Springer Monographs in Mathematics
Toshitsune Miyake
Modular Forms With 11 Figures
Springer
Toshitsune Miyake Hokkaido University Department of Mathematics 0600810 Sapporo, Japan emflz7;
[email protected].jp Translator Yoshitaka Maeda Hokkaido University Department of Mathematics 0600810 Sapporo, Japan emflz7:
[email protected] The first English edition was published as a monograph in 1989 by SpringerVerlag (ISBN 3540502688). Title of the original Japanese edition: "Hokei keishiki to seisuron" © Kinokuniya Company Ltd., Tokyo, 1976.
Library of Congress Control Number: 2005936235
Mathematics Subject Classification (2000): 11F03,11F06, iiFii, 11F12,11F27,11F30,11F66, 11G05,11M06,11R42,11R52,11R56 ISSN 14397382 ISBN10 3540295925 Springer Berlin Heidelberg New York ISBN13 9783540295921 Springer Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9,1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable for prosecution under the German Copyright Law. Springer is a part of Springer Science+Business Media springeronline.com © SpringerVerlag Berlin Heidelberg 1989, 2006 Printed in Germany The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Typesetting: Macmillan India Ltd., Bangalore Production: LETgK Jelonek, Schmidt & Vockler GbR, Leipzig Cover design: Erich Kirchner, Heidelberg Printed on acidfree paper
46/3142YL  5 4 3 2 1 0
Preface
Modular forms play an essential role in Number Theory. Furthermore the importance of modular forms has continued to grow in many areas of mathematics including the infinite dimensional representation theory of Lie groups and finite group theory. The aim of this book is to introduce some basic theory of modular forms of one variable. Originally this book was written in Japanese under the title "Automorphic forms and Number Theory" by Koji Doi and myself and pubHshed by Kinokuniya, Tokyo, in 1976. When the English translation was planned, the first named author proposed that only the chapters written mainly by me be translated together with some additional material and pubHshed under my sole authorship. In Chapters 1 and 2, the general theory of Fuchsian groups, automorphic forms and Hecke algebras is discussed. In Chapter 3,1 summarize some basic results on Dirichlet series which are necessary in the succeeding chapters. In Chapter 4, the classical theories of modular groups and modular forms are studied. Here the usefulness of Hecke operators as well as the remarkable relation between modular forms and Dirichlet series obtained by Hecke and Weil have been emphasized. Chapter 5 briefly reviews quaternion algebras and their unit groups, which are also Fuchsian groups and which play a role similar to that of modular groups in their application to number theory. Chapter 6 is devoted to the trace formulae of Hecke operators by Eichler and Selberg. The formulae have been generalized by many people including H. Shimizu, H. Hijikata and H. Saito. A formula computable by them is also offered. In our Japanese edition, as an introduction to the automorphic forms of several variables. Chapter 7 deals with Eisenstein series of Hilbert modular groups and the appHcation to values of zetafunctions (following Siegel). As a result of important series of recent work by Shimura on Eisenstein series, I decided to rewrite it to introduce some of his results on Eisenstein series restricting it to only the case of one variable. I should like to express my deepest gratitude to Professor Goro Shimura, who constructed the essential part of the arithmetic theory of automorphic functions, for his valuable suggestions and encouragement. The translation of Chapters 1 through 6 was prepared by my colleague Professor Yoshitaka Maeda. He also corrected mistakes in the original text, and gave me many appropriate suggestions. I express my deep and sincere thanks to him for his collaboration. I also express my hearty thanks to Professor Haruzo Hida
VI
Preface
whose lectures at Hokkaido University during 198384 were very helpful for the preparation of the present volume, and to Professor Hiroshi Saito and Dr. Masaru Ueda who kindly read the manuscript very carefully as a whole or in part and made many valuable suggestions. Sapporo, February 1989
Toshitsune Miyake
Contents
Notation and Terminology Chapter 1. The Upper Half Plane and Fuchsian Groups §1.1. §1.2. §1.3. §1.4. §1.5. §1.6. §1.7. §1.8. §1.9.
The Group of Automorphisms of the Upper Half Plane Actions of Groups Classification of Linear Fractional Transformations The Invariant Metric and Measure on H Fuchsian Groups Fundamental Domains Quotient Spaces r \ H * The Structure of r \ H * as a Riemann Surface Fuchsian Groups of the First Kind
IX 1 1 4 7 10 17 20 24 28 31
Chapter 2. Automorphic Forms
37
§2.1. §2.2. §2.3. §2.4. §2.5. §2.6. §2.7. §2.8.
37 45 48 53 57 61 69 74
Automorphic Forms Differentials on Compact Riemann Surfaces Automorphic Forms and Differentials The Measure of r \ H * Dimensions of ^^(r) and ^^iH Poincare Series and Eisenstein Series Hecke Algebras Hecke Operators on the Space of Automorphic Forms
Chapter 3. LFunctions
79
§3.1. Dirichlet Characters. §3.2. The Riemann ZetaFunction §3.3. Hecke LFunctions
79 84 90
Chapter 4. Modular Groups and Modular Forms
96
§4.1. SL2{Z) §4.2. Congruence Modular Groups §4.3. Modular Forms and Dirichlet Series
96 103 114
VIII
§4.4. §4.5. §4.6. §4.7. §4.8. §4.9.
Contents
A(z) and r](z) Hecke Algebras of Modular Groups Primitive Forms Dirichlet LFunctions and Modular Forms LFunctions of Quadratic Fields and Cusp Forms Theta Functions
129 131 153 175 182 185
Chapter 5. Unit Groups of Quaternion Algebras
195
§5.1. Algebras over Q and Adelization §5.2. Quaternion Algebras §5.3. Hecke Algebras of Unit Groups of Quaternion Algebras
195 198 210
Chapter 6. Traces of Hecke Operators
219
§6.1. §6.2. §6.3. §6.4. §6.5. §6.6. §6.7. §6.8.
219 225 228 231 243 248 256 259
Spaces of Functions on H The Projection of Lf (H) onto Hf (H) Function Spaces Consisting of Automorphic Forms Traces of Hecke Operators (Calculation of Integrals) Traces of Hecke Operators (Algebraic Calculation) Local Conjugacy Classes Class Numbers of Orders of Q [ a ] An Explicit Formula for ir(T(n))
Chapter 7. Eisenstein Series
268
§7.1. Eisenstein Series of Weight k^3 §7.2. Analytic Continuation of Eisenstein Series
268 274
Numerical Tables
295
References
315
List of Symbols
317
Subject Index
333
Notation and Terminology
1. We denote by / , Q, U and C, the ring of rational integers, the rational number field, the real number field and the complex number field, respectively. For a rational prime p, Z^ and Qp denote the ring of padic integers and the field of padic numbers, respectively. We also denote by U + , !R_ and C \ the set of positive real numbers, the set of negative real numbers and the set of complex numbers with absolute value 1, respectively: R+ = {xelR
I x > 0 } , [R_ = {xeR
 x < 0 } , C^ = {zeC

z = 1}.
2. For a complex number z, we denote by Re(z) and Im(z), the real part and the imaginary part of z, respectively. When z is a nonzero complex number, we denote by arg(z) the argument of z, which we specify by — TT < arg(z) ^ n. For a real number x, we denote by [x] the largest integer not exceeding x. When x is a nonzero real number, sgn(x) denotes + 1 or — 1 according as x > 0 or x < 0. 3. For a ring R with unity 1, we denote by R"" the group of invertible elements in R. Further we write M„{R) = the set of square matrices of degree n over R, GL„{R) = {oieM„(R)
\ det(a)eR^},
SL„(R) = {aeM„{R)
\ det(a) = 1}.
4. We denote by II the disjoint union of sets. For a finite set A,\A\ denotes the number of elements in A. We also denote by # { }, the number of the elements of the set given by { }. 5. When ^ i , . . . , g^ are elements of a group G, Q w+1/ lwp
(weK),
p gives an analytic isomorphism of H onto K.
D
We are interested in functions on H which satisfy certain transformation equations for automorphisms of H. (We say that they have automorphy) We first study automorphisms of H. We denote by Aut(H) and Aut(K) the groups of all (complex analytic) automorphisms of H and K, respectively. If a =
e GL2(U\
§1.1. The Group of Automorphisms of the Upper Half Plane
3
and z e H, then /I 1 ax
T / .
(1.1.7)
i„(„.)=___^.
^et(a) Im(z)
In particular, if det(a)>0, then we have lm(az)>0, and therefore, "zi—•az" induces an automorphism of H. We put
GmU) = {(xeGL2{U)\ det(a)>0}, and denote by I{OL) (aeGL2 (IR)) the automorphism "zi^az" of H. Then it follows from (1.1.4) that this mapping i: GLj(IR)9ai)>i(a)eAut(H) is a grouphomomorphism. Now we put (_ —sm ^ COS0JI We identify aeW" with
^
J
eGL^(IR). If for a = r
eGL2([R), Ka) is the
identity of H, then a belongs to R"", since cz^ + (d — a)z —b = 0 for any zeH. Now we have Theorem 1.1.3. (1) For any zeH, there exists an element a in SLjCIR) satisfying OLi = Z.
(2) The homomorphism i induces an isomorphism GLU^)/^''
^ SL2(U)/{± 1} ^ Aut(H).
(3) S02(U) = {oceSL2iU)\(xi = i} and U" • S02{U) = {a G GL2'" (IR) I (xi = i}. Proof. For any z = x + yiell, put
'y'[i:
Then a belongs to SL2iU) and
ai = z; this proves (1). The first isomorphism of (2) is obvious. To see the second isomorphism, we have only to verify the surjectivity. For this purpose, it is sufficient to show that if an element ij/ of Aut(H) satisfies \l/(i) = i, then there exists an element P in S02(IR) such that ^ = i(p). In fact, for each element (/>eAut(H), we get an element aeSL2(U) satisfying OL~^ (j){i) = i by (1). Then taking i(a" ^)(/> in place of ^, we have (J) = I(OLP) for some j5eS02(IR); this implies i is surjective. Now let xj/ be an element of Aut(H) such that ^(i) = i. We put p(z) = (z0/(2 + 0
(zeH),
which is an isomorphism of H onto K. Since p(0 = 0, tj^pif/p'^ is an automorphism of K such that ;/(0) = 0. Applying Schwarz's theorem to rj and ^ " S we see that ^(w) = w (weK).
4
1. The Upper Half Plane and Fuchsian Groups
A further application of Schwarz's theorem shows that there exists 6(0S60)
r  z + ai (thefixedpoint = 0, a 7^ d). First let a = r . ^ ^ , and B be an _0 f/J L —sin^ COS0J element of Z(a). Since ocpt = j?ai = PU Pl is also a fixed point of a. Hence fit = ± i. If det(j5) > 0, then ^i = i and this implies P e GL2 {U)i. Assume that det(j5) < 0. Then (1.1.7) implies that pi = i. Put P' =
~
U. Then det(j5') > 0 and p'i = i;
therefore, P' e GL^ ([R)f. In particular, both P and ^' are commutative with a, hence r  1 01 so is . On the other hand, we have fl 01 f  l Oli fcos^ L 0 i J l 0 ij =^Lsin^
sin^l cos^J^"'
since a is not scalar. This is a contradiction; thus Z{cc) = Z{cc) n GL2 (U). This proves (1). The other cases can be shown by direct calculation. D For any subgroup G of GL2{U), we denote by N{G) the normalizer of G in GL2(IR); namely (1.3.4)
N{G) =
{oLeGL2(U)\aGoLy=G}.
Then we obtain
Lemma 1.3.4. Let G he GL^([R), ( Z G H ) , GL^([R)Sf) (xeR u {00}), or GL2'"(IR), (x, x ' e R u { o o } , X T^x'). Then iN(G):N(G)r\GL^(U)'] = 2 and further (1) iV(G) n GL2^ (R) = G 1/ G = GL2' (U\, (2) Ar(G) n GL2'" (IR) = GL2' (R), 1/ G = GL2^ {U)^^\ (3) [iV(G) n GL2' (R): G] = 2 1/ G = GL2'' (fS). x
10
1. The Upper Half Plane and Fuchsian Groups
Proof. Since N{p~^Gp) = p~^N(G)p for p e G L j (IR), we may assume that z = i, x = oo and x ' = 0. Therefore S IGL2(U): GL^ (IR)] = 2, and
~
UN(G).
ri
0 eN(G) by Lemma 1.3.2(1), we see
L
Since
lNiG):N(G)nGL^(U)^
0 1.
[N(G):iV(G) n GL^ (R)] = 2. Let G = GL2'"(lR)i. liPeN{G) n GL^ (IR), then j^i = U since j5aj5~^ has a fixed point ^/ for any OCGG. This implies j8eG, and therefore, N(G)nGL^ (U) = G. A similar argument is also applicable to G = GLJ([R)!^\ Next assume that G = GL2 (R)ao,o» and let PeN(G) n GL2 (K). Then by a similar argument as above, p induces a permutation of {00,0}. Then we have [N(G)n GL2'"(IR):G] ^ 2. Since we see easily that but
^
~
UN(G) n GL^(U)
~ ^ U G , we have [iV(G)n GL2'"(R):G] = 2 for G = GL2'" Woo.o
•
Lemma L3.5. (1) /4n}; two distinct elliptic elements in GL2 (RL (zeH) are not conjugate in GLj(IR). (2) If two distinct parabolic elements ofGLjW are conjugate by a matrix of negative determinant, then they are not conjugate in GL2iU). Proof (1): Let a and p be elhptic elements such that P = SocS'^ with ^eGLj(IR). Then Sz is also a fixed point of j5, and therefore, Sz = z. This implies that SeGL2(U)^. Since a and S are commutative, we get a = j5. (2): Let a and p be distinct paraboHc elements. Assume that P = yocy~^ = SocS~^ with y, SEGL2iU) such that det(y) < 0 and det(^) > 0. Then S~^yeZ(oi). Since det(')i^(x, —y\ Ci and C2 are symmetric with respect to the real axis. Therefore, the intersection points of C^ and C2 He on the real axis. Conversely, by a similar argument as above, every point on the real axis has equal distance from z^ and Z2.
n Let a be an elliptic element of 5L2([R) which fixes ZQ, and p an isomorphism of H onto K such that P{ZQ) = 0. Then we see pap'
= P,
p
0 e
Since (1.4.4)
CCZQ
=
ZQ,
id
{—
n 0. Then by Lemma L4.3(2), (1.4.5) and (1.4.6), we see that ^ y ^ dx = ^ y ^dx°(x— j 2i/[arg(7(a, z))] Zl
Zl
'^'/(ai)" is also measurable on SL2(U) and if{z)dv{z)= H
] SL2im
f(ai)da.
§1.5. Fuchsian Groups
17
§1.5. Fuchsian Groups Let r be a group acting on a topological space X. We say that a group F acts properly discontinuously on A^ if for any two points x, y of X, there exist neighborhoods 1/ of X and V oi y such that *{yer\yVnV^0] 1. Then the universal covering of 9? is isomorphic to H. We fix this isomorphism and consider the homotopy group 7Ci(9?) of 9i as a subgroup of Aut(H). Then the subgroup F = i~^(;ri(9l)) of SLjCIR) is a Fuchsian group. Moreover F contains neither elliptic elements nor parabolic elements. (We shall give the definitions of the Riemann surfaces and their genera in §1.8.) Example L5.3. Let F = SL2(Z)
{[::]
eSL^
a, b, c, deZ
Since SLjiZ) = M2(I.)nSL2(U) and M2(Z) is discrete in MjCK), SLjCZ) is a discrete subgroup of SL2(IR), namely, a Fuchsian group. We call F or its subgroups of finite index modular groups. We shall study these groups in detail in Chapter 4. Example 1.5.4. Let B be an indefinite quaternion algebra over Q and R an order of B. We may identify B(S>QR with M2(IR), and then R is discrete in M2(IR). Therefore F =^ Rn SL2(U) is a Fuchsian group. Example 1.5.3 is a special case of this. (We shall discuss quaternion algebras in Chapter 5.)
§1.6. Fundamental Domains To study the quotient space r \ H for a Fuchsian group T, it is useful if there exists a connected domain of H which represents F\H. Such a domain is called a fundamental domain of F. Let r be a Fuchsian group and F a connected domain of H. We call F a fundamental domain of F ii F satisfies the following three conditions: (ii) F = U with an open set U consisting of all the interior points of F; (iii) yUnU = 0foranyyeFZ{F). Put, for example.
r cos7r/3 sin7r/3~\ _—sin7c/3 c o s 7 r / 3 j / '
§1.6. Fundamental Domains
21
Then we can take as fundamental domains of Tj and r2 the domains F^ and Fj shown by the figures below.
Fig. 1.6.1
Now let us show that any Fuchsian group F has a fundamental domain. We fix a point ZQEH which is not an elliptic point of F, For yeF — Z(F\ we put Fy = {z6Hrf(z, Zo) ^ d(z, yzo)}, L/, = {zeHrf(z,Zo) 0, where ZQ is the point taken above. Assume that M^nCy^ have yzQ e M2,., since
0. Then we
d(zo, JZQ) ^ (i(zo, z) + d(z, yzo) ^ 2r for zeM^nCy. Since M2r is compact, the number of such a 7 is finite. This implies the assertion since M is contained in M, for some r. D We define the subsets F and 1/ of H by (1.6.1)
F=[]Fy yer
= {zeHId{z, ZQ) S d{z, yzo) for all yeF}, (1.6.2)
U=
f]
Uy
yerZ(r)
= {zeHM(z,Zo) < d(z,yzo)
for all
yeFZ{F)}.
22
1. The Upper Half Plane and Fuchsian Groups
Let us show that F is a fundamental domain of F. First we are going to show that F satisfies condition (i) of the definition of the fundamental domain. Let Zj be any point of H. Since F is a Fuchsian group, {yzQ y e F } has no accumulation point in H. Hence there exists the minimum in {d(zi,yzQ)\yer}, which we denote by d{zi, yo^o) with some yo^T. Since d(zi, yo^o) ^ ^(^i» T^o) for any y e F , we have ZiEyoF. Next let us show that U is open and F = U. Take an open neighborhood V^ for each point z e H so that V^ is compact. Then
zeH
yerZin
We easily see that if Uy^V,^ 0 , then U^nV^^ V^ or CynV,^ 0. Therefore it follows from Lemma 1.6.1 that f]yerz{r)i^y^^ K) is an intersection of finitely many open subsets. Therefore it is an open set, so that U is open. By a similar argument, we have that
zeH
yerZin
zeH yerZ{r)
.
Since obviously F ^ U, we get F = U. Lastly let us verify condition (iii). Let yeF — Z(F). Suppose that UnyU ^ 0 and ZIGUnyU. Since Zje U, we see that ^(zi, Zo) < f/(zi, yzo) = i/(y"^Zi, Zo). On the other hand, since Zj ey[/, we see that d(y~^Zi,Zo); ( 1 ^ / c ^ / ) ,
y^x:^y
(/+l^/c^m).
We can take neighborhoods Wy_ of y^x and V^ oiyiovk>l Put k=l+1
such that W^r\V^ = 0.
k=l+1
§1.7. Quotient spaces r \ H *
25
Then U and V are neighborhoods of x and y, respectively, and we see that yUnV^0ye{y^,,
. . ,yi}oyx
= y.
Therefore Lemma 1.7.1 implies that r \ X is a Hausdorff space.
D
In particular, the quotient space T \ H of a Fuchsian group T is a Hausdorff space. If r \ H is compact or has a suitable compactification, then we can apply the theory of compact Riemann surfaces to study r \ H . We begin by adding cusps of r to H to compactify r \ H . Let Pp be the set of all cusps of F and put (1.7.1)
H*=H* = HuP^.
When r has no cusps, Pr = 0 and H* = H. We put (1.7.2)
I/, = { z G H  I m ( z ) > / } ,
Uf = Ui^{oo},
/ > 0.
Now we define the topology on H* as follows: (i) for z e H, we take as the fundamental neighborhood system at z in H* that at z in H; (ii) for xePp, we take as the fundamental neighborhood system at x the family {(7~^ Uf\l> 0}, where (TESL2iU) such that ax = oo. Then H* is also a Hausdorff space under this topology. In fact, put and X = — d/c. Then we see that
"[::]
(7'Ui = {zeHIm(z)/cz + d\^ > 1} and this is the inside of a circle with the radius (2/c^)" ^ tangent to the real axis at x. For xePp, we call a~^Ui a neighborhood of x in H. Since the action of T on H is a conformal mapping which maps circles or lines to those, F also acts on the topological space H*. Therefore the quotient space T \ H* can be defined and we may regard T \ H naturally as a subspace of T \ H*. When F has no cusps, we have
r\H* = r\H.
Fig. 1.7.1
26
1. The Upper Half Plane and Fuchsian Groups
Now let us show that T \ H* is also a Hausdorff space. Lemma 1.7.3. Assume that oo is a cusp of F and put
Lety = \^
e r . If\ch\ < 1, then
yeT^.
Proof. Assume that \ch\ < 1. We define inductively y„er'{± yo = y,
y« + i = >''»
Q
1} by
i V"
We write
'•=[:: ythen we see that a„+i = 1 a„(c„h\
b„^i = a^K
?„ + i = ciK
d„ + i = l \a„{c„h).
Hence we get that c„=c{chr\ and \a„^,1 = \d„^,
1 = \a„\\c„h\ S i\a\ + n)\ch\K
Therefore y„ converges to
. Since F is discrete, there exists n such that
y« ~
n
fl„^a + n,
(n>0),
1 • ^^^^ implies c = 0.
D
Lemma L7.4. Let Xj, X2 be two cusps off, and (T^, (TJ elements ofSL2(U) such that (TiXi = (72^2 = 00. Put
a,r,Mi''{±
'^A[o iTh^^} ^'='^'^'
Take two positive numbers /j and Ij satisfying /1/2 > I ^1^21 Then for y(Ji^Ui^na2^Ui^
=0
if
yef,
yx^^X2.
