4) F(a,b;t) 3. The analytic character of F(a,b;t). Thus far we have been operating in a purely formal fashion, and a few remarks on convergence and analyticity are in order. The partial products (aq)», converge for all values of a, as may easily be seen from the absolute convergence of 3> 4". Hence, if b is not one of the values g~1, q~?,..., the coefficients (aq), /(bq)n are bounded, and the series (1.1') converges for all t inside the unit circle, and represents an analytic function there. Hence the function on the right of (2.4) is regular in |t| < |g[~! except, (possibly) for a simple pole at t = 1. Thus we obtain the continuation of F to the larger circle. It is then easy to apply (2.4) again to continue F to the cirele lt] < Jal-?, and in this way we reach the conclusion that for 6 # q-", n> 1, the only possible singularities of F are at the points ¢ = 9-", n > 0, which are simple poles in general. As a function of , F is regular except possibly at the simple poles b = q~", n > 1, provided that t # q~", n > 0. Finally, as a function of a, F is entire, provided that 6 and ¢ do not have one of the singular values mentioned above. 4. More transformations. By combining (2.2) and (2.4) (with a, b replaced by ag, bq in the latter), we obtain easily Le atg , (1=ag)(b~ ata) (4.1) F(a,bit) = (1 = 6q)(1 = 4) tqF (aq, bg: tq) Let us write (4.2) (1= 0)F(a,b;t) = 1+ SO (An - SI Recalling that Ao = 1 and 1 bg" An- in 2 15 pagan (021) we have ‘THE METHOD OF ITERATION 3 Hence, from (4.2) (1- QF (a,b;t) =146 (a9)n=1 (yn (ea nas that is, (4.3) F(a,b;t) = + baad va. bg; tq). ~ 1=t* (1=bg)(1-9) Combining (4.3) with (2.4), we obtain (4.4) F(a,b;t) = and from (4.4) and (2.2) (= Hag, (A= aa) — ata) (45) F(a,b5t) = aa = F(aq.b;t). Finally, (4.5) and (2.4) yield 1-6 (b= atg) (=) ( ~ rant} 4, (= aa)(b~ ata) ~ ata?) (0 - ag) It is obvious that we can now obtain any transformation of the form (a, ,t) > (ag™, bq”, tq?), where m, n, p are arbitrary integers. (4.6) F(a,b;t) F(aq,6;tq). 5. The method of iteration. In this section we indicate an elementary method for solving difference equations of the form (5.1) Jn =Ln+MnJnti (n> 0), not for fy but for fo. This method will be useful in obtaining new forms for F(a,b;t). It is clear that by use of (5.1) for n = 0, 1, 2,...,.N, we may eliminate successively fi, fo,.--, iv, to get (5.2) fo=Gnt+Hwfwir (N20), where (5.3) Go=Lo, Ho = Mo By (5.2) and (5.1) (with n = N +1), we have fo = Gy + Hy (Lw41 + My+ifiv42) = (Gn + Hw bi) + (MvsiHw)ing2 = Gui + Anyi fne2 Hence (5.4) Gy =Gy + LyyiHn, 4 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES (5.5) Hy41 = Mysilty. From (5.5) and (5.3) we find (56) Hy = MoMiM2---Mv, and summing (5.4), wat (5.7) Gy = Lo + Yo brs (MoMi ++ My) 70 Now suppose that fy + J, Gv + G, Hw + H. Then (5.8) fo=G+Hf. It will sometimes happen that fy, Gy, Hy do not converge separately; it may then be possible to throw the combination (5.2) into such a form that its limit (which obviously exists) is readily recognizable, or to allow N to increase over some particular set of integers. 6. Application of iteration (t — tq). Let fo = F(a.b:t), fn = F(a,b;tq"), Ln = (1=)/(1=tq"), My = (b—atg"*")/(1—-tg"). Replacing t by tq” in (2.4), we obtain (5.1). We find that Ly + 1—b, My — 6, fn ~ F(a,b;0) = 1, and " ()— atgr}) Ha = [[ 4. Uf (1= tq") We must now consider several cases, depending on the value of b. First, suppose that 6 = 0. Then Hy + 0,80 F(a,0;t) = lity oe Gni that is, (nn) /2 (6.1) F(a,0;t) = reer ave a 30 Next, suppose that b= 1. Then Ly = 0, Gn = 0, and F(a,1;t) = litinoo Hn that is, (6.2) F(a, 1;t) = fie Equation (6.2) contains as special cases a number of well-known identities, three of which we state for later reference. The case a = ° yields (6.21) F(0,1;t) = > — a a os Now put a= q", m > 0. Then, recalling (1.3), we have (6.22) Fant = ["*"| m= n nz0 Omri 2% or by a slight a (6.221) =>[ Tc a” (N21). was se ITERATION OF 6 — bq 5 Finally, putting @ N+1)N > 0, replacing q by q~!, and simplifying, we get (6.23) (ta) ¥(% Ye tg 4H2 (N20). t=O Equations (6.221) and (6.23) are the basic analogues of the binomial series, to which they reduce termwise for q = 1. The condition for convergence in (6.22) is |t| <1, and in (6.221), [ul < |gl~*. The series in (6.1) converges for all a and t.t#q-™ m>0 Returning to our discussion of cases, suppose that 0 < |b <1. Then Hy — 0, and factoring 6" from the nth term in the resulting series yields (63) F(a,b;t) = ae (F ts °) Thus the function (1— £)F(a,b;2) is invariant under the involution a! = at/b, b' = t,t! = 6, Equation (6.3), proved for 0 < |b] < 1, holds for all values for which both sides are regular. If we multiply both sides of (6.3) by (1 t) and let t + 1, we get (29) 20 (69) 00 by virtue of (6.2). This could have been proved directly, of course. (6.31) Jim (1 — 1) F(a, 6:4) = (1— 8) F(a/, 138) = 7, Iteration of b — bg. Here we assume at # 0, and take fy = F(a,bq": -_" _ (6 = a)t In grat’ Mu= GTigntty(bgr Sal’ ‘Then equation (4.4), with 6 replaced by bg", assumes the form (5.1). We find (= (b/a)q") (= ba H(A = (B/athary" _(bfat) ss _(b/a)e On =F fat) 2s Toau(ba/eDe ‘Also Jn — F(a,0;1). If, by a slight change in notation, we set _ 7 = —(b/A)e0 (7.1) S= F000, HG eae (fat) ys (ln pera T= G7 Waalbafats Tanloajan,?? P= Flabo, then we have (7.2) F+Ge= ‘Two special cases are of interest here. First, multiply (7.2) by 1 — (b/at) and set a= b/t to get . 7.3) = CS Tales (Vee es _ F(o/t,0:t). 6 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMBTRIC SERIES Recalling from (6.1) that To (-0)" (7.31) F(b/t,0;t) (an and letting t + 0 in (7.3) and (7.31), we obtain PL yng int tns2) a Lien” Waa ren For 6 = -1, the left side becomes F(0, 1;q : q?), which can be expressed as a product by (6.21). Simplifying yields (7.321) Yar ene 30 a well-known formula due to Gauss. For b = 1, we have mo 2 (7.322) gL (-1ytghs*+/2, x (QR (a) x This result appears to be new. It is related to a similar formula due to Auluck [4]. If we put b = q~'/? in (7.32) and then replace q by q?, we get 2n q 1 gn? 7.323} = yeyng”. aaa) Loe Wa n20 n>0 This has an interesting arithmetic interpretation, which we shall discuss later (22.21). It is easy to see that the left side of (7.323) can be written as 1 Lf 1 (F(0.1:9) + FO,1;—a)} = las + ca}: 2 Thus, eventually, we get $20 gn? 1-q" en 2n—1)2 (7.324) (<1 a =[Ja-?a- y, or, on replacing q by —a, $00 (7.325) Mo’ = Ta-eatgry. SS onbt ‘These last two are standard results in the theory of the elliptic theta-funetions, and are special cases of an identity of Jacobi’s that we shall prove later (§17). ‘The second special case of (7.2) is obtained by setting a = 6?/t; then ngangin?+n)/2 S = F(t /t,0;t) = Ot aes __ 1 Galea Jox" oo (Dn T= 2s ann (i)rms F=F(P/t,b; CONSEQUENCES OF §7 7 the second form of F being derived from (6.3). Now t — 0 leads to 8 = Cepnengr, og, 1 gq T= Xa F =(1-6)F(b,0;6) It is possible to derive a simple power series expansion for h(b) = (1—b)F(b,0;.). Indeed, (4.6) yields (7.6) h(b) = 1 — bq — bq? h(bq). If we substitute a power series for h(b) in (7.6) and equate coefficients, we obtain (7.7). (1 = B)F(b, 050) = So (—1)% gl" 08" 4 J (Hyg nV B33, n50 SI An important limiting case is b+ 1. Using (6.31), we get too (78) Tle -a) = Sayre, nei = a famous identity due to Euler. 