Modular forms and functions ROBERTA, RANKIN Professor of Mathematiesin the University of Glasgow CAMBRIDGE UNIVERSITY PRESS CAMBRIDGE LONDON: NEW YORK: MELBOURNEPublished by the Syndis of the Cambridge University Press The Pitt Building, Trumpington Street, Cambridge CB2 1RP Bentley House, 200 Euston Road, London NW12DB. ‘32 East 57th Street, New York, NY 10022, USA 296 Beaconsfield Parade, Middle Park, Melbourne 3206, Australia © Cambridge University Press 1977 First published 197 Printed in Great Britain atthe University Press, Cambridge 9 publication dat Library of Congress catalogui Rankin, Robert Alexander; 1915- ‘Modular formsand functions Bibliography: p.361 Includes indexes 1, Forms,modular 2. Functions, modular I. Title (QA243.R36 512.948 76-1089 ISBN 0 521 21212 X TO MY WIFEAd 12 13 14 1s 16 17 2 22 23 24 25 a4 32 34 35 44 42 43 4a 45 46 Contents Preface page xi Groups of matrices and bilinear mappings Notation ‘The modular group ‘The subgroups 1, F, I and (1) The level of a subgroup; congruence subgroups Groups of level 2 Groups of level 3 Further results Mapping properties Conformal mappings Fixed points Fundamental regions Construction of fundamental regions for A(1) ‘and its subgroups Further results Automorphic factors and multiplier systems Introduction ‘The functions a(S, 7) and (5, 7) Automorphic factors on subgroups of the ‘modular group Multiplier systems on (1), 7? and [° Further results General properties of modular forms Definitions and general theorems Dimensions of spaces of modular forms Relations between modular forms Modular forms of weight 2 Orders of magnitude Further results 15 19 28 30 32 39 42 47 51 66 70 3 nn 82 86 88 102 108 122 128, 10vii sa 52 53 5a 55 56 57 58 61 62 63 64 65 66 1 72 13 14 18 81 82 83 84 85 86 87 oA 9.2 93 Contents Construction of modular forms Poincaré series ‘The Hilbert space of cusp forms ‘The Fourier coefficients of Poincaré series Kloosterman sums Poincaré series belonging to FN) Poincaré series on P(N") Modular forms on ["(N) of weight 2 Further results Functions belonging to the full modular group Modular forms of even weight with constant multiplier system Poincaré series The case 8, =1 Modular forms of any real w Modular equations Further remarks Groups of level 2 and sums of squares Hermite and theta functions ‘Modular functions of level 2 Eisenstein series belonging to (2) Functions belonging to F'v(2); sums of squares Further remarks Modular forms of level N Forms of fixed character or divisor ‘The interaction of the operators R¥ and D, ‘The operator J, ‘The operator H, ‘The operator L, ‘The conjugate linear map K Historical remarks Hecke operators and congruence groups Double coset modules Definition and properties of Hecke operators, “The eflect of Hecke operators on Poincaré series 134 144 155 164 172 179 183, 191 194 201 202 205 209 212 2s 226 232 238 243 245 250 254 260 264 269 am 273 289 307 o4 95 10 10.1 10.2 10.3 10.4 10.5 Contents: Eigenforms Historical and other remarks Applications Dirichlet series Eigenforms for the full modular group Bigenforms for I'y(2) and (2) Eigenforms of level 4 Final remarks Bibliography Index of special symbols Index of authors Subject index 317 333 337 342 343 350 359 361 369 380 382Preface This book has grown out of lectures given in 1963-64 in Indiana University and before that in the University of Cambridge. It aims to provide a reasonably elementary introduction to the theory of elliptic modular functions and forms. Chapter 1 is concerned with the study of the modular group SL(Q2, Z) and its more important subgroups. These are studied mainly as matrix groups but also as groups of mappings, although it is in the second chapter that the mapping properties are most closely investigated and it is there that fundamental regions are constructed, Since our concern is not with the more general theory of automorphic functions on Fuchsian groups, itis possible to give a relatively simple account. In another respect our treatment of the theory is fuller than is customary in a textbook on automorphic or modular forms, since it is not confined solely to modular forms of integral weight (dimen- sion). Multiplier systems of arbitrary real weight possess a compli- cated structure. We have therefore presented the elementary theory in chapter 3 ina more general form than is necessary for our purposes, so that it can be applied to SL(2, R) and its subgroups if desired; it is only in §3.3 that the theory is restricted to subgoups of the modular group. Chapter 4 contains the basic definitions and properties of modu- lar forms of arbitrary real weight with corresponding multiplier systems. Itis important to know the dimensions of vector spaces of modular forms belonging to various subgroups, Where possible we have, in this chapter and later, determined these dimensions hy elementary arguments and ‘without recourse to the Riemann— Roch theorem. This is, in fact, always possible when the gervus is zero, and in certain other cases also ‘There are many ways of constructing modular forms. In chapter 5 we have followed Petersson in using as our building blocks @ special family of Poincaré series. This method is particularly convenient for forms of real weight k >2.and arbitrary multiplierxii Preface system, The Fourier coefficients of these series, although compli- cated in form, are of interest and involve Klosterman sums of various types. In particular cases these sums are considered in detail and are estimated with sufficient accuracy to yield results in the limiting case when the weight k=2. The Petersson inner product is introduced in this chapter. This product is of impor- tance, not only because it allows us to regard the space of cusp forms as a Hilbert space, but also because it can be used to provide explicit formulae for the Fourier coefficients of an arbitrary cusp form. Chapter 6 is concerned with the simplest situation when the ‘group considered is the full modular group. The standard fune- tions, such as the modular discriminant 4, Klein's invariant J and the Eisenstein series B, (k = 4) are introduced and studied. In the fast section the six different multiplier systems of real weight are found by considering real powers of 4. ‘Modular forms of level 2 are discussed in chapter 7. The most important of these are the three simple theta functions 32, ds, 9s and certain polynomials in the graded ring they generate. The ‘corresponding multiplier systems are described and dimensions of various subspaces found. Modular equations are dbtained in a few very simple cases. The Eisenstein series of level 2 and integral weight k=3 are studied and their Fourier coefficients are evaluated. The theory is then applied to the problem of finding the number of representations of a number as the sum of 2k squares, In chapter 8 we return to a more general situation and discuss forms of level N and integral weight k with simple multiplier systems. The vector space of forms of this type can be expressed, as was done by Hecke, as a direct sum of subspaces each correspond- ing to a different character y modulo N and divisor t of N. The action on these subspaces of various linear operators is studied in preparation for the following chapter, which is concerned with the theory of Hecke operators. These operators and their effect on Fourier coefficients and on Poincaré series are examined in detail. ‘The Hecke-Petersson theory of eigenforms and the more recent Atkin-Lehner theory of newforms are developed and extended. ‘The final chapter begins by describing the connexion between these theories and Dirichlet series having Euler products, and concludes with a number of applications to spaces of levels 1 and 2, particular attention being paid to the eigenforms that arise in the Preface xiii study of the representations of a number as a sum of squares. When the weight is odd, these eigenforms have coefficients with rather curious multiplicative properties. ‘The material in the book is taken in the main from articles in the research journals and relies heavily on the fundamental work of Hecke and Petersson, but includes also some new results, particu- larly in chapters 3, 9 and 10. Modular function theory is now so vast a subject that a book of reasonable length cannot include every aspect and so necessarily reflects the particular current interests of the author. A quarter of a century ago, when the idea of writing a book on the subject was first suggested to me by my friend MrR.J.L. Kingsford, who was then Secretary to the Syndics of the Cambridge University Press, I had visions of writing a comprehen- sive treatise that would give a complete account of every partof the subject. Such a task was difficult then and would be virtually impossible today. Thus in the present work there will be found only a brief account of order results concerning the Fourier coefficients ‘of modular forms (an earlier interest of the author) and even briefer reference to the important recent work of Deligne and Serre in that area. Moreover, there is no mention of such subjects as the Selberg trace formula, congruence properties of coefficients, transcendentality of values of coefficients, complex multiplication and elliptic curves In conclusion, itis a pleasure to acknowledge my indebtedness to the individuals and institutions that have aided me in the writing of this book and of its earlier versions. In particular, my thanks are due to the Universities of Cambridge, Glasgow and Indiana, and to Clare Hall, Cambridge, where the major part of chapters 5-7 was completed during my three months’ tenure of a Visiting Fellowship there. I am grateful to Dr Bruce C. Berndt and Dr W. Wilson Stothers for their detailed comments and criticisms of the manu- script; it is scarcely necessary to state that the imperfections and inaccuracies that remain are the sole responsibility of the author. Finally, I wish to express my gratitude to Miss Doris M. Caldwell, who typed the greater part of the manuscript, for her skill and patience, RAR, Glasgow March 1976Groups of matrices and bilinear mappings 1.1. Notation. Modular functions and forms will first be defined in chapter 4. In this chapter we study the groups on which these functions are defined. We write € for the set of all (finite) complex numbers, with the usual topotogy. B for the set of all (finite) real numbers. @Q for the set of all rational numbers. Z for the set of all (rational) integers. Z* for the set of all positive integers. H=(2:2€C, Im z>0}, the upper half-plane. E=Cufeo}, the extended complex plane. This is the one-point compactification of C. In the topology on €, aset A is openif either ()A isanopensubset of C, or ii)-o€ A and € ~ A is compactin€. With this topology, € is homeomorphic to the two-sphere, ie. to the Riemann sphere (x, y, z)¢R': x7 +y?4+2?= 1}. R=RU {2}, the one-point compactification of P. Other subsets of € will be defined later. We write throughout ; rele ak ly 8 ea for a, b, c,d, a, B, y, 5€C and put [T]=det T= ad ~be. Let O=(T:a,b,¢,deC,|TI=1}, 1.2) and N={T: a,b,c, d eR, |T]=1) (1.1.3)2 Groups of matrices and bilinear mappings Then @ is a group under matrix multiplication with the identity element 10 1) th (1.4) and @ contains ? asa subgroup. In group theory the groups @ and A are referred to as the special linear groups SL(2,€) and SL(2, R), respectively. It is easily verified that, for each Te 0, [ ‘ (5) -c a r ‘The subgroup consisting of I and -1 a) I [ oo (1.1.6) is denoted by A; itis the centre of @ and 2. For each Te @ we write T= where 6y is any complex number for which this equation is valid. Then it is easily shown by induction that, for any q 2", T2RT-Fsh, aa.) +d=2.c0s Or where (1.1.8) This is a polynomial of degree q~ 1 in cos 6 and so is defined even when sin @r =0. We note also that u(T™) cos q6;. (L.1.9) Further, if 8, = #k/q for some integer k, then Tt=C (1.1.10) With cach Te @ we associate a bilineart mapping, which we also call T, defined on € by aztb w=Tla) cz+d (eo), 4 Other terms used are linear, linear fractional and M@bius. No confusion should arise with the use of the word bilinear in multilinear algebra, Se AEE SN I RO Lt Notation 3 and we also, for brevity, write Tz in place of T(z). For example, T(-dfe)=0 and Too=a/e. ‘These relations hold even when c=0, for then a and d are non-zero, so that —d/e and a/c both mean %; thisis a consequence of the rules zta=otz=0, 0 (ze0), zoo=coz=0, Z=0 (26€~{0). ‘The mapping T is a bijective mapping of € onto itself, the inverse mapping being given by dw-b 22D Ws if Fis any subgroup of @, the mappings T defined by matrices Te” form a group F under composition as group operation; ie., if SeP, TEI, then (ST)(z) means ${T(z)}. This is easily checked. For, by the definitions of § and T, sr-[%@ + Be ab + pd yat8e yb+édds" a1.) while arth Waetd ” (aatBe)z+(ab+Bd) azth, 5 (at bc)z Hy +A) cat S(T) y, ‘The group fis called the inhomogeneous group associated with which is called a homogeneous group. Let ¢ denote the mapping: 6: matrix T}bilinear mapping T. ‘Then the above remarks show that is a homomorphism of ‘onto @. Let I” be a subgroup of , so that ¢ is a homomorphism of T onto I. The subgroups /* that we consider will usually act not on the whole of C, but on some subset 0. We suppose that Dis a subset of Csuch that 7D=B, i.e. TD=D forall T¢ f° We suppose further4 Groups of matrices and bilinear mappings that D contains more than two points; usually D will be €, €, Hor Hs The identity mapping in fis (2 €D), and we have aztb Ta) ed =z forallzeD if and only if az +5 =c2*+ dz for all z €. Since D contains more than two points, this gives ie.a=d=+1,b=c=0,s0 that T= 41, Thus the kernel of d is A if Ie and is I if -I¢ 17. Hence we have fara (if-leD, f= (it-1éP), (1.1.12) In particular, 6 @/A and (= 0/A. The groups 6 and 3 are referred to in group theory as the linear fractional groups LF(2, ©) and LF(2, R), respectively. When —I¢ 1° we can adjoin —/ to I, obtaining an overgroup F(=LA = AD) of F having Fas a subgroup of index 2, and then farafiA. (1.1.13) By writing I” for the homogeneous group and /* for the associated inhomogencous group we indicate that we regard the latter as being determined by the former. This point of view is especially convenient when we are concerned with algebraic prop- erties of groups and, in particular, with multiplier systems. On the other hand, a different point of view can be taken when analytic properties are under discussion, For we shall be concerned with classes of functions f defined on H for which the quotient f(Tz)/fl(z) is the same for each member of the class, when T belongs to a certain given group of bilinear transformations. Since this quotient takes the same value for ~T' as for T, it is often convenient to assume that —J belongs to the associated matrix group. Accordingly, if we start with a group fof bilinear transfor mations, we may define the associated homogeneous group I” to consist of all matrices T such that the associated bilinear transfor- mation belongs to f°; it then follows that ~T'e whenever Te I It is easily checked that /”is in fact a group. (Authors who adopt this analytic point of view commonly write I” for the inhomogeneous ‘group and F for the homogeneous group.) ae Ree nN PE or 11 Notation 5 We now