Proof By taking a^Fa^^ and a^ya^^'m place of T and a, we may assume Xj = 00 and (Tj = 1. Assume that yC/ji n c ^ ^ C/j^ ^ 0  P^t ^ = 027 and write ^~ ^ =
J • Since dUi. n C/,, ^ 0 , we see that
Ic
d] IJ2 < Im(^" ^z)Im(z) = \m{zfl\cz
+ ^1^ ^ c  ^
§ 1.7. Quotient Spaces T \ H*
27
where zeSUi^n Ui^. Put 7i = K. Sy^S'^
. ei"oo * { ± 1 }• Then we see
= (T2yyiy~^ 0, and
define the function ij/ on Uf by ^Inizjh
^iz) = 0
for zeUi, for z = 00.
Then on(z)^. y
dxdy To see that M(
Let a =
J P ^ ^^ element of GL2 (U). We put; (a, z) = cz + ^ as in (1.1.2). This is
a holomorphic function on H without zero. We recall the basic properties ofj(a, z), since they are often used. Let a, PeOmU) and zeH. Then (2.1.1)
7(aft z) =7(a, MJiP, 4
(2.1.2)
rf(az)/Jz
(2.L3)
7(a"S z) =j(a,
oi'z)';
= det(a)//(a,z)2;
Im(az) = det(a) Im(z)/7(a, z)\\
Moreover, by (2.1.1), we see that (2.1.4)
7(a, z) = const.;()?, z)^(xr'^
GLUmoo •
Let k be an integer. For any function /(z) on H, we define the action of an element a of GL2 (K) by (2.1.5)
(Aa)(z) = det(a)'^/27(a,z)V(a^)
(zeU).
By (2.1.1), we have (2.1.6)
A«P = ifMkP
(a,peGmU)).
For scalar matrices, we see in particular, (2.1.7)
/  , P ^1 = sgn(a)V
(a # 0).
Let r be a Fuchsian group and k an integer. A meromorphic function/(z) on H is called an automorphic form of weight k with respect to F or simply a
38
2. Automorphic Forms
rautomorphic form of weight /c, if it satisfies f\ky=f for all elements yeP. We denote by QkiP) the set of all automorphic forms of weight k with respect to r. Then Qk(r) is a vector space over C and (2.1.8)
ifr ID r\ then Q^iF) a Q^{r');
(2.1.9)
iffeQkin
and a6GLj(IR), then f\kOCGQk(oi~^roc);
(2.1.10)
iffeQ,(r)
and geQ,(n
thenfgeQ,^i(r).
Furthermore by (2.1.7), we see that (2.1.11)
ifk is odd and 1 eT, then Q j r ) = {0}.
Let Q{r)bQ the module generated by all ^^(r); namely
Q(r)=
X
Q,{r).
k=oo
Then we have Lemma 2.1.1. The module Q(r) is a graded ring. Proof By (2.1.10), Q(r) is obviously a ring. Now it follows from (2.1.4) that for elements a, j5 of T, 7(a,z)=7(i5,z)oai5^er^. Since we have i;(r\H)/} for a sufficiently large /. Next suppose that k is odd and — 1 ^ r . From the above result,/^12k0 satisfying (2.1.12), and put / = [r^• {± 1}: r;• { ± 1}]. By assumption, / is finite and we get
,Tr>{±l} = ±[J''Jm6zJ. Let/be an element of ^n(r), then we have the Fourier expansion of the form
I«.
^Ininlzjlh
which is the Fourier expansion of / at x as an element of Qk(^'). Hence by Corollary 1.5.5, we obtain/e^uiT'). Conversely, suppose/e Qj,{F) n ^'^(r'). Then it has the Fourier expansion at x: (/,(7i)(z)= f
a„e2/'\
Since/ belongs to Qk(F\ we have (/lkO~^)(z + /i) = (/fc(T~^)(z). Hence a„ = 0 unless n is divisible by /, and we have n=l
which is the Fourier expansion of/at x as an element of Q^i^). This implies that/ belongs to ^i,(F). D Theorem 2.1.4. Assume thatfeQ^JX) is holomorphic on H. If there exists a positive real number v such that /(z) = 0(lm(z)^)
(Im(z)^O)
uniformly with respect to Re(z), then f belongs to ^fc(r). Moreover if we can take v so that v 0 such that \g(z)\ ^ MIm(z)~*/^. Therefore we get 2ft
1""! = ^ jg(x + yi)e'''"^''^y'^'^dx\ 0
< My'^'^e^"^''*. In particular, taking y = 2/n, we obtain \a„\ ^ Ln''^^ with L = Me^"/''2~*/^.
D
We note that the estimate of the Fourier coefficients a„ in Corollary 2.1.6 is not best possible. In fact, when T is a congruence modular group, it is proved that a„ = p(«(*^)/2) (c.f. Theorem 4.5.17). It is convenient to generalize the notion of automorphic forms to automorphic forms with a character. Let T be a Fuchsian group, and x a character of F of finite order. Put F^ = Ker(x). Since it is a subgroup of F of finite index, it is also a Fuchsian group. We put Odr.x) = {feQ,ir,)\f\,y
= xW/for any yeF},
%(F,x) = ^kir,x)r^%{F,l In particular, if x = 1 , then Q,(F, x) = Ouin ^kir. x) = ^ . ( r ) , %(r, x) = %(n ^k(r, X) = ^kiry We call the elements of ^fc(r, x) automorphic forms of F with xBy definition, we easily see the following (2.1.27) through (2.1.31). (2.1.27)
Suppose  1 e r . / / x (  1 ) # (  l)^ then Q,(F, x) = {0}.
(2.1.28)
Let F' be a subgroup of F^^ offinite index. Then we have:
^,(r,x) = r2,(r,x)n^,(r'), ^,(r,x) = Ofc(r,x)n^,(n, ^,(r,x) = D,(r,z)n5^,(r'). (2.1.29)
Let F' be a subgroup of F of finite index. We also denote by x the
44
2. Automorphic Forms
restriction of x to F'. Then we have
^,(r, X) c: ^,(r\ x\ %{r, x) ^ %ir\ x\ ^^(n x) ^ ^kir\ xl (2.1.30)
Iffe^.ir,
x) and ge^i{r,
i/zl thenfgGs/.^^ir,
#).
(2.1.31) Let a be an element ofGL2 (U). Put F' = cc~^F<x, and define the character X' of F' by x'(^~^yo(.) = x(y) Then we obtain the following isomorphisms by the correspondence "/ •—•/I ^ a ":
j^,(F,x)^^kir\x'l
^,iF,x)^'^kir\x'l
^,{F.x)^^kir\x'y
Let /(z), g(z)E^k{F,x) Assume that either /(z) or g(z) is a cusp form. Then/(z)sf(z)ey'2fc(Az^), and therefore, /(z)^(z)Im(z)* is bounded on H by Theorem 2.1.5. Furthermore if yeT, then f{yz)g(yz) \m{yzf =f(z) g{z) lm{z)\ Hence jr\H/(^)^(^) ^^{zfdv{z) is meaningful and finite. We put (2.1.32)
(/, g) = v(F\U)'
J /(z)^Im(z)*rft;(z), r\H
and call it the Petersson inner product (though it is defined only when either/(z) or g(z) is a cusp form). It induces an Hermitian inner product on ^ ^ ( r , x)Let Fi and F2 be two Fuchsian groups, and XnXi characters of T j , r 2 of finite order, respectively. Assume that there exists a Fuchsian group F' contained in Ker(xi)nKer(/2). For/(z)G5^fc(ri,/i) and ^(z)e^fc(r2, Z2X we can define the Petersson inner product (/, g) by considering/(z) G ^^i^') and g(z) e ^^(r'). We see easily that (2.1.33)
(/, g) is independent of the choice of F'.
We put ^,(F)
= {ge^,(F)U
^) = 0
for a l l / e ^ , ( r ) } ,
and
^fc(A X) = {g^^kir. x)U^) = 0 for all/G^,(r, x)}Theorem 2.1.7. (1) ^ , ( r , x) = ^kir, x) 0 ^ . ( r , x)(2) Let F' d F, and x! be the restriction of x to F. Then
^k(r,x) = ^kir\xlr^%{F,xy Proof ([Shimura 12]) The first assertion is an easy exercise of Linear Algebra. Let us prove (2). There exists a normal subgroup F" of F of finite index included in Ker(x). We are going to show that J^ki^, x) ^ ^fc(^"). Let giz)Gjr,^(F, x) Since gf(z)e^jfc(r"), we can write giz) = g,{z) + / i (z),
g,(z)e ^ , ( r " ) , / i (z) e ^ , ( r " ) .
Let yeF. Then g\ky = xiy)d Since ^fc(r") is stable under y and (/ijy, ^2) = (/ii, /1217 ~ ^) for any /ij e ^ ^ ( r " ) , /12 e ^^(r"), the space J^ki^") is also stable under y. Thus
§2.2. Differentials on Compact Riemann Surfaces
45
Then gi\ky = x{y)Qi ^ind fi\ky = xiy)fi by applying (1) for T' and the trivial character. This implies/i(z)G^fc(r,x), and
(/i J i ) = (9giJi)
= (gJi)  (91 Ji) = 0.
Therefore/i (z) = 0 and g{z) = giiz)ejr^(r"). {f,g) = 0 This implies g{z)eJ^kir\ hand side.
Since ^^Cr', /) ^ ^kiri,
we have
forany/(z)e5^,(r,/).
xl Obviously the righthand side is included in the leftD
§ 2.2. Differentials on Compact Riemann Surfaces Let 91 be a Riemann surface, and m any integer. The set {(0^, K^, t^)} consisting of the triples (0^, K^, r^) of a coordinate neighborhood V^ of % a local coordinate r^ on V^ and a meromorphic function 0^ on V^ is called the local expression of a differential of degree m, if the following conditions are satisfied: (i) {(Vfi, tf,)} is a coordinate system of^; (ii) (t)^(a)(dtjdt,)"'{a) = 0,3 for any aG F^n F„ provided V^nV,^
0.
Two local expressions {((p^, F^, r^)} and {(;\V,,t,)}.
For a differential co = {(,(a) = (dc^/dt,){a)
(aeV^).
Then {(0^, K^, t^)} is a differential of % and we denote it by d(l). The definition is obviously independent of the choice of a local coordinate system, and 0 is a constant if and only if ^0 = 0. Hereafter we assume that 9? is compact. Then D^(9?)^{0}, and therefore, D'"(9i) 7^ {0} for any integer m, since (^(Z))"* eD"'(9?). If co and co' are two differentials of D'"(9{) and co 9^ 0, then we have co" ^ co' e D^(9?) (= K(9t)); therefore we obtain (2.2.6)
dimj,(9j)D'"(9?) = 1.
We denote by Div(9?) the free module generated by all points of 9^, or Div(9^) = {J^ggg^ CgalCgeZ and 0^= 0except for finitely many points a}, and call it the divisor group of 9^. We call the elements of Div(9?) divisors of 91. For a divisor a = ^ g ^a^ of 9?, we define the degree ofsi by cieg(a) = Xca
(e/).
§2.2. Differentials on Compact Riemann Surfaces
47
We are going to define the divisors for functions and differentials of 9?. For an element (T^O) of X(iR), we define the divisor of cj) by (2.2.7)
div(0) = X Va()a,
where Va() is the order of at a defined by (1.8.4). Since Va(0) = 0 except for finitely many points a, div((/)) is welldefined and an element of Div(9?). Moreover by virtue of (1.8.5), we see (2.2.8)
deg(div((/>)) = 0.
If 0, ^AeX(9l)^ then we see (2.2.9)
div((/)(A) = cliv((/)) + div(i/^).
Therefore, if we put Div,(9?) = {div((/))g o niz) = cf(z)(d(Gz)ldzr
"'{t, o n(z))"'
= c/(z)7(a,z)2(fg0 7r(z))'" =
c(f\2mCT')iaz){t,on{z))
on U^^. Since / is meromorphic at ZQ, so is (/>g at a. Moreover, if b is a point of ^p, and Kg n Fjj # 0 , then we see easily that (2.3.6) also holds in this case. Thus we have proved that {((/)g, Kg, tg)} is a differential of degree m of 91/, which we denote by coy^ or o){f). Then we easily see that (Ojg = (Of(Og for
fes^2m(r\
^ / + » = % + .X.A,n=
I
z(?)((/>l,7)«
izeW).
yeA\r
It is obvious by the definition of the series and condition (i) that if F(z) is convergent, then it is independent of the choice of the representatives of / l \ r , and F\ky = Xiy)F for any y e r . We call this series the Poincare series. Theorem 2.6.6. Under the same notation and assumptions as above, wie have: (1) F{z) is convergent absolutely and uniformly on any compact subset of H — {yz^lyer, 1 ^ /i ^ m}, and F(z) belongs to Qk(F, x)\ (2) if (i){z) has a pole at w e H and is holomorphic at yw for allyeF — {±\}' A, then F{z) has also a pole at w and its order is equal to that of (piz). Proof Let ZQ be any element of H — {yz^lyeF, neighborhood W of ZQ SO that (2.6.3)
yWnU^ = 0,
(2.6.4)
yWnV,
=0
1 ^ /iiSfn}. Take t/^, V^ and a
(yeT, 1 ^ / i ^ m, 1 ^ v ^ r),
{yer\yWnW^0} = r,^.
Then (2.6.5)
j
X M\ky)iz)\lm(zr'dv{z)=
^
W yeA\r
y€A\r
=
X yeA\r
j \(t>iyz)\lm{yzr'dv(z) W
j \cl>{z)\lm{zr'dv{z). yW
We see that yW^ /).
Theorem 2.6.7. Assume that T, A, (/)(z) satisfy the foregoing conditions (i) through (iii). Let XQ be a cusp ofF. If conditions (iv) and (v) are satisfied for every cusp x which is Fequivalent to XQ, then F(z) is holomorphic at XQ. Moreover, if we can take e in (v) positive for all x which are Fequivalent to XQ, then F{z) has a zero at XQ. Proof First we note that conditions (iv) and (v) are independent of the choice of a. (M, /, 8 depend of course on the choice of a.) Let {a} be a complete set of representatives of A\F/Fx^. For each a, let {j5} be a complete set of representatives of (ai / l a n r , , ) \ r , , . Then T = U „ .^ocj? Put
Then (2.6.6)
F(z) = X I x ( a i ^ ) ( 0 l , a ^ ) W = Z.(4
and
(PoclkJ = xiy)(l>a
(yer^j.
By condition (ii), there exists a neighborhood U of Xo in H such that F(z) is convergent uniformly on any compact subset of U. In other words, by taking (TQeSLjiU) so that ^0^0 = ^ ' a
is convergent uniformly on any compact subset of the domain {zeHIm(z) > b} for some positive number b. Thus it is sufficient to show that (palk^o^ is holomorphic at 00, and moreover, has a zero at 00 if 8 > 0.
66
2. Automorphic Forms
r Assume that OCXQ is not a cusp of A. Since a" M a n T^^ coincides with {1} or {±1}, we have
with C = 1 or 1/2, respectively. Applying condition (iv) for x = OLXQ, a = CFQCC'^, we take M, /, e ( > 0 ) so that ((/>,a(7o^)(z)^Mz^
(Im(z)>/),
so that ((/)J,(7o^)(z)^2M X z + m / i   ^ 
(Im(z)>/),
meZ
where /i is a positive number such that OQF^^GQ ^ • {± 1}
1 /i"'
0 ^i j• 0
^'"^
Comparing the series on the righthand side with the series Im~^~^, we see it is convergent uniformly on any compact subset of the domain Im(z) > /, so that 0aL^o ^ is holomorphic at oo. Moreover if z tends to oo along the imaginary axis, {(t>a\k^o^)i^) does to 0; namely, (/>alik^o ^ ^^s a zero at oo. 2° Suppose OLXQ is a cusp of A. Then a~ Ma n r,^ is a subgroup of F^^ of finite index, so that the summation ^ ^ is a finite sum and
From condition (v) for x = axo, c = OQ^ ^ there exist M, /, e such that (2.6.7)
Ml.aco')iz)\
Noting that (JOPCTQ
^ Mz S M
1 m/i"
0 ij
(Im(z) > /).
for some integer m, we see that 0 , then we see from (2.6.7) that (^^IfcCTo at 00.
has a zero D
We can prove the existence of nonzero automorphic forms using the Poincare series. Here we prove that if T^ — 1, then .^kiF) ^ {0} for any odd integer k. Theorem 2.6.8. IfT^X,
then s^ki^) i^ {0} for any odd integer k.
Proof. Let ZQ be a nonelHptic point of H and put 0, f n=l
\n'\S^\]x^'^'^dx 1. Let us show that C(s) has the infinite product expansion
as) = Ui^pT\
§3.2. The Riemann ZetaFunction
85
where p runs over all prime numbers. In fact, let s be a positive number. Then for any positive integer m, we see
Ids) n i^p~r'\s
I n^'^^>
on R e ( s ) ^ l + e . Since the righthand side converges to 0 when m^ oo, np<m(^ ~P^y^ converges to C(s) on Re(s) ^ 1 + £. Since a is arbitrary, the infinite product expansion holds on Re(s)> 1. More generally, if a Dirichlet series (/)(s) is convergent on a complex domain and has an infinite product expansion of the form cl>is) = U4>pisl p
with a meromorphic function 0). / / Y,^= 1 cin^~^ = X*= 1 ^n^'^ on Re(s) ^ OQ, then a„ = b„ for all n. Proof. It is sufficient to show that if Y,T= i ^n^^ = 0, then a„ = 0 for all n. Since Zn°=i^n"^ is absolutely convergent at S = (TQ, it converges absolutely and uniformly on Re(s) ^ OQ. Let m be the smallest integer such that a^^O. Since we have (n/m)"*' < n'*^^^ for n>m^ and o > 0, we see \aJS
f
\a„\(n/m)^
n = m H 1
g
X
l«J(n/m)''+
£
a„n°'^
((S)= X « « " " ' n = 1
with a sequence {«„} of/C. Let  1 ) ,
r{s) becomes a meromorphic function defined on Re(s) > — 1. Repeating this procedure, r ( s ) is continued to a meromorphic function on the whole splane. We call r{s) the Ffunction. For complex numbers z( 7^ 0) and 5, we put (3.2.1)
z' = ^^oogll + 'arg(z))
(n
0 } , put (3.2.11)
X{s)=^\f(t)t'^dt. 0
If this integral is convergent on some domain, we call X(s) the Mellin transform of fit). By (3.2.10), yl(25) is the Mellin transform of i{co{t)  1). Let now t = e^ and put 5 = (T + it, then we can rewrite A((7 + iT)= J /(e*)e^'''">*^x, — 00
that is, X{a \ iz) is nothing but the Fourier transform of /(e*)^''* for a fixed a. Therefore the Mellin transform of f{t) is meaningful for a such that F(x) =f(e^)e*'^ is a Schwarz function; in other words, F(x) is a C°°function and satisfies for any positive integers m and n. (3.2.12)
„d'"F(x) sup x" dx'
i)
is odd.
Now we see easily (3.2.19)
zcot(z) = iz r,
= iz +
Comparing this with the well known partial fraction expansion (3.2.20)
cot(z) = + z
t
2 '^\
2.
„ = 1 z^ — n^TC
90
3. LFunctions
and putting t = — 2iz, we have t
. *
t^
'^'^='^2^\l.?Ti2^
0, and has the Euler product
(3.3.4)
L(s,0 = Wiip)N(vyT\
92
3. LFunctions
where p runs over all prime ideals of r^. In particular, when ^ is the trivial character {o, we write (3.3.5)
CF{s) = L(s,io) = lN{a)^ =
Yli^N{p)T\
p
a
and call it the Dedekind zetafunction of F. When F = Q, it is nothing but the Riemann zetafunction. Hecke obtained the functional equation for any Hecke Lfunction by generalizing the proof for the Riemann zetafunction described in the previous section ([Hecke]). Before we state the functional equation, let us define the Gauss sum of a Hecke character. We denote by tr = tr^ the trace mapping of F into Q, and put T)~^ = {aeF\tr(ab)eZ
for all
bexp}.
Then X)" Ms a fractional ideal of F and the inverse D = X) (F) of T) ~ Ms an integral ideal. We call T)(F) the different of F. For a basis {aj, . . . , a^} of r^ over Z, we put 4 = det[tr(a,aJ]
(eZ),
and call it the discriminant of F. Now let f be a primitive Hecke character of conductor m with M^, t^^. We put, for a e F (a^O), (3.3.6)
(^^(a) = 'n'(«v/l«vlM«vh, v=l
(3.3.7)
if{a)=H(a))i^(a)K
lfa = b mod m, then if (a) = if{b). Let c be an integral ideal of F such that T)mc is principal and (m, c)= 1. Take an element b of Xp so that T)mc = (b), and define the Gauss sum W{i) of ^ by (3.3.8)
W{^) = % ^
I^^(a)e^""«'>,
where a runs over a complete set of representatives for c/mc. The value W{i) is independent of the choice of c, b and a set of representatives. By a similar argument as in Lemma 3.LI, we have: (3.3.9)
\W(i)\' = N(m);
(3.3.10) let i, rj be two primitive Hecke characters of conductor m and n, respectively, //(m, n)= 1, then (3.3.11)
Wm = i(n)rj{m) W(OW{rj).
For a Hecke character { mod m, we put Ua)=m
(ae/),
then I is also a Hecke character modm. To state the functional equation for a
§3.3. Hecke LFunctions
93
Hecke Lfunction L(s, ^), we put (3.3.12)
A(,A)=y
p^),
j
^ j y
^
jZ.(..?),
where 1 (v^ri), 2 (ri + l ^ v ) .