8. Consequences of §7. In (7.1) and (7.2), set a= —b/ut to obtain D> (oY ont emia, S = F(-b/ut,0;1) om (8.1) F = F(-b/ut,b;t) = (=) F(-u"',t;6), with F +G@ = HS; the second forms of $ and F are applications of (6.1) and (6.3) respectively. Now let t + 0; we get S =D (burtynge ena, nd0 H = ((bq)o0()e0) (8.2) F =(1-6)F(-u"1,0;6). 8 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES For u = 6, (8.2) simplifies to S= Yarn, n50 1 1 a iol = Bgan’ 6 2. 2 pep OP a0), (Pn) /2 F =(1-0)F(-b-!,0;6) = > 2 (1 H)F( ) x Tas ‘The equation F + G = HS now reduces to (n?¢n)/2 —b-!.0:b) = q F(-b 0) = 5 ea (8.4) me =s Pog: q? = ST mgm + op (0.0 0:4"). If we expand the second member of (8.4) in powers of b by (6.221), we get (8.41) sb) = vist, #50 where ntk] (nt = (n?+n)/2 (8.42) %-D| n ]« . On expanding the third member of (8.4) by (6.21), (6.3), and (1.1), we obtain (8.43) F(-6-1,0;6) = 8 > nB0 Equating coefficients of b* in (8.41) and (8.43) yields pen Pe (8.5a) Vaz = am Yan Toya) 0 It is easy to see that Ve converges as k + 00; this yields another proof of Gauss’s formula (7.321). ‘The results of this section will be used later to derive a theorem on partitions, stated in [16] 9. Further consequences of §7. In (7.1) and (7.2), replace 6 by aq to obtain (9.1) tF(a,0;t) = deel dee { F(a,aq;t) + > mae net Ad)n| (Yoo FURTHER CONSEQUENCES OF §7 9 Now put a =~ ‘The infinite product on the right is obviously an even function of t, and so is the series. Hence (9.12) t{F(—t71, 054) + F(t71, 0; -1)} ao (07? Whoo { (9)o0 Now by (6.3), t T+ the last expansion being valid for |t| > |g|. If we subtract the similar expression with ¢ replaced by —t, we get + . tg™ 1+tgr } Hence, in the ring |g] < |t| < |gl~?, we can expand in a double series tat Eaten = (9 kemB1 2+ Dek Seay yh, NB1 0 kn=N a ote do eeeq(-ay" = (-)*}, mast NBA since the inner sum vanishes unless n and k are of opposite parity, so that their product must be even. This inner sum may be written 20 ee k n=2N oda Letting w denote an odd divisor of 2, therefore of N, we get 2 L(B New = antsy, ein 10 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES Hence (9.13) eC aot (7797: 47 Joo. aN 2N, -2N, = -2 Soon — po awhan Ge eee ‘The quantity in brackets on the left side of (9.13) is twice the even part of F(-t71,0;1); by (8.4) this is 28(1 — t?)~1(t?q?; q?)52. Recalling the value of S from (8.3) and (7.321), we have S(q)oo = (q?: q?)3,; finally, 2t 1 oo (t?4?3 4? Joo ye Tees = NB1 GIN (9.14) 2Njusw) the formula being valid in the ring |q| < |¢| < |g|~!. Replacing q? by q through- out, we have, for |q| <|t\? < |a|-?. 2 (9.2) (Woo. a1e 0-0) Da LOMO — ae), ai ow If we let ¢ =e, u real, this becomes Ms gg wy, (2N (93) Fa 29 costa ay tsinu Dod re Naa) This formula contains a wealth of information, some of which we shall unearth later. It is essentially the Fourier expansion of #4(0 | r)/#:(u |r), with q = exp(2rir). 10. A product-series identity. We shall now prove an identity that con- neets certain infinite products with series whose coefficients have arithmetic sig- nificance. Let p be an arbitrary integer greater than 1, and let r be chosen relatively prime to 2p, with 0 gl2emtnark-p-n) eg, ket where M = 2pn +r (n> 1), N = 2p(k—1)+p—r (k> 1). That is, Mer (mod 2p), M > r; N = p~r (mod2p), N > 0. Taking into account the term (1-47)? in (10.21), we get (10.3) O1(q?) = 4) 1 (10.4) ger + o1(q?) 530 gM, where now the sum runs over all M, N > O satisfying M = r, —r (mod 2p). Collecting all the terms which involve q?”", 2pn = MN —r(p—r), n > 0, we see that M runs over all the positive divisors of 2pn + r(p — r) for which the complementary divisor is congruent to p—r, and M = r (mod 2p). The latter condition automatically implies the former, since (r,2p) = 1. Hence Late =Dem N30 diapnctr(p-n) d=r (mod 2p) see] q (10.5) 12 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES Similarly, it is easy to show that (10.51) (7) =o ak. nB0 —dl2pn-¢r(p=r) d=—r (mod 2p) Finally, replacing q?? by q, we have, for 0 1; an an =r (mod m) + (mod m) in general, (10.71) jase2(Npm) = E,(Nym) + E,(N;m) + +--+ E,(N;m). Formula (10.6) is the one we set out to obtain. It is clear that it should be capable of many uses. Two types immediately present themselves: (a) those cases in which the product has another expansion, so that we are led to an arithmetic identity by equating coefficients, and (b) those cases in which the coefficients EZ lend themselves to further transformations, which are then reflected in identities among infinite products. We believe that (10.6) is new. 11. Iteration of a + ag. A routine application of iteration to equation (4.5) yields (11.1) F+G=HS, where S = F(0,b:t), — (ag)so(ata/) B= (cain (11.2) F = F(a,bt), 5 = U= Dealt) 5 (aa)n(ata/Dn “(= ag/6) (atq?/b)n For a = b, this reduces to Ly Waltln n — (Odeo) 20 py (11.3) = Tent = lee Flu b) An interesting special case is obtained if we set 6 = ¢~! in (11.3) (ua) (lee orn Wales ” ITERATION OF (6, t) — (ba, ta) 13 12, Iteration of (6,1) — (bg,tq). If we write R(a,b;t) = (1-1) F(a,;t), equation (4.3) becomes (b= a)tq (1 = bg)(1 = ta) Carrying through the iteration method, and observing that Hn > 0, fn = R(a,bq"; tq") — 1, we obtain (12.1) R(a,b;t) = 1+ R(a, bq; t9) ft) = (b/a)n Jn ginten)/2 (12.2) (1=t)F(a,b;t) = (-at} . DFlosbt) = Galata 04 This is a generalization of (6.1), to which it reduces for b = 0. If we let a — however, we obtain (12.3) (1-H F00,h1) =o nB0 Now t + 1 yields, in view of (6.32), (12.31) a a eam Oa" oF Tf we let b= ¢-1 78, (12.3) reduces to (1 —e') F(0, ee) (12.32) _ =14+00 (= 24 c080 +g) (1 — 29" cos +g") nel Put 6=¢ in (12.3). Then (12.33) (1-1) FO.t:t) = > Let r and m be positive integers, and replace q by q™, t by q” in (12. using definition (1.1) for the left side, we obtain the curious identity 3). Then, “ (12.331) mn®-+2rn, q of which (12.311) is a special case. 14. FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES Let us define (12.4) 9) 4 Faia) => 7 neh It is well-known and easy to prove that (q) is the generating function for d(N), the number of divisors of N: (12.41) (a) = Do a(n). we Putting a = 1, b= = in (12.2), we find (12.42) We may expand the extraneous factor (1 — q")~! to get the double series kr gin? +n)/2 (12.43) = Day Fortes To evaluate the inner sum, let t + 1 in (6.1) and use (6.31) to derive the useful formula sa)nginttn)/2 (12.44) (a7)20 = 0 (a)rg 530 (Q)n Replacing a by q* reduces (12.43) to the form (12.45) (9) = Da - at) ¥(!- Ho-m) R20 20 n>k 13. Iteration of (a,t) — (ag,tq). For 0 < |b| <1, the iteration method applied to (4.6) yields 1-t) ,,, = (aq)n(atq/b)an pm GS) Fla) = x (nl (aa/Bn Y nB0 = ag > Letelata/Pan. gg Xe Wnlaa/Pnns (13.1) By (6.3), the left side of (13.1) is F(at/b, t;b); making the substitution a’ = at/b, b' =, =6, then dropping the primes, we obtain a new form: (atq/b)n(aq)an jn ZX Calea/bn _ tg 5 (ata/P)n(ad)anss gyn © Le Galatea valid for |t| < 1. Letting +0 in (13.1), we find F(ab; (13.2) (13.3) (1-t)F(a,0;t) = yr Cale (-at?)"(1 = atg?t? )gl3n?+n)/2 t50 (Hn ITERATION OF (a,b, t) — (ag, bq, ta) 15 If we now let t = a, we recover (7.7), and if we let + 1, we get (ca) x (Qn nB0 (13.4) (aq) 00 (1 agin +h) g(9n?