introduce a notation that we shall find useful when dealing with coset representatives. Let 5 be a set closed under an operation, which we call multiplication, and let A and B be subsets of S. Then we write, as is customary, AB={xeS:x=ab,aeA, be B), An element x € AB may be expressible in more than one way as a product ab for a € A, b € B. If, however, each x € AB isexpressible in exactly one way as a product ab for a € A, b¢ B, we write AB=A-B. Itis easily verified that A - (B- C)=(A - B) - C, provided that one side is defined, in which case the other is also. IfA isa finite set, we write [A | for the number of its elements, Now suppose that I< 1's @, I’ and I being subgroups of Then the statements Mya0,- 8, M=£-T, are equivalent to the statements that J is a set of right coset representatives of 1", modulo I>, and that # is a set of left coset representatives, respectively. We call 9a right transversal, and 2 a left transversal, of Fy in Ty. If |R| is finite, so is |¥| and la\=|2)=(eTs), the index of Fin I. Note that, if Te ® and —l¢ I, then -T¢ Ht. Similar notations can be used with inhomogeneous groups. We note alo habit Telscly2Oand y= Ts then = PsA, where there is a one-to-one correspondence between matrices in % and transformations in 9; for this reason we shall usually write not only [=I +R but also fy =f 8. Theorem 1.1.1. Let l, and I, be subgroups of O with P< 1". Then, if, =T,: Rand $ is any member of 1, P= - (RS). A s:milar result holds in the inhomogeneous case. Proof. For y= I, = (I> 2)S (RS). Theorem 1.1.2. Let I’, be a subgroup of finite index w in a group I, and let S be a fixed member of Is. Then there exista finite number of6 Groups of matrices and bilinear mappings elements Ly, Las, Lm Say, in Ty and m disjoint sets F=ULS':05ko1 we take any L; not belonging to %, and define a by (1.1.14). As before, the o2 elements L,S* (0k o. tor, we take an LyéI'(F,U%;) and proceed similarly. Since pis finite and ¢, >-0 for each i, there exists a positive integer ‘m such that (1.1.15) holds and the process then terminates, giving the required result, The two final sentences in the enunciation are immediate consequences. Note that, when Li, L3y..., L, (¢|b,| we form a 0 Mf, By Hf a mo] ie = teevtive[ | ale djl ol La a -[2 4] “es ds and note that we then have |c|<[es] and tr T=, We now form T,:=U-"T,U", choosing m as above to make Jas~4¢] = [ds $e] sles] =Ieo| = sorb, and stop the process here if |e] $|Bs|; if not, then [e,|> 0s] and we proceed as above, obtaining a matrix T. with |es|2, we can put 12 ‘The modular group MW Table 1 1 0 “1 “1 Saeed tg 2 1 ok O14 2 <1 -k 0-1 1 =2 cosh @, for some @>0, so that, by (1.1.9), tr(S*)=2 cosh qo >2 for all qe Z°. It follows that, if | >2, then S*# £1 for all qeZ". ‘Theorem 1.2.4. I(1) is generated by U and V; every element TeT (1) can be written in the form T=U"VU"V ©» VU% 1.2.8) where q,€2 (0: i=); this representation is not unique. Proof. Let Te (1) and take S=L"'TL. as in theorem 1.2.2. By theorem 1.2.3, we may suppose that |1|>2, where = tr T: Then Se Ty, and so we can choose q¢Z so that, iff) = tr U%S, lnl={etavisalyls this is possible since y #0. Then Inlsalylsitle 4)? 0 ( 0 and that 1 is the least positive integer for which this holds. Then (1 VPP VP «Vpn (1.2.12) and p, + po#0 (mod 3). For if p, +p)=0 (mod 3), then n= 2 and T= (DPV VP, which is a word of shorter length n—2; hence n = 2 and so p= 0, which is false. Since P’ =I, we may assume that p, +po= 1 oF 2, 80 that we deduce from (1.2.12) that +/ can be tepresented as a product of n factors each of which is either fl 0 ie [; i} (1.2.13) at ver=u=[) 1 tube Since the entries in U and W are all non-negative the same is true of the product, which must be +/. But, if either T=US or I= WS, then $ has one negative entry, and this gives the desired contradic- tion. It follows from theorem 1.2.5 that Paya spr ws 1P VN waaay 12 ‘The modular group 13 with the notation introduced at the end of §1.1. Further, since P*=—P, itis clear that theorem 1.2.5 remains true, possibly with a different value of r, if each exponent 2p, (Oi =n) in (1.2.9) is replaced by p, (O=isn), Since the. matrices T and ~T give rise to the same bilinear mapping, itis clear that we have Theorem 1.2.6. (1) is generated by the mappings V and P, which have orders 2 and 3, respectively; i.e. Fay=0 (OJ°'TJ is closely associated with the non-analytic map zret(z): ToT (1.2.26) where the bar denotes the complex conjugate. For UTI) =I TH) = Tez¥d 1.3. The subgroups T?, I°, Mand ["(). Let T¢ (1), and suppose that T is expressed as in theorem 1.2.5. We define A(T)=n+2r, p(T) =potpites-+py (13.1) It then follows from (1.2.4) that, for any 7), Tre /(1), A(T,T.)=A(T,) + h(T:)mod 4) (1.3.2) and p(T. T2)= p(T) + p(T;) (mod 3). (13.3) ‘Thus h and p are homomorphisms of (1) onto the additive groups of residue classes modulis 4 and 3, respectively. The kernels of these homomorphisms, namely P:=({T: Te Ll), AD) {mod 4)} (13.4) and I={T: Te M1), p(T) =0 (mod 3)}, 0.3.5)16 Groups of matrices and bilinear mappings are therefore normal subgroups of (1) of indices 4 and 3, respec- tively. We also write P°:={T: TeT (1), h(T)=0 (mod 2)} 1.3.6) sothat “cI” I(1), and [is a normal subgroup of (1) of index 2. “From (1.2.9) and the definitions of the three subgroups it is easily seen that I” is generated by P and Pp, ‘ev-[I alk (13.7) thus I?=(P, Py) Similarly, T*=(P, Pi), (1.3.8) and P=(V, Vi Va), (1.3.9) where ih a (1.3.10) Pvp = pvp =| | ih Note that -/ belongs to J? and ° but not to J*. Thus Mara. Itis clear from theorem 1.2.5 that T*=(P?) * (Pi). (3.11) We now consider the commutator group I™(1) of (1). This group is generated by the commutators [s, Th=sTS"T, (13.12) for $, Te I). We prove Theorem 1.3.1. The commutator group I”(1) is a free group of index 12 in F(1). (1) has rank 2 and is generated by UW and WU. 13 ‘The subgroups 1?, 7, Mand (1) W Further (1) = T° F*, so that TeI"() if and only if A(T)=0 (mod 4) and p(T)=0(mod3). (1.3.13) The factor group F(A)/I"(A) is a eyelic group of order 12 and is generated by the coset I™(A)U. Also, writing I” for ["(1), we have —lePU’, Veru*®, PePu", Weru". (3.14) In fact, U*=—[UW, (WUY"'], Finally, PrOaPh, and P=Por. Proof. For any S, Te F() it follows from (1.3.2, 3, 12) that AUS, T) =O (mod 4), pS, T])=0 (mod 3). It follows from this that every Te I"(1) satisfies (1.3.13) Conversely, if T¢"(1) and T satisfies (1.3.13), we can express T in the form (1.2.9), where (1.2.10) holds. Write m= Put pit :*+P_, (O=msn). ‘Then n is even and it is easily verified that T=[P', VIEV, PLP™, VILV, PP). (PH, VIV, Pe, so that Te J~(1). Now (YP]=[¥ P= PY Vy and (V, P?J=[P", VP '= Wu, so that ["(1) is generated by UW and WU. From these facts it is clear that ["(I) is a free group of rank 2 having UW and WU as generators; for otherwise we could express Tas.a product of the form (1.2.9) with 1 > 0, and this is impossible ‘The index of I"(1) in F(1) is 12 since each coset corresponds to a different pair (h, p) of residue classes and consists of those Te 1) for which A(T) =h (mod. 4) and p(T) =p (mod 3).18 Groups of matrices and bilinear mappings In fact PA)=F0)- ®, where ® consists of the 12 matrices P*V" (0=p<3,0h <4). If we use ~ to denote the equivalence relation of belonging to the same coset of (1) in I'(1), we deduce from v=-U{U>, VIV,U"') that I~ V?~U%, V~U", P~VU~U" and W~ WUU"'~ U"' In fact we have Peyt~ urs (Osp<3,05h<4). Finally, it is obvious from the earlier part of the proof that PA)=Pals, and so P)=Ar. Also P(1)=L°T* since T°T* contains the generators V and P?. We now state the corresponding results for the associated inhomogeneous groups. For this purpose we use the results already obtained, together with (1.1.12, 13) Theorems 1.3.2. The associated normal inhomogeneous groups have the following properties: Paft=(Pya(Py, Mar, (af =2, (1.3.15) Pa(Vye(Ve(V), P=P/A, Uf ]=3, (3.16) fy=Poherw, Fahl, (1.3.17) fw=hr. (1.3.18) ‘The factor group '(1)/1*(1) is acyclic group of order 6 generated by the coset containing U. We conclude by giving alternative definitions of ?, /? and £*(1). For this purpose we define a:=(w,uj=vur=[) 3]. (4.3.19) Theorem 1.3.3, For v=2, 3 and 6 define TeI():O*TQ=+T (mod »)}. (1.3.20) Then G?=I?, G?=I" and G*=F*(1). 14 The level of a subgroup; congruence subgrours 19 Proof. G? and G? are certainly subgroups of (1) and are proper subgroups since they do not contain U, Further, G? contains the generators P and P, of I”, while G contains the generators V, V and V; of °. We deduce that G?= I? and G’ =I", Accordingly G=G0@=lo=F0). 1.4, The level of a subgroup; congruence subgroups. Let "be any subgroup of (1), not necessarily of finite index. For each L €I(1), define n, to be the least positive integer such that UN e LIL" (1.4.1) ‘When I” has infinite index in (1), no such finite positive n, need exist and we put nm, = ©0in this case; on the other hand, when "has finite index in (1), m, always exists and is finite. Consider the set {n,:L € M(1)}. If the numbers in this set have a finite least common imultiple mn, we call n the level of and write lev F (14.2) Inall other cases we put lev "=o. In particular, lev I” is always finite when I” has finite index in (1), since then "has only a finite number of conjugate subgroups in (1). When (1.4.2) holds for finite n we have U"eLIL™ forallLeF(1). (1.4.3) We now write A(n) for the normal closure of the cyclic group ee A(n)=(L" UL: €) (1.4.4) and is the smallest normal subgroup containing (LU). Clearly n is, the smallest positive integer such that A(n)< I. ‘We make exactly similar definitions for inhomogeneous groups. ‘Thus, if fis any subgroup of F(1), lev fis defined and the group A(n) is the normal closure of the group (U") of mappings. By way of example we note that lev PU) =lev F()='1, tev A(n) = lev Aw and lev(U")=0 for both homogeneous and inhomogeneous groups (U"). Also, by theorems 1.3.1, 2, lev ™(1)=12, lev M1) =6. fn 4German: Stufe20 Groups of matrices and bilinear mappings ‘An important class of subgroups of the modular group consists, of what are called congruence subgroups. If 5 and Te (1), we write S=T (mod n), where n¢Z’, if and only if a=a, B=b, y=c and 5=d(modn), It follows at once from (1.1.5) and (1.1.11) that if S.=5(modn) and T\=T;(modn), then T;'=T;'(modn) and $,T,=5,T;(modn). We write P(n):={8 ¢P():S =1 (mod n)}; (1.4.6) this agrees with the previous definition when n = 1. Then I) isa group; for if $= T= (mod n), then ST's IT"'=1 (mod n). We also write Fin):=AP(n)=(SePU):$=2I (mod ny}, (1.4.7) and can prove similarly that this is a group. The two homogeneous groups I) and Fn) ive rise to the same inhomogeneous group, which we denote by F(n). Each of the groups Mn) and P(n) is called a principal congruence group,t and the same title is some~ times conferred on I(n). Since Fe F(n) if and only if n = 1 or 2, we have Pa=Mo/A=fa/A (n=1,2), (4.8) fi=ln=PsA (n=3). (1.4.9) Both F(n) and F(n) are normal subgroups of (1), and f(n) isa normal subgroup of f(1). For, if S¢I(n) and Te I(1), then TST =T"IT=1 (mod n), and the proof is similar in the other cases. + German: Hauptkongruenzgruppe. ae pee er ROE ATR RS ER ne 14 ‘The level of a subgroup; congruence subgroups 24 ‘Clearly lev P(n)=lev P(n) = lev Mn) =n, and also Acrin), Aime fin). Further, since J'U*J=U™ and (U")=n, it follows from (1.2.22) that A(m) and (mn) are invariant under the automorphism T +> J*TJ, but not under the automorphism (1.2.22) unless n is even. On the other hand, F(n) is invariant under all autorrorph- isms of (1), and A(n) and F(n) are invariant under all automorph- isms of FI). ‘The factor groups Gw):=TD/Pn), Gin):=P)/Fn) (1.4.10) and Gin):= fay fin) (14.11) are called modulary+ groups of level n. By (1.4.8-11) we have G(n)=G(n)=Gln) (n= 1,2) (1.4.12) and Gln) =Gin=Gin)/A_ (n=), (1.4.13) Since the number of incongruent matrices T modulo n is clearly less than or equal to n*, the three modulary groups are clearly finite and their orders are denoted by (rt), Z(n) and si(n), respectively; z2(n) should not be confused with the Mébius function. ), where the product is taken over all primes p dividing n, and Aln)=in)= wl) (=1,2), ln) =H) =Iuln) (1 =3) Theorem 1.4.1. wien ti (1 Proof. By (1.4.12, 13), itis enough to find y(n), ie, the number of incongruent matrices § modulo n. Our proof is similar to that given by Gunning (1962). We say that a pair of integers c,d is a primitive + German: Modulargruppe.2 Groups of matrices and bilinear mappings pair modulo n if and only if (c, d,n)=1, and denote by A(n) the number of incongruent primitive pairs modulo n. Lemma 1. Ifc, dis a primitive pair modulo n there exists an S€T(1) such that y =¢, 5 =d (mod n). Conversely, if S € (1), then y, 8 isa primitive pair modulo n. Proof. The last part is obvious since (y, 8) = 1, so that (y, 6,2) = 1. As usual we write q|r to mean that q divides r. Suppose that c, d is a primitive pair modulo n, so that (c,d, n)=1. Take y= and write c=etn, where c, is the largest divisor of ¢ that is prime ton. Take m € Zs that 6:=d+mn=1 (mod cx), which is possible since (n, ¢1) = 1. Then (y, 6) = 1. Forif pisa prime divisor of y and 6, then p divides both c and d + mn. If ple; then, since & = 1 (mod c,), p +8, which is false. If ples, then pln and so pid; thus p|(c, d, n), which is false. Hence we have found y, 6 with y=c, 6d (mod n) and (y, 8) = 1. We can now find a, 6 € Z such that ad By =1 and so SeT(1). Lemma 2. Foreach primitive pair, c,d of integers modulo n there are exactly n matrices $€I(1) modulo n for which ysc, 6=d (moda). Proof. Let S: and S; be two members of (1) with second rows congruent to [c, d] modulo n. Then af! 4 S82 {i 1) mod), forsome k € Z, Asthere are only 1 incongruent values of k modulo 1 it follows that there are exactly n incongruent matrices S € (1) with [y, 8]=[c, d] (mod n). Lemma 3. \(n) is a multiplicative function; i.e., if (m4, m2) = 1, then Alniynta) = A(m)A(na). ARRON R19 RE ETRE RR A ARS 14 The level of a subgroup; congruence subgroups 23 Proof. Let ;, 6; be a primitive pair modulo n, (j= 1,2). Then yoms+ Yolk, 8yM2+82m, is a primitive pair modulo mn, since (n;,m:)=1. Also, incongruent pairs for mand 1; lead to incon- gruent pairs for mrt; ie. if Vins Yt yas + yan (mod ny) and Binat Sim = 6yr2+ Sans (mod msn), then i= y1 (mod n,), 844; (mod n,), 5% y2 (mod n,) and 35= 8; (mod nz). Thus A(m,)A (112) SA (myn). Conversely, let y, 5 be a primitive pair modulo n,n,; then 7, 5 isa primitive pair modulo m, and modulo nz. Also, since (m;, 12) = 1, incongruent pairs modulo nym cannot give rise to congruent ones modulo 1, and modulo rn. Thus A(nyna)SA(m)A(n2). Lemma 4, If pis prime and k= 1, A(p* 1p), Proof. There are p*(1—1/p) incongruent integers ¢ modulo p* such that (c, p)=1. For any one of these, each of the p* incon- gruent values of d will give a primitive pair. Since these pairs are all incongruent modulo p* we have p7*(1~1/p) such pairs. There are p*~* values of c such that (c, p) =p. To each of these correspond pX(1~1/p) values of d incongruent modulo p* and such that (d, p) = 1. This gives p*~"(1 — 1/p) primitive pairs. Addition gives ape awp="(1-5) From lemmas 3 and 4 we obtain. aman (1 4 >) and, since (1) =nA(n), by lemma 2, the theorem follows. Theorem 1.4.2. If m and n are positive integers, Pim) olin) =Ptm, n}y (1.4.14) and Pon\F(n)= Pll, n)), (1.4.15) where {m,n} is the least common multiple of m and 124 Groups of matrices and bilinear mappings Proof. Write h =(m, n),1= (in, n}. We have Te Pm) a(n) if and only if T=J (mod m) and T= (mod n); this holds if and only if T= 1 (mod 1). This proves (1.4.14). If Te L(m)(n) then T= $,5; where S, =I (mod m) and $= 1 (mod n). Hence S;= (mod h) and S21 (mod k), so that T= (mod A), It follows that Pm)P(a) sh), (1.4.16) Now, by (1.4.14) and one of the isomorphism theorems, (Pony F(a) = Fen) /TO, (14.17) so that wn) =P =O: Em) ):P()] TU): Pom) PIC): PO) =P) nye (D/ em). Since ()e(h)=2(m)u(n) by the formula in theorem 1.4.1 it follows that IF: Pom) Po)] = nh) and this combined with (1.4.16) gives (1.4.15). Theorem 1.4.3. If (m,n)= 1, then Glmn) = G(m) x Ga) Proof. By (1.4.15, 17), P(D/T(n) = Pn) /P nn) (1.4.18) so that we can identify G(n) with M(m)/I'(mn) and G(m) with I'(n)/F(mn). These are both subgroups of (1)/Mmn) = G(mn). ‘We have, by theorem 1.4.2, G(m)G(n) = {1}, Glm)nG(n) = G(mn), where J is the identity in G(mn). From this the theorem follows; see theorem 2.5.1 of Hall (1959). By repeated applications of theorem 1.4.3 we see that, if, nape ae nN REN ET RE PE EER 14 The level of a subgroup; congruence subgroups 25 where the p, are different primes and a, >0(1 1, (f*:f"]2n” and may be infinite. See also Rankin (1969). In the further algebraic study of the subgroups of the modular group the following three group-theoretic theorems are useful. Theorem 1.7.1 (Kurosh). Let the group G be a free product of subgroups A,. We write this as G =|J3 A.. Then, if H is a subgroup of G, we have H=F*1I5 Bs where Fis a free group and, for each B, Bs is conjugate to one of the subgroups Az. ‘We note that the free product []$ By can be empty and that F can be the trivial group. However, if [1 Bp is not empty, then F must, have infinite index in G; for otherwise some power of an element of [I B, (other than the identity) would belong to F. Theorem 1.7.2 (Schreier), Let H be a subgroup of finite index p ina free group Gof finite rank R. Then the rank r of His also finite and +u(R-1). We note that the rank of a free group is the number of free generators of the group. For proofs of both these theorems see Kurosh (1960). Theorem 1.7.3 (Nielsen). Let the group G have identity element e and be the free product of cyclic groups G, of orders m, (Sin). Then the commutator group G' is a free group of index m= mina... a in Gand the rank of Gis 1+ 1+ $(1 17 Further results 35 Gi is generated by the set of all commutators [xa’ acG,beG,ate#b, lsisjsnand My. M4 Miey Gay for any asin Gy (15k 51). The factor group G/G' is isomorphic 10 the direct product of the cyclic groups Gx, Gay... Gu See Nielsen (1948) and also Lyndon (1973); for the commutator notation see (1.3.12). Theorem 1.7.3 provides an alternative method of proving theorems 1.3.1 and 1.3.2. With the help of theorem 1.7.1 we easily derive the following result of Newman (1964). Theorem 1.7.4, A subgroup of [\(\) is free if and only if it contains no elements of finite order other than the identity, If fis a normal subgroup of 1(1) different from I'(1), [7 and £*, then fis a free group. The following theorem is a simple deduction from Schreier’s theorem. Theorem 1.7.5. Let f be a free subgroup of (1) of finite index p. Then u =0 (mod 6) and has rank 1+ hu. In particular, the index of any normal subgroup of £(1) other than F(1), f? and f? is divisible by 6. Proof. We shall prove the last sentence by analytic methods in chapter 2. The fact that 1+y is the rank of fis due to Mason (1969). write d = fr f(1) and put f:4\ v=[h:4], so that By theorem 1.7.2 applied to d as a subgroup of (and (1) we have 14(r=1)A= 14H, where r is the rank of f, Thus, 4 =6(r~1), from which the theorem follows.36 Groups of matrices and bilinear mappings Fig. 3 displays all the normal subgroups between /(1) and (6). The two groups {' and /" are the commutator groups of /? and . By applying theorem 1.7.3 their indices in the groups immediately above them can be verified. ‘A complete study of all subgroups between /(1) and (4) has been made_by Petersson (1963). His list contains thirty groups, including (4) and f(1); among these thirty groups there are contained, of course, the six groups in fig. 1. Petersson (1953) has also studied subgroups of finite index in (1) for which fy=f Uo such a subgroup /*is called a cycloidal subgroup of f(1). Fig. 3, Normal subgroups between (1) and (6). 17 Further results 37 We have already remarked that G(2) is isomorphic to the dihedral group of order 6 and that G(3) is isomorphic to the alternating group on four symbols, which is the tetrahedral group. It can be shown that G(4) and G(5) are isomorphic to the octahedral and icosahedral groups, respectively. In fact, for n= 2, 3, 4, 5, Gin) is generated by u=UP(n),. v fin), where w = (ww)? e being the group identity. a A great deal of information about the modulary groups Gn) and G(p*) can be found in Klein and Fricke (1890, 1892) and in Vivanti (1906, 1910); see also Frasch (1933). A complete classifi- cation of all the normal subgroups of G(m) has been given by McQuillan (1965). ‘The definition of level given in §1.4 is due to Wohlfahrt (1964) Previously the concept of level was only defined for congruence groups. Thus a congruence subgroup J” of (1) was defined to be of level n if P(n)< I’ and ifn is the least positive integer for which this inclusion is valid. That the two definitions agree for congruence ‘groups is a consequence of the relations A(n)Piqn)=Tin), A(nfigny=F(nj, (1.7.10) which were essentially proved by Wohlfahrt; see also Rankin (1969) and Leutbecher (1970). Here q is any positive integer. If we now assume that lev "=n, s0 that A(n) SPC T'(1), and that Fis a congruence group, then, for some qeZ", Mq)cI° From (1.7.10) we deduce that (2) 0. We say that f is meromorphic at a point p €€ if there exists an integer q=0 such that A is holomorphic at p, where Alz)!=(2—py'fte)s this implies in particular that h is continuous at p and that, Alp) = im (z ~p)*flz) , and write We say that f is meromorphic at <0 if f* is meromorphic at 0. A point p at which f is meromorphic but not holomorphic (so that40 Mapping properties q> 0) is called a pole of f. The smallest value of g that satisfies the conditions is called the order of the pole. If we can take q = 1, the pole is simple. At a pole p of f we have f(p) =. More generally, if fis meromorphic at p €C, we define ord(f, p) tobe the greatest integer q such that (z ~p)“f(z) isholomorphic at p.Ifq>0,q is the order of the zero of f at pif ¢<0, —q is the order of the pole of f at p. 2 ‘Suppose that f maps a domain D onto a subset D, of C. we say that f maps D conformally onto D,, oF that f is conformal on O, if and only if f is meromorphic on and f is bijective map of D onto B,. Iffis conformal on the domain D, it follows that its derivative /" exists and is non-zero at all points of D that are not poles of f. Also, if p isa pole of f in, then p isa simple pole. Further, the inverse function f-* of a conformal mapping f is again conformal. The proofs of these facts involve the use of Rouché’s theorem. Since a holomorphic function is continuous, it follows that every conformal mapping f is a homeomorphism of its domain D onto (0), which is also a domain, Conformal mappings have the property that angles between intersecting curves in the domain D are preserved in f(D) both as regards magnitude and direction. ‘We now consider the particular case when where Te 6. As we saw in §1.1, Tis a bijection of € onto itself. Ifc=0, then Aaya Get 8 ‘The mapping T is holomorphic on € and there is a simple pole at 0. Ife #0, the mapping T is holomorphic on € — {—d/c} and there isa simple pole at ~d/c. That is, in every case T is holomorphic on —{-d/c} with a simple pole at ~d/c. Hence T maps € confor- mally onto €. Note also that, for z #-d/c and z #0, pee las (ata ‘The converse result that, if f maps E conformally onto €, then fe 6, also holds, T= QAM) 24 Conformal mappings a It can be shown similarly that f maps C conformally onto C if and only if fe 6 and f(c0) = 0, Further, f maps H conformally onto Hif and only if fe 42. In these results the ‘if’ parts are straightforward, while the ‘only if’ parts require the use of Schwarz's lemma. Since we shall not require the ‘only if” parts, we content our- selves by showing that, if T'€ (3, then TH=H. For this purpose we write az+b atd where u, v, x and y are real. Now tiv = T(2) (2.1.2) and conversely ena (2.1.3) It follows that w €H if and only if 7 €H. ‘We note also that, if Te #2, then and, if Te (1), TR=P, TH=H Further, circles and straight lines are mapped by T onto circles or straight lines. Itissometimes convenient to write the transformation Te 6 ina different form, Let z and z, be two different finite complex numbers having finite images w, = Tz, (j= 1, 2). Then the transfor- mation arth can be written in the form (2.1.4) wow, Z—zy where K isa constant, Further, if z, is another finite point havinga Known finite image ws, then K is uniquely determined, namely42 Mapping properties Alternatively, since co = T(—d/c) we have weed (2.1.5) and this holds also if ¢ = 0. 2.2, Fixed points. For any Te @ and z€€ we define re=ce+d. (22.1) Observe that, although Tz and (~T)z are the same, (-T):z =~(T:2) In particular, f:2 = 1, (~I):z =—1, and, more generally, 4U%:2=41 (neZ). By (1.1.4), for any S, Te 6 and z #00, T'00, ST:2=(ya+ Sc)2 + yb + 6d =(yTz +8)(cz +d) =(S:Tz)(T:2). We thus have the important identity ST:z =(S:Tz)(T:2). (2.2.2) A particular case of this is = (1: Tz\(T:2). (2.2.3) ‘The equation (2.2.2) holds, in particular, for any S, T¢ andall zeH If Te and zeH, T:z is finite and non-zero. ‘A point z €C is called a fixed point of a mapping Te O if and only if Tz =z. We suppose in the first place that c #0, so that z #0, It then follows by induction from (2.2.2, 3) that Tiz=(Tiz!" (2.2.4) for any ne Z, where T is the associated matrix. The equation Tz =z is equivalent to cz?+(d—-a)z~b=0, which has two, not necessarily distinct, roots, namely (a-d)+[(a +ay'—4} 2¢ : yz (2.2.5) 22 Fixed points 43 We note also that z2a=-b/e, zt22=(a-d)fe, so that (Tiz(T:22) = (cz 4 d(czr+d) 1 (2.2.6) Itis clear that, when T’€ 0, the nature of the roots 21, 22 depends upon the sign of the real number (a +d)’—4. If (i) |tr 7] <2, then (a+d)?—4<0 and z, and 22 are conjugate complex numbers, one of which, say 2), lies in H. T'is then called an elliptic transformation and 2; and 2, are called elliptic fixed points. If (i) |tr 7}=2, then 2; =2; and we have, one real fixed point. T is called a parabolic transformation and 2, is called a parabolic fixed point. Finally (ii), if te T]>2, then z, and zz are distinct real numbers and T is called a hyperbolic transformation and z, and z> are called hype rbolic fixed points. ‘We now examine these possibilities in greater detail, for the case when Te I(1), still making the assumption that c #0, Our objectis to express the mapping 7 in the form (2.1.4), ie. (2.2.7) gna Tit) 2-27 where this is possible; see (2.1.5). Here w=Tz and w, Tz; (j= 1, 2), Note that, although 7: z, and T: we replace the matrix T by ~T, their ratio rem hange sign when unchanged (i) Elliptic transformations. Here {tr T1<2, there are two possibilities: (a) tr T=0, T=£L ‘VE for some Le TU), or (b) te T= 41, T=4L "PL for s=1,2 and some L eI). In case (a) Tz is equivalent to V(Lz)=Lz:ie. Lz isa fixed point for V and so L2, =i, Lz, =~i, Hence ind, by theorem 1.2.3, nab, 2=L (i) =%, (2.2.8) Here the bar denotes the complex conjugate. Since T we have, by (2.2.4), (T:z)'=—1 (j= 1,2) and therefore, by4a Mapping properties (2.2.6), K =(T:2,)/(T:21)=-1. Hence wor 7-2 woz, 2-2 (2.2.9) ‘The transformation T is of order 2. Incase (b), Tz =z isequivalent to P*(Lz) = Lz, wheres = 1 or 2. ‘An elementary calculation shows that both P and P* have fixed points prem (2.2.10) and A, so that Lz, =p, Lz, =. Hence nal p n=l e (2.2.11) Since T?=4P™=4I, (2.2.4) gives T:z,= +p or +p* and so T:z2= 4p" or £p, by (2.2.6). Hence K =p or p*, and the transfor- mation takes the form = Kt (2.2.12) z It is a transformation of order 3. 2 (ii) Parabolic transformations. Here tr T= +2 and, by theorem 1.23, T=£L7U"L for some eZ (q#0) and Le/(1). Thus Tz equivalent to U"(Lz)=Lz, ie. (Lz)-+q=Lz. Hence Lz =c and the single fixed point is 2=L 0. (2.2.13) Since ¢ #0, z1 is a finite rational number. Puta +d =2e, where e = +1, Then, by (2.2.5), z1=(a—d)/(2c) so that T:z) =e =itr T. If w= Te, etd e(z-z)+e woz (ance )iz—2) e(z—2) © Hence the transformation T can be expressed in the form (2.2.14) woz 2-2 So far we have assumed that ¢ #0. When c=0, T=+U® for some q€Z and tr T= £2. If q = Owe obtain the identical transfor- mation under which every point is fixed. When q#0, 22 Fixed points 45 £L™U"L (with L =D and we include T among the set of parabolic transformations. There is just one fixed point, namely 2;= 0, In place of (2.2.14), the transformation 7 takes the canonical form waztq (2.2.18) (iii) Hyperbolic transformations. Here |tt T|>2 and, since 22 is real, (2.1.5) and (2.2.6) give K =(T:2,)>0. The transformation has the canonical form (2.2.7) with K>0; clearly K #1, since nF From the above analysis we see that the transformations Tef\(1) can be divided into five classes: 1. The identity transformation, whose matrix T= +1. 2. Blliptic transformations of order 2. The matrix T of such a mapping is conjugate to + V and satisfies T? = —I, We denote by B,=(zeC:z=L7i,L ef} (2.2.16) the set of all elliptic fixed points of order 2 in #1 3. Elliptic transformations of order 3. The matsix T of such a mapping is conjugate to +P or +P? and satisfies T’= +1. We denote by Ey=(zeC:2=Lp,Le f(t} (2.2.17) the set of all elliptic fixed points of order 3 in. Here p is defined by (2.2.10). 