""
Now the functional equation for a Hecke Lfunction is as follows. Theorem 3.3.1. Let ^ be a primitive Hecke character of conductor m. (1) A(s, i) is analytically continued to a meromorphic function on the whole splane, and satisfies the functional equation A(ls,i)=m)A{s,i), where T(i) =
2'T"W(0/N{mYi\ ri lr2
ri +r2
v=l
v=l
(2) Ifi is the trivial character, then A(s, i) is holomorphic except for simple poles at s = 0 and s = 1; otherwise A(s, ^) is entire. The function A(s, ^) is bounded on any set of the form {seCfl^Re(s)^fe, Im(s)'^c} {a y/3/2. Thus we get c ^ 1 noting that c is an integer. Suppose \c\ = 1. Then z ± fi ^ 1 by (4.1.2). On the other hand, we have \z ±d\> 1 for any integer d, since zeU. This is a contradiction; thus c = 0, y= ±
L and yz = z +fe.For an integer b, both z and z + ft belong to U if
and only if ft = 0. This imphes y = ±
. Consequently, F is a fundamental
domain of F. (2): Since v{F\H) = v{FX we get v(F\H) = n/3 by Lemma L4.4. D
Theorem 4.1.3. (1) Any elliptic point ofF is equivalent toi = ^ — iorC = e"^'^ The point i is an elliptic point of order 2 and
[I :]• < : ;]} The point C is an elliptic point of order 3 and
.={4;:],.[; ;], 4::;]}. (2) The set Ppofthe cusps of F is Q u {00}, and all cusps of F are equivalent. (3) The genus of ^pis equal to 0. Proof (1): It is obvious that interior points of a fundamental domain are ordinary points. Thus any elliptic point must be equivalent to a boundary point of the fundamental domain F defined in Theorem 4.1.2. Since F contains
=[i I] """^ "=[? i} the boundary points off, other than the three points r, f, and f ( = e^"'^^), are also ordinary points. Observing that the interior angle of F at i is TT, we see the order of i is at most 2. Since oi = i, and a>^ = — 1, the point i is indeed an elliptic point of order 2. Since xC = C and the interior angles of F at C and C are both 7r/3, the order of C is at most 3. Now we note
=[:  ; } 2, i//c#2modl2.
Ain.o,Xo,r^.F),
where (/)o = 1 and Xo is the trivial character of F. Then by Lemma 4.1.5, (4.1.5)
G,(z) = ^ X {mz + n)\ •^ m,neZ {m,n)= 1
zeH,
D
100
4. Modular Groups and Modular Forms
and (4.1.6)
E,{z) = 2C(fc)G,(z),
where C(5) is the Riemann zetafunction. Lemma 4.1.6. Let k be an even integer ^ 4. (1) Ef^{z) is convergent absolutely and uniformly on any compact subset o/H, and belongs to ^jt(r). (2) £fc(z) has the following Fourier expansion at oo: £,(z) = 2C(/c) + 2  ^ ^ X cT,.An)e'^'"^ where 0 /c, and has an Euler product
p
The analytical continuity and the functional equation of ((s) induce those of L^is).
§4.2. Congruence Modular Groups For a positive integer N, we define subgroups FoiNX F^iN) and F{N) of 5L2(Z) by ro(Ar) = ]
^
ri(iV) =   ^ r(N):=\\^
eSL2(/)c = 0modiV MeSL2(Z)c = 0,
a=d
=lmodN
\eSL2(Z)\b = c = 0, a = d=
ImodN
We note SL2{Z) = Fo(l) = F,il) = F(n and SL2{Z)^Fo{N):=>F,iN)zDF(N). Further if MIN, then FO{M)ZDFO(NI
F,{M)Z,F,(NX
F(M)^F(N).
These subgroups are modular groups since IF(1):F{N)^ < oo, which will be proved below. We call F(N) a principal congruence modular group, and FQ(N) and
104
4. Modular Groups and Modular Forms
ri(iV) modular groups of Heche type. We call N the level ofToiNX r^iN), and r(N). A modular group containing a principal congruence modular group is called a congruence modular group. For an element y = \
, eM2(Z), we define an
Ic d] element Xf^iy) or y mod AT of M2(^1 NT) by A^(7) = 7mod
[^ :l
where a = a mod AT, h = bmodN, c = cmodN, d = dmodN. Then >ijv induces a homomorphism of SLjil) into SLjCZ/N/). We note that for integers m, w, (4.2.1)
if(m,n,N)=
1, t/ien f/iere exis^ integers m\ n' such that
m' = mmodiV,
n' = nmodN
and (m\ n') = 1.
This can be easily shown, for example, by the Dirichlet theorem on arithmetic progression. Theorem 4.2.1. Let Xf^ he the homomorphism of SL2{T.) into SL2(Z/NZ) defined above. (1) A^ is surjective. (2) Ker(Ajv) = r(N); in particular, r(N) is a normal subgroup of r{l).
C 2>
Proof. (2) is obvious. Let us show (1). Let  _ j  be an element of SL2(Z/NZ), and take integers a^, b^, c^, d^ so that
[:;:;][:'] Then a^d^ — biCi = l mod N, so that (cj, i/j, N) = 1. By (4.2.1), we may assume (ci,di)= 1. Let n be an integer such that a^di —b^Ci = 1 + nN. Since (ci, f/J = 1, we can take integers fl2» ^2 so that a2^i — ^2(N)\SL2(Z/NZ)\;
(2) \SL2(Z/NZ)\
=
N'Y\(l\/p'). P\N
Here <j) is the Euler function.
106
4. Modular Groups and Modular Forms
Proof. The first assertion is obvious. We have only to verify (2) for a prime power N — p^ by Lemma 4.2.3. First suppose e=\. Then Z/pZ is a field, so that GL2(Z/pZ) is equal to the number of all ordered basis of the two dimensional vector space Z/pZ 0 Z/pZ over Z/pZ. Then it is given by p(p — l)(p^ — 1), thus, (1) implies \SL^(ZlpZ)\ = p(p^  1) = p^(l  1/p^). Next assume e> \. Let /l':SL2(Z/p^Z)SL2(Z/pZ) be the canonical homomorphism. Then X' is surjective by Theorem 4.2.1(1) and
: :]'.«^'H[: a[; :]"«^4
Ker(/l') =
For any elements h, c, d of ZIp^Z such that 6 = c = 0 mod p, d = I mod p, there exists a unique element a of Z/p^Z such that ad — hc=\. Therefore Ker(A')= p^ 0, then the righthand side i^ convergent absolutely and uniformly on any compact subset o/H, and f{z) is holomorphic on H. Moreover, /(z) = 0(lm(z)^^)
(Im(z)^O),
/(z) ao = 0(e 2«^™)
(Im(z) ^ oo)
uniformly on Re(z). Proof By (3.2.9), we have for v > 0, limn7(l)''
(  :  ) =
=r(v+l).
Hence there exists L > 0 such that (  ;  ' ) for all n ^ 0. Put z = x + iy, then
(4.3.9)
f \a„\\e''"^ g if f ( Iff ~ ' ~ ' )e''A n=0
\n=0
\
n
J
J
= L(le2"^)^^
This impHes that /(z) is convergent absolutely and uniformly on any compact subset of H. Since (1 — e"^"^) = 0(>^) asy^0, we see /(z) = 0(y"''~^). Moreover (4.3.9) implies that /(z) is bounded when y • oo. Put g{z)= X «« + i^Itiinz n = 0
Since g{z) also satisfies the assumption, it is bounded on a neighborhood of oo. Therefore we obtain f{z)ao
= e^''''g(z) = Oie^^y)
(y^oo).
D
118
4. Modular Groups and Modular Forms
By the above lemma, all holomorphic functions f(z) on H satisfying (4.3.7) correspond bijectively to all sequences {a„}^=Q of complex numbers such that an = 0{n^) with V > 0. As a preliminary to Hecke's theorem, we shall prove the following Lemma 4.3.4. (PhragmenLindelof) For two real numbers Vj, V2 (vi < V2), put F = {seCvi ^ R e ( s ) ^ V2}. Let (j) be a holomorphic function on a domain containing F satisfying \(j>(s)\ = 0{e\^\')
(ITI^OO),
s = (j + ix,
uniformly on F with d > Q. For a real number b, if (/>(s) = 0 (  T  ^ )
(T  • 00)
on Re(s) = Vj and Re(s) = V2, '
then 10(5)1 = 0(111")
(TOO)
uniformly on
F.
Proof By assumption, there exists L > 0 such that \(j>(s)\^ Le\'^\\ First we consider the case when 6 = 0. Then there exists M > 0 such that (/)(s) ^ M on the lines Re(s) = Vi and Re(s) = V2. Let m be a positive integer such that m = 2 mod 4. Put 5 = (7 + ix. Since Re(s'") = Re((o + ii)'") is a polynomial of a and T, and the highest term of T is — T"*, we have (4.3.10)
Re(s'")=  T ' " + 0 (  T r  ^ )
(TOO),
uniformly on F, so that Re(s'") has an upper bound on F. Taking m and N so that m> 6 and Re(s'") ^ N, we have, for any 8 > 0, \(t>(s)e''"'\ ^ Me'^ on Re(s) = Vj and Re(s) = V2, and 0(5)^"'" = 0(^1^'^^^'" + ^'^''"") > 0,
(r > 00)
uniformly on F. By the maximum principle, we see
\(j)(s)e''"'\^Me'^,
(seF).
Letting e tend to 0, we obtain , where log takes the principal value. Since R e ( l o g ( s  V i + l)) = l o g  s  V i 4 1,
Next assume
§4.3. Modular Forms and Dirichlet Series
119
we have uniformly on F \il/(s)\ = \sv,^lf^\r\'
(ITHO)).
Put 01 (s) = (l)(s)/il/(s). Then 1 + V, so that it is holomorphic on Re(s) > 1 + v. We call L(s;f) the Dirichlet series associated with f. For N > 0, we put (4.3.12)
A^(s;f) =
{2n/^)^r(s)L{s;f).
Theorem 4.3.5. (Hecke) Let / ( z ) = Xr= o «n^^""" «"^ ^(^) = Zr=o^«^^"'"' be holomorphic functions on H satisfyirtg (4.3.7). For positive numbers k and N, the following conditions (A) and (B) are equivalent. (A) g{z) = (i^z)'f{l/Nz). (B) Both Aj^{s;f) and Apf(s; g)can he analytically continued to the whole splane, satisfy the functional equation ^Nis'J) =
Aff(ks;g),
and s
k —s
is holomorphic on the whole splane and bounded on any vertical strip. Proof (A)=>(B): Since there exists v > 0 such that a„ = 0(n'') and b„ = O(n'),
n= 1
and f
J fljre2^"^/^r^rfr
n= 1 0
(a>v+l)
120
4. Modular Groups and Modular Forms
are convergent. Therefore we see, for Re(s) > v f 1, AAs;f)= t a„(2nn/^)^]eU^'dt n= 1
0
n= 1 0
=
]t^{fiit/^)ao)t'dt 0 n
00
00
= __» + ^ t^f(il^t)t'dt ^
+ \ f(f(itl^)
1
ao)t'dt.
1
Since g(z) = (— i.Jl^z)~^f( — l/Nz), we obtain (4.3.13)
A^{s'J)=
^^^]t'^{g(it/^)bo)t'dt S
K
+
^
S
]f{f(it/^)ao)t'dt 1
on Re(5) > Max{fe, v + 1}. By Lemma 4.3.3, when t tends to oo, we have f{it)ao = Oie''"), g{it)bo = 0(e'"), so that J('(/(it/v^)ao)r'A and
Jt'»(3(,t/VJv)b„)r'dt 1
are convergent absolutely and uniformly on any vertical strip. Therefore they are holomorphic on the whole splane. If we define Af^{s; f) for any s e C by (4.3.13), it is a meromorphic function on the whole splane, and
is an entire function and bounded on any vertical strip. Similarly Af^{s; g) is also analytically continued to the whole splane, and satisfies (4.3.14)
AAks;g) =  ^   ^ + ] t'^{giit/^)bo)t'dt S
+
K — S
J
]fmtlsf^)ao)t'dt.
§4.3. Modular Forms and Dirichlet Series
121
Comparing (4.3.13) and (4.3.14), we obtain vl^(s;/) = Af^{ks; g), (B)=>(A): By (3.2.14), we have fiiy) = ^ . l a „ Zni „ = 1
j
{2nny)^r(s)ds +««
Re (s) = a
for any a > 0. If a > v + 1, then L(s;f) = ^^=1 ^n"~* is uniformly convergent and bounded on Re(s) = a, so that by Stirling's estitnate (3.2.8), Aj^(s;f) = {2n/y/N)~^ r(s)L(s;f) is absolutely integrable. Therefore we can exchange the order of summation and integration, and f(iy) = ^
I
(^y)'A^(s;f)ds
+ a^.
^'^^ Re(s) = a
Since L(s;f) is bounded on Re(s) = a, we see, for any // > 0, (4.3.15)
\AAs'J)\ = 0(Im(5)r^)
(Im(5) ^ ^ )
on Re(s) = a by Stirling's estimate. Next take P so that k — p > v{l. A similar argument implies that for any /x > 0, \AAs;f)\ = \A^iks;g)\
= 0(\lm(s)\'^)
(Im(s) ^ ex,)
on Re(s) = p. By assumption,
is bounded on the domain P ^ Re(s) ^ a. Hence for any /z > 0, we see by Lemma 4.3.4, (4.3.16)
(4.3.15) holds uniformly on the domain P ^ Re(s) ^ a.
Furthermore we assume that a > /c and p 1). Then the following conditions (A^) and (B^) are equivalent. (^^) f^\MNm^) = C^Qf(B^)Af^(s;f ij/) can be holomorphically continued to the whole splane, bounded on any vertical strip, and satisfies the functional equation: Aj^(s;f il/) = i^C^Af^(k  s; g, ij/) for a constant C^. Proof. Apply Theorem 4.3.6 for /^, C^gf, and Nm^ in place of /, g, and AT, respectively. Then we obtain the theorem by (4.3.20). D We put
«(«)= I \
(aeW).
Lemma 4.3.10. (1) Let f be a holomorphic function on H satisfying (4.3.7), and \jf a primitive Dirichlet character of conductor m. Then for an integer /c > 0, we have
u=wm' t m(f\,{N)Wm/lV{n
126
4. Modular Groups and Modular Forms
then glkixi^)y(^^ u))oc(u/m) = g\k(x(m)y(m, v))(x{v/m) for any integers u and v prime to m. Proof. By the assumption {A^) and Lemma 4.3.10, we see m
(4.3.25)
_
m
X mf\Mulm)m(Nm')
= z(M)"/'(.iV) ^ ^(u)3La(u/m).
M = l
U = l
For each integer u prime to m, take an integer v so that — uvN = 1 mod m. Then we have by (4.3.23) (4.3.26)
f\k(x(u/m)o)(Nm^) = g\ky{m, v)(x(v/m).
Since the lefthand side is independent of the choice of a representative of u mod m, so is the righthand side of the choice of y(m, v). Using (4.3.26), we rewrite (4.3.25) into (4.3.27)
T.^{v)g\Mm)y(m,
v)Mv/m) = 0.
V
Here v runs over a complete set of representatives of Z/mZ. We note that (4.3.27) is independent of the choice of representatives of Z/mZ. Let v^, Vj be two integers prime to m. Multiply both sides of (4.3.27) by ^(vi) — \j/(v2) and take the summations with respect to all nontrivial Dirichlet characters i/^modm. Noting that •AC^i) ~ ^(^i) = 0 if i/^ is trivial, we obtain ^lfc(z(w)  7(m, v^))(x{vjm) = gUx(m)  7(m, V2))cc(v2lm).
D
Lemma 4.3.14. Let m and n be odd prime numbers or 4. Assume both m and n are prime to N. Iff(z) and g{z) satisfy condition (A^) in Lemma 4.3.9 with the constant C^ = l(m^)^(—N)W(\l/)IW(\l/) for any primitive Dirichlet character ij/ whose conductor m^ = m or n, then g\ky = xiy)9 for all yefQiN) of the form y
luN
n]
.Jmvl Proof Put y = \[uN nj . Since we can take y and y' for y(m, v) and y{m, — v) in Lemma 4.3.13, respectively, we have glkixim)  y')oc{ v/m) = g\k(x{m)  y)oc{v/mX so that (4.3.28)
g\Mm)  YM2v/m)
We can also take 7'~ ^ = and obtain(4.3.29)
^, \_ — uN
g\kixin)y')
= g\Mm)  y).
and y " ^ = ^, for 7(n, v) and y(w,  v\ m_\ \_uN m_\ =
g\Mn)y'M2v/n).
§4.3. Modular Forms and Dirichlet Series
127
Since ;f(n);f(m) = 1, we see
x(n)y'~^ = xin)(x{^)y')y'~\ (4.3.30) (z(«)y ^M2v/n)=
x(n){xim)y)y
^Gc{2v/n)
as elements of C[GL^([R)]. Rewriting (4.3.29) by using (4.3.30), we have (4.3.31)
9\Mm)Y)
=
g\Mm)y)y'oi(2v/n)y\
so that from (4.3.28) (4.3.32)
g\^^^(rn)y)(ly'oi(2v/n)Yoc(2v/m))
= 0.
Put
h = g\kix(m) y) = x(m)g g\kyThe function h(z) is holomorphic on H and satisfies (4.3.33) h\,p = K ^ = ,  a (  2 . / n ) / a (  2 . M ) = [^^;/^
4/;'^]
Since tr()5) = \4/mn — 2 < 2, and tr(^) 7^ 0,1 by the assumption on m and n, p is elliptic and any eigenvalue of p is not a root of unity. In fact, since the entries of p are all rational numbers, the eigenvalues of P belong to a quadratic field. Hence if an eigenvalue of jS is a root of unity, then it must be + 1, ± i, + e"'/^ or ± e^^'^^, so that tr(j?) = 0, ± 1, or +2, which is a contradiction. Let now ZQ be a point of H fixed by p. Put p = (zozo)T[
"HeGL2(C),
P(w) = {h\kp')M = j(p\
wr'h(p'w)
(weK).
Then p(w) is holomorphic on K. Since we can write ppp~^ ~\ a ri
^^^^ C^C,
we see by (4.3.33) (4.3.34)
p(C'w) = CV(w).
Let p(w) = X"°=o^«^" t>e the Taylor expansion at w = 0, then we get
for all n. Since C is an eigenvalue of P which is not a root of unity, we obtainfl„= 0 for all n. Therefore h(z) = 0, so that g\ky = X(^)Q = x(y)9' D Now we are ready to state the main theorem in this section. For two coprime positive integers a and h, we put (4.3.35)
A(a,h) = {a + nh\neZ}.
128
4. Modular Groups and Modular Forms
Let M be a set of odd prime numbers or 4 satisfying the following two conditions: (i) any element of M is prime to N; (ii) N\nA(a,b)^ 0 for any A{a, b). There exists such a set M. We can take as M, for example, the set of all odd prime numbers prime to N by Dirichlet's theorem on arithmetic progression. Theorem 4.3.15. (Weil) Let k and N be two positive integers, and % ^ Dirichlet character modN such that x(— 1) = (— 1)^ For two sequences {a„}^=Q and {b„}^=Q of complex numbers such that a„ = O(n^), b„ = Oin") (v > 0), put
m = I a„e' «n ==0 o
and
g(z)= t^e'^'"'
(zeH).
«=o
Then f{z) e %{N, xX 9(z) ^ %(^^ x) ^^d g =f\^cOff, if the following two conditions are satisfied: (1) Aj^(s; f) and /l]v(s; g) satisfy condition {B) in Theorem 4.3.6; (2) for any primitive Dirichlet character \j/ whose conductor m^ belongs to M, yljv(s; /, ^) and Af^(s\ g, ij/) satisfy condition (B^) in Lemma 4.3.9 with the constant C^ = C^,^ =
xH)^{N)W(il^)/W{ii;).
Moreover, ifL{s;f) is absolutely convergent at s = k — 6 for ^ > 0, thenf(z) and g(z) are cusp forms. Proof We shall prove that g\i,y = x{y)g for 7 =
^^ . eToCN). If c = 0, then \_cN a J a = ii = ± 1, so that gfl^y = x(d)g = x(y)g since %(— 1) = (— If. Next assume c ^0. Since (a, cN) = (d, cN) = i, there exist integers s, t such that a + rciVe M, and d + scN e M by condition (ii). Put m = a\ tcN, n = d\scN, u = —c, then
r a bivi \_cN d\~[o
V = —{b\sm + stuN + nt),
tir
m viri si
ijLMiV
«J[o
Ij
Hence Lemma 4.3.14 implies
dlkJ = x(n)g = x(d)9 = xiy)gNow g(z) = Oiy'"'^) by Lemma 4.3.3, so that g{z)e%(N, x) by Theorem 2.1.4. Since/= (  lfg\,,o)ff by (1), we get/(z)e^fc(Ar, x) by Lemma 4.3.2(2). This implies the first half of the assertion. Next assume that L{s;f) is absolutely convergent at
§4.4. Aiz) and ti{z)
129
s = kS with ^ > 0. Put
c„= t \^J
(n^ll
then c < "'"'(_^EJ«J'»"'"') so that c„ = 0(n''"'). By Lemma 4.3.3, Xr=oC„e"^''"'' is convergent and Z"=oC„e^'"^ = 0(j'»*''') (y^O). Since a. = c,c„^„ we see
\f{z)a„\^(le''y)(yc„e''A, \n=l
SO that/(z) = 0(y ''•^^). Therefore/(z) is a cusp form by Theorem 2.1.4, and so is 9iz). • We note that when AT is not a prime number, /(z) and g{z) are not necessarily cusp forms even if a© = ^o = 0
§4.4. A{z) and f/(z) In this section, we give the proof of Theorem 4.1.7(2) which asserts that A{z) has no zeros on H. First we define the function ri(z) on H by ^(Z) = e«'^/12 ^ ( l  e ^ " ' " ^ ) .