-4m/2 On the other hand, if we let ¢ + 1 in (13.1), we get the more general (ad)oo _ > (29)n(a4/b)an pn he 2 (alulaa/Bn ” — ag LeAn(a/P)on +4 yon 0X anlaa/Pccr OO (13.5) For |a| < 1 and 6 = a, (13.5) becomes = [FR] tarnat = [PF Coaataay™. no 730 (13.51) (1-a)~ As a check, if we let q = 1, here, as we may, we obtain the not completely trivial identity (13.52) a-a'=¥ (7) oy ("7") 130 nB0 where we have set 2 = a(1—a). 14, Iteration of (a,b,t) > (ag, bq,tq). If we assume that b #0, iteration of (4.1) is easily seen to yield (14.1) (1 t)F(a,b:t) = > nb0 (29)n(atg/b)n - 2n+1 nan? (7 a This identity contains much information about the function F(a, ;), and many of our earlier results are easily derived from it. For example, (6.3) is an imme- diate consequence of the fact that the right side is formally invariant under the transformation (a,b,t) — (at/b,t,8). If we let t+ 1 in (14.1) and use (6.31), we get (29) 00 (2q)n(09/6)n 2 2 14.2) 1 —ag?"t*)b"q™, ad Gx 2 Chaln(de a of which (13.4) is a limiting ease, for b — 0. Equation (12.3) also follows directly from a = 0 in (14.1) Set t =a in (14.1) 2 2 (143) (=a) F(a,b,a) = oy EAD a gg? y(ad)ng” Ay Wn This simplifies if a2/b = 6, that is, if 6 = ta. For b= —a, we have (14.31) (1 = a)F(a,-a;a) = 14250 (-a?)"q™. not 16 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES Thus, a + 1 recovers for us equation (7.324); a = 7 yields $00 In (14.1), replace g by q? and set a = bg~!, t = bg: we find easily that (14.4) F (ba~*,b:bq : q?) = So org" *")/2, nz 's formula (7.321) by setting 6 = 1 and using (6.2). (14.32) (1-aF(,-é Here we can recover Gaus Another curious result is (14.41) Para aia’) = oa”, 730 that is, U-90-4) oy. (14.42) *a-pa-ey? + with (14.32): (i+a)a° 1+@ ° (L+@)(1+44) (1+ g)(1 + 9?)9® Teese A 1+ (14.43) nB0 We close this section with one more example of the versatility of (14.1). Put t= there: (ag)n)? antiyya (14.5) (1 = b)F(a,b; 6) = Lad VO — abg?h pn . x(a) ae 4 Now if we set a= 1, b= q and use (12.4) and (12.41), we obtain the well-known result of Clausen [10] 5 1+e (14.51) Yaya x -x; +0) 9" NEI which can also be proved directly by expanding the right side in powers of g. 15. A special development. ‘The function (15.1) G(t) = (HhooP (a,b;t) = YO Ant”, 230 with A, independent of t, is clearly entire. We shall determine the coefficients An. By using (1.1) and (12.44), we see directly that Nx =o eyeloas ae An = OCW Gano Oe git? -#/2 THE PARTIAL-FRACTION DECOMPOSITION 17 On the other hand, substitution into (2.4) shows that (15.3) G(t) = (1 6)(ta)a0 + (b= atg)G(ta), from which we deduce, by equating coefficients and using (12.44) again, that for n>1, inten (15.31) An (1 = bg") = =aal Anas + (1) = 0) ‘The obvious substitution ( —— Bn (bq)n reduces (15.31) to By = By t Ota, (Q)n whence, summing for n = 1,2,...,.V, we obtain Nn (On 15.4) By= ue) w= Do (Q)n a)! “oe ENy/2 (15.41) AN yo 5 an” ‘Thus we have the identity (cat)Mq(¥?+90/2 BM (by DD, (15.5) (YooF (a, b:t) = > On @n ‘Two special cases are of interest here: ( 0). ae. From (2.4) it is easy to see that (= a°")Cn = (b= ag ")Cn1 (m2 1), so that and from (6.31) (16.2) (n>0). a ororenernctenoeenerne lente neeaceeneraer sere Equations (16.1) and (16.2) suggest that . (aa)ee 5 (6/2) (00)" 16.3) F(a,b;t) = . : (abit) Galas Den T= and indeed this identity is easily verified, either directly, by expanding (1 = tq”)-}, interchanging the order of summation, and taking into account (6.2), or by observing that the difference of the two members of (16.3) is an entire function of t that tends to zero as |t| + 00. Similarly, by matching the singularities of 4(t) = (q)eo(1— at its poles t = g" (—co < n < oo), and comparing with 1 (ayn gh)qnttmv2 v(t) = + a ss = Goqatey * eae) it is not difficult to show that 4(t) = #(t). Thus we have established (1-t)F(0,t7!;t) (16.4) 1 py nant) Mt ago 972 - dz {i 10 Ta-eey f —1)-1. F(0, 47-4; t) 17. Jacobi’s triple product. Many proofs of this famous formula are known; we give one here that seems to fit in with the spirit of our approach. Consider (17.1) 9(@: 4) = (—a7*9)o0(—@) oo. By (12.44) we may write g(a: 4) It is clear that the inner sum Sy = S_yy, so we need consider only nonnegative values of N. For such values we have ; grghn 1 argh Se = 2 Galdmew ~ Tw Sy One a 1 1 ~ @x oo oo the next to the last step being a consequence of (12.31). Hence tee (17.2) (Q)co(—4"!g)a0(—a)o0 = J aNgl8?-807, A BILATERAL SERIES 19 Replacing q by q? and then a by aq, we obtain the usual form $0 ; ¥ ote”. An alternative approach is to derive a functional equation for the left side of (17.3) under a + ag?, assume a Laurent expansion with coefficients Dy, show that Dy = C(q)q%", and evaluate C(q) by taking a = 1 and using (7.325). (17.3) TL = 9") + ag?4)(1 + a 1g?""3) a3 18. A bilateral series, We shall now extend the definition of (aq)n to negative values of n. With this extension we can then define a bilateral basic series. Our definition is by induction: (18.1) (ag)n41 = (1—ag"**)(agjn_— (00 0), and for all n, =n gin?+n)/2 (18.11) (aq)-n = se Hence we may consider the function te0 ; (00) yn H(a,b; = Cale 8.2) mH oa = F(a,bt) + > cae (b/at)” n2i ™ = F(a, b;t) + F((bq)~*, (ag): b/at) = 1 ‘The possible singularities of H as a function of t are poles at t= 97", (b/a)q”™ (m > 0), with (in general) essential singularities at ¢ = 0,00. From (18.1) and (18.2) it is clear that H satisfies the equation (18.21) (1 = 0)H (a,b; t) = b(1 — atg/b)H (a,b; tq). Now, defining the auxiliary funetion (18.22) G(a,b;t) = C(t) = H(a,b;t) T] (1 -ta") (: - oa") m0 we have (18.23) Gt) = ~atgG(ta) The only singularity of G(t) in the finite plane is at t = 0. Hence, for t #0, we may write (18.24) Git) = 5 Bat", and from (18.23), (18.25) By = (—a)"ql""*")/) By (-00 0. Therefore (by (18.26) and (18.22)) G(1) = Bo T] (1 = 9™)(1 —a-4g™"4)(1 = ag™) mz1 = Te (- oer) Jim (1 ~ t)H(a,0;t) = T1e-# (1-3) Jim (1 ~ t)F(a,b5t) = Ty eg = bata) = ag “Il (1 — bg) . ‘Thus we have (827) Gt) = T] Q=a")a= mei ( or, reverting to H(t) and our standard notation, F (4)c0(b/@)o0 (a7! "oo (too 18.3) H(a,b;t) = eelblaoo(a Mt ")oo( ata) 0 : (0 = Cba)oo(O "aol Jeo(8/At) ac This formula is given by Ramanujan. It contains as special cases many interesting identities and has numerous applications. In (18.3), let a= bg~!. The left side, by definition, becomes H(bq7}, bt) = F(bq7, 8:0) + F(O2g7, “Es gl“2) — 1 (18.4) _ ” mT (gt-" =14(1-) > faa 01-8) > m2 i nz. 1 orige tetgn=ty(1 = atgh) (1 = bog Tat be") the two series converging for |g| < |t| < 1. If, in addition, |g| < [6] < lal, we may expand the denominators to get the double series bt (1=6)(1~t) kn (pk yn — p- kyon tne tas Be (oe — eRe) RBI (18.5) H(bg™*, b;t) = After simplifying the right side of (18.3) for a = bg~} and equating with (18.5), we get the identity vac aia 8 inten arte (18.6) rar (4)3o (bt) 20 (07 "#7 *4) 20 A BILATERAL SERIES 2 ‘The range of validity of (18.6) is easily seen to contain the region |g| < |b] < |al7', lal < (el < lal}. In particular, if we let 6 =e, t =e”, where u and v are real, we obtain essentially a formula from the theory of elliptic functions: We a" )2(1 = 2cos(u + v)a" +4") (1 = 29" cos