4, Parabolic transformations. The matrix T of such a mapping is conjugate to £U* (qeZ,q #0). The set of all parabolic fixed points is P, since P={z €€:z =L~'c0, Lef(1)}. Parabolic fixed points are also called cusps for a reason that will be clear later on. 5. Hyperbolic transformations. The fixed points of such transfor- mations are less important in the theory. It is easy to see that they are all irrational numbers. Hyperbolic transformations are of infinite order. Note that 7E, =E,, TE; =E; for all Te (1) We also write E-E,UE,. (2.2.18) We now suppose that "is a subgroup of 1). The mappings Te f can be divided into five classes in a similar way, but some of46 Mapping properties these classes may be empty. Thus f* will contain elliptic transfor- mations only if it contains a mapping conjugate to V or P. For m=2, 3 we denote by E, (J) the set of all fixed points in #0 of elliptic transformations of order m belonging to f- Itfollows that B(=(zeC:2 =", Lef(), Lb VLef}, (2.2.19) E(=(2eC:z =L'p, Le f(t), L“PLef}. (2.2.20) Note that, we can omit L~'P*L since L~'P*L ef if and only if LPL € f and has the same fixed points. We write ED) =E( VE. (2.2.21) If Fis normal in (1) itis clear that E,, (7) is either E,, or the null set © We now suppose that z is any point of H’, The stabilizer of z (mod I) is defined to be the subset I, of I” consisting of all Te I” for which Tz = 2. Clearly I’, is a subgroup of I”. The stabilizer of z (mod I(1))is denoted by [,(1); the corresponding inhomogene- ous groups are denoted by /, and (1). Evidently /; isa subgroup of £,(), Further, if Le /() then Lh L=(LtL),. (2.2.22) The preceding discussion of fixed points shows that fate in the notation of §1.2. Also (1) and f,(1) are the cyclic groups of orders 2 and 3 generated by the mappings V and P, respectively. It follows immediately from this and (2.2.22) that Lol, when z = L~'co, LORQ)L, whenz=L-%, fay= i (2.2.23) LPL, whenz =L"'p, A={1 otherwise. Here L is any member of f(1), and z eH’. Note that, in every case, £,(1) is a cyclic group. ‘We now define the order of z (mod I) to be ate, P= (haf) (2.2.24) ‘The index on the right is possibly infinite. However, if has fi 23 Fundamental regions a7 index in /(1) then n(z, 2) is finite, For if the cyclic group f,(1) is generated by S it follows that $” ¢ [for some finite positive integer 1n,and n(z, I) is in fact the least such integer n, We note also that, it we take [y= f(1), P= fin theorem 1.1.2, with $ as above then, for 1sism, oy =n(z, LIL) =n(Lz, 1). (2.2.25) The transformation S is parabolic, elliptic or the identity according as 7€P, E or W'—PUE, respectively; in fact, we can take S= L“UL, LVL, LPL and Fin the four cases listed in (2.2.23). In the latter case (2, r)= 1. When z €P, we also call n(z, 1) the width of the cusp z (mod I’); if z= L~‘oo, F. is generated by L“'U"L, where n = n(z,r). When z €€,, (m = 2, 3), n(2,P)= 1 or ‘m according as z does or does not belong to E,(I’) It can be shown similarly that, when f* has finite index in ((1), some positive power of every hyperbolic transformation belongs to ’ and is, of course, hyperbolic. Hence f* always contains kyper- bolic transformations. ‘We note, in conclusion, that if "= I"(1) or PWN) for N> 1, then EU) =@ andsontz, I) = mitz e€,, (m=2, 3), Further,ifz €P, n(z,P)=6 and n(z, P(N)=N. 2.3. Fundamental regions. Let "be a subgroup of (1), so that the mappings T in / map M1’ onto itself. In what follows we shall only be interested in subsets of Hf, so that we are not, for example, interested in hyperbolic fixed points. Two points z,, z2 in H' are said to be congruent, or equivalent, (mod 1), if there exists a Te F such that 22> Tz, Itis easily verified that this is an equivalence relation, and we write 2,52, (mod I). ‘The equivalence class containing a point z € MH is called the orbit of z (mod J’) and is denoted by f2 ; instead of /(1)z we may write [2 }. Thus E,=[/], E,=[p] and, if fis of finite index in F(1), P= Foo= {co}. Clearly n(z,, 1) =n(z2, F) when 2,2, (mod I). A subset F of is called a proper fundamental region for [it F contains exactly one point from each orbit fz. By giving F the quotient topology induced by the topology on Ht (compactified48, ‘Mapping properties suitably at points of P) and the equivalence relation, F can be made into a connected Hausdorff space, which is, in fact, the Riemann surface associated with the group [\ We shall not, however, use any Riemann surface theory, although we may mention this theory at various points. In practice itis usually convenient to impose further conditions on F, such as that it is a simply connected subset of 1 and is bounded by curves of a prescribed form. Theorem 2.3.1. Let F be a proper fundamental region for a subgroup of P(A) and suppose that F = Ira: Fm, where the sets F,, are disjoint. Then, if Tye for each n eZ", the set IP-y ToF. is also a proper fundamental region for f- This is obvious, as all we have done is to choose Tz as a representative of fy rather than z for some Te f- The theorem is useful since it enables us to piece together fundamental regions in alternative ways that may be convenient for special purposes. Usually the number of non-null regions F, is finite Theorem 2.3.2. Let F be a proper fundamental region for a subgroup f of F(t) and suppose that Te fl). Then (i) TE is a proper fundamental region for the conjugate group TIT (ii) Inparticular, if Tef and T 2-41, then TE isa proper fundamental region for * and F ATF is either empty or consists of a single point £, which is a fixed point for T. (iii) A fixed point { for a mapping TeT cannot be ‘an interior point of F. {iv) The regions TE for T ef cover ’ without overlapping; if T; and T; are different transformations in [, then TF and TF have at most one point in common. Proof. (i), Ize and Te F(A), then T-'z eH and so there exists an Sef such that ST-'zeF. Then TST'z€ TF. Further, if TS,T~'z and TS,T~'z are two points of TF, where S$, and S, are in Fithen $,T-?z and $,T~'z are points in the same orbit '7~"2 lying in and so are identical. The original points 7S, 7~'z and TST"'z are therefore also ide Me Gi) Let Te f, T#+I and suppose that £eF OTF. Then (eF and T"'Z€F and, since these points are congruent (mod), {= TC; i.e. Ty =f. Hence £ is a fixed point for T and there is only ‘one such point in H’. (iil) If Zis an interior point of F, then there exists a neighbour- hood Nof { withN¢F. But TNisa neighbourhood of Tf = { and so. 23 Fundamental regions 49 therefore is N=No IN. But’ CN¢FandN'c TN¢ TF, so that’ SE TF, which is false 6). ) That 1'SUrerF is obvious, Further £¢ TF TH if and only if Ti!f €F 9 T;' TF and the last part follows from thisand (i). When [=f is easy to find a fundamental region. Theorem 2.3.3. Let ne 2", 6 =0 and put S,(6):=(z eH: -}k0. Proof. Since f+ consists of the transformations w=ztkn (neZ), it is obvious that S, is a fundamental region for f.». Take {H' and 6 >0, and put {= &+in, where & 7 are real. If 1 =0, then [£] 9 $4(6) = @, so that we may assume that 1 >0. If Tg€§,(5) for any Te F(1), then, by (2.1.2), UE +n?) + Dede +d? = nf. For given &, n, 6 the number of pairs of integers c, d that satisfy this inequality is finite; for the inequality states that the point (c, d) lies ina certain ellipse, For each coprime pair of integers c, d satisfying the inequality we can find a matrix T’¢ (1) with second row fc, d}; any other such matrix T is given by T=U"T' for some n cE. if TLES,(6), then TE +n €S4(6) and this can hold for at most k different values of n. This completes the proof. Corollary 2.3.3. The set € is a countable subset of isolated points of H.50 Mapping properties It follows immediately from the theorem that each point of E is isolated. Further, OC tvtedos.a/m from which countability follows, as each subset ({iJU[p)0 S,(1/n) is finite. Let f be a subgroup of /(1). A subset F of H is called a fundamental region for f if F contains at least one point of every orbit fz (z €H') and exactly one point whenever :@EUF. A proper fundamental region is therefore a fundamental region, and a fundamental region only differs from a proper fundamental region in the possible inclusion of a countable number of fixed points of F(1). Theorem 2.3.4. Theorem 2.3.1 holds with the word ‘proper’ omitted. So does theorem 2.3.2, except that in (ii) and (iv) the two different fundamental regions may intersect in more than one point of EP. This follows immediately. We note that in the proof of part (ii) of theorem 2.3.2, FTF is a set of isolated points and so cannot contain Ni Theorem 2.3.5. Let I’, and f, be subgroups of [\1) and suppose that of\=f,-R. Then, if F, is a fundamental region for fy, F=UTR is a fundamental region for f,. Proof. Let z €W4;; then there exists an S,e/, such that S,z €F, Write Si'T, where Sef, and Te®. Then S2z = TRchs Conversely, suppose that z and z'¢F,, where 2'= S22 for Sze f,, and that neither z nor 2’ is a fixed point of (1). Then z€TF,, 2'eT'Fy, where T, Te R. Hence T-'z and Tz’ are congruent (mod I"), lie in F; and are not fixed points of /(1). They are therefore identical; ie. z= S3'2'=S;'T'T'z. Since zis not a fixed point, we must have S7'TT'=2H; ie. TeLsT. This implies that T’= T and $= #1; ie. 2 24 Construction of fundamental regions SI Corollary 2.3.5. Under similar assumptions, if fy=2 +f, Urex TF; is a fundamental region for f° This is proved similarly, 2.4. Construction of fundamental regions for /(1) and its sub- groups. We denote by F, the set Fs FM UE, 24.1) where =Rez50,|z|=1) (2.4.2) and FP:m{ze€:01). (2.4.3) We include © in F", but not in F®. The closure of any transform TE” or TE (Te F(A) we call a triangle. Theorem 2.4.1. F, is a proper fundamental region for (1). Proof. Let £ be any point of H. We show first that some memter of the orbit [¢] lies in F,. For this purpose we may assume that ¢ is not congruent to any point on the frontier aF; of F., since every such point z either lies in F,, or else one of the congruent points z ~ 1, =1/z does. We call 1 =Im ¢ the height of ¢. We observe first that, for some nm € Z, £,= U"G eS (see (2.3.2)) and has the same height as ¢. If {, 2F,, then |¢|. Since S(n) [is finite, the process ultimately terminates after a finite number k of stages in the finding of a point & €[Z] 6. It remains to prove that each orbit contains only one point of F,. For suppose that 2, and z, are different congruent points of Fr Then both are finite and we may assume that yaeimz,>y,=Im 2,52 ‘Mapping properties Let z= Tz, where Te (1), so that yn jez. +a Ya Hence |ez, +d|=1 We cannot have |e|22, since no circle of radius rs} and orthogonal to R meets F,; also, if c=0, then T= U*, which is clearly impossible. We may therefore assume that c = 1. The only circles of unit radius centred at points of Z that meet F, are the circles le|=1 and |e +1 ‘There are thus two cases: (i) c=1,d=0,|z, In case (i) we must have T= U*V and either k =0, 21 ior p. In case (ii), T=U*P and we must have p. Hence, in both cases z, = z2 and this completes the proof of the theorem. 1, (i) c=d=1La=e. It follows from theorem 2.3.1 that , =F GUE} (2.4.4) is also a proper fundamental region for (1). For each Te (1) we write Fr=TF, &,=TE,, (2.4.5) It follows from theorem 2.3.2 that H= U tre U fF, ref rel ‘We note that the boundary (frontier) of F, consists of four ‘sides’ fy, Ule, ly and Viy, where y= {zaxtyis W3sy} 246) and tyiz-}sxs0,lz/=1,y>0). (2.4.7) Jy and ly are contained in F, but Uly ~ {20} and Viy ~{i} are not. Fig. 4 shows how the regions fit together. The angles between sides contained in the regions in question are marked. That the 24 Construction of fundamental regions 33 Fig. regions cluster closer and closer to the real axis is shown by the following theorem. Theorem 2.4.2. If Fy contains a point £ with fed] = 1/6. Im g=6, then Proof. Suppose that ¢ = Tz'is such a point, so that z ef, Write r=|z|,z=re® where }7 56 <}m. If 72S we have, by (2.1.2), lez +d? = (cx +a) +e?y?<) (cr-+d}? cos*($@) + (cr ~ ad)? sin"(30) (7/5) sin @.54 Mapping properties Since cos*({4) 24, sin*(J8) =4, we have rersind cirttd? bil 7 =k from which the theorem follows. We note that since (c, d) = 1, the condition fed] 1/6 is satisfied by only a finite number of pairs c, d. By theorem 1.2.1, /(1) =.» &, where ® isa set of mappings T such that, for each SeI(1), there is exactly one TER [c, 4] = +[y, 6]. We choose such a set in the following way. We first take T= J, which corresponds to y = 0,6 = 1. Then, for cach pairc, d with c>0, (c,d) = 1 we choose a and b to satisty lel3 » Tel=2 ‘The equation a’= a +nc on p. 9 shows that this determines T uniquely except when c= we take a =sgn d, so that ad ~ bi is possible. This , in which case d #0 and it follows that —} and such vertices cannot be interior points of F, since they are fixed points for f°; further only two triangles in F can contain such a vertex. It follows that F =F, and that = 3. ‘The transformations L mapping congruent sides of F, into each other cannot map any one of these three elliptic fixed points i, }(i-1), i-1 into another, so that the six sides fall into three pairs AwAL(v = 1, 2,3) with L=LA,, L, ef (v= 1,2, 3), where Ly, Lz and Ls have i, i(i—1) and i—1 as fixed points, respectively. We take Li , Lyt=PVP'= Vy, Ly:=P?VP=V; (2.4.16) and it follows from (1.3.9) that these three mappings do in fact generate the normal: Subgroup” of index 3in f(1); further I, P,P? is a right transversal of /* in F(1). It can be shown similarly that, when n= I, the only groups are ty and f?. We already know that the latter is a normal subgroup of (1) of index 2, and arguments of a similar nature to those given for f? show that it has F, UFy as a fundamental region and is generated by P and Pi:= V-'PV. This completes the proof of the theorem. In fig. 5 the fundamental regions of some of the groups that we have discussed are shown. We note that, since {J, P, P*} is a right transversal for the groups f,,(2), (2), fw(2) and fy, the region F;, defined by (2.4.15), is a fundamental,region for each of these four groups. The transformations 1, (see theorem 2.4.4) that mapa side A, into a corresponding side A} are, however, different in each2 ‘Mapping properties 5 5 7 g aie ae a ane eeteenae & : Peay k we ) te Bye bE 2 FY zg ees z -t ° “1 ° g ean 52 ul ho dog bs Rete ° ebb e Eee: ers & : eg £ ees S etre ine 3 v i: eee eee A a eFelrtges F cs tua) | - -- 1 Fig 5 : 7 2 Table 3. Groups of small index of the four cases. The generators of the four groups, as given in the penultimate column of table 3, are derived from the mappings L, illustrated in fig. 5. Theorem 2.4.4 can be applied in a similar way to find the generators of other subgroups of /*(1) such as (2) and £4(1). Note that £(2) has schema {2, 3, 2} and that both (6) and ‘£°(1) have schema {2, 3, 6}. Table 4 contains a variety of information about the nine normal subgroups of /(1) lying between £(6) and (1). The rank of each subgroup is the minimum number of generators. The relations between the groups are illustrated in fig. 3. We conclude by observing that, for n > 1, the principle congru- ence group /(n) has schema {2, 3, n}, and that the number A (n) of x incongruent cusps is Order Vertex sets in 4a F a Fry 5 5 5 ' 2 2 5 3 3 6 t nw rp f.2) 2 hea fia 8 2) 3 (n=), ye (1 7) (n>2). im= (2.4.17)65 ome ne nu wy SY sett cer mg samt anes) nn en tcpow){t eee ty bt ve (pow | HI NY wens bare cexnomss teran[ re e € J Ost 295 (7 pou) d= Tee ieee a Nee aw souiBueD 17903 , we (9)y pue (1) J usamiaq sdnoxBqns [eULION “py AGeL66 ‘Mapping properties 2.5. Further results. The method that we have used for obtaining fundamental regions for (1) and its subgroups is convenient for the purposes we have in mind. However, if we had been consider ing the more general case of an arbitrary discontinuous subgroup /* of 2, we should probably have found it advantageous to introduce hyperbolic geometry and to adopt aslightly different definition of a fundamental region. A simple account of this more general theory can be found, for example, in the book by Lehner (1966), or the more advanced treatises by Fricke and Klein (1926) and Lehner (1964) may be consulted. We content ourselves here by sketching briefly how this general theory could be applied to a subgroup [* of finite index w in /*(1); in this sketch we shall introduce a number of concepts that we do not define, ‘The hyperbolic length ds of an element of arcin His defined by (ds:=y "(dx + (dy, (2.5.1) so that the hyperbolic length of a piecewise differentiable curve C inHis dx? | (dy\2}bde fof AGS esa Ina similar way, the hyperbolic area A(E) of a measurable subset E of His defined to be fforacas (2.5.3) AB this can be infinite even when the ordinary Euclidean area of E is finite, A curve in His called a hyperbolic