Since J] *= ^ e^"'"^ converges absolutely and uniformly on any compact subset of H, so does the above infinite product. We call ri(z) the Dedekind rjfunction. Theorem 4.4.1. The function ri(z) satisfies the transformation formula rji\/z)Hz/iY''rj(z), Proof We follow Weil's proof ([Weil 4]) which reduces a transformation formula for a holomorphic function on H to a functional equation of the corresponding Dirichlet series. Since rj{z) has no zeros on H, the assertion is equivalent to saying that (4.4.1)
log^(l/z)=logfy(z)+ilog(z/0.
By definition, we have \ogrj(z) = niz/12
f m^e^""'"""'.
130
4. Modular Groups and Modular Forms
Put f{z)=
f
m'e^'''""'' =
m,n=
niz/\2\ogrj(zl
I
then/(z) satisfies (4.3.7) and
L(s'j)=as)C{s+iy As in the previous section, put A{s;f) = (2n)^r(s)Lis;f) and A{s) = n^''r(s/2)i:{sl Since A{s;f)=iAis)Ais^l), Theorem 3.2.2 imphes that A(s;f) can be analytically continued to the whole splane and satisfies the functional equation A{s;f) = A(s'J). Moreover f ( — 1) = —1/12, and
is entire and bounded on any vertical strip. We note here that yl(s)4+ ^ S
1—5
is bounded on any vertical strip by a similar argument as in the proof of Theorem 4.3.5. Now m)=^
j
y'A{s;f)ds,
and we can change the integral paths from Re(5) = 2 to Re(5) =—2. By the holomorphy of (4.4.2), y~^A{s;f) has poles only at s = l , 0, —1 with the residues 7r/12y, ilog>', —Try/12, respectively. Hence m) = ^ =^.
J I
y'Ais;f)ds + fA{s;f)ds
+
n/12y+\\ogyny/n nn2y+\logyny/l2
= / (  l A » + 7r/12>'+log>'7r3;/12, by the functional equation for yl(s;/). Therefore (4.4.1) holds on the imaginary axis. Since both sides are holomorphic on H, (4.4.1) is satisfied everywhere on H. D
§4.5. Hecke Algebras of Modular Groups
131
Now we consider rj^'^iz). Since (4.4.3)
^2^(z) = e^"'^ 0 ( 1  ^^"''")^^ n=l
we have (4.4.4)
rj'\zhl) = rj'\zl
As r(l) is generated by
rj'\l/z)
and
= z''rj'^(z).
and rj^'^^iz) vanishes at oo by (4.4.3),
fj'\z)e^,2irWy Since dim 5^12(^(1)) = 1» we see ^^"^(z) = czl (z) with a constant c. Comparing the first Fourier coefficients of ^^"^(z) and zl(z), we get c= 1, hence A(z) = rj'^{z). Thus we obtain the following Theorem 4.4.2. The function A (z) has no zeros on H and has the infinite product expansion
A(z) = e^'''' f[ (le^'^'^'f^ n=l
which converges absolutely and uniformly on any compact subset o/H. Theorem 4.4.2 also implies the integrality of the Fourier coefficients T„ of A (z). The explicit forms of A (z) and ri(z) are quite useful to calculate modular forms. We can construct cusp forms of small level by using A{z) and rj(z). For example, we have ([Shimura], pp. 4950): 1° If N is 2, 3, 5, or 11, and k = 24/(N+1), then 6^kiroiN)) is onedimensional and is generated by (A {z)A (Nz))^'^^^^^ = (r]iz)rj(Nz))\ T If N is 2, 3, 4, 6 or 12, and /c= 12/AT, then ^fc(r(iV)) is also onedimensional and is generated by (J(z))^/^ = f/2'^(z).
§4.5. Hecke Algebras of Modular Groups Hecke operators play important roles in the study of modular forms. In this section, we explain, following Shimura, the relation between Fourier coefficients of modular forms and Hecke operatorsfirstobtained by Hecke. We begin by studying the structure of the Hecke algebras of modular groups. For a Fuchsian group T, we put f={QeGLim\gTQ^^T\
132
4. Modular Groups and Modular Forms
Lemma 4.5.1. For any modular group F, we have
Proof. Since FwF{l), we have only to prove the lemma for F = F{i) by Lemma 2.7.1(3). Let aeIR'' GL^CQ). Take ceW" so that j5 = caeM2(Z). Then oi~^F(l)(x=^p^F(l)p. Put m=det(j5), then m)5"^ eMjC/). For any yer(m), we see imr')yP^{mr')P
( = \^
^1)
mod m,
so that P~^yPeM2iZX and P~^F{m)P c= F. Therefore we have 6cF(x^nF =
pFp^nF^Fim).
"[: a
Since [ r : r ( m ) ] < o o , a belongs to T. Conversely, assume a =
eT. By
Corollary 1.5.5, the cusps aoo = a/c and aO = b/d of(xF(x~^ are also cusps of T. Since r «'a~ ^r'a, the points 'aoo = a/b and 'aO = c/d are also cusps of F. As any cusp of F is an element of Q u {oo}, the ratios of a, b, c, d are all rational numbers, and therefore a G R ^ • GL^ (O). D We define the subsemigroups AQ(N) and ^ J W of GL2 (Q) by (4.5.1)
^oW = 
^
^ \eM2{I.)\c = 0modN,{a,N)=l
adbc>oi,
and (4.5.2)
JJ(N) =  r
MeM2(Z) c = OmodAr,(^,iV)=l,arfftc>oi.
We see easily that (fa
bl
\c =
We are going to study Hecke algebras ^(Fo{N\ simply write
OmodN,{adbc,N)=U] AQ{N))
(4.5.4)
^(iV) = ^(ro(iV),zlo(Ar)),
(4.5.5)
^*(iV) = ^(ro(iV),zlS(iV)).
and ^{FQ(N\
Al(N)). We
Lemma 4.5.2. For any CL£AQ(N) {resp. zlJ(N)), there exist positive integers I and m such that /m, (/, iV)= 1 and
'o(N)aro(JV)=r„(JV)P
l\ro(N)
{0 n, ro(N)).
(resp. r„(JV)aro(N)=ro(JV) „ 77ie paJr (/, m) is uniquely determined by a.
133
§4.5. Hecke Algebras of Modular Groups
Proof. Put L=
M,rGz,
Lo =   ^ ^ " j  u , t ; G z  .
Then M2(/) acts on L by left multiplication and
CCLQ
a
LQ.
Put n = det(a), then
[L: aLo] = [ L : aL] [aL: aLo] = nAT. By the fundamental theorem of abehan groups, there exists a basis {wi, W2} of L such that (xLQ = Z{aw^)@Z{bw2) with positive integers a and b {ab = nN, a\b). Since OCLQ C/Z tL for any divisor t(> 1) of AT, we see that (a, Ar)= 1 and N\b. Then Zw^ © Z(Nw2) is a unique submodule of L of index iV containing OCLQ, and therefore, Lo = Zwi©Z(Nw2). Moreover if we put l = a and m = b/N, then Z(/wi)© Z(mw2) is a unique submodule of L of index n containing OLLQ. Therefore we get aL = Z(/wi)©Z(mw2). We define two elements 7i, 72 of M2(Z) by
Then
(wi,W2) = M ^ LL Ihu
N 0P
1 N = ('^i''"^2)72.
yi
We may assume det(7i) = det(72)=l by taking —W2 in place of W2, if necessary. Since Wj GLQ and al ^ UaLo, = det(7we see Ti, y2^^o(^) Therefore we obtain
ro(iV)aro(N)=ro(iV){ 0 : ] 'KoW.
Since
L/aLZ//Z©Z/mZ,
(4.5.6)
the pair (/, m) is uniquely determined by the fundamental theorem of abehan groups. A similar argument is also appUcable to /dg(iV). D We note by (4.5.6), (4.5.7)
ifl\m and {lm,N)=l
then UN)^^
ml^'^^^^^^^^f 0
^l^^^^^'
Theorem 4.5.3. (1) Both Hecke algebras M{N) and ^*{N) are commutative. (2) For any element ro{N)(xro{N) of M{N)KJ^*{N\ both ro{N)\roiN)oiro{N) and rQ{N)(xrQ{N)/rQ(N) have a common set of representatives.
134
4. Modular Groups and Modular Forms
Proof. For an element a=
ezloW (resp. ^oWX we put (iV) (resp. zl?(iV)).
Then the correspondence "ai*a'" satisfies the three conditions of Theorem 2.7.8. In fact, conditions (i) and (ii) are obvious by definition, and (iii) is proved by Lemma 4.5.2 and (i). Therefore ^(N) and ^*(N) are commutative. D Now let X be a Dirichlet character mod N. We put (4.5.8)
Z(«) = Z(^ for aJ"
%AoiN).
Then x is an extension of the character of ToCiV) defined by (4.3.2) to AQ(N). TO let Hecke algebra ^(N) act on %iN, /), we have to verify that x satisfies (2.8.1), or if oiyoL'eroiN) (yeToiNl
ae^oW)
then x(a7a"') = X(7).
In fact, by Lemma 4.5.2, we may assume oc = \
(/m, (/, Ar) = l). Put
U 3'^"
(AT), and y' = ocyoc ^ If 7' e ro(iV), then bl = 0 mod m and
_r a Y IcNm/l
bl/ml d J'
so that x(y) — xiy')' Therefore by Theorem 2.8.1, ^(N) acts on %{N,x) More precisely, if ro(N)aro(N) = Uro(iV)a, V
is an element of ^(N), then (4.5.9)
/I ro(iV)aro(iV) = det (a)*/^"'^/(a,)/,a,
(/e ^,(N, /)).
V
Next we extend the character x of rQ(N) to ^o(^) by (4.5.10)
X*io^) = Xid)
(oi=\''
%At{N)
Similarly as above, this extension also satisfies (2.8.1), and 3t*{N} also acts on %(N, X); namely, if ro(N)«ro(N)e^*(N) and ro(Af)aro(Ar)=Uro(N)a„ V
then (4.5.11)
/ro(N)aro(iV) = det(a)*/2^Xx*(av)/Lav
ife^iN^x))
§4.5. Hecke Algebras of Modular Groups
135
We note that when ro(N)oLro(N) is contained in ^(N)n^*(Nl the action of ro(iV)aro(iV) as an element of ^(AT) does not necessarily coincide with the action as an element of ^*(iV). The difference will be clarified in Theorem 4.5.4. We only note here that (4.5.12)
Ao(N)^AUN)
by
"a h^ w^ ^ aco^",
and (4.5.13)
x*icoN'oico^) = m
(aeJo(^)).
We define the elements T(n) and T{1 m) of ^(N) by Til m) = ro(iV)P
1\UN)
(/m, (/, iV)= 1),
(4.5.14) T(n)=
X
ro(iV)aro(iV),
det («) = «
where the summation is taken over all double cosets ro(N)aro(N) in ^(N) with det(a) = n. Similarly we put ,\ro(N) •.,..[: :]r„
r*(m,/) = ro(N) „
(l\m,(l,N)=l),
(4.5.15)
r*(n)= X r„(N)ar„(Ar), det(a) = /i
where the summation is taken over all double cosets ro{N)arQ{N) in ^*{N) with det(a) = n. When we emphasize the level iV, we write ^r(n),^r(/,m),^r*(n),
and
^r*(m,/),
respectively. Now Lemma 4.5.2 implies T(n)= X
niml
Im — n
(4.5.16) T*(n)= X r*(m,/), lm = n
and, in particular, for a prime number p we have (4.5.17)
T{p)=T(h
p\
T*(p)=T*(p,
1).
Moreover, if (n, iV)= 1, then
r(n,n)=ro(iV)[j ^jro(iV)=ro(iV)^J ^ ] ,
136
4. Modular Groups and Modular Forms
SO that
T(n, n)T(l m)=T{nl nm), (4.5.18) r*(n, n)T*{m, l)=T*(nm, nl). Theorem 4.5.4. (1) For any element f(z) of%(N, xX ^^ have /  r * ( m , 0 = z(/m)(/r(/,m)) ,/ (/m,JV)=l, /r*(n) = z(n)(/7'(n))
,/
(n,N)=l.
(2) T{1, m) and T*{m, I) (resp. T(n) and T*(n)) are the mutual adjoint operators with respect to the Petersson inner product on £^ki^, x)(3) y'ii(JV, x) has a basis consisting of common eigenfunctions of all T(n) and T{l,m)((n,N)=l,ilm,N)=l). Proof (1): Assume (Im, N) = 1. Put
By (4.5.7), we have
r and X*(a) = x(det(a))x(a) for any element a of ^o(^) '^ ^o (^) Hence we see /r*(m,/) = (/m)*/2^Xx*(«v)(Aav) = Z(M(/m)'^/2iXx(av)(/l.av) V
= X(lm)(f\T(hm)), Moreover, by (4.5.16) we get that/ T*(n) = x(n)(f\ T(n)) if (n, N) = 1. (2): For an element a =
, of GL2 (UX we put
Ic ^J , r a' = det(a)a"^ =
lc
d
bl
aj
The correspondence "an^a"' is an antiisomorphism of ^o(^) onto ^o(^) By Theorem 4.5.3(2), we can take a complete set of representatives {a^} so that (4.5.19)
ro{N)\^
^lro(iV) = IJro(iV)a, = IJa,ro(iV).
§4.5. Hecke Algebras of Modular Groups
137
Then
ro{N)\^ 5^1n(^) = LlnWa;. Let T* be the adjoint operator of T(l m) with respect to the Petersson inner product. Since xM = Z*(a'v)> we see by Theorem 2.8.2(1) (4.5.20) /I T* = (Imf^"' ^ z K ) {f\M
V
= /r*(m,/). Therefore T(l,m) and T*{mJ) are adjoint to each other. By (4.5.16), so are T(n) and T*(n). (3): By (1) and (2), if (/m, N) = 1, then T(n) and T(l m) are normal operators on 6^^(N,x)' Since ^{N) is commutative, they are simultaneously diagonalizable. D As was shown in Lemma 4.3.2, ^f,{N,x) is isomorphic to ^^(N,x) by the correspondence "/i^/lfcCo^". Theorem 4.5.5. The following diagram is commutative: ^AN,x)
^'"""^''•^"•"'"
. %(N,x) (ON
%(N,x)
^' ai L/(xL :ii L/ct2 L ^ r coincides with 7(1, p^""^) or T{p,p^). Hence aT{hp^^')
r(p)r(i,p^) = , _a'nip^^') , _ , ,^,
+ bT(p,f)
ifpJ^N, ifp\N,
with integers a, b and a'. By Lemma 4.5.6(1) and the definition of the multiplicatuon, we get a = a' = 1. Comparing the degrees of both sides using Lemma 4.5.6(2) and (2.7.7), we obtain [ P
ife>\.
This proves (1). When p\N, we see T(p^) = 7(1, p% so that (2) is nothing but the first assertion. Now assume that (p, AT) = 1. We prove the assertion by induction on e. First let e = 1. Since T{p) = 7(1, p), we see by (1) that 7(p)7(p)=7(l,p2) + (p+l)7(p,p) = 7(p2) + p7(p,p).
§4.5. Hecke Algebras of Modular Groups
141
Next assume e ^ 2 and the assertion is true for all positive integers less than e. Applying (4.5.16) and (4.5.18) to T{p% we see
np)T(p')=np){n\,p') + np,p)T{p'^)} =T(i,p'^')+pT{p,p)ni,p'') = nup'^') +
+
np,p)np)T(p'')
pT(p,p)T(ip''')
+ T(p,p){T(p'') + pT(p,p)T{p'')} =
T{ip'*')+T{p,p)nip'') +pnp,p){ni,p'')+T{p,p)T(p'')}
= T(p''') + pT{p,p)T{p''). Here we understand r(p^) = 0 if e < 0. This is what we wanted to show.
D
Lemma 4,5.8. (1) //(/m, I'm') = 1, then T(lm)T{l',m') = T(li',mm'). (2) //(m,n) = 1, then T(m)Tin) = T(mn). Proof. (1): Let 7 = ro(Ar)aro(iV) be any constituent of T(lm)T(l',m'\ can write a = a^aj with a^ 6 T(l, m), 0L2 6 T{1'\ m'). Put
"11 .^^Jl
then we
n\ J
then LlaL^L^ZIlZ@TlmT, OL^LjaL  LI0L2L ^ Z/rZ e Z/m'Z. Since (/m, /'m') = 1, we get L/aL^Z///'ZeZ/mm'Z, namely r = T(ll', mm'). Therefore T(U m)T(l', m') = aT(U', mm') with some integer a. We are going to prove a = \. Let Vll' 0 1 ^ = L0 rnm'\
^^^^{0 ^]^o(iV)=LJ^oWa„ and We may assume
142
4. Modular Groups and Modular Forms
Now we assume ro(iV)a = ro(iV)a^j5v» ^^^ l^t a = ya^j^v with yeFQiN). Put M =
{lu,v^\u,veZ}
and define the submodule MQ of M by MO = {[NM,I;]M, i?eZ}.
We let M2(/) act on M by the right multipHcation of matrices. Then {Mp, + Mp,)/Mp,

MpJiMP.nMM
Since the lefthand side is a subgroup of M/MP^, the order is a divisor of det(jSi) = /'m'. On the other hand, the righthand side is the quotient group of MP^/M(x; hence the order is a divisor of det(a)/det(j5v) = /m. Since (/m, I'm') = 1, we get Mpi — Mp^. Therefore there exists an element yieGLjiZ) such that yiPi=Pv. Since det(^i) = det(j5y), we get y^eSLjil). A similar argument for MQ in place of M implies MQP^ = MQP^, SO that MoVi = Mo, namely 7 i e r o ( N ) . Since j?^ is a representative of left ro(N)cosets, we get yi = 1 and Py = Pi. Similarly we obtain a^ = a j . Consequently a — 1. The second assertion easily follows from (1). D Theorem 4.5.9. The Hecke algebra 0t (N) is the polynomial ring over Z generated by T(p\ T{p, p) with all prime numbers p prime to N, and T(q) with prime numbers q dividing N; namely, ^(N) = ZlT(plT(p,p\Tiq)
\
pJ^N,q\Nl
Proof. By Theorem 4.5.3, Lemma 4.5.7 and Lemma 4.5.8, ^(N) is commutative and any element of ^(AT) can be expressed as a polynomial with rational integral coefficients in r(p), T(p,p) and T{q). Moreover it can be easily verified by Lemma 4.5.7 and Lemma 4.5.8 that T(p\ T(p, p) and T{q) are algebraically independent over Q. D Take two positive integers /, m so that l\m and (/, AT) = 1. Let m/l = YlpP^ ^^ the prime decomposition. By Lemma 4.5.8, we see (4.5.23)
Tam)= T(ll)T(hm/l)= T(lJ)l\T(lpn. p
where p is taken over prime divisors of m/l, so that we obtain (4.5.24) (4.5.25)
T(lm)
= Uro{N)l''^
r(n) = [Jro(iV)h
J l (ad = lm,0 ^b < dAa,b,d)
= l),
M {ad = n,0 ^ b < d^a, N) = 1),
by Lemma 4.5.6. Therefore we can express the action of T(l m) and T{n) on ^ki^^ X) explicitly. In particular, for/(z)e^fc(iV, %), (4.5.26)
(f\T(nmz)
= n''
^ 00 throughout this section. For a positive integer /, we put (4.6.1)
5, = V
^lecmu).
Then for a function f(z) on H, (4.6.2)
filz) = l'''{f\,5,)(z).
Lemma 4.6.1. lff{z)e^f^{N, %), then for any positive integer /, f(lz) =
l''i\f\,5,)(z)e%(lN,x).
Moreover, iff(z) is a cusp form, so isf(lz). Proof Let f(z)e0).
By (i), (/, m) = 1, so that / = 1 and m> 1. Since
we get Let y be an element of ro(N) such that ocyoc'^ ^FoiN). Since det(a)a^ = r _ ^
"Mezlo(iV),
we see that det(a)a7a"^ ezlo(^) Thus we can take elements 73, y4^erQ{N) so that (4.6.4)
ciet(a)73aya"^74 = ^
^
(u\v,
u,v>0)
by Lemma 4.5.2. We see uv = det(a)^ by taking the determinants of both sides. If u = v, then ocyoc ~ ^ = 7^ ^74 ^ e ^o(^)? which contradicts the assumption on y. Therefore h — v/u>i. Acting (4.6.4) on/(z), we get
cf(z) =f(z/h)
(c = h'^'h{y3)xiy)xiy4) ^ o).
Let/(z) = YJ^^O «(")^^"'"^ be the Fourier expansion. Then a(n) = C^a{n/h) = C'a(n/h') for any positive integer t. Therefore we get a(n) = 0 if n > 0 . Since /c>0, we obtain f{z) = 0. D Theorem 4.6.4. Let I be a positive integer. If a holomorphic function f(z) on H satisfies the two conditions: (i)/(z+l)=/(z),
(ii)/(/z)e^,(iV,;f),
156
4. Modular Groups and Modular Forms
then (1) iflm^\N,thenf{z)G%{Nllx)'. (2) iflm^)(N,thenf(z) = Q. Here m^ is the conductor of x Furthermore iff(lz) is a cusp form, then so isf(z). Proof We may assume that / is a prime number. First let us show/(z)e^fc(iV, x)Put^(z)=/(/z)and
^' = { U y^oWb.Omod/}. Take an element y = \ ^ I of T'. Since \_cN d _\
we see or
f\Mr'ySi
= xid)f\,s,.
This implies/I ^y = x(d)f Let F" be the group generated by F' and
. Since
[r":r]^/ and
we see that F" =
FQ{N)
or
TQCAT) is
4i:]
generated by
over F'. Therefore by (i),
f(z) belongs to %(N, x) Next suppose / is prime to N. Then f(z) = 0 by taking oc = Si in Lemma 4.6.3. Now we assume l\N. For any element
^'"[cAT
Jero(N),wesee = X(d)f,
^^'[cN/l and in particular,
^^'[N/I ij Therefore, if we put m"r 1
C
' = [i i}\_N/l 1 ][: ;]• then/I^7 = / Let n be any integer such that niV//+l ^ O m o d / .