u + g2")(1 — 2g” cosv + q2") (18.61) - sin(u/2) sin(v/2) 1a ot ‘= sin(ku + nv). my Specializing still further, we let v = —u-+e and make ¢— 0. We get 0 1~q")t (T= 2g cosu PP nt =1—8sin? 2 Tg _ =1-8sin’ 309 DL neos(k nju. REL oats (18.62) A comparison of (18.62) and (9.3) yields the identity 2 sinu a Dosa (2% a) u as oe =1-8sin?u 0 aN SO neos2(k —nju. NBL nkaw (18.7) For 6 #0, 00, ” (—00 1. The right side of (18.6) becomes (19.1) Il a net PAL — at gPm—(r+9))(1 — grr YE = PPL = am 1gP=S\(T — gh 3) oe ‘The left side becomes (19.2) SD gpmmtemtingn ST gemmrmangnn = So A(N)Q™, mn20 mn>0 NEO TWO PRODUCT-SERIES IDENTITIES 23 where A(N) = i an — > PN+rs=(pm+s)(pntr) —— pN+rs=(pm—s)(pn—r) mn20 mn20 ‘The sums run over all factorizations of M = pN +rs into dé, where d =r, 6 =8 (mod p) in the first sum. , 6 = —s (mod p) in the second. Since (r,p) d=r (modp) implies 6 = s (modp), and similarly for the second pair. Hence (19.3) A(N) = DO xe(d; plat”? = y_-(d; pam tr/Py, am where x, (d;p) = 1 if d= r (modp), = 0 otherwise. Therefore ket A(N) = a” SL wléy- YD veld) mz dine i (@-ri/pem(e) =(a40 /r=m(b) dine d=mp+r(kp) kot = YP a Emp +2(M; kp) ‘Thus we have established the jee mates) = agen—-1—9) oem (19.4) Pa = aX a Empsr(DN + rs: kp). 550 mao By a slight modification of the derivation of (19.4) with t = a (s = 0), we can obtain the following identity, where p > 2: TE tah) = agro) (> ger Pr a= tgPr (T= agr) ai (19.5) koa = 14 (1-0) So a¥ Yo a Emp r(Ni kp): Xt mzo For p=r+5, (19.4) becomes Ul (= g?")P(1 = ag?) = a *gP") UD rr re 0g TGP (19.6) ka = aha) SP a Bmp te (PN + 1(p — 1): kp) N30 24 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES For p = 2, this reduces to Tl (1 = 4?")?(1 = 2cos(2na/k)q?" + 9% (1961) "* y eA eames 2N +1: 2k). OSm<(k~1)/2 u 20. A general transformation. Given a sequence An, n > 0, such that (20.1) a(t) = Yo Ant” 30 converges, we define the linear transformation (20.11) Te(g) = h(t) = > ee". 250 (Pan We shall show that A(t) may be represented as a series in g(t), g(qt),-.-, with coefficients that depend on 6 in a simple way. In fact, for g(t) = t" (n > 0), ” = = (boyz (bq"* oot Tog) = tan (04)e0 (b9"* "oot yt pm MA = (b9)ce al aq") Tr ey by using (12.44). This may be written in the form geen oa"). (20.12) Tela) = (ba)ae (8 730 Equation (20.12), proved for the special case, is valid in general, by the linearity of Tp, provided that the series (20.11) converges. In all our applications this will be easily verified. Choose An = (a9)n/(q)n, 80 that (ata). g(t) = F(a,1;t) = (Doo then (tr (ata), g(t) = a(t). and (20.12) yields (aa)at™ _ (aldose _(_pratetnve (202) X Walaa ~ Bede Ze Caecarye OE” This contains some of our earlier results as well as some new ones. For example, with a =0 and ¢ replaced by gt, we get a _ rigor a Sites ~ Gata ae A GENERAL TRANSFORMATION 25 Setting 6 = —1, the left side becomes F(0,1s qt: 4°) = (ta; so (20.22) 5 Mike (40/9 = (atti Pe = T] Cae) 0 net For ¢ = 1, we recover (7.321); t = q?* and ¢ = q+" yield (8.5a) and (8.5b). For t =q7'/?, with q replaced by q?, we find (20.23) 1+ a =4°) ( a-wa-ayt 4G an) Veo If we put t =q in (20.2), we recover (7.3) essentially. Suppose we write a= ¢~¥c in (20.2) and let t + 0. We obtain easily gl —48)(1 —4°) U-a- #0 @)" * gitenyl2 (odheo 32-9) tala nz0 = (ea)oo Yo (-b) SH @nlean’ (20.24) gh emit ‘Thus the function on the left is symmetric in 6 and c. It is easy to invert the transformation Ta. Taking h(t) =", n > 0, we have Tz '(h) = 9(t) = (ag)nt” = (a9)0(a9"* "a2 = (cg) 2 SS (20.3) Tz (h(t) = (ag)o0 > (oa) 1(q"t) 5S Again, (20.3) is valid for general h(t) by linearity. By applying 7; followed by T!, we get — (a9)n 4 pn — (@9)o0 (a9)" (=0)* | (3?48)/2,(or4s, Tr = Daa) Ant” = aye oy ee (oa) (0) a ‘9(q"*°t) = (ag)20 5 (09) K) (BY) eres Coa)ec 2s (ae 9 or L( a) oer. Se Using (6.23) for the inner sum, we have = (odes 5 W/a (a)oo 2s (We (20.4) Tz 'Te(a(t)) )*a(a*t). 26 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES If we apply (20.4) with An = (¢q)n/(q)n, we obtain, as before, (2q)n(04)n yn _ oe (P/a)n(tn (gay 41) alte Calman Dey lela Hence, if we define (20.42) F (4 7 i= (tq) 20 (b9' nztiee equation (20.41), with 6, t replaced by bg, tq becomes ae _ p (oot (20.43) F( A w)7F( ot a): ‘That is, the function (20.42) is invariant under the substitution (20.44) a’=ba', W=ct, =a. Thus, if we write a = q4, b = q8, ¢ = q©, t = q?, (20.44) becomes a linear homogeneous substitution S: A'=-A+B, [ B C+D, (20.45) a i DI=A Because of the obvious symmetry in A and C, F is also invariant under Vi A= Cc, B= B, (20.46) Cn A Di= Dz ‘The substitutions $ and V generate a group of order 12, with the relations (20.47) SS=V? = (SVP =I. The transforms of the vector (A, B,C, D) under the group are: (20.48) 1:(A,B,C,D)y V:(C,B,A,D), S$:(B-A,C+D,D,A), VS:(D,C+D,B-A,A), $?:(C+D-B+A,A+D,A,B-A), VS?:(A,A+D,C+D-B+A,B-A), S°:(B-C,B,B-A,C+D-B+A), VS9:(B-A,B,B-C,C+D-B+A), S4:(0,C+D,C+D-B+A,B-C), VS*:(C+D-B+A,C+D,C,B-O), 8°: (D,A+D,B-C,C), VS° :(B-C,A+D,D,C). Now if A = —1, F reduces to a simple product, and the same is true, therefore, if any one of the first elements in (20.48) is —1. For example, from S?, with C+D-B+A=-l, we get (after a— ag! b+ bg7!, c= eg7!) @)aleln 2)" _ (ba/A)(b/@)20 an Daten (2 2) = aes ‘This summation formula is well known. A GENERAL TRANSFORMATION 27 We can also derive the transformation (6.3) from al l,at/b F =F (a ww) ( tq? sbgu* obtained by S4 with C = 0. There are many interesting results which may be derived from (20.48), by applying the transformations to special cases. For example, if we start with (20.5) e(% ) “Fa oo) =F (ae yp) =F( “at aa by S, $8, and VS? with C = A, and let a — 0 in the first, second, and third, then replace 6,¢ by 697}, tg7}, we obtain » (t)20(b0) 00 “2 DeOoe oe) On, (ura? = a aynglett0V = gh “Lao. = (ooo Daan Similarly, if we let 6 + 0 in the first, second, and fourth forms of (20.5), and then replace t by tq~', we get 0 (09V% yn S30 (Mn = ()n vn (20.52) = (aq) ce (ata)oc ~ Wala = (atq)oo (—at)q("?+"/2, aaa (an oa A double application of the inverse transformation (20.3) yields — (9(0) = (@a)oo So ATE gighry oe. ela) = (oimlere tg woe (Fy. nO n50 ve Hence, if we define (20.6) Jn(u) = x [;] w and write a; = au, (20.61) al) = (aa)o(eua)se Yo EN” #50 (Mn In(u)g(a"t), 28 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES Certain special cases of the function fn(u) have been studied. We shall develop some of its properties here. Let t = 0 in (20.61). Then (20.62) oe ae *) (ag)" = (aq)z! (aug)s! = W(a,u)- n>0 In Several functional equations for W are of interest. The first three, of course, are not independent (20.63a) W (aq, u) = (1 ~ aq)(1 — aug)W (a, u), (20.64a) W (a, ug) = (1 - aug)W (a,u), (20.65a) W (aq, u) = (1 - aq)W(a,uq), (20.66a) W(a,u~!) = W(au~*,u). By equating coefficients of (aq)", we get, respectively, (20.63b) Snzi(u) = (1+ u)fn(u) + u(l — a" )fn—1(u) = 0, (20.64b) fn(uq) — fn(u) + u(l = 9") fn—1(u) = 0, (20.65b) a" Sn(u) — fn (ug) + (1 4") fra (ug) (20.66b) fa(ut) =u" fa(u). ‘This last equation can be derived directly from (20.6) Now fo(u) = 1, fi(u) = 1+ u, and if we set w= —1 in (20.63b), we get, Insa(-l) = (1 4" )fn-a(= Hence n(—1) = (1 g)(1- 1-9 5d ns a fnl-) = (1 a)(b 98) 4 aa) fons1(-1) = 0. This could have been established directly from (20.62), since (2q2)" W(a,-1 a : Fen nz0 Also, the second part of (20.67) is a simple consequence of (20.66). Similarly, putting u = q'/? in (20.62), we get 1 rae fala) won={ Tha -e0ya - vl n30 net (ag)" =Laa, nd0 A GENERAL TRANSFORMATION 29 Hence 1/2 ()n (20.68) ae (Q?,q¥?)n = (14 q¥?)