straight line if itis either a semicircle in H centred at a point of R, or is an ordinary straight line in H that is orthogonal to R. If z, and z, are distinct points of H, there is a unique hyperbolic straight line passing through z, and 2, and the arc of this hyperbolic straight line joining z, and z, is the curve C of smallest hyperbolic length joining z, to z>. The hyper- bolic distance d(z,, 2.) of z, fromz2isdefined to be L(C);if z1= 22, we put d(z,, z,)=0. It can be shown that d(z;, 22) defines a metric on H, and the corresponding metric topology on H is identical with the natural topology onH. Itis convenient to extend this topology from to H 25 Further results 67 by giving each point of P a suitable base of neighbourhoods; see Lehner (1964), chapter 4. ‘The hyperbolic metric has the useful property that L(7C)= L(C)and A(TE) = A(E) for every T¢ (2. Thus every fundamental region for /has the same hyperbolic area. Further, this hyperbolic area is easy to evaluate by using the Gauss-Bonnet formula, which states that the area of a triangle bounded by hyperbolic line segments is 7—(a+8+y), where a, B and y are the interior angles between the sides, This gives, for example, for a fundamen- tal region F for A= 2A) = kur, (2.5.4) since F has interior angles 0, tw and 7. Different definitions of a fundamental region are given by different authors. What we have called a proper fundamental region is called a fundamental set by Lehner (1966) and Schoeneberg (1974). In his two books Lehner defines a subset D of H to be a fundamental region for f*if (i) D is open in H, (ii) no two distinct points of D are congruent modulo I” and (iii) every point of His congruent to a point of the closure D of B in H. On the other hand, Macbeath (1961) takes D to be closed and modifies (ii) accordingly. The interior of either of the fundamental regions F or F of theorem 2.4.3 is a fundamental region according to Lehner. Let 29€H—E(F) and define D = D(2,) to be the set of all points ze such that (2, 29) m, w(S, T=} 0 if-a0. (3.2.7) In order to examine further the properties of w(S, T) we intro- duce the following two subgroups of £2, namely _[e ‘| =: {Ren [i o}e>acen} (3.2.8) and ‘cosé ~sin ing cos él" E={Xe:x a0 1 ife>0 orife=0,d<0 (3.2.18) wR, T)=w(T,-R76 Automorphic factors and multiplier systems and wT wrneft ife=0,d<0, 0 otherwise. iltateds ‘These hold for Re Sand Te 2. Further, if $= RiX;, T= X2R:, where R, and R; belong to S and X\, Xze&, then, by (3.2.5, 15), w(S, T)= W(X, X2) = Lion X,+ph X2—ph X.X3}. (3.2.20) ‘We use these results to prove that, for any Re ¥ and Te, w(TRT, T)=w(T, T'RT)=0. 3.2.21) It is enough to prove that w(TRT"', T)=0, Take T=X,R, as (3.2.10) so that, by (3.2.15), w(TRT-', T) = w(XaRsX3', X2), where R;=R,RRz'¢ 5. Now X;Rs=RsXs, where Rie X and X,¢E. By (3.2.14), ph(X.X3')= ph Xe Hence, by (3.2.15, 20), w(TRT, T) = w(X.X3',X2) ph X;, = EApM(XX5") + ph Xs ph Xa) =0. ‘The proof that w(T, T-'RT) = 0 is similar. We can use (3.2.21) to prove the following generalization: w(T,RT;', Ty Tz!) = w(TiTs, T:RTs') (3.2.22) for all T;, T3€ and Re. For, by (3.2.4), W(TIRT |, TTS!) = w(TR, T3!)+ (DRT T)—w(T, Ts) w(Ty RTs') + w(Ty', T2RT;') —w(Ty, Ts!) =w(TiT3', TRT:'). Automorphic factors on subgroups of the modular group 77 Finally, if L, T€ and R, € &, itfollows from (3.2.4, 15, 17, 21) that o(L, L'RyLT)o(L“'RyL, T)=o(L, 7). (3.2.23) 3.3. Automorphic factors on subgroups of the modular group. In this section we assume that J" is a subgroup of (1) of finite index and that v is an AF of weight k on I We also assume that ~Fe I ‘We consider the form of »(T, z)and the associated multiplier v(T), when T is a parabolic or elliptic matrix. We note first that, for any Te 12 and qeZ, B(TUY, 2)= (cz +d +cq)* = wT, U2). G31) Suppose now that £ = Looe P, where L ¢1°(1). Then [; is gener- ated by T:=LU"L", where n,, = n(L00, I) (see §2.2); for con- venience we write n for m,. Since U" eI" =L~'TL, we have, by 1.10), (U2) =v (UYU 2) = v(U") 28, (3.3.2) say, where Nx, <1. Hence, by (3.1.20), =0(To(LU"L", L)/o(L, U") o(T)=o(LU"L"), (3.3.3) by (3.2.1, 17, 21), Fora given AF », the number x, depends only on the orbit f¥ of the cusp ¢ = Lo. To show this, suppose that Moo also belongs to Tt, where MeT(1), so that M=SLU*, where qeZ and Sel” Then my =n, =n, and we have MU"M '= STS". Hence v(STS“') ST)o(S ')o(ST, S*') = v(T}v(S")v(S)o(ST, SoS, T) =n ZSLS 2S, T) eMaaarets gyi® Automorphic factors and multiplier systems Now, by (3.2.4), with S, = ST, $,=S~' and $ 7 w(ST, S-)+w(S, T)~w(S"', S)= w(S, T)~w(STS™, S) by (3.2.22) with R = U", T= SL and T, = L. It follows that ern mT) =e, $0 that ky = kr. We call «, the (cusp) parameter associated with the cusp Loo and the AF y, and write Kp =«(L%, Tv), By induction from (3.1.5) and (3.3.2, 3) we have, for any m eZ, ermine = we(U™, 2) =0(U™" = (LU L™). (3.3.4) We suppose that I has an elliptic fixed point ¢=Lp (see (2.2.10), so that T:= LPL“ F and Ty = ¢. By (3.1.1), (L: Pe\(P:p) = LP:p =(LPL™:Lp\(L:0) so that, since Pp =p, T:f= LPL": Lp = =ltp=e™. (3.3.5) By induction from (3.1.5), oT, Q=((T OY (rez). (3.3.6) Now T° unique and »(~f, z)=1, so that (3.3.5) shows that for some teger m, (0<=m, <3), p™ =u, 2) = (T:g)'0(T) =e" v7). 3.7) Hence T) = v(LPL)=p™e"™. G38) For any re Z, Tr (Tey Therefore by (3.3.6), (LPL) =o(T) =, OMT 0" pmerner, 83.9) provided that ~2sr<3; for then ph(7":¢)=~m1/3. 3.3 Automorphie factorson subgroups of the modular group 79 It can be shown similarly that, if Li is a fixed point for P then LVL“ eT and LVL) =(-1te™? (15752) (3.3.10) for some integer s, (05, <2). We call m, and s, the parameters associated with the automor- phic factors (or multiplier system) and the elliptic fixed points Lp and Li, respectively. It is easy to show that they depend only on the orbits FLp and Li, respectively. For example, if M = SLP*, where qeZ and Se/, then, by (3.3.7) and (3.1.17) (with z= Lp, L=S, and T replaced by STS"), p™ = v(STS', SLp) = v(T, Lp) = 0™, since »° = 1; then Min, =m, We now consider how many different multiplier systems of weight k can be defined on the group I° Suppose that v, and v, are two multiplier systems of the same dimension on I, and write Xx(T)= v(T)/vi(T) for all Te I. Then, by (3.1.14), X(ST)=x(S)x(T) for all S, Te” and, by (3.1.13), x(—) = L.It follows that x is a (linear) character of the group I° with the property that x(—7) = x(T) for all Te T°; i.e. x is a character on the inhomogeneous group f Conversely, if x is any character on the inhomogeneous group /* and if v, is any MS on I of weight k, then so is v2, where vx(T) =x(T)v(T) (TEL). The number of different multiplier systems of weight k on "is therefore either zero, or else is equal to fh, the number of characters on f-. Clearly h is equal to the order of, the factor group f/f”, where [* is the commutator subgroup of f°; A may be infinite. In particular, if k is an even integer, then wis an AF for every group [7 so that, in this case, the number of different multiplier systems is h. The MS v associated with the AF pis called a constant multiplier system, since o(7)=1 forall Tel Baa) For this MS, x, =0 for all LeJ{(1), by (3.3.3). Further, by 8.3.9, 10), m, = 1k (mod3) ifLpeEMM) (keven) (3.3.12)80 Automorphic factors and multiplier systems and s,=ik(mod2) if LicE(M) (keven). (3.3.13) We denote the constant MS by the figure 1. In certain cases we can draw similar conclusions when k is an odd integer, and we make this assumption for the remainder of the present section. If -1€T7, then w is an AF on I” and the associated MS v is constant on J” since it satisfies (3.3.11). If —I€J;, however, things are not quite so simple. This often arises when we are given the inhomogeneous group f and I is taken to be the sct of all matrices associated with mappings in f°. If wwe can find a subgroup I of index 2 in /, which does not contain =I, then [‘=/* and we can define an AF v and a corresponding MS v on Fas follows: W(T,z)=e(T,z)(Tel), v(T,z)=—n(T,2)(Tel-I™), v(T)=1(Tel), o(M)=-1(Tel-P), (3.3.14) It isclear that these equations do in fact define an AF and MS on I’; further v is a constant MS on I. Since -I is the square of every elliptic matrix +L'VL, it follows that a necessary condition for such a subgroup 1* to exist is that EWM=9. (3.3.15) Petersson (19382) (III, Satz 18, p. 566) has proved a general theorem from which it follows that (3.3.15) is also a sufficient condition. This can also be deduced, in the case in which we are interested, from theorem 1.7.1 with G=/(1) and H =f, For the groups By (if they exist) have order 3 and are generated by mappings L5'P?L», where L, € [(1),while the free group F (if it exists) is generated by mappings T;, Tz, ..., T» say. Accordingly, we can define I* to be the group generated by the matrices L,'P*L, and the matrices +T7,, +T2,...,£T,, for any fixed choice of the signs +. It is clear that —/¢ J* and that there are 2” different choices of I. For example, when f=, FQ), fin) (n>2), 3.3 Automorphic factors on subgroups of the modular group 81 we may take rer, PQ), Mn) (n>2), respectively. For f(n) (n> 2) there are 2"**’* different choices for T* of which '(n) is the simplest; here 4 = (n) and is given by theorem 1.4.1, In the case of (2), we note that T*(2)= (UL?) W") (3.3.16) {see (1.5.4)), and that there are four possible choices for I, namely T*(2), P“F*(2)P and P7*(2)P?, (3.3.17) and F(2)=(-U)X-W), (3.3.18) which is normal in (1). The core of F*(2), ic. the intersection of the conjugate groups (3.3.17), is '(4). An alternative definition of FQ) is P(2)={TeT():a=d = (mod 4), b-c =e ~1 (mod 4), = +1}, (3.3.19) so that the associated MS # is given by BD =-1 for TeV), (3.3.20) This multiplier system dis identical with each conjugate 6%, by (3.1.21). On the other hand, the three multiplier systems », v”, 0” defined on 1"(2) by (3.3.14) with [= /"*(2) are different and take the common value 1 on the core (4). Further properties of these ‘groups and their multiplier systems are given in §5.6. Fig. 6 displays the relationship between some of these groups of level 2 and 4, All the groups illustrated are normal subgroups of T(1), with the exception of 1°*(2), (2) and F(4). In order to avoid complicating the figute, these three groups have been selected from their families of conjugates, since they are of particu- lar importance in the study of the theta function ,(z). It may be noted that (2) is the normalizer in I'(1) of both 'v(4) and 17*(2) That f(2)¢I* follows because -U"= PP, —W'= PIPL82 Automorphic factors and multiplier systems Fig. 6. Homogeneous groups between I!) and (4). 3.4, Multiplier systems on 1'(1), and I, We now take "= (1) and assume that » is an AF of weight k for (1), where k ER, The associated MS we denote by v, and we put »(U, 2)=0(U) = B40 For all LeT(),v'=», ot =v and nm, =1. Further, «.=K, m, =m and s, =s, say, for all LeT(1), and et = y(U)=v(LUL“) =a. (3.4.2) Since P= VU, it follows from (3.1.5) that (P, z)= ar V, Uz), so that, by (3.1.10), (2+ 1)fo(P) = w(z + 1)*0(V). (3.4.3) By (3.3.9, 10), v(P)=p™e™7, o(V)=(~1)'e™? (3.4.4) 34 Multiplier systems on (1), 7? and P? 83 and therefore, by (3.4.3), we ~1)'p" =:x(U), G.4.5) where x(U) is a sixth root of unity; the reason for this notation will be explained later. Hence, by (3.1.13) and (3.3.4, 9, 10), v(-D=0%, v(U")=w" (mez), 6.46) o(P)=0 (2573), o(V)=a” (157s). G47) Since W=UP and, by (3.2.17), w(U, P)=0, we deduce that v(W) =" and it is easily shown by induction, with the help of (3.2.7), that o(W™)=0°" (mez). 3.4.8) Since o(S, T) is a power of e*** = «w, it follows that, for every TeT(\), there exists an integer r(7) such that (=o, (3.4.9) Equation (3.4.5) defines x(U) as the sixth root of unity (~1)'p”. Now, by theorem 1.3.2, the factor group [(1)/f*(1) is a cyclic group generated by the coset containing U. We use this to extend the domain of the function x to the whole group (1); we define T)=(U” when Te*U"T(1). (3.4.10) It follows that x is a character on (1) with the property that 1. Hence, if one MS of weight k exists, then exactly six x(-D such multiplier systems exist, one for each choice of character, ic. ‘one for each choice of x(U). For any such choice, »(7) is given by (3.4.9), where w is defined in terms of y(U) by (3.4.5). We shail show in §6.4 that a MS of weight k on J"(1) does exist, and it follows that there are exactly six such multiplier systems for each k. Similar arguments can be applied to subgroups of (1). For example, if =" =17(2), we obtain, from (3.3.9), P)=e"*¥(P), v(P)=e™ YP) (3.4.11) where P, = V"'PV. Here x(P) and y(P,) are cube roots of unity. Since any MS on (1) is also a MS on any subgroup, it follows that there are exactly nine different multiplier systems on 1. We note that the commutator factor group /°/f* is isomorphic to the direct product of two cyclic groups of order 3.84 Automorphic factors and multiplier systems It can be shown similarly from (3.3.10) that there are eight multiplier systems on I”, for each k €®, and that o(Vy=e™@2x(V), of Vee 2x(V;) (3.4.12) for j= 1,2 (see (1.3.10)), where x(V)=x(V))=x7(V2)= 1. In fact, [*//*" is isomorphic to the direct product of three cyclic groups of order 2. For the remaining groups in tables 3 and 4 the commutator subgroups have infinite index, so that there are infinitely many multiplier systems of weight k for each of these subgroups. We conclude this section by determining explicitly the integer 1(T) appearing in the formula (3.4.9) for v(T), where w= o(U) =e™(~1)'p™ = ey (U). (3.4.13) We note that, since VP?=—W and VP*=U, it follows from theorem 1.2.5 that every T I"(1) can be expressed uniquely in the form T=P"SV* (sps2,02q=3), (3.4.14) where S is of the form S=U"WhUeW UW (a, 20, b, 20; 1s fm). (34.15) ‘The matrices $ form a free semigroup with generators U and W; wwe denote this semigroup by I”. The entries of each matrix Se" are non-negative integers. Since it is easily verified that, if SEI”, then the matrix P?? $ V* in (3.4.14) has non-negative entries only when p =q =0, it follows that T=(SeP(1):@=0,B=0, 720,520}, (3.4.16) We therefore have TU) =(P?)-T* (V), (3.4.17) Instead of (3.4.14) we shall find it convenient to use the equiva- lent representation T=(-1)P°SV" (OSrs1,05s52,0<1=1), (3.4.18) where $ is given by (3.4.15), and we shall put (3.4.19) 34 ‘Multiplier systems on (1), 1? and 85 We note that, if S=R\X,=X2Ro, P= RX, P7=RXe R2V'RKRs, where Rye 3, X/e 3 (1=/=5), then ph X,e[0,}7], phXz¢[0, 49], ph X>=4n, phX.=in, ph V-'=—4, phX,e(-7, 0), so that, by (3.2.15, 20), w(Si, $,)=0 for all $,, $.