§4.6. Primitive Forms
157
Then we can take an integer m so that (4.6.5)
nil + mN/l) \m = n\ m{nN/l + 1) = 0 mod /.
Then
and hence Ay
=
Xi^+nN/l)f.
Therefore if/(z) ^ 0, then xi^ + nN/l) = 1 for any integer n satisfying (nN/l + 1, /) = 1. This impHes that x is defined mod AT//, and N is divisible by Im^. This impHes (2). Assume lm^\N. Since ro(N//) is generated by f{z)G%{N/l,
over ro(N), we get
x)' The last assertion is obvious.
D
Lemma 4.6.5. Let f{z) = Yj^=o^n^^'''"^ be an element of%{N, x) For a positive integer L, we put 6f(z)=
Z
a^'^
in,L)=l
Then g{z)e^k{M, x) ^^th
M = NIIP p\L p\N
UP'p\L pJ^N
Furthermore iff(z) is a cusp form, so is g{z). Proof. We have only to prove the assertion when L is a prime number p. Put N' = N or N' = pN according as p AT or not. Since pi AT', we have (4.6.6)
^o(N')\l l\ro{N')=Ur,{N')\l L^
Pj
m=o
""]
{_yj p J
by Lemma 4.5.6. Since ^^(iV, x) ^ %(N\ x\ we g e t /  '''T{p)e^j,iN\ we see
xl By (4.6.6),
(/I ^•r(p))(z) = p' t'^n ' Z e''"^^*"''" M=0
m=0
^2ninz n= 0
SO that 00
{f\'''T(pMpz)=
X a„,e"''"''\ n=0
which belongs to %{N'p, x) by Lemma 4.6.1. Therefore we obtain g{z) =f{z){f\ The last assertion is obvious.
'^'T{p)){pz)e^,{N'p,
x\ D
4. Modular Groups and Modular Forms
158
Lemma 4.6.6. Let x be a Dirichlet character mod iV, / a positive integer, and p a prime number prime to I. Put M = IN. Then the following diagrams are commutative: roipN) (1)
%(pN.x)
%(pM,x)
[I p]'«^^^
'^kiN.x)
il 'M^
ro(pM)\ \ro(.M) ^^ > %(M,x)
Here the vertical arrows indicate the natural embeddings. UpN)\. (2)
r„(jv) %(N, X)
'^,(PN,X)
''o(pM)r
lr„(M)
%(PM,X)
%(M, X)
Similar results as above hold for cusp forms. Proof. (1): By assumption, pliV if and only if p\M. Therefore the assertion is obvious from Lemma 4.5.11. (2): Let/(z) be an element of %{pN,x\ and put ^ =/U(5,. We put
^ ^ I p\ [ p
ifplM otherwise,
and yv=iQ
JJ
(Ogvgpi).
\i d = p, then take an element y,, of FQ^M) as in Lemma 4.5.11. Then (4.6.7)
r„(pM)^J
^ J r o ( M ) = Uro(pM)^J
°Jy„
SO that
(4.6.8) p.r„(pM)[J ^]ro(M)= E^,U[; 0]y.
Since (O^v^p1),
159
§4.6. Primitive Forms
and
modp, ^lyp^i
1 [:
modiV,
provided pJfN,wG see
by Lemma 4.5.11. Therefore it follows from (4.6.8) that
^ro(pM)^J ^]ro(M) = (^/ro(pJV)^J P ] ^ ^ ^ ^ ^
Si.
D
Ininz Lemma 4.6.7. Let l(>\)he a positive squarefree integer, andf{z) = ^^=0 ^«^ an element of%{N, x) V^n = ^fa^ ^'' ^ prime to /, then we can express
p\l
where p runs over all prime factors of I. Furthermore ifl\N, then we can take gp(z) from %{Nl, x) V f(^) is a cusp form, all gp(z) can be taken as cusp forms. Proof We prove the assertion by induction on the number of the prime factors of /. First suppose / is prime. Then g{z) = f{z/l) satisfies the conditions of Theorem 4.6.4, so that g(z)e%(N/l, x) or /(^) = 9(z) = 0 according as lm^\N or not. Consequently g(z) e %(Nl, x) and fiz) = gilzy Next assume that / is a composite number and the assertion is true for any proper divisor of /. Let p be a prime factor of / and /' = l/p. Put h(z)=
Y.
»ne"'"'.
(>l.p)=l
Then h{z)€%{,Np^, %) by Lemma 4.6.5. Put
f(z)h(z)= f b„e^''". n= 0
Then if (n, p) = 1, then h„ = 0. Put gp{z)=f{z/p)h(z/p), then gp(z)e^j^{Np, x) by the induction assumption, and (4.6.9)
f{z) = gp(pz) + h(z).
Moreover h(z) satisfies the assumption of the lemma with Np^ and /' in place of AT and /, respectively. Therefore we can find gq{z)e%{Np^r^,x) (= %iNl^,x)) for
160
4. Modular Groups and Modular Forms
each prime factor ^ of /' satisfying
Hz) = Z gM^y q\l'
This combined with (4.6.9) imphes the first half of the assertion. It is obvious from the above proof and Lemma 4.6.5 that if /iV, then we can take gp(z) from ^fc(N/, x). It is also obvious that we can take gp{z) from cusp forms when f{z) is a cusp form. D The following theorem is a generalization of Theorem 4.6.4. Theorem 4.6.8. Let I be a positive integer, andf(z) = ^^=0 ««^^"'"^ ^^ element of %(N, x)' Let m^ he the conductor of x Assume a„ = Ofor all n prime to I. (1) If{lN/m,) = l,thenf(z) = 0. (2) //(/, N/m^) 7^ 1, then there exist fp{z)e^j^(Nlp, x)for all prime factors p of (/, N/m^) such that p\il,N/m,)
Iff(z) is a cusp form, then we can take fp(z) from cusp forms. Proof We may assume that / is squarefree. We prove the assertion by induction on the number of the factors of/. When / is a prime number, the assertion is nothing but Theorem 4.6.4 for/(z//). Next assume that the assertion is true for any proper divisor of /. Let p be a prime factor of /, and put /' = I/p. We put
(4.6.10)
h(z)=
X
ay'"\
(n,/')^l
(4.6.11)
X a„e Ininz
g(z)=f(z)h(z)=
{n,l')=l
We see h{z) and g(z) belong to %(Nl'^, i) by Lemma 4.6.5. If (n,p) = 1, then the nth Fourier coefficient of g(z) vanishes. First assume pm^J( N, then pm^^XNl'^ and therefore g(z) = 0 by Theorem 4.6.4, namely, f(z) = h(z)=
X
a„e2.
This implies that f(z) satisfies the assumption for /' in place of /, and hence, the assertion is true by the induction assumption. Next assume pm^ \ N and put gpiz) = g{z/p). Then gp(z)e%{Nr^/p,x) by Theorem4.6.4. Put d = pl p^)(N. Take elements yA^^^^d) ofToiNr^/p) so that
if p^\N, or d = p ii
roiNn[, 3^o(ivrVP)=Uro(N/'^)[; J]
§4,6. Primitive Forms
161
(see Lemma 4.5.11). Then we see
d
v= 0
Thus we obtain (4.6.12)
g(z) = g,(pz) = p{d + l)'(g\ro(NnV^
^1ro(M'Vp))(pz).
Put
/,(z) = p(e a primitive form Lemma 4.5.15(1), f\T{n) = aJ for all n.
of ^?(iV,x). Then
by
Moreover since/(z) is an eigenfunction of r*(«), and T*(n) is the adjoint operator of T(n) with respect to the Petersson inner product by Theorem 4.5.4(2), we get f\T*(n)
= dJ,
Furthermore Theorem 4.5.4(1) impHes that (4.6.17)
a, = man
if
{n.N)=\.
Moreover we see by Theorem 4.5.5 (4^18)
{f\k(^fi)\T{n)=d„(f\k(^N\ ( / L c o ^ ) m « ) = «„(/kcoJ,
for all positive integers n.
166
4. Modular Groups and Modular Forms
Theorem 4.6.15. (1) By the action ofcOf^, ^ki^^ X) ^^d S^l(N, y) are isomorphic to ^?(Ar, X) and ^1{N, x\ respectively. (2) If f(z) is a primitive form of ^t(N, /), then fp(z) is a primitive form of 5^«(iV, X) and f\kCON = cfp(z) (ceC). Proof (1): First let us prove that co^ maps ^i(N, x) into ^{{N, x). We have only to show that if f(z)==(h \M(z) then f\k(Oj^e^l{N,
ih{z)E^,(M, x\ M ^ N, IM\N),
x) Put /' = N/IM. Since
we see = (h \kCOM) \k^l' '
Since h\kO)j^^6^k(M, xl we get f\,a)j,e 6^UN, x) Next assume/(z)e 5^]?(iV, /). ^^Q sec if\kCON.h) = (filfh\,oy^) =0 for all h{z)e^liN, x) by Theorem 2.8.2. Therefore /fc(w^e;, on both sides, we obtain by Theorem 4.6.16(3), (4.6.34)
a,^ = z;(9)9»^.
(3): By assumption, q divides N/q, so that
ro(N)[J °lr„(Jv/q) = ro(N)P ^lro(N) by Lemma 4.5.6 and Lemma 4.5.11. A similar argument as in the proof of (2) combined with this imphes f\T(q)=f\ro(N)[^^
^]ro{N/q) = 0.
This proves (3).
•
We restate (4.6.31) and (4.6.33) as Corollary 4.6.18. Under the notation and assumptions in Theorem 4.6.16 and Theorem 4.6.17, we have: (\) If N^ =^ m^, then fin,
= eg,,
c^N'l^W(x,)x,{\)d\,
with a primitive form gJz) of6^^{N, XaL)
{N,=q%
§4.6. Primitive Forms
173
(2) If N^ = q and m^ = 1, then flk^q =cf,
c=
q^^^\,
and %
=Xq(Q)ag
We are going to prove that two primitive forms are equal if almost all a^ are equal. More precisely, Theorem 4.6.19. Let f(z) = ^^=1 ^n^^""" ^^ « primitive form of ^?(iV, x\ and g(z) = ^^^ J hne^""'"^ an element of ^fc(M, X). If g{z) is a common eigenfunction of M(M) u ^*(M), 6i = 1, and a„ = b„for all n prime to some integer L, then N = M and f{z) = g(z). Proof We may assume that L is a common multiple of N and M. If p is a prime number prime to L, then we see Z(P)P*~'
=alap2
=
HP)P''
by Lemma 4.5.7(2), so that x(p) = Hp) Therefore x(n) = X(n) for all integers n prime to L. Let us prove M\N. By Corollary 4.3.7, we have .4 ^ 3^.
^iv(^; / ) ^ ^M(S;^)
Aj^(ks;f\k(o^) AMiks;g\kCOM)'
Since L(s; f) and L{s\ g) have Euler products by Theorem 4.5.16, we see
(4.6.36)
HP)P'~'~'' ^Ms\f) ^ = i ,^ rr;, / ^ fV7,sn r n ^r ^^P~'^  M Z(P)P' ,fc  1  2s
on Re(s) > / c / 2 + 1 . Since the righthand side is a meromorphic function on the whole splane, (4.6.36) holds on the whole 5plane. On the other hand, g\T(n) = b„g, and g{z) is a common eigenfunction of T*{n) by assumption. Since T*{n) is the adjoint operator of r(n), we see g\T*{n) = b„g, and i9\k(^M)\T(n) = b„(g\k(OM)
by Theorem 4.5.5. Therefore Theorem 4.5.16 implies that L{s; g\kCOM) also has an Euler product. This combined with Theorem 4.6.15(2) implies
with a constant c. Thus by (4.6.35), we get
=c( fNi
HAfu'Mi^i^Mi^::!:!.
174
4. Modular Groups and Modular Forms
We denote by Mp and Np the pcomponents of M and AT, respectively. Then we see, for any prime factor p of L,
with a constant c^ by Lemma 3.2.1. Let u and v be the degrees of ^app'hx{p)p''~'' and l6pp^H2(p)p''i2^ as polynomials of p~^ respectively. Then 0 ^u,v each case separately. From now on we put
^2. We are going to discuss
X = p~*
and
M,/Np = p\ r If M = i;, then Mp = Np by (4.6.38). 2° Suppose M = 1 and t; = 0. We can rewrite (4.6.38) into Cp{\ QpX) = x^{l  dpp~^x~^).
Up ^ 0,
so that \ap\^ = p*; which contradicts Theorem 4.6.17. If w = 0 and v = 1, then we see Np = pMp. 3° Suppose u = 2 and t; = 0. Then (4.6.38) can be rewritten into (4.6.39)
Cp{\apX + x{p)p''x^) = x^lapp'x''+
x{p)P~'~'x^),
Comparing the degrees of both sides, we get e = 2 and CpXiP)p'''^ = h
Cp =
xiP)P~^~\
so that lz(p)P = P^; which is a contradiction. If M = 0 and v = 2, then we see easily e =  2 , so that Np = p^Mp. 4° Suppose M = 2 and v = 1. We rewrite (4.6.38) into ..^.m
^ lapX\xip)p'~'x^ ^ 1fepX
_
l  V ' ^ x  ^ + x(rt/7^^x" 1V ^
Put r = x~^ Then the absolute values of the roots of
t'apt^x{p)p''=0 are p^^~^^/^ by (4.5.41) and the absolute values of the roots of
X(p)p''t'apph+l=x{p)p''{t'appt
+ Xip)p'^') = 0
are pMp. Consequently, in any case we obtain that M AT, and / is induced by X. Therefore Theorem 4.6.12 implies that f(z) = g{z), and so AT = M. D This theorem combined with Corollary 4.6.14 implies
§4.7. Dirichlet LFunctions and Modular Forms
175
Corollary 4.6.20. Let f{z) be a nonzero element of ^^(N, /). Iff\ T(n) = a„ffor all integers nprime to N, then there uniquely exist a divisor M ofN with my^\M and a primitive form g(z) of^^(M, x) such that g\T{n) = UnQ for all n prime to N. Moreover ^, . . n\\ IH^\\T\ f{z)eig{lz) I /MiV>. Corollary 4.6.21. Let f{z) = Xr=i a^e^""^' be a nonzero element of6e^(N, x). Then f(z) is a constant multiple of a primitive form of 9'^(N, x) if and only if f(z) is a common eigenfunction of M(N) and ^*(N). By Theorem 4.5.5 and 4.5.16, we see Corollary 4.6.22. Let f(z) = Y.^= ^ a„e^'''"' be an element of Sf^iN, x) ^ith a^ = 1, and put
00
(/i,co^)(z) = c £ v''"•"^
(fci = 1).
Thenf(z) is a primitive form o/yjj (iV, x) if and only ifL(s;f) and L{s;f\kO)ff) have the following Euler products:
ite/) = n(i«pp"^+z(p)p'"'"'rs p
L{s;f\kCOr,) =
cYl(lb,p^^Xip)p'''r'
§4.7. Dirichlet LFunctions and Modular Forms In §4.3, we proved that integral forms are characterized by functional equations of the associated Dirichlet series. It is natural to ask what such Dirichlet series are. We showed in §4.1 that the Dirichlet series associated with Eisenstein series with respect to SL2(Z) are nothing but the products of two Riemann zetafunctions. In this section, we are going to generalize this and prove that elements of J^^. {N, x) correspond to the products of two Dirichlet Lfunctions. This was proved by Hecke by constructing modular forms expHcitly using Eisenstein series. Here we start from Dirichlet series and construct modular forms as an application of Weil's theorem. We shall discuss on Eisenstein series again in Chapter 7. Now let Xi and Xi t>e Dirichlet characters mod Mj and mod Mj, respectively. Put (/)(s) = L{s,Xi)L{s  fe + 1, xi)^ for a positive integer k. Then
p 00
176
4. Modular Groups and Modular Forms
on Re(s) > k. Here (4.7.1)
a„=
I
xAn/d)x2{d)d'K
0 A)^X(m)iA(M) w =
,.^^W = x('w)iA(M)—^.
Therefore/(z)e^fc(M, x) by Theorem 4.3.15. Next we consider case (i). Then (4.7.13)
A{s,il/)A(slil/)
where 1 (4.7.14)
if
^(_i)=_i^
if
iA(l)=l.
Ks)= i ^
By the functional equation of A(s, ij/), we obtain the functional equation (4.7.15)
^ ^ ( 5 ; / ^) =  C^A^(2  s;  / , i/T).
Thus/(z)e^2(M,z).
D
Hereafter we fix a positive integer k, and let^(z: Xi^Xi) be the modular form f(z) defined by (4.7.6). By Theorem 4.5.16, we note that (4.7.16)
fkiziXi^Xi) is ^ common eigenfunction of^T{n)for all n'^ 1.
§4.7. Dirichlet LFunctions and Modular Forms
179
For a Dirichlet character % ^^o^ ^ satisfying /( — 1) = (— If, we put (4.7.17) ^,(iV,X) = . We note that ^^(iV, x) is stable by T(n) ((w, AT) = 1) and generated by common eigenfunctions of T{n) ((n, N) = 1) by (4.7.16). Theorem 4.7.2. ^,(iV, x) = ^/^(iV, z). Proo/ First we show that ^^(Ar, x) c= ^^(Ar, x). Let g{z)GS'k(N, x) be a common eigenfunction of all T{n) ((n, N) = 1) with eigenvalue t„. Since ^k{N,x)^kiN.x)@^kiN,x\
we can write
9 = 01+92.
(9ie6^,(N,xl 92^J^kiN.X))'
By Corollary 2.8.4 and Theorem 4.5.18, ^^(Ar, x) and ./^^(Ar, x) are stable by T(n). Therefore g^ \ T(n)e6^k{N, x\ and hence, g^\T(n) = t,g,. Now there exists/(z) = ^(^J Zi»Z2) which has the same eigenvalues of T(n) as those for g(z) for all n prime to AT. On the other hand, by Corollary 4.6.14, there exist a divisor AT' of N and a primitive form h(z) of6^,^{N\ x) such that gi(z\ h{z) and g{z) have the same eigenvalues for T(n) ((n, AT) = 1). Therefore L(s;f) and L(s; h) has the same Euler factors for all prime numbers prime to AT. First assume that Xi is trivial. Then ris)Lis,Xi)L(sk\lX2) has a simple pole at s = /c. Since h{z) is a cusp form, r(s)L{s; h) is an entire function. Put 00
L{s;h)^
X «««"'' n= 1
then r{s)L(s, Xi)L(s  /c + 1, Z2) r(s)L(s;h)
fcl2s
1  «pP"' + Z(P)P = Trn i^'si^Xi{p)p')i^X2iP)p''r
and it has a pole at s = /c. This is impossible from the form of the righthand side. Next suppose that X2 is not trivial. Then by taking the twisted modular form g^^ in place of g, a similar argument is applicable. Consequently we obtain ^j = 0 or gEjVf,{N, x\ To prove that Sj,(N, x) = ^ki.^^ x\ we shall calculate the dimension of^,(iV,z). 1° The case when either /c > 2 or /c = 2 and x is nontrivial. Let us show that f^(lz\ Xi> Z2)'s are linearly independent. Let Z
c(lXi.X2)fkilz\XuX2) = ^.
(c(/,Xi,Z2)eC).
Here / runs over all positive integers such that IM\N, and X\ and X2 run over all
180
4. Modular Groups and Modular Forms
Dirichlet characters such that X\li = X Put 00
L(s, Xi)L(s k\hx2)=
Z a„{xuX2)n~"^ «= i
Then a„(xi,X2) is the eigenvalue of T(n) fovfk(lz; Xi^Xi) provided n is prime to N. Moreover if «„(Zi, Z2) = ^n(^i^ ^2) for all n prime to N, then Xi = «Ai and X2 = ^2Therefore, for any fixed pair (/i, X2\ we see (4.7.18)
Ic(/,Zi,Z2)/fefeZi,Z2) = 0.
Since the first Fourier coefficient of^(z; %i, ;(2) is equal to 1, we get c(/, Zi» X2) = ^ for all /. This proves the hnear independence of^(/z; Xi, X2) ^^ particular, dim ^fe(N, x) = #{(/, Zi, Z2)/MiM2Ar, Xi, Z2 are primitive and ;fi;f2 = x}The set of the righthand side corresponds bijectively to the set of pairs (Xi, ^2) of (not necessarily primitive) Dirichlet characters {{x'uX2)\x\X2 = X. M\M'2 = N}, where x'l is defined modM;. The correspondence is given by (lXuX2)^{x'uX2l where X'l — the Dirichlet character defined mod IM^ induced by Xi^ X'2= the. Dirichlet character defined mod N/IM^ induced by X2' Thus we obtain dim^,{N,x)=
^{(XuX2)\XiX2 = X,
M.M^^N).
2° The case when k = 2 and x is the trivial character. Then N > 1. By a similar argument as in l°,^(/z; Zi, Z2) are linearly independent, and dim^2(A^,z)= #{(Xi,Z2) I %iX2 = Z, M^M2 = N, and if X2 = X^ then M2^1}. 3° The case when k=\A{^^ (AT, x) ¥^ {0}, then AT g 3 .and / (  1 ) =  1 . Since /fcfe Xu X2) =fk{z\ lAi, »A2) if and only if {xu X2} = {^u ^2}^ we obtain dim^i(A^, X) = H {iXuX2)\XiX2 = X. M.M^ = AT} by a similar argument as in 1 °. Now we put (4.7.19)
S,{r,(N)) =
®S,{N,x\
where x runs over all Dirichlet characters mod N satisfying / ( — 1) = ( — 1)*. Then by Theorem 2.1.7, ^fc(A(iV))c:A(ri(N)).