(1 + g)(1 + 9?) (1+ 9?) = (-q"/:q"?)n. It is easy to obtain other values. For example, by (20.66b), (20.69) fulq>¥/?) = a7"? fa(q’/?). We can also derive the exact generating function for f,(u). For this purpose, set An = 4"/(q)m in (20.61), so that g(t) = (tq)=? and g(q"t) = (tq)n g(t). Then TalTeHalt)) = Ye LDaaudn gay Xe = (ad) aad > PB fu), n>0 ™ and by definition (20.42), with b= 0, ¢ = au, (ta)n np (au _ p( ot (2q)20(aU4) 0 x Te tnentaa =F (SO =F (aut sa) the last by virtue of (20.43 we get (20.7) Y Ja(woa” = (1 - au)“ (0, au; a) = (1 — @)-1F(0,a; au) 30 . Setting ¢ = 1, simplifying, and replacing a by aq7}, As an application, set w= —1 and use (20.67): (14a), -a;a) = Y> fon(=1)a" = > (4;47)na™ nB0 nB0 = F(q7!,0;a? : q?). Thus (20.71) (1 +a)~1F(0,-a;a: q) = F(q71,0;a? : ?). If we put u = q!/? in (20.7) and use (20.68), we get (1=a)“*F(0,a;a94/?) = 37 Ju(g'/?)a" nz0 = Aa" g! pa" = FAL nd0 and replacing q by q?, (20.72) (1 a)F(-1,0;a: 9) = F(0,a;aq : g?). Both (20.71) and (20.72) appear to be new. The latter can be used to obtain transformations of some of the mock theta-functions of Ramanujan, and also to derive a theorem on partitions (23.91). 30 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES 21. The basic multinomial. The function f,(u) defined in the preceding section is of some interest as a generalization of the binomial (1+ u)", to which it reduces for q = 1. Similarly, we can define the homogeneous function of k variables (21.1) Qala eu) = mE ntegreen It is easy to see that Qn (ua. ue) 21.11 a = a a Two special cases are of immediate interest. First, let us = exp(2mis/k), 9 = 0,1,2,...,4—1. The product in (21.11) becomes ha = Gey, % iso an Equating coefficients, we find Qn(ur--.,ue) =0 for n #0 (mod k), Qn = aHe— for n =0 (modk). Fitna Next, let uy = 9")/? (r= 1,2,...,p). The product in (21.11) becomes 1/py= Ge a;qV?)5) = ; (aia'”?) zy (GPs qi*), so that, on equating coefficients, we get (21.2) (us = exp(2nis/k)). (21.21) Qn (1a? 97/?,....g?-0/P) = ween, A combination of these two special cases leads to a more general result. Let k,p be positive integers, and let (21.22) Ups = exp(2mis/k)g—Y/P (r= With these kp arguments we find, as above Qe am = (ate qtlPy-t = fa e221) Ean a = 2 Gaga, m0 nz0 ++5B} 8 =0,...,4- 1), Hence Qm =0 ifm £0 (modk), (21.23) (nk On = Gem, (M20) Now the function Q, is related to the N-fold application of T,-!. In fact Tants Town *** Taw (9(t)) = A(t) DL laug)n-- (aug) Ant™ (21.3) nz0 = (ety G)o0° (ot a)n > ZOE (anata) n>o Nn ‘THE BASIC MULTINOMIAL at Take An = (¢q)n/(a)n+ 80 9(t) = (¢la)oo/(Hoo and g(tg") = 9(t)(t)n/(cta)n. Then (21.3) becomes (aura) + (auva)n(a)n yn x @n = (MUrdoo "(aw d)o0(Ct doe " = Dee See eae ne0 (21.31) Now choose the u’s as in (21.22). Then Teewean = = ((ag)*q"/ np, so (21.31) becomes, after using (21.23) and replacing a by al/*q-1 (250°? \np(Ca)n yn _ (0 0!” Joo( Ct) 20 (Onka” 21.32 (esq dnpkea)n yn _ (Ga ool et)oe S% (nk _ 2) we We Rance Now replace */? by q1. Then (2:41) np(ca)n x (mn 541 )o0( CF) oo (Qnka” (Ooo £54 (15 G1)n(eta)ne” (21.33) a ‘This identity in q and gy, as it stands, has meaning only for q1 = if we use it may be written as (64:4) oe (9. )e0 x (agi?ian ——— © (ado S (21.34) Further, let y= qf = q*. Then we have (with e+ eq™!) (c:q)oo (g's doot™ edn 2a ay"; 41 )oo( C4": Doo (21.4) = nd. =! ye (915 91 Joo Sh Now we may consider the validity of (21.4) for independent values of g, q1, y. Since the special values q, = q*/?, y = q* are sufficiently numerous, an analytic continuation argument will work. Alternatively, we can expand the left side in powers of a and compare coefficients. Either way, (21.4) is established in full generality. It is an example of a bi-basic identity. 32 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES ‘The methods of §20 can be pushed a little further. Suppose that, for g(t) = Ant”, we define = h(t) = (bY" Joo pn (21.7) S(g) = h(t) = x ya It is then fairly easy to show that (21.8) at) = Clan amgiyn), 2 Wn similarly to (20.4). Choosing An = (cy:y)n/(y:y)n. we get DD: nzo cusu)n , (by")oo pn (ysu)n ay" Joo = (ety:v)o0 5 _(/a)n(t.n_gn (G)n(ctys¥)n (21.81) After replacing ¢ by cy~! and making appropriate changes in notation, we find another bi-basic identity: es pn $Y )00(@Y"; 90 (21.9) mth (q’ oo (ety’ Jan 2, CaP aca} v) a” Notes §1. The most natural question that occurs in studying this book is: Why does stig) = So adnt™ F(a,b;t:q) x Tale have so much structure and yield such diverse and interesting results in such a natural way? A vague philosophical question like this is impossible to answer fully. Indeed two other mathematicians, L. J. Rogers [26, pp. 334-335] and G. W. Starcher {28}, both saw the importance of F(a, b;f:); however neither pursued its proper- ties in anything like the exhaustive manner followed in this book. I feel something of the magic of F(a,b;t:q) must lie in the following observation: Let us consider a distribution of weights w at the points 2 = g' (i = 0,1,2,...), where 0 a Real aa (b/a)i_ aaa 2 gigi +Vi(6/a), (i by (6.2) of Chapter 1) — (9) c0(b9)o0. = Dfa)so(aaias :@). The orthogonal polynomials pp (x) connected with this distribution are = slb0") (20)? Pal) = estan These are q-analogs of the Jacobi polynomials sealed to the interval {0,1]. They were first. given by W. Hahn [18], and many details about them appear in [15] Since these polynomials can be specialized to all of the classical continuous or- thogonal polynomials and since their moment generating function effectively encodes enough information to define them, we should not be surprised at the richness and importance of F(a, b; t:4). The basic properties of Gauss polynomials are given in Chapter 3 of (8]. §4. Equation (4.1) was given by L. J. Rogers [26, p. 334, (1)]- §6. Equation (6.1) was given by Starcher [28, (3.16) with 4 = 1]. Equation (6.2) is simultaneously first the q-analog of the binomial series [18, (3.3)] and second the q-analog of the beta integral (18, (3.11)]. Thus this elementary result tums out to be immensely powerful in applications. Applications of (6.2) and its corollaries are given in [8, Chapter 2] Equation (6.3) is an elegant involution with a simple combinatorial interpre- tation and proof (6, (5.1), pp. 574-575]. §7. The main result of this section is identity (7.2). It is equivalent to Theorem 1 of [10] with B = 0; actually equation (6.3) is required to then identify the resulting identity with (7.2). Thus a result of immense importance in much of Ramanujan’s “Lost” Notebook is a natural component of this work, R. P. Agarwal [1] has generalized Theorem 1 of [10] and has shown that it follows from three-term relations for the more general 342 basic hypergeometric function. Equation (7.321) (due to Gauss) is easily derived from Jacobi’s Triple Product, Identity [8, p. 27, history, p. 31]. 