¢1™ and w(S, V") = w(P, S)= w(P?, S) = w(P, SV) = w(P*, SV (3.4.20) It follows by induction from (3.1.14), (3.2.7) and (3.4.6, 8, 19) that 1S) = ao~bo, if S is given by (3.4.15) and, by (3.4.7, 20), (P'S V"') = =25 +ao~ bot 31. It remains to evaluate r(—T), where T= P” SV‘. We have o(-T) = v(-Dv(To(-h, T)= 0 *o(-1, M(T).. By (3.2.18), o(~2, T) = 1 except when c >0 or when c =0,d<0, in which cases o(—I,7) =o". In table 5 we list the values of cd and o(-I, 7), in terms of the entries of S for the matrix T= P'S V™, We note, for example, that, for s=2, 1=1 we cannot have since then By =—I, which is impossible for Se I~ Table 5 Ot eite a tO eat ea fees issOssee OFe theta HL y aty a ~f ~(p+8)-B $ B46 B y aty a D Oe io Maat erie Re ol When 5 =1=0, we have o(-J,T)=w"? except for T=$ =U" (B= 0), when o(-1, T) = 1. We therefore have, for T expressed in the form (3.4.18), uny-{ 6-28 +ag—by ifr =1,05852,1=0,5+b,>0, ~6r~25 +a~bo+3¢ otherwise. B4.21)86 Automorphic factors and multiplier systems 3.5. Further results. The most complete treatment of multiplier systems is to be found in Petersson (1938a); see, in particular, | §2 and III, §§6, 7. Petersson’s theory includes the case when the dimension is complex and applies to any horocyclic group (Grenzkreisgruppe). An earlier less complete discussion for real dimension is given in Petersson (1930). Itis possible to tabulate the values of w(S, 7) for all signs of the parameters 7, 5, a, ¢ and ya + 8c; see Satz 4 on pp. 44-5 of the first paper. Other authors, such as Rademacher (1955), van Lint (1957, 1958) and Lehner (1968), have studied multiplier systems associated with particular groups and particular modular forms. The fact that (see (1.3.10) w(Va, V2)=0, w(V2, Vi shows that it is not necessarily the case that w(S, T)=w(T, S) 35.2) holds for all S, Tin (1). However, (3.5.2) holds whenever $ and T commute. To show this we may assume, in view of (3.2.17, 1 that neither § nor T is +1, Let ¢ be a fixed point of S lying in H. Since STE= TSC = TE, Tg is also a fixed point of S inf; so is 7, and clearly T° = {, since S has at most two fixed points. We show that Te=6. (3.5.3) This is obvious if S is elliptic or parabolic. Hence we may assume that S is hyperbolic, and therefore T*is hyperbolic since it has ¢ as a fixed point. It follows that T is hyperbolic and that ¢ is one of its fixed points; ie. (3.5.3) holds. Now, by (3.2.2), 1 3.5.1) w(S, = stloas: T2+log(T:z)—log(ST:2z)}, so that w(S, T)-w(T, 8) = ziilloats: Ta) +log(T:z)—log(T: Sz) —log(S:2)}. If we let z tend to { in Hi we deduce (3.5.2). 35 Further results a7 ‘The relation (3.2.4) can be expressed by stating that w is a 2-cocyele of the trivial I™-module Z, where Iisa subgroup of (0); ie. we ZF, Z), The cohomology class of w consists of all mem- bers w*e Z°5 Z) such that w*-we BU, Z), 3.5.4) . Here fe B°(F, Z) when {(ST) = @(S)+e(T)- 9ST) > Z. We call w* symmetric it w(S,T)=w(T, 8) for all $ and T in I, and semisymmetric if this holds whenever S and T commute. We recall that we have just shown that w is semisymmetric. Leutbecher (1970) has shown, in a rather more general setting, that the cohomology class of an element of Z°(V",Z) contains a symmetric cocycle if and only if the element is semisymmetric; see also Wohlfahrt (1972). It follows that we can find a symmetric w* such that (3.5.4) holds, where w is the function studied in §§3.2-4. The multiplier system v* associated with w* has the convenient property that o*(ST) = v*(TS) for all S and T, but this is offset by the disadvantage that, in the definitions of o* and w®, it is necessary to choose ph(T:z) to depend upon the first row of T as well as on the second the group of 2-coboundari for some ¢:4: General properties of modular forms 4.1, Definitions and general theorems. The object of this chapter is to define what is meant by modular forms and functions and to discuss the distribution of their zeros and poles in a fundamental region. The dimensions of various vector spaces of modular forms. are investigated and simple estimates for the magnitude of their Fourier coefficients are given. Algebraic, as well as linear, relations between modular functions are also considered. ‘Throughout this chapter we assume that "is a subgroup I'(1) of finite index y: and let » be an automorphic factor of weight k on where k ER. We write v for the associated multiplier system and assume that ~Ie I’ Note that (7(1):]= a =(/(W: Sf). We denote by M'(I’, k, v) the set of all functions f that possess the following two properties: (D) f is a meromorphic function on H. (II) For all Te and all z eM (including poles of f) f(Tz)= »(T, z)f(2). (4.1.1) We recall that »(T,z)=(T:z)'o(T), (4.1.2) where the multiplier »(7) is of unit modulus and depends on T and k but not on z, and »(T, z) is holomorphic on H. ‘The zero function 0, which takes the value zero at all points of H, clearly belongs to M’(/, k, v) for all k €Rand all multiplier systems v. A function f belonging to M'(I, k, v) is called an unrestricted modular form of weightt k for the group I’; the adjective ‘unre- stricted’ is used since we shall find it convenient to impose further conditions on f later on. We do not yet know whether any + We follow Ogg (1973) in sing this terminology. The classical usage isto call fa form of dimension ~K (rather than weight k), but this conflicts with the use of the word dimension in vector-space theory. On the other hand, the use of the ‘word weight does not avoid all ambiguity, since some authors call fa form of weight 4 4t Definitions and general theorems 89 unrestricted modular forms, other than the zero form, exist and it will only be in the next chapter that we shall construct such forms. IffeM'(D, k, v) and L€ PU), the L-transform f, of fis defined fulz)= (IL ={u(L, 2)F 'f(L2). (4.1.3) Here, as in (3.1.8), w(L,2) operator |depends on k. Theorem 4.1.1. Suppose that fe M'(I,k,v) and that L, Ly, Lr€ T(1). Then we have: @ feM(L“TL, k,v"). (ii) IE La)=o(L, La f\L)|L2, where o(L,, L,) is defined by G.L15). TED fn emf, (iv) If (= Loo, then fiz tm) =e f.(2) (4.1.4) forall z <1, where m= n(L.00, I). the width of the cusp ¢ (mod T), ‘and x, is its parameter. by )’. Observe that the stroke =o(T, L)o(T)fc; in particular, f= v(T)f, and Proof. (i) Fx is clearly meromorphic on H, For Tel” and zeH, f(L“'TLhz)={u(L, L'TL2)}"f(TLz) stall, L 'TL2)p'0(T, Lz fz) = (L"TL, afl), by (3.1.17) i) By (4.1.1, 3) and (3.1.45), (FO MEM| Le = fe Lr = {a (La, 2) fi Laz) = fw (Le, z)e (Li, Liz) f(LiLiz) (Likes 2 ie (Lay 2) (Lr, Lab fry) fisesl2)/er(L, La). (iii) This follows from (ii) since fle) =MalT ay 'f(T2) = eT, 2)/a 7, 20 fl2) = o(Tfl2), The result for f_, uses (3.1.13) and (3.2.18).90 General properties of modular forms (iv) Since Ue L“'IL, we have, by (i) and (3.3.2), fala tm) =fi(U"z) = 9 (U, 2)fu2) = er fi.l2). Part (iii) of the theorem shows that, if constant factors of unit modulus are neglected, there is one transform of f for each right coset of fin (1). We saw in the proof of theorem 2.4.3 that these right cosets can be grouped into A=A(I’) families fLU* (0=qn} for some 0, then F,, is holomorphic on {1:0<|1|0 and an integer N, such that F(O= & a.(by" (4.1.10) for 0<|t|<6,; hence fiz)=ermmrm F ag(L) eon for Im z >{n, log(1/6:)}/(27) =: ‘The class of all functions f satisfying (1), (II) and (III) we denote by M(Pk, 0). Our earlier remarks show that essentially only A functions f, are affected by the condition (III). The condition affects the behaviour of f, near the point «0; i.e., condition (IID isa restriction on the behaviour of the unrestricted modular form f near each cusp. Since the behaviour of f,, at L00 is determined by the behaviour of fir, at ©, we note that, if feM(Uk,v), then fie M(L~'TL, k, v') for each L €I(1). We note that, in particular, 0¢ MUU’ k, v) where 0 denotes the zero form. If fe MU k, v) and f #0, we can take N, in (4.1.10) to be such that ay, (L) #0. The power series for f, (2) then begins with anon-zero constant multiple of "°°: and we call x: +N, the order of f at the cusp Loo (mod I) and write 11D, +N, = 0rd(f, L20, 1). (4.1.12) If eH, we let ord(f, &.P te (4.1.13) where /; is the stabilizer of { (mod J). It follows that 1 F meh © iffeb.un, ord f. 61) = (4.1.14) jord(f,g) if feb. (1), 26M.92 General properties of modular forms ‘Theorem 4.1.2. Suppose that fe MUI, k, v) and that f #0. Then we have: (i) if (i) =a (mod N), ord, £141) = 01d ffs. ie), where L eT (1), then s.+2ord(f, £1) #0 (mod 2), where 5, is defined in (3.3.10). If (LVL) = 1, then 4k €Z and Ak +2 ord(f, , P)=0 (mod 2). Gi) 1f = Lp €B(1), where Le T(), then my +3 ordi f, & F)=0 (mod 3), where m, is defined in (3.3.9). If v (LPL™)= 1, then tk €Z and tk +3 ord(f, £2) =0 (mod 3). Proof. (i) lf £1 € P the result follows, since N; and «;, depend on the orbit £%. Suppose therefore that ¢2= TZ,eH, where Te, and put w = Tz. Then for z ina neighbourhood of Z, by (2.1.4, 5), (8) "hm= we bi (4.1.15) for any ne Z. It follows from this that ord(f, {,) = ord(f, ¢) and so ord( fg, 1) = ord f, ¢2. P)- We now assume that {=LpeE,(I"), where Le T(1), so that T.=LPL“ eT and Tf =¢. Write 4.1.16) Put (4.1.17) er arr errereennrmenar En —enteneenann en 41 Definitions and general theorems 93 where ~ord(f, 2), so that is holomorphic and non-zero on a neighbourhood N of r=0. We may choose Ni sufficiently small so, that, for reN, ph (4.1.18) Now so that, by (4.1.18), A It follows from (4.1.15-17) and (3.3.9) that (o'r) —1, When k is an even integer and the MS is constant, rm, and s,, can be replaced by k/2 in the congruences and it follows that e@={E(1-4)} Getty, 4.120) where the braces denote the fractional part. ‘We denote by HI, k, v) the subset of M(F, k, v) consisting of all forms f that are holomorphic on H. If fe H(F k, v) and L € I(1), the expansion (4.1.11) is valid for all z €H. If fe M(I; k, v) and ord(f, ¢, 1) 20 for all eH! we call f an entire modular form and denote by {I', k, v} the class of all such forms, including the zero form in this class. If, in addition, ord(f, £7) >0 for all EP, or if {=0, we call fa cusp form; the class of all cusp forms is denoted by {I k, vlp If {= Looe P andord(f, {, F)= 4+ Nz =n, wesay that f has zero at if n >0 anda pole at £ if n <0; the values taken by f at £ in the two cases are 0 and co, respectively. If n = 0 we can say that f does not vanish and is not infinite at {. It may not, however, be meaningful to say that f takes some particular value ay at ¢ when n =0. For we have defined the behaviour of f at £ in terms of the behaviour of f, at «0. Different choices of L for which ¢ = Leo AI ceR, then {x}:=x—[2], where [x] is the integral part of x; ie. (r] is the greatest integer that does not exceed x. 41 Definitions and general theorems 95 sive, in general, different constant terms ao(Z.) in the power series (4.1:11) differing by factors of unit modulus, Note that AD k, voS AP k, v}S HUF, k, 0) SMU, k, v). If fe MIL k, v) and F is a fundamental region for / given by F=UFor UF, where [(1)=/*- &, then f has only a finite number of zeros and poles in F. It is enough to prove that, for each L € ®, f has only a finite number of zeros and poles in F,, i.e. that f, has only a finite number of zeros and poles in F;. Now the expansions (4.1.10, 11) show that f,(z) has only a finite number of zeros and poles in the part of F, for which Im z=2n;, while the part of F, for which Imz=2n, is compact and so can contain only a finite number of zeros and poles. Hence ord(f, ¢,") is non-zero at only a finite number of points ¢ belonging to a proper fundamental region F* for f. Wedefine the total order ord(f, 1) of f (mod I) to be the sum of ord(f, £1) for all points ¢¢F*. It is clear that ord(f,) is independent of the choice of F*. We shall show that it depends only on the index 4 of fin (1) and on the weight k. We have ord(f, = LY ord(f.g. 1. (4.1.21) Aah, eA] If ref, and § generates £,(1), then, by theorem 1.1.2, with fa), n=f, tel=No2=fU x2 =f Le, where the m points L,z (1m) are incongruent (mod I). Put nm=n(Lz, 0). Then E ord f= F ordlf. L219 center i =f EF oaiprs.19 rz ord(f, Lz, 7) vem Lz.) (4.1.22)96 General properties of modular forms where ((1) = f- ®. This follows, in the first place, for 2 = but is true for any transversal 2, In particular, if z =o, so that [2]=P, we have by (4.1.12, 22) E ord D= E (ee +N/ns (4.1.23) where, as usual, n, = (Leo, P). When z ¢H, aaa! pee IED = Rai Era Le). (4.1.24) since PialniLz, 1) == Dh Theorem 4.1.3. Suppose that f MUL, k, v),f #0 and that (1) = TR. Let &):= II fl). (4.1.25) Then ge M(I(), wk, 04), where pis the index of Fin (1) and the MS v* is associated with the AF v* defined by oS, Liz)u(L2) en! Serra) where LT = SL, and SEF, Lye. g(z)=g(25 (4.1.26) Proof. For any Te I'(1) we have g(T2)= M1 f.(T2)= I] a(L, Pe) fLT2). Now LT=SL,, where S¢I" and L,€% are uniquely determined by this equation. Hence a(Tz)= T] v(S, Liz Mu(L, Tzy'f (Liz) ten = a v(S, Liz Mull 2)/m(L, Te) file). Now L, runs through ® as L does, and so g(Tz)=v(T, z)g(z), (4.1.27) 4t Definitions and general theorems 7 where We note that v*(T, 2) is uniquely determined by T; for fix> and that by (3.1.10, 15), »*(7, 2)=0*(T)u(T, 2), where soe 7 USIAS Ls) OO Toe Thus |»*(T)| = 1; to verify that p* is an AF for ["(1) it suffices to prove that v*(ST, z)=v*(S, Tz)v*(T, z) for all S, Te T(1). By (4.1.27), WMST, z)gl2) = glST2) =¥"4s, Te)Q(T2) =v"(S, Tz w(T, z)g(z), and the required result follows since g #0. Since g is clearly meromorphic on H. it rema satisfies condition (II). By (4.1.11, 25), R, (TeP(y. to prove that g aley=e™ SF baer" (by #0), where q is real, 1 is the least common multi ny (L€&), and Imz > >0, say. Write t le of the A numbers 207" and put C= E bt" =e e(2), the series being convergent for sufficiently small. Since g¢ M'(I"(1), ku, v*) it follows from theorem 4.1.1 {iv) that there exists a « with O<« <1 such that glz+1)=er"p(z) (26H) Hence etna) = gle +1) =e?" PG (re? ‘This shows that q=« +N, where Ne Z, and that b,, (mod n), Hence G(0) is, in fact, a power series in” have Oifm#0 and we atzy=ermre SF Be (By #0)98, General properties of modular forms for Im z >. Itfollows that g¢ M(I(1), ku, 0%) and this completes the proof of theorem 4.1.3, Further, from (4.1.11, 12, 23) we have =X (ee tN )/m. = Lod. ord(g, 0, M(1))=« +N = where F*is a proper fundamental region for f. Also, when z € Ht, ord(g, z) ord(g, z, (1) = TRG a5 aA Le) © a TOT = £ ordi coheed by (4.1.13, 24, 25). It follows from (4.1.21) that ord(g, P(1)) = ord(f. 1), (4.1.28) We are now in a position to prove Theorem 4.1.4. If fe MUI, k,v) and f #0, then ord(f, 7) = wk/12, where w is the index of fin L(1). Note that the value of ord(f,) is independent of the AF v although it depends on &. Proof. It follows from (4, 1.28) that it is only necessary to prove that ord(f, P(1)) = k/12 for a function fe M(F(1), ky). We prove the theorem by integrating /"(z)/f(z) round the boundary of F,, which we first modify as follows. We replace F, by the compact subset F, defined for sufficiently small e >0 by zzeF,ysl/e|e—plee le to'lzele—ilze), so that all fixed points have been excluded. The boundary of F, then consists (see (2.4.6, 7)) of parts of fu, fv, Ulu, Viv and arcs Ai, Aay As, Aa on Which yel/e, |z-pl=e, |z+el=s, Iz-il=e, 4d Definitions and general theorems 99 respectively. We can take e sufficiently small so that ord(f, ¢)=0 for all { om Ay, As, As, Ae and in F, ~E, ~ (00, p, ~p?, i}. However, ord(f, ¢) may be non-zero at a finite number of points on the parts. of lus fy, Ulu, Viv that are sides of F,. Each such ¢ on ly: or fy has a congruent point on Ul or Viy, and conversely. We include the former and exclude the latter by indenting the boundary along semicircular arcs of radius ¢, and e can be chosen small enough so that none of these semicircles overlap each other or have points in ‘common with more than one side of F,; also we can arrange that no zeros or poles lie on these semicircles. The resulting region we denote by F; (see fig. 7). Its boundary is composed of ares Ay, li, Aas100 General properties of modular forms Tuy Au Wiig As, Ulex which we suppose described in that order, keeping the interior of F. on the left. Then, clearly ord(f, F(1)) = ord(f, 0, M1) +ord(fi, (Y)) + ord(f, e, PY) +16) KAN +h tins tTF), (4.1.29) where K+N=ord(f,0, (1) OS <1), n,=ord(f,i), ms =ord(f,p) and py sete ane) nese | Ge ia =so| 4 . 