§4.7. Dirichlet LFunctions and Modular Forms
181
From r , T and 3°, we see J^(f>(M)(l>iN/M),
if k^3,
N^3,
Nj^4,
M\N
~ Z ! {M){N/M)  1, if A: = 2, A^ ^ 3, AT # 4, M\N
dim^,(ri(iV)) =
Y,{M)(t>{N/M),
if /: = 1, A^ ^ 3, iV 7^ 4,
3 2
if AT = 4, ife(even) ^ 4, if AT = 4, A;(odd) ^ 3 or /: = 2; A^ = 2,fc(even)^ 4, if A^ = 4, ifc = l; A^ = 2, /: = 2; A^ = l, yfc(even)^4, otherwise
1 0
where 0 is the Euler function. Therefore by Theorem 2.1.7(1), Theorem 4.2.9 and Theorem 2.5.2, dim S^(r^ (N)) = dim jr^(r^ (N)). This implies ^^iri (N)) = jr^ir, (iV)), and therefore, ^^(iV, /) = ^^(iV, / ) . D Using expHcit Fourier expansions (4.7.1), we obtain the following estimate of Fourier coefficients of any element of ^^(N, x)Theorem 4.7.3. Letf(z) = i;r=o ««^^"'"^ ^^ «« element ofS^(N, x). Then 0(n''l 0(n''^%
if if
/c>2, /c=l,2,
for any e > 0. Proof For an integer n, we put (4.7.20) 0^2 p"^ £log2
If p ^ 2^^ then ^' ^ 2, so that (4.7.23) V
^
^
p«^
< ^ 
2^^
< 1. ~
By (4.7.22) or (4.7.23) according as Pj < 2^^^ or not, we see ^
^
n
^'/^'°^'^exp(2i/7elog2),
SO that Go(n) = 0{n^). These results combined with (4.7.1) prove the assertion.
D
Let us consider the special case when N = 1 and Xi = Z2 = Zo (the principal character). Assume /c ^ 4. From the remark mentioned at the end of §4.1, we see Aiz; Xo. Xo) = cE,{z)
(c = (k
l)\/2(2nin
where £^(2) is the Eisenstein series defined by (4.1.3). Even if iV > 1, it is known that if /c ^ 3, then fki^lXi^Xi) can be expressed as a linear combination of series similarly defined as Ek(z). In the case when /c = 1 or 2, a modification is necessary. Eisenstein series will be explained in §7.1 and §7.2 again.
§4.8. LFunctions of Quadratic Fields and Cusp Forms In the previous section, we showed that ^^(N, x) is generated by modular forms associated with products of two Dirichlet Lfunctions. It is natural to ask what kind of Dirichlet series are associated with cusp forms. If an Lfunction of a number field corresponds to a cusp form, then the field must be a quadratic field from the forms of the Euler products. Let X be a quadratic field, and ^ a Hecke character of K with My, v^ defined by (3.3.1). Assume that L{s — {k— l)/2, ^) corresponds to some cusp form. Then we see (i) K is imaginary, or (ii) K is real, and i;^ = 0 (v = 1, 2), M = MJ + M2 = 1, by comparing Tfactors which appear in the functional equations in Theorem 3.3.1 and Corollary 4.3.7. Here we use the same notation as in §3.3. Under the above situation, Hecke proved that L(s — (k— l)/2, i) corresponds to a cusp form by
§4.8. LFunctions of Quadratic Fields and Cusp Forms
183
using theta functions which we shall discuss in the next section. We are, however, going to prove this again by Weil's theorem. Let K = Q(y/d) be a quadratic field with discriminant d, p a. prime number with (p, d) = 1, and ij/ a primitive Dirichlet character of conductor p. We denote by r and / the ring of integers and the group of fractional ideals of X, respectively. We define the Hecke character il/° N^by (4.8.1)
{i^oN^)(a) = ik(NM)
(ael),
where Ng; is the norm mapping of/ into Q"". We note j/^ ° iV^^ is a primitive character of conductor px. Lemma 4.8.1. Let Xd = (~) ^^ ^^^ Kronecker symbol.
\iJ\d\ (2) If(d,p)=lthen
ifd 0, since L{s;f) is convergent for Re(s) > (M + 2)/2. Since L{s;f) has obviously the Euler product of the form in Theorem 4.5.16,/(z) is a common eigenfunction of all T{n). Assume that M = 0 and f(z) is not a cusp form. Since f{z) is a common eigenfunction of Hecke operators, L(s;f) =
Lis,Xi)Lis,X2)
with Dirichlet characters Xi^Xi t>y the proof of Theorem 4.7.2. But this can happen only when ^ is induced from a Dirichlet character. The last part of the assertion is obvious from Corollary 4.6.22. D For a real quadratic field K, a similar argument is apphcable and we obtain Theorem 4.8.3. Let K = Q{y/d) be a real quadratic field with discriminant d, and i a Hecke character mod n such that i{{a)) = a/\a\ = sgn(a)
(a=lmodn),
or ^(a)) = a'l\a'\ = sgn{a')
(a = 1 modn),
where a' is the conjugate of a over Q. We put fiz)=fiz;i)=l^iia)e'^^'^^K
§4.9. Theta Functions
185
where a runs over all integral ideals of K. Then where N = dNg^{n), and % is a Dirichlet character defined by x{m) = xAmKm)
(meZ).
Furthermore if ^ is primitive, then f{z) is a primitive form. Hereafter we assume that readers are familiar with Artin Lfunctions. Let us consider a special case when ^ in Theorem 4.8.2 and Theorem 4.8.3 is a class character. Then K is a real quadraticfieldor K is an imaginary quadraticfieldand ^({a)) = 1 provided a= \ modn. Let J be the subgroup of / defined by J = {ae/{(a)=l}. Let M be the abelian extension of K corresponding to J, and L the minimal Galois extension of Q containing M. Then G(L/K)/G(L/M)  G(M/K)  /(n)/J, where G(L/K) indicates the Galois group of an extension L/K, and /(n) = {a8/(a,n) = l}. Therefore ^ induces a character ^ of G(L/K). Let p be a representation of G(L/Q) induced by <J. Then we have L (s, p) = L(s, 0, so that L(s, (J) is an Artin Lfunction over Q. In other words, such an Artin Lfunction L(s, p) corresponds to a cusp form of weight L Therefore we may expect that there exists a correspondence between cusp forms of weight 1 and Artin Lfunctions associated with irreducible representations of degree 2 of Galois groups over Q. Indeed, [DeligneSerre] proved that cusp forms of weight 1 always correspond to such Artin Lfunctions. Conversely, it can be shown similarly to the above theorems that certain Artin Lfunctions satisfy functional equations of the type required in Weil's theorem. Therefore the above conjecture is equivalent to the Artin conjecture for such Artin Lfunctions.
§4.9. Theta Functions In §2.6, §4.7 and §4.8, we constructed modular forms by using Poincare series or Dirichlet series. In this section, we give another way of constructing modular forms which is a generalization of the theta function defined by (3.2.16). The advantage of theta functions is that we can explicitly calculate Fourier coefficients of modular forms. Furthermore, theta functions play essential roles in the theory of quadratic forms. Throughout this section, we put e(z) = eIniz
186
4. Modular Groups and Modular Forms
Let /I be a symmetric matrix of size r over Z. Assume that A is positive definite, namely, all the eigenvalues of A are positive. We put A [x] = ^xAx for a column vector x of size r, and put formally f(z)=
X e(Alm2z)
(zeU).
Hereafter we consider all elements of C , U\ Z** etc., as column vectors. For a nonnegative integer n, We put a{n,A)=
# { m 6 Z ' '  ^ [ m ] = n).
Then f(z) = f a{n, A)e^^'^\ n=0
We shall show that /(z) is convergent and is an integral form provided r is even. This gives an effective method of constructing integral forms. By this fact, we obtain an estimate of a{n. A) from that of Fourier coefficients of modular forms. The aim of this section is to prove that /(z), defined as above, is a modular form. We begin by defining theta functions from quadratic forms and studying the transformation formulae. The discussion below follows [Shimura 7]. We now define the differential operator A^ by ^A=
t
b,jd'/dx,dxj,
A' = lb,jl
Let P{x) be a homogeneous polynomial of degree v with complex coefficients in variables x^,. . . , x^. We call P(x) a spherical fiinction of degree v with respect to A if A^Pix) = 0. It is known that any spherical function of degree v is given by [ a constant P{x)=rqAx
(v = 0),
(qeC)
(v = 1),
( a linear combination of (^qAxY {qeC, X[^] = 0) (v > 1). We take an element /i e Z*" and a positive integer N satisfying (4.9.1)
NA^GM,(Z);
(4.9.2)
AheNr.
For the above A, N, h, and a spherical function P(x) of degree v with respect to A, we put for z e H
(4.9.3)
e{z;KAN,P)=
^ m = hmodN meZ'
=
^W^f^Sr^) \ ^^
/
f,^a(n,h,A,N,P)e(:^z]
§4.9. Theta Functions
187
where a(n, K A, iV, P) =
z
P(m).
A [m] = n m = h mod N
Since A is positive definite, there exist positive numbers c^ and Cj satisfying (4.9.4)
c^'xx^A
[x] S C2 'xx,
/or all x e 1R^
Furthermore by Schwarz's inequahty, there exists.C3 > 0 such that (4.9.5)
\P(x)\ S c^Cxxy/\
for all xeW.
Therefore ^
\P(m)\ = Oin^'^'^^'').
A [m] = n m = h mod N
Hence by Lemma 4.3.3, 6(z; /i. A, AT, P) is convergent absolutely and uniformly on any compact subset of H and is holomorphic on H. We call it a theta function. We put D = det(^), /c = r/2 + v. Lemma 4.9.L (The transformation formula) e{ l/z; k A, N, P) = {iyu^^^iizf
^
e{HAh/N^)Oiz; /, A, N, P).
Al = 0 mod N
Proof This is a generalization of (3.2.17), and is also proved by using the Poisson summation formula. For xeW and zeH, we put (4.9.6)
0(2, x) = Y. A
^— ^
and
We denote by/(«) the Fourier transform of/(x). Then and by (4.9.4), we can apply the Poisson summation formula and obtain (4.9.7)
e{z, x)= Y. /(m)e('mx)
We have only to prove the assertion for P(x) = (^qAxyiqeC). q = \qi, ..... Qr) and define the differential operator L by ^ = Z ^i^/^^i' i=l
Let
188
4. Modular Groups and Modular Forms
Acting U on both sides of (4.9.7), we see 'A [m + x] X P(m + x)ei 2
(4.9.8)
(_j)v/)i/2(^//)r/2v ^ {;qmye(A^lmyiz = iiYD''\z/iy''
+ 'mx)
X P{A'm)e{A^[rn\/2z
+ 'mx).
meZ'
This is justified by the uniform convergence of 6(z; h. A, N, P) on any compact subset of H. Note that if v > 1, then L^Alx^='qAq
= 0.
Put X = N~^h, and take — 1/z in place of z. Then the lefthand side is nothing but N^Oii/z; K A N, P). Put n = NA'^m. Then meT \i and only if «eZ'' and An = 0 mod N. Thus we see e(\lzKA,N,P) = (  iyD''\izf
X
Pin)e((A lny2N^)z
+ 'nAh/N"^).
An = 0 mod N neZ'
Since e(nAh/N^) is determined by the class of n mod AT, we see = {iyD"\izf
X
eClAh/N'')
/!/ = 0 mod N /eZ'/iVZ^
= (_/)v/)i/2(_^2)fc
Y
^
P(w)e((^M/2iV^)z)
n = l mod N
e{'lAh/N^)e(z;lA,N,P).
D
^/ = OmodiV /eZVNZ'
Hereafter we fix P{x) and write simply 9(z; h. A, N) in place of 6{z; h, A, N, P). The following equaHties are obvious by definition: (4.9.9)
e(z;KA,N)=
^
e(cz; g, cA, cN)
g = h modiV geZ'/cNZ''
for any positive integer c; (4.9.10) (4.9.11) Now let ^ =
d(z + 2; k A, N) = e(Alh']/N^)e(z; K A, N); e(z; h,A,N)
= ( iye(z; K A, N).
, \eSL2(Z). Suppose that c > 0 and a = d = 0 mod2. Since
cSz = a — (cz + d) ^ we obtain by applying (4.9.9), (4.9.10) and Lemma 4.9.1 e(dz; K A, N) = (iyD''^c''^i{czhd)y
X Al^O mod N ler/cNZ'
4>{k l)e(cz; /, cA, cN\
§4.9. Theta Functions
189
where 0(k /) =
Z
e(M[^] + TlAg + dAlQ)/2cN^).
g = hmodN geZ'/cNZ'
Since we can express it as
P
By taking 8pep = i into account
.(T^)= this isomorphism. Then, first 0 is an Fautomorphism of M and is compatible with the action of D, so that (/) is an element of Endj){M). Since Endj){M) = B, there exists an element P of B"" such that (j){m) = Pm
(meM).
Next for any yeC^ and any ij/eD, we see r, (3) N(a„)
^bmodp'"Zp.
Assuming we have constructed {aj, . . ., a„}, we shall show the existence of a„ + j . Since K is unramified over Qp, we have trx/QpW = Zp,
so that there exists an element y of r satisfying trK/Qp(7)= 1
Put a„^i = a„(l +C7)
(c = (bN(oc„))/N(a„)).
Since a„Gr"" by (1) and (2), we see c = Omodp*"Zp. Then a„ + i = a„modp'"'r.
§5.2. Quaternion Algebras
203
and iV(a„^i) = iV(a„)iV(l+cy) ^iV(aJ(l+c) =b mod/(" + ^>Zp. Therefore a„ + i is a desired one. Since {a„} is a Cauchy sequence by (2), it is convergent. So put a = lim a„. Then a satisfies (i) and (ii). If a ^ Z^, then j5 = a is what we want. Assume aeZp. Let {coi, CO2} be a basis of r over Z^, and (7 the generator of Gsi\{K/Qp). Put i/^ = l H  p ' ' ( n i C 0 i + W 2 < ^ 2 )
(^^i* " a ^ Z p ) .
If we can take n^ and ^2 so that then p = (x(il/''/il/) is what we seek. Since N{^''l^)= 1, ^°l^sZp if and only if •A^'/'A = ± 1 Since {ft)i, 0)2} is a basis of X over Q^, we can find n^ and «2 so that This completes the proof.
D
An element j5 of a quaternion algebra B over afieldF is called regular iff{X; p) has no multiple roots. Lemma 5.2.7. Let B be a quaternion algebra over Q^, R an order of B, and h a positive integer. For an element P of B satisfying NB(P)= 1, there exists a regular element a of B such that (i) iVfiCa) = 1,
(ii) oi = P modp^R.
Proof If P is regular, then the assertion is trivial. Assume p is not regular. First suppose B = M2(Qp). Then there exists an element y oi B"" such that
•[;:]• Taking a positive integer / so that pV"'M2(Z„)yc/i?, we put
«=±^"{'7' (ip')> Then a satisfies (i) and (ii). Next assume that B is a division quaternion algebra. Then j5 = ± 1, so that the assertion follows from Lemma 5.2.6. D
204
5. Unit Groups of Quaternion Algebras
Now let JB be a quaternion algebra over Q. We denote by Q^ the product of two copies of Q. For an element p of B, we put (5.2.2)
4>(P) =
(trB(PlN,(P))EQ\
The mapping (/) induces continuous mappings of B^ into Q^, and also of B^ into Q^, which are denoted by the same symbol {p) = (a, b). Next assume that f{X\ P) is reducible and put f{X;P) = {Xu)(Xv)
{u^veQp).
Since M 7^ t; by assumption, 5^ has a zero divisor, so that we may assume Bp — M2(Qp). A similar argument as in [Weil, XI, Lemma 1] implies that if (aj, b^) is sufficiently close to (a, b), then g{X;ai,bi) = 0 has two roots MJ, t?i in Op, which are close to M, V, respectively. In particular, Ui^v^. Therefore, taking yeBp so that
and putting
we get ae K and f(X; a) = g{X; a^, b^). This proves (2).
D
Theorem 5.2.9. Let B be an indefinite quaternion algebra over Q,Ra maximal order ofB, and N a positive integer. Assume that for each prime factor p ofN, an element Pp ofBp satisfying Ng(pp) = 1 is given. Then there exists an element p ofB satisfying the following three conditions: (i) NM = 1; (ii) p = P,modNR, (iii) peR, ifpJfN.
ifp\N;
§5.2. Quaternion Algebras
205
Proof. By Lemma 5.2.7, we may assume that (5.2.3)
Pp is regular for each p.
Let /c be a positive integer satisfying
for all prime factors p of N, and {aj,. . ., e j a complete set of representatives of (K/iV*R)\ Note that iR/NRr^Y\{Rp/N'Rp)\ p\N
since Z is dense in Ylp^p We see that (5.2.4)
(B^ X n Rp ) \ B A / ^ ' is a finite set. p
In fact, if B is a division quaternion algebra, then it follows from Lemma 5.2.4(5), Otherwise it can be verified directly. Let {^^^\ . . ., i^^^} he 3, complete set of representatives of (B^ xY\pRp)\B^/B"". As is easily seen, we can take all ^U) ^ (^(i)) so that i\/^ = 1 for all prime factors p of N. Now let L be a positive integer such that for all prime numbers p prime to L, (5.2.5)
Bp^M^iQp),
8,eR;(iSi^r\
^^J^ER;
[l^j
^s).
Moreover take a positive integer h so that (5.2.6) L'ii'Jhr'r'Rpil/hr'^Rp for all i andy and all prime factors p of L. For each prime number p which divides L but is prime to iV, we can take an element Pp of Bp by Lemma 5.2.6 so that (5.2.7)
NB(PP)
= 1, Pp is regular, and Pp = 1 mod L^Rp.
Hence a regular element Pp of Bp has been given for each prime factor p of LN. By Lemma 5.2.8(1), we can take a neighborhood Up of pp so that Up^pp\L'N'Rp and all elements of Up are regular. Take (a. l)siQ^ so that (5.2.8)
{p\LN),
Mi%,
and put g(X) = X^
aX^\.
By assumption, g(X) is a principal polynomial of an element of Bp for all p. Furthermore since B is indefinite, g(X) is also a principal polynomial of an element of B^ ( = M2(IR)). Therefore by Lemma 5.2.5, there exists an element a in 5 such thsit g(X)=f(X;oc). For each prime factor p of LN, there exists a^ e Up such that giX)=fiX;ap).
206
5. Unit Groups of Quaternion Algebras
Since a^eBp and aeB"", there exists an element yp of Bp such that
by Lemma 5.2.2. Let p be a prime number prime to LN. Since a e Z^, a is contained in a maximal order of Bp. Hence there exists an element jp of Bp such that ypay;^eRp, since all maximal orders of Bp are conjugate to each other. We may assume
for almost all p, since a is contained in Rp for almost all p. We put y = (yv) with the yp defined above and 700 =
1
Then 7 s BJJ, so that we can write
P
with some ^^^\ By the choice of {ej, there exists e^ such that (5.2.9)
rjpsr ^ e 1 + N^Rp
for all prime factors p of iV.
We put and shall prove that P satisfies the three conditions in the theorem, (i): Obviously NgiP) = 1. (ii): Suppose p\N. Since ^^/^ = 1, we have y^ = rjpS, so that
P = irips;')Hyp(^y;'){ripsr') = ii1psr^)^(xpirjpsr^). Therefore PPp =
{riper')'oCp(fjp8r')Pp
= 0 mod NRp by (5.2.9) and the choice of oCp and k. (iii): First assume pL and pJi'N. Then i5 = (e, m} has measure 0 and the set {zeH /(z)Im(z)*/^ > m —8} has nonzero measure for any positive number s. We denote by LJ(H) the space of all measurable functions / on H satisfying 1/ p < oo. In particular, Lg(H) is the L^space of functions on H in the usual sense. Hence it is a Banach space. Since LJ(H) is isomorphic to Lg(H) as normed spaces by the map y(z) i>/(z)Im(z)''^^", LJ(H) is also a Banach space for any k. We call two exponents p and q are conjugate, if they satisfy that
P
q
Here we understand l/p = 0 if p = oo. If p and q are conjugate exponents, we put
220
6. Traces of Hecke Operators
for/eLJ(H)and^eLl(H), (6.1.1)
(/, ^)H = J / ( z ) ^ lm{zfdv{z). H
By this pairing, an element g of LJ(H) can be identified with an element of the dual space of LJ(H), or the space of all continuous Hnear functional of LJ(H). Further if p 7^ 00, Ll(H) is considered as the dual space of LJ(H) by this identification. Further L^(H) is a Hilbert space with the inner product given by (6.1.1). We also denote by //f (H) the subspace consisting of holomorphic functions in Z/J(H). Theorem 6.1.1. The space HJ(H) is a closed subspace o/LJ(H). Proof. Let f{z) be the limit of a sequence {/„(z)} in HJ(H). Applying Corollary 2.6.4, we see that f(z) is also holomorphic. D Let H be a Hilbert space consisting of complex valued functions on a set X. We denote by (/, g) (fgeH) the inner product of H. We call a function K{x, y) on X X X a. kernel function of H if it satisfies the following two conditions: (6.1.2)
as a function ofx, K(x, y) belongs to H for each fixed y;
(6.1.3)
for any element f{x) of H, f(y) = {f{xlK(x,y))
(yeX),
where the righthand side is the inner product as functions ofx. If K(x, y) is a kernel function, then we see easily (6.1.4) K(x,y) = K(y,x). Kernel functions do not necessarily exist. However, if it exists, it is uniquely determined. Furthermore, if H isfinitedimensional, then it has the kernel function given by K{x,y)=
tfAx)fJj),
ix,yeX%
v= 1
where {/j,. . . ,/d} is an orthonormal basis of H. Now we shall show the existence of the kernel function of H^ (H) and calculate the expHcit form. By Corollary 2.6.2, there exists a constant C depending only on z 6 H and k such that /(z)^C/2 ifeHKH)). Therefore, for each fixed ZQEH, the map "/i—•/(ZQ)" is a continuous linear functional of H 1(H), Since Hl{H) is a Hilbert space, there exists a unique element ^^0 6/f^(H) satisfying /(2O) =
(/^2O)H
Put Xfc(zi, Z2) = ^22(^1) ^^^ Zi, Z2eH. Then it is the kernel function of Hl(H) and /(z) = J K,(z, Z2)fiz2)lm{z2fdv(z2) H
for any/e/f2(H).