34 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES Equation (7.7) is due to L. J. Rogers (26, p. 335, (4)] and (7.8) is Euler's celebrated Pentagonal Number Theorem (8, p. 11] §8. Identity (8.43) was derived in a different manner in [4, p. 137, Theorem Fl §9. and §10. Equation (9.3) is (as noted there) the expansion for the reciprocal ofa theta function. It has played an important part in the elucidation of certain Hecke modular form type identities [11]. Historical references are given in [11]. §11. Identity (11.1) is equivalent to Theorem 1 of [10] with A =0. Thus the comments on equation (7.2) are relevant here as well. §12. Equation (12.2) is a special case of an identity due to F. H. Jackson [21, p. 145, (4)}. Jackson's result is reproved in (7, §3] and applied to several summation theorems. The statement of Jackson’s identity in [7, p. 527, Lemma] reduces to (12.2) when a = g. There is a combinatorial proof of (12.2) in (6, (5.4), pp. 575-576). Equation (12.31) is probably due to Cauchy [8, pp. 20-21] Equation (12.44) is due to Euler (ef. (8, p. 19 (2.26))) §13. This section seems more than any in Chapter 1 to be begging for further study. Identities (13.1), (13.2), and (13.5) do not seem to fit into the known a-hypergeometric hierarchy of results §14. Equation (14.1) is due to L. J. Rogers (26, p. 334,(1)]. A combinatorial proof of (14.1) was given in [6, §4], where it was christened the “Rogers-Fine Identity.” It should be mentioned that it was also given by G. W. Starcher [28, p. 803, (1.12)] and that it plays a substantial role in proving many results in Ramanujan’s “Lost” Notebook [10]. §15. Equation (15.5) is another result unlike most in the literature. It is somewhat reminiscent of [10, p. 156, (42)] and [13, Lemma 1]. As with §13 we note that such results are not currently embedded in the g-hypergeometric hierarchy. §16. Identity (16.3) is the special case of Heine’s transformation (see equation (20.41) with ¢ = 1) [19, p. 106]. Equation (16.4) follows from Watson’s q-analog of Whipple's theorem (29; 27, p. 100, (3,4.1.5)] and is equivalent (modulo (12.3)) to the formula given by Watson to treat the third-order mock theta functions [30, p. 64; 9, 3.3)] §17. The proof of Jacobi’s triple product given here is closely related to that given originally by Jacobi [22, §64]. There have been numerous proofs of this result over the years (see [2; 8, pp. 30-31]). Also the many combinatorial proofs are examined in [12], and the related technique is applied to a wide class of combinatorial objects, §18. The main result of this section is (18.3), which is often referred to as “Ramanujan’s ;y-summation.” There are several proofs in the literature [5, 16, 18, 20]; of these the one by M. Ismail [20] is by far the most elegant and succinct. Ismail’s proof is given in Appendix C of [14], and numerous applications of it appear throughout [14] especially in Chapters 4 and 5. REFERENCES 35 §20. In this section, Heine's transformation [19, p. 106] of the q-hypergeometrie function is derived from the study of the linear operator Tp. Heine’s transforma- tion is (20.41). As an interesting corollary of this result, it is shown that Heine’s transformation and the natural symmetry of upper parameters generate a group of order 12. This was apparently understood by L. J. Rogers [23, p. 171], who isolated the three central transformations of this group. Equation (20.49) was also deduced by Heine (cf. [8, p. 20). ‘The section closes with a careful study of the polynomials J(u) defined by (20.6). These polynomials have come to be called the Rogers-Szegé polynomials, and the reader is referred to [8, pp. 49-50] for a discussion of their properties and the relevant literature. §21. The polynomials f,,(u) of §20 are generalized in (21.1) to Qn(t1,-.-. tte). which are a q-analog of the multinomial expansion. The Qn(ui,..., Ux) were extensively studied by L. J. Rogers [24] in his first memoir “On the Expansion of Some Infinite Products” and provided the foundations on which Rogers built up to the Rogers-Ramanujan identities [25]. The remainder of the section leads up to the identity (21.9), which is the case k = r = 1, s = 1 = 0 of the Fundamental Lemma in (3, p. 65]. It is clear from the results in Ramanujan’s “Lost” Notebook [9, 10] that he fully understood this technique also. References 1. R. P. Agarwal, On the paper “A ‘lost’ notebook of Ramanujan—Partial theta Junctions” of G. B. Andrews, Ady. in Math, 88 (1984), 201-300. 2. G.E. Andrews, A simple proof of Jacobi’s triple product identity, Proc. Amer. Math. Soc 16 (1965), 333-334. 3. ——, On basic hypergeometric functions, mock theta functions and partitions, (I), Quart. J. Math, Oxford Ser. (2) 17 (1966), 64-80. 4. —_. On basic hypergeometric functions, mock theta functions and partitions. (II), Quart J. Math. Oxford Ser. (2) 17 (1966), 132-143 5. —, On Ramanyjan’s summation of 11 (a;8;2), Proc. Amer. Math. Soc. 22 (1969), 552-558. 6. —_, Two theorems of Gauss and allied identities proved arithmetically, Pacific J. Math. 41 (1972), 563-568. 7. ——. On the q-analog of Kummer’s theorem and applications, Duke Math. J. 40 (1973), 525-528. 8. —_., The theory of partitions, Encyclopedia of Mathematics and Its Applications, Vol. 2, G.-C. Rota editor, Addison-Wesley, Reading, Mass.,1976; reprinted by Cambridge Univ. Press, 1984, 9. An introduction to Ramanujan’s ‘lost’ notebook, Amer. Math. Monthly 86 (1979), 89-108. 10, —_, Ramanujan's ‘lost’ notebook, 1. Partial #-functions, Adv. in Math. 41 (1981), 137-172, 11. ___, Herke modular forms and the Kac-Peterson identities, Trans. Amer. Math. Soc. 283 (1984), 451-458. 12, _, Generalized Frobenius partitions, Mem. Amer. Math, Soc. vol. 49, No. 301 (1984). 13, —_, Ramanujan’s ‘lost’ notebook. IV. Stacks and alternating parity in partitions, Adv. in Math. 58 (1984), 55-74. 36 FUNDAMENTAL PROPERTIES OF BASIC HYPERGEOMETRIC SERIES 14, __, q-Series: Their development and application in analysis, number theory, combina- tories, physics and computer algebra, CBMS Regional Conf. Ser. in Math., no. 66, Amer. Math. Soc., Providence, RL, 1986. 15. G. E. Andrews and R. Askey, Enumeration of partitions: The role of Bulerian series and 4g-orthogonal polynomials, Higher Combinatorics, M. Aigner, editor, Reidel, Dordrecht, Holland, 1977. 16. __., A simple proof of Ramanujan's 11, Aequationes Math. 18 (1978), 333-337. 17. R. Bellman, A brief introduction to theta functions, Holt, Rinehart and Winston, New ‘York, 1961 18 W. Hahn, Uber Orthogonalpolynome, die q-Differenzengleichungen geniigen, Math. Nach, 2 (1949), 4-34. 19. E. Heine, Handbuch der Kugelfunktionen, Vol. 1, Reimer, Berlin, 1878; reprinted by Physica-Verlag, Wurzburg, 1961 20. M. Ismail, A simple proof of Ramanujan's 11 sum, Proc. Amer. Math. Soc. 68 (1977), 185-186, 21. F. H. Jackson, Transformations of q-series, Mess. of Math. 39 (1910), 145-153. 22. C. G. J. Jacobi, Fundamenta nova theoriae funktionum ellipticarum, Regiomonti fratrum Borntriiger,1829; reprinted in Gesammelte Werke, Vol. 1, Reimer, Berlin, 1881, pp. 49-239. 23. L. J. Rogers, On a three-fold symmetry in the elements of Heine's series, Proc. London Math, Soc, 24 (1893), 171-179. 