2ni [, Pee ‘We now compute the contributions to /(F.) from the different sides, We require the following: Lemma, Let ys, 72 be the two smooth curves in C meeting at the origin O at an angle 0, where 0<8 2, and let y(e) be the arc of the circle {r:|r|=e} joining y, 10 y2 (see fig. 8). Suppose that, for O0,keven) (4.2.6) ‘Thus the upper bound 4, (1°) is attained only when g =0. For the remainder of the section we suppose that k is real, that I” is a normal subgroup of (1) and that 4 is a normal subgroup of ” with the property that [7/4 is abelian. We assume that ~J belongs to both A and F. We write h: =[I°: 4] and denote by ¥(I; 4) the group of all characters x on I” that are constant on 4; ie. X(ST)=x(S)x(T) forall, Tel, and x(T)=1 for all Ted (in particular, y(—J)=1). Clearly 2(F, A) has order h and is isomorphic to T/A. Let v be a given MS on I’ of weight k. Then, as y runs through (FA), it is clear that v= 1y runs through a complete set of multiplier systems v on I° of weight k for which v(T)=u(T) whenever Te 4. If M is a subspace of M(A, k, u), and L € (A), we write MIL: ={fILifeM), so that MIL isa subspace of M(L~'AL, k, u*), by theorem 4.1.1(i) In particular, for Le, u' =u, by $3.1, so that MIL is also a 4 See Gunning (1962), $8 42 Dimensions of spaces of modular forms 105 Table7. k even Pe ae es ey fee, PELE fy ttt GEER Set FJGHE roa to 2 2 offfsn ° Fue. For, i fu) 3 2 EO 2d (i]s o yk r 2130 3 4 3ff}tes ° ‘ fo 6 30032 be o k Plivesteeeeelee OntOit stra fey 1 fo 2 6 00 6 4 ke ° 6 00 9 6 ket 1 r © 02 Bom 1 fe) 0036 4 okHT 1 subspace of M(d, k, u). We can therefore always extend M so that MILEM forallLel (4.2.7) We are now in a position to provet Theorem 4.2.2. Let A and Ibe subgroups of finite index in P(1) such that -164, 4 is normal in I, and [/A is abelian of order h. Suppose further that M is a subspace of M(A, k, u), where w is a MS of real weight kon I, and that (4.2.7) holds. Then Mis the direct sum of h subspaces M., where M, © MUI k, 0), v= uy and xe X(0, A) Proof. For any fe M and v = uy, where x € X(T, A), write ly XY amMAN), Hren + Rankin (1967), theorem 2.106 General properties of modular forms where the bar denotes the conjugate complex number, and = 4-R. For Se 4, by theorem 4.1, 1ii) oS, 7), v(ST) oS, Dus) xSTutST) = a(THfIT. by (3.1.14), since y(S) = 1. Hence f. is independent of the choice of the right transversal ® Alsa f, eM since M|T'sM. Moreover, for any SET, a(ST)fIST = (AST AIT pis=j E aenueinis ty an i dao.) = 25) fITS L a(TsyiTs =v(SYor since 4 ='S = 4 «RS, Hence fe MUL, k, v). For each y ¢ XI; A) and v = uy, write M = (ffeM}. Then clearly M, isasubspace of M and M(I, k, v). Further, since f-¥ho where the summation is over all v = uy for ye X(T, 4), and since only the zero form can belong to different subspaces M,, it follows that M is the direct sum of the A subspaces M,. This completes the proof of the theorem. We note that the theorem does not apply to every MS u of weight k on A, but only to a MS w that can be extended to a MS on the bigger group I”. Corollary 4.2.2. If, in theorem 4.2.2, M&{A, k, u}, then dimM= J dimM,. 42 Dimensions of spaces of modular forms 107 In particular ir, 4, u)= S| dill kyu} and dim{d, k,ubo= E. dim{ k, uxt. For, if M={4, k, u} and fe {I k, v} for some v = uy, then feM and, since fe=f, (Mk, O}SMEI, k, v}. Thus M,={0 ko}; similarly for {7 k, obo. ‘We can apply theorem 4.2.2 and its corollary to "= (1), 4 = I°(1), We deduce that dim(P(), ku} = E dim(P(), k, uy") (4.2.8) where the character x is defined by (3.4.5, 10), A similar result holds for the corresponding spaces of cusp forms. Applications can also be made with f'= I”? and”? and A =I"? and I”?, respectively. We give more precise results about dimensions of certain spaces in chapter 6. ‘The proof of theorem 4.2.2 can also be applied to give Theorem 4.2.3, Let v be a MS of real weight k on a subgroup T of T(), and let be a subgroup of index h in I. For any fe M(A, k, 0), write 1 ha (DAD, (4.2.9) where [=A 2. Then f. is independent of the transversal R and Ae MIL, k, 0). If we write : TH, 4, vif: (4.2.10) then Tr(4, 0) is a linear operator from M(d,k,v) into M(I,k,v) and is called a trace operator; see Petersson (19675, 1973). In chapter 8 we shall make use of operators of this kind,108 General properties of modular forms 4.3. Relations between modular forms. We have seen in the preceding section that M(F,k,) is a vector space. We now investigate multiplication and division of modular forms. Let Ibe ‘a subgroup of I) of finite index and let vs and v2 be multiplier systems on "of weights k, and ka, respectively. Then itis clear that byvs and v,/v2 are multiplier systems on I” of weights k, +k; and k,—ks, respectively. We note also that v; is a MS on I” of weight 2m +k for every integer m; for, since w(S, 7) is an integer for all 5, Te [it follows from (3.2.1) that o(S, 7) is unaltered when k is replaced by k-+2m, while (3.1.13) shows that v(-U) =e" 7". Let fre M(F, ki, v1) and f2€ MU, ka, v2). Then, clearly, fifeeM(D kytks, 002) and fi/ fae MU. kik, 0/025 for all three conditions (I)-(II1) of $4.1 are easily verified. 'A modular form of weight zero is called a modular function. ‘Suppose that fe MUI, 0, v). Then the MS v is a character of the group I and so takes the constant value 1 on the commutator subgroup I”. Since, for all Te, f(Tz)=(Dflz) (ze 6, it follows from (2.1.1) that f(T) = 0(T)\T:2¥ 2) and also Since conditions (1)-(III) are easily verified, we have proved the first part of the following theorem. Theorem 4.3.1. (i) If fe M(F0, v) then fe M(E, 2, v) and f'/ fe MU, 2,1). (ii) Suppose that v is an MS of dimension zero and that, for each TeT, o(7) is a root of unity.t Suppose also that f #0 and that fe{F,0, v}. Then f is a non-zero constant, and v(T)=1 for all Tel. Git) If ge MUD, kv), where k #0, then his kgg"—(k + Die Pe MUL, 2k +4, 07). 1 This holds in particular for P= /'(1), P? and F°, since their commutator groups bhave finite index. 43 Relations between modular forms 109 Also, if #0, then h#0. Further, if ge(l,k,v}, then he {2k +4, 0p, Proof. (ii) I fe{F,0, 0}, f#0 and f is a constant, the relation f(Tz)= v(T)flz) shows that »(T) = 1 for all T 0. It follows from theorem 4.1.4 that the corresponding sum of £¢F for which ord(f, ¢,1')<0 is “vn. Also V(/, 1) isa positive integer, when fis nota constant, Foriff isnot a constant and V(f, [) =0, then ord, ¢, 7) = Ofor all eH. Hence, if g(2)=f(z)- fli), g MUI, 0, 1) and ord, £1) = 0 for all £eH’. By theorem 4.1.4, ord(g, £1) =0 for all Ce. This isa contradiction, since ord(g, i, P)>0.110 General properties of modular forms For any neZ, fe MI, 0, 1M f #0) and it follows that every rational function of fis alsoin M(, 0, 1). Thus M(I, 0, 1)isa field, and this enables us to discuss algebraic as well as linear dependence of modular functions. ‘Theorern 4.3.2. Let, fx€ M(I, 0, 1), neither being constant, and let their valences be q, and qz, respectively, ‘Then there exists an imeducible polynomial P(x,, x2) of degrees at most qs and q, in x, ‘and x, respectively, and with complex coefficients not all of which are zero, such that ®(f,, f.) = 0 identically. Proof. We note that the polynomial cannot have degree zero in either variable, or else f, or f» would reduce to a constant. ‘Take any positive integers m, and m, satisfying (my + 1)(m, + 1) = mqy + maga +2. (43.2) ‘The general polynomial P(x,, x2) of degrees m; and m,inx, and x, contains (m,+1)(m,+1) coefficients. For all choices of these coefficients (/,, f:)€ MU, 0, 1). By considering the poles of & we see that either V(b, TS mq t mde (4.3.3) or ©(f;, f2) is identically zero. Now we can choose the coefficients, not all zero, so that P(f,,f:) has zeros at (m,+1)0m2+ D1 assigned incongruent points of H~E at which f, and f_ are holo- morphic. Hence, if is not identically zero, 0 = ord(®, 2) = (m; + 1)m, + 1)— 1 VBP) > 0, which is a contradiction, by (4.3.2, 3). Thus (ffx course, possible that & may be reducible, but we can then take an irreducible factor with the same property. For each choice of m,, m; satisfying (4.3.2) we get an irreducible polynomial & with ®(/,, f,)identically zero. The inequality (4.3.2) shows that we cannot have m, 1. But oe M(I", 0, 1) and 1= Vw, A)=pV(o.T), which is a contradiction. ‘The last part of the theorem follows since, for Te I T:z)7w'(Tz) w@Rifo2)) = La(Te)= Note that itis always possible to choose the bilinear mappings so that they have determinant unity, but this may be at the expense of some simplicity. We given an illustration of the theorem in §7.2 for the case 4 =1(2) and = (1). ‘As an illustration of the kind of functional relationship that can hold between two different modular functions belonging to the same group we consider what are called transformations of order n, where n is a positive integer. Write (? b ole al where a, b, c, de Z and det T=ad—be =n. Then T is called matrix of order n, and the corresponding bilinear mapping aztb_ etd is called a transformation of order n. When h is an integer dividing a,b,c and d we may put au Baie tay A T=hT, =a" 4 Since the bilinear transformation associated with T; is the same as the mapping associated with T, we confirm our attention in this 43 Relations between modular forms 13 section to primitive matrices of order n, i.e. matrices T with det T=n, (a,b,c,d)=1. ‘The corresponding transformations are called primitive transfor- mations of order n. We write ab {r-[0 Fh det rans abed= tra,bedez} (4.3.6) for the set of all primitive matrices of order n, We also write (Y ={T: det T>0; a, b,c, d rational} (43.7) a ‘Then 2" is a group of matrices containing (1) as a subgrouy and QF asa subset. 2" is therefore a union of double cosets P(1) 71) (Te"), We are only interested in double cosets of this form for Te AS, and, in particular, in the double coset I(1)J,1"(1), where 10 j= [fh Oo] mez (438) Theorem 4.3.6. For any ne", OL= POSE =P) > T (4.3.9) The right transversal T, isa finite setof cardinality yn), and may be taken to be the set Ti=(TeMt:c =0,a>0, ad =n, b modulo d}. (43.10) Here ‘bh modulo d’ means that b runs through any complete set of residues modulo d satisfying the condition (a, 6, d) = 1. Similar results hold for left ransversals of 21% with respect tol (N), the condition ‘b modulo d’ in (4.3.10) being replaced by ‘b modulo a ‘ Proof. We call two matrices Ly and L; in 12% left-equivalent if L, = SL, forsome $¢I(1). This isan equivalence relation, and our object is to partition 2? into corresponding equivalence classes TU)L. Clearly Ls is primitive when Ly is. Right-equivalence is defined similarly.114 General properties of modular forms ILE N4, then PLE) E23, 0 that 2 is a union of double cosets in {2° and therefore a union of right cosets, from which the existence of a right transversal T, for which 2 = I"(1) - T, follows. ‘We now show that T, may be taken in the form (4.3.10). Suppose that AB ae L ie Bear and write (A.C), A=aa, C=ya, so that (a, y) = 1; we can therefore find a matrix $ € (1) with these values of a and y in its first column. Then “e-[2 4 stl Gf for some integers b and d. Note that a is positive and that itis the highest common factor of the elements in the first column of the original matrix L. Further, two matrices are left-equivalent if and only if a:=az, d,=d, and b= +, (mod d,), For every matrix left-equivalent to T; is of the form [em ab, +d, ST lve bit 6h for some $11). If T, = ST, we must have (since a; and az are positive) y=0, a=8=1, 6: =b,+Bds. Conversely, if these relations hold, then T; = U*T; and so T; and Ty are left-equivalent. “This shows that the right transversal 7, may be chosen to be the set Tt of (4.3.10). Clearly T is a finite set and all other right transversals must have the same number of elements. The canoni- cal form of the members of T? given in (4.3.10) is known as Hermite’s normal form; see MacDufee (1946), theorem 22.1. 43 Relations between modular forms 115 To prove that |T¢|=y(n) we put h =(b, d), so that IT=E E eld/h), where ¢(d/h) is Euler's function defined in §1.4, and where & and d are positive. This shows that [Thel=|TSI|T3] when (m,n so that it suffices to take m =p’, where p is prime and re 2", and prove that |T#]=W(p') =p! +p". Now there are p’ and @(p"") matrices of the types or) [5 pe] ose ( pl and |) pre] (ssn. respectively, and itis easily verified that +h or = pt ‘We shall see later on that it is not accidental that |, | is equal to the index of a(n) in (1); in §9.1, where we consider a rather more general situation, we shall give a group-theoretic, rather than 2 number-theoretic, proof of this fact. We now prove that 2%=I(1)J,'(1). This is equivalent to reducing the members of (2+ to Smith’s normal form; see Mac- Duffee (1946), theorem 26.2. We have already shown that any matrix L € is left-equivalent to one of the form (“0 5 0 D, Similarly, L is right-equivalent to a matrix of the form (*. B) 0 ] CG Ds Further, (A, C)s|A] and (A, B) [A]. This shows that, after a finite number of steps, as a result of alternate left and right ication by elements of F(1), L can be transformed into a u-[e be) matrix116 General properties of modular forms in which A* divides both B* and C*. Left and right multiplication by suitable powers of W and U then reduce L* to the form alo elo od ‘We now apply the process again to L'W, which is left-equivalent to a matrix with first entry (A’, D'). The process clearly terminates when we reach a matrix of the form @ a) Od in which a =(a, d), and this happens after a finite number of steps. le d, we may take a = 1, since the matrix is primitive. tis clear that simi by 1 results for right-equivalence can be proved r methods, This completes the proof of the theorem. We now take Ito be any subgroup of (1) of finite index and put M)=P-R, (4.3.11) so that, by (4.3.9), Mt=F)L LM) =P A- Ty (4.3.12) This double coset is a union of double cosets PLT, where Le 2%. We may therefore put, for any Le Q%, TLP=0-T), where TST, (4.3.13) and we also write Ta=LILAr (Lea). (4.3.14) T;, is called a transformation group of order n. Theorem 4.3.7. Let L€0% and let P’be a subgroup of I) of finite index y. Then, foreach T ¢ Ti, the group I'ris a conjugate subgroup of, in I. Conversely, each subgroup of T conjugate 10 Tis of the form I; for some Te T:. Moreover, wl): =(P TJS nv(n). (4.3.15) In particular, when I'=I'(1), I, is conjugate fo To(n) for each LEQ} and so w.(F(1)) = (an). 43 Relations between modular forms 117 Let Fe MU, 0, 1) and define Fon H by Fr(z)=F(Tz) (Te TL). (43.16) Then Er € MU 0, 1) and there exists an irreducible polynomial ®, such that (F,Fy)=0. forall eT: Proof. If Te TS, then T=Si'LS, for some S,, S2in I Hence, by (4.3.14), Py=S3L“TLS, NE =S3'T,S,. Conversely, if $:€ I then LS, = $,T for some $,¢ and Te Ts, and so $3'TS.