§6.1. Spaces of Functions on H
221
Let a be an element of SL2(^y For any feHl{ll), the function f{oLz)j(oi,z)~^ belongs to H^(H). Since the measure dv{z) of H is invariant under the action of SL2(Ul we see for any/eiff (H) JXfc(azi,az2)7(a,Zi) ''j{oc,Z2) ^f(z2)lm(z2fdv(z2) = ;(a,zi) * JiCfc(azi,Z2)/(a ^Z2);(a,a ^Z2)*Im(z2)*t/r(z2) H = ;(a,Zi)'/(zi);(«,Zi)''=/(zi)
By the uniqueness of the kernel function, we have (6.1.5)
K^iazi, az^) = K»(zi, Z2);(a, z.fjia, z^)'
(a6SL2(R)),
and (6.1.6) Kt(az„z,);(a,Zi)* = A:»(z.,a'z2);(a',Z2)''
(aeSL,(R)).
ing[J J](fc(fe e IR) as a, we see
In particular, taking (6.1.7)
X,(zi + 6, Z2 +fe)= X,(zi, Z2)
(beU).
We put M = {(zi, Z2)eC^ I Zi eH, zi  Z2 eH}, and /l(Zi,Z2) = Kfc(Zi,ZiZ2)
((Zi,Z2)eM).
Then /i(zi, Z2) is a holomorphic function on M. Since /i(zi + b, Z2) = /i(zi, Z2) for anyfoeIR,/i(zi, Z2) is independent of z^. For zeH, take z^eH so that {Z]^,z)eM and put Pf,{z) = h{z^, z) = Kf,{zi,zi  z). Then P^iz) is a holomorphic function on H and KkizuZ2)=^Pk{ziZ2) Take \^
(zi,Z2eH).
_A (aelR^) for a in (6.1.5), then we have Pk(a^z) = a'^'^Pkiz). In
particular, Pkiiy) = y"'n(0
(3^ > 0).
Since P^iz) is holomorphic on H, we see that Pt(z) = c,(z/2i)' with a constant c^. Thus we obtain the following
222
6. Traces of Hecke Operators
Theorem 6.1.2. The Hilbert space ^ ^ ( H ) has the kernel function
with a constant c^. Corollary 6.L3. H(H) c= H^CH). Proof. Let f{z) be an element in i4^(H). For any point Z Q ^ H , we see that l/(^o)l' =  ( / ( Z ) , K , ( Z , Z O ) ) H P ^ ll/lli lX.(z,Zo)i = /iX,(zo,Zo)=c,Im(zo)Ml/lll This implies /(zo)Im(zo)*^^ ^ ^Jc^ WfWi for any Z Q E H , and therefore/belongs toH^(H). D Let us calculate the constant c^ in Theorem 6.12. We denote by R+ and (R_ the set of all positive real numbers and the set of negative real numbers, respectively. For an element /(z) in //^(H) and a point 3;6 IR+, we put
fy(x)^f{x
+ iy)
(xeU).
Since  / 1  i = JH  / ( X + iy)\^y^~^ dxdy is finite, there exists a subset Sf of measure zero in IR+ such that/yGL^(IR) iiy^Sf. Denote by^(M) the Fourier transform oify for yiSf, or ^(M) = l.i.m.„^Qo J"_„^(x)e"^"*"''i/x, where l.i.m. implies the limit in L^{U). Theorem 6.L4. (1) For a function f{z)eHl{H% satisfying
there exists a function f(u) on U
and f(u) vanishes almost everywhere onU. (2) Ifk^hthenHl(Yi) = {Q}. Proof. For any positive integer n, we put /^^(M)=
J / ( x + i»e2"'"^^''»Jx = e2""^ ]
fy{x)e^'''"''dx.
—n
—n
Since^(w) = Li.m.^ „(M)e"^""^ for y^Sy, there exists a subset Ty of U of measure zero such that lim„^oo/y „(M) exists if M^ T^. Let y^ and ^2 t>e two positive numbers such that ^ 1 , ^ 2 ^ ^ / ^^^ yi ^i] Since /(z)e"^"*"^ is holomorphic on H, we have
{
n + iyi
n + iy2
—n + iy2
I
+ J +
J
 n + iyi 'J
+
J
^/(z)e2"'"^^z = 0.
Therefore lim„^^/y^,„(M) = l i m „ _ ^ ^ „l(w). This implies lim„^^^„(M) is independent of y. If we put/(M) = lim„_ ^ ^ „(«) then it satisfies the equation in (1).
§6.1. Spaces of Functions on H
223
Now we see that
(6.1.8) ^>\\f\\l
= ]y''dy 0

] \f,(x)\Ux = ] y''dy J I/MI^^^MM ~ 00
00
0
""00
0
The integral with respect to y isfiniteif and only if /c > ^ and M > 0. Then, if /c > 1, f(u) vanishes almost everywhere on [R_. If /c ^ 1, f(u) should vanish almost everywhere on R, and therefore f(z) = 0. This completes the proof. D For /(z)eiffc(H), we call the function f(u) on U the Fourier transform of/ Hereafter we assume /c> 1. We put G,(u) = j I y'~'^~^"'dy
= {nu)''r{kl)
(u > 0),
We denote by H^ the space of complex valued measurable functions satisfying the following two conditions: (6.1.9)
(/)(M)
on U
eHi).
0
By (6.1.8), we obtain (6.1.11)
  /   2 = J \f(u)\'G,(4u)du = ifjy
ifeHiiU)).
— 00
Therefore, if feH^iH), then/eH^. Conversely, let 0 be an element of H^. For zeH and a positive integer n, we put (6.1.12)
(^(z)= J (/>(M)e2"'""dM, (^„(z)= J (l)(u)e^'''"'du.
We shall prove that the map "/(z) ^^f(uf' gives an isomorphism of /ft^(H) onto Hj^. We need the following general Lemma 6.1.5. Let (X, dx) be a measure space and D a domain in C. Letf{z, x) he a function on D x X holomorphic in z for each fixed x. Assume that for any circle C contained in D,f(z, x) is integrahle onC x X. Then the function F (z) = J;^/(z, x) dx is holomorphic on D.
224
6. Traces of Hecke Operators
Proof. Let z be a point of D, and C a circle around z. Then 1 ./(C, x) /(Z,X) = —TJT^
C
Therefore
Thus F{z) is holomorphic on D.
D
Theorem 6.1.6. The correspondence 'J(z)\^f(uy' gives an isomorphism of H^iH) onto Hi. Proof. We have only to show that the map "(/) i—• 0" is an isomorphism of Hi into /ffc (H) and it is the inverse of *y(z) !—•/(«)". Let (u)\'G,(4u)du = ((l>,cl>}. 00
0
This implies "0 H^ (/>" is an isomorphism of H^ into HKH). It is easy to see that these two isomorphisms are inverse to each other. D Corollary 6.1.7. Hl(l{) ^ {0} if and only
ifk>\.
Proof It is obvious that if/c > 1 then Hi ^ {0}. Then by Theorem 6.1.6, we obtain the result. D
§6.2.
The Projection of L^H) onto //j^(H)
225
Now we are ready to calculate the constant c^ in Theorem 6.1.2. We denote by Kfc(M, z) the Fourier transform of K^iz^^z) considering it as a function of Zj for a fixed z. Then for any 4){u)eHl, we see that
= (
and it is bounded. By (2), we obtain (3).
D
Theorem 6.2.2. The integral operator K is a projection ofLKH) onto H£{H)for any p(\^p^co). Proof. By Theorem 6.2.1 (3), the integral in (6.2.1) isfinite.Now we define a function Kf{(x) on SL2(U) by (6.2.2)
X,*(a) = ICfc(ai, i); (a, 0"*
(ae5L,
Then for a, )SeSL2(lR), we see that (6.2.3)
X,(ai, pi)j(0L, i)'j{p,
i) ' =
Knr'oi)
and (6.2.4)
X,*(a^) = X*(^.
For a measurable function cp on 5L2(IR) with respect to the Haar measure da defined by (1.4.13), we put (
j
\(pioi)\'doX
ess. sup(/)(a)
(l0. Then we see by Corollary 1.7.5, 00 > j
\f(z)\Hm{zfdviz)
r\H
^\
ii ^ 1
\f(0), X
det(a)'cj'^
, and take the set {a}
so that \dj < \hCg,\. Take a point z so that ^.{a} d^t (a) \j (a, z)~* is convergent and \z\ ^ /i, then we obtain Xdet(aycJ'^ 0) tends to 0, we see j
X K{Z; (x) dv{z) = lim
j
YJ 'C(Z; a)Im(z)"*^i?(z).
We shall show that the order of integration and summation on the righthand side is exchangable. Put S(z,5)= X /c(z;a)Im(z)^
(5>0).
We have only to see that 5(z, s) is integrable on r^\U^, we have S(z,5)^Ci(Im(z)^^ + Im(z)^)
^ dcm''\j((x, aer„\r„
Similarly to (6.4.4), z)\'
§6.4. Traces of Hecke Operators (Calculation of Integrals)
235
with a constant C j . Since (Im(z)^~* + Im(z)~*) j(a, z)"* is integrable on r^\U^ and r^\T^ is a finite set, 5(z, s) is also integrable on r^\U^. D Let Pp be the set of all cusps of F. Take neighborhoods U^ of cusps x so that they satisfy (6.4.5)
Uy^ = yUAy^n
and
U^nU, = 0
ifx7«^x'.
Also take elements o^eSL2(IR) so that a^x = oo and Imiay^yz) = Im((TjfZ) iyeF). We put Z ( r ) = r n R " , r 2 = y (T^Z(T)) and T' = TT\ xePr
For a 6 r^, we put K(Z;OC)
(6.4.6)
(Z^
K:(Z; a, S) = l I K(z;a)Im(z);(ri},
(TI/2 = { Z G H  zir2llC 72) *()^ = Im(z)), this integral is not finite. \Z — Z + AL, J Therefore x must be a cusp of F and r(a) = T^. Denote by T^ the set of parabolic elements of T, and put 71=7^^ T^ for each cusp x of T. Since r , = r(a) for a e r  , any two elements are not Tconjugate. Therefore we can take the set (Jxerxp^^x ^s a complete set of representatives of T^l/F. Since r\Pj^ is a finite set, we see
^^^ Z
J '^^fe ^"sgn(yonm(a) + n r l  ^ n=—oo
and it has a pole of order at most 1 at s = 0 by Theorem 3.2.2. Further we see that  ( i ' " ' ' + (  i V ' ) =  s +fl2S^+ a3s3+ 71
Therefore we obtain the formula for t„.
D
§6.5. Traces of Hecke Operators (Algebraic Calculation)
243
For the further calculation, we arrange each term. For CLET^KJT^KJ
T^,
we put
(aer). fc(a) =
^g„°„).min{ICJ,l,jr' IC'/J sgn(a)*det(a)*/2i
{OLET^
and /(«) =
l/(2r(a)) l/Z(r)
(aen, (a e n
mZ(r)V\m(0L)r')
{oLETn.
Corollary 6.4.11. We have
te=
1 xm(oi)m\ 0Ler//r
(xeT^lir
and tp=  lim
X
Z(a)fc(a)/(a).
If r\H is compact, then t^ = tp = 0.
§6.5. Traces of Hecke Operators (Algebraic Calculation) In the rest of the chapter, we apply the general formula of traces of Hecke operators in §6.4 to Fuchsian groups obtained as unit groups of norm 1 of indefinite quaternion algebras. We use the notation in §5.3 for quaternion algebras. Let B be an indefinite quaternion algebra over Q with discriminant dg, R Sin order of B of level N, and rj, =
{yeR^\Ns{y)=l}.
Let r = Lir^iarjj be a (disjoint) union offinitelymany double T^jcosets included in AR. We identify T with the element SruaTu in ^(r^, A^). Moreover we put T^=Uj,TU^. It is a disjoint union of finitely many double URCOSQIS by Theorem 5.3.5. We identify T^ with the element ZC/jj a l/^ of ^(l/^j, Djj). Note that T corresponds to T^ through the isomorphism of ^(r^, zj^) onto M(Uj^,Dj^). Let ;f be a Dirichlet character mod N. Then x can be considered as a character of T^ by (5.3.6). The
6. Traces of Hecke Operators
244
purpose of the remaining sections of this chapter is to obtain a computable form of trace formulae of Hecke operators T= T{n) defined by (5.3.12) operating on ^ki^Ri x)' Hereafter we also denote Fj^ and U^ simply by F and 17, respectively. For the convenience of the reader, we recall the notation in the case of modular groups. In this case, B = M2(Q) and c = OmodAf/ UR = GL^(U)XY[R;
^B A
}•
»
where fffl
hi
e M 2 ( / , ) c = OmodNZ,
Therefore Tjj = L/^j n 5"" = FoiN), Moreover
ip\N)
D„ =
D^ = (GLUU)
X
llD,)nGL2iQA
As for Hecke operators, we see r(n) = {a6ZloWdet(a) = n}, T{n)^ = U^T{n)U^  {GmU)
xl[\T(n),\nGL2{QA
where r(n) = { a ^ E D J d e t ( a J e n Z ; } . Now we return to the initial situation in this section. We are going to calculate r V / r , T^//F and T^/fF, For an element a of B^ ni^ not contained in Q, we put C(a) = CBx(a) = {^a^i^eB^}. Lemma 6.5.1. (1) Assume OLET^^T^KJTP.
Then
an imaginary quadratic field K, Q[a]^< Q x Q , Q[£]
if cue T^, if
(8^ = 0),
(2) Let S be T\ T" or T^. Then for aeS^we have SnC(oL)=TnC(oi).
(xeT\
if oceT^,
§6.5. Traces of Hecke Operators (Algebraic Calculation)
245
Proof. (1): Leif(X) (eQ\_XJ) be the principal polynomial of a. Then Q[a]^Q[Z]/(/(X)). By definition, f(X) = 0 has two imaginary roots, two distinct real roots, or a multiple real root if a is elliptic, hyperbolic, or parabolic, respectively. This implies (1). (2): Since the type of any conjugate of a is the same as a, the assertion is true for 5 = r^ or TP. Suppose S = ^^ If T* 7^ 0 , then B = MjCQ), and therefore Pr = Q u {00}, and any element of B ^' = GLjCQ) maps Pf onto itself. Therefore if a is contained in T*, all fixed points of conjugates of a by elements of B"" are also cusps of r . This impHes the assertion for S = T*. D Since k((x) = k(S(xS^) for any SeB\ the above lemma impHes that if 5 = ^^ T" or T^ then
^6.5.1)
X xm{oi)m= X fc(a) s oceSf/r
aeSZ/B**
xmim
/56(rnC(a))//r
For any element ^ of 5"", (Q[a] n o.
Proof First suppose v = 0. Then Rp = M2(Zp). For an element a of GL2(Qp), there exist Ui,U2eGL2(Zp) such that WiaM2= Q
/
(eJeZ)
by elementary divisor theory. Therefore (xRp(x~^ = Rp if and only if e =/(namely
250
6. Traces of Hecke Operators
OLEQp Rp ). Next assume v > 0. Since there is a unique pair (R^p\ Rf^) of maximal orders of M2(Qp) such that
by Lemma 5.3.1, any element o{N(Rp) induces a permutation of {R^p\ Rf^} Since we can take
for (R'^K R^p^) in our case, we see [N(R,):Q;R;] = lor2. On the other hand, so that
^ is an element of N(Rp) but does not belong to Qp Rp , '•[:•;]" N{RP)^Q;R;.
This impHes the assertion.
D
Let a be a nonscalar element of Rp such that det(a) ^ 0, and
UX) = X^tX
+n
(t.neZp)
be the principal polynomial of a. Let r^ be an order of Op [a] including Tp[ix\. We put [rp:Zp[a]]=p^
(p ^ 0),
then Lemma 6.6.1 implies that r^ is uniquely determined by p. Put (6.6.5)
(2(a, Xp) = {^eZpUM) ^ 0 modp^^^^ tU^O
modp^}.
Furthermore, for an element { e Q(OL, Xp\ we put
(6.6.6)
^^=[P4,)
uy.
Since/,(fif^) = 0, wesee ^ ^ e C » = {xaxMxGGL2(Qp)}. Lemma 6.6.3. Let Rp he an order o/M2(Zp) defined by (6.6.3), a a nonscalar element of Rp, and Xp an order o/Op[a] such that tp^ZpM
and [rp:Zp[a]]=p^
(p^O).
For an element g of Cp((x\ the following four statements are equivalent. (1) geCp(ci,Xp). (2) geZp^ p'Rp, and g^Zp^ p^^' Rp.
§6.6. Local Conjugacy Classes
251
[:a
(3) Ifg = \ ^ ^ L then b = ad
= Omodp^,
c = 0 mod p^ •" \
and any one of the following three conditions is satisfied: (i) (ii) (iii)
b^Omodp^^'; c^Omodp^^'^'; ad^Omodp^^K (4) There exists an element {e^(a, r^) such that g is N{Rp)conjugate to g^.
Proof (l)o(2): Put g = hoch^ with heGL2(Qp). Then geCpioi, Xp)Qplg;\ nRp = hXph~^
oZplg:\
=
Zp¥Qplg]np'Rp
by Lemma 6.6.1. This implies the equivalence of (1) and (2). (2)o(3): This follows from the definition of Rp. (4)=>(2): Suppose that g is Ar(i?p'')conjugate to g^ with an element ieQ(ix,Xp). Since g^ always satisfies (3), it does (2), so that V n ^^
V A
^^^^ satisfies (2). This combined with Lemma 6.6.2 implies
that g satisfies (2). (3) => (4): Suppose that g satisfies (3). First assume condition (i), and put b = p^b^ (b^eZ;). Since g is conjugate to a, we seeX(S') = 0. In particular, t = a\d,
n = ad — bc,
so that / , ( a ) =  Z ) c = 0modp^•^2^ and r  2 a = 0modp^. Thus aeQ{<x, Xp). Moreover we see ugu ^ = g.
("[: :.]'•)
Next suppose that g satisfies condition (ii). Since
v ^ U' ^ satisfies [_P OJ LP OJ condition (i), it is 1?^conjugate to g^ with some ,r,)//K,   { ^ ^ , , , ^ ,  .  ^ , Q ( „ , , ^ ) } / / ; , ; ,
Lemma 6.6.4. Assume v ^ 1. TTien ^ =
(,^1).
eCp(a, r^) is Rpconjugate to some
g^ (^€0(a, rp)) if and only if fe^Omodp^"^
or arf # Omodp^'^
Proo/ The ifpart was already proved in the proof of the (3) => (4) of Lemma 6.6.3. Now assume that h = ad = Omodp''^\ and ugu'=g^
r
c' ^. Mp^^^^(0),
then [ r r : r n = P'. Proof We begin by proving (2). By Lemma 6.6.1, we have x = Zp\ P^XQ = {(a, b)eXo\a = b modp^}. Since
to" = z ; X z ; , we see
xs/x^^(z/p^zr
by the correspondence "(a, b) i> a/b'\ Therefore we get
Next let us prove (3). We can write ri = Z. + p^Z_8 and X2 = Zp +p'^^ZpS
§6.7. Class Numbers of Orders of Q [a]
257
with some integer/by Lemma 6.6.1. Thus rr=Z;+p^Z^£
and
x^ =Z;
h p'^^Z^e,
so that through the correspondence "a + p^bsh*b/a'\ Therefore we get
[tr:tn=P'. Lastly we are going to prove (1). Put If p splits in K, then so that this case is reduced to (2). Assume that p remains prime or ramifies in K. The Kp is a quadratic field over Q^, and
by Lemma 6.6.1. Then the sequence 1 ^ Z;/(1 + p^Zp) . ro^,/(l + p^xop)  toVr;  . 1 is exact, roV(l+P^WI = (ro>^ro,)^
{
p^^(l — p^^) (p: remains prime), p^^(l —p~^) (p: ramifies)
and i z ; / ( i + p^Zp)\ = i(z,/p^z,)^ I = so that we get (1).
p%ip'\ D
Class numbers are given by the following Theorem 6.7.2. (1) Let K = Q{^) be an imaginary quadratic field with discriminant d, XQ the maximal order of K. If x is an order of K and [ro:r] = n, then
Hx) =
''•"!?.('(?>"•) [ro^:t^]
where hj^ = /i(to) is the class number of K. (2) Let H = Q X Q, and x an order ofH. If [Z X Z:r] = n, then h(x)=nY\(lp'). p\n
258
6. Traces of Hecke Operators
(3) Let E = Q[£] (e^ = 0). Ifx is an order of E, then h(x)=L Proof. (1): By definition, we see that x^+ =x^ and
= VlroVr;i(K''nro^^)/(K^or;)r'. Since K is imaginary, we see X ^ n r o ' ^ = ro'
and
K"nr;=r\
This combined with Lemma 6.7.1(1) implies (1). (2): Put ro = / X Z. Similarly as above, we have h(x) =
\H:i/H^xSA^\'\xSAjxl.\'\{H''nx^^^)/{H^nx^^)\K
Lemma 6.6.1 implies r = {(a, b)exQ\a = b modn}, so that H^nro^^ = / / ' ^ n r ; = { ( l , l ) , (  l ,  1 ) } . Moreover, since Q;
= Q ^ z;,
we see Therefore we get p\n
by Lemma 6.7.1(2). (3): We see r = {a +
tbe\a,beZ}
with some tGQ"" by Lemma 6.6.1. Since £ J ; = Q ; ( I + Q^£), £^ = (Q^(HQe), and r;=Z;(l+tZ^8), we see that x^+=x2 and /i(r) =  £ ; / £ ^ r ; i = 1(1 + Q^6)/(l + Q8)(l + tZ^s)\ = I Q A / ( Q + ZA)I = 1.