24. __, On the expansion of some infinite products, Proc. London Math. Soc. 24 (1893), 837-352. 25. —__, Second memoir on the expansion of some infinite products, Proc. London Math. Soc. 25 (1894), 318-343. 26. —_, On two theorems of combinatory analysis and some allied identities, Proc. London Math. Soc., Ser 2 16 (1916), 315-336. 27. L. J. Slater, Generalized hypergeometric Junctions, Cambridge Univ. Press, 1966, 28. G. W. Starcher, On identities arising from solutions of q-difference equations and some interpretations in number theory, Amer. J. Math. 53 (1930), 801-816. 29. G. N. Watson, A new proof of the Rogers-Ramanujan identities, J. London Math. Soc. 4 (1929), 4-9. 30. —_, The final problem: An account of the mock-theta functions, J. London Math. Soc. 11 (1936), 55-80. CHAPTER 2 Partitions 22, Sums over partitions. We define a partition as a sequence 7 = (ki, ka...) of nonnegative integers, with the single restriction that all but a finite number of the ky are zero. The content of 7, ¢(7), is defined by (22.1) e(n) = Soak. We say that 7 is a partition of n = c(n). We may think of a partition of n as a “way” of writing n as a sum of positive integers j, each one oceurring with frequency or multiplicity kj, order being immaterial. An alternate notation is nm = (1*12k2...), where only the positive frequencies kj are mentioned. Thus the partition (1,0,2,0,0,...) with content 7 would be written x = (113%). We shall be dealing with sums of the form f(x), the summation being extended over all partitions of n. (We indicate this by writing Dyn) (7)-) ‘Two particular functions that we shall discuss are (22.11) Kn) = Ok, 7 the total number of parts in 7, and (22.12) Qn = 1 K>0 the number of distinct parts in . For example, the partition (113%) of 7 has ky = 1, ky =0, hy = 2, ka =--- =0, and &(n) = 3, Q(n) =2. We begin with an important though obvious remark. THEOREM 1. Let ¥3(9) = Dnzo Ci(K)g* (9 =1,2,...). Then TI vs) = Soa" Sr) Ca(kea)- pa 250 x) ‘There is no question of convergence here, since the power series and their product are purely formal. We give some examples. EXAMPLE 1. C,(k) =1 (j 21, & 2 0). Then ¥3(q) = (1 = 4)~ (22.13) T]G-@)2 =P@ =a" 1 = LY pln", 21 nB0 r(n) R20 where p(n) is the unrestricted partition function 37 38 PARTITIONS EXAMPLE 2. C;(k) = (-1)* (j > 1, k > 0). Then ¥j(q) = (1+ 4)", gam) Pasa = at eee = at NE pa nO a(n) n50 x(n) the generating function for pg(n) — po(n) where pp(n) is the number of partitions of n into an even number of parts, po(n) = p(n) — p(n). EXAMPLE 3. C;(k) = u* (j 21, k > 0). Then (22.15) T[@-e) 7 =a Suk a 130 eG) EXAMPLE 4. C;(k) = uf (j > 1, & 20). Then (22.16) Tla-4e) t= ad abup a nD0 x) EXAMPLE 5. ¥3(q) =e (j 2 1). Then C,(k) = t*/k!, so (22.17) If we sum over r, we get (nr) ) nat 22.18) =2. (2a) Ea. EXAMPLE 6. ¥,(q) = exp( wd This leads to > ae (22.19) kajer tt +1)---(t+n—-1) nl . = coefficient of t” in Summing over r, we obtain (22.2) x eR SUMS OVER PARTITIONS 39 EXAMPLE 7. From Examples | and 2, since = Dag yey rin) Fee Me-e (+211 (35)} =P(l-q+qt- P+). by (7.324). Equating coefficients, we have (22.21) p(n) = p(n) — p(n 1) + p(n — 2) ~ p(n — 37) +- Then EXAMPLE 8. (9) = (1+¢q)/(1-q) =1+ (1+t)g+(1+¢)g? + (22.22) H(#4)- Yat Das ye, eet n=O x(n) THEOREM 2. Let L(x) = L(ki,h2,...) be multilinear in the ky, that is, linear in each ky separately. Then, formally, (22.23) Ye DT Ue) =P@e (4 i 4 e ) nd0 x(n) It is sufficient to prove this in the case L(ks, ka, .+- TI. jes where S is any finite subset of the positive integers. Applying Theorem 1, we have (l=)? for g ¢5, gt 29? +392 +++ =q(1—q)-? for jes. ¥3(Q) = { Hence the generating function o@= Ta-Tas =Te-er Hata = rat (oe): As an immediate corollary, we have the following theorem. 40 PARTITIONS THEOREM 3. If L(m) = Dj>1 ajk; then (22.24) Liwm= Y wa. Ps n(n) ale In fact, applying Theorem 2, we get 9) = POD 22 =Pade Ds i pl 1 le equating coefficients leads to the required result. AMPLE 9. Put aj = 1 (j > 1) in Theorem 3. Then L(x) = k(r), 80 (22.25) Yew) x) where d(v) is the number of divisors of v. EXAMPLE 10. aj =j (j 21). Then L(n) = Djky =n, so (22.26) where o(v) is the sum of the divisors of v. EXAMPLE 11. a; = (J) (j > 1), where is the Mébius function. Since Sa={) ifv=1, ce 0 ifv>1, (22.27) X Lavy = lr - 0). ro 331 EXAMPLE 12. a; = 1 (j =1),a; 0 (j > 1). Then 1 (22.28) Ya=Yow. $ nm) ae EXAMPLE 13. a; = (J) (j > 1), where ¢ is the Euler function. Then, since Diy $9) =, (22.29) LVLewe= YL we) a(n) 321 THEOREM 4. For all m > 0, (22.30) DS hike kim =O (C a) a(n) a(n) Consider the generating function Le De ke = n30 mz0 x(n) SUMS OVER PARTITIONS aa where, by Theorem 2, L= F(t) = Ag+ Ait + Aol? +++ (-ta)oo = [] 0 +40"), gal 150 by (12.44). Thus the generating function is porto =T1 (FZ) = Dat Das, an nB0 x(n) by Example 8, Expanding the binomial and equating coefficients of q"t”, we obtain the required result. THEOREM 5. Let f be an arbitrary function of Q. Then (22.31) YQ) = TY ba kA (0). a(n) a(n) m20 For each fixed n, there is a polynomial /* such that {*(Q) = f(Q) for Q = 0,1,2,..., n. Then r@= (S)arro. m0 Therefore, by Theorem 4, Liem=Creay=-oY (eo) an s*(0) a(n) n(n) a(n) m>0 =D Yh km" y"(0) Hin) m0 = Vien" 0), AG) M50 since ky -++km =0 for m > n and A™f*(0) = A™/(0) for m ki. a(n) a(n) It would be interesting to find a combinatorial proof of (22.32). EXAMPLE 15. Let /(0) = 1, {(Q) =0 for Q > 0. Then for n>0 (22.33) Dott hi + hake — kakeks ++} = 0. xn) ‘This is a check formula due to Sylvester [43]. EXAMPLE 16. Let f(1) = 1, f(Q) =0 for Q # 1. Since Q(n) = 1 if and only if r= (7), ie., n = jk, for some J, D>,in) f(Q(m)) counts the divisors of n. Hence (22.34) d(n) = Yo {hy — 2kike + Skikekg — +>}. zn) a PARTITIONS THEOREM 6. Let (j,k) be arbitrary, except that f(j,0) =0. Then (224) CVG.) =O Lawes. x(n) 321 x(n) 521 where my w= Ex EC (HH) -163-9) rh dlr For each fixed j > 1, by Theorem 1, Yr YG.) =Ta-a) DG. He* m(n) ye k>1 = P(a)(l-@) D0 FG, k)a* si = Pq) DUSG.k) — SG, Da. is Now eo 9= Hoe mat Substituting, we get Dea LY sinks) = P@) YO ulm)(SG.4) - 0.8 - D) ae n(n) kame a a Summing over j, we obtain Le VV) (ny 21 = Pl) YY mom\($G.8) — 1G Dee u(m)(f(9,k) = £9. = 1). ‘An application of Theorem 2 yields Le VV sGs)= Ce LD Vales n(n) j21 (nj 21 where 97) = SD wlm)(FG,k) = FG,k- 1) shmer =Le(2) Cuvee -s6.8-1) ar free EeQE(65) 165) SUMS OVER PARTITIONS 43 After a change of notation, this is the required result. An alternate form, deriv able directly or by an application of Theorem 3, is can CY )= YO ww (s(n2)-r (23-1). (nm) 324 iver Ge Now suppose that /(j, k) = h(;)w(k) in Theorem 6, with w(0) = 0, and define H(s) by AG)=OH). Hs) =u (3) Ali). slp ais Then we obtain THEOREM 7. With the above notation, 26) SL agw() = Ls New —wer- mya (2) a(n) 721 w(n)g2torly EXAMPLE 17. Let H(s) = 1, h(j) = d(j) in Theorem 7. Then (22.61) YL Vavwk) = Ye New) - wer - 9) w(n) p21 am(n)g2a rly EXAMPLE 18. Let w(k) =1 for k > 0 in Theorem 7. Then (22.62) LLAD=LVAHw= YL raw). a(n) JEL x(n) 321 in) 22), Km) utven > u20, v21 EXAMPLE 19. In Example 18, let h(j) = d(j), so that H(s) =1. Then (22.63) YY Yaw=Tew= a(n) 521 x(n) g50 EXAMPLE 20. Let H(s) = 8, h(j) = 0(j) in Example 18. Then (22.64) Y Le) =m) = YL rue). 