=Ty. If 7=F(1), the double coset [LI is just Q% and, for ST), sfe ale) SIs -[¢ 3 43.17) It follows that [,=T%n) and that the different transformation groups Fy, for Te @%, are just the conjugates of Fn) (and therefore of T()). Accordingly, mi (T(1) = o(n) ‘We now return to the general case and prove (4.3.15). We use the fact that, if G, and G, are subgroups of finite index in a group G then [6,:G.NG]=1G:G.]; see Hall (1959), theorem 1.5.5. Write F,(1): = L-“W()LOF(). Then (POF Sr) T= a) and ' (PAR) TIS NL POL PAL POLL Ty S[LTU)L: LVL) =(F): 11> w. Hence w= Js adn).118 General properties of modular forms Now take any S¢ Ty and write $*= TST", so that S*e 1” and therefore F,(Sz) = F(TSz) = F(S*Tz) = F(z) = Fal) Thus FreM'U'n 0, 1), since Fr is a meromorphic function. ‘We must therefore consider the behaviour of Fr(z) at the csp $10, where S,€(1). Choose 52¢1(1) so that TS:c0 = $00 an ia AB sprs.-t=[4 5]. Then, since Loo = 0, C=O and F(S:2), which is F,,{(Az +B)/D}, has an expansion of the required form at z=00. Hence freemen of the irreducible polynomial , for which O(F, Fi) =0 follows from theorem. 4.3.2. For any Te TE we take Sy and S; as above and find that UF, Fe |S = P.(FISs, FilS) LF, Far) = OLB. Fr) ‘The irreducible equation &;(F, F:)=0 is called a modular (or transformation) equation. ; ee Tr ong theorem enables us to obtain further information about the polynomial 4. AF, Fis.) “Theorem 4.3.8. Under the same assumptions asin theorert 4.3.7 we have, for L€0% and TeILT, VE Pr) = VF Pr) = wD VE 2 Ie follows thas the irreducible polynomial ®, has degree not exceed ing 4, (I) VCE, F) in each of its two variables. (4.3.18) i finite set of Proof, Choose @ finite constant C differing from the of ee of F(z) and F,(z) at points z EU, and put Ge)~ F(z)~C. It follows that, when computing the valences viG.D), 43 Relations between modular forms 119 V(G, Ix) and V(Gr, 'z), we need only add up the orders of G or Gr at its zeros in a fundamental region, We deduce at once that ViGTA=mNVG.M and VIG) =VGrTr), since there is a one-to-one correspondence between zeros of G and Gr ina fundamental region for fy. For if Tz, = Tz2 (mofi Is) then, for some Sef, Tz:=STz, and so 2,=7-'ST23, But TST €[y and therefore 2, =z: (mod I). Equation (4.3.18) follows since F and G have the same valences for I’ and for I'y. The last part follows immediately from theorem 43.2. We now consider the effect of transformations of order on modular forms. For this purpose we need to extend the notation T: 2, first defined in (2.2.1), from T¢ @ toany non-singular matrix ab rl. Hi with real entries, As before, we put T:z=cz+d (4.3.19) and we easily verify that (2.2.2) continues to hold; ie ST:2=(S:Tz\(T:2) when $ and T are non-singular and z #00, z # T-'00, Now let I” be a subgroup of 1) of finite index, let v be a MS of weight k eR on I and suppose that fe M(I, k, v). We extend the definition of the stroke operator f|L, given in §4.1, by putting f.lz)= fla) = (der L)* tu (L, 2) fz) (4.3.21) for any non-singular matrix L with real entries, where, as usual, w(L, 2) = (Liz) When L€I(1) this agrees with the previous notation. Observe that, if L = hL.y, where h >0, then fiL = lL. Accordingly, when L is a matrix of order neZ* we may, without loss of generality, assume that L is primitive. Note also that (4.3.21) agrees with (4.3.16) in the particular case when F=f, k =O and v = 1 Now suppose that Le2!, where neZ". Then we can show, exactly as in §3.1, that an AF y and MS v! may be defined as in (4.3.20)120 General properties of modular forms (3.1.17, 20) not onty on 7, but on the larger group Tt=LTLnri). (4.3.22) In fact, they can be defined on L~'I'L, but this group need not be contained in (1). For this purpose we can apply §3.1 directly, with L replaced by Ly: =n“'L, since then L602. Moreover, for non-singular matrices S and T with real entries, we may define o(S, 7) as in (3.1.15) and deduce, with the help of (4.3.20) that |o(S, T)|= 1; w(S, T) may also be defined as in §3.2 and has similar properties. ‘We now prove an extension of theorem 4.1.1 Theorem 4,3.9. Suppose that fe MUL, k, v) and that L, L, and Lz belong to: 2*(n), 2 *(n,) and 2*(n2), respectively, where n, ry and ny, are positive integers. Then we have: () feMUt, kv"), Abba =o (Ly LAL iLs Gi) PTET, f= o(T, Lyv( Tf, and fn = ef. (iv) ful fe MU;.0, v"/v). (v) If fis a holomorphic, entire or cusp form, so is fr. Proof. (i) Any member of Fz can be expressed as L~'TL where Te. Then, as in the proof of theorem 4.1.1(i), fALTTL2) = nu (L, LL 2)" f(TL2) =nl{u(L, L1TL2)} ‘oT, Lz)f(Lz) =v(LTL, 2)filz) so that f, €M'(t, k, v*). Parts (ii) and (iii) are proved by arguments similar in theorem 4.1.1. Note that LiL2 need not be pri and L; are. ‘We now have to examine the behaviour of f, at cusps. Take any se1(1) and consider the cusp Sa. Then L$0o is a cusp and so LSe = $,0 for some $,€I(1). Put T= S7'LS so that Te Q$ and those used ive when Ly 43 Relations between modular forms 121 Too =00, Accordingly ¢ AGS ={o(L, SP fislz) (uw (LS, z)o(L, S)}'f(LS2) 2S Te) ¢ ~ wlSy, Te)o(L, SY" n¥o(S:,T) aE et =(2)"2 Dy (43 + 6) a! o(L,s)"" From this itis clear that f, has an expansion of the required format ‘ro, and therefore fe M(t, k, v*). Part (v) now follows and part (iv) is obvious. 7 In theorem 4.3.7, 8 we have been concerned with a single double coset ILI and this gave rise to a subset T! of AT,; see (4.3.13). Each double coset LI’ in 12 determines a different subset of AT and these subsets are disjoint. There therefore exists a finite set $ of elements of 2% such that AT, =U TE (disjoint union), (4.3.23) 0, so that y(T, z)=d* and therefore (Tz) and we may clearly iake SCAT, if we wish. The number of elements of S is just the number of different double cosets [LT in 4, For each LES we derive an irreducible polynomial satisfying the conditions of theorems 4.3.7, 8. These polynomials will, in general, be different. ‘We have seen that the transformation groups I’, have finite index in F. We observe also that, when I'is a congruence group, s0 is I, For, if [(q)SI for some qeZ* then, for any Let, LP (ng)L € 1(q). It follows that Png) ST; Finally, take I" = (1) and suppose that Fe M(I‘(1), 0, 1). Then, as T runs through T,, we obtain (n) functions Fr, each satisfying the same modular equation OF, Fy) = ice F(n) < Fa(nt), we note that Fy € M(t), 0, 1) for all Te 22, We shall consider explicit examples of this theory in §§6.5 and 7.2,122 General properties of modular forms 4.4, Modular forms of weight 2. We know from theorem 4.3.1) that the derivative of a modular function is a modular form of weight 2 but there may exist modular forms of weight 2 that are not derivatives of modular functions. In this section we prove a theorem about the residues of modular forms of weight 2 and restrict our attention (o constant multiplier systems. For any function f meromorphic at a point fC we write res(f,¢) for the residue of f at ¢; res(f,¢) is the coefficient of @ ‘in the Laurent series expansion of f(z) about the point. Now suppose that fe M(F, 2, 1), where is a subgroup of (1) of finite index and let £ eH, We define the residue of fat (mod I") tobe £0 rest f.é.0°) EP. (4.4.1) where I is the stabilizer of £ (mod ["); this definition is analogous to that of ord(f, ¢1°) in (4.1.13). Now suppose that {= Loo, where L €/°(1); since the MS v is constant, «, = 0 and we have, by (4.1.6, 8, 9, 10), filz)= E an(fider™s where ny =n) (4.4.2) and where we have written aq(fz) in place of aa(L). We define rest fig DP): =nuaolfe)- (4.4.3) Observe that, by (4.1.5), res(f, g,/") depends only on £ and not on the choice of L. Theorem 4.4.1. Suppose thatf M(I,2, 1) and that L € (1). Then, forall €€| res(f, LEP) =restfis 6s LTE). (4.4.4) In particular, if ¢,=¢,(mod I), then res(f Lin) =re5(f fo. P- (4.4.5) Proof. Let I be any piecewise differentiable arcin H with endpoints Zz, and 23; in applications { will be a circular arc or a straight line 44 Modular forms of weight2 123 segment, Put w =Lz for z€/ Then, as z goes from z; to z, along L.w goes from Lz, to Lz» along Ll, Since ‘ aw fowy= (Liz fle), Fra (be2)%, (4.4.6) it follows that [ronaw=[neerds, (4.4.7) when fisholomorphic on /, In particular, we deduce from Cauchy's theorem that res(f, LE) = rest fas 6) (4.4.8) for any £€H. Now, by (2.2.22), rer Lf De so that f, and (L“'FL), have the same order. From this and (4.4.1, 8) we obtain (4.4.4) in the case when {€H. i Hee 00, for SET (1), we have, by (2.2.25) and theorem i), res(f, Lg, P) = n(LS00, Mao( fis) = n(S00, LL )ag( fy |S) = res(fis LL). This completes the proof of the theorem. We are now in a position to state and prove Theorem 4.4.2. Letf € M(I, 2, 1) and let F be a, idamental region for f. Then eeecaree L resto =0, (4.4.9) Proof. We note that the summation on the left of (4.4.9) is over @ finite number of points¢ at which f has singularities. By (4.4.5), the sum is independent of the choice of F. We assume, in the first place, that I’= (1). In this case the theorem is easy to prove; we have merely to adapt the proof of124 General properties of modular forms theorem 4.1.4 in an obvious way. We consider the integral sy fferae 4.4.10) taken around the boundary af, of a compact subset F, of a fundamental region, Here Fis chosen, asin theorem 4.1.4, so that the fixed points ©, p, ~p? and fare excluded by the straight line segment 4, and the indentations Aa, Aa, As Other indentations are made round poles of f that would otherwise be on the contour; see fig. 7. Then, for sufficiently small ¢ >0, the integral (4.4.10) is equal to Li reslf )= Er real, 7), where the dash denotes that the summation is over all points { of Fy with the exception of 00, p, —p” and i. ‘On the other hand, it follows from (4.4.7) and theorem 2.4.4(iv) that the integrals over the sides ly, li, Iv, contribute zero. From the remaining arcs Aj, A2, Ay and A, we obtain the contributions af, f(z) dz ==res(f,0, (1), 410 AL nerae+] trae} =-retto,r0, é : (4.4.12) and i fz) dz =—res(f, i, M1). (4.4.13) Since (4.4.11) is obvious, it suffices to prove (4.4.12); the proof of (4.4.13) is similar, but simpler. By (4.4.7), [fede= [fae so that the left-hand side of (4.4.12) is expressible as yf peas, ai where y consists of the arc U~'A, followed by Az. Let z, be the initial point of 7; then its final point is Pz). Further the arcs y, Py 44 Modular forms of weight 2 125 and Py form a simple closed curve y* encircling @ once in a clockwise direction. Hence, by (4.4.7), —lres(f.p) rest/,p,T(1)), so that (4.4.12) holds. ‘Combining all these results, we obtain E rest, 7) =0, so that the theorem is true for (1). We now return to the general case and suppose that fe MUF, 2, 1) and write Payal ae. Define FU Efile) (4.4.14) Clearly F(z) is independent of the choice of the right transversal %, Take any §€1(1). By theorem 1.1.1, 25 isalso. right transver- sal for so that 2S = % fiste)= Fez). It follows that Fe M(M(1), 2, 1), and so Y restF, PU) =0. (4.4.15) Now take any z €24 and suppose that § generates the stabilizer fy of ¢ (mod I). We apply theorem 1.1.2 with P=f(1) and y= [We may take By (4.4.8), tes(fixsts C= rest f, Lio)s so that, by (4.4.14), restF, = os tesif, Le,126 General properties of modular forms where otic fidh@:fie) (sism), by (1.1.14). On division by |/%(1)] we find that res(F, £, P= res(f, Lg, (4.4.16) We obtain a similar result when {= 0. Take $= U in theorem 1.1.2 and note that, by (2.2.25), a= n(L0,P) so that res(F, 0,10) = aolF) = F oral fi.) = 2 restf, Ld I), It follows that (4.4.16) holds for each {€F,; the number m depends, of course, on the choice of ¢. Now each proper funda- mental region F for f contains exactly m incongruent points Li (1Sism) belonging to the orbit {¢]. From this and (4.4.15), we obtain (4.4.9). This completes the proof of theorem 4.4.2. The theorem could have been proved in the general case by an extension of the method used for the case when "= (1). In principle this extension is straightforward, but a careful treatment of all the details is somewhat troublesome. It may be noted that the theorem is a consequence of a general theorem that asserts that the sum of the residues of a meromorphic differential (in this case f(z) dz) on a Riemann surface is zero; see, for example, Springer (1957), theorem 6-10. We conclude by considering briefly the ‘abelian integral’ * fu) du (ze), (4.4.17) where fe H(L, 2,1) and z is a given point of H. Since f is ‘on H, F(z) is independent of the path from zo to z. (7) = mT, Zo): = F(T20). (4.4.18) 44 Modular forms of weight 2 127 We call 1(T) a period of f. Then, for Te fy Fite)=[" foo due [”” fd du =F(z)+n(T) (4.4.19) for all z €H, by (4.4.7, 18). Tt follows at once from (4.4.19) that a(ST)=m(S)+m(T) (S, TEL), (4.4.20) so that the map 1:/°-> C is a homomorphism of f* into the additive group of complex numbers, In particular, we deduce from (4.4.20) that a(T)=0 forall Tel. We also deduce immediatly from (4.4.20) that #(7) = 0 for every elliptic transformation Te Now suppose that T is a parabolic member of f* with fixed point £=Le (LeS(1)). The stabilizer of ¢ (mod I’) is generated by S=LU'L", (4.4.21) where 1 =n, =n(Z, 1"). Hence T=S" forsome meZ. (4.4.22) We show that a(T) =mm(S) =m reslf, 6, P). (4.4.23) For, by (4.4.18, 7, 21), m1s)= |" fu du -| flu) dee =f he du =nasit. and (4.4.23) follows from this and (4.4.3, 22). It is possible to evaluate 7(S) for a hyperbolic transforma- tion S in terms of the constant term in the expansion of (2—E))'z-G)f (2) in powers‘of an appropriate variable; here ¢, and £; are the fixed points of S. See Hecke (1925), pp. 214-15 We note, in conclusion that, if fe{I,2, 1}o, it follows. that 1(T) =0 for every elliptic and parabolic transformation Te ft f128 General properties of modular forms is generated by elliptic and parabolic transformations, we can conclude that f must be the zero form, For then, by (4.4.19), F(Iz)=F(z) (Tef, zen) and it is easily shown that F satisfies condition (II) of $4.1, since / is a cusp form. Further, Fe M(V,0,1) and so, by Theorem 4.3.1(ii), F is a constant, It follows that f= 0. 4.5. The order of magnitude of entire forms and of their Fourier coefficients. In this section we shall consider a given entire form f, and the constants C, C;, etc. that we introduce will depend on f but not on the value of zeH considered, We write y=Imz throughout. Theorem 4.5.1. Suppose that fe{I,k, v}, where k>0. Then, for some positive constant C and all z €4, Velsc Gel), If@lscr* O0) (4.5.2) Proof. As pointed out in §4.1, there are essentially only A different transforms fr of f. It follows from (4.1.11) that there exists a constant C such that [fates Center sree (4.5.3) for all Te J (1), and all £eH for which q=Imf=l. Here ker-+Np=ord(f, Too, 1), as usual, and my is the width of the cusp Too, Let 6 be the minimum of the A different quantities In(xr+Nz)/my so that 6 =0, and 6 >0if fe{l k, vo If y=}, it follows that |f(z)|0 and so n'* e“* is bounded for all 7 >0. 45 The order of magnitude of entire forms 129 Theorem 4.5.2. Let f EF; k, vo (k>0), so that payee Faget (Ne >0) for all 2 €¥. Then, for large m, O(n), Gi) E a, =OGn"* tog m), () a Cid § JaP=Olm), ivy § Iayl =m!) Similar estimates hold for the Fourier coefficients of the transformed functions f Proof. Suppose that m =2 and put z=x-+i/m. Then, by uniform convergence, [revere a, ay and so, by theorem 4.5.1, lal Cob e200 Cyn, which proves (i). Also [revere dx Em 2dx en [ime 2” and observe that eP = -ae =(1#a)? = 14(6), Put 2mx =n0, a= F148) + (1 a)? cos*($A) Hence