D
§6.8. An Explicit Formula for tr(r(n))
259
Remark 6.7.1. As for the class number hfr, see [BorevichShafarevich]. When K is an imaginary quadratic field, we easily see {±'i, ± \ / 5 t 
if K = Q ( y ^ ) and r is maximal,
{ ± 1 , ± e " ' / ^ ±e2"'/3} { ± 1}
if K = Q ( y ^ ) and r is maximal, otherwise.
§6.8. An Explicit Formula for tr(r(n)) Finally in this section, we present an explicit form of tr( T(n)). First we calculate the terms for elliptic a's. Lemma 6.8.1. Let K = Q(^/5) be a quadratic field with discriminant d, and XQ the ring of integers ofK. For an element (xofxQ(oLtZ),letX^ — tX\n be the minimal polynomial of a. Ifm is the positive integer given by t^ 4n = m^d, then [ r o : Z [ a ] ] = m. Proof It is easily seen that XQ = Z\ZS with S = (d { y/d)/2. Put oc = I\mS (/, meZ), then r^  4n = m^d. Since Z [ a ] = Z[m^], we get [to: Z [ a ] ] = m. D Notation being as above, we suppose that K is imaginary. For any positive integer/, there uniquely exists an order r of K such that [roir] = / ; in fact, r = Z + / r o . We put h(pd)
= h{x\
w(/2^) =  r ^  .
Then Theorem 6.7.2 implies
Hpd) w(pd)
h(d) w(d)
^PX©'')
Let a be an element of T^. Then Q [ a ] is isomorphic to an imaginary quadratic field Q(>/5) with discriminant d. Let f^(X) = X^ — tX{nhQXhQ minimal polynomial of a. We define m ( > 0) by r^ — 4n = m^d. Then by Lemma 6.8.1 combined with (6.5.3) and (6.6.15), we obtain
(6.8.1)
X l{x)h(x) Z[a]c:r
X
x{g)
ge{T^nCA<x.,x))IIV
=2" z S ^ n i I z,(«)+ z x,(to 0i ifd=.0,N>l
p = ordp(m//), M = m a x { 0 , e  [ ( v + l)/2]}, /i' = m a x { 0 , e  [v/2]  1},
and Xp is the character of (Zp/NTpY induced from % ^e note that ip(a) = 0 if (a, N) ^ I for aeZp (even ifxp is trivial). Remark 6.8.1. Though we have assumed /c ^ 3, a similar consideration using certain limit process is appHcable to the case when k = 2 (see [Ishikawa]), but it needs the knowledge of infinite dimensional representation theory of SL2{U). Another method by Eichler enables us to calculate traces for the case when fc = 2 ([Eichler 4], [Saito]). In any case, we can also calculate tr(r(n)) for the case when k = 2. We state the results without proof. Ifx is not trivial, then the above formula holds. Otherwise, tr( r(n)) is given by the above formula with the additional term j + i
(N=l), p\n
S(N,n)=
y .1+1
^
UP'U p\n p\N
where n = I~[pP'.
p\n PXN
§6.8. An Explicit Formula for tr{T{n))
265
Remark 6.8.2. When B = M2(Q), we have FR = ToCAT), and T{n) is the Hecke operator defined by (4.5.4). Theorem 6.8.4 is still somewhat complicated to calculate in a general case. In many cases, we can still simplify it. As an example, we write down here, without proof, a computable formula for the special case when T = ToCA^) with N = pq" {p,q: odd prime numbers, v ^ 0) and (n,N)= 1. Furthermore we assume that x is induced by a Dirichlet character modp and satisfies x{— 1) = (— 1 )*• Then tr{T(n)\^,(Na)) (V = 0 ) ,
q^'iq+l) 1
xiy/nW''^^
X
0),
(v = 0),
^v/2_^^v/2i
(y:even>0),
Iq^'^^l^
(viodd), 1

Z
ix(t) +
xin/t))t''xi
^[v/2]_^^[(vl)/2]
t\n 0(r(Ar)) c: ^„(r(N)). Proof. By Theorem 7.1.3 and Lemma 7.1.4, all elements of (^i^\r{N)) are holomorphic at co. Let g{z) be an element of y Corollary 2.6.12. To express the summation of the righthand side of (7.1.27) more expHcitly, we need the following
:[* Lemma 7.1.6. The correspondence:
*^^(c,d)U \\^{c,d) induces the following bijections:
(1) ro{N)^\ro{N) ^ {(c,^)c ^ 0modN, ( c , d ) = l d > 0 } ; {(cj)\c = 0 mod N, d=lmodN, I {(c,d)\c = 0modN,
(c,d)=i} (N ^ 3), d= 1 modN, (c,d)= Ud>0}
(N = l,2).
[
Proof. It is easy to see that the mapping is welldefined and surjective in both cases. Let us prove the injectivity. Let y = \
\ and y' = \
j \^^ ^^^ elements of
ToCN) (resp. r(N)) having the second row (c, d) in common. Then by a direct calculation, we see
•=C:}
yy
which belongs to FoiN)^ (resp. r(N)^). This implies the injectivity.
D
This lemma implies that when ;f(— 1) = (— 1)*,
ElsU;x)=
I
x(d){cz + d)\
c = 0 mod TV
{c,d)=\,d>0
and therefore, for any Dirichlet character x niod N, we obtain (7.1.29)
El^(z;x) = \
Z
xid)(cz + dy\
•^ c = 0 mod N
ic,d)=\
Thus (7.1.30)
E,(Nz; Xoa) = 2L^(/c, x)El ^(z; x)
for the principal character Xo> and (7.1.31)
E,(z; 0,1; iV) =
^  ; L ^ ( / C , X ) £ , % ( Z ; X)
by (7.1.26), where x runs over all characters mod AT. Since £^,^(z;x) belongs to J^kir{N)) by Theorem 2.1.7(2), we see (7.1.32)
E,{z;OA;N)e^,ir(N)).
274
7. Eisenstein Series
Theorem 7.1.7. Ifk^3,
then
Proof. We see by (7.1.17) and (7.1.19) that ^[^^ {r(N)) is generated by {£,(z;0, 1;L),7  LiV, yeSL^i Since £,(z;.0, 1; L ) e ^ , ( r ( L ) ) c: ^,(r(iV)), we see Sj,^^ {r(N)) cz jr^{r(N)). Let us prove ^fc(r(iV)) c= ^^i)(r(N)). First by Lemma 7.1.6(2) and (7,1.29), we see (7.1.33)
F,{z; cj>^, Xo. r{N)^,
r(iV)) =C
^
(^^ + ^ ) " '
modN )=i
2C 0(iV) , where ^
'1 "2
(N^3), (N = l,2),
and X runs over all Dirichlet characters mod N. Since Ej^^ ^ (z; ;f) belongs to ^i^\r(N)) by (7.1.30) and Lemma 7.1.4, so does Fj,(z; (j)^, Xo> ^(A^)ao, r{N)). Since ^ J E ^ H ^ W ) is stable under SLjiZ), we see F,iz;^o.Xo,r(N)^,r(N))\,ye^l'\r(N))
for
yeSL2(Z).
By (2.6.11) and Corollary 2.6.12, we obtain J^k(r(N)) cz ^j^^\r(N)).
D
§ 7.2. Analytic Continuation of Eisenstein Series When k is smaller than 3, the righthand side of (7.1.1) is not convergent. Therefore we need some modification to discuss the case when k ^2. Though the main object is the case when /c ^ 2, we note that the arguments and the results here cover any integral weight k. Now for any integer k and a complex number s, we put (7.2.1)
£,(2, s; X, JA) =
T m,n=
x(m)il^(n){mz + n)'\mz + n\^^  00
The righthand side is uniformly and absolutely convergent for s on k\ 2Re(s) ^ 2 + e (e>0). Therefore it is holomorphic on /c 4 2Re(s)>2. We call £^(2, s; x, ^) an Eisenstein series with parameter s. Similarly to Lemma 7.1.1, we obtain that (7.2.2)
E,{yz, s; x, lA) = x{d)if(d)(cz + df\cz + d\^'E,(z, s; x. ^A)
[ : : ]  (L, M).
§7.2. Analytic Continuation of Eisenstein Series
275
As was mentioned above, iffe^ 2, then the righthand side of (7.2.1) is not defined for s — 0. But if £fc(z, s\x^^) is continued analytically to s = 0 and holomorphic at s = 0, then we will obtain a modular form even for weight /c = 1 and 2. The purpose of this section is the investigation of the analytic continuation of E^fe s; x,^) and its property as a function of z and s. Since we easily see that if x(l)H^)^i^)\
then JS.Cz, s; z, lA) = 0,
we assume (7.2.3)
;^(_l)^(_l) = (_l)fc
throughout this section. We also see (7.2.4)
E,(z,s;x,il/) = ximk{^.k\s;x,ik)
if /c + 2Re(s)>2.
When k + 2Re(s) > 2, we can exchange the order of summation of the righthand side of (7.2.1) and see (7.2.5) £,(z,5;x,iA) = C(s) + 2 f
z(m) f
i^{n)(mz +
ny'\mz^n\'\
n= — ao
m= 1
where 0
(ZT^ZOX
For complex numbers a and p, we put (7.2.7)
S(z;a,^)=
f W=
(z + n)'(z + n)i>
(zeH).
— 00
The righthand side is absolutely convergent for a, j? with Re(a + j?) > 1, and (7.2.8)
f il/(n)imz\n)''\mz\n\^' n=  00 ffiM  1
/ 2
fl
Further we put (7.2.9)
4>(r,a,p;x) = {x + iyr''ixiy)f
(xeU, y>0, a, peC).
If we write z = x + iy (eH), then (7.2.10)
S(z;a,/?)=
f
(y, a, p; x + n).
n = — 00
We wish to apply the Poisson summation formula to the righthand side. When Re(a + j?)> 1, ;; a, ft x) belongs to L^ (R) as a function of x. Therefore there exists the Fourier transform of O, Re(a + )5)> 1).
276
7. Eisenstein Series
This is a holomorphic function of a and P on the domain {(a, j?)GCxCRe(a + j5)>l}, since the righthand side is uniformly convergent on any compact subset of the domain. Now the Poisson summation formula implies (7.2.12)
f n=
(l>(y;oi,P;xhn)=  00
f
^^(^(y; a, ft n).
n=oo
To justify this formula, it is sufficient to verify the following two conditions: 00
(7.2.13)
YJ \^(y*'> oc, P; x + n)\
is uniformly convergent for x;
n = — 00 00
(7.2.14)
X
l^(>'' 0}
and call it the right half plane. If Zj, Z2 eH', then (z,Z2r = zlzl
(aeC)
by definition (see (3.2.1)). We define the confluent hypergeometric function G{Z; OC, P) by (7.2.16)
(T(Z; a, ft = J e'"(u\If^
u^''du
(z, j^eH', a e C )
0
and put (7.2.17)
r](y; oc, P;t) = ]
e'"(u^tY'{uty'du
\t\
(y>0,teU,oi,
PeC).
We see easily the following properties: (7.2.18)
ifriiy; a, ft t) is convergent, then so is rj{y; ft a; —t) and r](y;cc, P; t) == r](y; P, ot; t);
(7.2.19)
ifa>0
and r](ay; a, ft t) is convergent, then so is rj{y; a, ft at) and rjiay; a, ft t) = fl^"^ ri(y; a, ft at).
As for the convergence of the functions rj and 0, then rj(y; a, ft t) is uniformly convergent on any compact subset ofCxH'asa function of a and p. (2) The function G(Z\ OL, ft is uniformly convergent on any compact subset of H xCxH.
§7.2, Analytic Continuation of Eisenstein Series
277
(3) //(>;, a, P)eU+ x C x H', then Proof. First we prove (2). Let The an arbitrary compact subset of H' x C x H'. Put p = min{Re(z) I (z, a, i8)er}, ^ = min{Re(iS)(z,a,i?)er}. Let p' be a number satisfying 0 1. Now we see i(y;oi,P;t)=
J e^^^^'^x + iyy^^ixiyy^
dx
— 00
 00
Since y±ixe H', we see easily
(i{y±ix)y'' =
r%y±ixy\
Therefore (7.2.20)
^(y;oL,P;t) = i^'
J ^2""^(};ix)"(}; + ix)^f/x. — 00
Assume that Re (a) > 0 and Re(j?) > 1. We put s = oi, a = y±ix in (3.2.5), then (7.2.21)
(y±ix)« = r(a)i J e(M27rO^i ~ [0
(M>27rr), (M ^ Int).
Now we assume t ^ 0. Then by (7.2.23) and (7.2.27), we see for Re(a) > 0, Re(jS) > 1, (7.2.28)
^(y, a, ft t) = i^«27rr(a)^r(i5)^
J
e23'("«OM«i(M_27rf)^i^u
u> 2nt
= i^^'inriay^ripy^ j e^y"(u{ntyHunt)^'du nt
= if'2nr(a)'r(p)'ri(2r,
a, p; nt).
We note that if t > 0 then i(y,ix, P; t) is continued to a holomorphic function of a
§7.2. Analytic Continuation of Eisenstein Series
279
and i5 on C X H' by (7.2.28) and Lemma 7.2.1(1). If t < 0, then (7.2.29)
ay; oi. ft t) = (ly^^y;
ft
a; r)
as long as Re(a + j5) > 1 by definition. Therefore if we define ^y; a, ft t) for (a, j5)6H' X C by (7.2.29), it is a meromorphic function and (7.2.30)
Uy; a, ft t) = (\y'i^^'2nr(oc)'r(P)'rj{2y; = i^''2nr(oi)'r(P)'ri(2y;
ft
a; 7rr)
a, ft Tir).
Therefore the case when r < 0 is also proved.
D
When r = 0, (^ is explicitly calculated in the following Lemma 7.2.3. The function ^y; cc,ft0) is continued analytically to a meromorphic function on C x C and satisfies
ay;a,P;0) = i^^'2nr{o^)'r{p)'r(oi^pi)(2yy'p, Proof When Re (a + j5) > 1, we see rj(y;a,P;0)=
] e^^u^^^P^du 0
= r(a +
pl)y l  a  / J
Since the righthand side is a meromorphic function on C x C, we obtain the analytic continuation of the function by Lemma 7.2.2. D We shall calculate ^(y; a, ft t) for t ^ 0. The calculation of ^(y; a, ft t) is reduced to that of t](y; a, ft 1) by Lemma 7.2.2, (7.2.18) and (7.2.19). Furthermore by Lemma 7.2.1(3), we have only to consider o(z; a, P). Now we put (7.2.31)
co(z; a, P) = r(p)'2^(7(2;
a, p)
((z, a, p)eU' x C x H').
By Lemma 7.2.1(2), a)(z; a, p) is a holomorphic function on H' x C x H'. Theorem 7.2.4. The function co(z; oc, P) is continued analytically to a holomorphic function on H' x C x C and satisfies: (1) a ) ( z ; l  f t l  a ) = a;(z;a,i9); (2) a;(z;a,i5) = (  l ) W ^ ( e  ^ z  ^ c o ( z ; a  n , i 5 ) )
{neN).
Proof For a, )?eH' and Z6IR+ , we see (7.2.32)
r(P)G{z\
\P,OL)
e'T(p){u+\)^u^^du.
= ] 0
Since r{P)(u^\)^
= ] e'^^^'^^^v^^dv,
280
7. Eisenstein Series
we have (7.2.33)
r(p)(7{z; lj?, a) = J e'"u°''du ] e''^"'^^v^Uv. 0
0
The integral is absolutely convergent, so that we can exchange the order of integration, and therefore, 0
0
0
= r(a)z^«(7(z;la,i5). Since r(p)(7{z;l—p,(x) and r{c()z^~"(7{z;l—<x,p) are holomorphic on H' X H' X H', they coincide with each other there. Thus we obtain (7.2.34)
(o{z; 1  ft 1  a) = co{z; a, P) (zeH',Re(a)< 1, Re(j?) > 0).
Define (o(z; a, P) for a e C with Re(a) < 1 and j? e C by the lefthand side of (7.2.34). Then it is a holomorphic function on H' x {aRe(a) < 1} x C, which coincides with the original (o(z; OL, P) when Re(^) > 0. Now when (z, a, j5)eH' x H' x H', the integral of the righthand side of (7.2.16) is uniformly convergent with respect to z on any compact subset of H'. Therefore we obtain (7.2.35)
^(e'(Tiz; a, P)) = (\fe'G(z
oc + n,P)
((z,a,i5)eH' X H' x H'), so that the continued function ca(z; a, P) also satisfies (7.2.36)
^ (e^zPco(z; a, P)) = { Ife'zPcoiz; a + n, i?) ((z, a, P)eW X {oi\ Re(a) < 1} x C).
For any (z, a, j?)eH' x C x C, take a positive integer n so that Re(a) < n + 1 and define co(z; a, P) by (7.2.37)
co(z; a, P) = (  l)"e^z^ ^
(^^z^ca(z; a  n, ^
then CO is a holomorphic function on H' x C x C. Now we have
D
§7.2. Analytic Continuation of Eisenstein Series
281
Theorem 7.2.5. (1) For each y > 0, as a function ofoc and j5, ^(y; a, p; t) is analytically continued to a meromorphic function on C x C which is expressed as (i^%2nYr{oc)\2y)^t''e^''y'o)i4nyt;
a, p)
i{y;oi.P;t) =  i^'{2nr^^r(oi)'rm'rioi^pi)i4nyY^p [j^"(27cfr(j5)^(2y)«rK^^2">'l'lft)(4;r};r;fta)
{t > 0),
(r = 0), (r < 0).
(2) As a function of a and j5, ^{y; a, j5; t) is holomorphic onC x Cfor t ^0, and r(a + j5  1)" ^ i(y; a, P; 0) is holomorphic on C x C. Proof The expression of ^(y; cc, P;t) in (1) is straightforward from Lemma 7,2.1, Lemma 7.2.2, Lemma 7.2.3, (7.2.18), (7.2.19) and (7.2.31) on a certain subdomain. Since the righthand side is meromorphic on C x C in each case, we obtain the analytic continuation. The second assertion is obvious from this expression and Lemma 7.2.4. D Lemma 7.2.6. For (z, a) e H' x C, we have co(z;a,0) = l. Proof If Re(a) > 0, then Lemma 7.2.4 and (7.2.31) imply that a>{z; 1 — a, 0) = ci>(z; 1, a)
= r(a)^zM^;i,a) 0
= 1. Since (o(z; a, 0) is holomorphic on a, we get the assertion.
D
Theorem 7.2.7. For any compact subset TofC x C, there exist positive numbers A and B satisfying \oj(y;oi,P)\SA(l¥y")
({y,oc,P)EU^ x T).
Proof First we prove the assertion when Tis contained in C x H'. Take a positive integer n so that Re(a) ^ n + 1 for any (a, p)eT. Let (a, p) be any point in T. Since RQ(P) > 0, we see by (7.2.16) and (7.2.31) that \(o{y;oi,P)\ S
\r(p)\'y^'^^^]ey"(u+l)"u'^'^^^'du 0
Since r(fe + Re(^)) and  r ( ^ )  " ' are continuous functions of )?, there exists a positive constant Ai satisfying k=0
282
7. Eisenstein Series
Therefore putting A = A^n and B = n, we obtain the assertion. Next we have to remove the assumption on T. If j5 e H', then taking integration by parts of the righthand side of (7.2.16), we have (7.2.38)
c7{y;a,p^l) =
Muhiplying y^'^^FiP\1)~^ (7.2.39)
y'{oil)Giy;oihP+l)^y'PcT(y;oi,P). to both sides, we obtain
co(y; a, p) = co{y; oi, ^ + l)\{loi)y'co(y;
aI
p^ l),
((a, i9)6 C x H ' ) . Since both sides of (7.2.39) are holomorphic functions of a and j?, (7.2.39) holds on C X C. Using (7.2.39) repeatedly, we obtain (7.2.40)
co(y; a, P)= Y.(j^]
 « = ! . ( : )
>'"'(!  a ) ( 2  a) • • • (/c a)co(y; ocKp + m).
For any compact subset T of C x C, we take a positive integer m so that {(a,j? + m)(a,i?)er}c=C x H'. Then we see easily the existence of the desired constants A and B from the result when T is contained in C x H'. D Now we can justify the Poisson summation formula and prove (7.2.12). Theorem 7.2.8. Let S(z; a, P) be the function defined in (7.2.7). Then it is analytically continued to a meromorphic function on C x C with respect to (a, p), and is expressed as (7.2.41)
S(z;oc,p) = ay;oi,P;0)+
f
^2«(^();; a, ft m) {z = x + iyeU).
m = — 00
The ^immx^^y.^ a, P; m) is convergent uniformly and absolutely on any compact subset o/H x C x C. Further r ( a + P — l)~^S{z; a, P) is an entire function of IX and p. Proof Let T be a compact subset of H x C x C. By Theorem 7.2.5, there exist positive constants C j , /? and q such that (7.2.42)
\iiy; a, P; m)\ ^ C^m^e^'''''"\w{^nym\ a, P)\
for any point (z, oc, p)eT and any positive integer m. We also see by Theorem 7.2.7 the existence of a positive constant C2 satisfying \co(4nym; (x, p)\ < C2 for any (z, cc, P)eT and any positive integer m. Therefore we have (7.2.43)
f
e2"''"^(^(>;;a,ftm)^CiC2 £ m^e'^^^'".
§7.2. Analytic Continuation of Eisenstein Series
283
Since the righthand side is clearly convergent, the lefthand side is uniformly convergent on T. We can also similarly prove that Yum=ao [^^"'"""^(y; a, jS;'w)! is uniformly convergent on T. Hence (7.2.14) is verified for a and ^ with Re(a + jS) > 1. Therefore the Poisson summation formula is appHcable, and (7.2.10) and (7.2.12) imply that (7.2.41) holds for a and ^ with Re(a + j?)> 1. Since Zm= 1 ^^"'""'^(}^; a, P\ m) and ^ ^ i _ oo e^"'"''