0) Bh ufotn ™ £30 whl In Theorem 7, put H(s) = 6(s,1) (Kronecker delta). Then h(j) = 1, and we have THEOREM 8. If w(0) =0, then (22.7) Y Vek) = Vw) - wv - ks n(n) 721 a(n) 721 EXAMPLE 21. Let 6;() = the number of j such that kj(w) = i (i > 1) Then putting w(k) = 1 for k =i, w(k) =0 for k # i in Theorem 8, we have (22.71) Y 4) = oe - ba): a(n) a(n) 44 PARTITIONS EXAMPLE 22. Let w(k) =0 for k 4 (¢ > 0). Then (22.72) YY =vK r@) J31 x(n) RBs That is, the total number of frequencies > # in all partitions of n is equal to the total number of times that the part # occurs. This generalizes Example 14. EXAMPLE 23. Let w(k) = k(k + 1)/2. Then (22.73) YY blk + 0/2 = np(n). x(n) 321 In Theorem 7, take h(j) = 6(j, jo) (Kronecker delta) and let w(k) = \((k/m]), where m is a positive integer and [] denotes the greatest integer function. Then we obtain, after dropping the subscript in jo, EXAMPLE 24. Dag) A({ky/ml) = Dopiny M(Eom)> This holds even when A(0) #0. It may be proved without the use of generating functions or the preceding theorems by setting up a bijection x + x! of the partitions of n such that (22.74) [2] m kjm(n'). 23. Partitions with odd parts and with distinct parts. In (7.5), put i. Then 1 gt, HS =— 7 in T+i lta (23.1) F(0,-1;4: 4), P= (1=a)F(E,058) = m4? FP mgm, neo = the last by reason of (7.7). Multiply HS = F + G by (1 —4), equate real parts, and replace q by —9 to get (23.11) F(0,-1;-q: q?) Hb TL (quon ty — grtonmmy 4 (g(2"-GN—) _ g(2n—1)6n~2))) nl = 1+ Tog lontr0/2 — glSn?-r0/2) = 2F(—1,05-1), nel PARTITIONS WITH ODD PARTS AND WITH DISTINCT PARTS 45 again by (7.7). Now by (6.3), followed by (2.4), (23.12) 2) 2 2 F(0,-1;-9: 4?) = ——F(0,-q;-1: FO.-L a2) = FOG 1) 1 = pp gll +9 aF 0-0 4 (-1)"@" q ¢ Pn ita GF aaTH Fg + PT a) Also by (2.4), 2F(-1,0;-1) = 1 -gF(-1,0;-9) =l-qt(l+a)?- (+a) ta?) +- ing (23.11), (23.12), and (23.13), we have Lag+ (1+ q)q? = (-1)"(1 +a) (tag + a- Ss 2n—1 ¥ a ¢¥ +o (1? (+a) +49) CW ay pasa as 1-q+a —g +a" Now define Qa(n) = the number of partitions of n into distinct parts, the maximum part being =a (mod2), a =0,1; Q5(n) = the number of partitions of n into odd parts, the maximum part being = b (mod 4), 6 = 1,3. (23.3) ‘The well-known identity 1 nm (23.4) Toes = T10+0) n>i V nz. can obviously be paraphrased into the theorem which in our notation is (23.5) Qo(n) + Q1(n) = Qi(n) + Q3(n). Now the first series in (23.2) is clearly the generating function for Qo(n) — Qu (n). Examination will reveal that the second series generates (~1)"(Qj(n) — Q3(n))- Hence (23.6) Qo(n) — Q1(n) = (-1)"(QF(n) - Q3(n)). From the third series in (23.2) we obtain this common value explicitly: +1 ifn= tt, b> 0. (23.7) Qo(n) -Qiln)= 4 -1 ifn= 24k k>0, 0 otherwise. 46 PARTITIONS It follows readily from (23.5) and (23.6) that Qi(2n) = Qol2n); — Q5(2n) = Qi(2n), Qi(2n +1) =Qi(2n+1); — Q3(2n+ 1) = Qol2n + 1). ‘The results (23.2), (23.7), and (23.8) were stated in [16]. Afterwards, D. H. Lehmer communicated to me elegant graphical proofs of (23.7) and (23.8). Another theorem along these lines can be obtained from (20.72). In that identity, replace a by tq, multiply through by tg, and apply definition (1.1) to get (23.8) wo, ett ——_#e eg, (23.9) = gt (1 qt)(L=9%t) © (1 —gt)(1— 48t)(1 — gt) Hatt (1+ a)a'? + (1+ QV(l +a?) + By equating coefficients of g¥t™, we find: The number of partitions of N into distinct parts with maximum part M is equal to the number of partitions of N into odd parts such that the mazimum part plus twice the number of parts is 2M +1 It is not difficult to derive (23.8) from (23.91), by summing separately over even and odd M, and by taking into account the fact that the number of parts has the same parity as N when all the parts are odd. (23.91) 24, Continuation. From (8.4) and (8.41), with b replaced by ag, we obtain (24.1) F(-a~"q71,0;aq) = > a*a*Vi. Ford ‘The transformation (2.2) yields (24.11) 1+ (1-+a)gF(-0-} aq) = Yo akg Vp. 0 ‘The coefficient of a* on both sides of (24.11) can be interpreted arithmetically. Indeed, by (1.1), qF(-a~',0;aq) =q + (1+ a7"q)aq? + (24.2) ae + (1+a74g) + (1a7tg")a"gh tt + A typical term on the right of (24.2) is a*g% = a~ta"q%, where Neantltpitpet- tp, n+l >pi>m>->m>0. For t,n, and N fixed we obtain all the partitions of N into t+ 1 distinet parts of which m+ 1 is the largest; thus k = (n+ 1) ~ (t+ 1) is the difference between the largest part and the number of parts. We are therefore led to the definition, due to Dyson: The rank of a partition is the difference between the largest part 24.: (243) and the number of parts. ‘THE RANK OF A PARTITION a7 For arbitrary partitions the rank may be positive, negative, or zero; for partitions into distinct parts the rank cannot be negative. Denoting by 6,(N) the number of those partitions of V into distinct parts which have rank k, we see that the coefficient of a* in (24.2) is SX & (Nya, at Thus, (24.11) becomes (24.31) 1+ Doa® Y (6(N) + 6-1(N))a* = Doak gt Ve, k20 NBL 0 where 6_4(N) = 0. Hence (24.4) DO Gu() + e-a(N))Q" = ake (B> 0). wt Now for k odd, (8.5b) shows that. (24.5) Vi = YO we(N)Q", N21 where w(N) is the number of partitions of V into odd parts with maximum part, equal to k. Equating coefficients in (24.4) and (24.5), and writing k = 2r-+1, > 0, we have the result stated in [16]: (24.6) Ware1(N) = bar41(N) + bar(N) — (r 20, N 21). ‘This theorem gives us a refined correspondence between partitions into odd and distinct parts. It is possible to derive an arithmetic theorem for even values of k in (24.4), but the result is not very elegant. 25. The rank of a partition. In this section we shall study more closely the concept of rank (24.3). We define P,(n) as the number of partitions of n with rank r, and P,(n;Q) as the number of partitions of n with rank = r (modQ). We make the convention that P9(0) = 1, P,(n) =0 for r #0, n <0 and r =0, n<0. Let (25.1) K,(a) = So Pe(n)a", (-c0 aia 2 mi ?) ( in powers of t and q as follows So emt SE tte gestae tm, mz Pi20 Pm20 *a)( Qa 48 PARTITIONS The coefficient of t"q” is then the number of solutions of n=pit2m+--+(Pm+l)m — (p20), r=m—(py+prt-+(Pm+1))- Each solution corresponds to a partition of n into v = py + p2 + --: + (Pm +1) parts with maximum part m, so that r= m —v is the rank of the partition. Hence (25.12) may be written too $00 (25.13) Y SEP = YO eK @)- 1ST o0nSl 00 But, directly from definition, we have (25.12) equal to t~4F(0,t7!; tq) — 1 Thus, by (2.4), 00 (25.14) (1-H FO) = YO Kat (al < ltl < lal’) Now, by (16.4), (25.15) (1—t)F(0,t7!;t)(q)oo = 1+ (1-t)(1-t7* ae. (25.15) (1-8) F(0,-*:1)(@), (1-9 Liaw where (25.16) Cn = (=I) (1 + gq? t/2 After the denominators are expanded, the right side of (25.15) becomes $20 +(1-e-t) > Wy, ve where Wy =n E ques) = Clearly W_ = Wy, and for N > 0, Wy = Deng = 1 nei 830 neh pea +n)/24Nn TF Hence $20 (25.17) (L=8)F(0.t7t)(a)oo = 1+ St (QW, — Wyat — Wes) Comparing (25.17) with (25.14), we have (25.18) Kola) = e142 D(-yranttn (oo n>1 THE RANK OF A PARTITION 49 oo (oe Now, by summing over all r in a residue class (mod @), we find easily that = A) (25.21) Kea Q) = is yen (=) (G+ gl 2-7) q(@n?=m)/2, net (25.19) K-y(a) = Ke(q) Slee agnygon mam (> 0), nel (25.20) Ko(q:Q) = az {2B 1) net for 0

    You might also like